lustre/lfsck/lfsck_namespace.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2012, 2013, Intel Corporation.
  24  */
  25 /*
  26  * lustre/lfsck/lfsck_namespace.c
  27  *
  28  * Author: Fan, Yong <fan.yong@intel.com>
  29  */
  30
  31 #define DEBUG_SUBSYSTEM S_LFSCK
  32
  33 #include <lustre/lustre_idl.h>
  34 #include <lu_object.h>
  35 #include <dt_object.h>
  36 #include <md_object.h>
  37 #include <lustre_fid.h>
  38 #include <lustre_lib.h>
  39 #include <lustre_net.h>
  40 #include <lustre/lustre_user.h>
  41
  42 #include "lfsck_internal.h"
  43
  44 #define LFSCK_NAMESPACE_MAGIC   0xA0629D03
  45
  46 enum lfsck_nameentry_check {
  47         LFSCK_NAMEENTRY_DEAD            = 1, /* The object has been unlinked. */
  48         LFSCK_NAMEENTRY_REMOVED         = 2, /* The entry has been removed. */
  49         LFSCK_NAMEENTRY_RECREATED       = 3, /* The entry has been recreated. */
  50 };
  51
  52 static const char lfsck_namespace_name[] = "lfsck_namespace";
  53
  54 static struct lfsck_namespace_req *
  55 lfsck_namespace_assistant_req_init(struct lfsck_instance *lfsck,
  56                                    struct lu_dirent *ent, __u16 type)
  57 {
  58         struct lfsck_namespace_req *lnr;
  59         int                         size;
  60
  61         size = sizeof(*lnr) + (ent->lde_namelen & ~3) + 4;
  62         OBD_ALLOC(lnr, size);
  63         if (lnr == NULL)
  64                 return ERR_PTR(-ENOMEM);
  65
  66         INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
  67         lnr->lnr_obj = lfsck_object_get(lfsck->li_obj_dir);
  68         lnr->lnr_lmv = lfsck_lmv_get(lfsck->li_lmv);
  69         lnr->lnr_fid = ent->lde_fid;
  70         lnr->lnr_oit_cookie = lfsck->li_pos_current.lp_oit_cookie;
  71         lnr->lnr_dir_cookie = ent->lde_hash;
  72         lnr->lnr_attr = ent->lde_attrs;
  73         lnr->lnr_size = size;
  74         lnr->lnr_type = type;
  75         lnr->lnr_namelen = ent->lde_namelen;
  76         memcpy(lnr->lnr_name, ent->lde_name, ent->lde_namelen);
  77
  78         return lnr;
  79 }
  80
  81 static void lfsck_namespace_assistant_req_fini(const struct lu_env *env,
  82                                                struct lfsck_assistant_req *lar)
  83 {
  84         struct lfsck_namespace_req *lnr =
  85                         container_of0(lar, struct lfsck_namespace_req, lnr_lar);
  86
  87         if (lnr->lnr_lmv != NULL)
  88                 lfsck_lmv_put(env, lnr->lnr_lmv);
  89
  90         lu_object_put(env, &lnr->lnr_obj->do_lu);
  91         OBD_FREE(lnr, lnr->lnr_size);
  92 }
  93
  94 static void lfsck_namespace_le_to_cpu(struct lfsck_namespace *dst,
  95                                       struct lfsck_namespace *src)
  96 {
  97         dst->ln_magic = le32_to_cpu(src->ln_magic);
  98         dst->ln_status = le32_to_cpu(src->ln_status);
  99         dst->ln_flags = le32_to_cpu(src->ln_flags);
 100         dst->ln_success_count = le32_to_cpu(src->ln_success_count);
 101         dst->ln_run_time_phase1 = le32_to_cpu(src->ln_run_time_phase1);
 102         dst->ln_run_time_phase2 = le32_to_cpu(src->ln_run_time_phase2);
 103         dst->ln_time_last_complete = le64_to_cpu(src->ln_time_last_complete);
 104         dst->ln_time_latest_start = le64_to_cpu(src->ln_time_latest_start);
 105         dst->ln_time_last_checkpoint =
 106                                 le64_to_cpu(src->ln_time_last_checkpoint);
 107         lfsck_position_le_to_cpu(&dst->ln_pos_latest_start,
 108                                  &src->ln_pos_latest_start);
 109         lfsck_position_le_to_cpu(&dst->ln_pos_last_checkpoint,
 110                                  &src->ln_pos_last_checkpoint);
 111         lfsck_position_le_to_cpu(&dst->ln_pos_first_inconsistent,
 112                                  &src->ln_pos_first_inconsistent);
 113         dst->ln_items_checked = le64_to_cpu(src->ln_items_checked);
 114         dst->ln_items_repaired = le64_to_cpu(src->ln_items_repaired);
 115         dst->ln_items_failed = le64_to_cpu(src->ln_items_failed);
 116         dst->ln_dirs_checked = le64_to_cpu(src->ln_dirs_checked);
 117         dst->ln_objs_checked_phase2 = le64_to_cpu(src->ln_objs_checked_phase2);
 118         dst->ln_objs_repaired_phase2 =
 119                                 le64_to_cpu(src->ln_objs_repaired_phase2);
 120         dst->ln_objs_failed_phase2 = le64_to_cpu(src->ln_objs_failed_phase2);
 121         dst->ln_objs_nlink_repaired = le64_to_cpu(src->ln_objs_nlink_repaired);
 122         fid_le_to_cpu(&dst->ln_fid_latest_scanned_phase2,
 123                       &src->ln_fid_latest_scanned_phase2);
 124         dst->ln_dirent_repaired = le64_to_cpu(src->ln_dirent_repaired);
 125         dst->ln_linkea_repaired = le64_to_cpu(src->ln_linkea_repaired);
 126         dst->ln_mul_linked_checked = le64_to_cpu(src->ln_mul_linked_checked);
 127         dst->ln_mul_linked_repaired = le64_to_cpu(src->ln_mul_linked_repaired);
 128         dst->ln_unknown_inconsistency =
 129                                 le64_to_cpu(src->ln_unknown_inconsistency);
 130         dst->ln_unmatched_pairs_repaired =
 131                                 le64_to_cpu(src->ln_unmatched_pairs_repaired);
 132         dst->ln_dangling_repaired = le64_to_cpu(src->ln_dangling_repaired);
 133         dst->ln_mul_ref_repaired = le64_to_cpu(src->ln_mul_ref_repaired);
 134         dst->ln_bad_type_repaired = le64_to_cpu(src->ln_bad_type_repaired);
 135         dst->ln_lost_dirent_repaired =
 136                                 le64_to_cpu(src->ln_lost_dirent_repaired);
 137         dst->ln_striped_dirs_scanned =
 138                                 le64_to_cpu(src->ln_striped_dirs_scanned);
 139         dst->ln_striped_dirs_repaired =
 140                                 le64_to_cpu(src->ln_striped_dirs_repaired);
 141         dst->ln_striped_dirs_failed =
 142                                 le64_to_cpu(src->ln_striped_dirs_failed);
 143         dst->ln_striped_dirs_disabled =
 144                                 le64_to_cpu(src->ln_striped_dirs_disabled);
 145         dst->ln_striped_dirs_skipped =
 146                                 le64_to_cpu(src->ln_striped_dirs_skipped);
 147         dst->ln_striped_shards_scanned =
 148                                 le64_to_cpu(src->ln_striped_shards_scanned);
 149         dst->ln_striped_shards_repaired =
 150                                 le64_to_cpu(src->ln_striped_shards_repaired);
 151         dst->ln_striped_shards_failed =
 152                                 le64_to_cpu(src->ln_striped_shards_failed);
 153         dst->ln_striped_shards_skipped =
 154                                 le64_to_cpu(src->ln_striped_shards_skipped);
 155         dst->ln_name_hash_repaired = le64_to_cpu(src->ln_name_hash_repaired);
 156         dst->ln_local_lpf_scanned = le64_to_cpu(src->ln_local_lpf_scanned);
 157         dst->ln_local_lpf_moved = le64_to_cpu(src->ln_local_lpf_moved);
 158         dst->ln_local_lpf_skipped = le64_to_cpu(src->ln_local_lpf_skipped);
 159         dst->ln_local_lpf_failed = le64_to_cpu(src->ln_local_lpf_failed);
 160         dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size);
 161 }
 162
 163 static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
 164                                       struct lfsck_namespace *src)
 165 {
 166         dst->ln_magic = cpu_to_le32(src->ln_magic);
 167         dst->ln_status = cpu_to_le32(src->ln_status);
 168         dst->ln_flags = cpu_to_le32(src->ln_flags);
 169         dst->ln_success_count = cpu_to_le32(src->ln_success_count);
 170         dst->ln_run_time_phase1 = cpu_to_le32(src->ln_run_time_phase1);
 171         dst->ln_run_time_phase2 = cpu_to_le32(src->ln_run_time_phase2);
 172         dst->ln_time_last_complete = cpu_to_le64(src->ln_time_last_complete);
 173         dst->ln_time_latest_start = cpu_to_le64(src->ln_time_latest_start);
 174         dst->ln_time_last_checkpoint =
 175                                 cpu_to_le64(src->ln_time_last_checkpoint);
 176         lfsck_position_cpu_to_le(&dst->ln_pos_latest_start,
 177                                  &src->ln_pos_latest_start);
 178         lfsck_position_cpu_to_le(&dst->ln_pos_last_checkpoint,
 179                                  &src->ln_pos_last_checkpoint);
 180         lfsck_position_cpu_to_le(&dst->ln_pos_first_inconsistent,
 181                                  &src->ln_pos_first_inconsistent);
 182         dst->ln_items_checked = cpu_to_le64(src->ln_items_checked);
 183         dst->ln_items_repaired = cpu_to_le64(src->ln_items_repaired);
 184         dst->ln_items_failed = cpu_to_le64(src->ln_items_failed);
 185         dst->ln_dirs_checked = cpu_to_le64(src->ln_dirs_checked);
 186         dst->ln_objs_checked_phase2 = cpu_to_le64(src->ln_objs_checked_phase2);
 187         dst->ln_objs_repaired_phase2 =
 188                                 cpu_to_le64(src->ln_objs_repaired_phase2);
 189         dst->ln_objs_failed_phase2 = cpu_to_le64(src->ln_objs_failed_phase2);
 190         dst->ln_objs_nlink_repaired = cpu_to_le64(src->ln_objs_nlink_repaired);
 191         fid_cpu_to_le(&dst->ln_fid_latest_scanned_phase2,
 192                       &src->ln_fid_latest_scanned_phase2);
 193         dst->ln_dirent_repaired = cpu_to_le64(src->ln_dirent_repaired);
 194         dst->ln_linkea_repaired = cpu_to_le64(src->ln_linkea_repaired);
 195         dst->ln_mul_linked_checked = cpu_to_le64(src->ln_mul_linked_checked);
 196         dst->ln_mul_linked_repaired = cpu_to_le64(src->ln_mul_linked_repaired);
 197         dst->ln_unknown_inconsistency =
 198                                 cpu_to_le64(src->ln_unknown_inconsistency);
 199         dst->ln_unmatched_pairs_repaired =
 200                                 cpu_to_le64(src->ln_unmatched_pairs_repaired);
 201         dst->ln_dangling_repaired = cpu_to_le64(src->ln_dangling_repaired);
 202         dst->ln_mul_ref_repaired = cpu_to_le64(src->ln_mul_ref_repaired);
 203         dst->ln_bad_type_repaired = cpu_to_le64(src->ln_bad_type_repaired);
 204         dst->ln_lost_dirent_repaired =
 205                                 cpu_to_le64(src->ln_lost_dirent_repaired);
 206         dst->ln_striped_dirs_scanned =
 207                                 cpu_to_le64(src->ln_striped_dirs_scanned);
 208         dst->ln_striped_dirs_repaired =
 209                                 cpu_to_le64(src->ln_striped_dirs_repaired);
 210         dst->ln_striped_dirs_failed =
 211                                 cpu_to_le64(src->ln_striped_dirs_failed);
 212         dst->ln_striped_dirs_disabled =
 213                                 cpu_to_le64(src->ln_striped_dirs_disabled);
 214         dst->ln_striped_dirs_skipped =
 215                                 cpu_to_le64(src->ln_striped_dirs_skipped);
 216         dst->ln_striped_shards_scanned =
 217                                 cpu_to_le64(src->ln_striped_shards_scanned);
 218         dst->ln_striped_shards_repaired =
 219                                 cpu_to_le64(src->ln_striped_shards_repaired);
 220         dst->ln_striped_shards_failed =
 221                                 cpu_to_le64(src->ln_striped_shards_failed);
 222         dst->ln_striped_shards_skipped =
 223                                 cpu_to_le64(src->ln_striped_shards_skipped);
 224         dst->ln_name_hash_repaired = cpu_to_le64(src->ln_name_hash_repaired);
 225         dst->ln_local_lpf_scanned = cpu_to_le64(src->ln_local_lpf_scanned);
 226         dst->ln_local_lpf_moved = cpu_to_le64(src->ln_local_lpf_moved);
 227         dst->ln_local_lpf_skipped = cpu_to_le64(src->ln_local_lpf_skipped);
 228         dst->ln_local_lpf_failed = cpu_to_le64(src->ln_local_lpf_failed);
 229         dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size);
 230 }
 231
 232 static void lfsck_namespace_record_failure(const struct lu_env *env,
 233                                            struct lfsck_instance *lfsck,
 234                                            struct lfsck_namespace *ns)
 235 {
 236         struct lfsck_position pos;
 237
 238         ns->ln_items_failed++;
 239         lfsck_pos_fill(env, lfsck, &pos, false);
 240         if (lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent) ||
 241             lfsck_pos_is_eq(&pos, &ns->ln_pos_first_inconsistent) < 0) {
 242                 ns->ln_pos_first_inconsistent = pos;
 243
 244                 CDEBUG(D_LFSCK, "%s: namespace LFSCK hit first non-repaired "
 245                        "inconsistency at the pos ["LPU64", "DFID", "LPX64"]\n",
 246                        lfsck_lfsck2name(lfsck),
 247                        ns->ln_pos_first_inconsistent.lp_oit_cookie,
 248                        PFID(&ns->ln_pos_first_inconsistent.lp_dir_parent),
 249                        ns->ln_pos_first_inconsistent.lp_dir_cookie);
 250         }
 251 }
 252
 253 /**
 254  * Load the MDT bitmap from the lfsck_namespace trace file.
 255  *
 256  * \param[in] env       pointer to the thread context
 257  * \param[in] com       pointer to the lfsck component
 258  *
 259  * \retval              0 for success
 260  * \retval              negative error number on failure or data corruption
 261  */
 262 static int lfsck_namespace_load_bitmap(const struct lu_env *env,
 263                                        struct lfsck_component *com)
 264 {
 265         struct dt_object                *obj    = com->lc_obj;
 266         struct lfsck_assistant_data     *lad    = com->lc_data;
 267         struct lfsck_namespace          *ns     = com->lc_file_ram;
 268         cfs_bitmap_t                    *bitmap = lad->lad_bitmap;
 269         ssize_t                          size;
 270         __u32                            nbits;
 271         int                              rc;
 272         ENTRY;
 273
 274         if (com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size >
 275             ns->ln_bitmap_size)
 276                 nbits = com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size;
 277         else
 278                 nbits = ns->ln_bitmap_size;
 279
 280         if (unlikely(nbits < BITS_PER_LONG))
 281                 nbits = BITS_PER_LONG;
 282
 283         if (nbits > bitmap->size) {
 284                 __u32 new_bits = bitmap->size;
 285                 cfs_bitmap_t *new_bitmap;
 286
 287                 while (new_bits < nbits)
 288                         new_bits <<= 1;
 289
 290                 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
 291                 if (new_bitmap == NULL)
 292                         RETURN(-ENOMEM);
 293
 294                 lad->lad_bitmap = new_bitmap;
 295                 CFS_FREE_BITMAP(bitmap);
 296                 bitmap = new_bitmap;
 297         }
 298
 299         if (ns->ln_bitmap_size == 0) {
 300                 lad->lad_incomplete = 0;
 301                 CFS_RESET_BITMAP(bitmap);
 302
 303                 RETURN(0);
 304         }
 305
 306         size = (ns->ln_bitmap_size + 7) >> 3;
 307         rc = dt_xattr_get(env, obj,
 308                           lfsck_buf_get(env, bitmap->data, size),
 309                           XATTR_NAME_LFSCK_BITMAP, BYPASS_CAPA);
 310         if (rc != size)
 311                 RETURN(rc >= 0 ? -EINVAL : rc);
 312
 313         if (cfs_bitmap_check_empty(bitmap))
 314                 lad->lad_incomplete = 0;
 315         else
 316                 lad->lad_incomplete = 1;
 317
 318         RETURN(0);
 319 }
 320
 321 /**
 322  * \retval +ve: the lfsck_namespace is broken, the caller should reset it.
 323  * \retval 0: succeed.
 324  * \retval -ve: failed cases.
 325  */
 326 static int lfsck_namespace_load(const struct lu_env *env,
 327                                 struct lfsck_component *com)
 328 {
 329         int len = com->lc_file_size;
 330         int rc;
 331
 332         rc = dt_xattr_get(env, com->lc_obj,
 333                           lfsck_buf_get(env, com->lc_file_disk, len),
 334                           XATTR_NAME_LFSCK_NAMESPACE, BYPASS_CAPA);
 335         if (rc == len) {
 336                 struct lfsck_namespace *ns = com->lc_file_ram;
 337
 338                 lfsck_namespace_le_to_cpu(ns,
 339                                 (struct lfsck_namespace *)com->lc_file_disk);
 340                 if (ns->ln_magic != LFSCK_NAMESPACE_MAGIC) {
 341                         CDEBUG(D_LFSCK, "%s: invalid lfsck_namespace magic "
 342                                "%#x != %#x\n", lfsck_lfsck2name(com->lc_lfsck),
 343                                ns->ln_magic, LFSCK_NAMESPACE_MAGIC);
 344                         rc = 1;
 345                 } else {
 346                         rc = 0;
 347                 }
 348         } else if (rc != -ENODATA) {
 349                 CDEBUG(D_LFSCK, "%s: fail to load lfsck_namespace, "
 350                        "expected = %d: rc = %d\n",
 351                        lfsck_lfsck2name(com->lc_lfsck), len, rc);
 352                 if (rc >= 0)
 353                         rc = 1;
 354         }
 355         return rc;
 356 }
 357
 358 static int lfsck_namespace_store(const struct lu_env *env,
 359                                  struct lfsck_component *com)
 360 {
 361         struct dt_object                *obj    = com->lc_obj;
 362         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 363         struct lfsck_namespace          *ns     = com->lc_file_ram;
 364         struct lfsck_assistant_data     *lad    = com->lc_data;
 365         cfs_bitmap_t                    *bitmap = NULL;
 366         struct thandle                  *handle;
 367         __u32                            nbits  = 0;
 368         int                              len    = com->lc_file_size;
 369         int                              rc;
 370         ENTRY;
 371
 372         if (lad != NULL) {
 373                 bitmap = lad->lad_bitmap;
 374                 nbits = bitmap->size;
 375
 376                 LASSERT(nbits > 0);
 377                 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
 378         }
 379
 380         ns->ln_bitmap_size = nbits;
 381         lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk,
 382                                   ns);
 383         handle = dt_trans_create(env, lfsck->li_bottom);
 384         if (IS_ERR(handle))
 385                 GOTO(log, rc = PTR_ERR(handle));
 386
 387         rc = dt_declare_xattr_set(env, obj,
 388                                   lfsck_buf_get(env, com->lc_file_disk, len),
 389                                   XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
 390         if (rc != 0)
 391                 GOTO(out, rc);
 392
 393         if (bitmap != NULL) {
 394                 rc = dt_declare_xattr_set(env, obj,
 395                                 lfsck_buf_get(env, bitmap->data, nbits >> 3),
 396                                 XATTR_NAME_LFSCK_BITMAP, 0, handle);
 397                 if (rc != 0)
 398                         GOTO(out, rc);
 399         }
 400
 401         rc = dt_trans_start_local(env, lfsck->li_bottom, handle);
 402         if (rc != 0)
 403                 GOTO(out, rc);
 404
 405         rc = dt_xattr_set(env, obj,
 406                           lfsck_buf_get(env, com->lc_file_disk, len),
 407                           XATTR_NAME_LFSCK_NAMESPACE, 0, handle, BYPASS_CAPA);
 408         if (rc == 0 && bitmap != NULL)
 409                 rc = dt_xattr_set(env, obj,
 410                                   lfsck_buf_get(env, bitmap->data, nbits >> 3),
 411                                   XATTR_NAME_LFSCK_BITMAP, 0, handle,
 412                                   BYPASS_CAPA);
 413
 414         GOTO(out, rc);
 415
 416 out:
 417         dt_trans_stop(env, lfsck->li_bottom, handle);
 418
 419 log:
 420         if (rc != 0)
 421                 CDEBUG(D_LFSCK, "%s: fail to store lfsck_namespace: rc = %d\n",
 422                        lfsck_lfsck2name(lfsck), rc);
 423         return rc;
 424 }
 425
 426 static int lfsck_namespace_init(const struct lu_env *env,
 427                                 struct lfsck_component *com)
 428 {
 429         struct lfsck_namespace *ns = com->lc_file_ram;
 430         int rc;
 431
 432         memset(ns, 0, sizeof(*ns));
 433         ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
 434         ns->ln_status = LS_INIT;
 435         down_write(&com->lc_sem);
 436         rc = lfsck_namespace_store(env, com);
 437         up_write(&com->lc_sem);
 438         return rc;
 439 }
 440
 441 /**
 442  * Update the namespace LFSCK trace file for the given @fid
 443  *
 444  * \param[in] env       pointer to the thread context
 445  * \param[in] com       pointer to the lfsck component
 446  * \param[in] fid       the fid which flags to be updated in the lfsck
 447  *                      trace file
 448  * \param[in] add       true if add new flags, otherwise remove flags
 449  *
 450  * \retval              0 for succeed or nothing to be done
 451  * \retval              negative error number on failure
 452  */
 453 int lfsck_namespace_trace_update(const struct lu_env *env,
 454                                  struct lfsck_component *com,
 455                                  const struct lu_fid *fid,
 456                                  const __u8 flags, bool add)
 457 {
 458         struct lfsck_instance   *lfsck  = com->lc_lfsck;
 459         struct dt_object        *obj    = com->lc_obj;
 460         struct lu_fid           *key    = &lfsck_env_info(env)->lti_fid3;
 461         struct dt_device        *dev    = lfsck->li_bottom;
 462         struct thandle          *th     = NULL;
 463         int                      rc     = 0;
 464         __u8                     old    = 0;
 465         __u8                     new    = 0;
 466         ENTRY;
 467
 468         LASSERT(flags != 0);
 469
 470         down_write(&com->lc_sem);
 471         fid_cpu_to_be(key, fid);
 472         rc = dt_lookup(env, obj, (struct dt_rec *)&old,
 473                        (const struct dt_key *)key, BYPASS_CAPA);
 474         if (rc == -ENOENT) {
 475                 if (!add)
 476                         GOTO(unlock, rc = 0);
 477
 478                 old = 0;
 479                 new = flags;
 480         } else if (rc == 0) {
 481                 if (add) {
 482                         if ((old & flags) == flags)
 483                                 GOTO(unlock, rc = 0);
 484
 485                         new = old | flags;
 486                 } else {
 487                         if ((old & flags) == 0)
 488                                 GOTO(unlock, rc = 0);
 489
 490                         new = old & ~flags;
 491                 }
 492         } else {
 493                 GOTO(log, rc);
 494         }
 495
 496         th = dt_trans_create(env, dev);
 497         if (IS_ERR(th))
 498                 GOTO(log, rc = PTR_ERR(th));
 499
 500         if (old != 0) {
 501                 rc = dt_declare_delete(env, obj,
 502                                        (const struct dt_key *)key, th);
 503                 if (rc != 0)
 504                         GOTO(log, rc);
 505         }
 506
 507         if (new != 0) {
 508                 rc = dt_declare_insert(env, obj,
 509                                        (const struct dt_rec *)&new,
 510                                        (const struct dt_key *)key, th);
 511                 if (rc != 0)
 512                         GOTO(log, rc);
 513         }
 514
 515         rc = dt_trans_start_local(env, dev, th);
 516         if (rc != 0)
 517                 GOTO(log, rc);
 518
 519         if (old != 0) {
 520                 rc = dt_delete(env, obj, (const struct dt_key *)key,
 521                                th, BYPASS_CAPA);
 522                 if (rc != 0)
 523                         GOTO(log, rc);
 524         }
 525
 526         if (new != 0) {
 527                 rc = dt_insert(env, obj, (const struct dt_rec *)&new,
 528                                (const struct dt_key *)key, th, BYPASS_CAPA, 1);
 529                 if (rc != 0)
 530                         GOTO(log, rc);
 531         }
 532
 533         GOTO(log, rc);
 534
 535 log:
 536         if (th != NULL && !IS_ERR(th))
 537                 dt_trans_stop(env, dev, th);
 538
 539         CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the "
 540                "trace file, flags %x, old %x, new %x: rc = %d\n",
 541                lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid),
 542                (__u32)flags, (__u32)old, (__u32)new, rc);
 543
 544 unlock:
 545         up_write(&com->lc_sem);
 546
 547         return rc;
 548 }
 549
 550 static int lfsck_namespace_check_exist(const struct lu_env *env,
 551                                        struct dt_object *dir,
 552                                        struct dt_object *obj, const char *name)
 553 {
 554         struct lu_fid    *fid = &lfsck_env_info(env)->lti_fid;
 555         int               rc;
 556         ENTRY;
 557
 558         if (unlikely(lfsck_is_dead_obj(obj)))
 559                 RETURN(LFSCK_NAMEENTRY_DEAD);
 560
 561         rc = dt_lookup(env, dir, (struct dt_rec *)fid,
 562                        (const struct dt_key *)name, BYPASS_CAPA);
 563         if (rc == -ENOENT)
 564                 RETURN(LFSCK_NAMEENTRY_REMOVED);
 565
 566         if (rc < 0)
 567                 RETURN(rc);
 568
 569         if (!lu_fid_eq(fid, lfsck_dto2fid(obj)))
 570                 RETURN(LFSCK_NAMEENTRY_RECREATED);
 571
 572         RETURN(0);
 573 }
 574
 575 static int lfsck_declare_namespace_exec_dir(const struct lu_env *env,
 576                                             struct dt_object *obj,
 577                                             struct thandle *handle)
 578 {
 579         int rc;
 580
 581         /* For destroying all invalid linkEA entries. */
 582         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
 583         if (rc != 0)
 584                 return rc;
 585
 586         /* For insert new linkEA entry. */
 587         rc = dt_declare_xattr_set(env, obj,
 588                         lfsck_buf_get_const(env, NULL, DEFAULT_LINKEA_SIZE),
 589                         XATTR_NAME_LINK, 0, handle);
 590         return rc;
 591 }
 592
 593 int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
 594                        struct linkea_data *ldata)
 595 {
 596         int rc;
 597
 598         if (ldata->ld_buf->lb_buf == NULL)
 599                 return -ENOMEM;
 600
 601         if (!dt_object_exists(obj))
 602                 return -ENOENT;
 603
 604         rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK, BYPASS_CAPA);
 605         if (rc == -ERANGE) {
 606                 /* Buf was too small, figure out what we need. */
 607                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK,
 608                                   BYPASS_CAPA);
 609                 if (rc <= 0)
 610                         return rc;
 611
 612                 lu_buf_realloc(ldata->ld_buf, rc);
 613                 if (ldata->ld_buf->lb_buf == NULL)
 614                         return -ENOMEM;
 615
 616                 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK,
 617                                   BYPASS_CAPA);
 618         }
 619
 620         if (rc > 0)
 621                 rc = linkea_init(ldata);
 622
 623         return rc;
 624 }
 625
 626 /**
 627  * Remove linkEA for the given object.
 628  *
 629  * The caller should take the ldlm lock before the calling.
 630  *
 631  * \param[in] env       pointer to the thread context
 632  * \param[in] com       pointer to the lfsck component
 633  * \param[in] obj       pointer to the dt_object to be handled
 634  *
 635  * \retval              0 for repaired cases
 636  * \retval              negative error number on failure
 637  */
 638 static int lfsck_namespace_links_remove(const struct lu_env *env,
 639                                         struct lfsck_component *com,
 640                                         struct dt_object *obj)
 641 {
 642         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 643         struct dt_device                *dev    = lfsck->li_bottom;
 644         struct thandle                  *th     = NULL;
 645         int                              rc     = 0;
 646         ENTRY;
 647
 648         LASSERT(dt_object_remote(obj) == 0);
 649
 650         th = dt_trans_create(env, dev);
 651         if (IS_ERR(th))
 652                 GOTO(log, rc = PTR_ERR(th));
 653
 654         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
 655         if (rc != 0)
 656                 GOTO(stop, rc);
 657
 658         rc = dt_trans_start_local(env, dev, th);
 659         if (rc != 0)
 660                 GOTO(stop, rc);
 661
 662         dt_write_lock(env, obj, 0);
 663         if (unlikely(lfsck_is_dead_obj(obj)))
 664                 GOTO(unlock, rc = -ENOENT);
 665
 666         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
 667                 GOTO(unlock, rc = 0);
 668
 669         rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th, BYPASS_CAPA);
 670
 671         GOTO(unlock, rc);
 672
 673 unlock:
 674         dt_write_unlock(env, obj);
 675
 676 stop:
 677         dt_trans_stop(env, dev, th);
 678
 679 log:
 680         CDEBUG(D_LFSCK, "%s: namespace LFSCK remove invalid linkEA "
 681                "for the object "DFID": rc = %d\n",
 682                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
 683
 684         if (rc == 0) {
 685                 struct lfsck_namespace *ns = com->lc_file_ram;
 686
 687                 ns->ln_flags |= LF_INCONSISTENT;
 688         }
 689
 690         return rc;
 691 }
 692
 693 static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
 694                              struct linkea_data *ldata, struct thandle *handle)
 695 {
 696         const struct lu_buf *buf = lfsck_buf_get_const(env,
 697                                                        ldata->ld_buf->lb_buf,
 698                                                        ldata->ld_leh->leh_len);
 699
 700         return dt_xattr_set(env, obj, buf, XATTR_NAME_LINK, 0, handle,
 701                             BYPASS_CAPA);
 702 }
 703
 704 static void lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
 705                                                 struct lu_name *cname,
 706                                                 struct lu_fid *pfid,
 707                                                 char *buf)
 708 {
 709         linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
 710         /* To guarantee the 'name' is terminated with '0'. */
 711         memcpy(buf, cname->ln_name, cname->ln_namelen);
 712         buf[cname->ln_namelen] = 0;
 713         cname->ln_name = buf;
 714 }
 715
 716 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
 717                                                struct lu_name *cname,
 718                                                struct lu_fid *pfid,
 719                                                bool remove)
 720 {
 721         struct link_ea_entry    *oldlee;
 722         int                      oldlen;
 723         int                      repeated = 0;
 724
 725         oldlee = ldata->ld_lee;
 726         oldlen = ldata->ld_reclen;
 727         linkea_next_entry(ldata);
 728         while (ldata->ld_lee != NULL) {
 729                 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
 730                                    ldata->ld_lee->lee_reclen[1];
 731                 if (unlikely(ldata->ld_reclen == oldlen &&
 732                              memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
 733                         repeated++;
 734                         if (!remove)
 735                                 break;
 736
 737                         linkea_del_buf(ldata, cname);
 738                 } else {
 739                         linkea_next_entry(ldata);
 740                 }
 741         }
 742         ldata->ld_lee = oldlee;
 743         ldata->ld_reclen = oldlen;
 744
 745         return repeated;
 746 }
 747
 748 /**
 749  * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
 750  *
 751  * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
 752  * with the given type to generate the name, the detailed rules for name
 753  * have been described as following.
