lustre/mdd/mdd_lfsck.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2012, 2013, Intel Corporation.
  24  */
  25 /*
  26  * lustre/mdd/mdd_lfsck.c
  27  *
  28  * Top-level entry points into mdd module
  29  *
  30  * LFSCK controller, which scans the whole device through low layer
  31  * iteration APIs, drives all lfsck compeonents, controls the speed.
  32  *
  33  * Author: Fan Yong <yong.fan@whamcloud.com>
  34  */
  35
  36 #ifndef EXPORT_SYMTAB
  37 # define EXPORT_SYMTAB
  38 #endif
  39 #define DEBUG_SUBSYSTEM S_MDS
  40
  41 #include <lustre/lustre_idl.h>
  42 #include <lustre_fid.h>
  43 #include <obd_support.h>
  44
  45 #include "mdd_internal.h"
  46 #include "mdd_lfsck.h"
  47
  48 #define HALF_SEC                        (CFS_HZ >> 1)
  49 #define LFSCK_CHECKPOINT_INTERVAL       60
  50
  51 #define LFSCK_NAMEENTRY_DEAD            1 /* The object has been unlinked. */
  52 #define LFSCK_NAMEENTRY_REMOVED         2 /* The entry has been removed. */
  53 #define LFSCK_NAMEENTRY_RECREATED       3 /* The entry has been recreated. */
  54
  55 const char lfsck_bookmark_name[] = "lfsck_bookmark";
  56 const char lfsck_namespace_name[] = "lfsck_namespace";
  57
  58 static const char *lfsck_status_names[] = {
  59         "init",
  60         "scanning-phase1",
  61         "scanning-phase2",
  62         "completed",
  63         "failed",
  64         "stopped",
  65         "paused",
  66         "crashed",
  67         NULL
  68 };
  69
  70 static const char *lfsck_flags_names[] = {
  71         "scanned-once",
  72         "inconsistent",
  73         "upgrade",
  74         NULL
  75 };
  76
  77 static const char *lfsck_param_names[] = {
  78         "failout",
  79         "dryrun",
  80         NULL
  81 };
  82
  83 /* misc functions */
  84
  85 static inline struct mdd_device *mdd_lfsck2mdd(struct md_lfsck *lfsck)
  86 {
  87         return container_of0(lfsck, struct mdd_device, mdd_lfsck);
  88 }
  89
  90 static inline char *mdd_lfsck2name(struct md_lfsck *lfsck)
  91 {
  92         struct mdd_device *mdd = mdd_lfsck2mdd(lfsck);
  93
  94         return mdd2obd_dev(mdd)->obd_name;
  95 }
  96
  97 static inline void mdd_lfsck_component_get(struct lfsck_component *com)
  98 {
  99         atomic_inc(&com->lc_ref);
 100 }
 101
 102 static inline void mdd_lfsck_component_put(const struct lu_env *env,
 103                                            struct lfsck_component *com)
 104 {
 105         if (atomic_dec_and_test(&com->lc_ref)) {
 106                 if (com->lc_obj != NULL)
 107                         lu_object_put(env, &com->lc_obj->do_lu);
 108                 if (com->lc_file_ram != NULL)
 109                         OBD_FREE(com->lc_file_ram, com->lc_file_size);
 110                 if (com->lc_file_disk != NULL)
 111                         OBD_FREE(com->lc_file_disk, com->lc_file_size);
 112                 OBD_FREE_PTR(com);
 113         }
 114 }
 115
 116 static inline struct lfsck_component *
 117 __mdd_lfsck_component_find(struct md_lfsck *lfsck, __u16 type, cfs_list_t *list)
 118 {
 119         struct lfsck_component *com;
 120
 121         cfs_list_for_each_entry(com, list, lc_link) {
 122                 if (com->lc_type == type)
 123                         return com;
 124         }
 125         return NULL;
 126 }
 127
 128 static struct lfsck_component *
 129 mdd_lfsck_component_find(struct md_lfsck *lfsck, __u16 type)
 130 {
 131         struct lfsck_component *com;
 132
 133         spin_lock(&lfsck->ml_lock);
 134         com = __mdd_lfsck_component_find(lfsck, type, &lfsck->ml_list_scan);
 135         if (com != NULL)
 136                 goto unlock;
 137
 138         com = __mdd_lfsck_component_find(lfsck, type,
 139                                          &lfsck->ml_list_double_scan);
 140         if (com != NULL)
 141                 goto unlock;
 142
 143         com = __mdd_lfsck_component_find(lfsck, type, &lfsck->ml_list_idle);
 144
 145 unlock:
 146         if (com != NULL)
