lustre/lfsck/lfsck_namespace.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2013, 2017, Intel Corporation.
  24  */
  25 /*
  26  * lustre/lfsck/lfsck_namespace.c
  27  *
  28  * Author: Fan, Yong <fan.yong@intel.com>
  29  */
  30
  31 #define DEBUG_SUBSYSTEM S_LFSCK
  32
  33 #include <lu_object.h>
  34 #include <dt_object.h>
  35 #include <md_object.h>
  36 #include <lustre_fid.h>
  37 #include <lustre_lib.h>
  38 #include <lustre_net.h>
  39
  40 #include "lfsck_internal.h"
  41
  42 #define LFSCK_NAMESPACE_MAGIC_V1        0xA0629D03
  43 #define LFSCK_NAMESPACE_MAGIC_V2        0xA0621A0B
  44 #define LFSCK_NAMESPACE_MAGIC_V3        0xA06249FF
  45
  46 /* For Lustre-2.x (x <= 6), the namespace LFSCK used LFSCK_NAMESPACE_MAGIC_V1
  47  * as the trace file magic. When downgrade to such old release, the old LFSCK
  48  * will not recognize the new LFSCK_NAMESPACE_MAGIC_V2 in the new trace file,
  49  * then it will reset the whole LFSCK, and will not cause start failure. The
  50  * similar case will happen when upgrade from such old release. */
  51 #define LFSCK_NAMESPACE_MAGIC           LFSCK_NAMESPACE_MAGIC_V3
  52
  53 enum lfsck_nameentry_check {
  54         LFSCK_NAMEENTRY_DEAD            = 1, /* The object has been unlinked. */
  55         LFSCK_NAMEENTRY_REMOVED         = 2, /* The entry has been removed. */
  56         LFSCK_NAMEENTRY_RECREATED       = 3, /* The entry has been recreated. */
  57 };
  58
  59 static struct lfsck_namespace_req *
  60 lfsck_namespace_assistant_req_init(struct lfsck_instance *lfsck,
  61                                    struct lfsck_assistant_object *lso,
  62                                    struct lu_dirent *ent, __u16 type)
  63 {
  64         struct lfsck_namespace_req *lnr;
  65         int                         size;
  66
  67         size = sizeof(*lnr) + (ent->lde_namelen & ~3) + 4;
  68         OBD_ALLOC(lnr, size);
  69         if (lnr == NULL)
  70                 return ERR_PTR(-ENOMEM);
  71
  72         INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
  73         lnr->lnr_lar.lar_parent = lfsck_assistant_object_get(lso);
  74         lnr->lnr_lmv = lfsck_lmv_get(lfsck->li_lmv);
  75         lnr->lnr_fid = ent->lde_fid;
  76         lnr->lnr_dir_cookie = ent->lde_hash;
  77         lnr->lnr_attr = ent->lde_attrs;
  78         lnr->lnr_size = size;
  79         lnr->lnr_type = type;
  80         lnr->lnr_namelen = ent->lde_namelen;
  81         memcpy(lnr->lnr_name, ent->lde_name, ent->lde_namelen);
  82
  83         return lnr;
  84 }
  85
  86 static void lfsck_namespace_assistant_req_fini(const struct lu_env *env,
  87                                                struct lfsck_assistant_req *lar)
  88 {
  89         struct lfsck_namespace_req *lnr =
  90                 container_of(lar, struct lfsck_namespace_req, lnr_lar);
  91
  92         if (lnr->lnr_lmv != NULL)
  93                 lfsck_lmv_put(env, lnr->lnr_lmv);
  94
  95         lfsck_assistant_object_put(env, lar->lar_parent);
  96         OBD_FREE(lnr, lnr->lnr_size);
  97 }
  98
  99 static void lfsck_namespace_le_to_cpu(struct lfsck_namespace *dst,
 100                                       struct lfsck_namespace *src)
 101 {
 102         dst->ln_magic = le32_to_cpu(src->ln_magic);
 103         dst->ln_status = le32_to_cpu(src->ln_status);
 104         dst->ln_flags = le32_to_cpu(src->ln_flags);
 105         dst->ln_success_count = le32_to_cpu(src->ln_success_count);
 106         dst->ln_run_time_phase1 = le64_to_cpu(src->ln_run_time_phase1);
 107         dst->ln_run_time_phase2 = le64_to_cpu(src->ln_run_time_phase2);
 108         dst->ln_time_last_complete = le64_to_cpu(src->ln_time_last_complete);
 109         dst->ln_time_latest_start = le64_to_cpu(src->ln_time_latest_start);
 110         dst->ln_time_last_checkpoint =
 111                                 le64_to_cpu(src->ln_time_last_checkpoint);
 112         lfsck_position_le_to_cpu(&dst->ln_pos_latest_start,
 113                                  &src->ln_pos_latest_start);
 114         lfsck_position_le_to_cpu(&dst->ln_pos_last_checkpoint,
 115                                  &src->ln_pos_last_checkpoint);
 116         lfsck_position_le_to_cpu(&dst->ln_pos_first_inconsistent,
 117                                  &src->ln_pos_first_inconsistent);
 118         dst->ln_items_checked = le64_to_cpu(src->ln_items_checked);
 119         dst->ln_items_repaired = le64_to_cpu(src->ln_items_repaired);
 120         dst->ln_items_failed = le64_to_cpu(src->ln_items_failed);
 121         dst->ln_dirs_checked = le64_to_cpu(src->ln_dirs_checked);
 122         dst->ln_objs_checked_phase2 = le64_to_cpu(src->ln_objs_checked_phase2);
 123         dst->ln_objs_repaired_phase2 =
 124                                 le64_to_cpu(src->ln_objs_repaired_phase2);
 125         dst->ln_objs_failed_phase2 = le64_to_cpu(src->ln_objs_failed_phase2);
 126         dst->ln_objs_nlink_repaired = le64_to_cpu(src->ln_objs_nlink_repaired);
 127         fid_le_to_cpu(&dst->ln_fid_latest_scanned_phase2,
 128                       &src->ln_fid_latest_scanned_phase2);
 129         dst->ln_dirent_repaired = le64_to_cpu(src->ln_dirent_repaired);
 130         dst->ln_linkea_repaired = le64_to_cpu(src->ln_linkea_repaired);
 131         dst->ln_mul_linked_checked = le64_to_cpu(src->ln_mul_linked_checked);
 132         dst->ln_mul_linked_repaired = le64_to_cpu(src->ln_mul_linked_repaired);
 133         dst->ln_unknown_inconsistency =
 134                                 le64_to_cpu(src->ln_unknown_inconsistency);
 135         dst->ln_unmatched_pairs_repaired =
 136                                 le64_to_cpu(src->ln_unmatched_pairs_repaired);
 137         dst->ln_dangling_repaired = le64_to_cpu(src->ln_dangling_repaired);
 138         dst->ln_mul_ref_repaired = le64_to_cpu(src->ln_mul_ref_repaired);
 139         dst->ln_bad_type_repaired = le64_to_cpu(src->ln_bad_type_repaired);
 140         dst->ln_lost_dirent_repaired =
 141                                 le64_to_cpu(src->ln_lost_dirent_repaired);
 142         dst->ln_striped_dirs_scanned =
 143                                 le64_to_cpu(src->ln_striped_dirs_scanned);
 144         dst->ln_striped_dirs_repaired =
 145                                 le64_to_cpu(src->ln_striped_dirs_repaired);
 146         dst->ln_striped_dirs_failed =
 147                                 le64_to_cpu(src->ln_striped_dirs_failed);
 148         dst->ln_striped_dirs_disabled =
 149                                 le64_to_cpu(src->ln_striped_dirs_disabled);
 150         dst->ln_striped_dirs_skipped =
 151                                 le64_to_cpu(src->ln_striped_dirs_skipped);
 152         dst->ln_striped_shards_scanned =
 153                                 le64_to_cpu(src->ln_striped_shards_scanned);
 154         dst->ln_striped_shards_repaired =
 155                                 le64_to_cpu(src->ln_striped_shards_repaired);
 156         dst->ln_striped_shards_failed =
 157                                 le64_to_cpu(src->ln_striped_shards_failed);
 158         dst->ln_striped_shards_skipped =
 159                                 le64_to_cpu(src->ln_striped_shards_skipped);
 160         dst->ln_name_hash_repaired = le64_to_cpu(src->ln_name_hash_repaired);
 161         dst->ln_local_lpf_scanned = le64_to_cpu(src->ln_local_lpf_scanned);
 162         dst->ln_local_lpf_moved = le64_to_cpu(src->ln_local_lpf_moved);
 163         dst->ln_local_lpf_skipped = le64_to_cpu(src->ln_local_lpf_skipped);
 164         dst->ln_local_lpf_failed = le64_to_cpu(src->ln_local_lpf_failed);
 165         dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size);
 166         dst->ln_time_latest_reset = le64_to_cpu(src->ln_time_latest_reset);
 167         dst->ln_linkea_overflow_cleared =
 168                                 le64_to_cpu(src->ln_linkea_overflow_cleared);
 169         dst->ln_agent_entries_repaired =
 170                                 le64_to_cpu(src->ln_agent_entries_repaired);
 171 }
 172
 173 static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
 174                                       struct lfsck_namespace *src)
 175 {
 176         dst->ln_magic = cpu_to_le32(src->ln_magic);
 177         dst->ln_status = cpu_to_le32(src->ln_status);
 178         dst->ln_flags = cpu_to_le32(src->ln_flags);
 179         dst->ln_success_count = cpu_to_le32(src->ln_success_count);
 180         dst->ln_run_time_phase1 = cpu_to_le64(src->ln_run_time_phase1);
 181         dst->ln_run_time_phase2 = cpu_to_le64(src->ln_run_time_phase2);
 182         dst->ln_time_last_complete = cpu_to_le64(src->ln_time_last_complete);
 183         dst->ln_time_latest_start = cpu_to_le64(src->ln_time_latest_start);
 184         dst->ln_time_last_checkpoint =
 185                                 cpu_to_le64(src->ln_time_last_checkpoint);
 186         lfsck_position_cpu_to_le(&dst->ln_pos_latest_start,
 187                                  &src->ln_pos_latest_start);
 188         lfsck_position_cpu_to_le(&dst->ln_pos_last_checkpoint,
 189                                  &src->ln_pos_last_checkpoint);
 190         lfsck_position_cpu_to_le(&dst->ln_pos_first_inconsistent,
 191                                  &src->ln_pos_first_inconsistent);
 192         dst->ln_items_checked = cpu_to_le64(src->ln_items_checked);
 193         dst->ln_items_repaired = cpu_to_le64(src->ln_items_repaired);
 194         dst->ln_items_failed = cpu_to_le64(src->ln_items_failed);
 195         dst->ln_dirs_checked = cpu_to_le64(src->ln_dirs_checked);
 196         dst->ln_objs_checked_phase2 = cpu_to_le64(src->ln_objs_checked_phase2);
 197         dst->ln_objs_repaired_phase2 =
 198                                 cpu_to_le64(src->ln_objs_repaired_phase2);
 199         dst->ln_objs_failed_phase2 = cpu_to_le64(src->ln_objs_failed_phase2);
 200         dst->ln_objs_nlink_repaired = cpu_to_le64(src->ln_objs_nlink_repaired);
 201         fid_cpu_to_le(&dst->ln_fid_latest_scanned_phase2,
 202                       &src->ln_fid_latest_scanned_phase2);
 203         dst->ln_dirent_repaired = cpu_to_le64(src->ln_dirent_repaired);
 204         dst->ln_linkea_repaired = cpu_to_le64(src->ln_linkea_repaired);
 205         dst->ln_mul_linked_checked = cpu_to_le64(src->ln_mul_linked_checked);
 206         dst->ln_mul_linked_repaired = cpu_to_le64(src->ln_mul_linked_repaired);
 207         dst->ln_unknown_inconsistency =
 208                                 cpu_to_le64(src->ln_unknown_inconsistency);
 209         dst->ln_unmatched_pairs_repaired =
 210                                 cpu_to_le64(src->ln_unmatched_pairs_repaired);
 211         dst->ln_dangling_repaired = cpu_to_le64(src->ln_dangling_repaired);
 212         dst->ln_mul_ref_repaired = cpu_to_le64(src->ln_mul_ref_repaired);
 213         dst->ln_bad_type_repaired = cpu_to_le64(src->ln_bad_type_repaired);
 214         dst->ln_lost_dirent_repaired =
 215                                 cpu_to_le64(src->ln_lost_dirent_repaired);
 216         dst->ln_striped_dirs_scanned =
 217                                 cpu_to_le64(src->ln_striped_dirs_scanned);
 218         dst->ln_striped_dirs_repaired =
 219                                 cpu_to_le64(src->ln_striped_dirs_repaired);
 220         dst->ln_striped_dirs_failed =
 221                                 cpu_to_le64(src->ln_striped_dirs_failed);
 222         dst->ln_striped_dirs_disabled =
 223                                 cpu_to_le64(src->ln_striped_dirs_disabled);
 224         dst->ln_striped_dirs_skipped =
 225                                 cpu_to_le64(src->ln_striped_dirs_skipped);
 226         dst->ln_striped_shards_scanned =
 227                                 cpu_to_le64(src->ln_striped_shards_scanned);
 228         dst->ln_striped_shards_repaired =
 229                                 cpu_to_le64(src->ln_striped_shards_repaired);
 230         dst->ln_striped_shards_failed =
 231                                 cpu_to_le64(src->ln_striped_shards_failed);
 232         dst->ln_striped_shards_skipped =
 233                                 cpu_to_le64(src->ln_striped_shards_skipped);
 234         dst->ln_name_hash_repaired = cpu_to_le64(src->ln_name_hash_repaired);
 235         dst->ln_local_lpf_scanned = cpu_to_le64(src->ln_local_lpf_scanned);
 236         dst->ln_local_lpf_moved = cpu_to_le64(src->ln_local_lpf_moved);
 237         dst->ln_local_lpf_skipped = cpu_to_le64(src->ln_local_lpf_skipped);
 238         dst->ln_local_lpf_failed = cpu_to_le64(src->ln_local_lpf_failed);
 239         dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size);
 240         dst->ln_time_latest_reset = cpu_to_le64(src->ln_time_latest_reset);
 241         dst->ln_linkea_overflow_cleared =
 242                                 cpu_to_le64(src->ln_linkea_overflow_cleared);
 243         dst->ln_agent_entries_repaired =
 244                                 cpu_to_le64(src->ln_agent_entries_repaired);
 245 }
 246
 247 static void lfsck_namespace_record_failure(const struct lu_env *env,
 248                                            struct lfsck_instance *lfsck,
 249                                            struct lfsck_namespace *ns)
 250 {
 251         struct lfsck_position pos;
 252
 253         ns->ln_items_failed++;
 254         lfsck_pos_fill(env, lfsck, &pos, false);
 255         if (lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent) ||
 256             lfsck_pos_is_eq(&pos, &ns->ln_pos_first_inconsistent) < 0) {
 257                 ns->ln_pos_first_inconsistent = pos;
 258
 259                 CDEBUG(D_LFSCK, "%s: namespace LFSCK hit first non-repaired "
 260                        "inconsistency at the pos [%llu, "DFID", %#llx]\n",
 261                        lfsck_lfsck2name(lfsck),
 262                        ns->ln_pos_first_inconsistent.lp_oit_cookie,
 263                        PFID(&ns->ln_pos_first_inconsistent.lp_dir_parent),
 264                        ns->ln_pos_first_inconsistent.lp_dir_cookie);
 265         }
 266 }
 267
 268 /**
 269  * Load the MDT bitmap from the lfsck_namespace trace file.
 270  *
 271  * \param[in] env       pointer to the thread context
 272  * \param[in] com       pointer to the lfsck component
 273  *
 274  * \retval              0 for success
 275  * \retval              negative error number on failure or data corruption
 276  */
 277 static int lfsck_namespace_load_bitmap(const struct lu_env *env,
 278                                        struct lfsck_component *com)
 279 {
 280         struct dt_object                *obj    = com->lc_obj;
 281         struct lfsck_assistant_data     *lad    = com->lc_data;
 282         struct lfsck_namespace          *ns     = com->lc_file_ram;
 283         struct cfs_bitmap                       *bitmap = lad->lad_bitmap;
 284         ssize_t                          size;
 285         __u32                            nbits;
 286         int                              rc;
 287         ENTRY;
 288
 289         if (com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size >
 290             ns->ln_bitmap_size)
 291                 nbits = com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size;
 292         else
 293                 nbits = ns->ln_bitmap_size;
 294
 295         if (unlikely(nbits < BITS_PER_LONG))
 296                 nbits = BITS_PER_LONG;
 297
 298         if (nbits > bitmap->size) {
 299                 __u32 new_bits = bitmap->size;
 300                 struct cfs_bitmap *new_bitmap;
 301
 302                 while (new_bits < nbits)
 303                         new_bits <<= 1;
 304
 305                 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
 306                 if (new_bitmap == NULL)
 307                         RETURN(-ENOMEM);
 308
 309                 lad->lad_bitmap = new_bitmap;
 310                 CFS_FREE_BITMAP(bitmap);
 311                 bitmap = new_bitmap;
 312         }
 313
 314         if (ns->ln_bitmap_size == 0) {
 315                 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
 316                 CFS_RESET_BITMAP(bitmap);
 317
 318                 RETURN(0);
 319         }
 320
 321         size = (ns->ln_bitmap_size + 7) >> 3;
 322         rc = dt_xattr_get(env, obj,
 323                           lfsck_buf_get(env, bitmap->data, size),
 324                           XATTR_NAME_LFSCK_BITMAP);
 325         if (rc != size)
 326                 RETURN(rc >= 0 ? -EINVAL : rc);
 327
 328         if (cfs_bitmap_check_empty(bitmap))
 329                 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
 330         else
 331                 set_bit(LAD_INCOMPLETE, &lad->lad_flags);
 332
 333         RETURN(0);
 334 }
 335
 336 /**
 337  * Load namespace LFSCK statistics information from the trace file.
 338  *
 339  * \param[in] env       pointer to the thread context
 340  * \param[in] com       pointer to the lfsck component
 341  *
 342  * \retval              0 for success
 343  * \retval              negative error number on failure
 344  */
 345 static int lfsck_namespace_load(const struct lu_env *env,
 346                                 struct lfsck_component *com)
 347 {
 348         int len = com->lc_file_size;
 349         int rc;
 350
 351         rc = dt_xattr_get(env, com->lc_obj,
 352                           lfsck_buf_get(env, com->lc_file_disk, len),
 353                           XATTR_NAME_LFSCK_NAMESPACE);
 354         if (rc == len) {
 355                 struct lfsck_namespace *ns = com->lc_file_ram;
 356
 357                 lfsck_namespace_le_to_cpu(ns,
 358                                 (struct lfsck_namespace *)com->lc_file_disk);
 359                 if (ns->ln_magic != LFSCK_NAMESPACE_MAGIC) {
 360                         CDEBUG(D_LFSCK, "%s: invalid lfsck_namespace magic "
 361                                "%#x != %#x\n", lfsck_lfsck2name(com->lc_lfsck),
 362                                ns->ln_magic, LFSCK_NAMESPACE_MAGIC);
 363                         rc = -ESTALE;
 364                 } else {
 365                         rc = 0;
 366                 }
 367         } else if (rc != -ENODATA) {
 368                 CDEBUG(D_LFSCK, "%s: fail to load lfsck_namespace, "
 369                        "expected = %d: rc = %d\n",
 370                        lfsck_lfsck2name(com->lc_lfsck), len, rc);
 371                 if (rc >= 0)
 372                         rc = -ESTALE;
 373         }
 374
 375         return rc;
 376 }
 377
 378 static int lfsck_namespace_store(const struct lu_env *env,
 379                                  struct lfsck_component *com)
 380 {
 381         struct dt_object                *obj    = com->lc_obj;
 382         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 383         struct lfsck_namespace          *ns     = com->lc_file_ram;
 384         struct lfsck_assistant_data     *lad    = com->lc_data;
 385         struct dt_device                *dev    = lfsck_obj2dev(obj);
 386         struct cfs_bitmap               *bitmap = NULL;
 387         struct thandle                  *handle;
 388         __u32                            nbits  = 0;
 389         int                              len    = com->lc_file_size;
 390         int                              rc;
 391         ENTRY;
 392
 393         if (lad != NULL) {
 394                 bitmap = lad->lad_bitmap;
 395                 nbits = bitmap->size;
 396
 397                 LASSERT(nbits > 0);
 398                 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
 399         }
 400
 401         ns->ln_bitmap_size = nbits;
 402         lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk,
 403                                   ns);
 404         handle = dt_trans_create(env, dev);
 405         if (IS_ERR(handle))
 406                 GOTO(log, rc = PTR_ERR(handle));
 407
 408         rc = dt_declare_xattr_set(env, obj,
 409                                   lfsck_buf_get(env, com->lc_file_disk, len),
 410                                   XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
 411         if (rc != 0)
 412                 GOTO(out, rc);
 413
 414         if (bitmap != NULL) {
 415                 rc = dt_declare_xattr_set(env, obj,
 416                                 lfsck_buf_get(env, bitmap->data, nbits >> 3),
 417                                 XATTR_NAME_LFSCK_BITMAP, 0, handle);
 418                 if (rc != 0)
 419                         GOTO(out, rc);
 420         }
 421
 422         rc = dt_trans_start_local(env, dev, handle);
 423         if (rc != 0)
 424                 GOTO(out, rc);
 425
 426         rc = dt_xattr_set(env, obj,
 427                           lfsck_buf_get(env, com->lc_file_disk, len),
 428                           XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
 429         if (rc == 0 && bitmap != NULL)
 430                 rc = dt_xattr_set(env, obj,
 431                                   lfsck_buf_get(env, bitmap->data, nbits >> 3),
 432                                   XATTR_NAME_LFSCK_BITMAP, 0, handle);
 433
 434         GOTO(out, rc);
 435
 436 out:
 437         dt_trans_stop(env, dev, handle);
 438
 439 log:
 440         if (rc != 0)
 441                 CDEBUG(D_LFSCK, "%s: fail to store lfsck_namespace: rc = %d\n",
 442                        lfsck_lfsck2name(lfsck), rc);
 443         return rc;
 444 }
 445
 446 static int lfsck_namespace_init(const struct lu_env *env,
 447                                 struct lfsck_component *com)
 448 {
 449         struct lfsck_namespace *ns = com->lc_file_ram;
 450         int rc;
 451
 452         memset(ns, 0, sizeof(*ns));
 453         ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
 454         ns->ln_status = LS_INIT;
 455         ns->ln_time_latest_reset = ktime_get_real_seconds();
 456         down_write(&com->lc_sem);
 457         rc = lfsck_namespace_store(env, com);
 458         if (rc == 0)
 459                 rc = lfsck_load_sub_trace_files(env, com,
 460                         &dt_lfsck_namespace_features, LFSCK_NAMESPACE, true);
 461         up_write(&com->lc_sem);
 462
 463         return rc;
 464 }
 465
 466 /**
 467  * Update the namespace LFSCK trace file for the given @fid
 468  *
 469  * \param[in] env       pointer to the thread context
 470  * \param[in] com       pointer to the lfsck component
 471  * \param[in] fid       the fid which flags to be updated in the lfsck
 472  *                      trace file
 473  * \param[in] add       true if add new flags, otherwise remove flags
 474  *
 475  * \retval              0 for success or nothing to be done
 476  * \retval              negative error number on failure
 477  */
 478 int lfsck_namespace_trace_update(const struct lu_env *env,
 479                                  struct lfsck_component *com,
 480                                  const struct lu_fid *fid,
 481                                  const __u8 flags, bool add)
 482 {
 483         struct lfsck_instance   *lfsck  = com->lc_lfsck;
 484         struct dt_object        *obj;
 485         struct lu_fid           *key    = &lfsck_env_info(env)->lti_fid3;
 486         struct dt_device        *dev;
 487         struct thandle          *th     = NULL;
 488         int                      idx;
 489         int                      rc     = 0;
 490         __u8                     old    = 0;
 491         __u8                     new    = 0;
 492         ENTRY;
 493
 494         LASSERT(flags != 0);
 495
 496         if (unlikely(!fid_is_sane(fid)))
 497                 RETURN(0);
 498
 499         idx = lfsck_sub_trace_file_fid2idx(fid);
 500         mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
 501         obj = com->lc_sub_trace_objs[idx].lsto_obj;
 502         if (unlikely(obj == NULL)) {
 503                 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
 504                 RETURN(0);
 505         }
 506
 507         lfsck_object_get(obj);
 508         dev = lfsck_obj2dev(obj);
 509         fid_cpu_to_be(key, fid);
 510         rc = dt_lookup(env, obj, (struct dt_rec *)&old,
 511                        (const struct dt_key *)key);
 512         if (rc == -ENOENT) {
 513                 if (!add)
 514                         GOTO(unlock, rc = 0);
 515
 516                 old = 0;
 517                 new = flags;
 518         } else if (rc == 0) {
 519                 if (add) {
 520                         if ((old & flags) == flags)
 521                                 GOTO(unlock, rc = 0);
 522
 523                         new = old | flags;
 524                 } else {
 525                         if ((old & flags) == 0)
 526                                 GOTO(unlock, rc = 0);
 527
 528                         new = old & ~flags;
 529                 }
 530         } else {
 531                 GOTO(log, rc);
 532         }
 533
 534         th = dt_trans_create(env, dev);
 535         if (IS_ERR(th))
 536                 GOTO(log, rc = PTR_ERR(th));
 537
 538         if (old != 0) {
 539                 rc = dt_declare_delete(env, obj,
 540                                        (const struct dt_key *)key, th);
 541                 if (rc != 0)
 542                         GOTO(log, rc);
 543         }
 544
 545         if (new != 0) {
 546                 rc = dt_declare_insert(env, obj,
 547                                        (const struct dt_rec *)&new,
 548                                        (const struct dt_key *)key, th);
 549                 if (rc != 0)
 550                         GOTO(log, rc);
 551         }
 552
 553         rc = dt_trans_start_local(env, dev, th);
 554         if (rc != 0)
 555                 GOTO(log, rc);
 556
 557         if (old != 0) {
 558                 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
 559                 if (rc != 0)
 560                         GOTO(log, rc);
 561         }
 562
 563         if (new != 0) {
 564                 rc = dt_insert(env, obj, (const struct dt_rec *)&new,
 565                                (const struct dt_key *)key, th);
 566                 if (rc != 0)
 567                         GOTO(log, rc);
 568         }
 569
 570         GOTO(log, rc);
 571
 572 log:
 573         if (th != NULL && !IS_ERR(th))
 574                 dt_trans_stop(env, dev, th);
 575
 576         CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the "
 577                "trace file, flags %x, old %x, new %x: rc = %d\n",
 578                lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid),
 579                (__u32)flags, (__u32)old, (__u32)new, rc);
 580
 581 unlock:
 582         mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
 583         lfsck_object_put(env, obj);
 584
 585         return rc;
 586 }
 587
 588 int lfsck_namespace_check_exist(const struct lu_env *env,
 589                                 struct dt_object *dir,
 590                                 struct dt_object *obj, const char *name)
 591 {
 592         struct lu_fid    *fid = &lfsck_env_info(env)->lti_fid;
 593         int               rc;
 594         ENTRY;
 595
 596         if (unlikely(lfsck_is_dead_obj(obj)))
 597                 RETURN(LFSCK_NAMEENTRY_DEAD);
 598
 599         rc = dt_lookup_dir(env, dir, name, fid);
 600         if (rc == -ENOENT)
 601                 RETURN(LFSCK_NAMEENTRY_REMOVED);
 602
 603         if (rc < 0)
 604                 RETURN(rc);
 605
 606         if (!lu_fid_eq(fid, lfsck_dto2fid(obj)))
 607                 RETURN(LFSCK_NAMEENTRY_RECREATED);
 608
 609         RETURN(0);
 610 }
 611
 612 static int lfsck_declare_namespace_exec_dir(const struct lu_env *env,
 613                                             struct dt_object *obj,
 614                                             struct thandle *handle)
 615 {
 616         int rc;
 617
 618         /* For remote updating LINKEA, there may be further LFSCK action
 619          * on remote MDT after the updating, so update the LINKEA ASAP. */
 620         if (dt_object_remote(obj))
 621                 handle->th_sync = 1;
 622
 623         /* For destroying all invalid linkEA entries. */
 624         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
 625         if (rc == 0)
 626                 /* For insert new linkEA entry. */
 627                 rc = dt_declare_xattr_set(env, obj,
 628                         lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
 629                         XATTR_NAME_LINK, 0, handle);
 630         return rc;
 631 }
 632
 633 int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
 634                        struct linkea_data *ldata, bool with_rec)
 635 {
 636         int rc;
 637
 638         if (ldata->ld_buf->lb_buf == NULL)
 639                 return -ENOMEM;
 640
 641         if (!dt_object_exists(obj))
 642                 return -ENOENT;
 643
 644         rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
 645         if (rc == -ERANGE) {
 646                 /* Buf was too small, figure out what we need. */
 647                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK);
 648                 if (unlikely(rc == 0))
 649                         return -ENODATA;
 650
 651                 if (rc < 0)
 652                         return rc;
 653
 654                 lu_buf_realloc(ldata->ld_buf, rc);
 655                 if (ldata->ld_buf->lb_buf == NULL)
 656                         return -ENOMEM;
 657
 658                 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK);
 659         }
 660
 661         if (unlikely(rc == 0))
 662                 return -ENODATA;
 663
 664         if (rc > 0) {
 665                 if (with_rec)
 666                         rc = linkea_init_with_rec(ldata);
 667                 else
 668                         rc = linkea_init(ldata);
 669         }
 670
 671         return rc;
 672 }
 673
 674 /**
 675  * Remove linkEA for the given object.
 676  *
 677  * The caller should take the ldlm lock before the calling.
 678  *
 679  * \param[in] env       pointer to the thread context
 680  * \param[in] com       pointer to the lfsck component
 681  * \param[in] obj       pointer to the dt_object to be handled
 682  *
 683  * \retval              0 for repaired cases
 684  * \retval              negative error number on failure
 685  */
 686 static int lfsck_namespace_links_remove(const struct lu_env *env,
 687                                         struct lfsck_component *com,
 688                                         struct dt_object *obj)
 689 {
 690         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 691         struct dt_device                *dev    = lfsck_obj2dev(obj);
 692         struct thandle                  *th     = NULL;
 693         int                              rc     = 0;
 694         ENTRY;
 695
 696         LASSERT(dt_object_remote(obj) == 0);
 697
 698         th = dt_trans_create(env, dev);
 699         if (IS_ERR(th))
 700                 GOTO(log, rc = PTR_ERR(th));
 701
 702         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
 703         if (rc != 0)
 704                 GOTO(stop, rc);
 705
 706         rc = dt_trans_start_local(env, dev, th);
 707         if (rc != 0)
 708                 GOTO(stop, rc);
 709
 710         dt_write_lock(env, obj, 0);
 711         if (unlikely(lfsck_is_dead_obj(obj)))
 712                 GOTO(unlock, rc = -ENOENT);
 713
 714         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
 715                 GOTO(unlock, rc = 0);
 716
 717         rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
 718
 719         GOTO(unlock, rc);
 720
 721 unlock:
 722         dt_write_unlock(env, obj);
 723
 724 stop:
 725         dt_trans_stop(env, dev, th);
 726
 727 log:
 728         CDEBUG(D_LFSCK, "%s: namespace LFSCK remove invalid linkEA "
 729                "for the object "DFID": rc = %d\n",
 730                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
 731
 732         if (rc == 0) {
 733                 struct lfsck_namespace *ns = com->lc_file_ram;
 734
 735                 ns->ln_flags |= LF_INCONSISTENT;
 736         }
 737
 738         return rc;
 739 }
 740
 741 static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
 742                              struct linkea_data *ldata, struct thandle *handle)
 743 {
 744         struct lu_buf buf;
 745         int rc;
 746
 747         lfsck_buf_init(&buf, ldata->ld_buf->lb_buf, ldata->ld_leh->leh_len);
 748
 749 again:
 750         rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LINK, 0, handle);
 751         if (unlikely(rc == -ENOSPC)) {
 752                 rc = linkea_overflow_shrink(ldata);
 753                 if (likely(rc > 0)) {
 754                         buf.lb_len = rc;
 755                         goto again;
 756                 }
 757         }
 758
 759         return rc;
 760 }
 761
 762 static inline bool linkea_reclen_is_valid(const struct linkea_data *ldata)
 763 {
 764         if (ldata->ld_reclen <= 0)
 765                 return false;
 766
 767         if ((char *)ldata->ld_lee + ldata->ld_reclen >
 768             (char *)ldata->ld_leh + ldata->ld_leh->leh_len)
 769                 return false;
 770
 771         return true;
 772 }
 773
 774 static inline bool linkea_entry_is_valid(const struct linkea_data *ldata,
 775                                          const struct lu_name *cname,
 776                                          const struct lu_fid *pfid)
 777 {
 778         if (!linkea_reclen_is_valid(ldata))
 779                 return false;
 780
 781         if (cname->ln_namelen <= 0 || cname->ln_namelen > NAME_MAX)
 782                 return false;
 783
 784         if (!fid_is_sane(pfid))
 785                 return false;
 786
 787         return true;
 788 }
 789
 790 static int lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
 791                                                struct lu_name *cname,
 792                                                struct lu_fid *pfid,
 793                                                char *buf, const int buflen)
 794 {
 795         linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
 796         if (unlikely(!linkea_entry_is_valid(ldata, cname, pfid)))
 797                 return -EINVAL;
 798
 799         /* To guarantee the 'name' is terminated with '0'. */
 800         memcpy(buf, cname->ln_name, cname->ln_namelen);
 801         buf[cname->ln_namelen] = 0;
 802         cname->ln_name = buf;
 803
 804         return 0;
 805 }
 806
 807 static void lfsck_linkea_del_buf(struct linkea_data *ldata,
 808                                  const struct lu_name *lname)
 809 {
 810         LASSERT(ldata->ld_leh != NULL && ldata->ld_lee != NULL);
 811
 812         /* If current record is corrupted, all the subsequent
 813          * records will be dropped. */
 814         if (unlikely(!linkea_reclen_is_valid(ldata))) {
 815                 void *ptr = ldata->ld_lee;
 816
 817                 ldata->ld_leh->leh_len = sizeof(struct link_ea_header);
 818                 ldata->ld_leh->leh_reccount = 0;
 819                 linkea_first_entry(ldata);
 820                 while (ldata->ld_lee != NULL &&
 821                        (char *)ldata->ld_lee < (char *)ptr) {
 822                         int reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
 823                                      ldata->ld_lee->lee_reclen[1];
 824
 825                         ldata->ld_leh->leh_len += reclen;
 826                         ldata->ld_leh->leh_reccount++;
 827                         ldata->ld_lee = (struct link_ea_entry *)
 828                                         ((char *)ldata->ld_lee + reclen);
 829                 }
 830
 831                 ldata->ld_lee = NULL;
 832         } else {
 833                 linkea_del_buf(ldata, lname);
 834         }
 835 }
 836
 837 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
 838                                                struct lu_name *cname,
 839                                                struct lu_fid *pfid,
 840                                                bool remove)
 841 {
 842         struct link_ea_entry    *oldlee;
 843         int                      oldlen;
 844         int                      repeated = 0;
 845
 846         oldlee = ldata->ld_lee;
 847         oldlen = ldata->ld_reclen;
 848         linkea_next_entry(ldata);
 849         while (ldata->ld_lee != NULL) {
 850                 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
 851                                    ldata->ld_lee->lee_reclen[1];
 852                 if (unlikely(!linkea_reclen_is_valid(ldata))) {
 853                         lfsck_linkea_del_buf(ldata, NULL);
 854                         LASSERT(ldata->ld_lee == NULL);
 855                 } else if (unlikely(ldata->ld_reclen == oldlen &&
 856                              memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
 857                         repeated++;
 858                         if (!remove)
 859                                 break;
 860
 861                         lfsck_linkea_del_buf(ldata, cname);
 862                 } else {
 863                         linkea_next_entry(ldata);
 864                 }
 865         }
 866         ldata->ld_lee = oldlee;
 867         ldata->ld_reclen = oldlen;
 868
 869         return repeated;
 870 }
 871
 872 /**
 873  * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
 874  *
 875  * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
 876  * with the given type to generate the name, the detailed rules for name
 877  * have been described as following.