 754  *
 755  * The function also generates the linkEA corresponding to the name entry
 756  * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
 757  *
 758  * \param[in] env       pointer to the thread context
 759  * \param[in] com       pointer to the lfsck component
 760  * \param[in] orphan    pointer to the orphan MDT-object
 761  * \param[in] infix     additional information for the orphan name, such as
 762  *                      the FID for original
 763  * \param[in] type      the type for describing why the orphan MDT-object is
 764  *                      created. The rules are as following:
 765  *
 766  *  type "D":           The MDT-object is a directory, it may knows its parent
 767  *                      but because there is no valid linkEA, the LFSCK cannot
 768  *                      know where to put it back to the namespace.
 769  *  type "O":           The MDT-object has no linkEA, and there is no name
 770  *                      entry that references the MDT-object.
 771  *
 772  *  type "S":           The orphan MDT-object is a shard of a striped directory
 773  *
 774  * \see lfsck_layout_recreate_parent() for more types.
 775  *
 776  * The orphan name will be like:
 777  * ${FID}-${infix}-${type}-${conflict_version}
 778  *
 779  * \param[out] count    if some others inserted some linkEA entries by race,
 780  *                      then return the linkEA entries count.
 781  *
 782  * \retval              positive number for repaired cases
 783  * \retval              0 if needs to repair nothing
 784  * \retval              negative error number on failure
 785  */
 786 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
 787                                          struct lfsck_component *com,
 788                                          struct dt_object *orphan,
 789                                          const char *infix, const char *type,
 790                                          int *count)
 791 {
 792         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 793         struct lu_name                  *cname  = &info->lti_name;
 794         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
 795         struct lu_fid                   *tfid   = &info->lti_fid5;
 796         struct lu_attr                  *la     = &info->lti_la3;
 797         const struct lu_fid             *cfid   = lfsck_dto2fid(orphan);
 798         const struct lu_fid             *pfid;
 799         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 800         struct dt_device                *dev    = lfsck->li_bottom;
 801         struct dt_object                *parent;
 802         struct thandle                  *th     = NULL;
 803         struct lustre_handle             plh    = { 0 };
 804         struct lustre_handle             clh    = { 0 };
 805         struct linkea_data               ldata  = { 0 };
 806         struct lu_buf                    linkea_buf;
 807         int                              namelen;
 808         int                              idx    = 0;
 809         int                              rc     = 0;
 810         bool                             exist  = false;
 811         ENTRY;
 812
 813         cname->ln_name = NULL;
 814         /* Create .lustre/lost+found/MDTxxxx when needed. */
 815         if (unlikely(lfsck->li_lpf_obj == NULL)) {
 816                 rc = lfsck_create_lpf(env, lfsck);
 817                 if (rc != 0)
 818                         GOTO(log, rc);
 819         }
 820
 821         parent = lfsck->li_lpf_obj;
 822         pfid = lfsck_dto2fid(parent);
 823
 824         /* Hold update lock on the parent to prevent others to access. */
 825         rc = lfsck_ibits_lock(env, lfsck, parent, &plh,
 826                               MDS_INODELOCK_UPDATE, LCK_EX);
 827         if (rc != 0)
 828                 GOTO(log, rc);
 829
 830         do {
 831                 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
 832                                    PFID(cfid), infix, type, idx++);
 833                 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
 834                                (const struct dt_key *)info->lti_key,
 835                                BYPASS_CAPA);
 836                 if (rc != 0 && rc != -ENOENT)
 837                         GOTO(log, rc);
 838
 839                 if (unlikely(rc == 0 && lu_fid_eq(cfid, tfid)))
 840                         exist = true;
 841         } while (rc == 0 && !exist);
 842
 843         cname->ln_name = info->lti_key;
 844         cname->ln_namelen = namelen;
 845         rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
 846         if (rc != 0)
 847                 GOTO(log, rc);
 848
 849         rc = linkea_add_buf(&ldata, cname, pfid);
 850         if (rc != 0)
 851                 GOTO(log, rc);
 852
 853         rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
 854                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP,
 855                               LCK_EX);
 856         if (rc != 0)
 857                 GOTO(log, rc);
 858
 859         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
 860                        ldata.ld_leh->leh_len);
 861         th = dt_trans_create(env, dev);
 862         if (IS_ERR(th))
 863                 GOTO(log, rc = PTR_ERR(th));
 864
 865         if (S_ISDIR(lfsck_object_type(orphan))) {
 866                 rc = dt_declare_delete(env, orphan,
 867                                        (const struct dt_key *)dotdot, th);
 868                 if (rc != 0)
 869                         GOTO(stop, rc);
 870
 871                 rec->rec_type = S_IFDIR;
 872                 rec->rec_fid = pfid;
 873                 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
 874                                        (const struct dt_key *)dotdot, th);
 875                 if (rc != 0)
 876                         GOTO(stop, rc);
 877         }
 878
 879         rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
 880                                   XATTR_NAME_LINK, 0, th);
 881         if (rc != 0)
 882                 GOTO(stop, rc);
 883
 884         if (!exist) {
 885                 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
 886                 rec->rec_fid = cfid;
 887                 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
 888                                        (const struct dt_key *)cname->ln_name,
 889                                        th);
 890                 if (rc != 0)
 891                         GOTO(stop, rc);
 892
 893                 if (S_ISDIR(rec->rec_type)) {
 894                         rc = dt_declare_ref_add(env, parent, th);
 895                         if (rc != 0)
 896                                 GOTO(stop, rc);
 897                 }
 898         }
 899
 900         memset(la, 0, sizeof(*la));
 901         la->la_ctime = cfs_time_current_sec();
 902         la->la_valid = LA_CTIME;
 903         rc = dt_declare_attr_set(env, orphan, la, th);
 904         if (rc != 0)
 905                 GOTO(stop, rc);
 906
 907         rc = dt_trans_start_local(env, dev, th);
 908         if (rc != 0)
 909                 GOTO(stop, rc);
 910
 911         dt_write_lock(env, orphan, 0);
 912         rc = lfsck_links_read(env, orphan, &ldata);
 913         if (likely((rc == -ENODATA) || (rc == -EINVAL) ||
 914                    (rc == 0 && ldata.ld_leh->leh_reccount == 0))) {
 915                 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
 916                         GOTO(unlock, rc = 1);
 917
 918                 if (S_ISDIR(lfsck_object_type(orphan))) {
 919                         rc = dt_delete(env, orphan,
 920                                        (const struct dt_key *)dotdot, th,
 921                                        BYPASS_CAPA);
 922                         if (rc != 0)
 923                                 GOTO(unlock, rc);
 924
 925                         rec->rec_type = S_IFDIR;
 926                         rec->rec_fid = pfid;
 927                         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
 928                                        (const struct dt_key *)dotdot, th,
 929                                        BYPASS_CAPA, 1);
 930                         if (rc != 0)
 931                                 GOTO(unlock, rc);
 932                 }
 933
 934                 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
 935                                   th, BYPASS_CAPA);
 936         } else {
 937                 if (rc == 0 && count != NULL)
 938                         *count = ldata.ld_leh->leh_reccount;
 939
 940                 GOTO(unlock, rc);
 941         }
 942         dt_write_unlock(env, orphan);
 943
 944         if (rc == 0 && !exist) {
 945                 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
 946                 rec->rec_fid = cfid;
 947                 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
 948                                (const struct dt_key *)cname->ln_name,
 949                                th, BYPASS_CAPA, 1);
 950                 if (rc == 0 && S_ISDIR(rec->rec_type)) {
 951                         dt_write_lock(env, parent, 0);
 952                         rc = dt_ref_add(env, parent, th);
 953                         dt_write_unlock(env, parent);
 954                 }
 955         }
 956
 957         if (rc != 0)
 958                 GOTO(unlock, rc);
 959
 960         rc = dt_attr_set(env, orphan, la, th, BYPASS_CAPA);
 961
 962         GOTO(stop, rc = (rc == 0 ? 1 : rc));
 963
 964 unlock:
 965         dt_write_unlock(env, orphan);
 966
 967 stop:
 968         dt_trans_stop(env, dev, th);
 969
 970 log:
 971         lfsck_ibits_unlock(&clh, LCK_EX);
 972         lfsck_ibits_unlock(&plh, LCK_EX);
 973         CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
 974                "object "DFID", name = %s: rc = %d\n",
 975                lfsck_lfsck2name(lfsck), PFID(cfid),
 976                cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
 977
 978         if (rc != 0) {
 979                 struct lfsck_namespace *ns = com->lc_file_ram;
 980
 981                 ns->ln_flags |= LF_INCONSISTENT;
 982         }
 983
 984         return rc;
 985 }
 986
 987 /**
 988  * Add the specified name entry back to namespace.
 989  *
 990  * If there is a linkEA entry that back references a name entry under
 991  * some parent directory, but such parent directory does not have the
 992  * claimed name entry. On the other hand, the linkEA entries count is
 993  * not larger than the MDT-object's hard link count. Under such case,
 994  * it is quite possible that the name entry is lost. Then the LFSCK
 995  * should add the name entry back to the namespace.
 996  *
 997  * \param[in] env       pointer to the thread context
 998  * \param[in] com       pointer to the lfsck component
 999  * \param[in] parent    pointer to the directory under which the name entry
1000  *                      will be inserted into
1001  * \param[in] child     pointer to the object referenced by the name entry
1002  *                      that to be inserted into the parent
1003  * \param[in] name      the name for the child in the parent directory
1004  *
1005  * \retval              positive number for repaired cases
1006  * \retval              0 if nothing to be repaired
1007  * \retval              negative error number on failure
1008  */
1009 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1010                                          struct lfsck_component *com,
1011                                          struct dt_object *parent,
1012                                          struct dt_object *child,
1013                                          const char *name)
1014 {
1015         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1016         struct lu_attr                  *la     = &info->lti_la;
1017         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1018         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1019         struct dt_device                *dev    = lfsck->li_next;
1020         struct thandle                  *th     = NULL;
1021         struct lustre_handle             lh     = { 0 };
1022         int                              rc     = 0;
1023         ENTRY;
1024
1025         if (unlikely(!dt_try_as_dir(env, parent)))
1026                 GOTO(log, rc = -ENOTDIR);
1027
1028         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1029                 GOTO(log, rc = 1);
1030
1031         /* Hold update lock on the parent to prevent others to access. */
1032         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
1033                               MDS_INODELOCK_UPDATE, LCK_EX);
1034         if (rc != 0)
1035                 GOTO(log, rc);
1036
1037         th = dt_trans_create(env, dev);
1038         if (IS_ERR(th))
1039                 GOTO(unlock, rc = PTR_ERR(th));
1040
1041         rec->rec_type = lfsck_object_type(child) & S_IFMT;
1042         rec->rec_fid = lfsck_dto2fid(child);
1043         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1044                                (const struct dt_key *)name, th);
1045         if (rc != 0)
1046                 GOTO(stop, rc);
1047
1048         if (S_ISDIR(rec->rec_type)) {
1049                 rc = dt_declare_ref_add(env, parent, th);
1050                 if (rc != 0)
1051                         GOTO(stop, rc);
1052         }
1053
1054         memset(la, 0, sizeof(*la));
1055         la->la_ctime = cfs_time_current_sec();
1056         la->la_valid = LA_CTIME;
1057         rc = dt_declare_attr_set(env, parent, la, th);
1058         if (rc != 0)
1059                 GOTO(stop, rc);
1060
1061         rc = dt_declare_attr_set(env, child, la, th);
1062         if (rc != 0)
1063                 GOTO(stop, rc);
1064
1065         rc = dt_trans_start_local(env, dev, th);
1066         if (rc != 0)
1067                 GOTO(stop, rc);
1068
1069         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1070                        (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1071         if (rc != 0)
1072                 GOTO(stop, rc);
1073
1074         if (S_ISDIR(rec->rec_type)) {
1075                 dt_write_lock(env, parent, 0);
1076                 rc = dt_ref_add(env, parent, th);
1077                 dt_write_unlock(env, parent);
1078                 if (rc != 0)
1079                         GOTO(stop, rc);
1080         }
1081
1082         la->la_ctime = cfs_time_current_sec();
1083         rc = dt_attr_set(env, parent, la, th, BYPASS_CAPA);
1084         if (rc != 0)
1085                 GOTO(stop, rc);
1086
1087         rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
1088
1089         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1090
1091 stop:
1092         dt_trans_stop(env, dev, th);
1093
1094 unlock:
1095         lfsck_ibits_unlock(&lh, LCK_EX);
1096
1097 log:
1098         CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1099                "the name %s and type %o to the parent "DFID": rc = %d\n",
1100                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(child)), name,
1101                lfsck_object_type(child) & S_IFMT,
1102                PFID(lfsck_dto2fid(parent)), rc);
1103
1104         if (rc != 0) {
1105                 struct lfsck_namespace *ns = com->lc_file_ram;
1106
1107                 ns->ln_flags |= LF_INCONSISTENT;
1108                 if (rc > 0)
1109                         ns->ln_lost_dirent_repaired++;
1110         }
1111
1112         return rc;
1113 }
1114
1115 /**
1116  * Create the specified orphan MDT-object on remote MDT.
1117  *
1118  * The LFSCK instance on this MDT will send LFSCK RPC to remote MDT to
1119  * ask the remote LFSCK instance to create the specified orphan object
1120  * under .lustre/lost+found/MDTxxxx/ directory with the name:
1121  * ${FID}-P-${conflict_version}.
1122  *
1123  * \param[in] env       pointer to the thread context
1124  * \param[in] com       pointer to the lfsck component
1125  * \param[in] orphan    pointer to the orphan MDT-object
1126  * \param[in] type      the orphan's type to be created
1127  *
1128  *  type "P":           The orphan object to be created was a parent directory
1129  *                      of some MDT-object which linkEA shows that the @orphan
1130  *                      object is missing.
1131  *
1132  * \see lfsck_layout_recreate_parent() for more types.
1133  *
1134  * \param[in] lmv       pointer to master LMV EA that will be set to the orphan
1135  *
1136  * \retval              positive number for repaired cases
1137  * \retval              0 if needs to repair nothing
1138  * \retval              negative error number on failure
1139  */
1140 static int lfsck_namespace_create_orphan_remote(const struct lu_env *env,
1141                                                 struct lfsck_component *com,
1142                                                 struct dt_object *orphan,
1143                                                 __u32 type,
1144                                                 struct lmv_mds_md_v1 *lmv)
1145 {
1146         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1147         struct lfsck_request            *lr     = &info->lti_lr;
1148         struct lu_seq_range             *range  = &info->lti_range;
1149         const struct lu_fid             *fid    = lfsck_dto2fid(orphan);
1150         struct lfsck_namespace          *ns     = com->lc_file_ram;
1151         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1152         struct seq_server_site          *ss     =
1153                         lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site);
1154         struct lfsck_tgt_desc           *ltd    = NULL;
1155         struct ptlrpc_request           *req    = NULL;
1156         int                              rc;
1157         ENTRY;
1158
1159         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1160                 GOTO(out, rc = 1);
1161
1162         fld_range_set_mdt(range);
1163         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
1164         if (rc != 0)
1165                 GOTO(out, rc);
1166
1167         ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, range->lsr_index);
1168         if (ltd == NULL) {
1169                 ns->ln_flags |= LF_INCOMPLETE;
1170
1171                 GOTO(out, rc = -ENODEV);
1172         }
1173
1174         req = ptlrpc_request_alloc(class_exp2cliimp(ltd->ltd_exp),
1175                                    &RQF_LFSCK_NOTIFY);
1176         if (req == NULL)
1177                 GOTO(out, rc = -ENOMEM);
1178
1179         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
1180         if (rc != 0) {
1181                 ptlrpc_request_free(req);
1182
1183                 GOTO(out, rc);
1184         }
1185
1186         lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
1187         memset(lr, 0, sizeof(*lr));
1188         lr->lr_event = LE_CREATE_ORPHAN;
1189         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1190         lr->lr_active = LFSCK_TYPE_NAMESPACE;
1191         lr->lr_fid = *fid;
1192         lr->lr_type = type;
1193         if (lmv != NULL) {
1194                 lr->lr_hash_type = lmv->lmv_hash_type;
1195                 lr->lr_stripe_count = lmv->lmv_stripe_count;
1196                 lr->lr_layout_version = lmv->lmv_layout_version;
1197                 memcpy(lr->lr_pool_name, lmv->lmv_pool_name,
1198                        sizeof(lr->lr_pool_name));
1199         }
1200
1201         ptlrpc_request_set_replen(req);
1202         rc = ptlrpc_queue_wait(req);
1203         ptlrpc_req_finished(req);
1204
1205         if (rc == 0)
1206                 rc = 1;
1207         else if (rc == -EEXIST)
1208                 rc = 0;
1209
1210         GOTO(out, rc);
1211
1212 out:
1213         CDEBUG(D_LFSCK, "%s: namespace LFSCK create object "
1214                DFID" on the MDT %x remotely: rc = %d\n",
1215                lfsck_lfsck2name(lfsck), PFID(fid),
1216                ltd != NULL ? ltd->ltd_index : -1, rc);
1217
1218         if (ltd != NULL)
1219                 lfsck_tgt_put(ltd);
1220
1221         return rc;
1222 }
1223
1224 /**
1225  * Create the specified orphan MDT-object locally.
1226  *
1227  * For the case that the parent MDT-object stored in some MDT-object's
1228  * linkEA entry is lost, the LFSCK will re-create the parent object as
1229  * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1230  * with the name ${FID}-P-${conflict_version}.
1231  *
1232  * \param[in] env       pointer to the thread context
1233  * \param[in] com       pointer to the lfsck component
1234  * \param[in] orphan    pointer to the orphan MDT-object to be created
1235  * \param[in] type      the orphan's type to be created
1236  *
1237  *  type "P":           The orphan object to be created was a parent directory
1238  *                      of some MDT-object which linkEA shows that the @orphan
1239  *                      object is missing.
1240  *
1241  * \see lfsck_layout_recreate_parent() for more types.
1242  *
1243  * \param[in] lmv       pointer to master LMV EA that will be set to the orphan
1244  *
1245  * \retval              positive number for repaired cases
1246  * \retval              negative error number on failure
1247  */
1248 static int lfsck_namespace_create_orphan_local(const struct lu_env *env,
1249                                                struct lfsck_component *com,
1250                                                struct dt_object *orphan,
1251                                                __u32 type,
1252                                                struct lmv_mds_md_v1 *lmv)
1253 {
1254         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1255         struct lu_attr                  *la     = &info->lti_la;
1256         struct dt_allocation_hint       *hint   = &info->lti_hint;
1257         struct dt_object_format         *dof    = &info->lti_dof;
1258         struct lu_name                  *cname  = &info->lti_name2;
1259         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1260         struct lu_fid                   *tfid   = &info->lti_fid;
1261         struct lmv_mds_md_v1            *lmv2   = &info->lti_lmv2;
1262         const struct lu_fid             *cfid   = lfsck_dto2fid(orphan);
1263         const struct lu_fid             *pfid;
1264         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1265         struct dt_device                *dev    = lfsck->li_bottom;
1266         struct dt_object                *parent = NULL;
1267         struct dt_object                *child  = NULL;
1268         struct thandle                  *th     = NULL;
1269         struct lustre_handle             lh     = { 0 };
1270         struct linkea_data               ldata  = { 0 };
1271         struct lu_buf                    linkea_buf;
1272         struct lu_buf                    lmv_buf;
1273         char                             name[32];
1274         int                              namelen;
1275         int                              idx    = 0;
1276         int                              rc     = 0;
1277         ENTRY;
1278
1279         LASSERT(!dt_object_exists(orphan));
1280         LASSERT(!dt_object_remote(orphan));
1281
1282         /* @orphan maybe not attached to lfsck->li_bottom */
1283         child = lfsck_object_find_by_dev(env, dev, cfid);
1284         if (IS_ERR(child))
1285                 GOTO(log, rc = PTR_ERR(child));
1286
1287         cname->ln_name = NULL;
1288         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1289                 GOTO(log, rc = 1);
1290
1291         /* Create .lustre/lost+found/MDTxxxx when needed. */
1292         if (unlikely(lfsck->li_lpf_obj == NULL)) {
1293                 rc = lfsck_create_lpf(env, lfsck);
1294                 if (rc != 0)
1295                         GOTO(log, rc);
1296         }
1297
1298         parent = lfsck->li_lpf_obj;
1299         pfid = lfsck_dto2fid(parent);
1300
1301         /* Hold update lock on the parent to prevent others to access. */
1302         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
1303                               MDS_INODELOCK_UPDATE, LCK_EX);
1304         if (rc != 0)
1305                 GOTO(log, rc);
1306
1307         do {
1308                 namelen = snprintf(name, 31, DFID"-P-%d",
1309                                    PFID(cfid), idx++);
1310                 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
1311                                (const struct dt_key *)name, BYPASS_CAPA);
1312                 if (rc != 0 && rc != -ENOENT)
1313                         GOTO(unlock1, rc);
1314         } while (rc == 0);
1315
1316         cname->ln_name = name;
1317         cname->ln_namelen = namelen;
1318
1319         memset(la, 0, sizeof(*la));
1320         la->la_mode = type | (S_ISDIR(type) ? 0700 : 0600);
1321         la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1322                        LA_ATIME | LA_MTIME | LA_CTIME;
1323
1324         child->do_ops->do_ah_init(env, hint, parent, child,
1325                                   la->la_mode & S_IFMT);
1326
1327         memset(dof, 0, sizeof(*dof));
1328         dof->dof_type = dt_mode_to_dft(type);
1329
1330         rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
1331         if (rc != 0)
1332                 GOTO(unlock1, rc);
1333
1334         rc = linkea_add_buf(&ldata, cname, pfid);
1335         if (rc != 0)
1336                 GOTO(unlock1, rc);
1337
1338         th = dt_trans_create(env, dev);
1339         if (IS_ERR(th))
1340                 GOTO(unlock1, rc = PTR_ERR(th));
1341
1342         rc = dt_declare_create(env, child, la, hint, dof, th);
1343         if (rc == 0 && S_ISDIR(type))
1344                 rc = dt_declare_ref_add(env, child, th);
1345
1346         if (rc != 0)
1347                 GOTO(stop, rc);
1348
1349         if (lmv != NULL) {
1350                 lmv->lmv_magic = LMV_MAGIC;
1351                 lmv->lmv_master_mdt_index = lfsck_dev_idx(dev);
1352                 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1353                 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1354                 rc = dt_declare_xattr_set(env, child, &lmv_buf,
1355                                           XATTR_NAME_LMV, 0, th);
1356                 if (rc != 0)
1357                         GOTO(stop, rc);
1358         }
1359
1360         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1361                        ldata.ld_leh->leh_len);
1362         rc = dt_declare_xattr_set(env, child, &linkea_buf,
1363                                   XATTR_NAME_LINK, 0, th);
1364         if (rc != 0)
1365                 GOTO(stop, rc);
1366
1367         rec->rec_type = type;
1368         rec->rec_fid = cfid;
1369         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1370                                (const struct dt_key *)name, th);
1371         if (rc == 0 && S_ISDIR(type))
1372                 rc = dt_declare_ref_add(env, parent, th);
1373
1374         if (rc != 0)
1375                 GOTO(stop, rc);
1376
1377         rc = dt_trans_start_local(env, dev, th);
1378         if (rc != 0)
1379                 GOTO(stop, rc);
1380
1381         dt_write_lock(env, child, 0);
1382         rc = dt_create(env, child, la, hint, dof, th);
1383         if (rc != 0)
1384                 GOTO(unlock2, rc);
1385
1386         if (S_ISDIR(type)) {
1387                 if (unlikely(!dt_try_as_dir(env, child)))
1388                         GOTO(unlock2, rc = -ENOTDIR);
1389
1390                 rec->rec_type = S_IFDIR;
1391                 rec->rec_fid = cfid;
1392                 rc = dt_insert(env, child, (const struct dt_rec *)rec,
1393                                (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
1394                 if (rc != 0)
1395                         GOTO(unlock2, rc);
1396
1397                 rec->rec_fid = pfid;
1398                 rc = dt_insert(env, child, (const struct dt_rec *)rec,
1399                                (const struct dt_key *)dotdot, th,
1400                                BYPASS_CAPA, 1);
1401                 if (rc != 0)
1402                         GOTO(unlock2, rc);
1403
1404                 rc = dt_ref_add(env, child, th);
1405                 if (rc != 0)
1406                         GOTO(unlock2, rc);
1407         }
1408
1409         if (lmv != NULL) {
1410                 rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV, 0,
1411                                   th, BYPASS_CAPA);
1412                 if (rc != 0)
1413                         GOTO(unlock2, rc);
1414         }
1415
1416         rc = dt_xattr_set(env, child, &linkea_buf,
1417                           XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
1418         dt_write_unlock(env, child);
1419         if (rc != 0)
1420                 GOTO(stop, rc);
1421
1422         rec->rec_type = type;
1423         rec->rec_fid = cfid;
1424         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1425                        (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1426         if (rc == 0 && S_ISDIR(type)) {
1427                 dt_write_lock(env, parent, 0);
1428                 rc = dt_ref_add(env, parent, th);
1429                 dt_write_unlock(env, parent);
1430         }
1431
1432         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1433
1434 unlock2:
1435         dt_write_unlock(env, child);
1436
1437 stop:
1438         dt_trans_stop(env, dev, th);
1439
1440 unlock1:
1441         lfsck_ibits_unlock(&lh, LCK_EX);
1442
1443 log:
1444         CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan locally for "
1445                "the object "DFID", name = %s, type %o: rc = %d\n",
1446                lfsck_lfsck2name(lfsck), PFID(cfid),
1447                cname->ln_name != NULL ? cname->ln_name : "<NULL>", type, rc);
1448
1449         if (child != NULL && !IS_ERR(child))
1450                 lfsck_object_put(env, child);
1451
1452         return rc;
1453 }
1454
1455 /**
1456  * Create the specified orphan MDT-object.
1457  *
1458  * For the case that the parent MDT-object stored in some MDT-object's
1459  * linkEA entry is lost, the LFSCK will re-create the parent object as
1460  * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1461  * with the name: ${FID}-P-${conflict_version}.
1462  *
1463  * \param[in] env       pointer to the thread context
1464  * \param[in] com       pointer to the lfsck component
1465  * \param[in] orphan    pointer to the orphan MDT-object
1466  *
1467  *  type "P":           The orphan object to be created was a parent directory
1468  *                      of some MDT-object which linkEA shows that the @orphan
1469  *                      object is missing.
1470  *
1471  * \see lfsck_layout_recreate_parent() for more types.
1472  *
1473  * \param[in] lmv       pointer to master LMV EA that will be set to the orphan
1474  *
1475  * \retval              positive number for repaired cases
1476  * \retval              0 if needs to repair nothing
1477  * \retval              negative error number on failure
1478  */
1479 static int lfsck_namespace_create_orphan(const struct lu_env *env,
1480                                          struct lfsck_component *com,
1481                                          struct dt_object *orphan,
1482                                          struct lmv_mds_md_v1 *lmv)
1483 {
1484         struct lfsck_namespace *ns = com->lc_file_ram;
1485         int                     rc;
1486
1487         if (dt_object_remote(orphan))
1488                 rc = lfsck_namespace_create_orphan_remote(env, com, orphan,
1489                                                           S_IFDIR, lmv);
1490         else
1491                 rc = lfsck_namespace_create_orphan_local(env, com, orphan,
1492                                                          S_IFDIR, lmv);
1493
1494         if (rc != 0)
1495                 ns->ln_flags |= LF_INCONSISTENT;
1496
1497         return rc;
1498 }
1499
1500 /**
1501  * Remove the specified entry from the linkEA.
1502  *
1503  * Locate the linkEA entry with the given @cname and @pfid, then
1504  * remove this entry or the other entries those are repeated with
1505  * this entry.