 147                 mdd_lfsck_component_get(com);
 148         spin_unlock(&lfsck->ml_lock);
 149         return com;
 150 }
 151
 152 static void mdd_lfsck_component_cleanup(const struct lu_env *env,
 153                                         struct lfsck_component *com)
 154 {
 155         if (!cfs_list_empty(&com->lc_link))
 156                 cfs_list_del_init(&com->lc_link);
 157         if (!cfs_list_empty(&com->lc_link_dir))
 158                 cfs_list_del_init(&com->lc_link_dir);
 159
 160         mdd_lfsck_component_put(env, com);
 161 }
 162
 163 static int lfsck_bits_dump(char **buf, int *len, int bits, const char *names[],
 164                            const char *prefix)
 165 {
 166         int save = *len;
 167         int flag;
 168         int rc;
 169         int i;
 170
 171         rc = snprintf(*buf, *len, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
 172         if (rc <= 0)
 173                 return -ENOSPC;
 174
 175         *buf += rc;
 176         *len -= rc;
 177         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
 178                 if (flag & bits) {
 179                         bits &= ~flag;
 180                         rc = snprintf(*buf, *len, "%s%c", names[i],
 181                                       bits != 0 ? ',' : '\n');
 182                         if (rc <= 0)
 183                                 return -ENOSPC;
 184
 185                         *buf += rc;
 186                         *len -= rc;
 187                 }
 188         }
 189         return save - *len;
 190 }
 191
 192 static int lfsck_time_dump(char **buf, int *len, __u64 time, const char *prefix)
 193 {
 194         int rc;
 195
 196         if (time != 0)
 197                 rc = snprintf(*buf, *len, "%s: "LPU64" seconds\n", prefix,
 198                               cfs_time_current_sec() - time);
 199         else
 200                 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
 201         if (rc <= 0)
 202                 return -ENOSPC;
 203
 204         *buf += rc;
 205         *len -= rc;
 206         return rc;
 207 }
 208
 209 static int lfsck_pos_dump(char **buf, int *len, struct lfsck_position *pos,
 210                           const char *prefix)
 211 {
 212         int rc;
 213
 214         if (fid_is_zero(&pos->lp_dir_parent)) {
 215                 if (pos->lp_oit_cookie == 0)
 216                         rc = snprintf(*buf, *len, "%s: N/A, N/A, N/A\n",
 217                                       prefix);
 218                 else
 219                         rc = snprintf(*buf, *len, "%s: "LPU64", N/A, N/A\n",
 220                                       prefix, pos->lp_oit_cookie);
 221         } else {
 222                 rc = snprintf(*buf, *len, "%s: "LPU64", "DFID", "LPU64"\n",
 223                               prefix, pos->lp_oit_cookie,
 224                               PFID(&pos->lp_dir_parent), pos->lp_dir_cookie);
 225         }
 226         if (rc <= 0)
 227                 return -ENOSPC;
 228
 229         *buf += rc;
 230         *len -= rc;
 231         return rc;
 232 }
 233
 234 static void mdd_lfsck_pos_fill(const struct lu_env *env, struct md_lfsck *lfsck,
 235                                struct lfsck_position *pos, bool init)
 236 {
 237         const struct dt_it_ops *iops = &lfsck->ml_obj_oit->do_index_ops->dio_it;
 238
 239         spin_lock(&lfsck->ml_lock);
 240         if (unlikely(lfsck->ml_di_oit == NULL)) {
 241                 spin_unlock(&lfsck->ml_lock);
 242                 memset(pos, 0, sizeof(*pos));
 243                 return;
 244         }
 245
 246         pos->lp_oit_cookie = iops->store(env, lfsck->ml_di_oit);
 247         if (!lfsck->ml_current_oit_processed && !init)
 248                 pos->lp_oit_cookie--;
 249
 250         LASSERT(pos->lp_oit_cookie > 0);
 251
 252         if (lfsck->ml_di_dir != NULL) {
 253                 struct dt_object *dto = lfsck->ml_obj_dir;
 254
 255                 pos->lp_dir_cookie = dto->do_index_ops->dio_it.store(env,