 878  *
 879  * The function also generates the linkEA corresponding to the name entry
 880  * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
 881  *
 882  * \param[in] env       pointer to the thread context
 883  * \param[in] com       pointer to the lfsck component
 884  * \param[in] orphan    pointer to the orphan MDT-object
 885  * \param[in] infix     additional information for the orphan name, such as
 886  *                      the FID for original
 887  * \param[in] type      the type for describing why the orphan MDT-object is
 888  *                      created. The rules are as following:
 889  *
 890  *  type "D":           The MDT-object is a directory, it may knows its parent
 891  *                      but because there is no valid linkEA, the LFSCK cannot
 892  *                      know where to put it back to the namespace.
 893  *  type "O":           The MDT-object has no linkEA, and there is no name
 894  *                      entry that references the MDT-object.
 895  *
 896  *  type "S":           The orphan MDT-object is a shard of a striped directory
 897  *
 898  * \see lfsck_layout_recreate_parent() for more types.
 899  *
 900  * The orphan name will be like:
 901  * ${FID}-${infix}-${type}-${conflict_version}
 902  *
 903  * \param[out] count    if some others inserted some linkEA entries by race,
 904  *                      then return the linkEA entries count.
 905  *
 906  * \retval              positive number for repaired cases
 907  * \retval              0 if needs to repair nothing
 908  * \retval              negative error number on failure
 909  */
 910 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
 911                                          struct lfsck_component *com,
 912                                          struct dt_object *orphan,
 913                                          const char *infix, const char *type,
 914                                          int *count)
 915 {
 916         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 917         struct lu_name                  *cname  = &info->lti_name;
 918         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
 919         struct lu_attr                  *la     = &info->lti_la2;
 920         const struct lu_fid             *cfid   = lfsck_dto2fid(orphan);
 921         const struct lu_fid             *pfid;
 922         struct lu_fid                    tfid;
 923         struct lfsck_instance           *lfsck  = com->lc_lfsck;
 924         struct dt_device                *dev    = lfsck_obj2dev(orphan);
 925         struct dt_object                *parent;
 926         struct thandle                  *th     = NULL;
 927         struct lfsck_lock_handle        *pllh   = &info->lti_llh;
 928         struct lustre_handle             clh    = { 0 };
 929         struct linkea_data               ldata2 = { NULL };
 930         struct lu_buf                    linkea_buf;
 931         int                              namelen;
 932         int                              idx    = 0;
 933         int                              rc     = 0;
 934         bool                             exist  = false;
 935         ENTRY;
 936
 937         cname->ln_name = NULL;
 938         if (unlikely(lfsck->li_lpf_obj == NULL))
 939                 GOTO(log, rc = -ENXIO);
 940
 941         parent = lfsck->li_lpf_obj;
 942         pfid = lfsck_dto2fid(parent);
 943
 944 again:
 945         do {
 946                 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
 947                                    PFID(cfid), infix, type, idx++);
 948                 rc = dt_lookup_dir(env, parent, info->lti_key, &tfid);
 949                 if (rc != 0 && rc != -ENOENT)
 950                         GOTO(log, rc);
 951
 952                 if (unlikely(rc == 0 && lu_fid_eq(cfid, &tfid)))
 953                         exist = true;
 954         } while (rc == 0 && !exist);
 955
 956         rc = lfsck_lock(env, lfsck, parent, info->lti_key, pllh,
 957                         MDS_INODELOCK_UPDATE, LCK_PW);
 958         if (rc != 0)
 959                 GOTO(log, rc);
 960
 961         /* Re-check whether the name conflict with othrs after taken
 962          * the ldlm lock. */
 963         rc = dt_lookup_dir(env, parent, info->lti_key, &tfid);
 964         if (rc == 0) {
 965                 if (!lu_fid_eq(cfid, &tfid)) {
 966                         exist = false;
 967                         lfsck_unlock(pllh);
 968                         goto again;
 969                 }
 970
 971                 exist = true;
 972         } else if (rc != -ENOENT) {
 973                 GOTO(log, rc);
 974         } else {
 975                 exist = false;
 976         }
 977
 978         cname->ln_name = info->lti_key;
 979         cname->ln_namelen = namelen;
 980         rc = linkea_links_new(&ldata2, &info->lti_linkea_buf2,
 981                               cname, pfid);
 982         if (rc != 0)
 983                 GOTO(log, rc);
 984
 985         rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
 986                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
 987                               MDS_INODELOCK_XATTR, LCK_EX);
 988         if (rc != 0)
 989                 GOTO(log, rc);
 990
 991         lfsck_buf_init(&linkea_buf, ldata2.ld_buf->lb_buf,
 992                        ldata2.ld_leh->leh_len);
 993         th = dt_trans_create(env, dev);
 994         if (IS_ERR(th))
 995                 GOTO(log, rc = PTR_ERR(th));
 996
 997         if (S_ISDIR(lfsck_object_type(orphan))) {
 998                 rc = dt_declare_delete(env, orphan,
 999                                        (const struct dt_key *)dotdot, th);
1000                 if (rc != 0)
1001                         GOTO(stop, rc);
1002
1003                 rec->rec_type = S_IFDIR;
1004                 rec->rec_fid = pfid;
1005                 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1006                                        (const struct dt_key *)dotdot, th);
1007                 if (rc != 0)
1008                         GOTO(stop, rc);
1009         }
1010
1011         rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1012                                   XATTR_NAME_LINK, 0, th);
1013         if (rc != 0)
1014                 GOTO(stop, rc);
1015
1016         if (!exist) {
1017                 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1018                 rec->rec_fid = cfid;
1019                 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1020                                        (const struct dt_key *)cname->ln_name,
1021                                        th);
1022                 if (rc != 0)
1023                         GOTO(stop, rc);
1024
1025                 if (S_ISDIR(rec->rec_type)) {
1026                         rc = dt_declare_ref_add(env, parent, th);
1027                         if (rc != 0)
1028                                 GOTO(stop, rc);
1029                 }
1030         }
1031
1032         memset(la, 0, sizeof(*la));
1033         la->la_ctime = ktime_get_real_seconds();
1034         la->la_valid = LA_CTIME;
1035         rc = dt_declare_attr_set(env, orphan, la, th);
1036         if (rc != 0)
1037                 GOTO(stop, rc);
1038
1039         rc = dt_trans_start_local(env, dev, th);
1040         if (rc != 0)
1041                 GOTO(stop, rc);
1042
1043         dt_write_lock(env, orphan, 0);
1044         rc = lfsck_links_read2_with_rec(env, orphan, &ldata2);
1045         if (likely(rc == -ENODATA || rc == -EINVAL)) {
1046                 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1047                         GOTO(unlock, rc = 1);
1048
1049                 if (S_ISDIR(lfsck_object_type(orphan))) {
1050                         rc = dt_delete(env, orphan,
1051                                        (const struct dt_key *)dotdot, th);
1052                         if (rc != 0)
1053                                 GOTO(unlock, rc);
1054
1055                         rec->rec_type = S_IFDIR;
1056                         rec->rec_fid = pfid;
1057                         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1058                                        (const struct dt_key *)dotdot, th);
1059                         if (rc != 0)
1060                                 GOTO(unlock, rc);
1061                 }
1062
1063                 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
1064                                   th);
1065         } else {
1066                 if (rc == 0 && count != NULL)
1067                         *count = ldata2.ld_leh->leh_reccount;
1068
1069                 GOTO(unlock, rc);
1070         }
1071         dt_write_unlock(env, orphan);
1072
1073         if (rc == 0 && !exist) {
1074                 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
1075                 rec->rec_fid = cfid;
1076                 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1077                                (const struct dt_key *)cname->ln_name, th);
1078                 if (rc == 0 && S_ISDIR(rec->rec_type)) {
1079                         dt_write_lock(env, parent, 0);
1080                         rc = dt_ref_add(env, parent, th);
1081                         dt_write_unlock(env, parent);
1082                 }
1083         }
1084
1085         if (rc == 0)
1086                 rc = dt_attr_set(env, orphan, la, th);
1087
1088         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1089
1090 unlock:
1091         dt_write_unlock(env, orphan);
1092
1093 stop:
1094         dt_trans_stop(env, dev, th);
1095
1096 log:
1097         lfsck_ibits_unlock(&clh, LCK_EX);
1098         lfsck_unlock(pllh);
1099         CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
1100                "object "DFID", name = %s: rc = %d\n",
1101                lfsck_lfsck2name(lfsck), PFID(cfid),
1102                cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1103
1104         if (rc != 0) {
1105                 struct lfsck_namespace *ns = com->lc_file_ram;
1106
1107                 ns->ln_flags |= LF_INCONSISTENT;
1108         }
1109
1110         return rc;
1111 }
1112
1113 static int lfsck_lmv_set(const struct lu_env *env,
1114                          struct lfsck_instance *lfsck,
1115                          struct dt_object *obj,
1116                          struct lmv_mds_md_v1 *lmv)
1117 {
1118         struct dt_device *dev = lfsck->li_next;
1119         struct thandle *th = NULL;
1120         struct lu_buf buf = { lmv, sizeof(*lmv) };
1121         int rc;
1122
1123         ENTRY;
1124
1125         th = dt_trans_create(env, dev);
1126         if (IS_ERR(th))
1127                 RETURN(PTR_ERR(th));
1128
1129         rc = dt_declare_xattr_set(env, obj, &buf, XATTR_NAME_LMV, 0, th);
1130         if (rc)
1131                 GOTO(stop, rc);
1132
1133         rc = dt_trans_start_local(env, dev, th);
1134         if (rc != 0)
1135                 GOTO(stop, rc);
1136
1137         rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LMV, 0, th);
1138         if (rc)
1139                 GOTO(stop, rc);
1140
1141         EXIT;
1142 stop:
1143         dt_trans_stop(env, dev, th);
1144
1145         return rc;
1146 }
1147
1148 static int lfsck_lmv_delete(const struct lu_env *env,
1149                             struct lfsck_instance *lfsck,
1150                             struct dt_object *obj)
1151 {
1152         struct dt_device *dev = lfsck->li_next;
1153         struct thandle *th = NULL;
1154         int rc;
1155
1156         ENTRY;
1157
1158         th = dt_trans_create(env, dev);
1159         if (IS_ERR(th))
1160                 RETURN(PTR_ERR(th));
1161
1162         rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LMV, th);
1163         if (rc)
1164                 GOTO(stop, rc);
1165
1166         rc = dt_trans_start_local(env, dev, th);
1167         if (rc != 0)
1168                 GOTO(stop, rc);
1169
1170         rc = dt_xattr_del(env, obj, XATTR_NAME_LMV, th);
1171         if (rc)
1172                 GOTO(stop, rc);
1173
1174         EXIT;
1175 stop:
1176         dt_trans_stop(env, dev, th);
1177
1178         return rc;
1179 }
1180
1181 static inline int lfsck_object_is_shard(const struct lu_env *env,
1182                                         struct lfsck_instance *lfsck,
1183                                         struct dt_object *obj,
1184                                         const struct lu_name *lname)
1185 {
1186         struct lfsck_thread_info *info = lfsck_env_info(env);
1187         struct lmv_mds_md_v1 *lmv = &info->lti_lmv;
1188         int rc;
1189
1190         rc = lfsck_shard_name_to_index(env, lname->ln_name, lname->ln_namelen,
1191                                        lfsck_object_type(obj),
1192                                        lfsck_dto2fid(obj));
1193         if (rc < 0)
1194                 return 0;
1195
1196         rc = lfsck_read_stripe_lmv(env, lfsck, obj, lmv);
1197         if (rc == -ENODATA)
1198                 return 0;
1199
1200         if (!rc && lmv->lmv_magic == LMV_MAGIC_STRIPE)
1201                 return 1;
1202
1203         return rc;
1204 }
1205
1206 /**
1207  * Add the specified name entry back to namespace.
1208  *
1209  * If there is a linkEA entry that back references a name entry under
1210  * some parent directory, but such parent directory does not have the
1211  * claimed name entry. On the other hand, the linkEA entries count is
1212  * not larger than the MDT-object's hard link count. Under such case,
1213  * it is quite possible that the name entry is lost. Then the LFSCK
1214  * should add the name entry back to the namespace.
1215  *
1216  * If \a child is shard, which means \a parent is a striped directory,
1217  * if \a parent has LMV, we need to delete it before insertion because
1218  * now parent's striping is broken and can't be parsed correctly.
1219  *
1220  * \param[in] env       pointer to the thread context
1221  * \param[in] com       pointer to the lfsck component
1222  * \param[in] parent    pointer to the directory under which the name entry
1223  *                      will be inserted into
1224  * \param[in] child     pointer to the object referenced by the name entry
1225  *                      that to be inserted into the parent
1226  * \param[in] lname     the name for the child in the parent directory
1227  *
1228  * \retval              positive number for repaired cases
1229  * \retval              0 if nothing to be repaired
1230  * \retval              negative error number on failure
1231  */
1232 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1233                                          struct lfsck_component *com,
1234                                          struct dt_object *parent,
1235                                          struct dt_object *child,
1236                                          const struct lu_name *lname)
1237 {
1238         struct lfsck_thread_info *info = lfsck_env_info(env);
1239         struct lu_attr *la = &info->lti_la;
1240         struct dt_insert_rec *rec = &info->lti_dt_rec;
1241         struct lfsck_instance *lfsck = com->lc_lfsck;
1242         /* The child and its name may be on different MDTs. */
1243         const struct lu_fid *pfid = lfsck_dto2fid(parent);
1244         const struct lu_fid *cfid = lfsck_dto2fid(child);
1245         struct dt_device *dev = lfsck->li_next;
1246         struct thandle *th = NULL;
1247         struct lfsck_lock_handle *llh = &info->lti_llh;
1248         struct lmv_mds_md_v1 *lmv = &info->lti_lmv;
1249         struct lu_buf buf = { lmv, sizeof(*lmv) };
1250         /* whether parent's LMV is deleted before insertion */
1251         bool parent_lmv_deleted = false;
1252         /* whether parent's LMV is missing */
1253         bool parent_lmv_lost = false;
1254         int rc = 0;
1255
1256         ENTRY;
1257
1258         /* @parent/@child may be based on lfsck->li_bottom,
1259          * but here we need the object based on the lfsck->li_next. */
1260
1261         parent = lfsck_object_locate(dev, parent);
1262         if (IS_ERR(parent))
1263                 GOTO(log, rc = PTR_ERR(parent));
1264
1265         child = lfsck_object_locate(dev, child);
1266         if (IS_ERR(child))
1267                 GOTO(log, rc = PTR_ERR(child));
1268
1269         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1270                 GOTO(log, rc = 1);
1271
1272         rc = lfsck_lock(env, lfsck, parent, lname->ln_name, llh,
1273                         MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
1274                         MDS_INODELOCK_XATTR, LCK_EX);
1275         if (rc)
1276                 GOTO(log, rc);
1277
1278         rc = lfsck_object_is_shard(env, lfsck, child, lname);
1279         if (rc < 0)
1280                 GOTO(unlock, rc);
1281
1282         if (rc == 1) {
1283                 rc = lfsck_read_stripe_lmv(env, lfsck, parent, lmv);
1284                 if (!rc) {
1285                         /*
1286                          * To add a shard, we need to convert parent to a
1287                          * plain directory by deleting its LMV, and after
1288                          * insertion set it back.
1289                          */
1290                         rc = lfsck_lmv_delete(env, lfsck, parent);
1291                         if (rc)
1292                                 GOTO(unlock, rc);
1293                         parent_lmv_deleted = true;
1294                         lmv->lmv_layout_version++;
1295                         lfsck_lmv_header_cpu_to_le(lmv, lmv);
1296                 } else if (rc == -ENODATA) {
1297                         struct lu_seq_range *range = &info->lti_range;
1298                         struct seq_server_site *ss = lfsck_dev_site(lfsck);
1299
1300                         rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
1301                         if (rc)
1302                                 GOTO(unlock, rc);
1303
1304                         fld_range_set_mdt(range);
1305                         rc = fld_server_lookup(env, ss->ss_server_fld,
1306                                        fid_seq(lfsck_dto2fid(parent)), range);
1307                         if (rc)
1308                                 GOTO(unlock, rc);
1309
1310                         parent_lmv_lost = true;
1311                         lmv->lmv_magic = LMV_MAGIC;
1312                         lmv->lmv_master_mdt_index = range->lsr_index;
1313                         lmv->lmv_layout_version++;
1314                         lfsck_lmv_header_cpu_to_le(lmv, lmv);
1315                 } else {
1316                         GOTO(unlock, rc);
1317                 }
1318         }
1319
1320         if (unlikely(!dt_try_as_dir(env, parent)))
1321                 GOTO(unlock, rc = -ENOTDIR);
1322
1323         th = dt_trans_create(env, dev);
1324         if (IS_ERR(th))
1325                 GOTO(unlock, rc = PTR_ERR(th));
1326
1327         rec->rec_type = lfsck_object_type(child) & S_IFMT;
1328         rec->rec_fid = cfid;
1329         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1330                                (const struct dt_key *)lname->ln_name, th);
1331         if (rc != 0)
1332                 GOTO(stop, rc);
1333
1334         if (S_ISDIR(rec->rec_type)) {
1335                 rc = dt_declare_ref_add(env, parent, th);
1336                 if (rc != 0)
1337                         GOTO(stop, rc);
1338         }
1339
1340         if (parent_lmv_lost) {
1341                 rc = dt_declare_xattr_set(env, parent, &buf, XATTR_NAME_LMV,
1342                                           0, th);
1343                 if (rc)
1344                         GOTO(stop, rc);
1345         }
1346
1347         la->la_ctime = ktime_get_real_seconds();
1348         la->la_valid = LA_CTIME;
1349         rc = dt_declare_attr_set(env, parent, la, th);
1350         if (rc != 0)
1351                 GOTO(stop, rc);
1352
1353         rc = dt_declare_attr_set(env, child, la, th);
1354         if (rc != 0)
1355                 GOTO(stop, rc);
1356
1357         rc = dt_trans_start_local(env, dev, th);
1358         if (rc != 0)
1359                 GOTO(stop, rc);
1360
1361         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1362                        (const struct dt_key *)lname->ln_name, th);
1363         if (rc != 0)
1364                 GOTO(stop, rc);
1365
1366         if (S_ISDIR(rec->rec_type)) {
1367                 dt_write_lock(env, parent, 0);
1368                 rc = dt_ref_add(env, parent, th);
1369                 dt_write_unlock(env, parent);
1370                 if (rc != 0)
1371                         GOTO(stop, rc);
1372         }
1373
1374         if (parent_lmv_lost) {
1375                 rc = dt_xattr_set(env, parent, &buf, XATTR_NAME_LMV, 0, th);
1376                 if (rc)
1377                         GOTO(stop, rc);
1378         }
1379
1380         rc = dt_attr_set(env, parent, la, th);
1381         if (rc != 0)
1382                 GOTO(stop, rc);
1383
1384         rc = dt_attr_set(env, child, la, th);
1385
1386         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1387
1388 stop:
1389         dt_trans_stop(env, dev, th);
1390
1391 unlock:
1392         if (parent_lmv_deleted)
1393                 lfsck_lmv_set(env, lfsck, parent, lmv);
1394
1395         lfsck_unlock(llh);
1396
1397 log:
1398         CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1399                "the name %s and type %o to the parent "DFID": rc = %d\n",
1400                lfsck_lfsck2name(lfsck), PFID(cfid), lname->ln_name,
1401                lfsck_object_type(child) & S_IFMT, PFID(pfid), rc);
1402
1403         if (rc != 0) {
1404                 struct lfsck_namespace *ns = com->lc_file_ram;
1405
1406                 ns->ln_flags |= LF_INCONSISTENT;
1407                 if (rc > 0)
1408                         ns->ln_lost_dirent_repaired++;
1409         }
1410
1411         return rc;
1412 }
1413
1414 /**
1415  * Create the specified orphan directory.
1416  *
1417  * For the case that the parent MDT-object stored in some MDT-object's
1418  * linkEA entry is lost, the LFSCK will re-create the parent object as
1419  * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1420  * with the name ${FID}-P-${conflict_version}.
1421  *
1422  * \param[in] env       pointer to the thread context
1423  * \param[in] com       pointer to the lfsck component
1424  * \param[in] orphan    pointer to the orphan MDT-object to be created
1425  * \param[in] lmv       pointer to master LMV EA that will be set to the orphan
1426  *
1427  * \retval              positive number for repaired cases
1428  * \retval              negative error number on failure
1429  */
1430 static int lfsck_namespace_create_orphan_dir(const struct lu_env *env,
1431                                              struct lfsck_component *com,
1432                                              struct dt_object *orphan,
1433                                              struct lmv_mds_md_v1 *lmv)
1434 {
1435         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1436         struct lu_attr                  *la     = &info->lti_la;
1437         struct dt_allocation_hint       *hint   = &info->lti_hint;
1438         struct dt_object_format         *dof    = &info->lti_dof;
1439         struct lu_name                  *cname  = &info->lti_name2;
1440         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1441         struct lmv_mds_md_v1            *lmv2   = &info->lti_lmv2;
1442         const struct lu_fid             *cfid   = lfsck_dto2fid(orphan);
1443         struct lu_fid                    tfid;
1444         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1445         struct lfsck_namespace          *ns     = com->lc_file_ram;
1446         struct dt_device                *dev    = lfsck_obj2dev(orphan);
1447         struct dt_object                *parent = NULL;
1448         struct thandle                  *th     = NULL;
1449         struct lfsck_lock_handle        *llh    = &info->lti_llh;
1450         struct linkea_data               ldata  = { NULL };
1451         struct lu_buf                    linkea_buf;
1452         struct lu_buf                    lmv_buf;
1453         char                             name[32];
1454         int                              namelen;
1455         int                              idx    = 0;
1456         int                              rc     = 0;
1457         int                              rc1    = 0;
1458         ENTRY;
1459
1460         LASSERT(!dt_object_exists(orphan));
1461
1462         cname->ln_name = NULL;
1463         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1464                 GOTO(log, rc = 1);
1465
1466         if (dt_object_remote(orphan)) {
1467                 LASSERT(lfsck->li_lpf_root_obj != NULL);
1468
1469                 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, cfid);
1470                 if (idx < 0)
1471                         GOTO(log, rc = idx);
1472
1473                 snprintf(name, 8, "MDT%04x", idx);
1474                 rc = dt_lookup_dir(env, lfsck->li_lpf_root_obj, name, &tfid);
1475                 if (rc != 0)
1476                         GOTO(log, rc = (rc == -ENOENT ? -ENXIO : rc));
1477
1478                 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
1479                 if (IS_ERR(parent))
1480                         GOTO(log, rc = PTR_ERR(parent));
1481
1482                 if (unlikely(!dt_try_as_dir(env, parent)))
1483                         GOTO(log, rc = -ENOTDIR);
1484         } else {
1485                 if (unlikely(lfsck->li_lpf_obj == NULL))
1486                         GOTO(log, rc = -ENXIO);
1487
1488                 parent = lfsck->li_lpf_obj;
1489         }
1490
1491         dev = lfsck_find_dev_by_fid(env, lfsck, cfid);
1492         if (IS_ERR(dev))
1493                 GOTO(log, rc = PTR_ERR(dev));
1494
1495         idx = 0;
1496
1497 again:
1498         do {
1499                 namelen = snprintf(name, 31, DFID"-P-%d",
1500                                    PFID(cfid), idx++);
1501                 rc = dt_lookup_dir(env, parent, name, &tfid);
1502                 if (rc != 0 && rc != -ENOENT)
1503                         GOTO(log, rc);
1504         } while (rc == 0);
1505
1506         rc = lfsck_lock(env, lfsck, parent, name, llh,
1507                         MDS_INODELOCK_UPDATE, LCK_PW);
1508         if (rc != 0)
1509                 GOTO(log, rc);
1510
1511         /* Re-check whether the name conflict with othrs after taken
1512          * the ldlm lock. */
1513         rc = dt_lookup_dir(env, parent, name, &tfid);
1514         if (unlikely(rc == 0)) {
1515                 lfsck_unlock(llh);
1516                 goto again;
1517         }
1518
1519         if (rc != -ENOENT)
1520                 GOTO(unlock1, rc);
1521
1522         cname->ln_name = name;
1523         cname->ln_namelen = namelen;
1524
1525         memset(la, 0, sizeof(*la));
1526         la->la_mode = S_IFDIR | 0700;
1527         la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1528                        LA_ATIME | LA_MTIME | LA_CTIME;
1529
1530         orphan->do_ops->do_ah_init(env, hint, parent, orphan,
1531                                    la->la_mode & S_IFMT);
1532
1533         memset(dof, 0, sizeof(*dof));
1534         dof->dof_type = dt_mode_to_dft(S_IFDIR);
1535
1536         rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
1537                               cname, lfsck_dto2fid(parent));
1538         if (rc != 0)
1539                 GOTO(unlock1, rc);
1540
1541         th = dt_trans_create(env, dev);
1542         if (IS_ERR(th))
1543                 GOTO(unlock1, rc = PTR_ERR(th));
1544
1545         /* Sync the remote transaction to guarantee that the subsequent
1546          * lock against the @orphan can find the @orphan in time. */
1547         if (dt_object_remote(orphan))
1548                 th->th_sync = 1;
1549
1550         rc = dt_declare_create(env, orphan, la, hint, dof, th);
1551         if (rc != 0)
1552                 GOTO(stop, rc);
1553
1554         if (unlikely(!dt_try_as_dir(env, orphan)))
1555                 GOTO(stop, rc = -ENOTDIR);
1556
1557         rc = dt_declare_ref_add(env, orphan, th);
1558         if (rc != 0)
1559                 GOTO(stop, rc);
1560
1561         rec->rec_type = S_IFDIR;
1562         rec->rec_fid = cfid;
1563         rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1564                                (const struct dt_key *)dot, th);
1565         if (rc != 0)
1566                 GOTO(stop, rc);
1567
1568         rec->rec_fid = lfsck_dto2fid(parent);
1569         rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
1570                                (const struct dt_key *)dotdot, th);
1571         if (rc != 0)
1572                 GOTO(stop, rc);
1573
1574         if (lmv != NULL) {
1575                 lmv->lmv_magic = LMV_MAGIC;
1576                 lmv->lmv_master_mdt_index = lfsck_dev_idx(lfsck);
1577                 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1578                 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1579                 rc = dt_declare_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV,
1580                                           0, th);
1581                 if (rc != 0)
1582                         GOTO(stop, rc);
1583         }
1584
1585         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1586                        ldata.ld_leh->leh_len);
1587         rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
1588                                   XATTR_NAME_LINK, 0, th);
1589         if (rc != 0)
1590                 GOTO(stop, rc);
1591
1592         rec->rec_fid = cfid;
1593         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1594                                (const struct dt_key *)name, th);
1595         if (rc == 0)
1596                 rc = dt_declare_ref_add(env, parent, th);
1597
1598         if (rc != 0)
1599                 GOTO(stop, rc);
1600
1601         rc = dt_trans_start_local(env, dev, th);
1602         if (rc != 0)
1603                 GOTO(stop, rc);
1604
1605         dt_write_lock(env, orphan, 0);
1606         rc = dt_create(env, orphan, la, hint, dof, th);
1607         if (rc != 0)
1608                 GOTO(unlock2, rc);
1609
1610         rc = dt_ref_add(env, orphan, th);
1611         if (rc != 0)
1612                 GOTO(unlock2, rc);
1613
1614         rec->rec_fid = cfid;
1615         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1616                        (const struct dt_key *)dot, th);
1617         if (rc != 0)
1618                 GOTO(unlock2, rc);
1619
1620         rec->rec_fid = lfsck_dto2fid(parent);
1621         rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
1622                        (const struct dt_key *)dotdot, th);
1623         if (rc != 0)
1624                 GOTO(unlock2, rc);
1625
1626         if (lmv != NULL) {
1627                 rc = dt_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV, 0, th);
1628                 if (rc != 0)
1629                         GOTO(unlock2, rc);
1630         }
1631
1632         rc = dt_xattr_set(env, orphan, &linkea_buf,
1633                           XATTR_NAME_LINK, 0, th);
1634         dt_write_unlock(env, orphan);
1635         if (rc != 0)
1636                 GOTO(stop, rc);
1637
1638         rec->rec_fid = cfid;
1639         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1640                        (const struct dt_key *)name, th);
1641         if (rc == 0) {
1642                 dt_write_lock(env, parent, 0);
1643                 rc = dt_ref_add(env, parent, th);
1644                 dt_write_unlock(env, parent);
1645         }
1646
1647         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1648
1649 unlock2:
1650         dt_write_unlock(env, orphan);
1651
1652 stop:
1653         rc1 = dt_trans_stop(env, dev, th);
1654         if (rc1 != 0 && rc > 0)
1655                 rc = rc1;
1656
1657 unlock1:
1658         lfsck_unlock(llh);
1659
1660 log:
1661         CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan dir for "
1662                "the object "DFID", name = %s: rc = %d\n",
1663                lfsck_lfsck2name(lfsck), PFID(cfid),
1664                cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
1665
1666         if (parent != NULL && !IS_ERR(parent) && parent != lfsck->li_lpf_obj)
1667                 lfsck_object_put(env, parent);
1668
1669         if (rc != 0)
1670                 ns->ln_flags |= LF_INCONSISTENT;
1671
1672         return rc;
1673 }
1674
1675 /**
1676  * Remove the specified entry from the linkEA.
1677  *
1678  * Locate the linkEA entry with the given @cname and @pfid, then
1679  * remove this entry or the other entries those are repeated with
1680  * this entry.
1681  *
1682  * \param[in] env       pointer to the thread context
1683  * \param[in] com       pointer to the lfsck component
1684  * \param[in] obj       pointer to the dt_object to be handled
1685  * \param[in,out]ldata  pointer to the buffer that holds the linkEA
1686  * \param[in] cname     the name for the child in the parent directory
1687  * \param[in] pfid      the parent directory's FID for the linkEA
1688  * \param[in] next      if true, then remove the first found linkEA
1689  *                      entry, and move the ldata->ld_lee to next entry
1690  *
1691  * \retval              positive number for repaired cases
1692  * \retval              0 if nothing to be repaired
1693  * \retval              negative error number on failure
1694  */
1695 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1696                                          struct lfsck_component *com,
1697                                          struct dt_object *obj,
1698                                          struct linkea_data *ldata,
1699                                          struct lu_name *cname,
1700                                          struct lu_fid *pfid,
1701                                          bool next)
1702 {
1703         struct lfsck_instance           *lfsck     = com->lc_lfsck;
1704         struct dt_device                *dev       = lfsck_obj2dev(obj);
1705         struct lfsck_bookmark           *bk        = &lfsck->li_bookmark_ram;
1706         struct thandle                  *th        = NULL;
1707         struct lustre_handle             lh        = { 0 };
1708         struct linkea_data               ldata_new = { NULL };
1709         struct lu_buf                    linkea_buf;
1710         int                              buflen    = 0;
1711         int                              rc        = 0;
1712         ENTRY;
1713
1714         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1715                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
1716                               LCK_EX);
1717         if (rc != 0)
1718                 GOTO(log, rc);
1719
1720         if (next)
1721                 lfsck_linkea_del_buf(ldata, cname);
1722         else
1723                 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1724                                                     true);
1725         if (ldata->ld_leh->leh_reccount > 0 ||
1726             unlikely(ldata->ld_leh->leh_overflow_time)) {
1727                 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1728                                ldata->ld_leh->leh_len);
1729                 buflen = linkea_buf.lb_len;
1730         }
1731
1732 again:
1733         th = dt_trans_create(env, dev);
1734         if (IS_ERR(th))
1735                 GOTO(unlock1, rc = PTR_ERR(th));
1736
1737         if (buflen != 0)
1738                 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1739                                           XATTR_NAME_LINK, 0, th);
1740         else
1741                 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
1742         if (rc != 0)
1743                 GOTO(stop, rc);
1744
1745         rc = dt_trans_start_local(env, dev, th);
1746         if (rc != 0)
1747                 GOTO(stop, rc);
1748
1749         dt_write_lock(env, obj, 0);
1750         if (unlikely(lfsck_is_dead_obj(obj)))
1751                 GOTO(unlock2, rc = -ENOENT);
1752
1753         rc = lfsck_links_read2_with_rec(env, obj, &ldata_new);
1754         if (rc)
1755                 GOTO(unlock2, rc = (rc == -ENODATA ? 0 : rc));
1756
1757         /* The specified linkEA entry has been removed by race. */
1758         rc = linkea_links_find(&ldata_new, cname, pfid);
1759         if (rc != 0)
1760                 GOTO(unlock2, rc = 0);
1761
1762         if (bk->lb_param & LPF_DRYRUN)
1763                 GOTO(unlock2, rc = 1);
1764
1765         if (next)
1766                 lfsck_linkea_del_buf(&ldata_new, cname);
1767         else
1768                 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1769                                                     true);
1770
1771         /*
1772          * linkea may change because it doesn't take lock in the first read, if
1773          * it becomes larger, restart from beginning.