1506  *
1507  * \param[in] env       pointer to the thread context
1508  * \param[in] com       pointer to the lfsck component
1509  * \param[in] obj       pointer to the dt_object to be handled
1510  * \param[in,out]ldata  pointer to the buffer that holds the linkEA
1511  * \param[in] cname     the name for the child in the parent directory
1512  * \param[in] pfid      the parent directory's FID for the linkEA
1513  * \param[in] next      if true, then remove the first found linkEA
1514  *                      entry, and move the ldata->ld_lee to next entry
1515  *
1516  * \retval              positive number for repaired cases
1517  * \retval              0 if nothing to be repaired
1518  * \retval              negative error number on failure
1519  */
1520 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1521                                          struct lfsck_component *com,
1522                                          struct dt_object *obj,
1523                                          struct linkea_data *ldata,
1524                                          struct lu_name *cname,
1525                                          struct lu_fid *pfid,
1526                                          bool next)
1527 {
1528         struct lfsck_instance           *lfsck     = com->lc_lfsck;
1529         struct dt_device                *dev       = lfsck->li_bottom;
1530         struct lfsck_bookmark           *bk        = &lfsck->li_bookmark_ram;
1531         struct thandle                  *th        = NULL;
1532         struct lustre_handle             lh        = { 0 };
1533         struct linkea_data               ldata_new = { 0 };
1534         struct lu_buf                    linkea_buf;
1535         int                              rc        = 0;
1536         ENTRY;
1537
1538         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1539                               MDS_INODELOCK_UPDATE |
1540                               MDS_INODELOCK_XATTR, LCK_EX);
1541         if (rc != 0)
1542                 GOTO(log, rc);
1543
1544         if (next)
1545                 linkea_del_buf(ldata, cname);
1546         else
1547                 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1548                                                     true);
1549         lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1550                        ldata->ld_leh->leh_len);
1551
1552 again:
1553         th = dt_trans_create(env, dev);
1554         if (IS_ERR(th))
1555                 GOTO(unlock1, rc = PTR_ERR(th));
1556
1557         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1558                                   XATTR_NAME_LINK, 0, th);
1559         if (rc != 0)
1560                 GOTO(stop, rc);
1561
1562         rc = dt_trans_start_local(env, dev, th);
1563         if (rc != 0)
1564                 GOTO(stop, rc);
1565
1566         dt_write_lock(env, obj, 0);
1567         if (unlikely(lfsck_is_dead_obj(obj)))
1568                 GOTO(unlock2, rc = -ENOENT);
1569
1570         rc = lfsck_links_read2(env, obj, &ldata_new);
1571         if (rc != 0)
1572                 GOTO(unlock2, rc);
1573
1574         /* The specified linkEA entry has been removed by race. */
1575         rc = linkea_links_find(&ldata_new, cname, pfid);
1576         if (rc != 0)
1577                 GOTO(unlock2, rc = 0);
1578
1579         if (bk->lb_param & LPF_DRYRUN)
1580                 GOTO(unlock2, rc = 1);
1581
1582         if (next)
1583                 linkea_del_buf(&ldata_new, cname);
1584         else
1585                 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1586                                                     true);
1587
1588         if (linkea_buf.lb_len < ldata_new.ld_leh->leh_len) {
1589                 dt_write_unlock(env, obj);
1590                 dt_trans_stop(env, dev, th);
1591                 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1592                                ldata_new.ld_leh->leh_len);
1593                 goto again;
1594         }
1595
1596         lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1597                        ldata_new.ld_leh->leh_len);
1598         rc = dt_xattr_set(env, obj, &linkea_buf,
1599                           XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
1600
1601         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1602
1603 unlock2:
1604         dt_write_unlock(env, obj);
1605
1606 stop:
1607         dt_trans_stop(env, dev, th);
1608
1609 unlock1:
1610         lfsck_ibits_unlock(&lh, LCK_EX);
1611
1612 log:
1613         CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1614                "for the object: "DFID", parent "DFID", name %.*s\n",
1615                lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1616                PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1617                cname->ln_name);
1618
1619         if (rc != 0) {
1620                 struct lfsck_namespace *ns = com->lc_file_ram;
1621
1622                 ns->ln_flags |= LF_INCONSISTENT;
1623         }
1624
1625         return rc;
1626 }
1627
1628 /**
1629  * Conditionally remove the specified entry from the linkEA.
1630  *
1631  * Take the parent lock firstly, then check whether the specified
1632  * name entry exists or not: if yes, do nothing; otherwise, call
1633  * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1634  *
1635  * \param[in] env       pointer to the thread context
1636  * \param[in] com       pointer to the lfsck component
1637  * \param[in] parent    pointer to the parent directory
1638  * \param[in] child     pointer to the child object that holds the linkEA
1639  * \param[in,out]ldata  pointer to the buffer that holds the linkEA
1640  * \param[in] cname     the name for the child in the parent directory
1641  * \param[in] pfid      the parent directory's FID for the linkEA
1642  *
1643  * \retval              positive number for repaired cases
1644  * \retval              0 if nothing to be repaired
1645  * \retval              negative error number on failure
1646  */
1647 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1648                                               struct lfsck_component *com,
1649                                               struct dt_object *parent,
1650                                               struct dt_object *child,
1651                                               struct linkea_data *ldata,
1652                                               struct lu_name *cname,
1653                                               struct lu_fid *pfid)
1654 {
1655         struct lu_fid           *cfid   = &lfsck_env_info(env)->lti_fid3;
1656         struct lustre_handle     lh     = { 0 };
1657         int                      rc;
1658         ENTRY;
1659
1660         rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
1661                               MDS_INODELOCK_UPDATE, LCK_EX);
1662         if (rc != 0)
1663                 RETURN(rc);
1664
1665         dt_read_lock(env, parent, 0);
1666         if (unlikely(lfsck_is_dead_obj(parent))) {
1667                 dt_read_unlock(env, parent);
1668                 lfsck_ibits_unlock(&lh, LCK_EX);
1669                 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1670                                                    cname, pfid, true);
1671
1672                 RETURN(rc);
1673         }
1674
1675         rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1676                        (const struct dt_key *)cname->ln_name,
1677                        BYPASS_CAPA);
1678         dt_read_unlock(env, parent);
1679
1680         /* It is safe to release the ldlm lock, because when the logic come
1681          * here, we have got all the needed information above whether the
1682          * linkEA entry is valid or not. It is not important that others
1683          * may add new linkEA entry after the ldlm lock released. If other
1684          * has removed the specified linkEA entry by race, then it is OK,
1685          * because the subsequent lfsck_namespace_shrink_linkea() can handle
1686          * such case. */
1687         lfsck_ibits_unlock(&lh, LCK_EX);
1688         if (rc == -ENOENT) {
1689                 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1690                                                    cname, pfid, true);
1691
1692                 RETURN(rc);
1693         }
1694
1695         if (rc != 0)
1696                 RETURN(rc);
1697
1698         /* The LFSCK just found some internal status of cross-MDTs
1699          * create operation. That is normal. */
1700         if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1701                 linkea_next_entry(ldata);
1702
1703                 RETURN(0);
1704         }
1705
1706         rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1707                                            pfid, true);
1708
1709         RETURN(rc);
1710 }
1711
1712 /**
1713  * Conditionally replace name entry in the parent.
1714  *
1715  * As required, the LFSCK may re-create the lost MDT-object for dangling
1716  * name entry, but such repairing may be wrong because of bad FID in the
1717  * name entry. As the LFSCK processing, the real MDT-object may be found,
1718  * then the LFSCK should check whether the former re-created MDT-object
1719  * has been modified or not, if not, then destroy it and update the name
1720  * entry in the parent to reference the real MDT-object.
1721  *
1722  * \param[in] env       pointer to the thread context
1723  * \param[in] com       pointer to the lfsck component
1724  * \param[in] parent    pointer to the parent directory
1725  * \param[in] child     pointer to the MDT-object that may be the real
1726  *                      MDT-object corresponding to the name entry in parent
1727  * \param[in] cfid      the current FID in the name entry
1728  * \param[in] cname     contains the name of the child in the parent directory
1729  *
1730  * \retval              positive number for repaired cases
1731  * \retval              0 if nothing to be repaired
1732  * \retval              negative error number on failure
1733  */
1734 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1735                                         struct lfsck_component *com,
1736                                         struct dt_object *parent,
1737                                         struct dt_object *child,
1738                                         const struct lu_fid *cfid,
1739                                         const struct lu_name *cname)
1740 {
1741         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1742         struct lu_fid                   *tfid   = &info->lti_fid5;
1743         struct lu_attr                  *la     = &info->lti_la;
1744         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1745         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1746         struct dt_device                *dev    = lfsck->li_next;
1747         const char                      *name   = cname->ln_name;
1748         struct dt_object                *obj    = NULL;
1749         struct lustre_handle             plh    = { 0 };
1750         struct lustre_handle             clh    = { 0 };
1751         struct linkea_data               ldata  = { 0 };
1752         struct thandle                  *th     = NULL;
1753         bool                             exist  = true;
1754         int                              rc     = 0;
1755         ENTRY;
1756
1757         rc = lfsck_ibits_lock(env, lfsck, parent, &plh,
1758                               MDS_INODELOCK_UPDATE, LCK_EX);
1759         if (rc != 0)
1760                 GOTO(log, rc);
1761
1762         if (!fid_is_sane(cfid)) {
1763                 exist = false;
1764                 goto replace;
1765         }
1766
1767         obj = lfsck_object_find(env, lfsck, cfid);
1768         if (IS_ERR(obj)) {
1769                 rc = PTR_ERR(obj);
1770                 if (rc == -ENOENT) {
1771                         exist = false;
1772                         goto replace;
1773                 }
1774
1775                 GOTO(log, rc);
1776         }
1777
1778         if (!dt_object_exists(obj)) {
1779                 exist = false;
1780                 goto replace;
1781         }
1782
1783         rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
1784                        (const struct dt_key *)name, BYPASS_CAPA);
1785         if (rc == -ENOENT) {
1786                 exist = false;
1787                 goto replace;
1788         }
1789
1790         if (rc != 0)
1791                 GOTO(log, rc);
1792
1793         /* Someone changed the name entry, cannot replace it. */
1794         if (!lu_fid_eq(cfid, tfid))
1795                 GOTO(log, rc = 0);
1796
1797         /* lock the object to be destroyed. */
1798         rc = lfsck_ibits_lock(env, lfsck, obj, &clh,
1799                               MDS_INODELOCK_UPDATE |
1800                               MDS_INODELOCK_XATTR, LCK_EX);
1801         if (rc != 0)
1802                 GOTO(log, rc);
1803
1804         if (unlikely(lfsck_is_dead_obj(obj))) {
1805                 exist = false;
1806                 goto replace;
1807         }
1808
1809         rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
1810         if (rc != 0)
1811                 GOTO(log, rc);
1812
1813         /* The object has been modified by other(s), or it is not created by
1814          * LFSCK, the two cases are indistinguishable. So cannot replace it. */
1815         if (la->la_ctime != 0)
1816                 GOTO(log, rc);
1817
1818         if (S_ISREG(la->la_mode)) {
1819                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV,
1820                                   BYPASS_CAPA);
1821                 /* If someone has created related OST-object(s),
1822                  * then keep it. */
1823                 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
1824                         GOTO(log, rc = (rc > 0 ? 0 : rc));
1825         }
1826
1827 replace:
1828         dt_read_lock(env, child, 0);
1829         rc = lfsck_links_read2(env, child, &ldata);
1830         dt_read_unlock(env, child);
1831
1832         /* Someone changed the child, no need to replace. */
1833         if (rc == -ENODATA)
1834                 GOTO(log, rc = 0);
1835
1836         if (rc != 0)
1837                 GOTO(log, rc);
1838
1839         rc = linkea_links_find(&ldata, cname, lfsck_dto2fid(parent));
1840         /* Someone moved the child, no need to replace. */
1841         if (rc != 0)
1842                 GOTO(log, rc = 0);
1843
1844         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1845                 GOTO(log, rc = 1);
1846
1847         th = dt_trans_create(env, dev);
1848         if (IS_ERR(th))
1849                 GOTO(log, rc = PTR_ERR(th));
1850
1851         if (exist) {
1852                 rc = dt_declare_destroy(env, obj, th);
1853                 if (rc != 0)
1854                         GOTO(stop, rc);
1855         }
1856
1857         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
1858         if (rc != 0)
1859                 GOTO(stop, rc);
1860
1861         rec->rec_type = S_IFDIR;
1862         rec->rec_fid = lfsck_dto2fid(child);
1863         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1864                                (const struct dt_key *)name, th);
1865         if (rc != 0)
1866                 GOTO(stop, rc);
1867
1868         rc = dt_trans_start(env, dev, th);
1869         if (rc != 0)
1870                 GOTO(stop, rc);
1871
1872         if (exist) {
1873                 rc = dt_destroy(env, obj, th);
1874                 if (rc != 0)
1875                         GOTO(stop, rc);
1876         }
1877
1878         /* The old name entry maybe not exist. */
1879         dt_delete(env, parent, (const struct dt_key *)name, th,
1880                   BYPASS_CAPA);
1881
1882         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1883                        (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1884
1885         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1886
1887 stop:
1888         dt_trans_stop(env, dev, th);
1889
1890 log:
1891         lfsck_ibits_unlock(&clh, LCK_EX);
1892         lfsck_ibits_unlock(&plh, LCK_EX);
1893         if (obj != NULL && !IS_ERR(obj))
1894                 lfsck_object_put(env, obj);
1895
1896         CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
1897                "object "DFID" because of conflict with the object "DFID
1898                " under the parent "DFID" with name %s: rc = %d\n",
1899                lfsck_lfsck2name(lfsck), PFID(cfid),
1900                PFID(lfsck_dto2fid(child)), PFID(lfsck_dto2fid(parent)),
1901                name, rc);
1902
1903         return rc;
1904 }
1905
1906 /**
1907  * Overwrite the linkEA for the object with the given ldata.
1908  *
1909  * The caller should take the ldlm lock before the calling.
1910  *
1911  * \param[in] env       pointer to the thread context
1912  * \param[in] com       pointer to the lfsck component
1913  * \param[in] obj       pointer to the dt_object to be handled
1914  * \param[in] ldata     pointer to the new linkEA data
1915  *
1916  * \retval              positive number for repaired cases
1917  * \retval              0 if nothing to be repaired
1918  * \retval              negative error number on failure
1919  */
1920 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
1921                                    struct lfsck_component *com,
1922                                    struct dt_object *obj,
1923                                    struct linkea_data *ldata)
1924 {
1925         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1926         struct dt_device                *dev    = lfsck->li_bottom;
1927         struct thandle                  *th     = NULL;
1928         struct lu_buf                    linkea_buf;
1929         int                              rc     = 0;
1930         ENTRY;
1931
1932         LASSERT(!dt_object_remote(obj));
1933
1934         th = dt_trans_create(env, dev);
1935         if (IS_ERR(th))
1936                 GOTO(log, rc = PTR_ERR(th));
1937
1938         lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1939                        ldata->ld_leh->leh_len);
1940         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1941                                   XATTR_NAME_LINK, 0, th);
1942         if (rc != 0)
1943                 GOTO(stop, rc);
1944
1945         rc = dt_trans_start_local(env, dev, th);
1946         if (rc != 0)
1947                 GOTO(stop, rc);
1948
1949         dt_write_lock(env, obj, 0);
1950         if (unlikely(lfsck_is_dead_obj(obj)))
1951                 GOTO(unlock, rc = 0);
1952
1953         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1954                 GOTO(unlock, rc = 1);
1955
1956         rc = dt_xattr_set(env, obj, &linkea_buf,
1957                           XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
1958
1959         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
1960
1961 unlock:
1962         dt_write_unlock(env, obj);
1963
1964 stop:
1965         dt_trans_stop(env, dev, th);
1966
1967 log:
1968         CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
1969                "object "DFID": rc = %d\n",
1970                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
1971
1972         if (rc != 0) {
1973                 struct lfsck_namespace *ns = com->lc_file_ram;
1974
1975                 ns->ln_flags |= LF_INCONSISTENT;
1976         }
1977
1978         return rc;
1979 }
1980
1981 /**
1982  * Repair invalid name entry.
1983  *
1984  * If the name entry contains invalid information, such as bad file type
1985  * or (and) corrupted object FID, then either remove the name entry or
1986  * udpate the name entry with the given (right) information.
1987  *
1988  * \param[in] env       pointer to the thread context
1989  * \param[in] com       pointer to the lfsck component
1990  * \param[in] parent    pointer to the parent directory
1991  * \param[in] child     pointer to the object referenced by the name entry
1992  * \param[in] name      the old name of the child under the parent directory
1993  * \param[in] name2     the new name of the child under the parent directory
1994  * \param[in] type      the type claimed by the name entry
1995  * \param[in] update    update the name entry if true; otherwise, remove it
1996  * \param[in] dec       decrease the parent nlink count if true
1997  *
1998  * \retval              positive number for repaired successfully
1999  * \retval              0 if nothing to be repaired
2000  * \retval              negative error number on failure
2001  */
2002 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2003                                   struct lfsck_component *com,
2004                                   struct dt_object *parent,
2005                                   struct dt_object *child,
2006                                   const char *name, const char *name2,
2007                                   __u16 type, bool update, bool dec)
2008 {
2009         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2010         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
2011         const struct lu_fid             *cfid   = lfsck_dto2fid(child);
2012         struct lu_fid                   *tfid   = &info->lti_fid5;
2013         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2014         struct dt_device                *dev    = lfsck->li_next;
2015         struct thandle                  *th     = NULL;
2016         struct lustre_handle             lh     = { 0 };
2017         int                              rc     = 0;
2018         ENTRY;
2019
2020         if (unlikely(!dt_try_as_dir(env, parent)))
2021                 GOTO(log, rc = -ENOTDIR);
2022
2023         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2024                               MDS_INODELOCK_UPDATE, LCK_EX);
2025         if (rc != 0)
2026                 GOTO(log, rc);
2027
2028         th = dt_trans_create(env, dev);
2029         if (IS_ERR(th))
2030                 GOTO(unlock1, rc = PTR_ERR(th));
2031
2032         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2033         if (rc != 0)
2034                 GOTO(stop, rc);
2035
2036         if (update) {
2037                 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2038                 rec->rec_fid = cfid;
2039                 rc = dt_declare_insert(env, parent,
2040                                        (const struct dt_rec *)rec,
2041                                        (const struct dt_key *)name2, th);
2042                 if (rc != 0)
2043                         GOTO(stop, rc);
2044         }
2045
2046         if (dec) {
2047                 rc = dt_declare_ref_del(env, parent, th);
2048                 if (rc != 0)
2049                         GOTO(stop, rc);
2050         }
2051
2052         rc = dt_trans_start(env, dev, th);
2053         if (rc != 0)
2054                 GOTO(stop, rc);
2055
2056         dt_write_lock(env, parent, 0);
2057         rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
2058                        (const struct dt_key *)name, BYPASS_CAPA);
2059         /* Someone has removed the bad name entry by race. */
2060         if (rc == -ENOENT)
2061                 GOTO(unlock2, rc = 0);
2062
2063         if (rc != 0)
2064                 GOTO(unlock2, rc);
2065
2066         /* Someone has removed the bad name entry and reused it for other
2067          * object by race. */
2068         if (!lu_fid_eq(tfid, cfid))
2069                 GOTO(unlock2, rc = 0);
2070
2071         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2072                 GOTO(unlock2, rc = 1);
2073
2074         rc = dt_delete(env, parent, (const struct dt_key *)name, th,
2075                        BYPASS_CAPA);
2076         if (rc != 0)
2077                 GOTO(unlock2, rc);
2078
2079         if (update) {
2080                 rc = dt_insert(env, parent,
2081                                (const struct dt_rec *)rec,
2082                                (const struct dt_key *)name2, th,
2083                                BYPASS_CAPA, 1);
2084                 if (rc != 0)
2085                         GOTO(unlock2, rc);
2086         }
2087
2088         if (dec) {
2089                 rc = dt_ref_del(env, parent, th);
2090                 if (rc != 0)
2091                         GOTO(unlock2, rc);
2092         }
2093
2094         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2095
2096 unlock2:
2097         dt_write_unlock(env, parent);
2098
2099 stop:
2100         dt_trans_stop(env, dev, th);
2101
2102         /* We are not sure whether the child will become orphan or not.
2103          * Record it in the LFSCK trace file for further checking in
2104          * the second-stage scanning. */
2105         if (!update && !dec && rc == 0)
2106                 lfsck_namespace_trace_update(env, com, cfid,
2107                                              LNTF_CHECK_LINKEA, true);
2108
2109 unlock1:
2110         lfsck_ibits_unlock(&lh, LCK_EX);
2111
2112 log:
2113         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2114                "entry for: parent "DFID", child "DFID", name %s, type "
2115                "in name entry %o, type claimed by child %o. repair it "
2116                "by %s with new name2 %s: rc = %d\n", lfsck_lfsck2name(lfsck),
2117                PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
2118                name, type, update ? lfsck_object_type(child) : 0,
2119                update ? "updating" : "removing", name2, rc);
2120
2121         if (rc != 0) {
2122                 struct lfsck_namespace *ns = com->lc_file_ram;
2123
2124                 ns->ln_flags |= LF_INCONSISTENT;
2125         }
2126
2127         return rc;
2128 }
2129
2130 /**
2131  * Update the ".." name entry for the given object.
2132  *
2133  * The object's ".." is corrupted, this function will update the ".." name
2134  * entry with the given pfid, and the linkEA with the given ldata.
2135  *
2136  * The caller should take the ldlm lock before the calling.
2137  *
2138  * \param[in] env       pointer to the thread context
2139  * \param[in] com       pointer to the lfsck component
2140  * \param[in] obj       pointer to the dt_object to be handled
2141  * \param[in] pfid      the new fid for the object's ".." name entry
2142  * \param[in] cname     the name for the @obj in the parent directory
2143  *
2144  * \retval              positive number for repaired cases
2145  * \retval              0 if nothing to be repaired
2146  * \retval              negative error number on failure
2147  */
2148 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2149                                                   struct lfsck_component *com,
2150                                                   struct dt_object *obj,
2151                                                   const struct lu_fid *pfid,
2152                                                   struct lu_name *cname)
2153 {
2154         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2155         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
2156         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2157         struct dt_device                *dev    = lfsck->li_bottom;
2158         struct thandle                  *th     = NULL;
2159         struct linkea_data               ldata  = { 0 };
2160         struct lu_buf                    linkea_buf;
2161         int                              rc     = 0;
2162         ENTRY;
2163
2164         LASSERT(!dt_object_remote(obj));
2165         LASSERT(S_ISDIR(lfsck_object_type(obj)));
2166
2167         rc = linkea_data_new(&ldata, &info->lti_big_buf);
2168         if (rc != 0)
2169                 GOTO(log, rc);
2170
2171         rc = linkea_add_buf(&ldata, cname, pfid);
2172         if (rc != 0)
2173                 GOTO(log, rc);
2174
2175         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2176                        ldata.ld_leh->leh_len);
2177
2178         th = dt_trans_create(env, dev);
2179         if (IS_ERR(th))
2180                 GOTO(log, rc = PTR_ERR(th));
2181
2182         rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2183         if (rc != 0)
2184                 GOTO(stop, rc);
2185
2186         rec->rec_type = S_IFDIR;
2187         rec->rec_fid = pfid;
2188         rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2189                                (const struct dt_key *)dotdot, th);
2190         if (rc != 0)
2191                 GOTO(stop, rc);
2192
2193         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2194                                   XATTR_NAME_LINK, 0, th);
2195         if (rc != 0)
2196                 GOTO(stop, rc);
2197
2198         rc = dt_trans_start_local(env, dev, th);
2199         if (rc != 0)
2200                 GOTO(stop, rc);
2201
2202         dt_write_lock(env, obj, 0);
2203         if (unlikely(lfsck_is_dead_obj(obj)))
2204                 GOTO(unlock, rc = 0);
2205
2206         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2207                 GOTO(unlock, rc = 1);
2208
2209         /* The old ".." name entry maybe not exist. */
2210         dt_delete(env, obj, (const struct dt_key *)dotdot, th,
2211                   BYPASS_CAPA);
2212
2213         rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2214                        (const struct dt_key *)dotdot, th, BYPASS_CAPA, 1);
2215         if (rc != 0)
2216                 GOTO(unlock, rc);
2217
2218         rc = dt_xattr_set(env, obj, &linkea_buf,
2219                           XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
2220
2221         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2222
2223 unlock:
2224         dt_write_unlock(env, obj);
2225
2226 stop:
2227         dt_trans_stop(env, dev, th);
2228
2229 log:
2230         CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2231                "the object "DFID", new parent "DFID": rc = %d\n",
2232                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2233                PFID(pfid), rc);
2234
2235         if (rc != 0) {
2236                 struct lfsck_namespace *ns = com->lc_file_ram;
2237
2238                 ns->ln_flags |= LF_INCONSISTENT;
2239         }
2240
2241         return rc;
2242 }
2243
2244 /**
2245  * Handle orphan @obj during Double Scan Directory.
2246  *
2247  * Remove the @obj's current (invalid) linkEA entries, and insert
2248  * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2249  * ${FID}-${PFID}-D-${conflict_version}
2250  *
2251  * The caller should take the ldlm lock before the calling.
2252  *
2253  * \param[in] env       pointer to the thread context
2254  * \param[in] com       pointer to the lfsck component
2255  * \param[in] obj       pointer to the orphan object to be handled
2256  * \param[in] pfid      the new fid for the object's ".." name entry
2257  * \param[in,out] lh    ldlm lock handler for the given @obj
2258  * \param[out] type     to tell the caller what the inconsistency is
2259  *
2260  * \retval              positive number for repaired cases
2261  * \retval              0 if nothing to be repaired
2262  * \retval              negative error number on failure
2263  */
2264 static int
2265 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2266                            struct lfsck_component *com,
2267                            struct dt_object *obj,
2268                            const struct lu_fid *pfid,
2269                            struct lustre_handle *lh,
2270                            enum lfsck_namespace_inconsistency_type *type)
2271 {
2272         struct lfsck_thread_info *info = lfsck_env_info(env);
2273         struct lfsck_namespace   *ns   = com->lc_file_ram;
2274         int                       rc;
2275         ENTRY;
2276
2277         /* Remove the unrecognized linkEA. */
2278         rc = lfsck_namespace_links_remove(env, com, obj);
2279         lfsck_ibits_unlock(lh, LCK_EX);
2280         if (rc < 0 && rc != -ENODATA)
2281                 RETURN(rc);
2282
2283         *type = LNIT_MUL_REF;
2284
2285         /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2286          * ever tried to verify some remote MDT-object that resides on this
2287          * MDT, but this MDT failed to respond such request. So means there
2288          * may be some remote name entry on other MDT that references this
2289          * object with another name, so we cannot know whether this linkEA
2290          * is valid or not. So keep it there and maybe resolved when next
2291          * LFSCK run. */
2292         if (ns->ln_flags & LF_INCOMPLETE)
2293                 RETURN(0);
2294
2295         /* The unique linkEA is invalid, even if the ".." name entry may be
2296          * valid, we still cannot know via which name entry this directory
2297          * will be referenced. Then handle it as pure orphan. */
2298         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2299                  "-"DFID, PFID(pfid));
2300         rc = lfsck_namespace_insert_orphan(env, com, obj,
2301                                            info->lti_tmpbuf, "D", NULL);
2302
2303         RETURN(rc);
2304 }
2305
2306 /**
2307  * Double Scan Directory object for single linkEA entry case.
2308  *
2309  * The given @child has unique linkEA entry. If the linkEA entry is valid,
2310  * then check whether the name is in the namespace or not, if not, add the
2311  * missing name entry back to namespace. If the linkEA entry is invalid,
2312  * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2313  * as an orphan.
2314  *
2315  * \param[in] env       pointer to the thread context
2316  * \param[in] com       pointer to the lfsck component
2317  * \param[in] child     pointer to the directory to be double scanned
2318  * \param[in] pfid      the FID corresponding to the ".." entry
2319  * \param[in] ldata     pointer to the linkEA data for the given @child
2320  * \param[in,out] lh    ldlm lock handler for the given @child
2321  * \param[out] type     to tell the caller what the inconsistency is
2322  * \param[in] retry     if found inconsistency, but the caller does not hold
2323  *                      ldlm lock on the @child, then set @retry as true
2324  *
2325  * \retval              positive number for repaired cases
2326  * \retval              0 if nothing to be repaired
2327  * \retval              negative error number on failure
2328  */
2329 static int
2330 lfsck_namespace_dsd_single(const struct lu_env *env,
2331                            struct lfsck_component *com,
2332                            struct dt_object *child,
2333                            const struct lu_fid *pfid,
2334                            struct linkea_data *ldata,
2335                            struct lustre_handle *lh,
2336                            enum lfsck_namespace_inconsistency_type *type,
2337                            bool *retry)
2338 {
2339         struct lfsck_thread_info *info          = lfsck_env_info(env);
2340         struct lu_name           *cname         = &info->lti_name;
2341         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2342         struct lu_fid            *tfid          = &info->lti_fid3;
2343         struct lfsck_namespace   *ns            = com->lc_file_ram;
2344         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2345         struct dt_object         *parent        = NULL;
2346         struct lmv_mds_md_v1     *lmv;
2347         int                       rc            = 0;
2348         ENTRY;
2349
2350         lfsck_namespace_unpack_linkea_entry(ldata, cname, tfid, info->lti_key);
2351         /* The unique linkEA entry with bad parent will be handled as orphan. */
2352         if (!fid_is_sane(tfid)) {
2353                 if (!lustre_handle_is_used(lh) && retry != NULL)
2354                         *retry = true;
2355                 else
2356                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2357                                                         pfid, lh, type);
2358
2359                 GOTO(out, rc);
2360         }
2361
2362         parent = lfsck_object_find_bottom(env, lfsck, tfid);
2363         if (IS_ERR(parent))
2364                 GOTO(out, rc = PTR_ERR(parent));
2365
2366         /* We trust the unique linkEA entry in spite of whether it matches the
2367          * ".." name entry or not. Because even if the linkEA entry is wrong
2368          * and the ".." name entry is right, we still cannot know via which
2369          * name entry the child will be referenced, since all known entries
2370          * have been verified during the first-stage scanning. */
2371         if (!dt_object_exists(parent)) {
2372                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2373                  * has ever tried to verify some remote MDT-object that resides
2374                  * on this MDT, but this MDT failed to respond such request. So
2375                  * means there may be some remote name entry on other MDT that
2376                  * references this object with another name, so we cannot know
2377                  * whether this linkEA is valid or not. So keep it there and
2378                  * maybe resolved when next LFSCK run. */
2379                 if (ns->ln_flags & LF_INCOMPLETE)
2380                         GOTO(out, rc = 0);
2381
2382                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2383                         *retry = true;
2384
2385                         GOTO(out, rc = 0);
2386                 }
2387
2388                 lfsck_ibits_unlock(lh, LCK_EX);
2389
2390 lost_parent:
2391                 lmv = &info->lti_lmv;
2392                 rc = lfsck_read_stripe_lmv(env, child, lmv);
2393                 if (rc != 0 && rc != -ENODATA)
2394                         GOTO(out, rc);
2395
2396                 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2397                         lmv = NULL;
2398                 } else if (lfsck_shard_name_to_index(env,
2399                                         cname->ln_name, cname->ln_namelen,
2400                                         S_IFDIR, cfid) < 0) {
2401                         /* It is an invalid name entry, we
2402                          * cannot trust the parent also. */
2403                         rc = lfsck_namespace_shrink_linkea(env, com, child,
2404                                                 ldata, cname, tfid, true);
2405                         if (rc < 0)
2406                                 GOTO(out, rc);
2407
2408                         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2409                                  "-"DFID, PFID(pfid));
2410                         rc = lfsck_namespace_insert_orphan(env, com, child,
2411                                                 info->lti_tmpbuf, "S", NULL);
2412
2413                         GOTO(out, rc);
2414                 }
2415
2416                 /* Create the lost parent as an orphan. */
2417                 rc = lfsck_namespace_create_orphan(env, com, parent, lmv);
2418                 if (rc >= 0) {
2419                         /* Add the missing name entry to the parent. */
2420                         rc = lfsck_namespace_insert_normal(env, com, parent,
2421                                                         child, cname->ln_name);
2422                         if (unlikely(rc == -EEXIST)) {
2423                                 /* Unfortunately, someone reused the name
2424                                  * under the parent by race. So we have
2425                                  * to remove the linkEA entry from
2426                                  * current child object. It means that the
2427                                  * LFSCK cannot recover the system
2428                                  * totally back to its original status,
2429                                  * but it is necessary to make the
2430                                  * current system to be consistent. */
2431                                 rc = lfsck_namespace_shrink_linkea(env,
2432                                                 com, child, ldata,
2433                                                 cname, tfid, true);
2434                                 if (rc >= 0) {
2435                                         snprintf(info->lti_tmpbuf,
2436                                                  sizeof(info->lti_tmpbuf),
2437                                                  "-"DFID, PFID(pfid));
2438                                         rc = lfsck_namespace_insert_orphan(env,
2439                                                 com, child, info->lti_tmpbuf,
2440                                                 "D", NULL);
2441                                 }
2442                         }
2443                 }
2444
2445                 GOTO(out, rc);
2446         }
2447
2448         /* The unique linkEA entry with bad parent will be handled as orphan. */
2449         if (unlikely(!dt_try_as_dir(env, parent))) {
2450                 if (!lustre_handle_is_used(lh) && retry != NULL)
2451                         *retry = true;
2452                 else
2453                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2454                                                         pfid, lh, type);
2455
2456                 GOTO(out, rc);
2457         }
2458
2459         rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
2460                        (const struct dt_key *)cname->ln_name, BYPASS_CAPA);
2461         if (rc == -ENOENT) {
2462                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2463                  * has ever tried to verify some remote MDT-object that resides
2464                  * on this MDT, but this MDT failed to respond such request. So
2465                  * means there may be some remote name entry on other MDT that
2466                  * references this object with another name, so we cannot know
2467                  * whether this linkEA is valid or not. So keep it there and
2468                  * maybe resolved when next LFSCK run. */
2469                 if (ns->ln_flags & LF_INCOMPLETE)
2470                         GOTO(out, rc = 0);
2471
2472                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2473                         *retry = true;
2474
2475                         GOTO(out, rc = 0);
2476                 }
2477
2478                 lfsck_ibits_unlock(lh, LCK_EX);
2479                 rc = lfsck_namespace_check_name(env, parent, child, cname);
2480                 if (rc == -ENOENT)
2481                         goto lost_parent;
2482
2483                 if (rc < 0)
2484                         GOTO(out, rc);
2485
2486                 /* It is an invalid name entry, drop it. */
2487                 if (unlikely(rc > 0)) {
2488                         rc = lfsck_namespace_shrink_linkea(env, com, child,
2489                                                 ldata, cname, tfid, true);
2490                         if (rc >= 0) {
2491                                 snprintf(info->lti_tmpbuf,
2492                                          sizeof(info->lti_tmpbuf),
2493                                          "-"DFID, PFID(pfid));
2494                                 rc = lfsck_namespace_insert_orphan(env, com,
2495                                         child, info->lti_tmpbuf, "D", NULL);
2496                         }
2497
2498                         GOTO(out, rc);
2499                 }
2500
2501                 /* Add the missing name entry back to the namespace. */
2502                 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2503                                                    cname->ln_name);
2504                 if (unlikely(rc == -ESTALE))
2505                         /* It may happen when the remote object has been
2506                          * removed, but the local MDT is not aware of that. */
2507                         goto lost_parent;
2508
2509                 if (unlikely(rc == -EEXIST)) {
2510                         /* Unfortunately, someone reused the name under the
2511                          * parent by race. So we have to remove the linkEA
2512                          * entry from current child object. It means that the
2513                          * LFSCK cannot recover the system totally back to
2514                          * its original status, but it is necessary to make
2515                          * the current system to be consistent.