 256                                                         lfsck->ml_di_dir);
 257
 258                 if (pos->lp_dir_cookie >= MDS_DIR_END_OFF) {
 259                         fid_zero(&pos->lp_dir_parent);
 260                         pos->lp_dir_cookie = 0;
 261                 } else {
 262                         pos->lp_dir_parent = *lu_object_fid(&dto->do_lu);
 263                 }
 264         } else {
 265                 fid_zero(&pos->lp_dir_parent);
 266                 pos->lp_dir_cookie = 0;
 267         }
 268         spin_unlock(&lfsck->ml_lock);
 269 }
 270
 271 static inline void mdd_lfsck_pos_set_zero(struct lfsck_position *pos)
 272 {
 273         memset(pos, 0, sizeof(*pos));
 274 }
 275
 276 static inline int mdd_lfsck_pos_is_zero(const struct lfsck_position *pos)
 277 {
 278         return pos->lp_oit_cookie == 0 && fid_is_zero(&pos->lp_dir_parent);
 279 }
 280
 281 static inline int mdd_lfsck_pos_is_eq(const struct lfsck_position *pos1,
 282                                       const struct lfsck_position *pos2)
 283 {
 284         if (pos1->lp_oit_cookie < pos2->lp_oit_cookie)
 285                 return -1;
 286
 287         if (pos1->lp_oit_cookie > pos2->lp_oit_cookie)
 288                 return 1;
 289
 290         if (fid_is_zero(&pos1->lp_dir_parent) &&
 291             !fid_is_zero(&pos2->lp_dir_parent))
 292                 return -1;
 293
 294         if (!fid_is_zero(&pos1->lp_dir_parent) &&
 295             fid_is_zero(&pos2->lp_dir_parent))
 296                 return 1;
 297
 298         if (fid_is_zero(&pos1->lp_dir_parent) &&
 299             fid_is_zero(&pos2->lp_dir_parent))
 300                 return 0;
 301
 302         LASSERT(lu_fid_eq(&pos1->lp_dir_parent, &pos2->lp_dir_parent));
 303
 304         if (pos1->lp_dir_cookie < pos2->lp_dir_cookie)
 305                 return -1;
 306
 307         if (pos1->lp_dir_cookie > pos2->lp_dir_cookie)
 308                 return 1;
 309
 310         return 0;
 311 }
 312
 313 static void mdd_lfsck_close_dir(const struct lu_env *env,
 314                                 struct md_lfsck *lfsck)
 315 {
 316         struct dt_object        *dir_obj  = lfsck->ml_obj_dir;
 317         const struct dt_it_ops  *dir_iops = &dir_obj->do_index_ops->dio_it;
 318         struct dt_it            *dir_di   = lfsck->ml_di_dir;
 319
 320         spin_lock(&lfsck->ml_lock);
 321         lfsck->ml_di_dir = NULL;
 322         spin_unlock(&lfsck->ml_lock);
 323
 324         dir_iops->put(env, dir_di);
 325         dir_iops->fini(env, dir_di);
 326         lfsck->ml_obj_dir = NULL;
 327         lu_object_put(env, &dir_obj->do_lu);
 328 }
 329
 330 static void __mdd_lfsck_set_speed(struct md_lfsck *lfsck, __u32 limit)
 331 {
 332         lfsck->ml_bookmark_ram.lb_speed_limit = limit;
 333         if (limit != LFSCK_SPEED_NO_LIMIT) {
 334                 if (limit > CFS_HZ) {
 335                         lfsck->ml_sleep_rate = limit / CFS_HZ;
 336                         lfsck->ml_sleep_jif = 1;
 337                 } else {
 338                         lfsck->ml_sleep_rate = 1;
 339                         lfsck->ml_sleep_jif = CFS_HZ / limit;
 340                 }
 341         } else {
 342                 lfsck->ml_sleep_jif = 0;
 343                 lfsck->ml_sleep_rate = 0;
 344         }
 345 }
 346
 347 static void mdd_lfsck_control_speed(struct md_lfsck *lfsck)
 348 {
 349         struct ptlrpc_thread *thread = &lfsck->ml_thread;
 350         struct l_wait_info    lwi;
 351
 352         if (lfsck->ml_sleep_jif > 0 &&
 353             lfsck->ml_new_scanned >= lfsck->ml_sleep_rate) {
 354                 spin_lock(&lfsck->ml_lock);
 355                 if (likely(lfsck->ml_sleep_jif > 0 &&
 356                            lfsck->ml_new_scanned >= lfsck->ml_sleep_rate)) {
 357                         lwi = LWI_TIMEOUT_INTR(lfsck->ml_sleep_jif, NULL,
 358                                                LWI_ON_SIGNAL_NOOP, NULL);
 359                         spin_unlock(&lfsck->ml_lock);