1774          */
1775         if ((ldata_new.ld_leh->leh_reccount > 0 ||
1776              unlikely(ldata_new.ld_leh->leh_overflow_time)) &&
1777             buflen < ldata_new.ld_leh->leh_len) {
1778                 dt_write_unlock(env, obj);
1779                 dt_trans_stop(env, dev, th);
1780                 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1781                                ldata_new.ld_leh->leh_len);
1782                 buflen = linkea_buf.lb_len;
1783                 goto again;
1784         }
1785
1786         if (buflen)
1787                 rc = lfsck_links_write(env, obj, &ldata_new, th);
1788         else
1789                 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
1790
1791         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1792
1793 unlock2:
1794         dt_write_unlock(env, obj);
1795
1796 stop:
1797         dt_trans_stop(env, dev, th);
1798
1799 unlock1:
1800         lfsck_ibits_unlock(&lh, LCK_EX);
1801
1802 log:
1803         CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1804                "for the object: "DFID", parent "DFID", name %.*s\n",
1805                lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1806                PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1807                cname->ln_name);
1808
1809         if (rc != 0) {
1810                 struct lfsck_namespace *ns = com->lc_file_ram;
1811
1812                 ns->ln_flags |= LF_INCONSISTENT;
1813         }
1814
1815         return rc;
1816 }
1817
1818 /**
1819  * Conditionally remove the specified entry from the linkEA.
1820  *
1821  * Take the parent lock firstly, then check whether the specified
1822  * name entry exists or not: if yes, do nothing; otherwise, call
1823  * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1824  *
1825  * \param[in] env       pointer to the thread context
1826  * \param[in] com       pointer to the lfsck component
1827  * \param[in] parent    pointer to the parent directory
1828  * \param[in] child     pointer to the child object that holds the linkEA
1829  * \param[in,out]ldata  pointer to the buffer that holds the linkEA
1830  * \param[in] cname     the name for the child in the parent directory
1831  * \param[in] pfid      the parent directory's FID for the linkEA
1832  *
1833  * \retval              positive number for repaired cases
1834  * \retval              0 if nothing to be repaired
1835  * \retval              negative error number on failure
1836  */
1837 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1838                                               struct lfsck_component *com,
1839                                               struct dt_object *parent,
1840                                               struct dt_object *child,
1841                                               struct linkea_data *ldata,
1842                                               struct lu_name *cname,
1843                                               struct lu_fid *pfid)
1844 {
1845         struct lfsck_thread_info *info  = lfsck_env_info(env);
1846         struct lu_fid            *cfid  = &info->lti_fid3;
1847         struct lfsck_lock_handle *llh   = &info->lti_llh;
1848         int                       rc;
1849         ENTRY;
1850
1851         rc = lfsck_lock(env, com->lc_lfsck, parent, cname->ln_name, llh,
1852                         MDS_INODELOCK_UPDATE, LCK_PR);
1853         if (rc != 0)
1854                 RETURN(rc);
1855
1856         dt_read_lock(env, parent, 0);
1857         if (unlikely(lfsck_is_dead_obj(parent))) {
1858                 dt_read_unlock(env, parent);
1859                 lfsck_unlock(llh);
1860                 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1861                                                    cname, pfid, true);
1862
1863                 RETURN(rc);
1864         }
1865
1866         rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1867                        (const struct dt_key *)cname->ln_name);
1868         dt_read_unlock(env, parent);
1869
1870         /* It is safe to release the ldlm lock, because when the logic come
1871          * here, we have got all the needed information above whether the
1872          * linkEA entry is valid or not. It is not important that others
1873          * may add new linkEA entry after the ldlm lock released. If other
1874          * has removed the specified linkEA entry by race, then it is OK,
1875          * because the subsequent lfsck_namespace_shrink_linkea() can handle
1876          * such case. */
1877         lfsck_unlock(llh);
1878         if (rc == -ENOENT) {
1879                 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1880                                                    cname, pfid, true);
1881
1882                 RETURN(rc);
1883         }
1884
1885         if (rc != 0)
1886                 RETURN(rc);
1887
1888         /* The LFSCK just found some internal status of cross-MDTs
1889          * create operation. That is normal. */
1890         if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1891                 linkea_next_entry(ldata);
1892
1893                 RETURN(0);
1894         }
1895
1896         rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1897                                            pfid, true);
1898
1899         RETURN(rc);
1900 }
1901
1902 /**
1903  * Conditionally replace name entry in the parent.
1904  *
1905  * As required, the LFSCK may re-create the lost MDT-object for dangling
1906  * name entry, but such repairing may be wrong because of bad FID in the
1907  * name entry. As the LFSCK processing, the real MDT-object may be found,
1908  * then the LFSCK should check whether the former re-created MDT-object
1909  * has been modified or not, if not, then destroy it and update the name
1910  * entry in the parent to reference the real MDT-object.
1911  *
1912  * \param[in] env       pointer to the thread context
1913  * \param[in] com       pointer to the lfsck component
1914  * \param[in] parent    pointer to the parent directory
1915  * \param[in] child     pointer to the MDT-object that may be the real
1916  *                      MDT-object corresponding to the name entry in parent
1917  * \param[in] cfid      the current FID in the name entry
1918  * \param[in] cname     contains the name of the child in the parent directory
1919  *
1920  * \retval              positive number for repaired cases
1921  * \retval              0 if nothing to be repaired
1922  * \retval              negative error number on failure
1923  */
1924 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1925                                         struct lfsck_component *com,
1926                                         struct dt_object *parent,
1927                                         struct dt_object *child,
1928                                         const struct lu_fid *cfid,
1929                                         const struct lu_name *cname)
1930 {
1931         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1932         struct lu_attr                  *la     = &info->lti_la;
1933         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
1934         struct lu_fid                    tfid;
1935         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1936         /* The child and its name may be on different MDTs. */
1937         struct dt_device                *dev    = lfsck->li_next;
1938         const char                      *name   = cname->ln_name;
1939         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
1940         struct dt_object                *cobj   = NULL;
1941         struct lfsck_lock_handle        *pllh   = &info->lti_llh;
1942         struct lustre_handle             clh    = { 0 };
1943         struct linkea_data               ldata  = { NULL };
1944         struct thandle                  *th     = NULL;
1945         bool                             exist  = true;
1946         int                              rc     = 0;
1947         ENTRY;
1948
1949         /* @parent/@child may be based on lfsck->li_bottom,
1950          * but here we need the object based on the lfsck->li_next. */
1951
1952         parent = lfsck_object_locate(dev, parent);
1953         if (IS_ERR(parent))
1954                 GOTO(log, rc = PTR_ERR(parent));
1955
1956         if (unlikely(!dt_try_as_dir(env, parent)))
1957                 GOTO(log, rc = -ENOTDIR);
1958
1959         rc = lfsck_lock(env, lfsck, parent, name, pllh,
1960                         MDS_INODELOCK_UPDATE, LCK_PW);
1961         if (rc != 0)
1962                 GOTO(log, rc);
1963
1964         if (!fid_is_sane(cfid)) {
1965                 exist = false;
1966                 goto replace;
1967         }
1968
1969         cobj = lfsck_object_find_by_dev(env, dev, cfid);
1970         if (IS_ERR(cobj)) {
1971                 rc = PTR_ERR(cobj);
1972                 if (rc == -ENOENT) {
1973                         exist = false;
1974                         goto replace;
1975                 }
1976
1977                 GOTO(log, rc);
1978         }
1979
1980         if (!dt_object_exists(cobj)) {
1981                 exist = false;
1982                 goto replace;
1983         }
1984
1985         rc = dt_lookup_dir(env, parent, name, &tfid);
1986         if (rc == -ENOENT) {
1987                 exist = false;
1988                 goto replace;
1989         }
1990
1991         if (rc != 0)
1992                 GOTO(log, rc);
1993
1994         /* Someone changed the name entry, cannot replace it. */
1995         if (!lu_fid_eq(cfid, &tfid))
1996                 GOTO(log, rc = 0);
1997
1998         /* lock the object to be destroyed. */
1999         rc = lfsck_ibits_lock(env, lfsck, cobj, &clh,
2000                               MDS_INODELOCK_UPDATE |
2001                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_XATTR,
2002                               LCK_EX);
2003         if (rc != 0)
2004                 GOTO(log, rc);
2005
2006         if (unlikely(lfsck_is_dead_obj(cobj))) {
2007                 exist = false;
2008                 goto replace;
2009         }
2010
2011         rc = dt_attr_get(env, cobj, la);
2012         if (rc != 0)
2013                 GOTO(log, rc);
2014
2015         /* The object has been modified by other(s), or it is not created by
2016          * LFSCK, the two cases are indistinguishable. So cannot replace it. */
2017         if (la->la_ctime != 0)
2018                 GOTO(log, rc);
2019
2020         if (S_ISREG(la->la_mode)) {
2021                 rc = dt_xattr_get(env, cobj, &LU_BUF_NULL, XATTR_NAME_LOV);
2022                 /* If someone has created related OST-object(s),
2023                  * then keep it. */
2024                 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
2025                         GOTO(log, rc = (rc > 0 ? 0 : rc));
2026         }
2027
2028 replace:
2029         dt_read_lock(env, child, 0);
2030         rc = lfsck_links_read2_with_rec(env, child, &ldata);
2031         dt_read_unlock(env, child);
2032
2033         /* Someone changed the child, no need to replace. */
2034         if (rc == -ENODATA)
2035                 GOTO(log, rc = 0);
2036
2037         if (rc != 0)
2038                 GOTO(log, rc);
2039
2040         rc = linkea_links_find(&ldata, cname, pfid);
2041         /* Someone moved the child, no need to replace. */
2042         if (rc != 0)
2043                 GOTO(log, rc = 0);
2044
2045         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2046                 GOTO(log, rc = 1);
2047
2048         th = dt_trans_create(env, dev);
2049         if (IS_ERR(th))
2050                 GOTO(log, rc = PTR_ERR(th));
2051
2052         if (exist) {
2053                 rc = dt_declare_destroy(env, cobj, th);
2054                 if (rc != 0)
2055                         GOTO(stop, rc);
2056         }
2057
2058         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2059         if (rc != 0)
2060                 GOTO(stop, rc);
2061
2062         rec->rec_type = S_IFDIR;
2063         rec->rec_fid = lfsck_dto2fid(child);
2064         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
2065                                (const struct dt_key *)name, th);
2066         if (rc != 0)
2067                 GOTO(stop, rc);
2068
2069         rc = dt_trans_start_local(env, dev, th);
2070         if (rc != 0)
2071                 GOTO(stop, rc);
2072
2073         if (exist) {
2074                 rc = dt_destroy(env, cobj, th);
2075                 if (rc != 0)
2076                         GOTO(stop, rc);
2077         }
2078
2079         /* The old name entry maybe not exist. */
2080         rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2081         if (rc != 0 && rc != -ENOENT)
2082                 GOTO(stop, rc);
2083
2084         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
2085                        (const struct dt_key *)name, th);
2086
2087         GOTO(stop, rc = (rc == 0 ? 1 : rc));
2088
2089 stop:
2090         dt_trans_stop(env, dev, th);
2091
2092 log:
2093         lfsck_ibits_unlock(&clh, LCK_EX);
2094         lfsck_unlock(pllh);
2095
2096         if (cobj != NULL && !IS_ERR(cobj))
2097                 lfsck_object_put(env, cobj);
2098
2099         CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
2100                "object "DFID" because of conflict with the object "DFID
2101                " under the parent "DFID" with name %s: rc = %d\n",
2102                lfsck_lfsck2name(lfsck), PFID(cfid),
2103                PFID(lfsck_dto2fid(child)), PFID(pfid), name, rc);
2104
2105         return rc;
2106 }
2107
2108 /**
2109  * Overwrite the linkEA for the object with the given ldata.
2110  *
2111  * The caller should take the ldlm lock before the calling.
2112  *
2113  * \param[in] env       pointer to the thread context
2114  * \param[in] com       pointer to the lfsck component
2115  * \param[in] obj       pointer to the dt_object to be handled
2116  * \param[in] ldata     pointer to the new linkEA data
2117  *
2118  * \retval              positive number for repaired cases
2119  * \retval              0 if nothing to be repaired
2120  * \retval              negative error number on failure
2121  */
2122 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
2123                                    struct lfsck_component *com,
2124                                    struct dt_object *obj,
2125                                    struct linkea_data *ldata)
2126 {
2127         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2128         struct dt_device                *dev    = lfsck_obj2dev(obj);
2129         struct thandle                  *th     = NULL;
2130         struct lu_buf                    linkea_buf;
2131         int                              rc     = 0;
2132         ENTRY;
2133
2134         th = dt_trans_create(env, dev);
2135         if (IS_ERR(th))
2136                 GOTO(log, rc = PTR_ERR(th));
2137
2138         lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
2139                        ldata->ld_leh->leh_len);
2140         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2141                                   XATTR_NAME_LINK, 0, th);
2142         if (rc != 0)
2143                 GOTO(stop, rc);
2144
2145         rc = dt_trans_start_local(env, dev, th);
2146         if (rc != 0)
2147                 GOTO(stop, rc);
2148
2149         dt_write_lock(env, obj, 0);
2150         if (unlikely(lfsck_is_dead_obj(obj)))
2151                 GOTO(unlock, rc = 0);
2152
2153         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2154                 GOTO(unlock, rc = 1);
2155
2156         rc = dt_xattr_set(env, obj, &linkea_buf,
2157                           XATTR_NAME_LINK, 0, th);
2158
2159         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2160
2161 unlock:
2162         dt_write_unlock(env, obj);
2163
2164 stop:
2165         dt_trans_stop(env, dev, th);
2166
2167 log:
2168         CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
2169                "object "DFID": rc = %d\n",
2170                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
2171
2172         if (rc != 0) {
2173                 struct lfsck_namespace *ns = com->lc_file_ram;
2174
2175                 ns->ln_flags |= LF_INCONSISTENT;
2176         }
2177
2178         return rc;
2179 }
2180
2181 /**
2182  * Repair invalid name entry.
2183  *
2184  * If the name entry contains invalid information, such as bad file type
2185  * or (and) corrupted object FID, then either remove the name entry or
2186  * udpate the name entry with the given (right) information.
2187  *
2188  * \param[in] env       pointer to the thread context
2189  * \param[in] com       pointer to the lfsck component
2190  * \param[in] parent    pointer to the parent directory
2191  * \param[in] child     pointer to the object referenced by the name entry
2192  * \param[in] name      the old name of the child under the parent directory
2193  * \param[in] name2     the new name of the child under the parent directory
2194  * \param[in] type      the type claimed by the name entry
2195  * \param[in] update    update the name entry if true; otherwise, remove it
2196  * \param[in] dec       decrease the parent nlink count if true
2197  *
2198  * \retval              positive number for repaired successfully
2199  * \retval              0 if nothing to be repaired
2200  * \retval              negative error number on failure
2201  */
2202 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2203                                   struct lfsck_component *com,
2204                                   struct dt_object *parent,
2205                                   struct dt_object *child,
2206                                   const char *name, const char *name2,
2207                                   __u16 type, bool update, bool dec)
2208 {
2209         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2210         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
2211         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
2212         struct lu_fid                   cfid    = {0};
2213         struct lu_fid                    tfid;
2214         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2215         struct dt_device                *dev    = lfsck->li_next;
2216         struct thandle                  *th     = NULL;
2217         struct lfsck_lock_handle        *llh    = &info->lti_llh;
2218         struct lustre_handle             lh     = { 0 };
2219         int                              rc     = 0;
2220         ENTRY;
2221
2222         if (child)
2223                 cfid = *lfsck_dto2fid(child);
2224         parent = lfsck_object_locate(dev, parent);
2225         if (IS_ERR(parent))
2226                 GOTO(log, rc = PTR_ERR(parent));
2227
2228         if (unlikely(!dt_try_as_dir(env, parent)))
2229                 GOTO(log, rc = -ENOTDIR);
2230
2231         if (!update || strcmp(name, name2) == 0)
2232                 rc = lfsck_lock(env, lfsck, parent, name, llh,
2233                                 MDS_INODELOCK_UPDATE, LCK_PW);
2234         else
2235                 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2236                                       MDS_INODELOCK_UPDATE, LCK_PW);
2237         if (rc != 0)
2238                 GOTO(log, rc);
2239
2240         th = dt_trans_create(env, dev);
2241         if (IS_ERR(th))
2242                 GOTO(unlock1, rc = PTR_ERR(th));
2243
2244         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2245         if (rc != 0)
2246                 GOTO(stop, rc);
2247
2248         if (update) {
2249                 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2250                 LASSERT(!fid_is_zero(&cfid));
2251                 rec->rec_fid = &cfid;
2252                 rc = dt_declare_insert(env, parent,
2253                                        (const struct dt_rec *)rec,
2254                                        (const struct dt_key *)name2, th);
2255                 if (rc != 0)
2256                         GOTO(stop, rc);
2257         }
2258
2259         if (dec && S_ISDIR(type)) {
2260                 rc = dt_declare_ref_del(env, parent, th);
2261                 if (rc != 0)
2262                         GOTO(stop, rc);
2263         }
2264
2265         rc = dt_trans_start_local(env, dev, th);
2266         if (rc != 0)
2267                 GOTO(stop, rc);
2268
2269
2270         dt_write_lock(env, parent, 0);
2271         rc = dt_lookup_dir(env, dt_object_child(parent), name, &tfid);
2272         /* Someone has removed the bad name entry by race. */
2273         if (rc == -ENOENT)
2274                 GOTO(unlock2, rc = 0);
2275
2276         if (rc != 0)
2277                 GOTO(unlock2, rc);
2278
2279         /* Someone has removed the bad name entry and reused it for other
2280          * object by race. */
2281         if (!lu_fid_eq(&tfid, &cfid))
2282                 GOTO(unlock2, rc = 0);
2283
2284         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2285                 GOTO(unlock2, rc = 1);
2286
2287         rc = dt_delete(env, parent, (const struct dt_key *)name, th);
2288         if (rc != 0)
2289                 GOTO(unlock2, rc);
2290
2291         if (update) {
2292                 rc = dt_insert(env, parent,
2293                                (const struct dt_rec *)rec,
2294                                (const struct dt_key *)name2, th);
2295                 if (rc != 0)
2296                         GOTO(unlock2, rc);
2297         }
2298
2299         if (dec && S_ISDIR(type)) {
2300                 rc = dt_ref_del(env, parent, th);
2301                 if (rc != 0)
2302                         GOTO(unlock2, rc);
2303         }
2304
2305         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2306
2307 unlock2:
2308         dt_write_unlock(env, parent);
2309
2310 stop:
2311         dt_trans_stop(env, dev, th);
2312
2313         /* We are not sure whether the child will become orphan or not.
2314          * Record it in the LFSCK trace file for further checking in
2315          * the second-stage scanning. */
2316         if (!update && !dec && child && rc == 0)
2317                 lfsck_namespace_trace_update(env, com, &cfid,
2318                                              LNTF_CHECK_LINKEA, true);
2319
2320 unlock1:
2321         /* It is harmless even if unlock the unused lock_handle */
2322         lfsck_ibits_unlock(&lh, LCK_PW);
2323         lfsck_unlock(llh);
2324
2325 log:
2326         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2327                "entry for: parent "DFID", child "DFID", name %s, type "
2328                "in name entry %o, type claimed by child %o. repair it "
2329                "by %s with new name2 %s: rc = %d\n",
2330                lfsck_lfsck2name(lfsck), PFID(pfid), PFID(&cfid),
2331                name, type, update ? lfsck_object_type(child) : 0,
2332                update ? "updating" : "removing", name2, rc);
2333
2334         if (rc != 0) {
2335                 struct lfsck_namespace *ns = com->lc_file_ram;
2336
2337                 ns->ln_flags |= LF_INCONSISTENT;
2338         }
2339
2340         return rc;
2341 }
2342
2343 /**
2344  * Update the ".." name entry for the given object.
2345  *
2346  * The object's ".." is corrupted, this function will update the ".." name
2347  * entry with the given pfid, and the linkEA with the given ldata.
2348  *
2349  * The caller should take the ldlm lock before the calling.
2350  *
2351  * \param[in] env       pointer to the thread context
2352  * \param[in] com       pointer to the lfsck component
2353  * \param[in] obj       pointer to the dt_object to be handled
2354  * \param[in] pfid      the new fid for the object's ".." name entry
2355  * \param[in] cname     the name for the @obj in the parent directory
2356  *
2357  * \retval              positive number for repaired cases
2358  * \retval              0 if nothing to be repaired
2359  * \retval              negative error number on failure
2360  */
2361 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2362                                                   struct lfsck_component *com,
2363                                                   struct dt_object *obj,
2364                                                   const struct lu_fid *pfid,
2365                                                   struct lu_name *cname)
2366 {
2367         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2368         struct dt_insert_rec            *rec    = &info->lti_dt_rec;
2369         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2370         struct dt_device                *dev    = lfsck_obj2dev(obj);
2371         struct thandle                  *th     = NULL;
2372         struct linkea_data               ldata  = { NULL };
2373         struct lu_buf                    linkea_buf;
2374         int                              rc     = 0;
2375         ENTRY;
2376
2377         LASSERT(!dt_object_remote(obj));
2378         LASSERT(S_ISDIR(lfsck_object_type(obj)));
2379
2380         rc = linkea_links_new(&ldata, &info->lti_big_buf, cname, pfid);
2381         if (rc != 0)
2382                 GOTO(log, rc);
2383
2384         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2385                        ldata.ld_leh->leh_len);
2386
2387         th = dt_trans_create(env, dev);
2388         if (IS_ERR(th))
2389                 GOTO(log, rc = PTR_ERR(th));
2390
2391         rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2392         if (rc != 0)
2393                 GOTO(stop, rc);
2394
2395         rec->rec_type = S_IFDIR;
2396         rec->rec_fid = pfid;
2397         rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2398                                (const struct dt_key *)dotdot, th);
2399         if (rc != 0)
2400                 GOTO(stop, rc);
2401
2402         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2403                                   XATTR_NAME_LINK, 0, th);
2404         if (rc != 0)
2405                 GOTO(stop, rc);
2406
2407         rc = dt_trans_start_local(env, dev, th);
2408         if (rc != 0)
2409                 GOTO(stop, rc);
2410
2411         dt_write_lock(env, obj, 0);
2412         if (unlikely(lfsck_is_dead_obj(obj)))
2413                 GOTO(unlock, rc = 0);
2414
2415         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2416                 GOTO(unlock, rc = 1);
2417
2418         /* The old ".." name entry maybe not exist. */
2419         dt_delete(env, obj, (const struct dt_key *)dotdot, th);
2420
2421         rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2422                        (const struct dt_key *)dotdot, th);
2423         if (rc != 0)
2424                 GOTO(unlock, rc);
2425
2426         rc = lfsck_links_write(env, obj, &ldata, th);
2427
2428         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2429
2430 unlock:
2431         dt_write_unlock(env, obj);
2432
2433 stop:
2434         dt_trans_stop(env, dev, th);
2435
2436 log:
2437         CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2438                "the object "DFID", new parent "DFID": rc = %d\n",
2439                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2440                PFID(pfid), rc);
2441
2442         if (rc != 0) {
2443                 struct lfsck_namespace *ns = com->lc_file_ram;
2444
2445                 ns->ln_flags |= LF_INCONSISTENT;
2446         }
2447
2448         return rc;
2449 }
2450
2451 /**
2452  * Handle orphan @obj during Double Scan Directory.
2453  *
2454  * Remove the @obj's current (invalid) linkEA entries, and insert
2455  * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2456  * ${FID}-${PFID}-D-${conflict_version}
2457  *
2458  * The caller should take the ldlm lock before the calling.
2459  *
2460  * \param[in] env       pointer to the thread context
2461  * \param[in] com       pointer to the lfsck component
2462  * \param[in] obj       pointer to the orphan object to be handled
2463  * \param[in] pfid      the new fid for the object's ".." name entry
2464  * \param[in,out] lh    ldlm lock handler for the given @obj
2465  * \param[out] type     to tell the caller what the inconsistency is
2466  *
2467  * \retval              positive number for repaired cases
2468  * \retval              0 if nothing to be repaired
2469  * \retval              negative error number on failure
2470  */
2471 static int
2472 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2473                            struct lfsck_component *com,
2474                            struct dt_object *obj,
2475                            const struct lu_fid *pfid,
2476                            struct lustre_handle *lh,
2477                            enum lfsck_namespace_inconsistency_type *type)
2478 {
2479         struct lfsck_thread_info *info = lfsck_env_info(env);
2480         struct lfsck_namespace   *ns   = com->lc_file_ram;
2481         int                       rc;
2482         ENTRY;
2483
2484         /* Remove the unrecognized linkEA. */
2485         rc = lfsck_namespace_links_remove(env, com, obj);
2486         lfsck_ibits_unlock(lh, LCK_EX);
2487         if (rc < 0 && rc != -ENODATA)
2488                 RETURN(rc);
2489
2490         *type = LNIT_MUL_REF;
2491
2492         /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2493          * ever tried to verify some remote MDT-object that resides on this
2494          * MDT, but this MDT failed to respond such request. So means there
2495          * may be some remote name entry on other MDT that references this
2496          * object with another name, so we cannot know whether this linkEA
2497          * is valid or not. So keep it there and maybe resolved when next
2498          * LFSCK run. */
2499         if (ns->ln_flags & LF_INCOMPLETE)
2500                 RETURN(0);
2501
2502         /* The unique linkEA is invalid, even if the ".." name entry may be
2503          * valid, we still cannot know via which name entry this directory
2504          * will be referenced. Then handle it as pure orphan. */
2505         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2506                  "-"DFID, PFID(pfid));
2507         rc = lfsck_namespace_insert_orphan(env, com, obj,
2508                                            info->lti_tmpbuf, "D", NULL);
2509
2510         RETURN(rc);
2511 }
2512
2513 /**
2514  * Double Scan Directory object for single linkEA entry case.
2515  *
2516  * The given @child has unique linkEA entry. If the linkEA entry is valid,
2517  * then check whether the name is in the namespace or not, if not, add the
2518  * missing name entry back to namespace. If the linkEA entry is invalid,
2519  * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2520  * as an orphan.
2521  *
2522  * \param[in] env       pointer to the thread context
2523  * \param[in] com       pointer to the lfsck component
2524  * \param[in] child     pointer to the directory to be double scanned
2525  * \param[in] pfid      the FID corresponding to the ".." entry
2526  * \param[in] ldata     pointer to the linkEA data for the given @child
2527  * \param[in,out] lh    ldlm lock handler for the given @child
2528  * \param[out] type     to tell the caller what the inconsistency is
2529  * \param[in] retry     if found inconsistency, but the caller does not hold
2530  *                      ldlm lock on the @child, then set @retry as true
2531  * \param[in] unknown   set if does not know how to repair the inconsistency
2532  *
2533  * \retval              positive number for repaired cases
2534  * \retval              0 if nothing to be repaired
2535  * \retval              negative error number on failure
2536  */
2537 static int
2538 lfsck_namespace_dsd_single(const struct lu_env *env,
2539                            struct lfsck_component *com,
2540                            struct dt_object *child,
2541                            const struct lu_fid *pfid,
2542                            struct linkea_data *ldata,
2543                            struct lustre_handle *lh,
2544                            enum lfsck_namespace_inconsistency_type *type,
2545                            bool *retry, bool *unknown)
2546 {
2547         struct lfsck_thread_info *info          = lfsck_env_info(env);
2548         struct lu_name           *cname         = &info->lti_name;
2549         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2550         struct lu_fid             tfid;
2551         struct lfsck_namespace   *ns            = com->lc_file_ram;
2552         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2553         struct dt_object         *parent        = NULL;
2554         struct lmv_mds_md_v1     *lmv;
2555         int                       rc            = 0;
2556         ENTRY;
2557
2558         rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2559                                                  info->lti_key,
2560                                                  sizeof(info->lti_key));
2561         /* The unique linkEA entry with bad parent will be handled as orphan. */
2562         if (rc != 0) {
2563                 if (!lustre_handle_is_used(lh) && retry != NULL)
2564                         *retry = true;
2565                 else
2566                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2567                                                         pfid, lh, type);
2568
2569                 GOTO(out, rc);
2570         }
2571
2572         parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2573         if (IS_ERR(parent))
2574                 GOTO(out, rc = PTR_ERR(parent));
2575
2576         /* We trust the unique linkEA entry in spite of whether it matches the
2577          * ".." name entry or not. Because even if the linkEA entry is wrong
2578          * and the ".." name entry is right, we still cannot know via which
2579          * name entry the child will be referenced, since all known entries
2580          * have been verified during the first-stage scanning. */
2581         if (!dt_object_exists(parent)) {
2582                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2583                  * has ever tried to verify some remote MDT-object that resides
2584                  * on this MDT, but this MDT failed to respond such request. So
2585                  * means there may be some remote name entry on other MDT that
2586                  * references this object with another name, so we cannot know
2587                  * whether this linkEA is valid or not. So keep it there and
2588                  * maybe resolved when next LFSCK run. */
2589                 if (ns->ln_flags & LF_INCOMPLETE)
2590                         GOTO(out, rc = 0);
2591
2592                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2593                         *retry = true;
2594
2595                         GOTO(out, rc = 0);
2596                 }
2597
2598                 lfsck_ibits_unlock(lh, LCK_EX);
2599
2600 lost_parent:
2601                 lmv = &info->lti_lmv;
2602                 rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
2603                 if (rc != 0 && rc != -ENODATA)
2604                         GOTO(out, rc);
2605
2606                 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2607                         lmv = NULL;
2608                 } else if (lfsck_shard_name_to_index(env,
2609                                         cname->ln_name, cname->ln_namelen,
2610                                         S_IFDIR, cfid) < 0) {
2611                         /* It is an invalid name entry, we
2612                          * cannot trust the parent also. */
2613                         rc = lfsck_namespace_shrink_linkea(env, com, child,
2614                                                 ldata, cname, &tfid, true);
2615                         if (rc < 0)
2616                                 GOTO(out, rc);
2617
2618                         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2619                                  "-"DFID, PFID(pfid));
2620                         rc = lfsck_namespace_insert_orphan(env, com, child,
2621                                                 info->lti_tmpbuf, "S", NULL);
2622
2623                         GOTO(out, rc);
2624                 }
2625
2626                 /* Create the lost parent as an orphan. */
2627                 rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
2628                 if (rc >= 0) {
2629                         /* Add the missing name entry to the parent. */
2630                         rc = lfsck_namespace_insert_normal(env, com, parent,
2631                                                            child, cname);
2632                         if (unlikely(rc == -EEXIST)) {
2633                                 /* Unfortunately, someone reused the name
2634                                  * under the parent by race. So we have
2635                                  * to remove the linkEA entry from
2636                                  * current child object. It means that the
2637                                  * LFSCK cannot recover the system
2638                                  * totally back to its original status,
2639                                  * but it is necessary to make the
2640                                  * current system to be consistent. */
2641                                 rc = lfsck_namespace_shrink_linkea(env,
2642                                                 com, child, ldata,
2643                                                 cname, &tfid, true);
2644                                 if (rc >= 0) {
2645                                         snprintf(info->lti_tmpbuf,
2646                                                  sizeof(info->lti_tmpbuf),
2647                                                  "-"DFID, PFID(pfid));
2648                                         rc = lfsck_namespace_insert_orphan(env,
2649                                                 com, child, info->lti_tmpbuf,
2650                                                 "D", NULL);
2651                                 }
2652                         }
2653                 }
2654
2655                 GOTO(out, rc);
2656         } /* !dt_object_exists(parent) */
2657
2658         /* The unique linkEA entry with bad parent will be handled as orphan. */
2659         if (unlikely(!dt_try_as_dir(env, parent))) {
2660                 if (!lustre_handle_is_used(lh) && retry != NULL)
2661                         *retry = true;
2662                 else
2663                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2664                                                         pfid, lh, type);
2665
2666                 GOTO(out, rc);
2667         }
2668
2669         rc = dt_lookup_dir(env, parent, cname->ln_name, &tfid);
2670         if (rc == -ENOENT) {
2671                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2672                  * has ever tried to verify some remote MDT-object that resides
2673                  * on this MDT, but this MDT failed to respond such request. So
2674                  * means there may be some remote name entry on other MDT that
2675                  * references this object with another name, so we cannot know
2676                  * whether this linkEA is valid or not. So keep it there and
2677                  * maybe resolved when next LFSCK run. */
2678                 if (ns->ln_flags & LF_INCOMPLETE)
2679                         GOTO(out, rc = 0);
2680
2681                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2682                         *retry = true;
2683
2684                         GOTO(out, rc = 0);
2685                 }
2686
2687                 lfsck_ibits_unlock(lh, LCK_EX);
2688                 rc = lfsck_namespace_check_name(env, lfsck, parent, child,
2689                                                 cname);
2690                 if (rc == -ENOENT)
2691                         goto lost_parent;
2692
2693                 if (rc < 0)
2694                         GOTO(out, rc);
2695
2696                 /* It is an invalid name entry, drop it. */
2697                 if (unlikely(rc > 0)) {
2698                         rc = lfsck_namespace_shrink_linkea(env, com, child,
2699                                                 ldata, cname, &tfid, true);
2700                         if (rc >= 0) {
2701                                 snprintf(info->lti_tmpbuf,
2702                                          sizeof(info->lti_tmpbuf),
2703                                          "-"DFID, PFID(pfid));
2704                                 rc = lfsck_namespace_insert_orphan(env, com,
2705                                         child, info->lti_tmpbuf, "D", NULL);
2706                         }
2707
2708                         GOTO(out, rc);
2709                 }
2710
2711                 /* Add the missing name entry back to the namespace. */
2712                 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2713                                                    cname);
2714                 if (unlikely(rc == -ESTALE))
2715                         /* It may happen when the remote object has been
2716                          * removed, but the local MDT is not aware of that. */
2717                         goto lost_parent;
2718
2719                 if (unlikely(rc == -EEXIST)) {
2720                         /* Unfortunately, someone reused the name under the
2721                          * parent by race. So we have to remove the linkEA
2722                          * entry from current child object. It means that the
2723                          * LFSCK cannot recover the system totally back to
2724                          * its original status, but it is necessary to make
2725                          * the current system to be consistent.