2516                          *
2517                          * It also may be because of the LFSCK found some
2518                          * internal status of create operation. Under such
2519                          * case, nothing to be done. */
2520                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
2521                                         parent, child, ldata, cname, tfid);
2522                         if (rc >= 0) {
2523                                 snprintf(info->lti_tmpbuf,
2524                                          sizeof(info->lti_tmpbuf),
2525                                          "-"DFID, PFID(pfid));
2526                                 rc = lfsck_namespace_insert_orphan(env, com,
2527                                         child, info->lti_tmpbuf, "D", NULL);
2528                         }
2529                 }
2530
2531                 GOTO(out, rc);
2532         }
2533
2534         if (rc != 0)
2535                 GOTO(out, rc);
2536
2537         if (!lu_fid_eq(tfid, cfid)) {
2538                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2539                         *retry = true;
2540
2541                         GOTO(out, rc = 0);
2542                 }
2543
2544                 lfsck_ibits_unlock(lh, LCK_EX);
2545                 /* The name entry references another MDT-object that
2546                  * may be created by the LFSCK for repairing dangling
2547                  * name entry. Try to replace it. */
2548                 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2549                                                   tfid, cname);
2550                 if (rc == 0)
2551                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2552                                                         pfid, lh, type);
2553
2554                 GOTO(out, rc);
2555         }
2556
2557         /* The ".." name entry is wrong, update it. */
2558         if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2559                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2560                         *retry = true;
2561
2562                         GOTO(out, rc = 0);
2563                 }
2564
2565                 *type = LNIT_UNMATCHED_PAIRS;
2566                 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2567                                                 lfsck_dto2fid(parent), cname);
2568         }
2569
2570         GOTO(out, rc);
2571
2572 out:
2573         if (parent != NULL && !IS_ERR(parent))
2574                 lfsck_object_put(env, parent);
2575
2576         return rc;
2577 }
2578
2579 /**
2580  * Double Scan Directory object for multiple linkEA entries case.
2581  *
2582  * The given @child has multiple linkEA entries. There is at most one linkEA
2583  * entry will be valid, all the others will be removed. Firstly, the function
2584  * will try to find out the linkEA entry for which the name entry exists under
2585  * the given parent (@pfid). If there is no linkEA entry that matches the given
2586  * ".." name entry, then tries to find out the first linkEA entry that both the
2587  * parent and the name entry exist to rebuild a new ".." name entry.
2588  *
2589  * \param[in] env       pointer to the thread context
2590  * \param[in] com       pointer to the lfsck component
2591  * \param[in] child     pointer to the directory to be double scanned
2592  * \param[in] pfid      the FID corresponding to the ".." entry
2593  * \param[in] ldata     pointer to the linkEA data for the given @child
2594  * \param[in,out] lh    ldlm lock handler for the given @child
2595  * \param[out] type     to tell the caller what the inconsistency is
2596  * \param[in] lpf       true if the ".." entry is under lost+found/MDTxxxx/
2597  *
2598  * \retval              positive number for repaired cases
2599  * \retval              0 if nothing to be repaired
2600  * \retval              negative error number on failure
2601  */
2602 static int
2603 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2604                              struct lfsck_component *com,
2605                              struct dt_object *child,
2606                              const struct lu_fid *pfid,
2607                              struct linkea_data *ldata,
2608                              struct lustre_handle *lh,
2609                              enum lfsck_namespace_inconsistency_type *type,
2610                              bool lpf)
2611 {
2612         struct lfsck_thread_info *info          = lfsck_env_info(env);
2613         struct lu_name           *cname         = &info->lti_name;
2614         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2615         struct lu_fid            *tfid          = &info->lti_fid3;
2616         struct lu_fid            *pfid2         = &info->lti_fid4;
2617         struct lfsck_namespace   *ns            = com->lc_file_ram;
2618         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2619         struct lfsck_bookmark    *bk            = &lfsck->li_bookmark_ram;
2620         struct dt_object         *parent        = NULL;
2621         struct linkea_data        ldata_new     = { 0 };
2622         int                       count         = 0;
2623         int                       rc            = 0;
2624         bool                      once          = true;
2625         ENTRY;
2626
2627 again:
2628         while (ldata->ld_lee != NULL) {
2629                 lfsck_namespace_unpack_linkea_entry(ldata, cname, tfid,
2630                                                     info->lti_key);
2631                 /* Drop repeated linkEA entries. */
2632                 lfsck_namespace_filter_linkea_entry(ldata, cname, tfid, true);
2633                 /* Drop invalid linkEA entry. */
2634                 if (!fid_is_sane(tfid)) {
2635                         linkea_del_buf(ldata, cname);
2636                         continue;
2637                 }
2638
2639                 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2640                  * then it is possible that: the directry object has ever
2641                  * been lost, but its name entry was there. In the former
2642                  * LFSCK run, during the first-stage scanning, the LFSCK
2643                  * found the dangling name entry, but it did not recreate
2644                  * the lost object, and when moved to the second-stage
2645                  * scanning, some children objects of the lost directory
2646                  * object were found, then the LFSCK recreated such lost
2647                  * directory object as an orphan.
2648                  *
2649                  * When the LFSCK runs again, if the dangling name is still
2650                  * there, the LFSCK should move the orphan directory object
2651                  * back to the normal namespace. */
2652                 if (!lpf && !lu_fid_eq(pfid, tfid) && once) {
2653                         linkea_next_entry(ldata);
2654                         continue;
2655                 }
2656
2657                 parent = lfsck_object_find_bottom(env, lfsck, tfid);
2658                 if (IS_ERR(parent))
2659                         RETURN(PTR_ERR(parent));
2660
2661                 if (!dt_object_exists(parent)) {
2662                         lfsck_object_put(env, parent);
2663                         if (ldata->ld_leh->leh_reccount > 1) {
2664                                 /* If it is NOT the last linkEA entry, then
2665                                  * there is still other chance to make the
2666                                  * child to be visible via other parent, then
2667                                  * remove this linkEA entry. */
2668                                 linkea_del_buf(ldata, cname);
2669                                 continue;
2670                         }
2671
2672                         break;
2673                 }
2674
2675                 /* The linkEA entry with bad parent will be removed. */
2676                 if (unlikely(!dt_try_as_dir(env, parent))) {
2677                         lfsck_object_put(env, parent);
2678                         linkea_del_buf(ldata, cname);
2679                         continue;
2680                 }
2681
2682                 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
2683                                (const struct dt_key *)cname->ln_name,
2684                                BYPASS_CAPA);
2685                 *pfid2 = *lfsck_dto2fid(parent);
2686                 if (rc == -ENOENT) {
2687                         lfsck_object_put(env, parent);
2688                         linkea_next_entry(ldata);
2689                         continue;
2690                 }
2691
2692                 if (rc != 0) {
2693                         lfsck_object_put(env, parent);
2694
2695                         RETURN(rc);
2696                 }
2697
2698                 if (lu_fid_eq(tfid, cfid)) {
2699                         lfsck_object_put(env, parent);
2700                         if (!lu_fid_eq(pfid, pfid2)) {
2701                                 *type = LNIT_UNMATCHED_PAIRS;
2702                                 rc = lfsck_namespace_repair_unmatched_pairs(env,
2703                                                 com, child, pfid2, cname);
2704
2705                                 RETURN(rc);
2706                         }
2707
2708 rebuild:
2709                         /* It is the most common case that we find the
2710                          * name entry corresponding to the linkEA entry
2711                          * that matches the ".." name entry. */
2712                         rc = linkea_data_new(&ldata_new, &info->lti_big_buf);
2713                         if (rc != 0)
2714                                 RETURN(rc);
2715
2716                         rc = linkea_add_buf(&ldata_new, cname, pfid2);
2717                         if (rc != 0)
2718                                 RETURN(rc);
2719
2720                         rc = lfsck_namespace_rebuild_linkea(env, com, child,
2721                                                             &ldata_new);
2722                         if (rc < 0)
2723                                 RETURN(rc);
2724
2725                         linkea_del_buf(ldata, cname);
2726                         linkea_first_entry(ldata);
2727                         /* There may be some invalid dangling name entries under
2728                          * other parent directories, remove all of them. */
2729                         while (ldata->ld_lee != NULL) {
2730                                 lfsck_namespace_unpack_linkea_entry(ldata,
2731                                                 cname, tfid, info->lti_key);
2732                                 if (!fid_is_sane(tfid))
2733                                         goto next;
2734
2735                                 parent = lfsck_object_find_bottom(env, lfsck,
2736                                                                   tfid);
2737                                 if (IS_ERR(parent)) {
2738                                         rc = PTR_ERR(parent);
2739                                         if (rc != -ENOENT &&
2740                                             bk->lb_param & LPF_FAILOUT)
2741                                                 RETURN(rc);
2742
2743                                         goto next;
2744                                 }
2745
2746                                 if (!dt_object_exists(parent)) {
2747                                         lfsck_object_put(env, parent);
2748                                         goto next;
2749                                 }
2750
2751                                 rc = lfsck_namespace_repair_dirent(env, com,
2752                                         parent, child, cname->ln_name,
2753                                         cname->ln_name, S_IFDIR, false, true);
2754                                 lfsck_object_put(env, parent);
2755                                 if (rc < 0) {
2756                                         if (bk->lb_param & LPF_FAILOUT)
2757                                                 RETURN(rc);
2758
2759                                         goto next;
2760                                 }
2761
2762                                 count += rc;
2763
2764 next:
2765                                 linkea_del_buf(ldata, cname);
2766                         }
2767
2768                         ns->ln_dirent_repaired += count;
2769
2770                         RETURN(rc);
2771                 }
2772
2773                 lfsck_ibits_unlock(lh, LCK_EX);
2774                 /* The name entry references another MDT-object that may be
2775                  * created by the LFSCK for repairing dangling name entry.
2776                  * Try to replace it. */
2777                 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2778                                                   tfid, cname);
2779                 lfsck_object_put(env, parent);
2780                 if (rc < 0)
2781                         RETURN(rc);
2782
2783                 if (rc > 0)
2784                         goto rebuild;
2785
2786                 linkea_del_buf(ldata, cname);
2787         }
2788
2789         if (ldata->ld_leh->leh_reccount == 1) {
2790                 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
2791                                                 lh, type, NULL);
2792
2793                 RETURN(rc);
2794         }
2795
2796         /* All linkEA entries are invalid and removed, then handle the @child
2797          * as an orphan.*/
2798         if (ldata->ld_leh->leh_reccount == 0) {
2799                 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
2800                                                 type);
2801
2802                 RETURN(rc);
2803         }
2804
2805         linkea_first_entry(ldata);
2806         /* If the dangling name entry for the orphan directory object has
2807          * been remvoed, then just check whether the directory object is
2808          * still under the .lustre/lost+found/MDTxxxx/ or not. */
2809         if (lpf) {
2810                 lpf = false;
2811                 goto again;
2812         }
2813
2814         /* There is no linkEA entry that matches the ".." name entry. Find
2815          * the first linkEA entry that both parent and name entry exist to
2816          * rebuild a new ".." name entry. */
2817         if (once) {
2818                 once = false;
2819                 goto again;
2820         }
2821
2822         RETURN(rc);
2823 }
2824
2825 /**
2826  * Repair the object's nlink attribute.
2827  *
2828  * If all the known name entries have been verified, then the object's hard
2829  * link attribute should match the object's linkEA entries count unless the
2830  * object's has too much hard link to be recorded in the linkEA. Such cases
2831  * should have been marked in the LFSCK trace file. Otherwise, trust the
2832  * linkEA to update the object's nlink attribute.
2833  *
2834  * \param[in] env       pointer to the thread context
2835  * \param[in] com       pointer to the lfsck component
2836  * \param[in] obj       pointer to the dt_object to be handled
2837  * \param[in,out] nlink pointer to buffer to object's hard lock count before
2838  *                      and after the repairing
2839  *
2840  * \retval              positive number for repaired cases
2841  * \retval              0 if nothing to be repaired
2842  * \retval              negative error number on failure
2843  */
2844 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
2845                                         struct lfsck_component *com,
2846                                         struct dt_object *obj, __u32 *nlink)
2847 {
2848         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2849         struct lu_attr                  *la     = &info->lti_la3;
2850         struct lu_fid                   *tfid   = &info->lti_fid3;
2851         struct lfsck_namespace          *ns     = com->lc_file_ram;
2852         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2853         struct dt_device                *dev    = lfsck->li_bottom;
2854         const struct lu_fid             *cfid   = lfsck_dto2fid(obj);
2855         struct dt_object                *child  = NULL;
2856         struct thandle                  *th     = NULL;
2857         struct linkea_data               ldata  = { 0 };
2858         struct lustre_handle             lh     = { 0 };
2859         __u32                            old    = *nlink;
2860         int                              rc     = 0;
2861         __u8                             flags;
2862         ENTRY;
2863
2864         LASSERT(!dt_object_remote(obj));
2865         LASSERT(S_ISREG(lfsck_object_type(obj)));
2866
2867         child = lfsck_object_find_by_dev(env, dev, cfid);
2868         if (IS_ERR(child))
2869                 GOTO(log, rc = PTR_ERR(child));
2870
2871         rc = lfsck_ibits_lock(env, lfsck, child, &lh,
2872                               MDS_INODELOCK_UPDATE |
2873                               MDS_INODELOCK_XATTR, LCK_EX);
2874         if (rc != 0)
2875                 GOTO(log, rc);
2876
2877         th = dt_trans_create(env, dev);
2878         if (IS_ERR(th))
2879                 GOTO(log, rc = PTR_ERR(th));
2880
2881         la->la_valid = LA_NLINK;
2882         rc = dt_declare_attr_set(env, child, la, th);
2883         if (rc != 0)
2884                 GOTO(stop, rc);
2885
2886         rc = dt_trans_start_local(env, dev, th);
2887         if (rc != 0)
2888                 GOTO(stop, rc);
2889
2890         dt_write_lock(env, child, 0);
2891         /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2892          * ever tried to verify some remote MDT-object that resides on this
2893          * MDT, but this MDT failed to respond such request. So means there
2894          * may be some remote name entry on other MDT that references this
2895          * object with another name, so we cannot know whether this linkEA
2896          * is valid or not. So keep it there and maybe resolved when next
2897          * LFSCK run. */
2898         if (ns->ln_flags & LF_INCOMPLETE)
2899                 GOTO(unlock, rc = 0);
2900
2901         fid_cpu_to_be(tfid, cfid);
2902         rc = dt_lookup(env, com->lc_obj, (struct dt_rec *)&flags,
2903                        (const struct dt_key *)tfid, BYPASS_CAPA);
2904         if (rc != 0)
2905                 GOTO(unlock, rc);
2906
2907         if (flags & LNTF_SKIP_NLINK)
2908                 GOTO(unlock, rc = 0);
2909
2910         rc = lfsck_links_read2(env, child, &ldata);
2911         if (rc == -ENODATA)
2912                 GOTO(unlock, rc = 0);
2913
2914         if (rc != 0)
2915                 GOTO(unlock, rc);
2916
2917         if (*nlink == ldata.ld_leh->leh_reccount)
2918                 GOTO(unlock, rc = 0);
2919
2920         la->la_nlink = *nlink = ldata.ld_leh->leh_reccount;
2921         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2922                 GOTO(unlock, rc = 1);
2923
2924         rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
2925
2926         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2927
2928 unlock:
2929         dt_write_unlock(env, child);
2930
2931 stop:
2932         dt_trans_stop(env, dev, th);
2933
2934 log:
2935         lfsck_ibits_unlock(&lh, LCK_EX);
2936         if (child != NULL && !IS_ERR(child))
2937                 lfsck_object_put(env, child);
2938
2939         CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
2940                "nlink count from %u to %u: rc = %d\n",
2941                lfsck_lfsck2name(lfsck), PFID(cfid), old, *nlink, rc);
2942
2943         if (rc != 0)
2944                 ns->ln_flags |= LF_INCONSISTENT;
2945
2946         return rc;
2947 }
2948
2949 /**
2950  * Double scan the directory object for namespace LFSCK.
2951  *
2952  * This function will verify the <parent, child> pairs in the namespace tree:
2953  * the parent references the child via some name entry that should be in the
2954  * child's linkEA entry, the child should back references the parent via its
2955  * ".." name entry.
2956  *
2957  * The LFSCK will scan every linkEA entry in turn until find out the first
2958  * matched pairs. If found, then all other linkEA entries will be dropped.
2959  * If all the linkEA entries cannot match the ".." name entry, then there
2960  * are serveral possible cases:
2961  *
2962  * 1) If there is only one linkEA entry, then trust it as long as the PFID
2963  *    in the linkEA entry is valid.
2964  *
2965  * 2) If there are multiple linkEA entries, then try to find the linkEA
2966  *    that matches the ".." name entry. If found, then all other entries
2967  *    are invalid; otherwise, it is quite possible that the ".." name entry
2968  *    is corrupted. Under such case, the LFSCK will rebuild the ".." name
2969  *    entry according to the first valid linkEA entry (both the parent and
2970  *    the name entry should exist).
2971  *
2972  * 3) If the directory object has no (valid) linkEA entry, then the
2973  *    directory object will be handled as pure orphan and inserted
2974  *    in the .lustre/lost+found/MDTxxxx/ with the name:
2975  *    ${self_FID}-${PFID}-D-${conflict_version}
2976  *
2977  * \param[in] env       pointer to the thread context
2978  * \param[in] com       pointer to the lfsck component
2979  * \param[in] child     pointer to the directory object to be handled
2980  * \param[in] flags     to indicate the specical checking on the @child
2981  *
2982  * \retval              positive number for repaired cases
2983  * \retval              0 if nothing to be repaired
2984  * \retval              negative error number on failure
2985  */
2986 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
2987                                            struct lfsck_component *com,
2988                                            struct dt_object *child, __u8 flags)
2989 {
2990         struct lfsck_thread_info *info          = lfsck_env_info(env);
2991         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2992         struct lu_fid            *pfid          = &info->lti_fid2;
2993         struct lfsck_namespace   *ns            = com->lc_file_ram;
2994         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2995         struct lustre_handle      lh            = { 0 };
2996         struct linkea_data        ldata         = { 0 };
2997         bool                      unknown       = false;
2998         bool                      lpf           = false;
2999         bool                      retry         = false;
3000         enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3001         int                       rc            = 0;
3002         ENTRY;
3003
3004         LASSERT(!dt_object_remote(child));
3005
3006         if (flags & LNTF_UNCERTAIN_LMV) {
3007                 if (flags & LNTF_RECHECK_NAME_HASH) {
3008                         rc = lfsck_namespace_scan_shard(env, com, child);
3009                         if (rc < 0)
3010                                 RETURN(rc);
3011
3012                         ns->ln_striped_shards_scanned++;
3013                 } else {
3014                         ns->ln_striped_shards_skipped++;
3015                 }
3016         }
3017
3018         flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
3019         if (flags == 0)
3020                 RETURN(0);
3021
3022         if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
3023             !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
3024                 CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the"
3025                        "the namespace LFSCK, then the LFSCK cannot guarantee"
3026                        "all the name entries have been verified in first-stage"
3027                        "scanning. So have to skip orphan related handling for"
3028                        "the directory object "DFID" with remote name entry\n",
3029                        lfsck_lfsck2name(lfsck), PFID(cfid));
3030
3031                 RETURN(0);
3032         }
3033
3034         if (unlikely(!dt_try_as_dir(env, child)))
3035                 GOTO(out, rc = -ENOTDIR);
3036
3037         /* We only take ldlm lock on the @child when required. When the
3038          * logic comes here for the first time, it is always false. */
3039         if (0) {
3040
3041 lock:
3042                 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
3043                                       MDS_INODELOCK_UPDATE |
3044                                       MDS_INODELOCK_XATTR, LCK_EX);
3045                 if (rc != 0)
3046                         GOTO(out, rc);
3047         }
3048
3049         dt_read_lock(env, child, 0);
3050         if (unlikely(lfsck_is_dead_obj(child))) {
3051                 dt_read_unlock(env, child);
3052
3053                 GOTO(out, rc = 0);
3054         }
3055
3056         rc = dt_lookup(env, child, (struct dt_rec *)pfid,
3057                        (const struct dt_key *)dotdot, BYPASS_CAPA);
3058         if (rc != 0) {
3059                 if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
3060                         dt_read_unlock(env, child);
3061
3062                         GOTO(out, rc);
3063                 }
3064
3065                 if (!lustre_handle_is_used(&lh)) {
3066                         dt_read_unlock(env, child);
3067                         goto lock;
3068                 }
3069
3070                 fid_zero(pfid);
3071         } else if (lfsck->li_lpf_obj != NULL &&
3072                    lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
3073                 lpf = true;
3074         }
3075
3076         rc = lfsck_links_read(env, child, &ldata);
3077         dt_read_unlock(env, child);
3078         if (rc != 0) {
3079                 if (rc != -ENODATA && rc != -EINVAL)
3080                         GOTO(out, rc);
3081
3082                 if (!lustre_handle_is_used(&lh))
3083                         goto lock;
3084
3085                 if (rc == -EINVAL && !fid_is_zero(pfid)) {
3086                         /* Remove the corrupted linkEA. */
3087                         rc = lfsck_namespace_links_remove(env, com, child);
3088                         if (rc == 0)
3089                                 /* Here, because of the crashed linkEA, we
3090                                  * cannot know whether there is some parent
3091                                  * that references the child directory via
3092                                  * some name entry or not. So keep it there,
3093                                  * when the LFSCK run next time, if there is
3094                                  * some parent that references this object,
3095                                  * then the LFSCK can rebuild the linkEA;
3096                                  * otherwise, this object will be handled
3097                                  * as orphan as above. */
3098                                 unknown = true;
3099                 } else {
3100                         /* 1. If we have neither ".." nor linkEA,
3101                          *    then it is an orphan.
3102                          *
3103                          * 2. If we only have the ".." name entry,
3104                          *    but no parent references this child
3105                          *    directory, then handle it as orphan. */
3106                         lfsck_ibits_unlock(&lh, LCK_EX);
3107                         type = LNIT_MUL_REF;
3108
3109                         /* If the LFSCK is marked as LF_INCOMPLETE,
3110                          * then means some MDT has ever tried to
3111                          * verify some remote MDT-object that resides
3112                          * on this MDT, but this MDT failed to respond
3113                          * such request. So means there may be some
3114                          * remote name entry on other MDT that
3115                          * references this object with another name,
3116                          * so we cannot know whether this linkEA is
3117                          * valid or not. So keep it there and maybe
3118                          * resolved when next LFSCK run. */
3119                         if (ns->ln_flags & LF_INCOMPLETE)
3120                                 GOTO(out, rc = 0);
3121
3122                         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
3123                                  "-"DFID, PFID(pfid));
3124                         rc = lfsck_namespace_insert_orphan(env, com, child,
3125                                                 info->lti_tmpbuf, "D", NULL);
3126                 }
3127
3128                 GOTO(out, rc);
3129         }
3130
3131         linkea_first_entry(&ldata);
3132         /* This is the most common case: the object has unique linkEA entry. */
3133         if (ldata.ld_leh->leh_reccount == 1) {
3134                 rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
3135                                                 &lh, &type, &retry);
3136                 if (retry) {
3137                         LASSERT(!lustre_handle_is_used(&lh));
3138
3139                         retry = false;
3140                         goto lock;
3141                 }
3142
3143                 GOTO(out, rc);
3144         }
3145
3146         if (!lustre_handle_is_used(&lh))
3147                 goto lock;
3148
3149         if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
3150                 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
3151                                                 &type);
3152
3153                 GOTO(out, rc);
3154         }
3155
3156         /* When we come here, the cases usually like that:
3157          * 1) The directory object has a corrupted linkEA entry. During the
3158          *    first-stage scanning, the LFSCK cannot know such corruption,
3159          *    then it appends the right linkEA entry according to the found
3160          *    name entry after the bad one.
3161          *
3162          * 2) The directory object has a right linkEA entry. During the
3163          *    first-stage scanning, the LFSCK finds some bad name entry,
3164          *    but the LFSCK cannot aware that at that time, then it adds
3165          *    the bad linkEA entry for further processing. */
3166         rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
3167                                           &lh, &type, lpf);
3168
3169         GOTO(out, rc);
3170
3171 out:
3172         lfsck_ibits_unlock(&lh, LCK_EX);
3173         if (rc > 0) {
3174                 switch (type) {
3175                 case LNIT_BAD_LINKEA:
3176                         ns->ln_linkea_repaired++;
3177                         break;
3178                 case LNIT_UNMATCHED_PAIRS:
3179                         ns->ln_unmatched_pairs_repaired++;
3180                         break;
3181                 case LNIT_MUL_REF:
3182                         ns->ln_mul_ref_repaired++;
3183                         break;
3184                 default:
3185                         break;
3186                 }
3187         }
3188
3189         if (unknown)
3190                 ns->ln_unknown_inconsistency++;
3191
3192         return rc;
3193 }
3194
3195 /**
3196  * Double scan the MDT-object for namespace LFSCK.
3197  *
3198  * If the MDT-object contains invalid or repeated linkEA entries, then drop
3199  * those entries from the linkEA; if the linkEA becomes empty or the object
3200  * has no linkEA, then it is an orphan and will be added into the directory
3201  * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
3202  * the remote parent; if the name entry corresponding to some linkEA entry
3203  * is lost, then add the name entry back to the namespace.