 360
 361                         l_wait_event(thread->t_ctl_waitq,
 362                                      !thread_is_running(thread),
 363                                      &lwi);
 364                         lfsck->ml_new_scanned = 0;
 365                 } else {
 366                         spin_unlock(&lfsck->ml_lock);
 367                 }
 368         }
 369 }
 370
 371 /* lfsck_bookmark file ops */
 372
 373 static void inline mdd_lfsck_bookmark_to_cpu(struct lfsck_bookmark *des,
 374                                              struct lfsck_bookmark *src)
 375 {
 376         des->lb_magic = le32_to_cpu(src->lb_magic);
 377         des->lb_version = le16_to_cpu(src->lb_version);
 378         des->lb_param = le16_to_cpu(src->lb_param);
 379         des->lb_speed_limit = le32_to_cpu(src->lb_speed_limit);
 380 }
 381
 382 static void inline mdd_lfsck_bookmark_to_le(struct lfsck_bookmark *des,
 383                                             struct lfsck_bookmark *src)
 384 {
 385         des->lb_magic = cpu_to_le32(src->lb_magic);
 386         des->lb_version = cpu_to_le16(src->lb_version);
 387         des->lb_param = cpu_to_le16(src->lb_param);
 388         des->lb_speed_limit = cpu_to_le32(src->lb_speed_limit);
 389 }
 390
 391 static int mdd_lfsck_bookmark_load(const struct lu_env *env,
 392                                    struct md_lfsck *lfsck)
 393 {
 394         loff_t pos = 0;
 395         int    len = sizeof(struct lfsck_bookmark);
 396         int    rc;
 397
 398         rc = dt_record_read(env, lfsck->ml_bookmark_obj,
 399                             mdd_buf_get(env, &lfsck->ml_bookmark_disk, len),
 400                             &pos);
 401         if (rc == 0) {
 402                 struct lfsck_bookmark *bm = &lfsck->ml_bookmark_ram;
 403
 404                 mdd_lfsck_bookmark_to_cpu(bm, &lfsck->ml_bookmark_disk);
 405                 if (bm->lb_magic != LFSCK_BOOKMARK_MAGIC) {
 406                         CWARN("%.16s: invalid lfsck_bookmark magic "
 407                               "0x%x != 0x%x\n", mdd_lfsck2name(lfsck),
 408                               bm->lb_magic, LFSCK_BOOKMARK_MAGIC);
 409                         /* Process it as new lfsck_bookmark. */
 410                         rc = -ENODATA;
 411                 }
 412         } else {
 413                 if (rc == -EFAULT && pos == 0)
 414                         /* return -ENODATA for empty lfsck_bookmark. */
 415                         rc = -ENODATA;
 416                 else
 417                         CERROR("%.16s: fail to load lfsck_bookmark, "
 418                                "expected = %d, rc = %d\n",
 419                                mdd_lfsck2name(lfsck), len, rc);
 420         }
 421         return rc;
 422 }
 423
 424 static int mdd_lfsck_bookmark_store(const struct lu_env *env,
 425                                     struct md_lfsck *lfsck)
 426 {
 427         struct mdd_device *mdd    = mdd_lfsck2mdd(lfsck);
 428         struct thandle    *handle;
 429         struct dt_object  *obj    = lfsck->ml_bookmark_obj;
 430         loff_t             pos    = 0;
 431         int                len    = sizeof(struct lfsck_bookmark);
 432         int                rc;
 433         ENTRY;
 434
 435         mdd_lfsck_bookmark_to_le(&lfsck->ml_bookmark_disk,
 436                                  &lfsck->ml_bookmark_ram);
 437         handle = dt_trans_create(env, mdd->mdd_bottom);
 438         if (IS_ERR(handle)) {
 439                 rc = PTR_ERR(handle);
 440                 CERROR("%.16s: fail to create trans for storing "
 441                        "lfsck_bookmark: %d\n,", mdd_lfsck2name(lfsck), rc);
 442                 RETURN(rc);
 443         }
 444
 445         rc = dt_declare_record_write(env, obj, len, 0, handle);
 446         if (rc != 0) {
 447                 CERROR("%.16s: fail to declare trans for storing "
 448                        "lfsck_bookmark: %d\n,", mdd_lfsck2name(lfsck), rc);