2726                          *
2727                          * It also may be because of the LFSCK found some
2728                          * internal status of create operation. Under such
2729                          * case, nothing to be done. */
2730                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
2731                                         parent, child, ldata, cname, &tfid);
2732                         if (rc >= 0) {
2733                                 snprintf(info->lti_tmpbuf,
2734                                          sizeof(info->lti_tmpbuf),
2735                                          "-"DFID, PFID(pfid));
2736                                 rc = lfsck_namespace_insert_orphan(env, com,
2737                                         child, info->lti_tmpbuf, "D", NULL);
2738                         }
2739                 }
2740
2741                 GOTO(out, rc);
2742         } /* rc == -ENOENT */
2743
2744         if (rc != 0)
2745                 GOTO(out, rc);
2746
2747         if (!lu_fid_eq(&tfid, cfid)) {
2748                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2749                         *retry = true;
2750
2751                         GOTO(out, rc = 0);
2752                 }
2753
2754                 lfsck_ibits_unlock(lh, LCK_EX);
2755                 /* The name entry references another MDT-object that
2756                  * may be created by the LFSCK for repairing dangling
2757                  * name entry. Try to replace it. */
2758                 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2759                                                   &tfid, cname);
2760                 if (rc == 0)
2761                         rc = lfsck_namespace_dsd_orphan(env, com, child,
2762                                                         pfid, lh, type);
2763
2764                 GOTO(out, rc);
2765         }
2766
2767         /* Zero FID may because the remote directroy object has invalid linkEA,
2768          * or lost linkEA. Under such case, the LFSCK on this MDT does not know
2769          * how to repair the inconsistency, but the namespace LFSCK on the MDT
2770          * where its name entry resides may has more information (name, FID) to
2771          * repair such inconsistency. So here, keep the inconsistency to avoid
2772          * some imporper repairing. */
2773         if (fid_is_zero(pfid)) {
2774                 if (unknown)
2775                         *unknown = true;
2776
2777                 GOTO(out, rc = 0);
2778         }
2779
2780         /* The ".." name entry is wrong, update it. */
2781         if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2782                 if (!lustre_handle_is_used(lh) && retry != NULL) {
2783                         *retry = true;
2784
2785                         GOTO(out, rc = 0);
2786                 }
2787
2788                 *type = LNIT_UNMATCHED_PAIRS;
2789                 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2790                                                 lfsck_dto2fid(parent), cname);
2791         }
2792
2793         GOTO(out, rc);
2794
2795 out:
2796         if (parent != NULL && !IS_ERR(parent))
2797                 lfsck_object_put(env, parent);
2798
2799         return rc;
2800 }
2801
2802 /**
2803  * Double Scan Directory object for multiple linkEA entries case.
2804  *
2805  * The given @child has multiple linkEA entries. There is at most one linkEA
2806  * entry will be valid, all the others will be removed. Firstly, the function
2807  * will try to find out the linkEA entry for which the name entry exists under
2808  * the given parent (@pfid). If there is no linkEA entry that matches the given
2809  * ".." name entry, then tries to find out the first linkEA entry that both the
2810  * parent and the name entry exist to rebuild a new ".." name entry.
2811  *
2812  * \param[in] env       pointer to the thread context
2813  * \param[in] com       pointer to the lfsck component
2814  * \param[in] child     pointer to the directory to be double scanned
2815  * \param[in] pfid      the FID corresponding to the ".." entry
2816  * \param[in] ldata     pointer to the linkEA data for the given @child
2817  * \param[in,out] lh    ldlm lock handler for the given @child
2818  * \param[out] type     to tell the caller what the inconsistency is
2819  * \param[in] lpf       true if the ".." entry is under lost+found/MDTxxxx/
2820  * \param[in] unknown   set if does not know how to repair the inconsistency
2821  *
2822  * \retval              positive number for repaired cases
2823  * \retval              0 if nothing to be repaired
2824  * \retval              negative error number on failure
2825  */
2826 static int
2827 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2828                              struct lfsck_component *com,
2829                              struct dt_object *child,
2830                              const struct lu_fid *pfid,
2831                              struct linkea_data *ldata,
2832                              struct lustre_handle *lh,
2833                              enum lfsck_namespace_inconsistency_type *type,
2834                              bool lpf, bool *unknown)
2835 {
2836         struct lfsck_thread_info *info          = lfsck_env_info(env);
2837         struct lu_name           *cname         = &info->lti_name;
2838         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
2839         struct lu_fid            *pfid2         = &info->lti_fid3;
2840         struct lu_fid             tfid;
2841         struct lfsck_namespace   *ns            = com->lc_file_ram;
2842         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2843         struct lfsck_bookmark    *bk            = &lfsck->li_bookmark_ram;
2844         struct dt_object         *parent        = NULL;
2845         struct linkea_data        ldata_new     = { NULL };
2846         int                       dirent_count  = 0;
2847         int                       rc            = 0;
2848         bool                      once          = true;
2849         ENTRY;
2850
2851 again:
2852         while (ldata->ld_lee != NULL) {
2853                 rc = lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
2854                                                          info->lti_key,
2855                                                          sizeof(info->lti_key));
2856                 /* Drop invalid linkEA entry. */
2857                 if (rc != 0) {
2858                         lfsck_linkea_del_buf(ldata, cname);
2859                         continue;
2860                 }
2861
2862                 /* Drop repeated linkEA entries. */
2863                 lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
2864
2865                 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2866                  * then it is possible that: the directry object has ever
2867                  * been lost, but its name entry was there. In the former
2868                  * LFSCK run, during the first-stage scanning, the LFSCK
2869                  * found the dangling name entry, but it did not recreate
2870                  * the lost object, and when moved to the second-stage
2871                  * scanning, some children objects of the lost directory
2872                  * object were found, then the LFSCK recreated such lost
2873                  * directory object as an orphan.
2874                  *
2875                  * When the LFSCK runs again, if the dangling name is still
2876                  * there, the LFSCK should move the orphan directory object
2877                  * back to the normal namespace. */
2878                 if (!lpf && !fid_is_zero(pfid) &&
2879                     !lu_fid_eq(pfid, &tfid) && once) {
2880                         linkea_next_entry(ldata);
2881                         continue;
2882                 }
2883
2884                 parent = lfsck_object_find_bottom(env, lfsck, &tfid);
2885                 if (IS_ERR(parent))
2886                         RETURN(PTR_ERR(parent));
2887
2888                 if (!dt_object_exists(parent)) {
2889                         lfsck_object_put(env, parent);
2890                         if (ldata->ld_leh->leh_reccount > 1) {
2891                                 /* If it is NOT the last linkEA entry, then
2892                                  * there is still other chance to make the
2893                                  * child to be visible via other parent, then
2894                                  * remove this linkEA entry. */
2895                                 lfsck_linkea_del_buf(ldata, cname);
2896                                 continue;
2897                         }
2898
2899                         break;
2900                 }
2901
2902                 /* The linkEA entry with bad parent will be removed. */
2903                 if (unlikely(!dt_try_as_dir(env, parent))) {
2904                         lfsck_object_put(env, parent);
2905                         lfsck_linkea_del_buf(ldata, cname);
2906                         continue;
2907                 }
2908
2909                 rc = dt_lookup_dir(env, parent, cname->ln_name, &tfid);
2910                 *pfid2 = *lfsck_dto2fid(parent);
2911                 if (rc == -ENOENT) {
2912                         lfsck_object_put(env, parent);
2913                         linkea_next_entry(ldata);
2914                         continue;
2915                 }
2916
2917                 if (rc != 0) {
2918                         lfsck_object_put(env, parent);
2919
2920                         RETURN(rc);
2921                 }
2922
2923                 if (lu_fid_eq(&tfid, cfid)) {
2924                         lfsck_object_put(env, parent);
2925                         /* If the parent (that is declared via linkEA entry)
2926                          * directory contains the specified child, but such
2927                          * parent does not match the dotdot name entry, then
2928                          * trust the linkEA. */
2929                         if (!fid_is_zero(pfid) && !lu_fid_eq(pfid, pfid2)) {
2930                                 *type = LNIT_UNMATCHED_PAIRS;
2931                                 rc = lfsck_namespace_repair_unmatched_pairs(env,
2932                                                 com, child, pfid2, cname);
2933
2934                                 RETURN(rc);
2935                         }
2936
2937 rebuild:
2938                         /* It is the most common case that we find the
2939                          * name entry corresponding to the linkEA entry
2940                          * that matches the ".." name entry. */
2941                         rc = linkea_links_new(&ldata_new, &info->lti_big_buf,
2942                                               cname, pfid2);
2943                         if (rc != 0)
2944                                 RETURN(rc);
2945
2946                         rc = lfsck_namespace_rebuild_linkea(env, com, child,
2947                                                             &ldata_new);
2948                         if (rc < 0)
2949                                 RETURN(rc);
2950
2951                         lfsck_linkea_del_buf(ldata, cname);
2952                         linkea_first_entry(ldata);
2953                         /* There may be some invalid dangling name entries under
2954                          * other parent directories, remove all of them. */
2955                         while (ldata->ld_lee != NULL) {
2956                                 rc = lfsck_namespace_unpack_linkea_entry(ldata,
2957                                                 cname, &tfid, info->lti_key,
2958                                                 sizeof(info->lti_key));
2959                                 if (rc != 0)
2960                                         goto next;
2961
2962                                 parent = lfsck_object_find_bottom(env, lfsck,
2963                                                                   &tfid);
2964                                 if (IS_ERR(parent)) {
2965                                         rc = PTR_ERR(parent);
2966                                         if (rc != -ENOENT &&
2967                                             bk->lb_param & LPF_FAILOUT)
2968                                                 RETURN(rc);
2969
2970                                         goto next;
2971                                 }
2972
2973                                 if (!dt_object_exists(parent)) {
2974                                         lfsck_object_put(env, parent);
2975                                         goto next;
2976                                 }
2977
2978                                 rc = lfsck_namespace_repair_dirent(env, com,
2979                                         parent, child, cname->ln_name,
2980                                         cname->ln_name, S_IFDIR, false, true);
2981                                 lfsck_object_put(env, parent);
2982                                 if (rc < 0) {
2983                                         if (bk->lb_param & LPF_FAILOUT)
2984                                                 RETURN(rc);
2985
2986                                         goto next;
2987                                 }
2988
2989                                 dirent_count += rc;
2990
2991 next:
2992                                 lfsck_linkea_del_buf(ldata, cname);
2993                         }
2994
2995                         ns->ln_dirent_repaired += dirent_count;
2996
2997                         RETURN(rc);
2998                 } /* lu_fid_eq(&tfid, lfsck_dto2fid(child)) */
2999
3000                 lfsck_ibits_unlock(lh, LCK_EX);
3001                 /* The name entry references another MDT-object that may be
3002                  * created by the LFSCK for repairing dangling name entry.
3003                  * Try to replace it. */
3004                 rc = lfsck_namespace_replace_cond(env, com, parent, child,
3005                                                   &tfid, cname);
3006                 lfsck_object_put(env, parent);
3007                 if (rc < 0)
3008                         RETURN(rc);
3009
3010                 if (rc > 0)
3011                         goto rebuild;
3012
3013                 lfsck_linkea_del_buf(ldata, cname);
3014         } /* while (ldata->ld_lee != NULL) */
3015
3016         /* If there is still linkEA overflow, return. */
3017         if (unlikely(ldata->ld_leh->leh_overflow_time))
3018                 RETURN(0);
3019
3020         linkea_first_entry(ldata);
3021         if (ldata->ld_leh->leh_reccount == 1) {
3022                 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
3023                                                 lh, type, NULL, unknown);
3024
3025                 RETURN(rc);
3026         }
3027
3028         /* All linkEA entries are invalid and removed, then handle the @child
3029          * as an orphan.*/
3030         if (ldata->ld_leh->leh_reccount == 0) {
3031                 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
3032                                                 type);
3033
3034                 RETURN(rc);
3035         }
3036
3037         /* If the dangling name entry for the orphan directory object has
3038          * been remvoed, then just check whether the directory object is
3039          * still under the .lustre/lost+found/MDTxxxx/ or not. */
3040         if (lpf) {
3041                 lpf = false;
3042                 goto again;
3043         }
3044
3045         /* There is no linkEA entry that matches the ".." name entry. Find
3046          * the first linkEA entry that both parent and name entry exist to
3047          * rebuild a new ".." name entry. */
3048         if (once) {
3049                 once = false;
3050                 goto again;
3051         }
3052
3053         RETURN(rc);
3054 }
3055
3056 /**
3057  * Repair the object's nlink attribute.
3058  *
3059  * If all the known name entries have been verified, then the object's hard
3060  * link attribute should match the object's linkEA entries count unless the
3061  * object's has too many hard link to be recorded in the linkEA. Such cases
3062  * should have been marked in the LFSCK trace file. Otherwise, trust the
3063  * linkEA to update the object's nlink attribute.
3064  *
3065  * \param[in] env       pointer to the thread context
3066  * \param[in] com       pointer to the lfsck component
3067  * \param[in] obj       pointer to the dt_object to be handled
3068  * \param[in,out] la    pointer to buffer to object's attribute before
3069  *                      and after the repairing
3070  *
3071  * \retval              positive number for repaired cases
3072  * \retval              0 if nothing to be repaired
3073  * \retval              negative error number on failure
3074  */
3075 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
3076                                         struct lfsck_component *com,
3077                                         struct dt_object *obj,
3078                                         struct lu_attr *la)
3079 {
3080         struct lfsck_namespace          *ns     = com->lc_file_ram;
3081         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3082         struct dt_device                *dev    = lfsck_obj2dev(obj);
3083         const struct lu_fid             *cfid   = lfsck_dto2fid(obj);
3084         struct thandle                  *th     = NULL;
3085         struct linkea_data               ldata  = { NULL };
3086         struct lustre_handle             lh     = { 0 };
3087         __u32                            old    = la->la_nlink;
3088         int                              rc     = 0;
3089         ENTRY;
3090
3091         LASSERT(!dt_object_remote(obj));
3092
3093         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3094                               MDS_INODELOCK_UPDATE, LCK_PW);
3095         if (rc != 0)
3096                 GOTO(log, rc);
3097
3098         th = dt_trans_create(env, dev);
3099         if (IS_ERR(th))
3100                 GOTO(log, rc = PTR_ERR(th));
3101
3102         la->la_valid = LA_NLINK;
3103         rc = dt_declare_attr_set(env, obj, la, th);
3104         if (rc != 0)
3105                 GOTO(stop, rc);
3106
3107         rc = dt_trans_start_local(env, dev, th);
3108         if (rc != 0)
3109                 GOTO(stop, rc);
3110
3111         dt_write_lock(env, obj, 0);
3112         /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
3113          * ever tried to verify some remote MDT-object that resides on this
3114          * MDT, but this MDT failed to respond such request. So means there
3115          * may be some remote name entry on other MDT that references this
3116          * object with another name, so we cannot know whether this linkEA
3117          * is valid or not. So keep it there and maybe resolved when next
3118          * LFSCK run. */
3119         if (ns->ln_flags & LF_INCOMPLETE)
3120                 GOTO(unlock, rc = 0);
3121
3122         rc = dt_attr_get(env, obj, la);
3123         if (rc != 0)
3124                 GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
3125
3126         rc = lfsck_links_read2_with_rec(env, obj, &ldata);
3127         if (rc)
3128                 GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
3129
3130         /* XXX: Currently, we only update the nlink attribute if the known
3131          *      linkEA entries is larger than the nlink attribute. That is
3132          *      safe action. */
3133         if (la->la_nlink >= ldata.ld_leh->leh_reccount ||
3134             unlikely(la->la_nlink == 0 ||
3135                      ldata.ld_leh->leh_overflow_time))
3136                 GOTO(unlock, rc = 0);
3137
3138         la->la_nlink = ldata.ld_leh->leh_reccount;
3139         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3140                 GOTO(unlock, rc = 1);
3141
3142         rc = dt_attr_set(env, obj, la, th);
3143
3144         GOTO(unlock, rc = (rc == 0 ? 1 : rc));
3145
3146 unlock:
3147         dt_write_unlock(env, obj);
3148
3149 stop:
3150         dt_trans_stop(env, dev, th);
3151
3152 log:
3153         lfsck_ibits_unlock(&lh, LCK_PW);
3154         CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
3155                "nlink count from %u to %u: rc = %d\n",
3156                lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc);
3157
3158         if (rc != 0)
3159                 ns->ln_flags |= LF_INCONSISTENT;
3160
3161         return rc;
3162 }
3163
3164 /**
3165  * Double scan the directory object for namespace LFSCK.
3166  *
3167  * This function will verify the <parent, child> pairs in the namespace tree:
3168  * the parent references the child via some name entry that should be in the
3169  * child's linkEA entry, the child should back references the parent via its
3170  * ".." name entry.
3171  *
3172  * The LFSCK will scan every linkEA entry in turn until find out the first
3173  * matched pairs. If found, then all other linkEA entries will be dropped.
3174  * If all the linkEA entries cannot match the ".." name entry, then there
3175  * are serveral possible cases:
3176  *
3177  * 1) If there is only one linkEA entry, then trust it as long as the PFID
3178  *    in the linkEA entry is valid.
3179  *
3180  * 2) If there are multiple linkEA entries, then try to find the linkEA
3181  *    that matches the ".." name entry. If found, then all other entries
3182  *    are invalid; otherwise, it is quite possible that the ".." name entry
3183  *    is corrupted. Under such case, the LFSCK will rebuild the ".." name
3184  *    entry according to the first valid linkEA entry (both the parent and
3185  *    the name entry should exist).
3186  *
3187  * 3) If the directory object has no (valid) linkEA entry, then the
3188  *    directory object will be handled as pure orphan and inserted
3189  *    in the .lustre/lost+found/MDTxxxx/ with the name:
3190  *    ${self_FID}-${PFID}-D-${conflict_version}
3191  *
3192  * \param[in] env       pointer to the thread context
3193  * \param[in] com       pointer to the lfsck component
3194  * \param[in] child     pointer to the directory object to be handled
3195  * \param[in] flags     to indicate the specical checking on the @child
3196  *
3197  * \retval              positive number for repaired cases
3198  * \retval              0 if nothing to be repaired
3199  * \retval              negative error number on failure
3200  */
3201 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
3202                                            struct lfsck_component *com,
3203                                            struct dt_object *child, __u8 flags)
3204 {
3205         struct lfsck_thread_info *info          = lfsck_env_info(env);
3206         const struct lu_fid      *cfid          = lfsck_dto2fid(child);
3207         struct lu_fid            *pfid          = &info->lti_fid2;
3208         struct lfsck_namespace   *ns            = com->lc_file_ram;
3209         struct lfsck_instance    *lfsck         = com->lc_lfsck;
3210         struct lustre_handle      lh            = { 0 };
3211         struct linkea_data        ldata         = { NULL };
3212         bool                      unknown       = false;
3213         bool                      lpf           = false;
3214         bool                      retry         = false;
3215         enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3216         int                       rc            = 0;
3217         ENTRY;
3218
3219         LASSERT(!dt_object_remote(child));
3220
3221         if (flags & LNTF_UNCERTAIN_LMV) {
3222                 if (flags & LNTF_RECHECK_NAME_HASH) {
3223                         rc = lfsck_namespace_scan_shard(env, com, child);
3224                         if (rc < 0)
3225                                 RETURN(rc);
3226
3227                         ns->ln_striped_shards_scanned++;
3228                 } else {
3229                         ns->ln_striped_shards_skipped++;
3230                 }
3231         }
3232
3233         flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
3234         if (flags == 0)
3235                 RETURN(0);
3236
3237         if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
3238             !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
3239                 CDEBUG(D_LFSCK,
3240                        "%s: some MDT(s) maybe NOT take part in the the namespace LFSCK, then the LFSCK cannot guarantee all the name entries have been verified in first-stage scanning. So have to skip orphan related handling for the directory object "DFID" with remote name entry\n",
3241                        lfsck_lfsck2name(lfsck), PFID(cfid));
3242
3243                 RETURN(0);
3244         }
3245
3246         if (unlikely(!dt_try_as_dir(env, child)))
3247                 GOTO(out, rc = -ENOTDIR);
3248
3249         /* We only take ldlm lock on the @child when required. When the
3250          * logic comes here for the first time, it is always false. */
3251         if (0) {
3252
3253 lock:
3254                 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
3255                                       MDS_INODELOCK_UPDATE |
3256                                       MDS_INODELOCK_XATTR, LCK_EX);
3257                 if (rc != 0)
3258                         GOTO(out, rc);
3259         }
3260
3261         dt_read_lock(env, child, 0);
3262         if (unlikely(lfsck_is_dead_obj(child))) {
3263                 dt_read_unlock(env, child);
3264
3265                 GOTO(out, rc = 0);
3266         }
3267
3268         rc = dt_lookup_dir(env, child, dotdot, pfid);
3269         if (rc != 0) {
3270                 if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
3271                         dt_read_unlock(env, child);
3272
3273                         GOTO(out, rc);
3274                 }
3275
3276                 if (!lustre_handle_is_used(&lh)) {
3277                         dt_read_unlock(env, child);
3278                         goto lock;
3279                 }
3280
3281                 fid_zero(pfid);
3282         } else if (lfsck->li_lpf_obj != NULL &&
3283                    lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
3284                 lpf = true;
3285         } else if (unlikely(!fid_is_sane(pfid))) {
3286                 fid_zero(pfid);
3287         }
3288
3289         rc = lfsck_links_read(env, child, &ldata);
3290         dt_read_unlock(env, child);
3291         if (rc != 0) {
3292                 if (rc != -ENODATA && rc != -EINVAL)
3293                         GOTO(out, rc);
3294
3295                 if (!lustre_handle_is_used(&lh))
3296                         goto lock;
3297
3298                 if (rc == -EINVAL && !fid_is_zero(pfid)) {
3299                         /* Remove the corrupted linkEA. */
3300                         rc = lfsck_namespace_links_remove(env, com, child);
3301                         if (rc == 0)
3302                                 /* Here, because of the crashed linkEA, we
3303                                  * cannot know whether there is some parent
3304                                  * that references the child directory via
3305                                  * some name entry or not. So keep it there,
3306                                  * when the LFSCK run next time, if there is
3307                                  * some parent that references this object,
3308                                  * then the LFSCK can rebuild the linkEA;
3309                                  * otherwise, this object will be handled
3310                                  * as orphan as above. */
3311                                 unknown = true;
3312                 } else {
3313                         /* 1. If we have neither ".." nor linkEA,
3314                          *    then it is an orphan.
3315                          *
3316                          * 2. If we only have the ".." name entry,
3317                          *    but no parent references this child
3318                          *    directory, then handle it as orphan. */
3319                         lfsck_ibits_unlock(&lh, LCK_EX);
3320                         type = LNIT_MUL_REF;
3321
3322                         /* If the LFSCK is marked as LF_INCOMPLETE,
3323                          * then means some MDT has ever tried to
3324                          * verify some remote MDT-object that resides
3325                          * on this MDT, but this MDT failed to respond
3326                          * such request. So means there may be some
3327                          * remote name entry on other MDT that
3328                          * references this object with another name,
3329                          * so we cannot know whether this linkEA is
3330                          * valid or not. So keep it there and maybe
3331                          * resolved when next LFSCK run. */
3332                         if (ns->ln_flags & LF_INCOMPLETE)
3333                                 GOTO(out, rc = 0);
3334
3335                         snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
3336                                  "-"DFID, PFID(pfid));
3337                         rc = lfsck_namespace_insert_orphan(env, com, child,
3338                                                 info->lti_tmpbuf, "D", NULL);
3339                 }
3340
3341                 GOTO(out, rc);
3342         } /* rc != 0 */
3343
3344         linkea_first_entry(&ldata);
3345         /* This is the most common case: the object has unique linkEA entry. */
3346         if (ldata.ld_leh->leh_reccount == 1) {
3347                 rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
3348                                                 &lh, &type, &retry, &unknown);
3349                 if (retry) {
3350                         LASSERT(!lustre_handle_is_used(&lh));
3351
3352                         retry = false;
3353                         goto lock;
3354                 }
3355
3356                 GOTO(out, rc);
3357         }
3358
3359         if (!lustre_handle_is_used(&lh))
3360                 goto lock;
3361
3362         if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
3363                 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
3364                                                 &type);
3365
3366                 GOTO(out, rc);
3367         }
3368
3369         /* When we come here, the cases usually like that:
3370          * 1) The directory object has a corrupted linkEA entry. During the
3371          *    first-stage scanning, the LFSCK cannot know such corruption,
3372          *    then it appends the right linkEA entry according to the found
3373          *    name entry after the bad one.
3374          *
3375          * 2) The directory object has a right linkEA entry. During the
3376          *    first-stage scanning, the LFSCK finds some bad name entry,
3377          *    but the LFSCK cannot aware that at that time, then it adds
3378          *    the bad linkEA entry for further processing. */
3379         rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
3380                                           &lh, &type, lpf, &unknown);
3381
3382         GOTO(out, rc);
3383
3384 out:
3385         lfsck_ibits_unlock(&lh, LCK_EX);
3386         if (rc > 0) {
3387                 switch (type) {
3388                 case LNIT_BAD_LINKEA:
3389                         ns->ln_linkea_repaired++;
3390                         break;
3391                 case LNIT_UNMATCHED_PAIRS:
3392                         ns->ln_unmatched_pairs_repaired++;
3393                         break;
3394                 case LNIT_MUL_REF:
3395                         ns->ln_mul_ref_repaired++;
3396                         break;
3397                 default:
3398                         break;
3399                 }
3400         }
3401
3402         if (unknown)
3403                 ns->ln_unknown_inconsistency++;
3404
3405         return rc;
3406 }
3407
3408 static inline bool
3409 lfsck_namespace_linkea_stale_overflow(struct linkea_data *ldata,
3410                                       struct lfsck_namespace *ns)
3411 {
3412         /* Both the leh_overflow_time and ln_time_latest_reset are
3413          * local time based, so need NOT to care about clock drift
3414          * among the servers. */
3415         return ldata->ld_leh->leh_overflow_time &&
3416                ldata->ld_leh->leh_overflow_time < ns->ln_time_latest_reset;
3417 }
3418
3419 /**
3420  * Clear the object's linkEA overflow timestamp.
3421  *
3422  * If the MDT-object has too many hard links as to the linkEA cannot hold
3423  * all of them, then overflow timestamp will be set in the linkEA header.
3424  * If some hard links are removed after that, then it is possible to hold
3425  * other missed linkEA entries. If the namespace LFSCK have added all the
3426  * related linkEA entries, then it will remove the overflow timestamp.
3427  *
3428  * \param[in] env       pointer to the thread context
3429  * \param[in] com       pointer to the lfsck component
3430  * \param[in] ldata     pointer to the linkEA data for the given @obj
3431  * \param[in] obj       pointer to the dt_object to be handled
3432  *
3433  * \retval              positive number for repaired cases
3434  * \retval              0 if nothing to be repaired
3435  * \retval              negative error number on failure
3436  */
3437 static int lfsck_namespace_linkea_clear_overflow(const struct lu_env *env,
3438                                                  struct lfsck_component *com,
3439                                                  struct linkea_data *ldata,
3440                                                  struct dt_object *obj)
3441 {
3442         struct lfsck_namespace *ns = com->lc_file_ram;
3443         struct lfsck_instance *lfsck = com->lc_lfsck;
3444         struct dt_device *dev = lfsck_obj2dev(obj);
3445         struct thandle *th = NULL;
3446         struct lustre_handle lh = { 0 };
3447         struct lu_buf linkea_buf;
3448         int rc = 0;
3449         ENTRY;
3450
3451         LASSERT(!dt_object_remote(obj));
3452
3453         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3454                               MDS_INODELOCK_UPDATE, LCK_PW);
3455         if (rc != 0)
3456                 GOTO(log, rc);
3457
3458         th = dt_trans_create(env, dev);
3459         if (IS_ERR(th))
3460                 GOTO(log, rc = PTR_ERR(th));
3461
3462         rc = dt_declare_xattr_set(env, obj,
3463                         lfsck_buf_get_const(env, NULL, MAX_LINKEA_SIZE),
3464                         XATTR_NAME_LINK, 0, th);
3465         if (rc != 0)
3466                 GOTO(stop, rc);
3467
3468         rc = dt_trans_start_local(env, dev, th);
3469         if (rc != 0)
3470                 GOTO(stop, rc);
3471
3472         dt_write_lock(env, obj, 0);
3473         rc = lfsck_links_read(env, obj, ldata);
3474         if (rc != 0)
3475                 GOTO(unlock, rc);
3476
3477         if (unlikely(!lfsck_namespace_linkea_stale_overflow(ldata, ns)))
3478                 GOTO(unlock, rc = 0);
3479
3480         ldata->ld_leh->leh_overflow_time = 0;
3481         if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3482                 GOTO(unlock, rc = 1);
3483
3484         /* If all known entries are in the linkEA, then the 'leh_reccount'
3485          * should NOT be zero. */
3486         LASSERT(ldata->ld_leh->leh_reccount > 0);
3487
3488         lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
3489                        ldata->ld_leh->leh_len);
3490         rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK, 0, th);
3491         if (unlikely(rc == -ENOSPC))
3492                 rc = 0;
3493         else if (!rc)
3494                 rc = 1;
3495
3496         GOTO(unlock, rc);
3497
3498 unlock:
3499         dt_write_unlock(env, obj);
3500
3501 stop:
3502         dt_trans_stop(env, dev, th);
3503
3504 log:
3505         lfsck_ibits_unlock(&lh, LCK_PW);
3506         CDEBUG(D_LFSCK, "%s: clear linkea overflow timestamp for the object "
3507                DFID": rc = %d\n",
3508                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
3509
3510         return rc;
3511 }
3512
3513 /**
3514  * Verify the object's agent entry.
3515  *
3516  * If the object claims to have agent entry but the linkEA does not contain
3517  * remote parent, then remove the agent entry. Otherwise, if the object has
3518  * no agent entry but its linkEA contains remote parent, then will generate
3519  * agent entry for it.