3204  *
3205  * \param[in] env       pointer to the thread context
3206  * \param[in] com       pointer to the lfsck component
3207  * \param[in] child     pointer to the dt_object to be handled
3208  * \param[in] flags     some hints to indicate how the @child should be handled
3209  *
3210  * \retval              positive number for repaired cases
3211  * \retval              0 if nothing to be repaired
3212  * \retval              negative error number on failure
3213  */
3214 static int lfsck_namespace_double_scan_one(const struct lu_env *env,
3215                                            struct lfsck_component *com,
3216                                            struct dt_object *child, __u8 flags)
3217 {
3218         struct lfsck_thread_info *info     = lfsck_env_info(env);
3219         struct lu_attr           *la       = &info->lti_la;
3220         struct lu_name           *cname    = &info->lti_name;
3221         struct lu_fid            *pfid     = &info->lti_fid;
3222         struct lu_fid            *cfid     = &info->lti_fid2;
3223         struct lfsck_instance    *lfsck    = com->lc_lfsck;
3224         struct lfsck_namespace   *ns       = com->lc_file_ram;
3225         struct dt_object         *parent   = NULL;
3226         struct linkea_data        ldata    = { 0 };
3227         bool                      repaired = false;
3228         int                       count    = 0;
3229         int                       rc;
3230         ENTRY;
3231
3232         dt_read_lock(env, child, 0);
3233         if (unlikely(lfsck_is_dead_obj(child))) {
3234                 dt_read_unlock(env, child);
3235
3236                 RETURN(0);
3237         }
3238
3239         if (S_ISDIR(lfsck_object_type(child))) {
3240                 dt_read_unlock(env, child);
3241                 rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
3242
3243                 RETURN(rc);
3244         }
3245
3246         rc = lfsck_links_read(env, child, &ldata);
3247         dt_read_unlock(env, child);
3248         if (rc != 0)
3249                 GOTO(out, rc);
3250
3251         linkea_first_entry(&ldata);
3252         while (ldata.ld_lee != NULL) {
3253                 lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
3254                                                     info->lti_key);
3255                 rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
3256                                                          false);
3257                 /* Found repeated linkEA entries */
3258                 if (rc > 0) {
3259                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3260                                                 &ldata, cname, pfid, false);
3261                         if (rc < 0)
3262                                 GOTO(out, rc);
3263
3264                         if (rc == 0)
3265                                 continue;
3266
3267                         repaired = true;
3268
3269                         /* fall through */
3270                 }
3271
3272                 /* Invalid PFID in the linkEA entry. */
3273                 if (!fid_is_sane(pfid)) {
3274                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3275                                                 &ldata, cname, pfid, true);
3276                         if (rc < 0)
3277                                 GOTO(out, rc);
3278
3279                         if (rc > 0)
3280                                 repaired = true;
3281
3282                         continue;
3283                 }
3284
3285                 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3286                 if (IS_ERR(parent))
3287                         GOTO(out, rc = PTR_ERR(parent));
3288
3289                 if (!dt_object_exists(parent)) {
3290
3291 lost_parent:
3292                         if (ldata.ld_leh->leh_reccount > 1) {
3293                                 /* If it is NOT the last linkEA entry, then
3294                                  * there is still other chance to make the
3295                                  * child to be visible via other parent, then
3296                                  * remove this linkEA entry. */
3297                                 rc = lfsck_namespace_shrink_linkea(env, com,
3298                                         child, &ldata, cname, pfid, true);
3299                         } else {
3300                                 /* If the LFSCK is marked as LF_INCOMPLETE,
3301                                  * then means some MDT has ever tried to
3302                                  * verify some remote MDT-object that resides
3303                                  * on this MDT, but this MDT failed to respond
3304                                  * such request. So means there may be some
3305                                  * remote name entry on other MDT that
3306                                  * references this object with another name,
3307                                  * so we cannot know whether this linkEA is
3308                                  * valid or not. So keep it there and maybe
3309                                  * resolved when next LFSCK run. */
3310                                 if (ns->ln_flags & LF_INCOMPLETE) {
3311                                         lfsck_object_put(env, parent);
3312
3313                                         GOTO(out, rc = 0);
3314                                 }
3315
3316                                 /* Create the lost parent as an orphan. */
3317                                 rc = lfsck_namespace_create_orphan(env, com,
3318                                                                 parent, NULL);
3319                                 if (rc < 0) {
3320                                         lfsck_object_put(env, parent);
3321
3322                                         GOTO(out, rc);
3323                                 }
3324
3325                                 if (rc > 0)
3326                                         repaired = true;
3327
3328                                 /* Add the missing name entry to the parent. */
3329                                 rc = lfsck_namespace_insert_normal(env, com,
3330                                                 parent, child, cname->ln_name);
3331                                 if (unlikely(rc == -EEXIST))
3332                                         /* Unfortunately, someone reused the
3333                                          * name under the parent by race. So we
3334                                          * have to remove the linkEA entry from
3335                                          * current child object. It means that
3336                                          * the LFSCK cannot recover the system
3337                                          * totally back to its original status,
3338                                          * but it is necessary to make the
3339                                          * current system to be consistent. */
3340                                         rc = lfsck_namespace_shrink_linkea(env,
3341                                                         com, child, &ldata,
3342                                                         cname, pfid, true);
3343                                 else
3344                                         linkea_next_entry(&ldata);
3345                         }
3346
3347                         lfsck_object_put(env, parent);
3348                         if (rc < 0)
3349                                 GOTO(out, rc);
3350
3351                         if (rc > 0)
3352                                 repaired = true;
3353
3354                         continue;
3355                 }
3356
3357                 /* The linkEA entry with bad parent will be removed. */
3358                 if (unlikely(!dt_try_as_dir(env, parent))) {
3359                         lfsck_object_put(env, parent);
3360                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3361                                                 &ldata, cname, pfid, true);
3362                         if (rc < 0)
3363                                 GOTO(out, rc);
3364
3365                         if (rc > 0)
3366                                 repaired = true;
3367
3368                         continue;
3369                 }
3370
3371                 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
3372                                (const struct dt_key *)cname->ln_name,
3373                                BYPASS_CAPA);
3374                 if (rc != 0 && rc != -ENOENT) {
3375                         lfsck_object_put(env, parent);
3376
3377                         GOTO(out, rc);
3378                 }
3379
3380                 if (rc == 0) {
3381                         if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
3382                                 /* It is the most common case that we
3383                                  * find the name entry corresponding
3384                                  * to the linkEA entry. */
3385                                 lfsck_object_put(env, parent);
3386                                 linkea_next_entry(&ldata);
3387                         } else {
3388                                 /* The name entry references another
3389                                  * MDT-object that may be created by
3390                                  * the LFSCK for repairing dangling
3391                                  * name entry. Try to replace it. */
3392                                 rc = lfsck_namespace_replace_cond(env, com,
3393                                                 parent, child, cfid, cname);
3394                                 lfsck_object_put(env, parent);
3395                                 if (rc < 0)
3396                                         GOTO(out, rc);
3397
3398                                 if (rc > 0) {
3399                                         repaired = true;
3400                                         linkea_next_entry(&ldata);
3401                                 } else {
3402                                         rc = lfsck_namespace_shrink_linkea(env,
3403                                                         com, child, &ldata,
3404                                                         cname, pfid, true);
3405                                         if (rc < 0)
3406                                                 GOTO(out, rc);
3407
3408                                         if (rc > 0)
3409                                                 repaired = true;
3410                                 }
3411                         }
3412
3413                         continue;
3414                 }
3415
3416                 rc = dt_attr_get(env, child, la, BYPASS_CAPA);
3417                 if (rc != 0)
3418                         GOTO(out, rc);
3419
3420                 /* If there is no name entry in the parent dir and the object
3421                  * link count is less than the linkea entries count, then the
3422                  * linkea entry should be removed. */
3423                 if (ldata.ld_leh->leh_reccount > la->la_nlink) {
3424                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
3425                                         parent, child, &ldata, cname, pfid);
3426                         lfsck_object_put(env, parent);
3427                         if (rc < 0)
3428                                 GOTO(out, rc);
3429
3430                         if (rc > 0)
3431                                 repaired = true;
3432
3433                         continue;
3434                 }
3435
3436                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3437                  * MDT has ever tried to verify some remote MDT-object that
3438                  * resides on this MDT, but this MDT failed to respond such
3439                  * request. So means there may be some remote name entry on
3440                  * other MDT that references this object with another name,
3441                  * so we cannot know whether this linkEA is valid or not.
3442                  * So keep it there and maybe resolved when next LFSCK run. */
3443                 if (ns->ln_flags & LF_INCOMPLETE) {
3444                         lfsck_object_put(env, parent);
3445
3446                         GOTO(out, rc = 0);
3447                 }
3448
3449                 rc = lfsck_namespace_check_name(env, parent, child, cname);
3450                 if (rc == -ENOENT)
3451                         goto lost_parent;
3452
3453                 if (rc < 0) {
3454                         lfsck_object_put(env, parent);
3455
3456                         GOTO(out, rc);
3457                 }
3458
3459                 /* It is an invalid name entry, drop it. */
3460                 if (unlikely(rc > 0)) {
3461                         lfsck_object_put(env, parent);
3462                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3463                                                 &ldata, cname, pfid, true);
3464                         if (rc < 0)
3465                                 GOTO(out, rc);
3466
3467                         if (rc > 0)
3468                                 repaired = true;
3469
3470                         continue;
3471                 }
3472
3473                 /* Add the missing name entry back to the namespace. */
3474                 rc = lfsck_namespace_insert_normal(env, com, parent, child,
3475                                                    cname->ln_name);
3476                 if (unlikely(rc == -ESTALE))
3477                         /* It may happen when the remote object has been
3478                          * removed, but the local MDT is not aware of that. */
3479                         goto lost_parent;
3480
3481                 if (unlikely(rc == -EEXIST))
3482                         /* Unfortunately, someone reused the name under the
3483                          * parent by race. So we have to remove the linkEA
3484                          * entry from current child object. It means that the
3485                          * LFSCK cannot recover the system totally back to
3486                          * its original status, but it is necessary to make
3487                          * the current system to be consistent.
3488                          *
3489                          * It also may be because of the LFSCK found some
3490                          * internal status of create operation. Under such
3491                          * case, nothing to be done. */
3492                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
3493                                         parent, child, &ldata, cname, pfid);
3494                 else
3495                         linkea_next_entry(&ldata);
3496
3497                 lfsck_object_put(env, parent);
3498                 if (rc < 0)
3499                         GOTO(out, rc);
3500
3501                 if (rc > 0)
3502                         repaired = true;
3503         }
3504
3505         GOTO(out, rc = 0);
3506
3507 out:
3508         if (rc < 0 && rc != -ENODATA)
3509                 return rc;
3510
3511         if (rc == 0) {
3512                 LASSERT(ldata.ld_leh != NULL);
3513
3514                 count = ldata.ld_leh->leh_reccount;
3515         }
3516
3517         /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3518          * MDT has ever tried to verify some remote MDT-object that
3519          * resides on this MDT, but this MDT failed to respond such
3520          * request. So means there may be some remote name entry on
3521          * other MDT that references this object with another name,
3522          * so we cannot know whether this linkEA is valid or not.
3523          * So keep it there and maybe resolved when next LFSCK run. */
3524         if (count == 0 && !(ns->ln_flags & LF_INCOMPLETE)) {
3525                 /* If the child becomes orphan, then insert it into
3526                  * the global .lustre/lost+found/MDTxxxx directory. */
3527                 rc = lfsck_namespace_insert_orphan(env, com, child, "", "O",
3528                                                    &count);
3529                 if (rc < 0)
3530                         return rc;
3531
3532                 if (rc > 0) {
3533                         ns->ln_mul_ref_repaired++;
3534                         repaired = true;
3535                 }
3536         }
3537
3538         rc = dt_attr_get(env, child, la, BYPASS_CAPA);
3539         if (rc != 0)
3540                 return rc;
3541
3542         if (la->la_nlink != count) {
3543                 rc = lfsck_namespace_repair_nlink(env, com, child,
3544                                                   &la->la_nlink);
3545                 if (rc > 0) {
3546                         ns->ln_objs_nlink_repaired++;
3547                         rc = 0;
3548                 }
3549         }
3550
3551         if (repaired) {
3552                 if (la->la_nlink > 1)
3553                         ns->ln_mul_linked_repaired++;
3554
3555                 if (rc == 0)
3556                         rc = 1;
3557         }
3558
3559         return rc;
3560 }
3561
3562 static void lfsck_namespace_dump_statistics(struct seq_file *m,
3563                                             struct lfsck_namespace *ns,
3564                                             __u64 checked_phase1,
3565                                             __u64 checked_phase2,
3566                                             __u32 time_phase1,
3567                                             __u32 time_phase2)
3568 {
3569         seq_printf(m, "checked_phase1: "LPU64"\n"
3570                       "checked_phase2: "LPU64"\n"
3571                       "updated_phase1: "LPU64"\n"
3572                       "updated_phase2: "LPU64"\n"
3573                       "failed_phase1: "LPU64"\n"
3574                       "failed_phase2: "LPU64"\n"
3575                       "directories: "LPU64"\n"
3576                       "dirent_repaired: "LPU64"\n"
3577                       "linkea_repaired: "LPU64"\n"
3578                       "nlinks_repaired: "LPU64"\n"
3579                       "multiple_linked_checked: "LPU64"\n"
3580                       "multiple_linked_repaired: "LPU64"\n"
3581                       "unknown_inconsistency: "LPU64"\n"
3582                       "unmatched_pairs_repaired: "LPU64"\n"
3583                       "dangling_repaired: "LPU64"\n"
3584                       "multiple_referenced_repaired: "LPU64"\n"
3585                       "bad_file_type_repaired: "LPU64"\n"
3586                       "lost_dirent_repaired: "LPU64"\n"
3587                       "local_lost_found_scanned: "LPU64"\n"
3588                       "local_lost_found_moved: "LPU64"\n"
3589                       "local_lost_found_skipped: "LPU64"\n"
3590                       "local_lost_found_failed: "LPU64"\n"
3591                       "striped_dirs_scanned: "LPU64"\n"
3592                       "striped_dirs_repaired: "LPU64"\n"
3593                       "striped_dirs_failed: "LPU64"\n"
3594                       "striped_dirs_disabled: "LPU64"\n"
3595                       "striped_dirs_skipped: "LPU64"\n"
3596                       "striped_shards_scanned: "LPU64"\n"
3597                       "striped_shards_repaired: "LPU64"\n"
3598                       "striped_shards_failed: "LPU64"\n"
3599                       "striped_shards_skipped: "LPU64"\n"
3600                       "name_hash_repaired: "LPU64"\n"
3601                       "success_count: %u\n"
3602                       "run_time_phase1: %u seconds\n"
3603                       "run_time_phase2: %u seconds\n",
3604                       checked_phase1,
3605                       checked_phase2,
3606                       ns->ln_items_repaired,
3607                       ns->ln_objs_repaired_phase2,
3608                       ns->ln_items_failed,
3609                       ns->ln_objs_failed_phase2,
3610                       ns->ln_dirs_checked,
3611                       ns->ln_dirent_repaired,
3612                       ns->ln_linkea_repaired,
3613                       ns->ln_objs_nlink_repaired,
3614                       ns->ln_mul_linked_checked,
3615                       ns->ln_mul_linked_repaired,
3616                       ns->ln_unknown_inconsistency,
3617                       ns->ln_unmatched_pairs_repaired,
3618                       ns->ln_dangling_repaired,
3619                       ns->ln_mul_ref_repaired,
3620                       ns->ln_bad_type_repaired,
3621                       ns->ln_lost_dirent_repaired,
3622                       ns->ln_local_lpf_scanned,
3623                       ns->ln_local_lpf_moved,
3624                       ns->ln_local_lpf_skipped,
3625                       ns->ln_local_lpf_failed,
3626                       ns->ln_striped_dirs_scanned,
3627                       ns->ln_striped_dirs_repaired,
3628                       ns->ln_striped_dirs_failed,
3629                       ns->ln_striped_dirs_disabled,
3630                       ns->ln_striped_dirs_skipped,
3631                       ns->ln_striped_shards_scanned,
3632                       ns->ln_striped_shards_repaired,
3633                       ns->ln_striped_shards_failed,
3634                       ns->ln_striped_shards_skipped,
3635                       ns->ln_name_hash_repaired,
3636                       ns->ln_success_count,
3637                       time_phase1,
3638                       time_phase2);
3639 }
3640
3641 static void lfsck_namespace_release_lmv(const struct lu_env *env,
3642                                         struct lfsck_component *com)
3643 {
3644         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3645         struct lfsck_namespace          *ns     = com->lc_file_ram;
3646
3647         while (!list_empty(&lfsck->li_list_lmv)) {
3648                 struct lfsck_lmv_unit   *llu;
3649                 struct lfsck_lmv        *llmv;
3650
3651                 llu = list_entry(lfsck->li_list_lmv.next,
3652                                  struct lfsck_lmv_unit, llu_link);
3653                 llmv = &llu->llu_lmv;
3654
3655                 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
3656                          "still in using: %u\n",
3657                          atomic_read(&llmv->ll_ref));
3658
3659                 ns->ln_striped_dirs_skipped++;
3660                 lfsck_lmv_put(env, llmv);
3661         }
3662 }
3663
3664 /* namespace APIs */
3665
3666 static int lfsck_namespace_reset(const struct lu_env *env,
3667                                  struct lfsck_component *com, bool init)
3668 {
3669         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3670         struct lfsck_namespace          *ns     = com->lc_file_ram;
3671         struct lfsck_assistant_data     *lad    = com->lc_data;
3672         struct dt_object                *root;
3673         struct dt_object                *dto;
3674         int                              rc;
3675         ENTRY;
3676
3677         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
3678         if (IS_ERR(root))
3679                 GOTO(log, rc = PTR_ERR(root));
3680
3681         if (unlikely(!dt_try_as_dir(env, root)))
3682                 GOTO(put, rc = -ENOTDIR);
3683
3684         down_write(&com->lc_sem);
3685         if (init) {
3686                 memset(ns, 0, sizeof(*ns));
3687         } else {
3688                 __u32 count = ns->ln_success_count;
3689                 __u64 last_time = ns->ln_time_last_complete;
3690
3691                 memset(ns, 0, sizeof(*ns));
3692                 ns->ln_success_count = count;
3693                 ns->ln_time_last_complete = last_time;
3694         }
3695         ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
3696         ns->ln_status = LS_INIT;
3697
3698         rc = local_object_unlink(env, lfsck->li_bottom, root,
3699                                  lfsck_namespace_name);
3700         if (rc != 0)
3701                 GOTO(out, rc);
3702
3703         lfsck_object_put(env, com->lc_obj);
3704         com->lc_obj = NULL;
3705         dto = local_index_find_or_create(env, lfsck->li_los, root,
3706                                          lfsck_namespace_name,
3707                                          S_IFREG | S_IRUGO | S_IWUSR,
3708                                          &dt_lfsck_features);
3709         if (IS_ERR(dto))
3710                 GOTO(out, rc = PTR_ERR(dto));
3711
3712         com->lc_obj = dto;
3713         rc = dto->do_ops->do_index_try(env, dto, &dt_lfsck_features);
3714         if (rc != 0)
3715                 GOTO(out, rc);
3716
3717         lad->lad_incomplete = 0;
3718         CFS_RESET_BITMAP(lad->lad_bitmap);
3719
3720         rc = lfsck_namespace_store(env, com);
3721
3722         GOTO(out, rc);
3723
3724 out:
3725         up_write(&com->lc_sem);
3726
3727 put:
3728         lu_object_put(env, &root->do_lu);
3729 log:
3730         CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n",
3731                lfsck_lfsck2name(lfsck), rc);
3732         return rc;
3733 }
3734
3735 static void
3736 lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
3737                      bool new_checked)
3738 {
3739         struct lfsck_namespace *ns = com->lc_file_ram;
3740
3741         down_write(&com->lc_sem);
3742         if (new_checked)
3743                 com->lc_new_checked++;
3744         lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
3745         up_write(&com->lc_sem);
3746 }
3747
3748 static void lfsck_namespace_close_dir(const struct lu_env *env,
3749                                       struct lfsck_component *com)
3750 {
3751         struct lfsck_namespace          *ns     = com->lc_file_ram;
3752         struct lfsck_assistant_data     *lad    = com->lc_data;
3753         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3754         struct lfsck_lmv                *llmv   = lfsck->li_lmv;
3755         struct lfsck_namespace_req      *lnr;
3756         __u32                            size   =
3757                                 sizeof(*lnr) + LFSCK_TMPBUF_LEN;
3758         bool                             wakeup = false;
3759         ENTRY;
3760
3761         if (llmv == NULL)
3762                 RETURN_EXIT;
3763
3764         OBD_ALLOC(lnr, size);
3765         if (lnr == NULL) {
3766                 ns->ln_striped_dirs_skipped++;
3767
3768                 RETURN_EXIT;
3769         }
3770
3771         /* Generate a dummy request to indicate that all shards' name entry
3772          * in this striped directory has been scanned for the first time. */
3773         INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
3774         lnr->lnr_obj = lfsck_object_get(lfsck->li_obj_dir);
3775         lnr->lnr_lmv = lfsck_lmv_get(llmv);
3776         lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
3777         lnr->lnr_oit_cookie = lfsck->li_pos_current.lp_oit_cookie;
3778         lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
3779         lnr->lnr_size = size;
3780
3781         spin_lock(&lad->lad_lock);
3782         if (lad->lad_assistant_status < 0) {
3783                 spin_unlock(&lad->lad_lock);
3784                 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
3785                 ns->ln_striped_dirs_skipped++;
3786
3787                 RETURN_EXIT;
3788         }
3789
3790         list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
3791         if (lad->lad_prefetched == 0)
3792                 wakeup = true;
3793
3794         lad->lad_prefetched++;
3795         spin_unlock(&lad->lad_lock);
3796         if (wakeup)
3797                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
3798
3799         EXIT;
3800 }
3801
3802 static int lfsck_namespace_open_dir(const struct lu_env *env,
3803                                     struct lfsck_component *com)
3804 {
3805         struct lfsck_instance   *lfsck  = com->lc_lfsck;
3806         struct lfsck_namespace  *ns     = com->lc_file_ram;
3807         struct lfsck_lmv        *llmv   = lfsck->li_lmv;
3808         int                      rc     = 0;
3809         ENTRY;
3810
3811         if (llmv == NULL)
3812                 RETURN(0);
3813
3814         if (llmv->ll_lmv_master) {
3815                 struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
3816
3817                 if (lmv->lmv_master_mdt_index !=
3818                     lfsck_dev_idx(lfsck->li_bottom)) {
3819                         lmv->lmv_master_mdt_index =
3820                                 lfsck_dev_idx(lfsck->li_bottom);
3821                         ns->ln_flags |= LF_INCONSISTENT;
3822                         llmv->ll_lmv_updated = 1;
3823                 }
3824         } else {
3825                 rc = lfsck_namespace_verify_stripe_slave(env, com,
3826                                         lfsck->li_obj_dir, llmv);
3827         }
3828
3829         RETURN(rc > 0 ? 0 : rc);
3830 }
3831
3832 static int lfsck_namespace_checkpoint(const struct lu_env *env,
3833                                       struct lfsck_component *com, bool init)
3834 {
3835         struct lfsck_instance   *lfsck = com->lc_lfsck;
3836         struct lfsck_namespace  *ns    = com->lc_file_ram;
3837         int                      rc;
3838
3839         if (!init) {
3840                 rc = lfsck_checkpoint_generic(env, com);
3841                 if (rc != 0)
3842                         goto log;
3843         }
3844
3845         down_write(&com->lc_sem);
3846         if (init) {
3847                 ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
3848         } else {
3849                 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
3850                 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
3851                                 HALF_SEC - lfsck->li_time_last_checkpoint);
3852                 ns->ln_time_last_checkpoint = cfs_time_current_sec();
3853                 ns->ln_items_checked += com->lc_new_checked;
3854                 com->lc_new_checked = 0;
3855         }
3856
3857         rc = lfsck_namespace_store(env, com);
3858         up_write(&com->lc_sem);
3859
3860 log:
3861         CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos ["LPU64
3862                ", "DFID", "LPX64"]: rc = %d\n", lfsck_lfsck2name(lfsck),
3863                lfsck->li_pos_current.lp_oit_cookie,
3864                PFID(&lfsck->li_pos_current.lp_dir_parent),
3865                lfsck->li_pos_current.lp_dir_cookie, rc);
3866
3867         return rc > 0 ? 0 : rc;
3868 }
3869
3870 static int lfsck_namespace_prep(const struct lu_env *env,
3871                                 struct lfsck_component *com,
3872                                 struct lfsck_start_param *lsp)
3873 {
3874         struct lfsck_instance   *lfsck  = com->lc_lfsck;
3875         struct lfsck_namespace  *ns     = com->lc_file_ram;
3876         struct lfsck_position   *pos    = &com->lc_pos_start;
3877         int                      rc;
3878
3879         rc = lfsck_namespace_load_bitmap(env, com);
3880         if (rc != 0 || ns->ln_status == LS_COMPLETED) {
3881                 rc = lfsck_namespace_reset(env, com, false);
3882                 if (rc == 0)
3883                         rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
3884
3885                 if (rc != 0) {
3886                         CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
3887                                "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
3888
3889                         return rc;
3890                 }
3891         }
3892
3893         down_write(&com->lc_sem);
3894         ns->ln_time_latest_start = cfs_time_current_sec();
3895         spin_lock(&lfsck->li_lock);
3896
3897         if (ns->ln_flags & LF_SCANNED_ONCE) {
3898                 if (!lfsck->li_drop_dryrun ||
3899                     lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
3900                         ns->ln_status = LS_SCANNING_PHASE2;
3901                         list_move_tail(&com->lc_link,
3902                                        &lfsck->li_list_double_scan);
3903                         if (!list_empty(&com->lc_link_dir))
3904                                 list_del_init(&com->lc_link_dir);
3905                         lfsck_pos_set_zero(pos);
3906                 } else {
3907                         ns->ln_status = LS_SCANNING_PHASE1;
3908                         ns->ln_run_time_phase1 = 0;
3909                         ns->ln_run_time_phase2 = 0;
3910                         ns->ln_items_checked = 0;
3911                         ns->ln_items_repaired = 0;
3912                         ns->ln_items_failed = 0;
3913                         ns->ln_dirs_checked = 0;
3914                         ns->ln_objs_checked_phase2 = 0;
3915                         ns->ln_objs_repaired_phase2 = 0;
3916                         ns->ln_objs_failed_phase2 = 0;
3917                         ns->ln_objs_nlink_repaired = 0;
3918                         ns->ln_dirent_repaired = 0;
3919                         ns->ln_linkea_repaired = 0;
3920                         ns->ln_mul_linked_checked = 0;
3921                         ns->ln_mul_linked_repaired = 0;
3922                         ns->ln_unknown_inconsistency = 0;
3923                         ns->ln_unmatched_pairs_repaired = 0;
3924                         ns->ln_dangling_repaired = 0;
3925                         ns->ln_mul_ref_repaired = 0;
3926                         ns->ln_bad_type_repaired = 0;
3927                         ns->ln_lost_dirent_repaired = 0;
3928                         ns->ln_striped_dirs_scanned = 0;
3929                         ns->ln_striped_dirs_repaired = 0;
3930                         ns->ln_striped_dirs_failed = 0;
3931                         ns->ln_striped_dirs_disabled = 0;
3932                         ns->ln_striped_dirs_skipped = 0;
3933                         ns->ln_striped_shards_scanned = 0;
3934                         ns->ln_striped_shards_repaired = 0;
3935                         ns->ln_striped_shards_failed = 0;
3936                         ns->ln_striped_shards_skipped = 0;
3937                         ns->ln_name_hash_repaired = 0;
3938                         fid_zero(&ns->ln_fid_latest_scanned_phase2);
3939                         if (list_empty(&com->lc_link_dir))
3940                                 list_add_tail(&com->lc_link_dir,
3941                                               &lfsck->li_list_dir);
3942                         *pos = ns->ln_pos_first_inconsistent;
3943                 }
3944         } else {
3945                 ns->ln_status = LS_SCANNING_PHASE1;
3946                 if (list_empty(&com->lc_link_dir))
3947                         list_add_tail(&com->lc_link_dir,
3948                                       &lfsck->li_list_dir);
3949                 if (!lfsck->li_drop_dryrun ||
3950                     lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
3951                         *pos = ns->ln_pos_last_checkpoint;
3952                         pos->lp_oit_cookie++;
3953                 } else {
3954                         *pos = ns->ln_pos_first_inconsistent;
3955                 }
3956         }
3957
3958         spin_unlock(&lfsck->li_lock);
3959         up_write(&com->lc_sem);
3960
3961         rc = lfsck_start_assistant(env, com, lsp);
3962
3963         CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos ["LPU64", "
3964                DFID", "LPX64"]: rc = %d\n",
3965                lfsck_lfsck2name(lfsck), pos->lp_oit_cookie,
3966                PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc);
3967
3968         return rc;
3969 }
3970
3971 static int lfsck_namespace_exec_oit(const struct lu_env *env,
3972                                     struct lfsck_component *com,
3973                                     struct dt_object *obj)
3974 {
3975         struct lfsck_thread_info *info  = lfsck_env_info(env);
3976         struct lfsck_namespace   *ns    = com->lc_file_ram;
3977         struct lfsck_instance    *lfsck = com->lc_lfsck;
3978         const struct lu_fid      *fid   = lfsck_dto2fid(obj);
3979         struct lu_attr           *la    = &info->lti_la;
3980         struct lu_fid            *pfid  = &info->lti_fid2;
3981         struct lu_name           *cname = &info->lti_name;
3982         struct lu_seq_range      *range = &info->lti_range;
3983         struct dt_device         *dev   = lfsck->li_bottom;
3984         struct seq_server_site   *ss    =
3985                                 lu_site2seq(dev->dd_lu_dev.ld_site);
3986         struct linkea_data        ldata = { 0 };
3987         __u32                     idx   = lfsck_dev_idx(dev);
3988         int                       rc;
3989         ENTRY;
3990
3991         rc = lfsck_links_read(env, obj, &ldata);
3992         if (rc == -ENOENT)
3993                 GOTO(out, rc = 0);
3994
3995         /* -EINVAL means crashed linkEA, should be verified. */
3996         if (rc == -EINVAL) {
3997                 rc = lfsck_namespace_trace_update(env, com, fid,
3998                                                   LNTF_CHECK_LINKEA, true);
3999                 if (rc == 0) {
4000                         struct lustre_handle lh = { 0 };
4001
4002                         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4003                                               MDS_INODELOCK_UPDATE |
4004                                               MDS_INODELOCK_XATTR, LCK_EX);
4005                         if (rc == 0) {
4006                                 rc = lfsck_namespace_links_remove(env, com,
4007                                                                   obj);
4008                                 lfsck_ibits_unlock(&lh, LCK_EX);
4009                         }
4010                 }
4011
4012                 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
4013         }
4014
4015         /* zero-linkEA object may be orphan, but it also maybe because
4016          * of upgrading. Currently, we cannot record it for double scan.