3520  *
3521  * \param[in] env       pointer to the thread context
3522  * \param[in] com       pointer to the lfsck component
3523  * \param[in] obj       pointer to the dt_object to be handled
3524  *
3525  * \retval              positive number for repaired cases
3526  * \retval              0 if nothing to be repaired
3527  * \retval              negative error number on failure
3528  */
3529 static int lfsck_namespace_check_agent_entry(const struct lu_env *env,
3530                                              struct lfsck_component *com,
3531                                              struct dt_object *obj)
3532 {
3533         struct linkea_data ldata = { NULL };
3534         struct lfsck_thread_info *info = lfsck_env_info(env);
3535         struct lfsck_namespace *ns = com->lc_file_ram;
3536         struct lfsck_instance *lfsck = com->lc_lfsck;
3537         struct lu_fid *pfid = &info->lti_fid2;
3538         struct lu_name *cname = &info->lti_name;
3539         struct lu_seq_range *range = &info->lti_range;
3540         struct seq_server_site *ss = lfsck_dev_site(lfsck);
3541         __u32 idx = lfsck_dev_idx(lfsck);
3542         int rc;
3543         bool remote = false;
3544         ENTRY;
3545
3546         if (!(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT))
3547                 RETURN(0);
3548
3549         rc = lfsck_links_read_with_rec(env, obj, &ldata);
3550         if (rc == -ENOENT || rc == -ENODATA)
3551                 RETURN(0);
3552
3553         if (rc && rc != -EINVAL)
3554                 GOTO(out, rc);
3555
3556         /* We check the agent entry again after verifying the linkEA
3557          * successfully. So invalid linkEA should be dryrun mode. */
3558         if (rc == -EINVAL || unlikely(!ldata.ld_leh->leh_reccount))
3559                 RETURN(0);
3560
3561         linkea_first_entry(&ldata);
3562         while (ldata.ld_lee != NULL && !remote) {
3563                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
3564                                     cname, pfid);
3565                 if (!linkea_entry_is_valid(&ldata, cname, pfid))
3566                         GOTO(out, rc = 0);
3567
3568                 fld_range_set_mdt(range);
3569                 rc = fld_server_lookup(env, ss->ss_server_fld,
3570                                        fid_seq(pfid), range);
3571                 if (rc)
3572                         GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
3573
3574                 if (range->lsr_index != idx)
3575                         remote = true;
3576                 else
3577                         linkea_next_entry(&ldata);
3578         }
3579
3580         if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
3581             (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
3582                 struct dt_device *dev = lfsck_obj2dev(obj);
3583                 struct linkea_data ldata2 = { NULL };
3584                 struct lustre_handle lh = { 0 };
3585                 struct lu_buf linkea_buf;
3586                 struct thandle *handle;
3587
3588                 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3589                         GOTO(out, rc = 1);
3590
3591                 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
3592                                       MDS_INODELOCK_UPDATE |
3593                                       MDS_INODELOCK_XATTR, LCK_EX);
3594                 if (rc)
3595                         GOTO(out, rc);
3596
3597                 handle = dt_trans_create(env, dev);
3598                 if (IS_ERR(handle))
3599                         GOTO(unlock, rc = PTR_ERR(handle));
3600
3601                 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
3602                                ldata.ld_leh->leh_len);
3603                 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
3604                                 XATTR_NAME_LINK, LU_XATTR_REPLACE, handle);
3605                 if (rc)
3606                         GOTO(stop, rc);
3607
3608                 rc = dt_trans_start_local(env, dev, handle);
3609                 if (rc)
3610                         GOTO(stop, rc);
3611
3612                 dt_write_lock(env, obj, 0);
3613                 rc = lfsck_links_read2_with_rec(env, obj, &ldata2);
3614                 if (rc) {
3615                         if (rc == -ENOENT || rc == -ENODATA)
3616                                 rc = 0;
3617                         GOTO(unlock2, rc);
3618                 }
3619
3620                 /* If someone changed linkEA by race, then the agent
3621                  * entry will be updated by lower layer automatically. */
3622                 if (ldata.ld_leh->leh_len != ldata2.ld_leh->leh_len ||
3623                     memcmp(ldata.ld_buf->lb_buf, ldata2.ld_buf->lb_buf,
3624                            ldata.ld_leh->leh_len) != 0)
3625                         GOTO(unlock2, rc = 0);
3626
3627                 rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK,
3628                                   LU_XATTR_REPLACE, handle);
3629                 if (!rc)
3630                         rc = 1;
3631
3632                 GOTO(unlock2, rc);
3633
3634 unlock2:
3635                 dt_write_unlock(env, obj);
3636 stop:
3637                 dt_trans_stop(env, dev, handle);
3638 unlock:
3639                 lfsck_ibits_unlock(&lh, LCK_EX);
3640         }
3641
3642         GOTO(out, rc);
3643
3644 out:
3645         if (rc > 0)
3646                 ns->ln_agent_entries_repaired++;
3647         if (rc)
3648                 CDEBUG(D_LFSCK, "%s: repair agent entry for "DFID": rc = %d\n",
3649                        lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
3650         return rc;
3651 }
3652
3653 /**
3654  * Double scan the MDT-object for namespace LFSCK.
3655  *
3656  * If the MDT-object contains invalid or repeated linkEA entries, then drop
3657  * those entries from the linkEA; if the linkEA becomes empty or the object
3658  * has no linkEA, then it is an orphan and will be added into the directory
3659  * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
3660  * the remote parent; if the name entry corresponding to some linkEA entry
3661  * is lost, then add the name entry back to the namespace.
3662  *
3663  * \param[in] env       pointer to the thread context
3664  * \param[in] com       pointer to the lfsck component
3665  * \param[in] child     pointer to the dt_object to be handled
3666  * \param[in] flags     some hints to indicate how the @child should be handled
3667  *
3668  * \retval              positive number for repaired cases
3669  * \retval              0 if nothing to be repaired
3670  * \retval              negative error number on failure
3671  */
3672 static int lfsck_namespace_double_scan_one(const struct lu_env *env,
3673                                            struct lfsck_component *com,
3674                                            struct dt_object *child, __u8 flags)
3675 {
3676         struct lfsck_thread_info *info     = lfsck_env_info(env);
3677         struct lu_attr           *la       = &info->lti_la;
3678         struct lu_name           *cname    = &info->lti_name;
3679         struct lu_fid            *pfid     = &info->lti_fid;
3680         struct lu_fid            *cfid     = &info->lti_fid2;
3681         struct lfsck_instance    *lfsck    = com->lc_lfsck;
3682         struct lfsck_namespace   *ns       = com->lc_file_ram;
3683         struct dt_object         *parent   = NULL;
3684         struct linkea_data        ldata    = { NULL };
3685         bool                      repaired = false;
3686         int                       count    = 0;
3687         int                       rc;
3688         ENTRY;
3689
3690         dt_read_lock(env, child, 0);
3691         if (unlikely(lfsck_is_dead_obj(child))) {
3692                 dt_read_unlock(env, child);
3693
3694                 RETURN(0);
3695         }
3696
3697         if (S_ISDIR(lfsck_object_type(child))) {
3698                 dt_read_unlock(env, child);
3699                 rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
3700                 if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
3701                         rc = lfsck_namespace_check_agent_entry(env, com, child);
3702
3703                 RETURN(rc);
3704         }
3705
3706         rc = lfsck_links_read(env, child, &ldata);
3707         dt_read_unlock(env, child);
3708
3709         if (rc == -EINVAL) {
3710                 struct lustre_handle lh = { 0 };
3711
3712                 rc = lfsck_ibits_lock(env, com->lc_lfsck, child, &lh,
3713                                       MDS_INODELOCK_UPDATE |
3714                                       MDS_INODELOCK_XATTR, LCK_EX);
3715                 if (rc == 0) {
3716                         rc = lfsck_namespace_links_remove(env, com, child);
3717                         lfsck_ibits_unlock(&lh, LCK_EX);
3718                 }
3719
3720                 GOTO(out, rc);
3721         }
3722
3723         if (rc != 0)
3724                 GOTO(out, rc);
3725
3726         if (!(ns->ln_flags & LF_INCOMPLETE) &&
3727             unlikely(lfsck_namespace_linkea_stale_overflow(&ldata, ns))) {
3728                 rc = lfsck_namespace_linkea_clear_overflow(env, com, &ldata,
3729                                                            child);
3730                 if (rc < 0)
3731                         GOTO(out, rc);
3732
3733                 if (rc > 0)
3734                         ns->ln_linkea_overflow_cleared++;
3735         }
3736
3737         linkea_first_entry(&ldata);
3738         while (ldata.ld_lee != NULL) {
3739                 rc = lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
3740                                                          info->lti_key,
3741                                                          sizeof(info->lti_key));
3742                 /* Invalid PFID in the linkEA entry. */
3743                 if (rc != 0) {
3744                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3745                                                 &ldata, cname, pfid, true);
3746                         if (rc < 0)
3747                                 GOTO(out, rc);
3748
3749                         if (rc > 0)
3750                                 repaired = true;
3751
3752                         continue;
3753                 }
3754
3755                 rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
3756                                                          false);
3757                 /* Found repeated linkEA entries */
3758                 if (rc > 0) {
3759                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3760                                                 &ldata, cname, pfid, false);
3761                         if (rc < 0)
3762                                 GOTO(out, rc);
3763
3764                         if (rc == 0)
3765                                 continue;
3766
3767                         repaired = true;
3768
3769                         /* fall through */
3770                 }
3771
3772                 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3773                 if (IS_ERR(parent))
3774                         GOTO(out, rc = PTR_ERR(parent));
3775
3776                 if (!dt_object_exists(parent)) {
3777
3778 lost_parent:
3779                         if (ldata.ld_leh->leh_reccount > 1) {
3780                                 /* If it is NOT the last linkEA entry, then
3781                                  * there is still other chance to make the
3782                                  * child to be visible via other parent, then
3783                                  * remove this linkEA entry. */
3784                                 rc = lfsck_namespace_shrink_linkea(env, com,
3785                                         child, &ldata, cname, pfid, true);
3786                         } else {
3787                                 /* If the LFSCK is marked as LF_INCOMPLETE,
3788                                  * then means some MDT has ever tried to
3789                                  * verify some remote MDT-object that resides
3790                                  * on this MDT, but this MDT failed to respond
3791                                  * such request. So means there may be some
3792                                  * remote name entry on other MDT that
3793                                  * references this object with another name,
3794                                  * so we cannot know whether this linkEA is
3795                                  * valid or not. So keep it there and maybe
3796                                  * resolved when next LFSCK run. */
3797                                 if (ns->ln_flags & LF_INCOMPLETE) {
3798                                         lfsck_object_put(env, parent);
3799
3800                                         GOTO(out, rc = 0);
3801                                 }
3802
3803                                 /* Create the lost parent as an orphan. */
3804                                 rc = lfsck_namespace_create_orphan_dir(env, com,
3805                                                                 parent, NULL);
3806                                 if (rc < 0) {
3807                                         lfsck_object_put(env, parent);
3808
3809                                         GOTO(out, rc);
3810                                 }
3811
3812                                 if (rc > 0)
3813                                         repaired = true;
3814
3815                                 /* Add the missing name entry to the parent. */
3816                                 rc = lfsck_namespace_insert_normal(env, com,
3817                                                         parent, child, cname);
3818                                 if (unlikely(rc == -EEXIST))
3819                                         /* Unfortunately, someone reused the
3820                                          * name under the parent by race. So we
3821                                          * have to remove the linkEA entry from
3822                                          * current child object. It means that
3823                                          * the LFSCK cannot recover the system
3824                                          * totally back to its original status,
3825                                          * but it is necessary to make the
3826                                          * current system to be consistent. */
3827                                         rc = lfsck_namespace_shrink_linkea(env,
3828                                                         com, child, &ldata,
3829                                                         cname, pfid, true);
3830                                 else
3831                                         linkea_next_entry(&ldata);
3832                         }
3833
3834                         lfsck_object_put(env, parent);
3835                         if (rc < 0)
3836                                 GOTO(out, rc);
3837
3838                         if (rc > 0)
3839                                 repaired = true;
3840
3841                         continue;
3842                 } /* !dt_object_exists(parent) */
3843
3844                 /* The linkEA entry with bad parent will be removed. */
3845                 if (unlikely(!dt_try_as_dir(env, parent))) {
3846                         lfsck_object_put(env, parent);
3847                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3848                                                 &ldata, cname, pfid, true);
3849                         if (rc < 0)
3850                                 GOTO(out, rc);
3851
3852                         if (rc > 0)
3853                                 repaired = true;
3854
3855                         continue;
3856                 }
3857
3858                 rc = dt_lookup_dir(env, parent, cname->ln_name, cfid);
3859                 if (rc != 0 && rc != -ENOENT) {
3860                         lfsck_object_put(env, parent);
3861
3862                         GOTO(out, rc);
3863                 }
3864
3865                 if (rc == 0) {
3866                         if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
3867                                 /* It is the most common case that we
3868                                  * find the name entry corresponding
3869                                  * to the linkEA entry. */
3870                                 lfsck_object_put(env, parent);
3871                                 linkea_next_entry(&ldata);
3872                         } else {
3873                                 /* The name entry references another
3874                                  * MDT-object that may be created by
3875                                  * the LFSCK for repairing dangling
3876                                  * name entry. Try to replace it. */
3877                                 rc = lfsck_namespace_replace_cond(env, com,
3878                                                 parent, child, cfid, cname);
3879                                 lfsck_object_put(env, parent);
3880                                 if (rc < 0)
3881                                         GOTO(out, rc);
3882
3883                                 if (rc > 0) {
3884                                         repaired = true;
3885                                         linkea_next_entry(&ldata);
3886                                 } else {
3887                                         rc = lfsck_namespace_shrink_linkea(env,
3888                                                         com, child, &ldata,
3889                                                         cname, pfid, true);
3890                                         if (rc < 0)
3891                                                 GOTO(out, rc);
3892
3893                                         if (rc > 0)
3894                                                 repaired = true;
3895                                 }
3896                         }
3897
3898                         continue;
3899                 }
3900
3901                 /* The following handles -ENOENT case */
3902
3903                 rc = dt_attr_get(env, child, la);
3904                 if (rc != 0)
3905                         GOTO(out, rc);
3906
3907                 /* If there is no name entry in the parent dir and the object
3908                  * link count is fewer than the linkea entries count, then the
3909                  * linkea entry should be removed. */
3910                 if (ldata.ld_leh->leh_reccount > la->la_nlink) {
3911                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
3912                                         parent, child, &ldata, cname, pfid);
3913                         lfsck_object_put(env, parent);
3914                         if (rc < 0)
3915                                 GOTO(out, rc);
3916
3917                         if (rc > 0)
3918                                 repaired = true;
3919
3920                         continue;
3921                 }
3922
3923                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3924                  * MDT has ever tried to verify some remote MDT-object that
3925                  * resides on this MDT, but this MDT failed to respond such
3926                  * request. So means there may be some remote name entry on
3927                  * other MDT that references this object with another name,
3928                  * so we cannot know whether this linkEA is valid or not.
3929                  * So keep it there and maybe resolved when next LFSCK run. */
3930                 if (ns->ln_flags & LF_INCOMPLETE) {
3931                         lfsck_object_put(env, parent);
3932
3933                         GOTO(out, rc = 0);
3934                 }
3935
3936                 rc = lfsck_namespace_check_name(env, lfsck, parent, child,
3937                                                 cname);
3938                 if (rc == -ENOENT)
3939                         goto lost_parent;
3940
3941                 if (rc < 0) {
3942                         lfsck_object_put(env, parent);
3943
3944                         GOTO(out, rc);
3945                 }
3946
3947                 /* It is an invalid name entry, drop it. */
3948                 if (unlikely(rc > 0)) {
3949                         lfsck_object_put(env, parent);
3950                         rc = lfsck_namespace_shrink_linkea(env, com, child,
3951                                                 &ldata, cname, pfid, true);
3952                         if (rc < 0)
3953                                 GOTO(out, rc);
3954
3955                         if (rc > 0)
3956                                 repaired = true;
3957
3958                         continue;
3959                 }
3960
3961                 /* Add the missing name entry back to the namespace. */
3962                 rc = lfsck_namespace_insert_normal(env, com, parent, child,
3963                                                    cname);
3964                 if (unlikely(rc == -ESTALE))
3965                         /* It may happen when the remote object has been
3966                          * removed, but the local MDT is not aware of that. */
3967                         goto lost_parent;
3968
3969                 if (unlikely(rc == -EEXIST))
3970                         /* Unfortunately, someone reused the name under the
3971                          * parent by race. So we have to remove the linkEA
3972                          * entry from current child object. It means that the
3973                          * LFSCK cannot recover the system totally back to
3974                          * its original status, but it is necessary to make
3975                          * the current system to be consistent.
3976                          *
3977                          * It also may be because of the LFSCK found some
3978                          * internal status of create operation. Under such
3979                          * case, nothing to be done. */
3980                         rc = lfsck_namespace_shrink_linkea_cond(env, com,
3981                                         parent, child, &ldata, cname, pfid);
3982                 else
3983                         linkea_next_entry(&ldata);
3984
3985                 lfsck_object_put(env, parent);
3986                 if (rc < 0)
3987                         GOTO(out, rc);
3988
3989                 if (rc > 0)
3990                         repaired = true;
3991         }
3992
3993         GOTO(out, rc = 0);
3994
3995 out:
3996         if (rc < 0 && rc != -ENODATA)
3997                 return rc;
3998
3999         if (rc == 0 && ldata.ld_leh != NULL)
4000                 count = ldata.ld_leh->leh_reccount;
4001
4002         if (count == 0) {
4003                 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
4004                  * MDT has ever tried to verify some remote MDT-object that
4005                  * resides on this MDT, but this MDT failed to respond such
4006                  * request. So means there may be some remote name entry on
4007                  * other MDT that references this object with another name,
4008                  * so we cannot know whether this linkEA is valid or not.
4009                  * So keep it there and maybe resolved when next LFSCK run. */
4010                 if (!(ns->ln_flags & LF_INCOMPLETE) &&
4011                     (ldata.ld_leh == NULL ||
4012                      !ldata.ld_leh->leh_overflow_time)) {
4013                         /* If the child becomes orphan, then insert it into
4014                          * the global .lustre/lost+found/MDTxxxx directory. */
4015                         rc = lfsck_namespace_insert_orphan(env, com, child,
4016                                                            "", "O", &count);
4017                         if (rc < 0)
4018                                 return rc;
4019
4020                         if (rc > 0) {
4021                                 ns->ln_mul_ref_repaired++;
4022                                 repaired = true;
4023                         }
4024                 }
4025         } else {
4026                 rc = dt_attr_get(env, child, la);
4027                 if (rc != 0)
4028                         return rc;
4029
4030                 if (la->la_nlink != 0 && la->la_nlink != count) {
4031                         if (unlikely(!S_ISREG(lfsck_object_type(child)) &&
4032                                      !S_ISLNK(lfsck_object_type(child)))) {
4033                                 CDEBUG(D_LFSCK, "%s: namespace LFSCK finds "
4034                                        "the object "DFID"'s nlink count %d "
4035                                        "does not match linkEA count %d, "
4036                                        "type %o, skip it.\n",
4037                                        lfsck_lfsck2name(lfsck),
4038                                        PFID(lfsck_dto2fid(child)),
4039                                        la->la_nlink, count,
4040                                        lfsck_object_type(child));
4041                         } else if (la->la_nlink < count &&
4042                                    likely(!ldata.ld_leh->leh_overflow_time)) {
4043                                 rc = lfsck_namespace_repair_nlink(env, com,
4044                                                                   child, la);
4045                                 if (rc > 0) {
4046                                         ns->ln_objs_nlink_repaired++;
4047                                         rc = 0;
4048                                 }
4049                         }
4050                 }
4051         }
4052
4053         if (repaired) {
4054                 if (la->la_nlink > 1)
4055                         ns->ln_mul_linked_repaired++;
4056
4057                 if (rc == 0)
4058                         rc = 1;
4059         }
4060
4061         if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
4062                 rc = lfsck_namespace_check_agent_entry(env, com, child);
4063
4064         return rc;
4065 }
4066
4067 static void lfsck_namespace_dump_statistics(struct seq_file *m,
4068                                             struct lfsck_namespace *ns,
4069                                             __u64 checked_phase1,
4070                                             __u64 checked_phase2,
4071                                             time64_t time_phase1,
4072                                             time64_t time_phase2, bool dryrun)
4073 {
4074         const char *postfix = dryrun ? "inconsistent" : "repaired";
4075
4076         seq_printf(m, "checked_phase1: %llu\n"
4077                    "checked_phase2: %llu\n"
4078                    "%s_phase1: %llu\n"
4079                    "%s_phase2: %llu\n"
4080                    "failed_phase1: %llu\n"
4081                    "failed_phase2: %llu\n"
4082                    "directories: %llu\n"
4083                    "dirent_%s: %llu\n"
4084                    "linkea_%s: %llu\n"
4085                    "nlinks_%s: %llu\n"
4086                    "multiple_linked_checked: %llu\n"
4087                    "multiple_linked_%s: %llu\n"
4088                    "unknown_inconsistency: %llu\n"
4089                    "unmatched_pairs_%s: %llu\n"
4090                    "dangling_%s: %llu\n"
4091                    "multiple_referenced_%s: %llu\n"
4092                    "bad_file_type_%s: %llu\n"
4093                    "lost_dirent_%s: %llu\n"
4094                    "local_lost_found_scanned: %llu\n"
4095                    "local_lost_found_moved: %llu\n"
4096                    "local_lost_found_skipped: %llu\n"
4097                    "local_lost_found_failed: %llu\n"
4098                    "striped_dirs_scanned: %llu\n"
4099                    "striped_dirs_%s: %llu\n"
4100                    "striped_dirs_failed: %llu\n"
4101                    "striped_dirs_disabled: %llu\n"
4102                    "striped_dirs_skipped: %llu\n"
4103                    "striped_shards_scanned: %llu\n"
4104                    "striped_shards_%s: %llu\n"
4105                    "striped_shards_failed: %llu\n"
4106                    "striped_shards_skipped: %llu\n"
4107                    "name_hash_%s: %llu\n"
4108                    "linkea_overflow_%s: %llu\n"
4109                    "agent_entries_%s: %llu\n"
4110                    "success_count: %u\n"
4111                    "run_time_phase1: %lld seconds\n"
4112                    "run_time_phase2: %lld seconds\n",
4113                    checked_phase1,
4114                    checked_phase2,
4115                    dryrun ? "inconsistent" : "updated",
4116                    ns->ln_items_repaired,
4117                    dryrun ? "inconsistent" : "updated",
4118                    ns->ln_objs_repaired_phase2,
4119                    ns->ln_items_failed,
4120                    ns->ln_objs_failed_phase2,
4121                    ns->ln_dirs_checked,
4122                    postfix, ns->ln_dirent_repaired,
4123                    postfix, ns->ln_linkea_repaired,
4124                    postfix, ns->ln_objs_nlink_repaired,
4125                    ns->ln_mul_linked_checked,
4126                    postfix, ns->ln_mul_linked_repaired,
4127                    ns->ln_unknown_inconsistency,
4128                    postfix, ns->ln_unmatched_pairs_repaired,
4129                    postfix, ns->ln_dangling_repaired,
4130                    postfix, ns->ln_mul_ref_repaired,
4131                    postfix, ns->ln_bad_type_repaired,
4132                    postfix, ns->ln_lost_dirent_repaired,
4133                    ns->ln_local_lpf_scanned,
4134                    ns->ln_local_lpf_moved,
4135                    ns->ln_local_lpf_skipped,
4136                    ns->ln_local_lpf_failed,
4137                    ns->ln_striped_dirs_scanned,
4138                    postfix, ns->ln_striped_dirs_repaired,
4139                    ns->ln_striped_dirs_failed,
4140                    ns->ln_striped_dirs_disabled,
4141                    ns->ln_striped_dirs_skipped,
4142                    ns->ln_striped_shards_scanned,
4143                    postfix, ns->ln_striped_shards_repaired,
4144                    ns->ln_striped_shards_failed,
4145                    ns->ln_striped_shards_skipped,
4146                    postfix, ns->ln_name_hash_repaired,
4147                    dryrun ? "inconsistent" : "cleared",
4148                    ns->ln_linkea_overflow_cleared,
4149                    postfix, ns->ln_agent_entries_repaired,
4150                    ns->ln_success_count,
4151                    time_phase1,
4152                    time_phase2);
4153 }
4154
4155 static void lfsck_namespace_release_lmv(const struct lu_env *env,
4156                                         struct lfsck_component *com)
4157 {
4158         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4159         struct lfsck_namespace          *ns     = com->lc_file_ram;
4160
4161         while (!list_empty(&lfsck->li_list_lmv)) {
4162                 struct lfsck_lmv_unit   *llu;
4163                 struct lfsck_lmv        *llmv;
4164
4165                 llu = list_entry(lfsck->li_list_lmv.next,
4166                                  struct lfsck_lmv_unit, llu_link);
4167                 llmv = &llu->llu_lmv;
4168
4169                 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
4170                          "still in using: %u\n",
4171                          atomic_read(&llmv->ll_ref));
4172
4173                 ns->ln_striped_dirs_skipped++;
4174                 lfsck_lmv_put(env, llmv);
4175         }
4176 }
4177
4178 static int lfsck_namespace_check_for_double_scan(const struct lu_env *env,
4179                                                  struct lfsck_component *com,
4180                                                  struct dt_object *obj)
4181 {
4182         struct lu_attr *la = &lfsck_env_info(env)->lti_la;
4183         int             rc;
4184
4185         rc = dt_attr_get(env, obj, la);
4186         if (rc != 0)
4187                 return rc;
4188
4189         /* zero-linkEA object may be orphan, but it also maybe because
4190          * of upgrading. Currently, we cannot record it for double scan.
4191          * Because it may cause the LFSCK trace file to be too large. */
4192
4193         /* "la_ctime" == 1 means that it has ever been removed from
4194          * backend /lost+found directory but not been added back to
4195          * the normal namespace yet. */
4196
4197         if ((S_ISREG(lfsck_object_type(obj)) && la->la_nlink > 1) ||
4198             unlikely(la->la_ctime == 1))
4199                 rc = lfsck_namespace_trace_update(env, com, lfsck_dto2fid(obj),
4200                                                   LNTF_CHECK_LINKEA, true);
4201
4202         return rc;
4203 }
4204
4205 /* namespace APIs */
4206
4207 static int lfsck_namespace_reset(const struct lu_env *env,
4208                                  struct lfsck_component *com, bool init)
4209 {
4210         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4211         struct lfsck_namespace          *ns     = com->lc_file_ram;
4212         struct lfsck_assistant_data     *lad    = com->lc_data;
4213         struct dt_object                *root;
4214         int                              rc;
4215         ENTRY;
4216
4217         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
4218         if (IS_ERR(root))
4219                 GOTO(log, rc = PTR_ERR(root));
4220
4221         if (unlikely(!dt_try_as_dir(env, root)))
4222                 GOTO(put, rc = -ENOTDIR);
4223
4224         down_write(&com->lc_sem);
4225         if (init) {
4226                 memset(ns, 0, sizeof(*ns));
4227         } else {
4228                 __u32 count = ns->ln_success_count;
4229                 time64_t last_time = ns->ln_time_last_complete;
4230
4231                 memset(ns, 0, sizeof(*ns));
4232                 ns->ln_success_count = count;
4233                 ns->ln_time_last_complete = last_time;
4234         }
4235         ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
4236         ns->ln_status = LS_INIT;
4237         ns->ln_time_latest_reset = ktime_get_real_seconds();
4238
4239         rc = lfsck_load_one_trace_file(env, com, root, &com->lc_obj,
4240                                        &dt_lfsck_namespace_features,
4241                                        LFSCK_NAMESPACE, true);
4242         if (rc)
4243                 GOTO(out, rc);
4244
4245         rc = lfsck_load_sub_trace_files(env, com, &dt_lfsck_namespace_features,
4246                                         LFSCK_NAMESPACE, true);
4247         if (rc != 0)
4248                 GOTO(out, rc);
4249
4250         clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
4251         CFS_RESET_BITMAP(lad->lad_bitmap);
4252
4253         rc = lfsck_namespace_store(env, com);
4254
4255         GOTO(out, rc);
4256
4257 out:
4258         up_write(&com->lc_sem);
4259
4260 put:
4261         lfsck_object_put(env, root);
4262 log:
4263         CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n",
4264                lfsck_lfsck2name(lfsck), rc);
4265         return rc;
4266 }
4267
4268 static void
4269 lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
4270                      bool new_checked)
4271 {
4272         struct lfsck_namespace *ns = com->lc_file_ram;
4273
4274         down_write(&com->lc_sem);
4275         if (new_checked)
4276                 com->lc_new_checked++;
4277         lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4278         up_write(&com->lc_sem);
4279 }
4280
4281 static void lfsck_namespace_close_dir(const struct lu_env *env,
4282                                       struct lfsck_component *com)
4283 {
4284         struct lfsck_namespace          *ns     = com->lc_file_ram;
4285         struct lfsck_assistant_data     *lad    = com->lc_data;
4286         struct lfsck_assistant_object   *lso    = NULL;
4287         struct lfsck_instance           *lfsck  = com->lc_lfsck;
4288         struct lfsck_lmv                *llmv   = lfsck->li_lmv;
4289         struct lfsck_namespace_req      *lnr;
4290         struct lu_attr *la = &lfsck_env_info(env)->lti_la2;
4291         __u32 size = sizeof(*lnr) + LFSCK_TMPBUF_LEN;
4292         int rc;
4293         bool wakeup = false;
4294         ENTRY;
4295
4296         if (llmv == NULL)
4297                 RETURN_EXIT;
4298
4299         rc = dt_attr_get(env, lfsck->li_obj_dir, la);
4300         if (rc)
4301                 RETURN_EXIT;
4302
4303         OBD_ALLOC(lnr, size);
4304         if (lnr == NULL) {
4305                 ns->ln_striped_dirs_skipped++;
4306
4307                 RETURN_EXIT;
4308         }
4309
4310         lso = lfsck_assistant_object_init(env, lfsck_dto2fid(lfsck->li_obj_dir),
4311                         la, lfsck->li_pos_current.lp_oit_cookie, true);
4312         if (IS_ERR(lso)) {
4313                 OBD_FREE(lnr, size);
4314                 ns->ln_striped_dirs_skipped++;
4315
4316                 RETURN_EXIT;
4317         }
4318
4319         /* Generate a dummy request to indicate that all shards' name entry
4320          * in this striped directory has been scanned for the first time. */
4321         INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
4322         lnr->lnr_lar.lar_parent = lso;
4323         lnr->lnr_lmv = lfsck_lmv_get(llmv);
4324         lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
4325         lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
4326         lnr->lnr_size = size;
4327         lnr->lnr_type = lso->lso_attr.la_mode;
4328
4329         spin_lock(&lad->lad_lock);
4330         if (lad->lad_assistant_status < 0 ||
4331             unlikely(!thread_is_running(&lfsck->li_thread) ||
4332                      !thread_is_running(&lad->lad_thread))) {
4333                 spin_unlock(&lad->lad_lock);
4334                 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4335                 ns->ln_striped_dirs_skipped++;
4336
4337                 RETURN_EXIT;
4338         }
4339
4340         list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4341         if (lad->lad_prefetched == 0)
4342                 wakeup = true;
4343
4344         lad->lad_prefetched++;
4345         spin_unlock(&lad->lad_lock);
4346         if (wakeup)
4347                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4348
4349         EXIT;
4350 }
4351
4352 static int lfsck_namespace_open_dir(const struct lu_env *env,
4353                                     struct lfsck_component *com)
4354 {
4355         struct lfsck_instance   *lfsck  = com->lc_lfsck;
4356         struct lfsck_namespace  *ns     = com->lc_file_ram;
4357         struct lfsck_lmv        *llmv   = lfsck->li_lmv;
4358         int                      rc     = 0;
4359         ENTRY;
4360
4361         if (llmv == NULL)
4362                 RETURN(0);
4363
4364         if (llmv->ll_lmv_master) {
4365                 struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
4366
4367                 if (lmv->lmv_master_mdt_index != lfsck_dev_idx(lfsck)) {
4368                         lmv->lmv_master_mdt_index =
4369                                 lfsck_dev_idx(lfsck);
4370                         ns->ln_flags |= LF_INCONSISTENT;
4371                         llmv->ll_lmv_updated = 1;
4372                 }
4373         } else {
4374                 rc = lfsck_namespace_verify_stripe_slave(env, com,
4375                                         lfsck->li_obj_dir, llmv);
4376         }
4377
4378         RETURN(rc > 0 ? 0 : rc);
4379 }
4380
4381 static int lfsck_namespace_checkpoint(const struct lu_env *env,
4382                                       struct lfsck_component *com, bool init)
4383 {
4384         struct lfsck_instance   *lfsck = com->lc_lfsck;
4385         struct lfsck_namespace  *ns    = com->lc_file_ram;
4386         int                      rc;
4387
4388         if (!init) {
4389                 rc = lfsck_checkpoint_generic(env, com);
4390                 if (rc != 0)
4391                         goto log;
4392         }
4393
4394         down_write(&com->lc_sem);
4395         if (init) {
4396                 ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
4397         } else {
4398                 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4399                 ns->ln_run_time_phase1 += ktime_get_seconds() -
4400                                           lfsck->li_time_last_checkpoint;
4401                 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
4402                 ns->ln_items_checked += com->lc_new_checked;
4403                 com->lc_new_checked = 0;
4404         }
4405
4406         rc = lfsck_namespace_store(env, com);
4407         up_write(&com->lc_sem);
4408
4409 log:
4410         CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos [%llu"
4411                ", "DFID", %#llx], status = %d: rc = %d\n",
4412                lfsck_lfsck2name(lfsck), lfsck->li_pos_current.lp_oit_cookie,
4413                PFID(&lfsck->li_pos_current.lp_dir_parent),
4414                lfsck->li_pos_current.lp_dir_cookie, ns->ln_status, rc);
4415
4416         return rc > 0 ? 0 : rc;
4417 }
4418
4419 static int lfsck_namespace_prep(const struct lu_env *env,
4420                                 struct lfsck_component *com,
4421                                 struct lfsck_start_param *lsp)
4422 {
4423         struct lfsck_instance   *lfsck  = com->lc_lfsck;
4424         struct lfsck_namespace  *ns     = com->lc_file_ram;
4425         struct lfsck_position   *pos    = &com->lc_pos_start;
4426         int                      rc;
4427
4428         rc = lfsck_namespace_load_bitmap(env, com);
4429         if (rc != 0 || ns->ln_status == LS_COMPLETED) {
4430                 rc = lfsck_namespace_reset(env, com, false);
4431                 if (rc == 0)
4432                         rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
4433
4434                 if (rc != 0) {
4435                         CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
4436                                "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
4437
4438                         return rc;
4439                 }
4440         }
4441
4442         down_write(&com->lc_sem);
4443         ns->ln_time_latest_start = ktime_get_real_seconds();
4444         spin_lock(&lfsck->li_lock);
4445
4446         if (ns->ln_flags & LF_SCANNED_ONCE) {
4447                 if (!lfsck->li_drop_dryrun ||
4448                     lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4449                         ns->ln_status = LS_SCANNING_PHASE2;
4450                         list_move_tail(&com->lc_link,
4451                                        &lfsck->li_list_double_scan);
4452                         if (!list_empty(&com->lc_link_dir))
4453                                 list_del_init(&com->lc_link_dir);
4454                         lfsck_pos_set_zero(pos);
4455                 } else {
4456                         ns->ln_status = LS_SCANNING_PHASE1;
4457                         ns->ln_run_time_phase1 = 0;
4458                         ns->ln_run_time_phase2 = 0;
4459                         ns->ln_items_checked = 0;
4460                         ns->ln_items_repaired = 0;
4461                         ns->ln_items_failed = 0;
4462                         ns->ln_dirs_checked = 0;
4463                         ns->ln_objs_checked_phase2 = 0;
4464                         ns->ln_objs_repaired_phase2 = 0;
4465                         ns->ln_objs_failed_phase2 = 0;
4466                         ns->ln_objs_nlink_repaired = 0;
4467                         ns->ln_dirent_repaired = 0;
4468                         ns->ln_linkea_repaired = 0;
4469                         ns->ln_mul_linked_checked = 0;
4470                         ns->ln_mul_linked_repaired = 0;
4471                         ns->ln_unknown_inconsistency = 0;
4472                         ns->ln_unmatched_pairs_repaired = 0;
4473                         ns->ln_dangling_repaired = 0;
4474                         ns->ln_mul_ref_repaired = 0;
4475                         ns->ln_bad_type_repaired = 0;
4476                         ns->ln_lost_dirent_repaired = 0;
4477                         ns->ln_striped_dirs_scanned = 0;
4478                         ns->ln_striped_dirs_repaired = 0;
4479                         ns->ln_striped_dirs_failed = 0;
4480                         ns->ln_striped_dirs_disabled = 0;
4481                         ns->ln_striped_dirs_skipped = 0;
4482                         ns->ln_striped_shards_scanned = 0;
4483                         ns->ln_striped_shards_repaired = 0;
4484                         ns->ln_striped_shards_failed = 0;
4485                         ns->ln_striped_shards_skipped = 0;
4486                         ns->ln_name_hash_repaired = 0;
4487                         fid_zero(&ns->ln_fid_latest_scanned_phase2);
4488                         if (list_empty(&com->lc_link_dir))
4489                                 list_add_tail(&com->lc_link_dir,
4490                                               &lfsck->li_list_dir);
4491                         *pos = ns->ln_pos_first_inconsistent;
4492                 }
4493         } else {
4494                 ns->ln_status = LS_SCANNING_PHASE1;
4495                 if (list_empty(&com->lc_link_dir))
4496                         list_add_tail(&com->lc_link_dir,
4497                                       &lfsck->li_list_dir);
4498                 if (!lfsck->li_drop_dryrun ||
4499                     lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
4500                         *pos = ns->ln_pos_last_checkpoint;
4501                         pos->lp_oit_cookie++;
4502                 } else {
4503                         *pos = ns->ln_pos_first_inconsistent;
4504                 }
4505         }
4506
4507         spin_unlock(&lfsck->li_lock);
4508         up_write(&com->lc_sem);
4509
4510         rc = lfsck_start_assistant(env, com, lsp);
4511
4512         CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos [%llu, "
4513                DFID", %#llx]: rc = %d\n",
4514                lfsck_lfsck2name(lfsck), pos->lp_oit_cookie,
4515                PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc);
4516
4517         return rc;
4518 }
4519
4520 static int lfsck_namespace_exec_oit(const struct lu_env *env,
4521                                     struct lfsck_component *com,
4522                                     struct dt_object *obj)
4523 {
4524         struct lfsck_thread_info *info = lfsck_env_info(env);
4525         struct lfsck_namespace *ns = com->lc_file_ram;
4526         struct lfsck_instance *lfsck = com->lc_lfsck;
4527         const struct lu_fid *fid = lfsck_dto2fid(obj);
4528         struct lu_fid *pfid = &info->lti_fid2;
4529         struct lu_name *cname = &info->lti_name;
4530         struct lu_seq_range *range = &info->lti_range;
4531         struct seq_server_site *ss = lfsck_dev_site(lfsck);
4532         struct linkea_data ldata = { NULL };
4533         __u32 idx = lfsck_dev_idx(lfsck);
4534         struct lu_attr la = { .la_valid = 0 };
4535         bool remote = false;
4536         int rc;
4537         ENTRY;
4538
4539         rc = dt_attr_get(env, obj, &la);
4540         if (unlikely(rc || (la.la_valid & LA_FLAGS &&
4541                             la.la_flags & LUSTRE_ORPHAN_FL))) {
4542                 CDEBUG(D_INFO,
4543                        "%s: skip orphan "DFID", %llx/%x: rc = %d\n",
4544                        lfsck_lfsck2name(lfsck), PFID(fid),
4545                        la.la_valid, la.la_flags, rc);
4546
4547                 return rc;
4548         }
4549
4550         rc = lfsck_links_read(env, obj, &ldata);
4551         if (rc == -ENOENT)
4552                 GOTO(out, rc = 0);
4553
4554         /* -EINVAL means crashed linkEA, should be verified. */
4555         if (rc == -EINVAL) {
4556                 rc = lfsck_namespace_trace_update(env, com, fid,
4557                                                   LNTF_CHECK_LINKEA, true);
4558                 if (rc == 0) {
4559                         struct lustre_handle lh = { 0 };
4560
4561                         rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4562                                               MDS_INODELOCK_UPDATE |
4563                                               MDS_INODELOCK_XATTR, LCK_EX);
4564                         if (rc == 0) {
4565                                 rc = lfsck_namespace_links_remove(env, com,
4566                                                                   obj);
4567                                 lfsck_ibits_unlock(&lh, LCK_EX);
4568                         }
4569                 }
4570
4571                 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
4572         }
4573
4574         if (rc && rc != -ENODATA)
4575                 GOTO(out, rc);
4576
4577         if (rc == -ENODATA || unlikely(!ldata.ld_leh->leh_reccount)) {
4578                 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4579
4580                 GOTO(out, rc);
4581         }
4582
4583         linkea_first_entry(&ldata);
4584         while (ldata.ld_lee != NULL) {
4585                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4586                                     cname, pfid);
4587                 if (!fid_is_sane(pfid)) {
4588                         rc = lfsck_namespace_trace_update(env, com, fid,
4589                                                   LNTF_CHECK_PARENT, true);
4590                 } else if (!linkea_entry_is_valid(&ldata, cname, pfid)) {
4591                         GOTO(out, rc);
4592                 } else {
4593                         fld_range_set_mdt(range);
4594                         rc = fld_server_lookup(env, ss->ss_server_fld,
4595                                                fid_seq(pfid), range);
4596                         if ((rc == -ENOENT) ||
4597                             (!rc && range->lsr_index != idx)) {
4598                                 remote = true;
4599                                 break;
4600                         }
4601                 }
4602                 if (rc)
4603                         GOTO(out, rc);
4604
4605                 linkea_next_entry(&ldata);
4606         }
4607
4608         if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
4609             (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
4610                 rc = lfsck_namespace_trace_update(env, com, fid,
4611                                                   LNTF_CHECK_AGENT_ENTRY, true);
4612                 if (rc)
4613                         GOTO(out, rc);
4614         }
4615
4616         /* Record multiple-linked object. */
4617         if (ldata.ld_leh->leh_reccount > 1) {
4618                 rc = lfsck_namespace_trace_update(env, com, fid,
4619                                                   LNTF_CHECK_LINKEA, true);
4620
4621                 GOTO(out, rc);
4622         }
4623
4624         if (remote)
4625                 rc = lfsck_namespace_trace_update(env, com, fid,
4626                                                   LNTF_CHECK_LINKEA, true);
4627         else
4628                 rc = lfsck_namespace_check_for_double_scan(env, com, obj);
4629
4630         GOTO(out, rc);
4631
4632 out:
4633         down_write(&com->lc_sem);
4634         if (S_ISDIR(lfsck_object_type(obj)))
4635                 ns->ln_dirs_checked++;
4636         if (rc != 0)
4637                 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4638         up_write(&com->lc_sem);
4639
4640         return rc;
4641 }
4642
4643 static int lfsck_namespace_exec_dir(const struct lu_env *env,
4644                                     struct lfsck_component *com,
4645                                     struct lfsck_assistant_object *lso,
4646                                     struct lu_dirent *ent, __u16 type)
4647 {
4648         struct lfsck_assistant_data     *lad     = com->lc_data;
4649         struct lfsck_instance           *lfsck   = com->lc_lfsck;
4650         struct lfsck_namespace_req      *lnr;
4651         struct lfsck_bookmark           *bk      = &lfsck->li_bookmark_ram;
4652         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
4653         struct ptlrpc_thread            *athread = &lad->lad_thread;
4654         bool                             wakeup  = false;
4655
4656         wait_event_idle(mthread->t_ctl_waitq,
4657                         lad->lad_prefetched < bk->lb_async_windows ||
4658                         !thread_is_running(mthread) ||
4659                         !thread_is_running(athread));
4660
4661         if (unlikely(!thread_is_running(mthread) ||
4662                      !thread_is_running(athread)))
4663                 return 0;
4664
4665         if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
4666                 return 0;
4667
4668         lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, lso, ent, type);
4669         if (IS_ERR(lnr)) {
4670                 struct lfsck_namespace *ns = com->lc_file_ram;
4671
4672                 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4673                 return PTR_ERR(lnr);
4674         }
4675
4676         spin_lock(&lad->lad_lock);
4677         if (lad->lad_assistant_status < 0 ||
4678             unlikely(!thread_is_running(mthread) ||
4679                      !thread_is_running(athread))) {
4680                 spin_unlock(&lad->lad_lock);
4681                 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4682                 return lad->lad_assistant_status;
4683         }
4684
4685         list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4686         if (lad->lad_prefetched == 0)
4687                 wakeup = true;
4688
4689         lad->lad_prefetched++;
4690         spin_unlock(&lad->lad_lock);
4691         if (wakeup)
4692                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4693
4694         down_write(&com->lc_sem);
4695         com->lc_new_checked++;
4696         up_write(&com->lc_sem);
4697
4698         return 0;
4699 }
4700
4701 static int lfsck_namespace_post(const struct lu_env *env,
4702                                 struct lfsck_component *com,
4703                                 int result, bool init)
4704 {
4705         struct lfsck_instance   *lfsck = com->lc_lfsck;
4706         struct lfsck_namespace  *ns    = com->lc_file_ram;
4707         int                      rc;
4708         ENTRY;
4709
4710         lfsck_post_generic(env, com, &result);
4711
4712         down_write(&com->lc_sem);
4713         lfsck_namespace_release_lmv(env, com);
4714
4715         spin_lock(&lfsck->li_lock);
4716         if (!init)
4717                 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4718         if (result > 0) {
4719                 ns->ln_status = LS_SCANNING_PHASE2;
4720                 ns->ln_flags |= LF_SCANNED_ONCE;
4721                 ns->ln_flags &= ~LF_UPGRADE;
4722                 list_del_init(&com->lc_link_dir);
4723                 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4724         } else if (result == 0) {
4725                 if (lfsck->li_status != 0)
4726                         ns->ln_status = lfsck->li_status;
4727                 else
4728                         ns->ln_status = LS_STOPPED;
4729                 if (ns->ln_status != LS_PAUSED) {
4730                         list_del_init(&com->lc_link_dir);
4731                         list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4732                 }
4733         } else {
4734                 ns->ln_status = LS_FAILED;
4735                 list_del_init(&com->lc_link_dir);
4736                 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4737         }
4738         spin_unlock(&lfsck->li_lock);
4739
4740         if (!init) {
4741                 ns->ln_run_time_phase1 += ktime_get_seconds() -
4742                                           lfsck->li_time_last_checkpoint;
4743                 ns->ln_time_last_checkpoint = ktime_get_real_seconds();
4744                 ns->ln_items_checked += com->lc_new_checked;
4745                 com->lc_new_checked = 0;
4746         }
4747
4748         rc = lfsck_namespace_store(env, com);
4749         up_write(&com->lc_sem);
4750
4751         CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
4752                lfsck_lfsck2name(lfsck), rc);
4753
4754         RETURN(rc);
4755 }
4756
4757 static void
4758 lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
4759                      struct seq_file *m)
4760 {
4761         struct lfsck_instance   *lfsck = com->lc_lfsck;
4762         struct lfsck_bookmark   *bk    = &lfsck->li_bookmark_ram;
4763         struct lfsck_namespace  *ns    = com->lc_file_ram;
4764
4765         down_read(&com->lc_sem);
4766         seq_printf(m, "name: lfsck_namespace\n"
4767                    "magic: %#x\n"
4768                    "version: %d\n"
4769                    "status: %s\n",
4770                    ns->ln_magic,
4771                    bk->lb_version,
4772                    lfsck_status2name(ns->ln_status));
4773
4774         lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
4775
4776         lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4777
4778         lfsck_time_dump(m, ns->ln_time_last_complete, "last_completed");
4779
4780         lfsck_time_dump(m, ns->ln_time_latest_start, "latest_start");
4781
4782         lfsck_time_dump(m, ns->ln_time_last_checkpoint, "last_checkpoint");
4783
4784         lfsck_pos_dump(m, &ns->ln_pos_latest_start, "latest_start_position");
4785
4786         lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
4787                        "last_checkpoint_position");
4788
4789         lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
4790                        "first_failure_position");
4791
4792         if (ns->ln_status == LS_SCANNING_PHASE1) {
4793                 struct lfsck_position pos;
4794                 time64_t duration = ktime_get_seconds() -
4795                                     lfsck->li_time_last_checkpoint;
4796                 u64 checked = ns->ln_items_checked + com->lc_new_checked;
4797                 u64 speed = checked;
4798                 u64 new_checked = com->lc_new_checked;
4799                 time64_t rtime = ns->ln_run_time_phase1 + duration;
4800
4801                 if (duration != 0)
4802                         new_checked = div64_s64(new_checked, duration);
4803
4804                 if (rtime != 0)
4805                         speed = div64_s64(speed, rtime);
4806
4807                 lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0,
4808                                                 bk->lb_param & LPF_DRYRUN);
4809                 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4810                            "average_speed_phase2: N/A\n"
4811                            "average_speed_total: %llu items/sec\n"
4812                            "real_time_speed_phase1: %llu items/sec\n"
4813                            "real_time_speed_phase2: N/A\n",
4814                            speed,
4815                            speed,
4816                            new_checked);
4817
4818                 if (likely(lfsck->li_di_oit)) {
4819                         const struct dt_it_ops *iops =
4820                                 &lfsck->li_obj_oit->do_index_ops->dio_it;
4821
4822                         /* The low layer otable-based iteration position may NOT
4823                          * exactly match the namespace-based directory traversal
4824                          * cookie. Generally, it is not a serious issue. But the
4825                          * caller should NOT make assumption on that. */
4826                         pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
4827                         if (!lfsck->li_current_oit_processed)
4828                                 pos.lp_oit_cookie--;
4829
4830                         spin_lock(&lfsck->li_lock);
4831                         if (lfsck->li_di_dir) {
4832                                 pos.lp_dir_cookie = lfsck->li_cookie_dir;
4833                                 if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
4834                                         fid_zero(&pos.lp_dir_parent);
4835                                         pos.lp_dir_cookie = 0;
4836                                 } else {
4837                                         pos.lp_dir_parent =
4838                                         *lfsck_dto2fid(lfsck->li_obj_dir);
4839                                 }
4840                         } else {
4841                                 fid_zero(&pos.lp_dir_parent);
4842                                 pos.lp_dir_cookie = 0;
4843                         }
4844                         spin_unlock(&lfsck->li_lock);
4845                 } else {
4846                         pos = ns->ln_pos_last_checkpoint;
4847                 }
4848
4849                 lfsck_pos_dump(m, &pos, "current_position");
4850         } else if (ns->ln_status == LS_SCANNING_PHASE2) {
4851                 time64_t duration = ktime_get_seconds() -
4852                                     com->lc_time_last_checkpoint;
4853                 __u64 checked = ns->ln_objs_checked_phase2 +
4854                                 com->lc_new_checked;
4855                 __u64 speed1 = ns->ln_items_checked;
4856                 __u64 speed2 = checked;
4857                 __u64 speed0 = speed1 + speed2;
4858                 __u64 new_checked = com->lc_new_checked;
4859                 time64_t rtime = ns->ln_run_time_phase2 + duration;
4860                 time64_t time0 = ns->ln_run_time_phase1 + rtime;
4861
4862                 if (duration != 0)
4863                         new_checked = div64_s64(new_checked, duration);
4864
4865                 if (ns->ln_run_time_phase1 != 0)
4866                         speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
4867                 else if (ns->ln_items_checked != 0)
4868                         time0++;
4869
4870                 if (rtime != 0)
4871                         speed2 = div64_s64(speed2, rtime);
4872                 else if (checked != 0)
4873                         time0++;
4874
4875                 if (time0 != 0)
4876                         speed0 = div64_s64(speed0, time0);
4877
4878                 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4879                                                 checked,
4880                                                 ns->ln_run_time_phase1, rtime,
4881                                                 bk->lb_param & LPF_DRYRUN);
4882                 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4883                            "average_speed_phase2: %llu objs/sec\n"
4884                            "average_speed_total: %llu items/sec\n"
4885                            "real_time_speed_phase1: N/A\n"
4886                            "real_time_speed_phase2: %llu objs/sec\n"
4887                            "current_position: "DFID"\n",
4888                            speed1,
4889                            speed2,
4890                            speed0,
4891                            new_checked,
4892                            PFID(&ns->ln_fid_latest_scanned_phase2));
4893         } else {
4894                 __u64 speed1 = ns->ln_items_checked;
4895                 __u64 speed2 = ns->ln_objs_checked_phase2;
4896                 __u64 speed0 = speed1 + speed2;
4897                 time64_t time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
4898
4899                 if (ns->ln_run_time_phase1 != 0)
4900                         speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
4901                 else if (ns->ln_items_checked != 0)
4902                         time0++;
4903
4904                 if (ns->ln_run_time_phase2 != 0)
4905                         speed2 = div64_s64(speed2, ns->ln_run_time_phase2);
4906                 else if (ns->ln_objs_checked_phase2 != 0)
4907                         time0++;
4908
4909                 if (time0 != 0)
4910                         speed0 = div64_s64(speed0, time0);
4911
4912                 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4913                                                 ns->ln_objs_checked_phase2,
4914                                                 ns->ln_run_time_phase1,
4915                                                 ns->ln_run_time_phase2,
4916                                                 bk->lb_param & LPF_DRYRUN);
4917                 seq_printf(m, "average_speed_phase1: %llu items/sec\n"
4918                            "average_speed_phase2: %llu objs/sec\n"
4919                            "average_speed_total: %llu items/sec\n"
4920                            "real_time_speed_phase1: N/A\n"
4921                            "real_time_speed_phase2: N/A\n"
4922                            "current_position: N/A\n",
4923                            speed1,
4924                            speed2,
4925                            speed0);
4926         }
4927
4928         up_read(&com->lc_sem);
4929 }
4930
4931 static int lfsck_namespace_double_scan(const struct lu_env *env,
4932                                        struct lfsck_component *com)
4933 {
4934         struct lfsck_namespace          *ns     = com->lc_file_ram;
4935         struct lfsck_assistant_data     *lad    = com->lc_data;
4936         struct lfsck_tgt_descs          *ltds   = &com->lc_lfsck->li_mdt_descs;
4937         struct lfsck_tgt_desc           *ltd;
4938         struct lfsck_tgt_desc           *next;
4939         int                              rc;
4940
4941         rc = lfsck_double_scan_generic(env, com, ns->ln_status);
4942         if (thread_is_stopped(&lad->lad_thread)) {
4943                 LASSERT(list_empty(&lad->lad_req_list));
4944                 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4945
4946                 spin_lock(&ltds->ltd_lock);
4947                 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4948                                          ltd_namespace_phase_list) {
4949                         list_del_init(&ltd->ltd_namespace_phase_list);
4950                 }
4951                 spin_unlock(&ltds->ltd_lock);
4952         }
4953
4954         return rc;
4955 }
4956
4957 static void lfsck_namespace_data_release(const struct lu_env *env,
4958                                          struct lfsck_component *com)
4959 {
4960         struct lfsck_assistant_data     *lad    = com->lc_data;
4961         struct lfsck_tgt_descs          *ltds   = &com->lc_lfsck->li_mdt_descs;
4962         struct lfsck_tgt_desc           *ltd;
4963         struct lfsck_tgt_desc           *next;
4964
4965         LASSERT(lad != NULL);
4966         LASSERT(thread_is_init(&lad->lad_thread) ||
4967                 thread_is_stopped(&lad->lad_thread));
4968         LASSERT(list_empty(&lad->lad_req_list));
4969
4970         com->lc_data = NULL;
4971         lfsck_namespace_release_lmv(env, com);
4972
4973         spin_lock(&ltds->ltd_lock);
4974         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4975                                  ltd_namespace_phase_list) {
4976                 list_del_init(&ltd->ltd_namespace_phase_list);
4977         }
4978         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4979                                  ltd_namespace_phase_list) {
4980                 list_del_init(&ltd->ltd_namespace_phase_list);
4981         }
4982         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
4983                                  ltd_namespace_list) {
4984                 list_del_init(&ltd->ltd_namespace_list);
4985         }
4986         spin_unlock(&ltds->ltd_lock);
4987
4988         if (likely(lad->lad_bitmap != NULL))
4989                 CFS_FREE_BITMAP(lad->lad_bitmap);
4990
4991         OBD_FREE_PTR(lad);
4992 }
4993
4994 static void lfsck_namespace_quit(const struct lu_env *env,
4995                                  struct lfsck_component *com)
4996 {
4997         struct lfsck_assistant_data     *lad    = com->lc_data;
4998         struct lfsck_tgt_descs          *ltds   = &com->lc_lfsck->li_mdt_descs;
4999         struct lfsck_tgt_desc           *ltd;
5000         struct lfsck_tgt_desc           *next;
5001
5002         LASSERT(lad != NULL);
5003
5004         lfsck_quit_generic(env, com);
5005
5006         LASSERT(thread_is_init(&lad->lad_thread) ||
5007                 thread_is_stopped(&lad->lad_thread));
5008         LASSERT(list_empty(&lad->lad_req_list));
5009
5010         lfsck_namespace_release_lmv(env, com);
5011
5012         spin_lock(&ltds->ltd_lock);
5013         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
5014                                  ltd_namespace_phase_list) {
5015                 list_del_init(&ltd->ltd_namespace_phase_list);
5016         }
5017         list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5018                                  ltd_namespace_phase_list) {
5019                 list_del_init(&ltd->ltd_namespace_phase_list);
5020         }
5021         spin_unlock(&ltds->ltd_lock);
5022 }
5023
5024 static int lfsck_namespace_in_notify(const struct lu_env *env,
5025                                      struct lfsck_component *com,
5026                                      struct lfsck_request *lr)
5027 {
5028         struct lfsck_instance *lfsck = com->lc_lfsck;
5029         struct lfsck_namespace *ns = com->lc_file_ram;
5030         struct lfsck_assistant_data *lad = com->lc_data;
5031         struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
5032         struct lfsck_tgt_desc *ltd;
5033         int rc = 0;
5034         bool fail = false;
5035         ENTRY;
5036
5037         switch (lr->lr_event) {
5038         case LE_SET_LMV_MASTER: {
5039                 struct dt_object        *obj;
5040
5041                 obj = lfsck_object_find_bottom(env, lfsck, &lr->lr_fid);
5042                 if (IS_ERR(obj))
5043                         RETURN(PTR_ERR(obj));
5044
5045                 if (likely(dt_object_exists(obj)))
5046                         rc = lfsck_namespace_notify_lmv_master_local(env, com,
5047                                                                      obj);
5048
5049                 lfsck_object_put(env, obj);
5050
5051                 RETURN(rc > 0 ? 0 : rc);
5052         }
5053         case LE_SET_LMV_SLAVE: {
5054                 if (!(lr->lr_flags & LEF_RECHECK_NAME_HASH))
5055                         ns->ln_striped_shards_repaired++;
5056
5057                 rc = lfsck_namespace_trace_update(env, com, &lr->lr_fid,
5058                                                   LNTF_RECHECK_NAME_HASH, true);
5059
5060                 RETURN(rc > 0 ? 0 : rc);
5061         }
5062         case LE_PHASE1_DONE:
5063         case LE_PHASE2_DONE:
5064         case LE_PEER_EXIT:
5065                 break;
5066         default:
5067                 RETURN(-EINVAL);
5068         }
5069
5070         CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
5071                "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
5072                lr->lr_index, lr->lr_status, lr->lr_flags2);
5073
5074         spin_lock(&ltds->ltd_lock);
5075         ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
5076         if (ltd == NULL) {
5077                 spin_unlock(&ltds->ltd_lock);
5078
5079                 RETURN(-ENXIO);
5080         }
5081
5082         list_del_init(&ltd->ltd_namespace_phase_list);
5083         switch (lr->lr_event) {
5084         case LE_PHASE1_DONE:
5085                 if (lr->lr_status <= 0) {
5086                         ltd->ltd_namespace_done = 1;
5087                         list_del_init(&ltd->ltd_namespace_list);
5088                         CDEBUG(D_LFSCK, "%s: MDT %x failed/stopped at "
5089                                "phase1 for namespace LFSCK: rc = %d.\n",
5090                                lfsck_lfsck2name(lfsck),
5091                                ltd->ltd_index, lr->lr_status);
5092                         ns->ln_flags |= LF_INCOMPLETE;
5093                         fail = true;
5094                         break;
5095                 }
5096
5097                 if (lr->lr_flags2 & LF_INCOMPLETE)
5098                         ns->ln_flags |= LF_INCOMPLETE;
5099
5100                 if (list_empty(&ltd->ltd_namespace_list))
5101                         list_add_tail(&ltd->ltd_namespace_list,
5102                                       &lad->lad_mdt_list);
5103                 list_add_tail(&ltd->ltd_namespace_phase_list,
5104                               &lad->lad_mdt_phase2_list);
5105                 break;
5106         case LE_PHASE2_DONE:
5107                 ltd->ltd_namespace_done = 1;
5108                 list_del_init(&ltd->ltd_namespace_list);
5109                 break;
5110         case LE_PEER_EXIT:
5111                 fail = true;
5112                 ltd->ltd_namespace_done = 1;
5113                 list_del_init(&ltd->ltd_namespace_list);
5114                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
5115                         CDEBUG(D_LFSCK,
5116                                "%s: the peer MDT %x exit namespace LFSCK\n",
5117                                lfsck_lfsck2name(lfsck), ltd->ltd_index);
5118                         ns->ln_flags |= LF_INCOMPLETE;
5119                 }
5120                 break;
5121         default:
5122                 break;
5123         }
5124         spin_unlock(&ltds->ltd_lock);
5125
5126         if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
5127                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5128
5129                 memset(stop, 0, sizeof(*stop));
5130                 stop->ls_status = lr->lr_status;
5131                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5132                 lfsck_stop(env, lfsck->li_bottom, stop);
5133         } else if (lfsck_phase2_next_ready(lad)) {
5134                 wake_up_all(&lad->lad_thread.t_ctl_waitq);
5135         }
5136
5137         RETURN(0);
5138 }
5139
5140 static void lfsck_namespace_repaired(struct lfsck_namespace *ns, __u64 *count)
5141 {
5142         *count += ns->ln_objs_nlink_repaired;
5143         *count += ns->ln_dirent_repaired;
5144         *count += ns->ln_linkea_repaired;
5145         *count += ns->ln_mul_linked_repaired;
5146         *count += ns->ln_unmatched_pairs_repaired;
5147         *count += ns->ln_dangling_repaired;
5148         *count += ns->ln_mul_ref_repaired;
5149         *count += ns->ln_bad_type_repaired;
5150         *count += ns->ln_lost_dirent_repaired;
5151         *count += ns->ln_striped_dirs_disabled;
5152         *count += ns->ln_striped_dirs_repaired;
5153         *count += ns->ln_striped_shards_repaired;
5154         *count += ns->ln_name_hash_repaired;
5155         *count += ns->ln_local_lpf_moved;
5156 }
5157
5158 static int lfsck_namespace_query_all(const struct lu_env *env,
5159                                      struct lfsck_component *com,
5160                                      __u32 *mdts_count, __u64 *repaired)
5161 {
5162         struct lfsck_namespace *ns = com->lc_file_ram;
5163         struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
5164         struct lfsck_tgt_desc *ltd;
5165         int idx;
5166         int rc;
5167         ENTRY;
5168
5169         rc = lfsck_query_all(env, com);
5170         if (rc != 0)
5171                 RETURN(rc);
5172
5173         down_read(&ltds->ltd_rw_sem);
5174         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
5175                 ltd = lfsck_ltd2tgt(ltds, idx);
5176                 LASSERT(ltd != NULL);
5177
5178                 mdts_count[ltd->ltd_namespace_status]++;
5179                 *repaired += ltd->ltd_namespace_repaired;
5180         }
5181         up_read(&ltds->ltd_rw_sem);
5182
5183         down_read(&com->lc_sem);
5184         mdts_count[ns->ln_status]++;
5185         lfsck_namespace_repaired(ns, repaired);
5186         up_read(&com->lc_sem);
5187
5188         RETURN(0);
5189 }
5190
5191 static int lfsck_namespace_query(const struct lu_env *env,
5192                                  struct lfsck_component *com,
5193                                  struct lfsck_request *req,
5194                                  struct lfsck_reply *rep,
5195                                  struct lfsck_query *que, int idx)
5196 {
5197         struct lfsck_namespace *ns = com->lc_file_ram;
5198         int rc = 0;
5199
5200         if (que != NULL) {
5201                 LASSERT(com->lc_lfsck->li_master);
5202
5203                 rc = lfsck_namespace_query_all(env, com,
5204                                                que->lu_mdts_count[idx],
5205                                                &que->lu_repaired[idx]);
5206         } else {
5207                 down_read(&com->lc_sem);
5208                 rep->lr_status = ns->ln_status;
5209                 if (req->lr_flags & LEF_QUERY_ALL)
5210                         lfsck_namespace_repaired(ns, &rep->lr_repaired);
5211                 up_read(&com->lc_sem);
5212         }
5213
5214         return rc;
5215 }
5216
5217 static const struct lfsck_operations lfsck_namespace_ops = {
5218         .lfsck_reset            = lfsck_namespace_reset,
5219         .lfsck_fail             = lfsck_namespace_fail,
5220         .lfsck_close_dir        = lfsck_namespace_close_dir,
5221         .lfsck_open_dir         = lfsck_namespace_open_dir,
5222         .lfsck_checkpoint       = lfsck_namespace_checkpoint,
5223         .lfsck_prep             = lfsck_namespace_prep,
5224         .lfsck_exec_oit         = lfsck_namespace_exec_oit,
5225         .lfsck_exec_dir         = lfsck_namespace_exec_dir,
5226         .lfsck_post             = lfsck_namespace_post,
5227         .lfsck_dump             = lfsck_namespace_dump,
5228         .lfsck_double_scan      = lfsck_namespace_double_scan,
5229         .lfsck_data_release     = lfsck_namespace_data_release,
5230         .lfsck_quit             = lfsck_namespace_quit,
5231         .lfsck_in_notify        = lfsck_namespace_in_notify,
5232         .lfsck_query            = lfsck_namespace_query,
5233 };
5234
5235 /**
5236  * Repair dangling name entry.
5237  *
5238  * For the name entry with dangling reference, we need to repare the
5239  * inconsistency according to the LFSCK sponsor's requirement:
5240  *
5241  * 1) Keep the inconsistency there and report the inconsistency case,
5242  *    then give the chance to the application to find related issues,
5243  *    and the users can make the decision about how to handle it with
5244  *    more human knownledge. (by default)
5245  *
5246  * 2) Re-create the missing MDT-object with the FID information.
5247  *
5248  * \param[in] env       pointer to the thread context
5249  * \param[in] com       pointer to the lfsck component
5250  * \param[in] parent    pointer to the dir object that contains the dangling
5251  *                      name entry
5252  * \param[in] child     pointer to the object corresponding to the dangling
5253  *                      name entry
5254  * \param[in] lnr       pointer to the namespace request that contains the
5255  *                      name's name, parent object, parent's LMV, and ect.
5256  *
5257  * \retval              positive number if no need to repair
5258  * \retval              zero for repaired successfully
5259  * \retval              negative error number on failure
5260  */
5261 int lfsck_namespace_repair_dangling(const struct lu_env *env,
5262                                     struct lfsck_component *com,
5263                                     struct dt_object *parent,
5264                                     struct dt_object *child,
5265                                     struct lfsck_namespace_req *lnr)
5266 {
5267         struct lfsck_thread_info *info = lfsck_env_info(env);
5268         struct lu_attr *la = &info->lti_la;
5269         struct dt_allocation_hint *hint = &info->lti_hint;
5270         struct dt_object_format *dof = &info->lti_dof;
5271         struct dt_insert_rec *rec = &info->lti_dt_rec;
5272         struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
5273         const struct lu_name *cname;
5274         const struct lu_fid *pfid = lfsck_dto2fid(parent);
5275         const struct lu_fid *cfid = lfsck_dto2fid(child);
5276         struct linkea_data ldata = { NULL };
5277         struct lfsck_lock_handle *llh = &info->lti_llh;
5278         struct lustre_handle rlh = { 0 };
5279         struct lustre_handle clh = { 0 };
5280         struct lu_buf linkea_buf;
5281         struct lu_buf lmv_buf;
5282         struct lfsck_instance *lfsck = com->lc_lfsck;
5283         struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5284         struct dt_device *dev = lfsck->li_next;
5285         struct thandle *th = NULL;
5286         int rc = 0;
5287         __u16 type = lnr->lnr_type;
5288         bool create;
5289         ENTRY;
5290
5291         cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5292         if (bk->lb_param & LPF_CREATE_MDTOBJ)
5293                 create = true;
5294         else
5295                 create = false;
5296
5297         if (!create || bk->lb_param & LPF_DRYRUN)
5298                 GOTO(log, rc = 0);
5299
5300         /* We may need to create the sub-objects of the @child via LOD,
5301          * so make the modification based on lfsck->li_next. */
5302
5303         parent = lfsck_object_locate(dev, parent);
5304         if (IS_ERR(parent))
5305                 GOTO(log, rc = PTR_ERR(parent));
5306
5307         if (unlikely(!dt_try_as_dir(env, parent)))
5308                 GOTO(log, rc = -ENOTDIR);
5309
5310         child = lfsck_object_locate(dev, child);
5311         if (IS_ERR(child))
5312                 GOTO(log, rc = PTR_ERR(child));
5313
5314         rc = linkea_links_new(&ldata, &info->lti_linkea_buf2,
5315                               cname, pfid);
5316         if (rc != 0)
5317                 GOTO(log, rc);
5318
5319         rc = lfsck_lock(env, lfsck, parent, lnr->lnr_name, llh,
5320                         MDS_INODELOCK_UPDATE, LCK_PW);
5321         if (rc != 0)
5322                 GOTO(log, rc);
5323
5324         rc = lfsck_namespace_check_exist(env, parent, child, lnr->lnr_name);
5325         if (rc != 0)
5326                 GOTO(log, rc);
5327
5328         if (dt_object_remote(child)) {
5329                 rc = lfsck_remote_lookup_lock(env, lfsck, parent, child, &rlh,
5330                                               LCK_EX);
5331                 if (rc != 0)
5332                         GOTO(log, rc);
5333         }
5334
5335         rc = lfsck_ibits_lock(env, lfsck, child, &clh,
5336                               MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
5337                               MDS_INODELOCK_XATTR, LCK_EX);
5338         if (rc != 0)
5339                 GOTO(unlock_remote_lookup, rc);
5340
5341         /* Set the ctime as zero, then others can know it is created for
5342          * repairing dangling name entry by LFSCK. And if the LFSCK made
5343          * wrong decision and the real MDT-object has been found later,
5344          * then the LFSCK has chance to fix the incosistency properly. */
5345         memset(la, 0, sizeof(*la));
5346         if (S_ISDIR(type))
5347                 la->la_mode = (type & S_IFMT) | 0700;
5348         else
5349                 la->la_mode = (type & S_IFMT) | 0600;
5350         la->la_valid = LA_TYPE | LA_MODE | LA_CTIME;
5351
5352         /*
5353          * if it's directory, skip do_ah_init() to create a plain directory
5354          * because it may have shards already, which will be inserted back
5355          * later, besides, it may be remote, and creating stripe directory
5356          * remotely is not supported.