4017          * Because it may cause the LFSCK trace file to be too large. */
4018         if (rc == -ENODATA) {
4019                 if (S_ISDIR(lfsck_object_type(obj)))
4020                         GOTO(out, rc = 0);
4021
4022                 rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
4023                 if (rc != 0)
4024                         GOTO(out, rc);
4025
4026                 /* "la_ctime" == 1 means that it has ever been removed from
4027                  * backend /lost+found directory but not been added back to
4028                  * the normal namespace yet. */
4029                 if (la->la_nlink > 1 || unlikely(la->la_ctime == 1))
4030                         rc = lfsck_namespace_trace_update(env, com, fid,
4031                                                 LNTF_CHECK_LINKEA, true);
4032
4033                 GOTO(out, rc);
4034         }
4035
4036         if (rc != 0)
4037                 GOTO(out, rc);
4038
4039         /* Record multiple-linked object. */
4040         if (ldata.ld_leh->leh_reccount > 1) {
4041                 rc = lfsck_namespace_trace_update(env, com, fid,
4042                                                   LNTF_CHECK_LINKEA, true);
4043
4044                 GOTO(out, rc);
4045         }
4046
4047         linkea_first_entry(&ldata);
4048         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
4049         if (!fid_is_sane(pfid)) {
4050                 rc = lfsck_namespace_trace_update(env, com, fid,
4051                                                   LNTF_CHECK_PARENT, true);
4052         } else {
4053                 fld_range_set_mdt(range);
4054                 rc = fld_local_lookup(env, ss->ss_server_fld,
4055                                       fid_seq(pfid), range);
4056                 if ((rc == -ENOENT) ||
4057                     (rc == 0 && range->lsr_index != idx)) {
4058                         rc = lfsck_namespace_trace_update(env, com, fid,
4059                                                 LNTF_CHECK_LINKEA, true);
4060                 } else {
4061                         if (S_ISDIR(lfsck_object_type(obj)))
4062                                 GOTO(out, rc = 0);
4063
4064                         rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
4065                         if (rc != 0)
4066                                 GOTO(out, rc);
4067
4068                         /* "la_ctime" == 1 means that it has ever been
4069                          * removed from backend /lost+found directory but
4070                          * not been added back to the normal namespace yet. */
4071                         if (la->la_nlink > 1 || unlikely(la->la_ctime == 1))
4072                                 rc = lfsck_namespace_trace_update(env, com,
4073                                                 fid, LNTF_CHECK_LINKEA, true);
4074                 }
4075         }
4076
4077         GOTO(out, rc);
4078
4079 out:
4080         down_write(&com->lc_sem);
4081         com->lc_new_checked++;
4082         if (S_ISDIR(lfsck_object_type(obj)))
4083                 ns->ln_dirs_checked++;
4084         if (rc != 0)
4085                 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4086         up_write(&com->lc_sem);
4087
4088         return rc;
4089 }
4090
4091 static int lfsck_namespace_exec_dir(const struct lu_env *env,
4092                                     struct lfsck_component *com,
4093                                     struct lu_dirent *ent, __u16 type)
4094 {
4095         struct lfsck_assistant_data     *lad    = com->lc_data;
4096         struct lfsck_namespace_req      *lnr;
4097         bool                             wakeup = false;
4098
4099         lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, ent, type);
4100         if (IS_ERR(lnr)) {
4101                 struct lfsck_namespace *ns = com->lc_file_ram;
4102
4103                 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4104                 return PTR_ERR(lnr);
4105         }
4106
4107         spin_lock(&lad->lad_lock);
4108         if (lad->lad_assistant_status < 0) {
4109                 spin_unlock(&lad->lad_lock);
4110                 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4111                 return lad->lad_assistant_status;
4112         }
4113
4114         list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4115         if (lad->lad_prefetched == 0)
4116                 wakeup = true;
4117
4118         lad->lad_prefetched++;
4119         spin_unlock(&lad->lad_lock);
4120         if (wakeup)
4121                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4122
4123         down_write(&com->lc_sem);
4124         com->lc_new_checked++;
4125         up_write(&com->lc_sem);
4126
4127         return 0;
4128 }
4129
4130 static int lfsck_namespace_post(const struct lu_env *env,
4131                                 struct lfsck_component *com,
4132                                 int result, bool init)
4133 {
4134         struct lfsck_instance   *lfsck = com->lc_lfsck;
4135         struct lfsck_namespace  *ns    = com->lc_file_ram;
4136         int                      rc;
4137         ENTRY;
4138
4139         lfsck_post_generic(env, com, &result);
4140
4141         down_write(&com->lc_sem);
4142         lfsck_namespace_release_lmv(env, com);
4143
4144         spin_lock(&lfsck->li_lock);
4145         if (!init)
4146                 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4147         if (result > 0) {
4148                 ns->ln_status = LS_SCANNING_PHASE2;
4149                 ns->ln_flags |= LF_SCANNED_ONCE;
4150                 ns->ln_flags &= ~LF_UPGRADE;
4151                 list_del_init(&com->lc_link_dir);
4152                 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4153         } else if (result == 0) {
4154                 if (lfsck->li_status != 0)
4155                         ns->ln_status = lfsck->li_status;
4156                 else
4157                         ns->ln_status = LS_STOPPED;
4158                 if (ns->ln_status != LS_PAUSED) {
4159                         list_del_init(&com->lc_link_dir);
4160                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4161                 }
4162         } else {
4163                 ns->ln_status = LS_FAILED;
4164                 list_del_init(&com->lc_link_dir);
4165                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4166         }
4167         spin_unlock(&lfsck->li_lock);
4168
4169         if (!init) {
4170                 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4171                                 HALF_SEC - lfsck->li_time_last_checkpoint);
4172                 ns->ln_time_last_checkpoint = cfs_time_current_sec();
4173                 ns->ln_items_checked += com->lc_new_checked;
4174                 com->lc_new_checked = 0;
4175         }
4176
4177         rc = lfsck_namespace_store(env, com);
4178         up_write(&com->lc_sem);
4179
4180         CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
4181                lfsck_lfsck2name(lfsck), rc);
4182
4183         RETURN(rc);
4184 }
4185
4186 static int
4187 lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
4188                      struct seq_file *m)
4189 {
4190         struct lfsck_instance   *lfsck = com->lc_lfsck;
4191         struct lfsck_bookmark   *bk    = &lfsck->li_bookmark_ram;
4192         struct lfsck_namespace  *ns    = com->lc_file_ram;
4193         int                      rc;
4194
4195         down_read(&com->lc_sem);
4196         seq_printf(m, "name: lfsck_namespace\n"
4197                    "magic: %#x\n"
4198                    "version: %d\n"
4199                    "status: %s\n",
4200                    ns->ln_magic,
4201                    bk->lb_version,
4202                    lfsck_status2names(ns->ln_status));
4203
4204         rc = lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
4205         if (rc < 0)
4206                 goto out;
4207
4208         rc = lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4209         if (rc < 0)
4210                 goto out;
4211
4212         rc = lfsck_time_dump(m, ns->ln_time_last_complete,
4213                              "time_since_last_completed");
4214         if (rc < 0)
4215                 goto out;
4216
4217         rc = lfsck_time_dump(m, ns->ln_time_latest_start,
4218                              "time_since_latest_start");
4219         if (rc < 0)
4220                 goto out;
4221
4222         rc = lfsck_time_dump(m, ns->ln_time_last_checkpoint,
4223                              "time_since_last_checkpoint");
4224         if (rc < 0)
4225                 goto out;
4226
4227         rc = lfsck_pos_dump(m, &ns->ln_pos_latest_start,
4228                             "latest_start_position");
4229         if (rc < 0)
4230                 goto out;
4231
4232         rc = lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
4233                             "last_checkpoint_position");
4234         if (rc < 0)
4235                 goto out;
4236
4237         rc = lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
4238                             "first_failure_position");
4239         if (rc < 0)
4240                 goto out;
4241
4242         if (ns->ln_status == LS_SCANNING_PHASE1) {
4243                 struct lfsck_position pos;
4244                 const struct dt_it_ops *iops;
4245                 cfs_duration_t duration = cfs_time_current() -
4246                                           lfsck->li_time_last_checkpoint;
4247                 __u64 checked = ns->ln_items_checked + com->lc_new_checked;
4248                 __u64 speed = checked;
4249                 __u64 new_checked = com->lc_new_checked * HZ;
4250                 __u32 rtime = ns->ln_run_time_phase1 +
4251                               cfs_duration_sec(duration + HALF_SEC);
4252
4253                 if (duration != 0)
4254                         do_div(new_checked, duration);
4255                 if (rtime != 0)
4256                         do_div(speed, rtime);
4257                 lfsck_namespace_dump_statistics(m, ns, checked,
4258                                                 ns->ln_objs_checked_phase2,
4259                                                 rtime, ns->ln_run_time_phase2);
4260
4261                 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4262                               "average_speed_phase2: N/A\n"
4263                               "real_time_speed_phase1: "LPU64" items/sec\n"
4264                               "real_time_speed_phase2: N/A\n",
4265                               speed,
4266                               new_checked);
4267
4268                 LASSERT(lfsck->li_di_oit != NULL);
4269
4270                 iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
4271
4272                 /* The low layer otable-based iteration position may NOT
4273                  * exactly match the namespace-based directory traversal
4274                  * cookie. Generally, it is not a serious issue. But the
4275                  * caller should NOT make assumption on that. */
4276                 pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
4277                 if (!lfsck->li_current_oit_processed)
4278                         pos.lp_oit_cookie--;
4279
4280                 spin_lock(&lfsck->li_lock);
4281                 if (lfsck->li_di_dir != NULL) {
4282                         pos.lp_dir_cookie = lfsck->li_cookie_dir;
4283                         if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
4284                                 fid_zero(&pos.lp_dir_parent);
4285                                 pos.lp_dir_cookie = 0;
4286                         } else {
4287                                 pos.lp_dir_parent =
4288                                         *lfsck_dto2fid(lfsck->li_obj_dir);
4289                         }
4290                 } else {
4291                         fid_zero(&pos.lp_dir_parent);
4292                         pos.lp_dir_cookie = 0;
4293                 }
4294                 spin_unlock(&lfsck->li_lock);
4295                 lfsck_pos_dump(m, &pos, "current_position");
4296         } else if (ns->ln_status == LS_SCANNING_PHASE2) {
4297                 cfs_duration_t duration = cfs_time_current() -
4298                                           lfsck->li_time_last_checkpoint;
4299                 __u64 checked = ns->ln_objs_checked_phase2 +
4300                                 com->lc_new_checked;
4301                 __u64 speed1 = ns->ln_items_checked;
4302                 __u64 speed2 = checked;
4303                 __u64 new_checked = com->lc_new_checked * HZ;
4304                 __u32 rtime = ns->ln_run_time_phase2 +
4305                               cfs_duration_sec(duration + HALF_SEC);
4306
4307                 if (duration != 0)
4308                         do_div(new_checked, duration);
4309                 if (ns->ln_run_time_phase1 != 0)
4310                         do_div(speed1, ns->ln_run_time_phase1);
4311                 if (rtime != 0)
4312                         do_div(speed2, rtime);
4313                 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4314                                                 checked,
4315                                                 ns->ln_run_time_phase1, rtime);
4316
4317                 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4318                               "average_speed_phase2: "LPU64" objs/sec\n"
4319                               "real_time_speed_phase1: N/A\n"
4320                               "real_time_speed_phase2: "LPU64" objs/sec\n"
4321                               "current_position: "DFID"\n",
4322                               speed1,
4323                               speed2,
4324                               new_checked,
4325                               PFID(&ns->ln_fid_latest_scanned_phase2));
4326         } else {
4327                 __u64 speed1 = ns->ln_items_checked;
4328                 __u64 speed2 = ns->ln_objs_checked_phase2;
4329
4330                 if (ns->ln_run_time_phase1 != 0)
4331                         do_div(speed1, ns->ln_run_time_phase1);
4332                 if (ns->ln_run_time_phase2 != 0)
4333                         do_div(speed2, ns->ln_run_time_phase2);
4334                 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4335                                                 ns->ln_objs_checked_phase2,
4336                                                 ns->ln_run_time_phase1,
4337                                                 ns->ln_run_time_phase2);
4338
4339                 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4340                               "average_speed_phase2: "LPU64" objs/sec\n"
4341                               "real_time_speed_phase1: N/A\n"
4342                               "real_time_speed_phase2: N/A\n"
4343                               "current_position: N/A\n",
4344                               speed1,
4345                               speed2);
4346         }
4347 out:
4348         up_read(&com->lc_sem);
4349         return 0;
4350 }
4351
4352 static int lfsck_namespace_double_scan(const struct lu_env *env,
4353                                        struct lfsck_component *com)
4354 {
4355         struct lfsck_namespace          *ns     = com->lc_file_ram;
4356         struct lfsck_assistant_data     *lad    = com->lc_data;
4357         struct lfsck_tgt_descs          *ltds   = &com->lc_lfsck->li_mdt_descs;
4358         struct lfsck_tgt_desc           *ltd;
4359         struct lfsck_tgt_desc           *next;
4360         int                              rc;
4361
4362         rc = lfsck_double_scan_generic(env, com, ns->ln_status);
4363         if (thread_is_stopped(&lad->lad_thread)) {
4364                 LASSERT(list_empty(&lad->lad_req_list));
4365                 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4366
4367                 spin_lock(&ltds->ltd_lock);
4368                 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4369                                          ltd_namespace_phase_list) {
4370                         list_del_init(&ltd->ltd_namespace_phase_list);
4371                 }
4372                 spin_unlock(&ltds->ltd_lock);
4373         }
4374
4375         return rc;
4376 }
4377
4378 static void lfsck_namespace_data_release(const struct lu_env *env,
4379                                          struct lfsck_component *com)
4380 {
4381         struct lfsck_assistant_data     *lad    = com->lc_data;
4382         struct lfsck_tgt_descs          *ltds   = &com->lc_lfsck->li_mdt_descs;
4383         struct lfsck_tgt_desc           *ltd;
4384         struct lfsck_tgt_desc           *next;
4385
4386         LASSERT(lad != NULL);
4387         LASSERT(thread_is_init(&lad->lad_thread) ||
4388                 thread_is_stopped(&lad->lad_thread));
4389         LASSERT(list_empty(&lad->lad_req_list));
4390
4391         com->lc_data = NULL;
4392         lfsck_namespace_release_lmv(env, com);
4393
4394         spin_lock(&ltds->ltd_lock);
4395         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4396                                  ltd_namespace_phase_list) {
4397                 list_del_init(&ltd->ltd_namespace_phase_list);
4398         }
4399         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4400                                  ltd_namespace_phase_list) {
4401                 list_del_init(&ltd->ltd_namespace_phase_list);
4402         }
4403         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
4404                                  ltd_namespace_list) {
4405                 list_del_init(&ltd->ltd_namespace_list);
4406         }
4407         spin_unlock(&ltds->ltd_lock);
4408
4409         if (likely(lad->lad_bitmap != NULL))
4410                 CFS_FREE_BITMAP(lad->lad_bitmap);
4411
4412         OBD_FREE_PTR(lad);
4413 }
4414
4415 static void lfsck_namespace_quit(const struct lu_env *env,
4416                                  struct lfsck_component *com)
4417 {
4418         struct lfsck_assistant_data     *lad    = com->lc_data;
4419         struct lfsck_tgt_descs          *ltds   = &com->lc_lfsck->li_mdt_descs;
4420         struct lfsck_tgt_desc           *ltd;
4421         struct lfsck_tgt_desc           *next;
4422
4423         LASSERT(lad != NULL);
4424
4425         lfsck_quit_generic(env, com);
4426
4427         LASSERT(thread_is_init(&lad->lad_thread) ||
4428                 thread_is_stopped(&lad->lad_thread));
4429         LASSERT(list_empty(&lad->lad_req_list));
4430
4431         lfsck_namespace_release_lmv(env, com);
4432
4433         spin_lock(&ltds->ltd_lock);
4434         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4435                                  ltd_namespace_phase_list) {
4436                 list_del_init(&ltd->ltd_namespace_phase_list);
4437         }
4438         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4439                                  ltd_namespace_phase_list) {
4440                 list_del_init(&ltd->ltd_namespace_phase_list);
4441         }
4442         spin_unlock(&ltds->ltd_lock);
4443 }
4444
4445 static int lfsck_namespace_in_notify(const struct lu_env *env,
4446                                      struct lfsck_component *com,
4447                                      struct lfsck_request *lr,
4448                                      struct thandle *th)
4449 {
4450         struct lfsck_instance           *lfsck = com->lc_lfsck;
4451         struct lfsck_namespace          *ns    = com->lc_file_ram;
4452         struct lfsck_assistant_data     *lad   = com->lc_data;
4453         struct lfsck_tgt_descs          *ltds  = &lfsck->li_mdt_descs;
4454         struct lfsck_tgt_desc           *ltd;
4455         int                              rc;
4456         bool                             fail  = false;
4457         ENTRY;
4458
4459         switch (lr->lr_event) {
4460         case LE_CREATE_ORPHAN: {
4461                 struct dt_object *orphan = NULL;
4462                 struct lmv_mds_md_v1 *lmv;
4463
4464                 CDEBUG(D_LFSCK, "%s: namespace LFSCK handling notify from "
4465                        "MDT %x to create orphan"DFID" with type %o\n",
4466                        lfsck_lfsck2name(lfsck), lr->lr_index,
4467                        PFID(&lr->lr_fid), lr->lr_type);
4468
4469                 orphan = lfsck_object_find(env, lfsck, &lr->lr_fid);
4470                 if (IS_ERR(orphan))
4471                         GOTO(out_create, rc = PTR_ERR(orphan));
4472
4473                 if (dt_object_exists(orphan))
4474                         GOTO(out_create, rc = -EEXIST);
4475
4476                 if (lr->lr_stripe_count > 0) {
4477                         lmv = &lfsck_env_info(env)->lti_lmv;
4478                         memset(lmv, 0, sizeof(*lmv));
4479                         lmv->lmv_hash_type = lr->lr_hash_type;
4480                         lmv->lmv_stripe_count = lr->lr_stripe_count;
4481                         lmv->lmv_layout_version = lr->lr_layout_version;
4482                         memcpy(lmv->lmv_pool_name, lr->lr_pool_name,
4483                                sizeof(lmv->lmv_pool_name));
4484                 } else {
4485                         lmv = NULL;
4486                 }
4487
4488                 rc = lfsck_namespace_create_orphan_local(env, com, orphan,
4489                                                          lr->lr_type, lmv);
4490
4491                 GOTO(out_create, rc = (rc == 1) ? 0 : rc);
4492
4493 out_create:
4494                 CDEBUG(D_LFSCK, "%s: namespace LFSCK handled notify from "
4495                        "MDT %x to create orphan"DFID" with type %o: rc = %d\n",
4496                        lfsck_lfsck2name(lfsck), lr->lr_index,
4497                        PFID(&lr->lr_fid), lr->lr_type, rc);
4498
4499                 if (orphan != NULL && !IS_ERR(orphan))
4500                         lfsck_object_put(env, orphan);
4501
4502                 return rc;
4503         }
4504         case LE_SKIP_NLINK_DECLARE: {
4505                 struct dt_object        *obj   = com->lc_obj;
4506                 struct lu_fid           *key   = &lfsck_env_info(env)->lti_fid3;
4507                 __u8                     flags = 0;
4508
4509                 LASSERT(th != NULL);
4510
4511                 rc = dt_declare_delete(env, obj,
4512                                        (const struct dt_key *)key, th);
4513                 if (rc == 0)
4514                         rc = dt_declare_insert(env, obj,
4515                                                (const struct dt_rec *)&flags,
4516                                                (const struct dt_key *)key, th);
4517
4518                 RETURN(rc);
4519         }
4520         case LE_SKIP_NLINK: {
4521                 struct dt_object        *obj   = com->lc_obj;
4522                 struct lu_fid           *key   = &lfsck_env_info(env)->lti_fid3;
4523                 __u8                     flags = 0;
4524                 bool                     exist = false;
4525                 ENTRY;
4526
4527                 LASSERT(th != NULL);
4528
4529                 fid_cpu_to_be(key, &lr->lr_fid);
4530                 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
4531                                (const struct dt_key *)key, BYPASS_CAPA);
4532                 if (rc == 0) {
4533                         if (flags & LNTF_SKIP_NLINK)
4534                                 RETURN(0);
4535
4536                         exist = true;
4537                 } else if (rc != -ENOENT) {
4538                         GOTO(log, rc);
4539                 }
4540
4541                 flags |= LNTF_SKIP_NLINK;
4542                 if (exist) {
4543                         rc = dt_delete(env, obj, (const struct dt_key *)key,
4544                                        th, BYPASS_CAPA);
4545                         if (rc != 0)
4546                                 GOTO(log, rc);
4547                 }
4548
4549                 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
4550                                (const struct dt_key *)key, th, BYPASS_CAPA, 1);
4551
4552                 GOTO(log, rc);
4553
4554 log:
4555                 CDEBUG(D_LFSCK, "%s: RPC service thread mark the "DFID
4556                        " to be skipped for namespace double scan: rc = %d\n",
4557                        lfsck_lfsck2name(com->lc_lfsck), PFID(&lr->lr_fid), rc);
4558
4559                 if (rc != 0)
4560                         /* If we cannot record this object in the LFSCK tracing,
4561                          * we have to mark the LFSC as LF_INCOMPLETE, then the
4562                          * LFSCK will skip nlink attribute verification for
4563                          * all objects. */
4564                         ns->ln_flags |= LF_INCOMPLETE;
4565
4566                 return 0;
4567         }
4568         case LE_SET_LMV_MASTER: {
4569                 struct dt_object        *obj;
4570
4571                 obj = lfsck_object_find_by_dev(env, lfsck->li_bottom,
4572                                                &lr->lr_fid);
4573                 if (IS_ERR(obj))
4574                         RETURN(PTR_ERR(obj));
4575
4576                 rc = lfsck_namespace_notify_lmv_master_local(env, com, obj);
4577                 lfsck_object_put(env, obj);
4578
4579                 RETURN(rc > 0 ? 0 : rc);
4580         }
4581         case LE_PHASE1_DONE:
4582         case LE_PHASE2_DONE:
4583         case LE_PEER_EXIT:
4584                 break;
4585         default:
4586                 RETURN(-EINVAL);
4587         }
4588
4589         CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
4590                "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
4591                lr->lr_index, lr->lr_status, lr->lr_flags2);
4592
4593         spin_lock(&ltds->ltd_lock);
4594         ltd = LTD_TGT(ltds, lr->lr_index);
4595         if (ltd == NULL) {
4596                 spin_unlock(&ltds->ltd_lock);
4597
4598                 RETURN(-ENXIO);
4599         }
4600
4601         list_del_init(&ltd->ltd_namespace_phase_list);
4602         switch (lr->lr_event) {
4603         case LE_PHASE1_DONE:
4604                 if (lr->lr_status <= 0) {
4605                         ltd->ltd_namespace_done = 1;
4606                         list_del_init(&ltd->ltd_namespace_list);
4607                         CDEBUG(D_LFSCK, "%s: MDT %x failed/stopped at "
4608                                "phase1 for namespace LFSCK: rc = %d.\n",
4609                                lfsck_lfsck2name(lfsck),
4610                                ltd->ltd_index, lr->lr_status);
4611                         ns->ln_flags |= LF_INCOMPLETE;
4612                         fail = true;
4613                         break;
4614                 }
4615
4616                 if (lr->lr_flags2 & LF_INCOMPLETE)
4617                         ns->ln_flags |= LF_INCOMPLETE;
4618
4619                 if (list_empty(&ltd->ltd_namespace_list))
4620                         list_add_tail(&ltd->ltd_namespace_list,
4621                                       &lad->lad_mdt_list);
4622                 list_add_tail(&ltd->ltd_namespace_phase_list,
4623                               &lad->lad_mdt_phase2_list);
4624                 break;
4625         case LE_PHASE2_DONE:
4626                 ltd->ltd_namespace_done = 1;
4627                 list_del_init(&ltd->ltd_namespace_list);
4628                 break;
4629         case LE_PEER_EXIT:
4630                 fail = true;
4631                 ltd->ltd_namespace_done = 1;
4632                 list_del_init(&ltd->ltd_namespace_list);
4633                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
4634                         CDEBUG(D_LFSCK,
4635                                "%s: the peer MDT %x exit namespace LFSCK\n",
4636                                lfsck_lfsck2name(lfsck), ltd->ltd_index);
4637                         ns->ln_flags |= LF_INCOMPLETE;
4638                 }
4639                 break;
4640         default:
4641                 break;
4642         }
4643         spin_unlock(&ltds->ltd_lock);
4644
4645         if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
4646                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
4647
4648                 memset(stop, 0, sizeof(*stop));
4649                 stop->ls_status = lr->lr_status;
4650                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
4651                 lfsck_stop(env, lfsck->li_bottom, stop);
4652         } else if (lfsck_phase2_next_ready(lad)) {
4653                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4654         }
4655
4656         RETURN(0);
4657 }
4658
4659 static int lfsck_namespace_query(const struct lu_env *env,
4660                                  struct lfsck_component *com)
4661 {
4662         struct lfsck_namespace *ns = com->lc_file_ram;
4663
4664         return ns->ln_status;
4665 }
4666
4667 static struct lfsck_operations lfsck_namespace_ops = {
4668         .lfsck_reset            = lfsck_namespace_reset,
4669         .lfsck_fail             = lfsck_namespace_fail,
4670         .lfsck_close_dir        = lfsck_namespace_close_dir,
4671         .lfsck_open_dir         = lfsck_namespace_open_dir,
4672         .lfsck_checkpoint       = lfsck_namespace_checkpoint,
4673         .lfsck_prep             = lfsck_namespace_prep,
4674         .lfsck_exec_oit         = lfsck_namespace_exec_oit,
4675         .lfsck_exec_dir         = lfsck_namespace_exec_dir,
4676         .lfsck_post             = lfsck_namespace_post,
4677         .lfsck_dump             = lfsck_namespace_dump,
4678         .lfsck_double_scan      = lfsck_namespace_double_scan,
4679         .lfsck_data_release     = lfsck_namespace_data_release,
4680         .lfsck_quit             = lfsck_namespace_quit,
4681         .lfsck_in_notify        = lfsck_namespace_in_notify,
4682         .lfsck_query            = lfsck_namespace_query,
4683 };
4684
4685 /**
4686  * Repair dangling name entry.
4687  *
4688  * For the name entry with dangling reference, we need to repare the
4689  * inconsistency according to the LFSCK sponsor's requirement:
4690  *
4691  * 1) Keep the inconsistency there and report the inconsistency case,
4692  *    then give the chance to the application to find related issues,
4693  *    and the users can make the decision about how to handle it with
4694  *    more human knownledge. (by default)
4695  *
4696  * 2) Re-create the missing MDT-object with the FID information.
4697  *
4698  * \param[in] env       pointer to the thread context
4699  * \param[in] com       pointer to the lfsck component
4700  * \param[in] child     pointer to the object corresponding to the dangling
4701  *                      name entry
4702  * \param[in] lnr       pointer to the namespace request that contains the
4703  *                      name's name, parent object, parent's LMV, and ect.