5357          */
5358         if (S_ISREG(type))
5359                 child->do_ops->do_ah_init(env, hint,  parent, child, type);
5360         else if (S_ISDIR(type))
5361                 child->do_ops->do_ah_init(env, hint,  NULL, child, type);
5362
5363         memset(dof, 0, sizeof(*dof));
5364         dof->dof_type = dt_mode_to_dft(type);
5365         /* If the target is a regular file, then the LFSCK will only create
5366          * the MDT-object without stripes (dof->dof_reg.striped = 0). related
5367          * OST-objects will be created when write open. */
5368
5369         th = dt_trans_create(env, dev);
5370         if (IS_ERR(th))
5371                 GOTO(unlock_child, rc = PTR_ERR(th));
5372
5373         /* 1a. create child. */
5374         rc = dt_declare_create(env, child, la, hint, dof, th);
5375         if (rc != 0)
5376                 GOTO(stop, rc);
5377
5378         if (S_ISDIR(type)) {
5379                 if (unlikely(!dt_try_as_dir(env, child)))
5380                         GOTO(stop, rc = -ENOTDIR);
5381
5382                 /* 2a. increase child nlink */
5383                 rc = dt_declare_ref_add(env, child, th);
5384                 if (rc != 0)
5385                         GOTO(stop, rc);
5386
5387                 /* 3a. insert dot into child dir */
5388                 rec->rec_type = S_IFDIR;
5389                 rec->rec_fid = cfid;
5390                 rc = dt_declare_insert(env, child,
5391                                        (const struct dt_rec *)rec,
5392                                        (const struct dt_key *)dot, th);
5393                 if (rc != 0)
5394                         GOTO(stop, rc);
5395
5396                 /* 4a. insert dotdot into child dir */
5397                 rec->rec_fid = pfid;
5398                 rc = dt_declare_insert(env, child,
5399                                        (const struct dt_rec *)rec,
5400                                        (const struct dt_key *)dotdot, th);
5401                 if (rc != 0)
5402                         GOTO(stop, rc);
5403
5404                 /* 5a. generate slave LMV EA. */
5405                 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5406                         int idx;
5407
5408                         idx = lfsck_shard_name_to_index(env,
5409                                         lnr->lnr_name, lnr->lnr_namelen,
5410                                         type, cfid);
5411                         if (unlikely(idx < 0))
5412                                 GOTO(stop, rc = idx);
5413
5414                         *lmv2 = lnr->lnr_lmv->ll_lmv;
5415                         lmv2->lmv_magic = LMV_MAGIC_STRIPE;
5416                         lmv2->lmv_master_mdt_index = idx;
5417
5418                         lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
5419                         lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
5420                         rc = dt_declare_xattr_set(env, child, &lmv_buf,
5421                                                   XATTR_NAME_LMV, 0, th);
5422                         if (rc != 0)
5423                                 GOTO(stop, rc);
5424                 }
5425         }
5426
5427         /* 6a. insert linkEA for child */
5428         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
5429                        ldata.ld_leh->leh_len);
5430         rc = dt_declare_xattr_set(env, child, &linkea_buf,
5431                                   XATTR_NAME_LINK, 0, th);
5432         if (rc != 0)
5433                 GOTO(stop, rc);
5434
5435         /* 7a. if child is remote, delete and insert to generate local agent */
5436         if (dt_object_remote(child)) {
5437                 rc = dt_declare_delete(env, parent,
5438                                        (const struct dt_key *)lnr->lnr_name,
5439                                        th);
5440                 if (rc)
5441                         GOTO(stop, rc);
5442
5443                 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
5444                                        (const struct dt_key *)lnr->lnr_name,
5445                                        th);
5446                 if (rc)
5447                         GOTO(stop, rc);
5448         }
5449
5450         rc = dt_trans_start_local(env, dev, th);
5451         if (rc != 0)
5452                 GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
5453
5454         dt_write_lock(env, child, 0);
5455         /* 1b. create child */
5456         rc = dt_create(env, child, la, hint, dof, th);
5457         if (rc != 0)
5458                 GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc));
5459
5460         if (S_ISDIR(type)) {
5461                 /* 2b. increase child nlink */
5462                 rc = dt_ref_add(env, child, th);
5463                 if (rc != 0)
5464                         GOTO(unlock, rc);
5465
5466                 /* 3b. insert dot into child dir */
5467                 rec->rec_type = S_IFDIR;
5468                 rec->rec_fid = cfid;
5469                 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5470                                (const struct dt_key *)dot, th);
5471                 if (rc != 0)
5472                         GOTO(unlock, rc);
5473
5474                 /* 4b. insert dotdot into child dir */
5475                 rec->rec_fid = pfid;
5476                 rc = dt_insert(env, child, (const struct dt_rec *)rec,
5477                                (const struct dt_key *)dotdot, th);
5478                 if (rc != 0)
5479                         GOTO(unlock, rc);
5480
5481                 /* 5b. generate slave LMV EA. */
5482                 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5483                         rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV,
5484                                           0, th);
5485                         if (rc != 0)
5486                                 GOTO(unlock, rc);
5487                 }
5488         }
5489
5490         /* 6b. insert linkEA for child. */
5491         rc = dt_xattr_set(env, child, &linkea_buf,
5492                           XATTR_NAME_LINK, 0, th);
5493         if (rc)
5494                 GOTO(unlock, rc);
5495
5496         /* 7b. if child is remote, delete and insert to generate local agent */
5497         if (dt_object_remote(child)) {
5498                 rc = dt_delete(env, parent,
5499                                (const struct dt_key *)lnr->lnr_name, th);
5500                 if (rc)
5501                         GOTO(unlock, rc);
5502
5503                 rec->rec_type = type;
5504                 rec->rec_fid = cfid;
5505                 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
5506                                (const struct dt_key *)lnr->lnr_name, th);
5507                 if (rc)
5508                         GOTO(unlock, rc);
5509         }
5510
5511         GOTO(unlock, rc);
5512
5513 unlock:
5514         dt_write_unlock(env, child);
5515
5516 stop:
5517         dt_trans_stop(env, dev, th);
5518
5519 unlock_child:
5520         lfsck_ibits_unlock(&clh, LCK_EX);
5521 unlock_remote_lookup:
5522         if (dt_object_remote(child))
5523                 lfsck_ibits_unlock(&rlh, LCK_EX);
5524 log:
5525         lfsck_unlock(llh);
5526         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
5527                "reference for: parent "DFID", child "DFID", type %u, "
5528                "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck),
5529                PFID(pfid), PFID(cfid), type, cname->ln_name,
5530                create ? "Create the lost MDT-object as required" :
5531                         "Keep the MDT-object there by default", rc);
5532
5533         if (rc <= 0) {
5534                 struct lfsck_namespace *ns = com->lc_file_ram;
5535
5536                 ns->ln_flags |= LF_INCONSISTENT;
5537         }
5538
5539         return rc;
5540 }
5541
5542 static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
5543                                                 struct lfsck_component *com,
5544                                                 struct lfsck_assistant_req *lar)
5545 {
5546         struct lfsck_thread_info   *info     = lfsck_env_info(env);
5547         struct lu_attr             *la       = &info->lti_la;
5548         struct lfsck_instance      *lfsck    = com->lc_lfsck;
5549         struct lfsck_bookmark      *bk       = &lfsck->li_bookmark_ram;
5550         struct lfsck_namespace     *ns       = com->lc_file_ram;
5551         struct lfsck_assistant_data *lad     = com->lc_data;
5552         struct linkea_data          ldata    = { NULL };
5553         const struct lu_name       *cname;
5554         struct thandle             *handle   = NULL;
5555         struct lfsck_namespace_req *lnr      =
5556                 container_of(lar, struct lfsck_namespace_req, lnr_lar);
5557         struct dt_object           *dir      = NULL;
5558         struct dt_object           *obj      = NULL;
5559         struct lfsck_assistant_object *lso   = lar->lar_parent;
5560         const struct lu_fid        *pfid     = &lso->lso_fid;
5561         struct dt_device           *dev      = NULL;
5562         struct lustre_handle        lh       = { 0 };
5563         bool                        repaired = false;
5564         bool                        dtlocked = false;
5565         bool                        remove;
5566         bool                        newdata;
5567         bool                        log      = false;
5568         bool                        bad_hash = false;
5569         bool                        bad_linkea = false;
5570         int                         idx      = 0;
5571         int                         count    = 0;
5572         int                         rc       = 0;
5573         enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
5574         ENTRY;
5575
5576         if (lso->lso_dead)
5577                 RETURN(0);
5578
5579         la->la_nlink = 0;
5580         if (lnr->lnr_attr & LUDA_UPGRADE) {
5581                 ns->ln_flags |= LF_UPGRADE;
5582                 ns->ln_dirent_repaired++;
5583                 repaired = true;
5584         } else if (lnr->lnr_attr & LUDA_REPAIR) {
5585                 ns->ln_flags |= LF_INCONSISTENT;
5586                 ns->ln_dirent_repaired++;
5587                 repaired = true;
5588         }
5589
5590         if (unlikely(fid_is_zero(&lnr->lnr_fid) &&
5591                      strcmp(lnr->lnr_name, dotdot) == 0)) {
5592                 rc = lfsck_namespace_trace_update(env, com, pfid,
5593                                                 LNTF_CHECK_PARENT, true);
5594
5595                 GOTO(out, rc);
5596         }
5597
5598         if (unlikely(!fid_is_sane(&lnr->lnr_fid))) {
5599                 CDEBUG(D_LFSCK, "%s: dir scan find invalid FID "DFID
5600                        " for the name entry %.*s under "DFID"\n",
5601                        lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5602                        lnr->lnr_namelen, lnr->lnr_name, PFID(pfid));
5603
5604                 if (strcmp(lnr->lnr_name, dotdot) != 0)
5605                         /* invalid FID means bad name entry, remove it. */
5606                         type = LNIT_BAD_DIRENT;
5607                 else
5608                         /* If the parent FID is invalid, we cannot remove
5609                          * the ".." entry directly. */
5610                         rc = lfsck_namespace_trace_update(env, com, pfid,
5611                                                 LNTF_CHECK_PARENT, true);
5612
5613                 GOTO(out, rc);
5614         }
5615
5616         if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
5617                 rc = lfsck_namespace_striped_dir_rescan(env, com, lnr);
5618
5619                 RETURN(rc);
5620         }
5621
5622         if (fid_seq_is_dot(fid_seq(&lnr->lnr_fid)))
5623                 GOTO(out, rc = 0);
5624
5625         if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5626                 rc = lfsck_namespace_handle_striped_master(env, com, lnr);
5627
5628                 RETURN(rc);
5629         }
5630
5631         idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
5632         if (idx < 0)
5633                 GOTO(out, rc = idx);
5634
5635         if (idx == lfsck_dev_idx(lfsck)) {
5636                 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0))
5637                         GOTO(out, rc = 0);
5638
5639                 dev = lfsck->li_bottom;
5640         } else {
5641                 struct lfsck_tgt_desc *ltd;
5642
5643                 /* Usually, some local filesystem consistency verification
5644                  * tools can guarantee the local namespace tree consistenct.
5645                  * So the LFSCK will only verify the remote directory. */
5646                 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) {
5647                         rc = lfsck_namespace_trace_update(env, com, pfid,
5648                                                 LNTF_CHECK_PARENT, true);
5649
5650                         GOTO(out, rc);
5651                 }
5652
5653                 ltd = lfsck_ltd2tgt(&lfsck->li_mdt_descs, idx);
5654                 if (unlikely(ltd == NULL)) {
5655                         CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
5656                                "did not join the namespace LFSCK\n",
5657                                lfsck_lfsck2name(lfsck), idx);
5658                         lfsck_lad_set_bitmap(env, com, idx);
5659
5660                         GOTO(out, rc = -ENODEV);
5661                 }
5662
5663                 dev = ltd->ltd_tgt;
5664         }
5665
5666         obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
5667         if (IS_ERR(obj))
5668                 GOTO(out, rc = PTR_ERR(obj));
5669
5670         cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5671         if (dt_object_exists(obj) == 0) {
5672
5673 dangling:
5674                 if (dir == NULL) {
5675                         dir = lfsck_assistant_object_load(env, lfsck, lso);
5676                         if (IS_ERR(dir)) {
5677                                 rc = PTR_ERR(dir);
5678
5679                                 GOTO(trace, rc == -ENOENT ? 0 : rc);
5680                         }
5681                 }
5682
5683                 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5684                 if (rc == 0) {
5685                         if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5686                                         lnr->lnr_name, lnr->lnr_namelen)) {
5687                                 type = LNIT_BAD_DIRENT;
5688
5689                                 GOTO(out, rc);
5690                         }
5691
5692                         type = LNIT_DANGLING;
5693                         rc = lfsck_namespace_repair_dangling(env, com, dir,
5694                                                              obj, lnr);
5695                         if (rc == 0)
5696                                 repaired = true;
5697                 }
5698
5699                 GOTO(out, rc);
5700         }
5701
5702         if (!(bk->lb_param & LPF_DRYRUN) && lad->lad_advance_lock) {
5703
5704 again:
5705                 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5706                                       MDS_INODELOCK_UPDATE |
5707                                       MDS_INODELOCK_XATTR, LCK_EX);
5708                 if (rc != 0)
5709                         GOTO(out, rc);
5710
5711                 handle = dt_trans_create(env, dev);
5712                 if (IS_ERR(handle))
5713                         GOTO(out, rc = PTR_ERR(handle));
5714
5715                 rc = lfsck_declare_namespace_exec_dir(env, obj, handle);
5716                 if (rc != 0)
5717                         GOTO(stop, rc);
5718
5719                 rc = dt_trans_start_local(env, dev, handle);
5720                 if (rc != 0)
5721                         GOTO(stop, rc);
5722
5723                 dt_write_lock(env, obj, 0);
5724                 dtlocked = true;
5725         }
5726
5727         rc = lfsck_links_read(env, obj, &ldata);
5728         if (unlikely(rc == -ENOENT)) {
5729                 if (handle != NULL) {
5730                         dt_write_unlock(env, obj);
5731                         dtlocked = false;
5732
5733                         dt_trans_stop(env, dev, handle);
5734                         handle = NULL;
5735
5736                         lfsck_ibits_unlock(&lh, LCK_EX);
5737                 }
5738
5739                 /* It may happen when the remote object has been removed,
5740                  * but the local MDT is not aware of that. */
5741                 goto dangling;
5742         } else if (rc == 0) {
5743                 count = ldata.ld_leh->leh_reccount;
5744                 rc = linkea_links_find(&ldata, cname, pfid);
5745                 if ((rc == 0) &&
5746                     (count == 1 || !S_ISDIR(lfsck_object_type(obj)))) {
5747                         if ((lfsck_object_type(obj) & S_IFMT) !=
5748                             lnr->lnr_type) {
5749                                 ns->ln_flags |= LF_INCONSISTENT;
5750                                 type = LNIT_BAD_TYPE;
5751                         }
5752
5753                         goto stop;
5754                 }
5755
5756                 /* If the name entry hash does not match the slave striped
5757                  * directory, and the name entry does not match also, then
5758                  * it is quite possible that name entry is corrupted. */
5759                 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5760                                         lnr->lnr_name, lnr->lnr_namelen)) {
5761                         ns->ln_flags |= LF_INCONSISTENT;
5762                         type = LNIT_BAD_DIRENT;
5763
5764                         GOTO(stop, rc = 0);
5765                 }
5766
5767                 /* If the file type stored in the name entry does not match
5768                  * the file type claimed by the object, and the object does
5769                  * not recognize the name entry, then it is quite possible
5770                  * that the name entry is corrupted. */
5771                 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type) {
5772                         ns->ln_flags |= LF_INCONSISTENT;
5773                         type = LNIT_BAD_DIRENT;
5774
5775                         GOTO(stop, rc = 0);
5776                 }
5777
5778                 /* For sub-dir object, we cannot make sure whether the sub-dir
5779                  * back references the parent via ".." name entry correctly or
5780                  * not in the LFSCK first-stage scanning. It may be that the
5781                  * (remote) sub-dir ".." name entry has no parent FID after
5782                  * file-level backup/restore and its linkEA may be wrong.
5783                  * So under such case, we should replace the linkEA according
5784                  * to current name entry. But this needs to be done during the
5785                  * LFSCK second-stage scanning. The LFSCK will record the name
5786                  * entry for further possible using. */
5787                 remove = false;
5788                 newdata = false;
5789                 goto nodata;
5790         } else if (unlikely(rc == -EINVAL)) {
5791                 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5792                         type = LNIT_BAD_TYPE;
5793
5794                 count = 1;
5795                 /* The magic crashed, we are not sure whether there are more
5796                  * corrupt data in the linkea, so remove all linkea entries. */
5797                 remove = true;
5798                 newdata = true;
5799                 goto nodata;
5800         } else if (rc == -ENODATA) {
5801                 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5802                         type = LNIT_BAD_TYPE;
5803
5804                 count = 1;
5805                 remove = false;
5806                 newdata = true;
5807
5808 nodata:
5809                 if (bk->lb_param & LPF_DRYRUN) {
5810                         if (rc == -ENODATA)
5811                                 ns->ln_flags |= LF_UPGRADE;
5812                         else
5813                                 ns->ln_flags |= LF_INCONSISTENT;
5814                         ns->ln_linkea_repaired++;
5815                         repaired = true;
5816                         log = true;
5817                         goto stop;
5818                 }
5819
5820                 if (!lustre_handle_is_used(&lh)) {
5821                         remove = false;
5822                         newdata = false;
5823                         type = LNIT_NONE;
5824
5825                         goto again;
5826                 }
5827
5828                 LASSERT(handle != NULL);
5829
5830                 if (dir == NULL) {
5831                         dir = lfsck_assistant_object_load(env, lfsck, lso);
5832                         if (IS_ERR(dir)) {
5833                                 rc = PTR_ERR(dir);
5834
5835                                 GOTO(stop, rc == -ENOENT ? 0 : rc);
5836                         }
5837                 }
5838
5839                 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5840                 if (rc != 0)
5841                         GOTO(stop, rc);
5842
5843                 bad_linkea = true;
5844                 if (!remove && newdata)
5845                         ns->ln_flags |= LF_UPGRADE;
5846                 else if (remove || !(ns->ln_flags & LF_UPGRADE))
5847                         ns->ln_flags |= LF_INCONSISTENT;
5848
5849                 if (remove) {
5850                         LASSERT(newdata);
5851
5852                         rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle);
5853                         if (rc != 0 && rc != -ENOENT && rc != -ENODATA)
5854                                 GOTO(stop, rc);
5855                 }
5856
5857                 if (newdata) {
5858                         rc = linkea_data_new(&ldata,
5859                                         &lfsck_env_info(env)->lti_linkea_buf);
5860                         if (rc != 0)
5861                                 GOTO(stop, rc);
5862                 }
5863
5864                 rc = linkea_add_buf(&ldata, cname, pfid);
5865                 if (rc == 0)
5866                         rc = lfsck_links_write(env, obj, &ldata, handle);
5867                 if (rc != 0)
5868                         GOTO(stop, rc);
5869
5870                 count = ldata.ld_leh->leh_reccount;
5871                 if (!S_ISDIR(lfsck_object_type(obj)) ||
5872                     !dt_object_remote(obj)) {
5873                         ns->ln_linkea_repaired++;
5874                         repaired = true;
5875                         log = true;
5876                 }
5877         } else {
5878                 GOTO(stop, rc);
5879         }
5880
5881 stop:
5882         if (dtlocked)
5883                 dt_write_unlock(env, obj);
5884
5885         if (handle != NULL && !IS_ERR(handle))
5886                 dt_trans_stop(env, dev, handle);
5887
5888 out:
5889         lfsck_ibits_unlock(&lh, LCK_EX);
5890
5891         if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) &&
5892             !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5893                                              lnr->lnr_name, lnr->lnr_namelen) &&
5894             type != LNIT_BAD_DIRENT) {
5895                 ns->ln_flags |= LF_INCONSISTENT;
5896
5897                 log = false;
5898                 if (dir == NULL) {
5899                         dir = lfsck_assistant_object_load(env, lfsck, lso);
5900                         if (IS_ERR(dir)) {
5901                                 rc = PTR_ERR(dir);
5902
5903                                 GOTO(trace, rc == -ENOENT ? 0 : rc);
5904                         }
5905                 }
5906
5907                 rc = lfsck_namespace_repair_bad_name_hash(env, com, dir,
5908                                                 lnr->lnr_lmv, lnr->lnr_name);
5909                 if (rc == 0)
5910                         bad_hash = true;
5911         }
5912
5913         if (rc >= 0) {
5914                 if (type != LNIT_NONE && dir == NULL) {
5915                         dir = lfsck_assistant_object_load(env, lfsck, lso);
5916                         if (IS_ERR(dir)) {
5917                                 rc = PTR_ERR(dir);
5918
5919                                 GOTO(trace, rc == -ENOENT ? 0 : rc);
5920                         }
5921                 }
5922
5923                 switch (type) {
5924                 case LNIT_BAD_TYPE:
5925                         log = false;
5926                         rc = lfsck_namespace_repair_dirent(env, com, dir,
5927                                         obj, lnr->lnr_name, lnr->lnr_name,
5928                                         lnr->lnr_type, true, false);
5929                         if (rc > 0)
5930                                 repaired = true;
5931                         break;
5932                 case LNIT_BAD_DIRENT:
5933                         log = false;
5934                         /* XXX: This is a bad dirent, we do not know whether
5935                          *      the original name entry reference a regular
5936                          *      file or a directory, then keep the parent's
5937                          *      nlink count unchanged here. */
5938                         rc = lfsck_namespace_repair_dirent(env, com, dir,
5939                                         obj, lnr->lnr_name, lnr->lnr_name,
5940                                         lnr->lnr_type, false, false);
5941                         if (rc > 0)
5942                                 repaired = true;
5943                         break;
5944                 default:
5945                         break;
5946                 }
5947
5948                 if (obj != NULL && count == 1 &&
5949                     S_ISREG(lfsck_object_type(obj)))
5950                         dt_attr_get(env, obj, la);
5951         }
5952
5953 trace:
5954         down_write(&com->lc_sem);
5955         if (rc < 0) {
5956                 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
5957                        "the entry: "DFID", parent "DFID", name %.*s: rc = %d\n",
5958                        lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid), PFID(pfid),
5959                        lnr->lnr_namelen, lnr->lnr_name, rc);
5960
5961                 lfsck_namespace_record_failure(env, lfsck, ns);
5962                 if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
5963                      rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
5964                      rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
5965                     dev != NULL && dev != lfsck->li_bottom)
5966                         lfsck_lad_set_bitmap(env, com, idx);
5967
5968                 if (!(bk->lb_param & LPF_FAILOUT))
5969                         rc = 0;
5970         } else {
5971                 if (repaired) {
5972                         ns->ln_items_repaired++;
5973                         if (log)
5974                                 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant "
5975                                        "repaired the entry: "DFID", parent "DFID
5976                                        ", name %.*s, type %d\n",
5977                                        lfsck_lfsck2name(lfsck),
5978                                        PFID(&lnr->lnr_fid), PFID(pfid),
5979                                        lnr->lnr_namelen, lnr->lnr_name, type);
5980
5981                         switch (type) {
5982                         case LNIT_DANGLING:
5983                                 ns->ln_dangling_repaired++;
5984                                 break;
5985                         case LNIT_BAD_TYPE:
5986                                 ns->ln_bad_type_repaired++;
5987                                 break;
5988                         case LNIT_BAD_DIRENT:
5989                                 ns->ln_dirent_repaired++;
5990                                 break;
5991                         default:
5992                                 break;
5993                         }
5994
5995                         if (bk->lb_param & LPF_DRYRUN &&
5996                             lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5997                                 lfsck_pos_fill(env, lfsck,
5998                                                &ns->ln_pos_first_inconsistent,
5999                                                false);
6000                 }
6001
6002                 if (bad_hash) {
6003                         ns->ln_name_hash_repaired++;
6004
6005                         /* Not count repeatedly. */
6006                         if (!repaired) {
6007                                 ns->ln_items_repaired++;
6008                                 if (log)
6009                                         CDEBUG(D_LFSCK, "%s: namespace LFSCK "
6010                                                "assistant repaired the entry: "
6011                                                DFID", parent "DFID
6012                                                ", name %.*s\n",
6013                                                lfsck_lfsck2name(lfsck),
6014                                                PFID(&lnr->lnr_fid), PFID(pfid),
6015                                                lnr->lnr_namelen, lnr->lnr_name);
6016                         }
6017
6018                         if (bk->lb_param & LPF_DRYRUN &&
6019                             lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
6020                                 lfsck_pos_fill(env, lfsck,
6021                                                &ns->ln_pos_first_inconsistent,
6022                                                false);
6023                 }
6024
6025                 rc = 0;
6026         }
6027
6028         if (count > 1 || la->la_nlink > 1)
6029                 ns->ln_mul_linked_checked++;
6030
6031         up_write(&com->lc_sem);
6032
6033         if (obj != NULL && !IS_ERR(obj))
6034                 lfsck_object_put(env, obj);
6035
6036         if (dir != NULL && !IS_ERR(dir))
6037                 lfsck_object_put(env, dir);
6038
6039         lad->lad_advance_lock = bad_linkea;
6040
6041         return rc;
6042 }
6043
6044 /**
6045  * Handle one orphan under the backend /lost+found directory
6046  *
6047  * Insert the orphan FID into the namespace LFSCK trace file for further
6048  * processing (via the subsequent namespace LFSCK second-stage scanning).
6049  * At the same time, remove the orphan name entry from backend /lost+found
6050  * directory. There is an interval between the orphan name entry removed
6051  * from the backend /lost+found directory and the orphan FID in the LFSCK
6052  * trace file handled. In such interval, the LFSCK can be reset, then
6053  * all the FIDs recorded in the namespace LFSCK trace file will be dropped.
6054  * To guarantee that the orphans can be found when LFSCK run next time
6055  * without e2fsck again, when remove the orphan name entry, the LFSCK
6056  * will set the orphan's ctime attribute as 1. Since normal applications
6057  * cannot change the object's ctime attribute as 1. Then when LFSCK run
6058  * next time, it can record the object (that ctime is 1) in the namespace
6059  * LFSCK trace file during the first-stage scanning.
6060  *
6061  * \param[in] env       pointer to the thread context
6062  * \param[in] com       pointer to the lfsck component
6063  * \param[in] parent    pointer to the object for the backend /lost+found
6064  * \param[in] ent       pointer to the name entry for the target under the
6065  *                      backend /lost+found
6066  *
6067  * \retval              positive for repaired
6068  * \retval              0 if needs to repair nothing
6069  * \retval              negative error number on failure
6070  */
6071 static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
6072                                               struct lfsck_component *com,
6073                                               struct dt_object *parent,
6074                                               struct lu_dirent *ent)
6075 {
6076         struct lfsck_thread_info        *info   = lfsck_env_info(env);
6077         struct lu_fid                   *key    = &info->lti_fid;
6078         struct lu_attr                  *la     = &info->lti_la;
6079         struct lfsck_instance           *lfsck  = com->lc_lfsck;
6080         struct dt_object                *obj;
6081         struct dt_device                *dev    = lfsck->li_bottom;
6082         struct dt_object                *child  = NULL;
6083         struct thandle                  *th     = NULL;
6084         int                              idx;
6085         int                              rc     = 0;
6086         __u8                             flags  = 0;
6087         bool                             exist  = false;
6088         ENTRY;
6089
6090         child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid);
6091         if (IS_ERR(child))
6092                 RETURN(PTR_ERR(child));
6093
6094         LASSERT(dt_object_exists(child));
6095         LASSERT(!dt_object_remote(child));
6096
6097         idx = lfsck_sub_trace_file_fid2idx(&ent->lde_fid);
6098         obj = com->lc_sub_trace_objs[idx].lsto_obj;
6099         fid_cpu_to_be(key, &ent->lde_fid);
6100         rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
6101                        (const struct dt_key *)key);
6102         if (rc == 0) {
6103                 exist = true;
6104                 flags |= LNTF_CHECK_ORPHAN;
6105         } else if (rc == -ENOENT) {
6106                 flags = LNTF_CHECK_ORPHAN;
6107         } else {
6108                 GOTO(out, rc);
6109         }
6110
6111         th = dt_trans_create(env, dev);
6112         if (IS_ERR(th))
6113                 GOTO(out, rc = PTR_ERR(th));
6114
6115         /* a1. remove name entry from backend /lost+found */
6116         rc = dt_declare_delete(env, parent,
6117                                (const struct dt_key *)ent->lde_name, th);
6118         if (rc != 0)
6119                 GOTO(stop, rc);
6120
6121         if (S_ISDIR(lfsck_object_type(child))) {
6122                 /* a2. decrease parent's nlink */
6123                 rc = dt_declare_ref_del(env, parent, th);
6124                 if (rc != 0)
6125                         GOTO(stop, rc);
6126         }
6127
6128         if (exist) {
6129                 /* a3. remove child's FID from the LFSCK trace file. */
6130                 rc = dt_declare_delete(env, obj,
6131                                        (const struct dt_key *)key, th);
6132                 if (rc != 0)
6133                         GOTO(stop, rc);
6134         } else {
6135                 /* a4. set child's ctime as 1 */
6136                 memset(la, 0, sizeof(*la));
6137                 la->la_ctime = 1;
6138                 la->la_valid = LA_CTIME;
6139                 rc = dt_declare_attr_set(env, child, la, th);
6140                 if (rc != 0)
6141                         GOTO(stop, rc);
6142         }
6143
6144         /* a5. insert child's FID into the LFSCK trace file. */
6145         rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
6146                                (const struct dt_key *)key, th);
6147         if (rc != 0)
6148                 GOTO(stop, rc);
6149
6150         rc = dt_trans_start_local(env, dev, th);
6151         if (rc != 0)
6152                 GOTO(stop, rc);
6153
6154         /* b1. remove name entry from backend /lost+found */
6155         rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th);
6156         if (rc != 0)
6157                 GOTO(stop, rc);
6158
6159         if (S_ISDIR(lfsck_object_type(child))) {
6160                 /* b2. decrease parent's nlink */
6161                 dt_write_lock(env, parent, 0);
6162                 rc = dt_ref_del(env, parent, th);
6163                 dt_write_unlock(env, parent);
6164                 if (rc != 0)
6165                         GOTO(stop, rc);
6166         }
6167
6168         if (exist) {
6169                 /* a3. remove child's FID from the LFSCK trace file. */
6170                 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
6171                 if (rc != 0)
6172                         GOTO(stop, rc);
6173         } else {
6174                 /* b4. set child's ctime as 1 */
6175                 rc = dt_attr_set(env, child, la, th);
6176                 if (rc != 0)
6177                         GOTO(stop, rc);
6178         }
6179
6180         /* b5. insert child's FID into the LFSCK trace file. */
6181         rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
6182                        (const struct dt_key *)key, th);
6183
6184         GOTO(stop, rc = (rc == 0 ? 1 : rc));
6185
6186 stop:
6187         dt_trans_stop(env, dev, th);
6188
6189 out:
6190         lfsck_object_put(env, child);
6191
6192         return rc;
6193 }
6194
6195 /**
6196  * Handle orphans under the backend /lost+found directory
6197  *
6198  * Some backend checker, such as e2fsck for ldiskfs may find some orphans
6199  * and put them under the backend /lost+found directory that is invisible
6200  * to client. The LFSCK will scan such directory, for the original client
6201  * visible orphans, add their fids into the namespace LFSCK trace file,
6202  * then the subsenquent namespace LFSCK second-stage scanning can handle
6203  * them as other objects to be double scanned: either move back to normal
6204  * namespace, or to the global visible orphan directory:
6205  * /ROOT/.lustre/lost+found/MDTxxxx/
6206  *
6207  * \param[in] env       pointer to the thread context
6208  * \param[in] com       pointer to the lfsck component
6209  */
6210 static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
6211                                            struct lfsck_component *com)
6212 {
6213         struct lfsck_thread_info        *info   = lfsck_env_info(env);
6214         struct lu_dirent                *ent    =
6215                                         (struct lu_dirent *)info->lti_key;
6216         struct lu_seq_range             *range  = &info->lti_range;
6217         struct lfsck_instance           *lfsck  = com->lc_lfsck;
6218         struct ptlrpc_thread            *thread = &lfsck->li_thread;
6219         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
6220         struct lfsck_namespace          *ns     = com->lc_file_ram;
6221         struct dt_object                *parent;
6222         const struct dt_it_ops          *iops;
6223         struct dt_it                    *di;
6224         struct seq_server_site          *ss     = lfsck_dev_site(lfsck);
6225         __u64                            cookie;
6226         __u32                            idx    = lfsck_dev_idx(lfsck);
6227         int                              rc     = 0;
6228         __u16                            type;
6229         ENTRY;
6230
6231         parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
6232                                           &LU_BACKEND_LPF_FID);
6233         if (IS_ERR(parent)) {
6234                 CERROR("%s: fail to find backend /lost+found: rc = %ld\n",
6235                        lfsck_lfsck2name(lfsck), PTR_ERR(parent));
6236                 RETURN_EXIT;
6237         }
6238
6239         /* It is normal that the /lost+found does not exist for ZFS backend. */
6240         if (!dt_object_exists(parent))
6241                 GOTO(out, rc = 0);
6242
6243         if (unlikely(!dt_try_as_dir(env, parent)))
6244                 GOTO(out, rc = -ENOTDIR);
6245
6246         CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n",
6247                lfsck_lfsck2name(lfsck));
6248
6249         com->lc_new_scanned = 0;
6250         iops = &parent->do_index_ops->dio_it;
6251         di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE);
6252         if (IS_ERR(di))
6253                 GOTO(out, rc = PTR_ERR(di));
6254
6255         rc = iops->load(env, di, 0);
6256         if (rc == 0)
6257                 rc = iops->next(env, di);
6258         else if (rc > 0)
6259                 rc = 0;
6260
6261         while (rc == 0) {
6262                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6263                     unlikely(!thread_is_running(thread)))
6264                         break;
6265
6266                 rc = iops->rec(env, di, (struct dt_rec *)ent,
6267                                LUDA_64BITHASH | LUDA_TYPE);
6268                 if (rc == 0)
6269                         rc = lfsck_unpack_ent(ent, &cookie, &type);
6270
6271                 if (unlikely(rc != 0)) {
6272                         CDEBUG(D_LFSCK, "%s: fail to iterate backend "
6273                                "/lost+found: rc = %d\n",
6274                                lfsck_lfsck2name(lfsck), rc);
6275
6276                         goto skip;
6277                 }
6278
6279                 /* skip dot and dotdot entries */
6280                 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
6281                         goto next;
6282
6283                 if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid)))
6284                         goto skip;
6285
6286                 if (fid_is_norm(&ent->lde_fid)) {
6287                         fld_range_set_mdt(range);
6288                         rc = fld_local_lookup(env, ss->ss_server_fld,
6289                                               fid_seq(&ent->lde_fid), range);
6290                         if (rc != 0)
6291                                 goto skip;
6292                 } else if (idx != 0) {
6293                         /* If the returned FID is IGIF, then there are three
6294                          * possible cases:
6295                          *
6296                          * 1) The object is upgraded from old Lustre-1.8 with
6297                          *    IGIF assigned to such object.