4704  *
4705  * \retval              positive number if no need to repair
4706  * \retval              zero for repaired successfully
4707  * \retval              negative error number on failure
4708  */
4709 int lfsck_namespace_repair_dangling(const struct lu_env *env,
4710                                     struct lfsck_component *com,
4711                                     struct dt_object *child,
4712                                     struct lfsck_namespace_req *lnr)
4713 {
4714         struct lfsck_thread_info        *info   = lfsck_env_info(env);
4715         struct lu_attr                  *la     = &info->lti_la;
4716         struct dt_allocation_hint       *hint   = &info->lti_hint;
4717         struct dt_object_format         *dof    = &info->lti_dof;
4718         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
4719         struct lmv_mds_md_v1            *lmv2   = &info->lti_lmv2;
4720         struct dt_object                *parent = lnr->lnr_obj;
4721         const struct lu_name            *cname;
4722         struct linkea_data               ldata  = { 0 };
4723         struct lustre_handle             lh     = { 0 };
4724         struct lu_buf                    linkea_buf;
4725         struct lu_buf                    lmv_buf;
4726         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4727         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
4728         struct dt_device                *dev    = lfsck_obj2dt_dev(child);
4729         struct thandle                  *th     = NULL;
4730         int                              rc     = 0;
4731         __u16                            type   = lnr->lnr_type;
4732         bool                             create;
4733         ENTRY;
4734
4735         cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
4736         if (bk->lb_param & LPF_CREATE_MDTOBJ)
4737                 create = true;
4738         else
4739                 create = false;
4740
4741         if (!create || bk->lb_param & LPF_DRYRUN)
4742                 GOTO(log, rc = 0);
4743
4744         rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
4745         if (rc != 0)
4746                 GOTO(log, rc);
4747
4748         rc = linkea_add_buf(&ldata, cname, lfsck_dto2fid(parent));
4749         if (rc != 0)
4750                 GOTO(log, rc);
4751
4752         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
4753                               MDS_INODELOCK_UPDATE, LCK_EX);
4754         if (rc != 0)
4755                 GOTO(log, rc);
4756
4757         rc = lfsck_namespace_check_exist(env, parent, child, lnr->lnr_name);
4758         if (rc != 0)
4759                 GOTO(log, rc);
4760
4761         th = dt_trans_create(env, dev);
4762         if (IS_ERR(th))
4763                 GOTO(log, rc = PTR_ERR(th));
4764
4765         /* Set the ctime as zero, then others can know it is created for
4766          * repairing dangling name entry by LFSCK. And if the LFSCK made
4767          * wrong decision and the real MDT-object has been found later,
4768          * then the LFSCK has chance to fix the incosistency properly. */
4769         memset(la, 0, sizeof(*la));
4770         la->la_mode = (type & S_IFMT) | 0600;
4771         la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
4772                         LA_ATIME | LA_MTIME | LA_CTIME;
4773
4774         child->do_ops->do_ah_init(env, hint, parent, child,
4775                                   la->la_mode & S_IFMT);
4776
4777         memset(dof, 0, sizeof(*dof));
4778         dof->dof_type = dt_mode_to_dft(type);
4779         /* If the target is a regular file, then the LFSCK will only create
4780          * the MDT-object without stripes (dof->dof_reg.striped = 0). related
4781          * OST-objects will be created when write open. */
4782
4783         /* 1a. create child. */
4784         rc = dt_declare_create(env, child, la, hint, dof, th);
4785         if (rc != 0)
4786                 GOTO(stop, rc);
4787
4788         if (S_ISDIR(type)) {
4789                 if (unlikely(!dt_try_as_dir(env, child)))
4790                         GOTO(stop, rc = -ENOTDIR);
4791
4792                 /* 2a. insert dot into child dir */
4793                 rec->rec_type = S_IFDIR;
4794                 rec->rec_fid = lfsck_dto2fid(child);
4795                 rc = dt_declare_insert(env, child,
4796                                        (const struct dt_rec *)rec,
4797                                        (const struct dt_key *)dot, th);
4798                 if (rc != 0)
4799                         GOTO(stop, rc);
4800
4801                 /* 3a. insert dotdot into child dir */
4802                 rec->rec_fid = lfsck_dto2fid(parent);
4803                 rc = dt_declare_insert(env, child,
4804                                        (const struct dt_rec *)rec,
4805                                        (const struct dt_key *)dotdot, th);
4806                 if (rc != 0)
4807                         GOTO(stop, rc);
4808
4809                 /* 4a. increase child nlink */
4810                 rc = dt_declare_ref_add(env, child, th);
4811                 if (rc != 0)
4812                         GOTO(stop, rc);
4813
4814                 /* 5a. generate slave LMV EA. */
4815                 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
4816                         int idx;
4817
4818                         idx = lfsck_shard_name_to_index(env,
4819                                         lnr->lnr_name, lnr->lnr_namelen,
4820                                         type, lfsck_dto2fid(child));
4821                         if (unlikely(idx < 0))
4822                                 GOTO(stop, rc = idx);
4823
4824                         *lmv2 = lnr->lnr_lmv->ll_lmv;
4825                         lmv2->lmv_magic = LMV_MAGIC_STRIPE;
4826                         lmv2->lmv_master_mdt_index = idx;
4827
4828                         lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
4829                         lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
4830                         rc = dt_declare_xattr_set(env, child, &lmv_buf,
4831                                                   XATTR_NAME_LMV, 0, th);
4832                         if (rc != 0)
4833                                 GOTO(stop, rc);
4834                 }
4835         }
4836
4837         /* 6a. insert linkEA for child */
4838         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
4839                        ldata.ld_leh->leh_len);
4840         rc = dt_declare_xattr_set(env, child, &linkea_buf,
4841                                   XATTR_NAME_LINK, 0, th);
4842         if (rc != 0)
4843                 GOTO(stop, rc);
4844
4845         rc = dt_trans_start(env, dev, th);
4846         if (rc != 0)
4847                 GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
4848
4849         dt_write_lock(env, child, 0);
4850         /* 1b. create child */
4851         rc = dt_create(env, child, la, hint, dof, th);
4852         if (rc != 0)
4853                 GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc));
4854
4855         if (S_ISDIR(type)) {
4856                 if (unlikely(!dt_try_as_dir(env, child)))
4857                         GOTO(unlock, rc = -ENOTDIR);
4858
4859                 /* 2b. insert dot into child dir */
4860                 rec->rec_type = S_IFDIR;
4861                 rec->rec_fid = lfsck_dto2fid(child);
4862                 rc = dt_insert(env, child, (const struct dt_rec *)rec,
4863                                (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
4864                 if (rc != 0)
4865                         GOTO(unlock, rc);
4866
4867                 /* 3b. insert dotdot into child dir */
4868                 rec->rec_fid = lfsck_dto2fid(parent);
4869                 rc = dt_insert(env, child, (const struct dt_rec *)rec,
4870                                (const struct dt_key *)dotdot, th,
4871                                BYPASS_CAPA, 1);
4872                 if (rc != 0)
4873                         GOTO(unlock, rc);
4874
4875                 /* 4b. increase child nlink */
4876                 rc = dt_ref_add(env, child, th);
4877                 if (rc != 0)
4878                         GOTO(unlock, rc);
4879
4880                 /* 5b. generate slave LMV EA. */
4881                 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
4882                         rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV,
4883                                           0, th, BYPASS_CAPA);
4884                         if (rc != 0)
4885                                 GOTO(unlock, rc);
4886                 }
4887         }
4888
4889         /* 6b. insert linkEA for child. */
4890         rc = dt_xattr_set(env, child, &linkea_buf,
4891                           XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
4892
4893         GOTO(unlock, rc);
4894
4895 unlock:
4896         dt_write_unlock(env, child);
4897
4898 stop:
4899         dt_trans_stop(env, dev, th);
4900
4901 log:
4902         lfsck_ibits_unlock(&lh, LCK_EX);
4903         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
4904                "reference for: parent "DFID", child "DFID", type %u, "
4905                "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck),
4906                PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
4907                type, cname->ln_name,
4908                create ? "Create the lost OST-object as required" :
4909                         "Keep the MDT-object there by default", rc);
4910
4911         if (rc <= 0) {
4912                 struct lfsck_namespace *ns = com->lc_file_ram;
4913
4914                 ns->ln_flags |= LF_INCONSISTENT;
4915         }
4916
4917         return rc;
4918 }
4919
4920 static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
4921                                                 struct lfsck_component *com,
4922                                                 struct lfsck_assistant_req *lar)
4923 {
4924         struct lfsck_thread_info   *info     = lfsck_env_info(env);
4925         struct lu_attr             *la       = &info->lti_la;
4926         struct lfsck_instance      *lfsck    = com->lc_lfsck;
4927         struct lfsck_bookmark      *bk       = &lfsck->li_bookmark_ram;
4928         struct lfsck_namespace     *ns       = com->lc_file_ram;
4929         struct linkea_data          ldata    = { 0 };
4930         const struct lu_name       *cname;
4931         struct thandle             *handle   = NULL;
4932         struct lfsck_namespace_req *lnr      =
4933                         container_of0(lar, struct lfsck_namespace_req, lnr_lar);
4934         struct dt_object           *dir      = lnr->lnr_obj;
4935         struct dt_object           *obj      = NULL;
4936         const struct lu_fid        *pfid     = lfsck_dto2fid(dir);
4937         struct dt_device           *dev      = NULL;
4938         struct lustre_handle        lh       = { 0 };
4939         bool                        repaired = false;
4940         bool                        dtlocked = false;
4941         bool                        remove;
4942         bool                        newdata;
4943         bool                        log      = false;
4944         bool                        bad_hash = false;
4945         int                         idx      = 0;
4946         int                         count    = 0;
4947         int                         rc;
4948         enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
4949         ENTRY;
4950
4951         if (lnr->lnr_attr & LUDA_UPGRADE) {
4952                 ns->ln_flags |= LF_UPGRADE;
4953                 ns->ln_dirent_repaired++;
4954                 repaired = true;
4955         } else if (lnr->lnr_attr & LUDA_REPAIR) {
4956                 ns->ln_flags |= LF_INCONSISTENT;
4957                 ns->ln_dirent_repaired++;
4958                 repaired = true;
4959         }
4960
4961         if (unlikely(fid_is_zero(&lnr->lnr_fid))) {
4962                 if (strcmp(lnr->lnr_name, dotdot) != 0)
4963                         LBUG();
4964                 else
4965                         rc = lfsck_namespace_trace_update(env, com, pfid,
4966                                                 LNTF_CHECK_PARENT, true);
4967
4968                 GOTO(out, rc);
4969         }
4970
4971         if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
4972                 rc = lfsck_namespace_striped_dir_rescan(env, com, lnr);
4973
4974                 RETURN(rc);
4975         }
4976
4977         if (lnr->lnr_name[0] == '.' &&
4978             (lnr->lnr_namelen == 1 || fid_seq_is_dot(fid_seq(&lnr->lnr_fid))))
4979                 GOTO(out, rc = 0);
4980
4981         idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
4982         if (idx < 0)
4983                 GOTO(out, rc = idx);
4984
4985         if (idx == lfsck_dev_idx(lfsck->li_bottom)) {
4986                 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0))
4987                         GOTO(out, rc = 0);
4988
4989                 dev = lfsck->li_next;
4990         } else {
4991                 struct lfsck_tgt_desc *ltd;
4992
4993                 /* Usually, some local filesystem consistency verification
4994                  * tools can guarantee the local namespace tree consistenct.
4995                  * So the LFSCK will only verify the remote directory. */
4996                 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) {
4997                         rc = lfsck_namespace_trace_update(env, com, pfid,
4998                                                 LNTF_CHECK_PARENT, true);
4999
5000                         GOTO(out, rc);
5001                 }
5002
5003                 ltd = LTD_TGT(&lfsck->li_mdt_descs, idx);
5004                 if (unlikely(ltd == NULL)) {
5005                         CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
5006                                "did not join the namespace LFSCK\n",
5007                                lfsck_lfsck2name(lfsck), idx);
5008                         lfsck_lad_set_bitmap(env, com, idx);
5009
5010                         GOTO(out, rc = -ENODEV);
5011                 }
5012
5013                 dev = ltd->ltd_tgt;
5014         }
5015
5016         obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
5017         if (IS_ERR(obj))
5018                 GOTO(out, rc = PTR_ERR(obj));
5019
5020         cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5021         if (dt_object_exists(obj) == 0) {
5022
5023 dangling:
5024                 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5025                 if (rc == 0) {
5026                         if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5027                                         lnr->lnr_name, lnr->lnr_namelen)) {
5028                                 type = LNIT_BAD_DIRENT;
5029
5030                                 GOTO(out, rc);
5031                         }
5032
5033                         type = LNIT_DANGLING;
5034                         rc = lfsck_namespace_repair_dangling(env, com,
5035                                                              obj, lnr);
5036                         if (rc == 0)
5037                                 repaired = true;
5038                 }
5039
5040                 GOTO(out, rc);
5041         }
5042
5043         if (!(bk->lb_param & LPF_DRYRUN) && repaired) {
5044
5045 again:
5046                 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5047                                       MDS_INODELOCK_UPDATE |
5048                                       MDS_INODELOCK_XATTR, LCK_EX);
5049                 if (rc != 0)
5050                         GOTO(out, rc);
5051
5052                 handle = dt_trans_create(env, dev);
5053                 if (IS_ERR(handle))
5054                         GOTO(out, rc = PTR_ERR(handle));
5055
5056                 rc = lfsck_declare_namespace_exec_dir(env, obj, handle);
5057                 if (rc != 0)
5058                         GOTO(stop, rc);
5059
5060                 rc = dt_trans_start(env, dev, handle);
5061                 if (rc != 0)
5062                         GOTO(stop, rc);
5063
5064                 dt_write_lock(env, obj, 0);
5065                 dtlocked = true;
5066         }
5067
5068         rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5069         if (rc != 0)
5070                 GOTO(stop, rc);
5071
5072         rc = lfsck_links_read(env, obj, &ldata);
5073         if (unlikely(rc == -ENOENT)) {
5074                 if (handle != NULL) {
5075                         dt_write_unlock(env, obj);
5076                         dtlocked = false;
5077
5078                         dt_trans_stop(env, dev, handle);
5079                         handle = NULL;
5080
5081                         lfsck_ibits_unlock(&lh, LCK_EX);
5082                 }
5083
5084                 /* It may happen when the remote object has been removed,
5085                  * but the local MDT is not aware of that. */
5086                 goto dangling;
5087         } else if (rc == 0) {
5088                 count = ldata.ld_leh->leh_reccount;
5089                 rc = linkea_links_find(&ldata, cname, pfid);
5090                 if ((rc == 0) &&
5091                     (count == 1 || !S_ISDIR(lfsck_object_type(obj)))) {
5092                         if ((lfsck_object_type(obj) & S_IFMT) !=
5093                             lnr->lnr_type) {
5094                                 ns->ln_flags |= LF_INCONSISTENT;
5095                                 type = LNIT_BAD_TYPE;
5096                         }
5097
5098                         goto record;
5099                 }
5100
5101                 ns->ln_flags |= LF_INCONSISTENT;
5102
5103                 /* If the name entry hash does not match the slave striped
5104                  * directory, and the name entry does not match also, then
5105                  * it is quite possible that name entry is corrupted. */
5106                 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5107                                         lnr->lnr_name, lnr->lnr_namelen)) {
5108                         type = LNIT_BAD_DIRENT;
5109
5110                         GOTO(stop, rc = 0);
5111                 }
5112
5113                 /* If the file type stored in the name entry does not match
5114                  * the file type claimed by the object, and the object does
5115                  * not recognize the name entry, then it is quite possible
5116                  * that the name entry is corrupted. */
5117                 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type) {
5118                         type = LNIT_BAD_DIRENT;
5119
5120                         GOTO(stop, rc = 0);
5121                 }
5122
5123                 /* For sub-dir object, we cannot make sure whether the sub-dir
5124                  * back references the parent via ".." name entry correctly or
5125                  * not in the LFSCK first-stage scanning. It may be that the
5126                  * (remote) sub-dir ".." name entry has no parent FID after
5127                  * file-level backup/restore and its linkEA may be wrong.
5128                  * So under such case, we should replace the linkEA according
5129                  * to current name entry. But this needs to be done during the
5130                  * LFSCK second-stage scanning. The LFSCK will record the name
5131                  * entry for further possible using. */
5132                 remove = false;
5133                 newdata = false;
5134                 goto nodata;
5135         } else if (unlikely(rc == -EINVAL)) {
5136                 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5137                         type = LNIT_BAD_TYPE;
5138
5139                 count = 1;
5140                 ns->ln_flags |= LF_INCONSISTENT;
5141                 /* The magic crashed, we are not sure whether there are more
5142                  * corrupt data in the linkea, so remove all linkea entries. */
5143                 remove = true;
5144                 newdata = true;
5145                 goto nodata;
5146         } else if (rc == -ENODATA) {
5147                 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5148                         type = LNIT_BAD_TYPE;
5149
5150                 count = 1;
5151                 ns->ln_flags |= LF_UPGRADE;
5152                 remove = false;
5153                 newdata = true;
5154
5155 nodata:
5156                 if (bk->lb_param & LPF_DRYRUN) {
5157                         ns->ln_linkea_repaired++;
5158                         repaired = true;
5159                         log = true;
5160                         goto record;
5161                 }
5162
5163                 if (!lustre_handle_is_used(&lh))
5164                         goto again;
5165
5166                 if (remove) {
5167                         LASSERT(newdata);
5168
5169                         rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle,
5170                                           BYPASS_CAPA);
5171                         if (rc != 0)
5172                                 GOTO(stop, rc);
5173                 }
5174
5175                 if (newdata) {
5176                         rc = linkea_data_new(&ldata,
5177                                         &lfsck_env_info(env)->lti_linkea_buf);
5178                         if (rc != 0)
5179                                 GOTO(stop, rc);
5180                 }
5181
5182                 rc = linkea_add_buf(&ldata, cname, pfid);
5183                 if (rc != 0)
5184                         GOTO(stop, rc);
5185
5186                 rc = lfsck_links_write(env, obj, &ldata, handle);
5187                 if (unlikely(rc == -ENOSPC) &&
5188                     S_ISREG(lfsck_object_type(obj)) && !dt_object_remote(obj)) {
5189                         if (handle != NULL) {
5190                                 LASSERT(dt_write_locked(env, obj));
5191
5192                                 dt_write_unlock(env, obj);
5193                                 dtlocked = false;
5194
5195                                 dt_trans_stop(env, dev, handle);
5196                                 handle = NULL;
5197
5198                                 lfsck_ibits_unlock(&lh, LCK_EX);
5199                         }
5200
5201                         rc = lfsck_namespace_trace_update(env, com,
5202                                         &lnr->lnr_fid, LNTF_SKIP_NLINK, true);
5203                         if (rc != 0)
5204                                 /* If we cannot record this object in the
5205                                  * LFSCK tracing, we have to mark the LFSCK
5206                                  * as LF_INCOMPLETE, then the LFSCK will
5207                                  * skip nlink attribute verification for
5208                                  * all objects. */
5209                                 ns->ln_flags |= LF_INCOMPLETE;
5210
5211                         GOTO(out, rc = 0);
5212                 }
5213
5214                 if (rc != 0)
5215                         GOTO(stop, rc);
5216
5217                 count = ldata.ld_leh->leh_reccount;
5218                 if (!S_ISDIR(lfsck_object_type(obj)) ||
5219                     !dt_object_remote(obj)) {
5220                         ns->ln_linkea_repaired++;
5221                         repaired = true;
5222                         log = true;
5223                 }
5224         } else {
5225                 GOTO(stop, rc);
5226         }
5227
5228 record:
5229         LASSERT(count > 0);
5230
5231         rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
5232         if (rc != 0)
5233                 GOTO(stop, rc);
5234
5235         if ((count == 1 && la->la_nlink == 1) ||
5236             S_ISDIR(lfsck_object_type(obj)))
5237                 /* Usually, it is for single linked object or dir, do nothing.*/
5238                 GOTO(stop, rc);
5239
5240         /* Following modification will be in another transaction.  */
5241         if (handle != NULL) {
5242                 dt_write_unlock(env, obj);
5243                 dtlocked = false;
5244
5245                 dt_trans_stop(env, dev, handle);
5246                 handle = NULL;
5247
5248                 lfsck_ibits_unlock(&lh, LCK_EX);
5249         }
5250
5251         ns->ln_mul_linked_checked++;
5252         rc = lfsck_namespace_trace_update(env, com, &lnr->lnr_fid,
5253                                           LNTF_CHECK_LINKEA, true);
5254
5255         GOTO(out, rc);
5256
5257 stop:
5258         if (dtlocked)
5259                 dt_write_unlock(env, obj);
5260
5261         if (handle != NULL && !IS_ERR(handle))
5262                 dt_trans_stop(env, dev, handle);
5263
5264 out:
5265         lfsck_ibits_unlock(&lh, LCK_EX);
5266
5267         if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) &&
5268             !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5269                                              lnr->lnr_name, lnr->lnr_namelen) &&
5270             type != LNIT_BAD_DIRENT) {
5271                 ns->ln_flags |= LF_INCONSISTENT;
5272
5273                 log = false;
5274                 rc = lfsck_namespace_repair_bad_name_hash(env, com, dir,
5275                                                 lnr->lnr_lmv, lnr->lnr_name);
5276                 if (rc >= 0)
5277                         bad_hash = true;
5278         }
5279
5280         if (rc >= 0) {
5281                 switch (type) {
5282                 case LNIT_BAD_TYPE:
5283                         log = false;
5284                         rc = lfsck_namespace_repair_dirent(env, com, dir,
5285                                         obj, lnr->lnr_name, lnr->lnr_name,
5286                                         lnr->lnr_type, true, false);
5287                         if (rc > 0)
5288                                 repaired = true;
5289                         break;
5290                 case LNIT_BAD_DIRENT:
5291                         log = false;
5292                         /* XXX: This is a bad dirent, we do not know whether
5293                          *      the original name entry reference a regular
5294                          *      file or a directory, then keep the parent's
5295                          *      nlink count unchanged here. */
5296                         rc = lfsck_namespace_repair_dirent(env, com, dir,
5297                                         obj, lnr->lnr_name, lnr->lnr_name,
5298                                         lnr->lnr_type, false, false);
5299                         if (rc > 0)
5300                                 repaired = true;
5301                         break;
5302                 default:
5303                         break;
5304                 }
5305         }
5306
5307         down_write(&com->lc_sem);
5308         if (rc < 0) {
5309                 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
5310                        "the entry: "DFID", parent "DFID", name %.*s: rc = %d\n",
5311                        lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5312                        PFID(lfsck_dto2fid(lnr->lnr_obj)),
5313                        lnr->lnr_namelen, lnr->lnr_name, rc);
5314
5315                 lfsck_namespace_record_failure(env, lfsck, ns);
5316                 if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
5317                      rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
5318                      rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
5319                     dev != NULL && dev != lfsck->li_next)
5320                         lfsck_lad_set_bitmap(env, com, idx);
5321
5322                 if (!(bk->lb_param & LPF_FAILOUT))
5323                         rc = 0;
5324         } else {
5325                 if (log)
5326                         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant "
5327                                "repaired the entry: "DFID", parent "DFID
5328                                ", name %.*s\n", lfsck_lfsck2name(lfsck),
5329                                PFID(&lnr->lnr_fid),
5330                                PFID(lfsck_dto2fid(lnr->lnr_obj)),
5331                                lnr->lnr_namelen, lnr->lnr_name);
5332
5333                 if (repaired) {
5334                         ns->ln_items_repaired++;
5335
5336                         switch (type) {
5337                         case LNIT_DANGLING:
5338                                 ns->ln_dangling_repaired++;
5339                                 break;
5340                         case LNIT_BAD_TYPE:
5341                                 ns->ln_bad_type_repaired++;
5342                                 break;
5343                         case LNIT_BAD_DIRENT:
5344                                 ns->ln_dirent_repaired++;
5345                                 break;
5346                         default:
5347                                 break;
5348                         }
5349
5350                         if (bk->lb_param & LPF_DRYRUN &&
5351                             lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5352                                 lfsck_pos_fill(env, lfsck,
5353                                                &ns->ln_pos_first_inconsistent,
5354                                                false);
5355                 }
5356
5357                 if (bad_hash) {
5358                         ns->ln_name_hash_repaired++;
5359
5360                         /* Not count repeatedly. */
5361                         if (!repaired)
5362                                 ns->ln_items_repaired++;
5363
5364                         if (bk->lb_param & LPF_DRYRUN &&
5365                             lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5366                                 lfsck_pos_fill(env, lfsck,
5367                                                &ns->ln_pos_first_inconsistent,
5368                                                false);
5369                 }
5370
5371
5372                 rc = 0;
5373         }
5374         up_write(&com->lc_sem);
5375
5376         if (obj != NULL && !IS_ERR(obj))
5377                 lfsck_object_put(env, obj);
5378
5379         return rc;
5380 }
5381
5382 /**
5383  * Handle one orphan under the backend /lost+found directory
5384  *
5385  * Insert the orphan FID into the namespace LFSCK trace file for further
5386  * processing (via the subsequent namespace LFSCK second-stage scanning).
5387  * At the same time, remove the orphan name entry from backend /lost+found
5388  * directory. There is an interval between the orphan name entry removed
5389  * from the backend /lost+found directory and the orphan FID in the LFSCK
5390  * trace file handled. In such interval, the LFSCK can be reset, then
5391  * all the FIDs recorded in the namespace LFSCK trace file will be dropped.
5392  * To guarantee that the orphans can be found when LFSCK run next time
5393  * without e2fsck again, when remove the orphan name entry, the LFSCK
5394  * will set the orphan's ctime attribute as 1. Since normal applications
5395  * cannot change the object's ctime attribute as 1. Then when LFSCK run
5396  * next time, it can record the object (that ctime is 1) in the namespace
5397  * LFSCK trace file during the first-stage scanning.
5398  *
5399  * \param[in] env       pointer to the thread context
5400  * \param[in] com       pointer to the lfsck component
5401  * \param[in] parent    pointer to the object for the backend /lost+found
5402  * \param[in] ent       pointer to the name entry for the target under the
5403  *                      backend /lost+found
5404  *
5405  * \retval              positive for repaired
5406  * \retval              0 if needs to repair nothing
5407  * \retval              negative error number on failure
5408  */
5409 static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
5410                                               struct lfsck_component *com,
5411                                               struct dt_object *parent,
5412                                               struct lu_dirent *ent)
5413 {
5414         struct lfsck_thread_info        *info   = lfsck_env_info(env);
5415         struct lu_fid                   *key    = &info->lti_fid;
5416         struct lu_attr                  *la     = &info->lti_la;
5417         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5418         struct dt_object                *obj    = com->lc_obj;
5419         struct dt_device                *dev    = lfsck->li_bottom;
5420         struct dt_object                *child  = NULL;
5421         struct thandle                  *th     = NULL;
5422         int                              rc     = 0;
5423         __u8                             flags  = 0;
5424         bool                             exist  = false;
5425         ENTRY;
5426
5427         child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid);
5428         if (IS_ERR(child))
5429                 RETURN(PTR_ERR(child));
5430
5431         LASSERT(dt_object_exists(child));
5432         LASSERT(!dt_object_remote(child));
5433
5434         fid_cpu_to_be(key, &ent->lde_fid);
5435         rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
5436                        (const struct dt_key *)key, BYPASS_CAPA);
5437         if (rc == 0) {
5438                 exist = true;
5439                 flags |= LNTF_CHECK_ORPHAN;
5440         } else if (rc == -ENOENT) {
5441                 flags = LNTF_CHECK_ORPHAN;
5442         } else {
5443                 GOTO(out, rc);
5444         }
5445
5446         th = dt_trans_create(env, dev);
5447         if (IS_ERR(th))
5448                 GOTO(out, rc = PTR_ERR(th));
5449
5450         /* a1. remove name entry from backend /lost+found */
5451         rc = dt_declare_delete(env, parent,
5452                                (const struct dt_key *)ent->lde_name, th);
5453         if (rc != 0)
5454                 GOTO(stop, rc);
5455
5456         if (S_ISDIR(lfsck_object_type(child))) {
5457                 /* a2. decrease parent's nlink */
5458                 rc = dt_declare_ref_del(env, parent, th);
5459                 if (rc != 0)
5460                         GOTO(stop, rc);
5461         }
5462
5463         if (exist) {
5464                 /* a3. remove child's FID from the LFSCK trace file. */
5465                 rc = dt_declare_delete(env, obj,
5466                                        (const struct dt_key *)key, th);
5467                 if (rc != 0)
5468                         GOTO(stop, rc);
5469         } else {
5470                 /* a4. set child's ctime as 1 */
5471                 memset(la, 0, sizeof(*la));
5472                 la->la_ctime = 1;
5473                 la->la_valid = LA_CTIME;
5474                 rc = dt_declare_attr_set(env, child, la, th);
5475                 if (rc != 0)
5476                         GOTO(stop, rc);
5477         }
5478
5479         /* a5. insert child's FID into the LFSCK trace file. */
5480         rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
5481                                (const struct dt_key *)key, th);
5482         if (rc != 0)
5483                 GOTO(stop, rc);
5484
5485         rc = dt_trans_start_local(env, dev, th);
5486         if (rc != 0)
5487                 GOTO(stop, rc);
5488
5489         /* b1. remove name entry from backend /lost+found */
5490         rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th,
5491                        BYPASS_CAPA);
5492         if (rc != 0)
5493                 GOTO(stop, rc);
5494
5495         if (S_ISDIR(lfsck_object_type(child))) {
5496                 /* b2. decrease parent's nlink */
5497                 dt_write_lock(env, parent, 0);
5498                 rc = dt_ref_del(env, parent, th);
5499                 dt_write_unlock(env, parent);
5500                 if (rc != 0)
5501                         GOTO(stop, rc);
5502         }
5503
5504         if (exist) {
5505                 /* a3. remove child's FID from the LFSCK trace file. */
5506                 rc = dt_delete(env, obj, (const struct dt_key *)key, th,
5507                                BYPASS_CAPA);
5508                 if (rc != 0)
5509                         GOTO(stop, rc);
5510         } else {
5511                 /* b4. set child's ctime as 1 */
5512                 rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
5513                 if (rc != 0)
5514                         GOTO(stop, rc);
5515         }
5516
5517         /* b5. insert child's FID into the LFSCK trace file. */
5518         rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
5519                        (const struct dt_key *)key, th, BYPASS_CAPA, 1);
5520
5521         GOTO(stop, rc = (rc == 0 ? 1 : rc));
5522
5523 stop:
5524         dt_trans_stop(env, dev, th);
5525
5526 out:
5527         lu_object_put(env, &child->do_lu);
5528
5529         return rc;
5530 }
5531
5532 /**
5533  * Handle orphans under the backend /lost+found directory
5534  *
5535  * Some backend checker, such as e2fsck for ldiskfs may find some orphans
5536  * and put them under the backend /lost+found directory that is invisible
5537  * to client. The LFSCK will scan such directory, for the original client
5538  * visible orphans, add their fids into the namespace LFSCK trace file,
5539  * then the subsenquent namespace LFSCK second-stage scanning can handle
5540  * them as other objects to be double scanned: either move back to normal
5541  * namespace, or to the global visible orphan directory:
5542  * /ROOT/.lustre/lost+found/MDTxxxx/
5543  *
5544  * \param[in] env       pointer to the thread context
5545  * \param[in] com       pointer to the lfsck component
5546  */
5547 static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
5548                                            struct lfsck_component *com)
5549 {
5550         struct lfsck_thread_info        *info   = lfsck_env_info(env);
5551         struct lu_dirent                *ent    =
5552                                         (struct lu_dirent *)info->lti_key;
5553         struct lu_seq_range             *range  = &info->lti_range;
5554         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5555         struct ptlrpc_thread            *thread = &lfsck->li_thread;
5556         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
5557         struct dt_device                *dev    = lfsck->li_bottom;
5558         struct lfsck_namespace          *ns     = com->lc_file_ram;
5559         struct dt_object                *parent;
5560         const struct dt_it_ops          *iops;
5561         struct dt_it                    *di;
5562         struct seq_server_site          *ss     =
5563                                         lu_site2seq(dev->dd_lu_dev.ld_site);
5564         __u64                            cookie;
5565         int                              rc     = 0;
5566         __u16                            type;
5567         ENTRY;
5568
5569         parent = lfsck_object_find_by_dev(env, dev, &LU_BACKEND_LPF_FID);
5570         if (IS_ERR(parent)) {
5571                 CERROR("%s: fail to find backend /lost+found: rc = %ld\n",
5572                        lfsck_lfsck2name(lfsck), PTR_ERR(parent));
5573                 RETURN_EXIT;
5574         }
5575
5576         /* It is normal that the /lost+found does not exist for ZFS backend. */
5577         if (!dt_object_exists(parent))
5578                 GOTO(out, rc = 0);
5579
5580         if (unlikely(!dt_try_as_dir(env, parent)))
5581                 GOTO(out, rc = -ENOTDIR);
5582
5583         CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n",
5584                lfsck_lfsck2name(lfsck));
5585
5586         com->lc_new_scanned = 0;
5587         iops = &parent->do_index_ops->dio_it;
5588         di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE, BYPASS_CAPA);
5589         if (IS_ERR(di))
5590                 GOTO(out, rc = PTR_ERR(di));
5591
5592         rc = iops->load(env, di, 0);
5593         if (rc == 0)
5594                 rc = iops->next(env, di);
5595         else if (rc > 0)
5596                 rc = 0;
5597
5598         while (rc == 0) {
5599                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY3) &&
5600                     cfs_fail_val > 0) {
5601                         struct l_wait_info lwi;
5602
5603                         lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
5604                                           NULL, NULL);
5605                         l_wait_event(thread->t_ctl_waitq,
5606                                      !thread_is_running(thread),
5607                                      &lwi);
5608
5609                         if (unlikely(!thread_is_running(thread)))
5610                                 break;
5611                 }
5612
5613                 rc = iops->rec(env, di, (struct dt_rec *)ent,
5614                                LUDA_64BITHASH | LUDA_TYPE);
5615                 if (rc == 0)
5616                         rc = lfsck_unpack_ent(ent, &cookie, &type);
5617
5618                 if (unlikely(rc != 0)) {
5619                         CDEBUG(D_LFSCK, "%s: fail to iterate backend "
5620                                "/lost+found: rc = %d\n",
5621                                lfsck_lfsck2name(lfsck), rc);
5622
5623                         goto skip;
5624                 }
5625
5626                 /* skip dot and dotdot entries */
5627                 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
5628                         goto next;
5629
5630                 if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid)))
5631                         goto skip;
5632
5633                 if (fid_is_norm(&ent->lde_fid)) {
5634                         fld_range_set_mdt(range);
5635                         rc = fld_local_lookup(env, ss->ss_server_fld,
5636                                               fid_seq(&ent->lde_fid), range);
5637                         if (rc != 0)
5638                                 goto skip;
5639                 } else if (lfsck_dev_idx(dev) != 0) {
5640                         /* If the returned FID is IGIF, then there are three
5641                          * possible cases:
5642                          *
5643                          * 1) The object is upgraded from old Lustre-1.8 with
5644                          *    IGIF assigned to such object.
5645                          * 2) The object is a backend local object and is
5646                          *    invisible to client.
5647                          * 3) The object lost its LMV EA, and since there is
5648                          *    no FID-in-dirent for the orphan in the backend
5649                          *    /lost+found directory, then the low layer will
5650                          *    return IGIF for such object.
5651                          *
5652                          * For MDTx (x != 0), it is either case 2) or case 3),
5653                          * but from the LFSCK view, they are indistinguishable.
5654                          * To be safe, the LFSCK will keep it there and report
5655                          * some message, then the adminstrator can handle that
5656                          * furtherly.