6298                          * 2) The object is a backend local object and is
6299                          *    invisible to client.
6300                          * 3) The object lost its LMV EA, and since there is
6301                          *    no FID-in-dirent for the orphan in the backend
6302                          *    /lost+found directory, then the low layer will
6303                          *    return IGIF for such object.
6304                          *
6305                          * For MDTx (x != 0), it is either case 2) or case 3),
6306                          * but from the LFSCK view, they are indistinguishable.
6307                          * To be safe, the LFSCK will keep it there and report
6308                          * some message, then the adminstrator can handle that
6309                          * furtherly.
6310                          *
6311                          * For MDT0, it is more possible the case 1). The LFSCK
6312                          * will handle the orphan as an upgraded object. */
6313                         CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID
6314                                "in the backend /lost+found on the MDT %04x, "
6315                                "to be safe, skip it.\n",
6316                                lfsck_lfsck2name(lfsck), ent->lde_namelen,
6317                                ent->lde_name, PFID(&ent->lde_fid), idx);
6318                         goto skip;
6319                 }
6320
6321                 rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent);
6322
6323 skip:
6324                 down_write(&com->lc_sem);
6325                 com->lc_new_scanned++;
6326                 ns->ln_local_lpf_scanned++;
6327                 if (rc > 0)
6328                         ns->ln_local_lpf_moved++;
6329                 else if (rc == 0)
6330                         ns->ln_local_lpf_skipped++;
6331                 else
6332                         ns->ln_local_lpf_failed++;
6333                 up_write(&com->lc_sem);
6334
6335                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6336                         break;
6337
6338 next:
6339                 lfsck_control_speed_by_self(com);
6340                 if (unlikely(!thread_is_running(thread))) {
6341                         rc = 0;
6342                         break;
6343                 }
6344
6345                 rc = iops->next(env, di);
6346         }
6347
6348         iops->put(env, di);
6349         iops->fini(env, di);
6350
6351         EXIT;
6352
6353 out:
6354         CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n",
6355                lfsck_lfsck2name(lfsck), rc);
6356
6357         lfsck_object_put(env, parent);
6358 }
6359
6360 /**
6361  * Rescan the striped directory after the master LMV EA reset.
6362  *
6363  * Sometimes, the master LMV EA of the striped directory maybe lost, so when
6364  * the namespace LFSCK engine scan the striped directory for the first time,
6365  * it will be regarded as a normal directory. As the LFSCK processing, some
6366  * other LFSCK instance on other MDT will find the shard of this striped dir,
6367  * and find that the master MDT-object of the striped directory lost its LMV
6368  * EA, then such remote LFSCK instance will regenerate the master LMV EA and
6369  * notify the LFSCK instance on this MDT to rescan the striped directory.
6370  *
6371  * \param[in] env       pointer to the thread context
6372  * \param[in] com       pointer to the lfsck component
6373  * \param[in] llu       the lfsck_lmv_unit that contains the striped directory
6374  *                      to be rescanned.
6375  *
6376  * \retval              positive number for success
6377  * \retval              0 for LFSCK stopped/paused
6378  * \retval              negative error number on failure
6379  */
6380 static int lfsck_namespace_rescan_striped_dir(const struct lu_env *env,
6381                                               struct lfsck_component *com,
6382                                               struct lfsck_lmv_unit *llu)
6383 {
6384         struct lfsck_thread_info        *info   = lfsck_env_info(env);
6385         struct lfsck_instance           *lfsck  = com->lc_lfsck;
6386         struct lfsck_assistant_data     *lad    = com->lc_data;
6387         struct dt_object                *dir;
6388         const struct dt_it_ops          *iops;
6389         struct dt_it                    *di;
6390         struct lu_dirent                *ent    =
6391                         (struct lu_dirent *)info->lti_key;
6392         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
6393         struct ptlrpc_thread            *thread = &lfsck->li_thread;
6394         struct lfsck_assistant_object   *lso    = NULL;
6395         struct lfsck_namespace_req      *lnr;
6396         struct lfsck_assistant_req      *lar;
6397         int                              rc;
6398         __u16                            type;
6399         ENTRY;
6400
6401         LASSERT(list_empty(&lad->lad_req_list));
6402
6403         lfsck->li_lmv = &llu->llu_lmv;
6404         lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
6405         rc = lfsck_open_dir(env, lfsck, 0);
6406         if (rc != 0)
6407                 RETURN(rc);
6408
6409         dir = lfsck->li_obj_dir;
6410         di = lfsck->li_di_dir;
6411         iops = &dir->do_index_ops->dio_it;
6412         do {
6413                 rc = iops->rec(env, di, (struct dt_rec *)ent,
6414                                lfsck->li_args_dir);
6415                 if (rc == 0)
6416                         rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
6417                                               &type);
6418
6419                 if (rc != 0) {
6420                         if (bk->lb_param & LPF_FAILOUT)
6421                                 GOTO(out, rc);
6422
6423                         goto next;
6424                 }
6425
6426                 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
6427                         goto next;
6428
6429                 if (lso == NULL) {
6430                         lso = lfsck_assistant_object_init(env,
6431                                 lfsck_dto2fid(dir), NULL,
6432                                 lfsck->li_pos_current.lp_oit_cookie, true);
6433                         if (IS_ERR(lso)) {
6434                                 if (bk->lb_param & LPF_FAILOUT)
6435                                         GOTO(out, rc = PTR_ERR(lso));
6436
6437                                 lso = NULL;
6438                                 goto next;
6439                         }
6440                 }
6441
6442                 lnr = lfsck_namespace_assistant_req_init(lfsck, lso, ent, type);
6443                 if (IS_ERR(lnr)) {
6444                         if (bk->lb_param & LPF_FAILOUT)
6445                                 GOTO(out, rc = PTR_ERR(lnr));
6446
6447                         goto next;
6448                 }
6449
6450                 lar = &lnr->lnr_lar;
6451                 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6452                 lfsck_namespace_assistant_req_fini(env, lar);
6453                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
6454                         GOTO(out, rc);
6455
6456                 if (unlikely(!thread_is_running(thread)))
6457                         GOTO(out, rc = 0);
6458
6459 next:
6460                 rc = iops->next(env, di);
6461         } while (rc == 0);
6462
6463 out:
6464         if (lso != NULL && !IS_ERR(lso))
6465                 lfsck_assistant_object_put(env, lso);
6466
6467         lfsck_close_dir(env, lfsck, rc);
6468         if (rc <= 0)
6469                 RETURN(rc);
6470
6471         /* The close_dir() may insert a dummy lnr in the lad->lad_req_list. */
6472         if (list_empty(&lad->lad_req_list))
6473                 RETURN(1);
6474
6475         spin_lock(&lad->lad_lock);
6476         lar = list_entry(lad->lad_req_list.next, struct lfsck_assistant_req,
6477                           lar_list);
6478         list_del_init(&lar->lar_list);
6479         spin_unlock(&lad->lad_lock);
6480
6481         rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
6482         lfsck_namespace_assistant_req_fini(env, lar);
6483
6484         RETURN(rc == 0 ? 1 : rc);
6485 }
6486
6487 static int
6488 lfsck_namespace_double_scan_one_trace_file(const struct lu_env *env,
6489                                            struct lfsck_component *com,
6490                                            struct dt_object *obj, bool first)
6491 {
6492         struct lfsck_instance   *lfsck  = com->lc_lfsck;
6493         struct ptlrpc_thread    *thread = &lfsck->li_thread;
6494         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
6495         struct lfsck_namespace  *ns     = com->lc_file_ram;
6496         const struct dt_it_ops  *iops   = &obj->do_index_ops->dio_it;
6497         struct dt_object        *target;
6498         struct dt_it            *di;
6499         struct dt_key           *key;
6500         struct lu_fid            fid;
6501         int                      rc;
6502         __u8                     flags  = 0;
6503         ENTRY;
6504
6505         di = iops->init(env, obj, 0);
6506         if (IS_ERR(di))
6507                 RETURN(PTR_ERR(di));
6508
6509         if (first)
6510                 fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
6511         else
6512                 fid_zero(&fid);
6513         rc = iops->get(env, di, (const struct dt_key *)&fid);
6514         if (rc < 0)
6515                 GOTO(fini, rc);
6516
6517         if (first) {
6518                 /* The start one either has been processed or does not exist,
6519                  * skip it. */
6520                 rc = iops->next(env, di);
6521                 if (rc != 0)
6522                         GOTO(put, rc);
6523         }
6524
6525         do {
6526                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
6527                     unlikely(!thread_is_running(thread)))
6528                         GOTO(put, rc = 0);
6529
6530                 key = iops->key(env, di);
6531                 if (IS_ERR(key)) {
6532                         rc = PTR_ERR(key);
6533                         if (rc == -ENOENT)
6534                                 GOTO(put, rc = 1);
6535
6536                         goto checkpoint;
6537                 }
6538
6539                 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
6540                 if (!fid_is_sane(&fid)) {
6541                         rc = 0;
6542                         goto checkpoint;
6543                 }
6544
6545                 target = lfsck_object_find_bottom(env, lfsck, &fid);
6546                 if (IS_ERR(target)) {
6547                         rc = PTR_ERR(target);
6548                         goto checkpoint;
6549                 }
6550
6551                 if (dt_object_exists(target)) {
6552                         rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
6553                         if (rc == 0) {
6554                                 rc = lfsck_namespace_double_scan_one(env, com,
6555                                                                 target, flags);
6556                                 if (rc == -ENOENT)
6557                                         rc = 0;
6558                         }
6559                 }
6560
6561                 lfsck_object_put(env, target);
6562
6563 checkpoint:
6564                 down_write(&com->lc_sem);
6565                 com->lc_new_checked++;
6566                 com->lc_new_scanned++;
6567                 if (rc >= 0)
6568                         ns->ln_fid_latest_scanned_phase2 = fid;
6569
6570                 if (rc > 0)
6571                         ns->ln_objs_repaired_phase2++;
6572                 else if (rc < 0)
6573                         ns->ln_objs_failed_phase2++;
6574                 up_write(&com->lc_sem);
6575
6576                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
6577                         GOTO(put, rc);
6578
6579                 if (unlikely(com->lc_time_next_checkpoint <=
6580                              ktime_get_seconds()) &&
6581                     com->lc_new_checked != 0) {
6582                         down_write(&com->lc_sem);
6583                         ns->ln_run_time_phase2 += ktime_get_seconds() -
6584                                                   com->lc_time_last_checkpoint;
6585                         ns->ln_time_last_checkpoint = ktime_get_real_seconds();
6586                         ns->ln_objs_checked_phase2 += com->lc_new_checked;
6587                         com->lc_new_checked = 0;
6588                         lfsck_namespace_store(env, com);
6589                         up_write(&com->lc_sem);
6590
6591                         com->lc_time_last_checkpoint = ktime_get_seconds();
6592                         com->lc_time_next_checkpoint =
6593                                 com->lc_time_last_checkpoint +
6594                                 LFSCK_CHECKPOINT_INTERVAL;
6595                 }
6596
6597                 lfsck_control_speed_by_self(com);
6598                 if (unlikely(!thread_is_running(thread)))
6599                         GOTO(put, rc = 0);
6600
6601                 rc = iops->next(env, di);
6602         } while (rc == 0);
6603
6604         GOTO(put, rc);
6605
6606 put:
6607         iops->put(env, di);
6608
6609 fini:
6610         iops->fini(env, di);
6611
6612         return rc;
6613 }
6614
6615 static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
6616                                                 struct lfsck_component *com)
6617 {
6618         struct lfsck_instance   *lfsck  = com->lc_lfsck;
6619         struct lfsck_namespace  *ns     = com->lc_file_ram;
6620         int                      rc;
6621         int                      i;
6622         ENTRY;
6623
6624         while (!list_empty(&lfsck->li_list_lmv)) {
6625                 struct lfsck_lmv_unit *llu;
6626
6627                 spin_lock(&lfsck->li_lock);
6628                 llu = list_entry(lfsck->li_list_lmv.next,
6629                                  struct lfsck_lmv_unit, llu_link);
6630                 list_del_init(&llu->llu_link);
6631                 spin_unlock(&lfsck->li_lock);
6632
6633                 rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
6634                 if (rc <= 0)
6635                         RETURN(rc);
6636         }
6637
6638         CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
6639                lfsck_lfsck2name(lfsck));
6640
6641         lfsck_namespace_scan_local_lpf(env, com);
6642
6643         com->lc_new_checked = 0;
6644         com->lc_new_scanned = 0;
6645         com->lc_time_last_checkpoint = ktime_get_seconds();
6646         com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6647                                        LFSCK_CHECKPOINT_INTERVAL;
6648
6649         i = lfsck_sub_trace_file_fid2idx(&ns->ln_fid_latest_scanned_phase2);
6650         rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6651                                 com->lc_sub_trace_objs[i].lsto_obj, true);
6652         while (rc > 0 && ++i < LFSCK_STF_COUNT)
6653                 rc = lfsck_namespace_double_scan_one_trace_file(env, com,
6654                                 com->lc_sub_trace_objs[i].lsto_obj, false);
6655
6656         CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop at the No. %d "
6657                "trace file: rc = %d\n", lfsck_lfsck2name(lfsck), i, rc);
6658
6659         RETURN(rc);
6660 }
6661
6662 static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env,
6663                                                struct lfsck_component *com,
6664                                                struct lfsck_position *pos)
6665 {
6666         struct lfsck_assistant_data     *lad = com->lc_data;
6667         struct lfsck_namespace_req      *lnr;
6668
6669         if (((struct lfsck_namespace *)(com->lc_file_ram))->ln_status !=
6670             LS_SCANNING_PHASE1)
6671                 return;
6672
6673         if (list_empty(&lad->lad_req_list))
6674                 return;
6675
6676         lnr = list_entry(lad->lad_req_list.next,
6677                          struct lfsck_namespace_req,
6678                          lnr_lar.lar_list);
6679         pos->lp_oit_cookie = lnr->lnr_lar.lar_parent->lso_oit_cookie;
6680         pos->lp_dir_cookie = lnr->lnr_dir_cookie - 1;
6681         pos->lp_dir_parent = lnr->lnr_lar.lar_parent->lso_fid;
6682 }
6683
6684 static int lfsck_namespace_double_scan_result(const struct lu_env *env,
6685                                               struct lfsck_component *com,
6686                                               int rc)
6687 {
6688         struct lfsck_instance   *lfsck  = com->lc_lfsck;
6689         struct lfsck_namespace  *ns     = com->lc_file_ram;
6690
6691         down_write(&com->lc_sem);
6692         ns->ln_run_time_phase2 += ktime_get_seconds() -
6693                                   com->lc_time_last_checkpoint;
6694         ns->ln_time_last_checkpoint = ktime_get_real_seconds();
6695         ns->ln_objs_checked_phase2 += com->lc_new_checked;
6696         com->lc_new_checked = 0;
6697
6698         if (rc > 0) {
6699                 if (ns->ln_flags & LF_INCOMPLETE)
6700                         ns->ln_status = LS_PARTIAL;
6701                 else
6702                         ns->ln_status = LS_COMPLETED;
6703                 ns->ln_flags &= ~LF_SCANNED_ONCE;
6704                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
6705                         ns->ln_flags &= ~LF_INCONSISTENT;
6706                 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
6707                 ns->ln_success_count++;
6708         } else if (rc == 0) {
6709                 if (lfsck->li_status != 0)
6710                         ns->ln_status = lfsck->li_status;
6711                 else
6712                         ns->ln_status = LS_STOPPED;
6713         } else {
6714                 ns->ln_status = LS_FAILED;
6715         }
6716
6717         rc = lfsck_namespace_store(env, com);
6718         up_write(&com->lc_sem);
6719
6720         return rc;
6721 }
6722
6723 static int
6724 lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
6725                                                   struct ptlrpc_request *req,
6726                                                   void *args, int rc)
6727 {
6728         if (rc == 0) {
6729                 struct lfsck_async_interpret_args *laia = args;
6730                 struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
6731
6732                 ltd->ltd_synced_failures = 1;
6733         }
6734
6735         return 0;
6736 }
6737
6738 /**
6739  * Notify remote LFSCK instances about former failures.
6740  *
6741  * The local LFSCK instance has recorded which MDTs have ever failed to respond
6742  * some LFSCK verification requests (maybe because of network issues or the MDT
6743  * itself trouble). During the respond gap the MDT may missed some name entries
6744  * verification, then the MDT cannot know whether related MDT-objects have been
6745  * referenced by related name entries or not, then in the second-stage scanning,
6746  * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
6747  * linkEA for back reference, then it will misguide the LFSCK to generate wrong
6748  * name entry for repairing the orphan.
6749  *
6750  * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
6751  * it will scan the bitmap for the ever failed MDTs, and notify them that they
6752  * have ever missed some name entries verification and should skip the handling
6753  * for orphan MDT-objects.
6754  *
6755  * \param[in] env       pointer to the thread context
6756  * \param[in] com       pointer to the lfsck component
6757  * \param[in] lr        pointer to the lfsck request
6758  */
6759 static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
6760                                                     struct lfsck_component *com,
6761                                                     struct lfsck_request *lr)
6762 {
6763         struct lfsck_async_interpret_args *laia  =
6764                                 &lfsck_env_info(env)->lti_laia2;
6765         struct lfsck_assistant_data       *lad   = com->lc_data;
6766         struct lfsck_namespace            *ns    = com->lc_file_ram;
6767         struct lfsck_instance             *lfsck = com->lc_lfsck;
6768         struct lfsck_tgt_descs            *ltds  = &lfsck->li_mdt_descs;
6769         struct lfsck_tgt_desc             *ltd;
6770         struct ptlrpc_request_set         *set;
6771         __u32                              idx;
6772         int                                rc    = 0;
6773         ENTRY;
6774
6775         if (!test_bit(LAD_INCOMPLETE, &lad->lad_flags))
6776                 RETURN_EXIT;
6777
6778         set = ptlrpc_prep_set();
6779         if (set == NULL)
6780                 GOTO(out, rc = -ENOMEM);
6781
6782         lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
6783         memset(laia, 0, sizeof(*laia));
6784         lad->lad_touch_gen++;
6785
6786         down_read(&ltds->ltd_rw_sem);
6787         cfs_foreach_bit(lad->lad_bitmap, idx) {
6788                 ltd = lfsck_ltd2tgt(ltds, idx);
6789                 if (unlikely(!ltd))
6790                         continue;
6791
6792                 laia->laia_ltd = ltd;
6793                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
6794                         lfsck_namespace_assistant_sync_failures_interpret,
6795                         laia, LFSCK_NOTIFY);
6796                 if (rc != 0)
6797                         CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6798                                "to sync failure with MDT %x: rc = %d\n",
6799                                lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
6800         }
6801         up_read(&ltds->ltd_rw_sem);
6802
6803         rc = ptlrpc_set_wait(env, set);
6804         ptlrpc_set_destroy(set);
6805
6806         GOTO(out, rc);
6807
6808 out:
6809         if (rc != 0)
6810                 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6811                        "to sync failure with MDTs, and related MDTs "
6812                        "may handle orphan improperly: rc = %d\n",
6813                        lfsck_lfsck2name(lfsck), rc);
6814
6815         EXIT;
6816 }
6817
6818 const struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
6819         .la_handler_p1          = lfsck_namespace_assistant_handler_p1,
6820         .la_handler_p2          = lfsck_namespace_assistant_handler_p2,
6821         .la_fill_pos            = lfsck_namespace_assistant_fill_pos,
6822         .la_double_scan_result  = lfsck_namespace_double_scan_result,
6823         .la_req_fini            = lfsck_namespace_assistant_req_fini,
6824         .la_sync_failures       = lfsck_namespace_assistant_sync_failures,
6825 };
6826
6827 /**
6828  * Verify the specified linkEA entry for the given directory object.
6829  * If the object has no such linkEA entry or it has more other linkEA
6830  * entries, then re-generate the linkEA with the given information.
6831  *
6832  * \param[in] env       pointer to the thread context
6833  * \param[in] obj       pointer to the dt_object to be handled
6834  * \param[in] cname     the name for the child in the parent directory
6835  * \param[in] pfid      the parent directory's FID for the linkEA
6836  *
6837  * \retval              0 for success
6838  * \retval              negative error number on failure
6839  */
6840 int lfsck_verify_linkea(const struct lu_env *env, struct dt_object *obj,
6841                         const struct lu_name *cname, const struct lu_fid *pfid)
6842 {
6843         struct dt_device        *dev    = lfsck_obj2dev(obj);
6844         struct linkea_data       ldata  = { NULL };
6845         struct lu_buf            linkea_buf;
6846         struct thandle          *th;
6847         int                      rc;
6848         int                      fl     = LU_XATTR_CREATE;
6849         bool                     dirty  = false;
6850         ENTRY;
6851
6852         LASSERT(S_ISDIR(lfsck_object_type(obj)));
6853
6854         rc = lfsck_links_read_with_rec(env, obj, &ldata);
6855         if (rc == -ENODATA) {
6856                 dirty = true;
6857         } else if (rc == 0) {
6858                 fl = LU_XATTR_REPLACE;
6859                 if (ldata.ld_leh->leh_reccount != 1) {
6860                         dirty = true;
6861                 } else {
6862                         rc = linkea_links_find(&ldata, cname, pfid);
6863                         if (rc != 0)
6864                                 dirty = true;
6865                 }
6866         }
6867
6868         if (!dirty)
6869                 RETURN(rc);
6870
6871         rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
6872                               cname, pfid);
6873         if (rc != 0)
6874                 RETURN(rc);
6875
6876         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
6877                        ldata.ld_leh->leh_len);
6878         th = dt_trans_create(env, dev);
6879         if (IS_ERR(th))
6880                 RETURN(PTR_ERR(th));
6881
6882         rc = dt_declare_xattr_set(env, obj, &linkea_buf,
6883                                   XATTR_NAME_LINK, fl, th);
6884         if (rc != 0)
6885                 GOTO(stop, rc);
6886
6887         rc = dt_trans_start_local(env, dev, th);
6888         if (rc != 0)
6889                 GOTO(stop, rc);
6890
6891         dt_write_lock(env, obj, 0);
6892         rc = dt_xattr_set(env, obj, &linkea_buf,
6893                           XATTR_NAME_LINK, fl, th);
6894         dt_write_unlock(env, obj);
6895
6896         GOTO(stop, rc);
6897
6898 stop:
6899         dt_trans_stop(env, dev, th);
6900         return rc;
6901 }
6902
6903 /**
6904  * Get the name and parent directory's FID from the first linkEA entry.
6905  *
6906  * \param[in] env       pointer to the thread context
6907  * \param[in] obj       pointer to the object which get linkEA from
6908  * \param[out] name     pointer to the buffer to hold the name
6909  *                      in the first linkEA entry
6910  * \param[out] pfid     pointer to the buffer to hold the parent
6911  *                      directory's FID in the first linkEA entry
6912  *
6913  * \retval              0 for success
6914  * \retval              negative error number on failure
6915  */
6916 int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
6917                           char *name, struct lu_fid *pfid)
6918 {
6919         struct lu_name           *cname = &lfsck_env_info(env)->lti_name;
6920         struct linkea_data        ldata = { NULL };
6921         int                       rc;
6922
6923         rc = lfsck_links_read_with_rec(env, obj, &ldata);
6924         if (rc)
6925                 return rc;
6926
6927         linkea_first_entry(&ldata);
6928         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
6929         if (!linkea_entry_is_valid(&ldata, cname, pfid))
6930                 return -EINVAL;
6931
6932         /* To guarantee the 'name' is terminated with '0'. */
6933         memcpy(name, cname->ln_name, cname->ln_namelen);
6934         name[cname->ln_namelen] = 0;
6935
6936         return 0;
6937 }
6938
6939 /**
6940  * Update the object's name entry with the given FID.
6941  *
6942  * \param[in] env       pointer to the thread context
6943  * \param[in] lfsck     pointer to the lfsck instance
6944  * \param[in] dir       pointer to the directory that holds
6945  *                      the name entry
6946  * \param[in] name      the name for the entry to be updated
6947  * \param[in] fid       the new FID for the name entry referenced
6948  * \param[in] type      the type for the name entry to be updated
6949  *
6950  * \retval              0 for success
6951  * \retval              negative error number on failure
6952  */
6953 int lfsck_update_name_entry(const struct lu_env *env,
6954                             struct lfsck_instance *lfsck,
6955                             struct dt_object *dir, const char *name,
6956                             const struct lu_fid *fid, __u32 type)
6957 {
6958         struct lfsck_thread_info *info   = lfsck_env_info(env);
6959         struct dt_insert_rec     *rec    = &info->lti_dt_rec;
6960         struct lfsck_lock_handle *llh    = &info->lti_llh;
6961         struct dt_device         *dev    = lfsck_obj2dev(dir);
6962         struct thandle           *th;
6963         int                       rc;
6964         bool                      exists = true;
6965         ENTRY;
6966
6967         rc = lfsck_lock(env, lfsck, dir, name, llh,
6968                         MDS_INODELOCK_UPDATE, LCK_PW);
6969         if (rc != 0)
6970                 RETURN(rc);
6971
6972         th = dt_trans_create(env, dev);
6973         if (IS_ERR(th))
6974                 GOTO(unlock, rc = PTR_ERR(th));
6975
6976         rc = dt_declare_delete(env, dir, (const struct dt_key *)name, th);
6977         if (rc != 0)
6978                 GOTO(stop, rc);
6979
6980         rec->rec_type = type;
6981         rec->rec_fid = fid;
6982         rc = dt_declare_insert(env, dir, (const struct dt_rec *)rec,
6983                                (const struct dt_key *)name, th);
6984         if (rc != 0)
6985                 GOTO(stop, rc);
6986
6987         rc = dt_declare_ref_add(env, dir, th);
6988         if (rc != 0)
6989                 GOTO(stop, rc);
6990
6991         rc = dt_trans_start_local(env, dev, th);
6992         if (rc != 0)
6993                 GOTO(stop, rc);
6994
6995         rc = dt_delete(env, dir, (const struct dt_key *)name, th);
6996         if (rc == -ENOENT) {
6997                 exists = false;
6998                 rc = 0;
6999         }
7000
7001         if (rc != 0)
7002                 GOTO(stop, rc);
7003
7004         rc = dt_insert(env, dir, (const struct dt_rec *)rec,
7005                        (const struct dt_key *)name, th);
7006         if (rc == 0 && S_ISDIR(type) && !exists) {
7007                 dt_write_lock(env, dir, 0);
7008                 rc = dt_ref_add(env, dir, th);
7009                 dt_write_unlock(env, dir);
7010         }
7011
7012         GOTO(stop, rc);
7013
7014 stop:
7015         dt_trans_stop(env, dev, th);
7016
7017 unlock:
7018         lfsck_unlock(llh);
7019         CDEBUG(D_LFSCK, "%s: update name entry "DFID"/%s with the FID "DFID
7020                " and the type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
7021                PFID(lfsck_dto2fid(dir)), name, PFID(fid), type, rc);
7022
7023         return rc;
7024 }
7025
7026 int lfsck_namespace_setup(const struct lu_env *env,
7027                           struct lfsck_instance *lfsck)
7028 {
7029         struct lfsck_component  *com;
7030         struct lfsck_namespace  *ns;
7031         struct dt_object        *root = NULL;
7032         struct dt_object        *obj;
7033         int                      i;
7034         int                      rc;
7035         ENTRY;
7036
7037         LASSERT(lfsck->li_master);
7038
7039         OBD_ALLOC_PTR(com);
7040         if (com == NULL)
7041                 RETURN(-ENOMEM);
7042
7043         INIT_LIST_HEAD(&com->lc_link);
7044         INIT_LIST_HEAD(&com->lc_link_dir);
7045         init_rwsem(&com->lc_sem);
7046         atomic_set(&com->lc_ref, 1);
7047         com->lc_lfsck = lfsck;
7048         com->lc_type = LFSCK_TYPE_NAMESPACE;
7049         com->lc_ops = &lfsck_namespace_ops;
7050         com->lc_data = lfsck_assistant_data_init(
7051                         &lfsck_namespace_assistant_ops,
7052                         LFSCK_NAMESPACE);
7053         if (com->lc_data == NULL)
7054                 GOTO(out, rc = -ENOMEM);
7055
7056         com->lc_file_size = sizeof(struct lfsck_namespace);
7057         OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
7058         if (com->lc_file_ram == NULL)
7059                 GOTO(out, rc = -ENOMEM);
7060
7061         OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
7062         if (com->lc_file_disk == NULL)
7063                 GOTO(out, rc = -ENOMEM);
7064
7065         for (i = 0; i < LFSCK_STF_COUNT; i++)
7066                 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
7067
7068         root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
7069         if (IS_ERR(root))
7070                 GOTO(out, rc = PTR_ERR(root));
7071
7072         if (unlikely(!dt_try_as_dir(env, root)))
7073                 GOTO(out, rc = -ENOTDIR);
7074
7075         obj = local_index_find_or_create(env, lfsck->li_los, root,
7076                                          LFSCK_NAMESPACE,
7077                                          S_IFREG | S_IRUGO | S_IWUSR,
7078                                          &dt_lfsck_namespace_features);
7079         if (IS_ERR(obj))
7080                 GOTO(out, rc = PTR_ERR(obj));
7081
7082         com->lc_obj = obj;
7083         rc = lfsck_namespace_load(env, com);
7084         if (rc == -ENODATA) {
7085                 rc = lfsck_namespace_init(env, com);
7086         } else if (rc < 0) {
7087                 rc = lfsck_namespace_reset(env, com, true);
7088         } else {
7089                 rc = lfsck_load_sub_trace_files(env, com,
7090                         &dt_lfsck_namespace_features, LFSCK_NAMESPACE, false);
7091                 if (rc)
7092                         rc = lfsck_namespace_reset(env, com, true);
7093         }
7094         if (rc != 0)
7095                 GOTO(out, rc);
7096
7097         ns = com->lc_file_ram;
7098         switch (ns->ln_status) {
7099         case LS_INIT:
7100         case LS_COMPLETED:
7101         case LS_FAILED:
7102         case LS_STOPPED:
7103                 spin_lock(&lfsck->li_lock);
7104                 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
7105                 spin_unlock(&lfsck->li_lock);
7106                 break;
7107         default:
7108                 CERROR("%s: unknown lfsck_namespace status %d\n",
7109                        lfsck_lfsck2name(lfsck), ns->ln_status);
7110                 /* fall through */
7111         case LS_SCANNING_PHASE1:
7112         case LS_SCANNING_PHASE2:
7113                 /* No need to store the status to disk right now.
7114                  * If the system crashed before the status stored,
7115                  * it will be loaded back when next time. */
7116                 ns->ln_status = LS_CRASHED;
7117                 /* fall through */
7118         case LS_PAUSED:
7119         case LS_CRASHED:
7120                 spin_lock(&lfsck->li_lock);
7121                 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
7122                 list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
7123                 spin_unlock(&lfsck->li_lock);
7124                 break;
7125         }
7126
7127         GOTO(out, rc = 0);
7128
7129 out:
7130         if (root != NULL && !IS_ERR(root))
7131                 lfsck_object_put(env, root);
7132         if (rc != 0) {
7133                 lfsck_component_cleanup(env, com);
7134                 CERROR("%s: fail to init namespace LFSCK component: rc = %d\n",
7135                        lfsck_lfsck2name(lfsck), rc);
7136         }
7137         return rc;
7138 }