5657                          *
5658                          * For MDT0, it is more possible the case 1). The LFSCK
5659                          * will handle the orphan as an upgraded object. */
5660                         CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID
5661                                "in the backend /lost+found on the MDT %04x, "
5662                                "to be safe, skip it.\n",
5663                                lfsck_lfsck2name(lfsck), ent->lde_namelen,
5664                                ent->lde_name, PFID(&ent->lde_fid),
5665                                lfsck_dev_idx(dev));
5666                         goto skip;
5667                 }
5668
5669                 rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent);
5670
5671 skip:
5672                 down_write(&com->lc_sem);
5673                 com->lc_new_scanned++;
5674                 ns->ln_local_lpf_scanned++;
5675                 if (rc > 0)
5676                         ns->ln_local_lpf_moved++;
5677                 else if (rc == 0)
5678                         ns->ln_local_lpf_skipped++;
5679                 else
5680                         ns->ln_local_lpf_failed++;
5681                 up_write(&com->lc_sem);
5682
5683                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5684                         break;
5685
5686 next:
5687                 lfsck_control_speed_by_self(com);
5688                 if (unlikely(!thread_is_running(thread))) {
5689                         rc = 0;
5690                         break;
5691                 }
5692
5693                 rc = iops->next(env, di);
5694         }
5695
5696         iops->put(env, di);
5697         iops->fini(env, di);
5698
5699         EXIT;
5700
5701 out:
5702         CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n",
5703                lfsck_lfsck2name(lfsck), rc);
5704
5705         lu_object_put(env, &parent->do_lu);
5706 }
5707
5708 /**
5709  * Rescan the striped directory after the master LMV EA reset.
5710  *
5711  * Sometimes, the master LMV EA of the striped directory maybe lost, so when
5712  * the namespace LFSCK engine scan the striped directory for the first time,
5713  * it will be reguarded as a normal directory. As the LFSCK processing, some
5714  * other LFSCK instance on other MDT will find the shard of this striped dir,
5715  * and find that the master MDT-object of the striped directory lost its LMV
5716  * EA, then such remote LFSCK instance will regenerate the master LMV EA and
5717  * notify the LFSCK instance on this MDT to rescan the striped directory.
5718  *
5719  * \param[in] env       pointer to the thread context
5720  * \param[in] com       pointer to the lfsck component
5721  * \param[in] llu       the lfsck_lmv_unit that contains the striped directory
5722  *                      to be rescanned.
5723  *
5724  * \retval              positive number for success
5725  * \retval              0 for LFSCK stopped/paused
5726  * \retval              negative error number on failure
5727  */
5728 static int lfsck_namespace_rescan_striped_dir(const struct lu_env *env,
5729                                               struct lfsck_component *com,
5730                                               struct lfsck_lmv_unit *llu)
5731 {
5732         struct lfsck_thread_info        *info   = lfsck_env_info(env);
5733         struct lfsck_instance           *lfsck  = com->lc_lfsck;
5734         struct lfsck_assistant_data     *lad    = com->lc_data;
5735         struct dt_object                *dir;
5736         const struct dt_it_ops          *iops;
5737         struct dt_it                    *di;
5738         struct lu_dirent                *ent    =
5739                         (struct lu_dirent *)info->lti_key;
5740         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
5741         struct ptlrpc_thread            *thread = &lfsck->li_thread;
5742         struct lfsck_namespace_req      *lnr;
5743         struct lfsck_assistant_req      *lar;
5744         int                              rc;
5745         __u16                            type;
5746         ENTRY;
5747
5748         LASSERT(list_empty(&lad->lad_req_list));
5749
5750         lfsck->li_lmv = &llu->llu_lmv;
5751         lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
5752         rc = lfsck_open_dir(env, lfsck, 0);
5753         if (rc != 0)
5754                 RETURN(rc);
5755
5756         dir = lfsck->li_obj_dir;
5757         di = lfsck->li_di_dir;
5758         iops = &dir->do_index_ops->dio_it;
5759         do {
5760                 rc = iops->rec(env, di, (struct dt_rec *)ent,
5761                                lfsck->li_args_dir);
5762                 if (rc == 0)
5763                         rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
5764                                               &type);
5765
5766                 if (rc != 0) {
5767                         if (bk->lb_param & LPF_FAILOUT)
5768                                 GOTO(out, rc);
5769
5770                         goto next;
5771                 }
5772
5773                 if (ent->lde_attrs & LUDA_IGNORE &&
5774                     strcmp(ent->lde_name, dotdot) != 0)
5775                         goto next;
5776
5777                 lnr = lfsck_namespace_assistant_req_init(lfsck, ent, type);
5778                 if (IS_ERR(lnr)) {
5779                         if (bk->lb_param & LPF_FAILOUT)
5780                                 GOTO(out, rc = PTR_ERR(lnr));
5781
5782                         goto next;
5783                 }
5784
5785                 lar = &lnr->lnr_lar;
5786                 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
5787                 lfsck_namespace_assistant_req_fini(env, lar);
5788                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
5789                         GOTO(out, rc);
5790
5791                 if (unlikely(!thread_is_running(thread)))
5792                         GOTO(out, rc = 0);
5793
5794 next:
5795                 rc = iops->next(env, di);
5796         } while (rc == 0);
5797
5798 out:
5799         lfsck_close_dir(env, lfsck, rc);
5800         if (rc <= 0)
5801                 RETURN(rc);
5802
5803         /* The close_dir() may insert a dummy lnr in the lad->lad_req_list. */
5804         if (list_empty(&lad->lad_req_list))
5805                 RETURN(1);
5806
5807         spin_lock(&lad->lad_lock);
5808         lar = list_entry(lad->lad_req_list.next, struct lfsck_assistant_req,
5809                           lar_list);
5810         list_del_init(&lar->lar_list);
5811         spin_unlock(&lad->lad_lock);
5812
5813         rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
5814         lfsck_namespace_assistant_req_fini(env, lar);
5815
5816         RETURN(rc == 0 ? 1 : rc);
5817 }
5818
5819 static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
5820                                                 struct lfsck_component *com)
5821 {
5822         struct lfsck_instance   *lfsck  = com->lc_lfsck;
5823         struct ptlrpc_thread    *thread = &lfsck->li_thread;
5824         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
5825         struct lfsck_namespace  *ns     = com->lc_file_ram;
5826         struct dt_object        *obj    = com->lc_obj;
5827         const struct dt_it_ops  *iops   = &obj->do_index_ops->dio_it;
5828         struct dt_object        *target;
5829         struct dt_it            *di;
5830         struct dt_key           *key;
5831         struct lu_fid            fid;
5832         int                      rc;
5833         __u8                     flags  = 0;
5834         ENTRY;
5835
5836         while (!list_empty(&lfsck->li_list_lmv)) {
5837                 struct lfsck_lmv_unit *llu;
5838
5839                 spin_lock(&lfsck->li_lock);
5840                 llu = list_entry(lfsck->li_list_lmv.next,
5841                                  struct lfsck_lmv_unit, llu_link);
5842                 list_del_init(&llu->llu_link);
5843                 spin_unlock(&lfsck->li_lock);
5844
5845                 rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
5846                 if (rc <= 0)
5847                         RETURN(rc);
5848         }
5849
5850         CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
5851                lfsck_lfsck2name(lfsck));
5852
5853         lfsck_namespace_scan_local_lpf(env, com);
5854
5855         com->lc_new_checked = 0;
5856         com->lc_new_scanned = 0;
5857         com->lc_time_last_checkpoint = cfs_time_current();
5858         com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
5859                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
5860
5861         di = iops->init(env, obj, 0, BYPASS_CAPA);
5862         if (IS_ERR(di))
5863                 RETURN(PTR_ERR(di));
5864
5865         fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
5866         rc = iops->get(env, di, (const struct dt_key *)&fid);
5867         if (rc < 0)
5868                 GOTO(fini, rc);
5869
5870         /* Skip the start one, which either has been processed or non-exist. */
5871         rc = iops->next(env, di);
5872         if (rc != 0)
5873                 GOTO(put, rc);
5874
5875         do {
5876                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY3) &&
5877                     cfs_fail_val > 0) {
5878                         struct l_wait_info lwi;
5879
5880                         lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
5881                                           NULL, NULL);
5882                         l_wait_event(thread->t_ctl_waitq,
5883                                      !thread_is_running(thread),
5884                                      &lwi);
5885
5886                         if (unlikely(!thread_is_running(thread)))
5887                                 GOTO(put, rc = 0);
5888                 }
5889
5890                 key = iops->key(env, di);
5891                 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
5892                 if (!fid_is_sane(&fid)) {
5893                         rc = 0;
5894                         goto checkpoint;
5895                 }
5896
5897                 target = lfsck_object_find_by_dev(env, lfsck->li_bottom, &fid);
5898                 if (IS_ERR(target)) {
5899                         rc = PTR_ERR(target);
5900                         goto checkpoint;
5901                 }
5902
5903                 if (dt_object_exists(target)) {
5904                         rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
5905                         if (rc == 0) {
5906                                 rc = lfsck_namespace_double_scan_one(env, com,
5907                                                                 target, flags);
5908                                 if (rc == -ENOENT)
5909                                         rc = 0;
5910                         }
5911                 }
5912
5913                 lfsck_object_put(env, target);
5914
5915 checkpoint:
5916                 down_write(&com->lc_sem);
5917                 com->lc_new_checked++;
5918                 com->lc_new_scanned++;
5919                 ns->ln_fid_latest_scanned_phase2 = fid;
5920                 if (rc > 0)
5921                         ns->ln_objs_repaired_phase2++;
5922                 else if (rc < 0)
5923                         ns->ln_objs_failed_phase2++;
5924                 up_write(&com->lc_sem);
5925
5926                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5927                         GOTO(put, rc);
5928
5929                 if (unlikely(cfs_time_beforeq(com->lc_time_next_checkpoint,
5930                                               cfs_time_current())) &&
5931                     com->lc_new_checked != 0) {
5932                         down_write(&com->lc_sem);
5933                         ns->ln_run_time_phase2 +=
5934                                 cfs_duration_sec(cfs_time_current() +
5935                                 HALF_SEC - com->lc_time_last_checkpoint);
5936                         ns->ln_time_last_checkpoint = cfs_time_current_sec();
5937                         ns->ln_objs_checked_phase2 += com->lc_new_checked;
5938                         com->lc_new_checked = 0;
5939                         rc = lfsck_namespace_store(env, com);
5940                         up_write(&com->lc_sem);
5941                         if (rc != 0)
5942                                 GOTO(put, rc);
5943
5944                         com->lc_time_last_checkpoint = cfs_time_current();
5945                         com->lc_time_next_checkpoint =
5946                                 com->lc_time_last_checkpoint +
5947                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
5948                 }
5949
5950                 lfsck_control_speed_by_self(com);
5951                 if (unlikely(!thread_is_running(thread)))
5952                         GOTO(put, rc = 0);
5953
5954                 rc = iops->next(env, di);
5955         } while (rc == 0);
5956
5957         GOTO(put, rc);
5958
5959 put:
5960         iops->put(env, di);
5961
5962 fini:
5963         iops->fini(env, di);
5964
5965         CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop: rc = %d\n",
5966                lfsck_lfsck2name(lfsck), rc);
5967
5968         return rc;
5969 }
5970
5971 static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env,
5972                                                struct lfsck_component *com,
5973                                                struct lfsck_position *pos)
5974 {
5975         struct lfsck_assistant_data     *lad = com->lc_data;
5976         struct lfsck_namespace_req      *lnr;
5977
5978         if (list_empty(&lad->lad_req_list))
5979                 return;
5980
5981         lnr = list_entry(lad->lad_req_list.next,
5982                          struct lfsck_namespace_req,
5983                          lnr_lar.lar_list);
5984         pos->lp_oit_cookie = lnr->lnr_oit_cookie;
5985         pos->lp_dir_cookie = lnr->lnr_dir_cookie - 1;
5986         pos->lp_dir_parent = *lfsck_dto2fid(lnr->lnr_obj);
5987 }
5988
5989 static int lfsck_namespace_double_scan_result(const struct lu_env *env,
5990                                               struct lfsck_component *com,
5991                                               int rc)
5992 {
5993         struct lfsck_instance   *lfsck  = com->lc_lfsck;
5994         struct lfsck_namespace  *ns     = com->lc_file_ram;
5995
5996         down_write(&com->lc_sem);
5997         ns->ln_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
5998                                 HALF_SEC - lfsck->li_time_last_checkpoint);
5999         ns->ln_time_last_checkpoint = cfs_time_current_sec();
6000         ns->ln_objs_checked_phase2 += com->lc_new_checked;
6001         com->lc_new_checked = 0;
6002
6003         if (rc > 0) {
6004                 if (ns->ln_flags & LF_INCOMPLETE)
6005                         ns->ln_status = LS_PARTIAL;
6006                 else
6007                         ns->ln_status = LS_COMPLETED;
6008                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
6009                         ns->ln_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
6010                 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
6011                 ns->ln_success_count++;
6012         } else if (rc == 0) {
6013                 if (lfsck->li_status != 0)
6014                         ns->ln_status = lfsck->li_status;
6015                 else
6016                         ns->ln_status = LS_STOPPED;
6017         } else {
6018                 ns->ln_status = LS_FAILED;
6019         }
6020
6021         rc = lfsck_namespace_store(env, com);
6022         up_write(&com->lc_sem);
6023
6024         return rc;
6025 }
6026
6027 static int
6028 lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
6029                                                   struct ptlrpc_request *req,
6030                                                   void *args, int rc)
6031 {
6032         return 0;
6033 }
6034
6035 /**
6036  * Notify remote LFSCK instances about former failures.
6037  *
6038  * The local LFSCK instance has recorded which MDTs have ever failed to respond
6039  * some LFSCK verification requests (maybe because of network issues or the MDT
6040  * itself trouble). During the respond gap the MDT may missed some name entries
6041  * verification, then the MDT cannot know whether related MDT-objects have been
6042  * referenced by related name entries or not, then in the second-stage scanning,
6043  * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
6044  * linkEA for back reference, then it will misguide the LFSCK to generate wrong
6045  * name entry for repairing the orphan.
6046  *
6047  * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
6048  * it will scan the bitmap for the ever failed MDTs, and notify them that they
6049  * have ever missed some name entries verification and should skip the handling
6050  * for orphan MDT-objects.
6051  *
6052  * \param[in] env       pointer to the thread context
6053  * \param[in] com       pointer to the lfsck component
6054  * \param[in] lr        pointer to the lfsck request
6055  */
6056 static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
6057                                                     struct lfsck_component *com,
6058                                                     struct lfsck_request *lr)
6059 {
6060         struct lfsck_async_interpret_args *laia  =
6061                                 &lfsck_env_info(env)->lti_laia2;
6062         struct lfsck_assistant_data       *lad   = com->lc_data;
6063         struct lfsck_namespace            *ns    = com->lc_file_ram;
6064         struct lfsck_instance             *lfsck = com->lc_lfsck;
6065         struct lfsck_tgt_descs            *ltds  = &lfsck->li_mdt_descs;
6066         struct lfsck_tgt_desc             *ltd;
6067         struct ptlrpc_request_set         *set;
6068         int                                rc    = 0;
6069         ENTRY;
6070
6071         set = ptlrpc_prep_set();
6072         if (set == NULL)
6073                 GOTO(out, rc = -ENOMEM);
6074
6075         lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
6076         memset(laia, 0, sizeof(*laia));
6077         lad->lad_touch_gen++;
6078
6079         spin_lock(&ltds->ltd_lock);
6080         while (!list_empty(&lad->lad_mdt_list)) {
6081                 ltd = list_entry(lad->lad_mdt_list.next,
6082                                  struct lfsck_tgt_desc,
6083                                  ltd_namespace_list);
6084                 if (ltd->ltd_namespace_gen == lad->lad_touch_gen)
6085                         break;
6086
6087                 ltd->ltd_namespace_gen = lad->lad_touch_gen;
6088                 list_move_tail(&ltd->ltd_namespace_list,
6089                                &lad->lad_mdt_list);
6090                 if (!lad->lad_incomplete ||
6091                     !cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
6092                         ltd->ltd_namespace_failed = 0;
6093                         continue;
6094                 }
6095
6096                 ltd->ltd_namespace_failed = 1;
6097                 spin_unlock(&ltds->ltd_lock);
6098                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
6099                         lfsck_namespace_assistant_sync_failures_interpret,
6100                         laia, LFSCK_NOTIFY);
6101                 if (rc != 0)
6102                         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6103                                "to sync failure with MDT %x: rc = %d\n",
6104                                lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
6105
6106                 spin_lock(&ltds->ltd_lock);
6107         }
6108         spin_unlock(&ltds->ltd_lock);
6109
6110         rc = ptlrpc_set_wait(set);
6111         ptlrpc_set_destroy(set);
6112
6113         GOTO(out, rc);
6114
6115 out:
6116         if (rc != 0)
6117                 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6118                        "to sync failure with MDTs, and related MDTs "
6119                        "may handle orphan improperly: rc = %d\n",
6120                        lfsck_lfsck2name(lfsck), rc);
6121
6122         EXIT;
6123 }
6124
6125 struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
6126         .la_handler_p1          = lfsck_namespace_assistant_handler_p1,
6127         .la_handler_p2          = lfsck_namespace_assistant_handler_p2,
6128         .la_fill_pos            = lfsck_namespace_assistant_fill_pos,
6129         .la_double_scan_result  = lfsck_namespace_double_scan_result,
6130         .la_req_fini            = lfsck_namespace_assistant_req_fini,
6131         .la_sync_failures       = lfsck_namespace_assistant_sync_failures,
6132 };
6133
6134 /**
6135  * Verify the specified linkEA entry for the given directory object.
6136  * If the object has no such linkEA entry or it has more other linkEA
6137  * entries, then re-generate the linkEA with the given information.
6138  *
6139  * \param[in] env       pointer to the thread context
6140  * \param[in] dev       pointer to the dt_device
6141  * \param[in] obj       pointer to the dt_object to be handled
6142  * \param[in] cname     the name for the child in the parent directory
6143  * \param[in] pfid      the parent directory's FID for the linkEA
6144  *
6145  * \retval              0 for success
6146  * \retval              negative error number on failure
6147  */
6148 int lfsck_verify_linkea(const struct lu_env *env, struct dt_device *dev,
6149                         struct dt_object *obj, const struct lu_name *cname,
6150                         const struct lu_fid *pfid)
6151 {
6152         struct linkea_data       ldata  = { 0 };
6153         struct lu_buf            linkea_buf;
6154         struct thandle          *th;
6155         int                      rc;
6156         int                      fl     = LU_XATTR_CREATE;
6157         bool                     dirty  = false;
6158         ENTRY;
6159
6160         LASSERT(S_ISDIR(lfsck_object_type(obj)));
6161
6162         rc = lfsck_links_read(env, obj, &ldata);
6163         if (rc == -ENODATA) {
6164                 dirty = true;
6165         } else if (rc == 0) {
6166                 fl = LU_XATTR_REPLACE;
6167                 if (ldata.ld_leh->leh_reccount != 1) {
6168                         dirty = true;
6169                 } else {
6170                         rc = linkea_links_find(&ldata, cname, pfid);
6171                         if (rc != 0)
6172                                 dirty = true;
6173                 }
6174         }
6175
6176         if (!dirty)
6177                 RETURN(rc);
6178
6179         rc = linkea_data_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf);
6180         if (rc != 0)
6181                 RETURN(rc);
6182
6183         rc = linkea_add_buf(&ldata, cname, pfid);
6184         if (rc != 0)
6185                 RETURN(rc);
6186
6187         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
6188                        ldata.ld_leh->leh_len);
6189         th = dt_trans_create(env, dev);
6190         if (IS_ERR(th))
6191                 RETURN(PTR_ERR(th));
6192
6193         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
6194                                   XATTR_NAME_LINK, fl, th);
6195         if (rc != 0)
6196                 GOTO(stop, rc);
6197
6198         rc = dt_trans_start_local(env, dev, th);
6199         if (rc != 0)
6200                 GOTO(stop, rc);
6201
6202         dt_write_lock(env, obj, 0);
6203         rc = dt_xattr_set(env, obj, &linkea_buf,
6204                           XATTR_NAME_LINK, fl, th, BYPASS_CAPA);
6205         dt_write_unlock(env, obj);
6206
6207         GOTO(stop, rc);
6208
6209 stop:
6210         dt_trans_stop(env, dev, th);
6211         return rc;
6212 }
6213
6214 /**
6215  * Get the name and parent directory's FID from the first linkEA entry.
6216  *
6217  * \param[in] env       pointer to the thread context
6218  * \param[in] obj       pointer to the object which get linkEA from
6219  * \param[out] name     pointer to the buffer to hold the name
6220  *                      in the first linkEA entry
6221  * \param[out] pfid     pointer to the buffer to hold the parent
6222  *                      directory's FID in the first linkEA entry
6223  *
6224  * \retval              0 for success
6225  * \retval              negative error number on failure
6226  */
6227 int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
6228                           char *name, struct lu_fid *pfid)
6229 {
6230         struct lu_name           *cname = &lfsck_env_info(env)->lti_name;
6231         struct linkea_data        ldata = { 0 };
6232         int                       rc;
6233
6234         rc = lfsck_links_read(env, obj, &ldata);
6235         if (rc != 0)
6236                 return rc;
6237
6238         linkea_first_entry(&ldata);
6239         if (ldata.ld_lee == NULL)
6240                 return -ENODATA;
6241
6242         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
6243         /* To guarantee the 'name' is terminated with '0'. */
6244         memcpy(name, cname->ln_name, cname->ln_namelen);
6245         name[cname->ln_namelen] = 0;
6246
6247         return 0;
6248 }
6249
6250 /**
6251  * Remove the name entry from the parent directory.
6252  *
6253  * No need to care about the object referenced by the name entry,
6254  * either the name entry is invalid or redundant, or the referenced
6255  * object has been processed has been or will be handled by others.
6256  *
6257  * \param[in] env       pointer to the thread context
6258  * \param[in] lfsck     pointer to the lfsck instance
6259  * \param[in] parent    pointer to the lost+found object
6260  * \param[in] name      the name for the name entry to be removed
6261  * \param[in] type      the type for the name entry to be removed
6262  *
6263  * \retval              0 for success
6264  * \retval              negative error number on failure
6265  */
6266 int lfsck_remove_name_entry(const struct lu_env *env,
6267                             struct lfsck_instance *lfsck,
6268                             struct dt_object *parent,
6269                             const char *name, __u32 type)
6270 {
6271         struct dt_device        *dev    = lfsck->li_next;
6272         struct thandle          *th;
6273         struct lustre_handle     lh     = { 0 };
6274         int                      rc;
6275         ENTRY;
6276
6277         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
6278                               MDS_INODELOCK_UPDATE, LCK_EX);
6279         if (rc != 0)
6280                 RETURN(rc);
6281
6282         th = dt_trans_create(env, dev);
6283         if (IS_ERR(th))
6284                 GOTO(unlock, rc = PTR_ERR(th));
6285
6286         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
6287         if (rc != 0)
6288                 GOTO(stop, rc);
6289
6290         if (S_ISDIR(type)) {
6291                 rc = dt_declare_ref_del(env, parent, th);
6292                 if (rc != 0)
6293                         GOTO(stop, rc);
6294         }
6295
6296         rc = dt_trans_start(env, dev, th);
6297         if (rc != 0)
6298                 GOTO(stop, rc);
6299
6300         rc = dt_delete(env, parent, (const struct dt_key *)name, th,
6301                        BYPASS_CAPA);
6302         if (rc != 0)
6303                 GOTO(stop, rc);
6304
6305         if (S_ISDIR(type)) {
6306                 dt_write_lock(env, parent, 0);
6307                 rc = dt_ref_del(env, parent, th);
6308                 dt_write_unlock(env, parent);
6309         }
6310
6311         GOTO(stop, rc);
6312
6313 stop:
6314         dt_trans_stop(env, dev, th);
6315
6316 unlock:
6317         lfsck_ibits_unlock(&lh, LCK_EX);
6318
6319         CDEBUG(D_LFSCK, "%s: remove name entry "DFID"/%s "
6320                "with type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
6321                PFID(lfsck_dto2fid(parent)), name, type, rc);
6322
6323         return rc;
6324 }
6325
6326 /**
6327  * Update the object's name entry with the given FID.
6328  *
6329  * \param[in] env       pointer to the thread context
6330  * \param[in] lfsck     pointer to the lfsck instance
6331  * \param[in] parent    pointer to the parent directory that holds
6332  *                      the name entry
6333  * \param[in] name      the name for the entry to be updated
6334  * \param[in] pfid      the new PFID for the name entry
6335  * \param[in] type      the type for the name entry to be updated
6336  *
6337  * \retval              0 for success
6338  * \retval              negative error number on failure
6339  */
6340 int lfsck_update_name_entry(const struct lu_env *env,
6341                             struct lfsck_instance *lfsck,
6342                             struct dt_object *parent, const char *name,
6343                             const struct lu_fid *pfid, __u32 type)
6344 {
6345         struct dt_insert_rec    *rec    = &lfsck_env_info(env)->lti_dt_rec;
6346         struct dt_device        *dev    = lfsck->li_next;
6347         struct lustre_handle     lh     = { 0 };
6348         struct thandle          *th;
6349         int                      rc;
6350         bool                     exists = true;
6351         ENTRY;
6352
6353         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
6354                               MDS_INODELOCK_UPDATE, LCK_EX);
6355         if (rc != 0)
6356                 RETURN(rc);
6357
6358         th = dt_trans_create(env, dev);
6359         if (IS_ERR(th))
6360                 GOTO(unlock, rc = PTR_ERR(th));
6361
6362         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
6363         if (rc != 0)
6364                 GOTO(stop, rc);
6365
6366         rec->rec_type = type;
6367         rec->rec_fid = pfid;
6368         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
6369                                (const struct dt_key *)name, th);
6370         if (rc != 0)
6371                 GOTO(stop, rc);
6372
6373         rc = dt_declare_ref_add(env, parent, th);
6374         if (rc != 0)
6375                 GOTO(stop, rc);
6376
6377         rc = dt_trans_start(env, dev, th);
6378         if (rc != 0)
6379                 GOTO(stop, rc);
6380
6381         rc = dt_delete(env, parent, (const struct dt_key *)name, th,
6382                        BYPASS_CAPA);
6383         if (rc == -ENOENT) {
6384                 exists = false;
6385                 rc = 0;
6386         }
6387
6388         if (rc != 0)
6389                 GOTO(stop, rc);
6390
6391         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
6392                        (const struct dt_key *)name, th, BYPASS_CAPA, 1);
6393         if (rc == 0 && S_ISDIR(type) && !exists) {
6394                 dt_write_lock(env, parent, 0);
6395                 rc = dt_ref_add(env, parent, th);
6396                 dt_write_unlock(env, parent);
6397         }
6398
6399         GOTO(stop, rc);
6400
6401 stop:
6402         dt_trans_stop(env, dev, th);
6403
6404 unlock:
6405         lfsck_ibits_unlock(&lh, LCK_EX);
6406
6407         CDEBUG(D_LFSCK, "%s: update name entry "DFID"/%s with the FID "DFID
6408                " and the type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
6409                PFID(lfsck_dto2fid(parent)), name, PFID(pfid), type, rc);
6410
6411         return rc;
6412 }
6413
6414 int lfsck_namespace_setup(const struct lu_env *env,
6415                           struct lfsck_instance *lfsck)
6416 {
6417         struct lfsck_component  *com;
6418         struct lfsck_namespace  *ns;
6419         struct dt_object        *root = NULL;
6420         struct dt_object        *obj;
6421         int                      rc;
6422         ENTRY;
6423
6424         LASSERT(lfsck->li_master);
6425
6426         OBD_ALLOC_PTR(com);
6427         if (com == NULL)
6428                 RETURN(-ENOMEM);
6429
6430         INIT_LIST_HEAD(&com->lc_link);
6431         INIT_LIST_HEAD(&com->lc_link_dir);
6432         init_rwsem(&com->lc_sem);
6433         atomic_set(&com->lc_ref, 1);
6434         com->lc_lfsck = lfsck;
6435         com->lc_type = LFSCK_TYPE_NAMESPACE;
6436         com->lc_ops = &lfsck_namespace_ops;
6437         com->lc_data = lfsck_assistant_data_init(
6438                         &lfsck_namespace_assistant_ops,
6439                         "lfsck_namespace");
6440         if (com->lc_data == NULL)
6441                 GOTO(out, rc = -ENOMEM);
6442
6443         com->lc_file_size = sizeof(struct lfsck_namespace);
6444         OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6445         if (com->lc_file_ram == NULL)
6446                 GOTO(out, rc = -ENOMEM);
6447
6448         OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6449         if (com->lc_file_disk == NULL)
6450                 GOTO(out, rc = -ENOMEM);
6451
6452         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6453         if (IS_ERR(root))
6454                 GOTO(out, rc = PTR_ERR(root));
6455
6456         if (unlikely(!dt_try_as_dir(env, root)))
6457                 GOTO(out, rc = -ENOTDIR);
6458
6459         obj = local_index_find_or_create(env, lfsck->li_los, root,
6460                                          lfsck_namespace_name,
6461                                          S_IFREG | S_IRUGO | S_IWUSR,
6462                                          &dt_lfsck_features);
6463         if (IS_ERR(obj))
6464                 GOTO(out, rc = PTR_ERR(obj));
6465
6466         com->lc_obj = obj;
6467         rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_features);
6468         if (rc != 0)
6469                 GOTO(out, rc);
6470
6471         rc = lfsck_namespace_load(env, com);
6472         if (rc > 0)
6473                 rc = lfsck_namespace_reset(env, com, true);
6474         else if (rc == -ENODATA)
6475                 rc = lfsck_namespace_init(env, com);
6476         if (rc != 0)
6477                 GOTO(out, rc);
6478
6479         ns = com->lc_file_ram;
6480         switch (ns->ln_status) {
6481         case LS_INIT:
6482         case LS_COMPLETED:
6483         case LS_FAILED:
6484         case LS_STOPPED:
6485                 spin_lock(&lfsck->li_lock);
6486                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6487                 spin_unlock(&lfsck->li_lock);
6488                 break;
6489         default:
6490                 CERROR("%s: unknown lfsck_namespace status %d\n",
6491                        lfsck_lfsck2name(lfsck), ns->ln_status);
6492                 /* fall through */
6493         case LS_SCANNING_PHASE1:
6494         case LS_SCANNING_PHASE2:
6495                 /* No need to store the status to disk right now.
6496                  * If the system crashed before the status stored,
6497                  * it will be loaded back when next time. */
6498                 ns->ln_status = LS_CRASHED;
6499                 /* fall through */
6500         case LS_PAUSED:
6501         case LS_CRASHED:
6502                 spin_lock(&lfsck->li_lock);
6503                 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6504                 list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
6505                 spin_unlock(&lfsck->li_lock);
6506                 break;
6507         }
6508
6509         GOTO(out, rc = 0);
6510
6511 out:
6512         if (root != NULL && !IS_ERR(root))
6513                 lu_object_put(env, &root->do_lu);
6514         if (rc != 0) {
6515                 lfsck_component_cleanup(env, com);
6516                 CERROR("%s: fail to init namespace LFSCK component: rc = %d\n",
6517                        lfsck_lfsck2name(lfsck), rc);
6518         }
6519         return rc;
6520 }