 449                 GOTO(out, rc);
 450         }
 451
 452         rc = dt_trans_start_local(env, mdd->mdd_bottom, handle);
 453         if (rc != 0) {
 454                 CERROR("%.16s: fail to start trans for storing "
 455                        "lfsck_bookmark: %d\n,", mdd_lfsck2name(lfsck), rc);
 456                 GOTO(out, rc);
 457         }
 458
 459         rc = dt_record_write(env, obj,
 460                              mdd_buf_get(env, &lfsck->ml_bookmark_disk, len),
 461                              &pos, handle);
 462         if (rc != 0)
 463                 CERROR("%.16s: fail to store lfsck_bookmark, expected = %d, "
 464                        "rc = %d\n", mdd_lfsck2name(lfsck), len, rc);
 465
 466         GOTO(out, rc);
 467
 468 out:
 469         dt_trans_stop(env, mdd->mdd_bottom, handle);
 470         return rc;
 471 }
 472
 473 static int mdd_lfsck_bookmark_init(const struct lu_env *env,
 474                                    struct md_lfsck *lfsck)
 475 {
 476         struct lfsck_bookmark *mb = &lfsck->ml_bookmark_ram;
 477         int rc;
 478
 479         memset(mb, 0, sizeof(*mb));
 480         mb->lb_magic = LFSCK_BOOKMARK_MAGIC;
 481         mb->lb_version = LFSCK_VERSION_V2;
 482         mutex_lock(&lfsck->ml_mutex);
 483         rc = mdd_lfsck_bookmark_store(env, lfsck);
 484         mutex_unlock(&lfsck->ml_mutex);
 485         return rc;
 486 }
 487
 488 /* lfsck_namespace file ops */
 489
 490 static void inline mdd_lfsck_position_to_cpu(struct lfsck_position *des,
 491                                              struct lfsck_position *src)
 492 {
 493         des->lp_oit_cookie = le64_to_cpu(src->lp_oit_cookie);
 494         fid_le_to_cpu(&des->lp_dir_parent, &src->lp_dir_parent);
 495         des->lp_dir_cookie = le64_to_cpu(src->lp_dir_cookie);
 496 }
 497
 498 static void inline mdd_lfsck_position_to_le(struct lfsck_position *des,
 499                                              struct lfsck_position *src)
 500 {
 501         des->lp_oit_cookie = cpu_to_le64(src->lp_oit_cookie);
 502         fid_cpu_to_le(&des->lp_dir_parent, &src->lp_dir_parent);
 503         des->lp_dir_cookie = cpu_to_le64(src->lp_dir_cookie);
 504 }
 505
 506 static void inline mdd_lfsck_namespace_to_cpu(struct lfsck_namespace *des,
 507                                               struct lfsck_namespace *src)
 508 {
 509         des->ln_magic = le32_to_cpu(src->ln_magic);
 510         des->ln_status = le32_to_cpu(src->ln_status);
 511         des->ln_flags = le32_to_cpu(src->ln_flags);
 512         des->ln_success_count = le32_to_cpu(src->ln_success_count);
 513         des->ln_run_time_phase1 = le32_to_cpu(src->ln_run_time_phase1);
 514         des->ln_run_time_phase2 = le32_to_cpu(src->ln_run_time_phase2);
 515         des->ln_time_last_complete = le64_to_cpu(src->ln_time_last_complete);
 516         des->ln_time_latest_start = le64_to_cpu(src->ln_time_latest_start);
 517         des->ln_time_last_checkpoint =
 518                                 le64_to_cpu(src->ln_time_last_checkpoint);
 519         mdd_lfsck_position_to_cpu(&des->ln_pos_latest_start,
 520                                   &src->ln_pos_latest_start);
 521         mdd_lfsck_position_to_cpu(&des->ln_pos_last_checkpoint,
 522                                   &src->ln_pos_last_checkpoint);
 523         mdd_lfsck_position_to_cpu(&des->ln_pos_first_inconsistent,
 524                                   &src->ln_pos_first_inconsistent);
 525         des->ln_items_checked = le64_to_cpu(src->ln_items_checked);
 526         des->ln_items_repaired = le64_to_cpu(src->ln_items_repaired);
 527         des->ln_items_failed = le64_to_cpu(src->ln_items_failed);
 528         des->ln_dirs_checked = le64_to_cpu(src->ln_dirs_checked);
 529         des->ln_mlinked_checked = le64_to_cpu(src->ln_mlinked_checked);
 530         des->ln_objs_checked_phase2 = le64_to_cpu(src->ln_objs_checked_phase2);
 531         des->ln_objs_repaired_phase2 =
 532                                 le64_to_cpu(src->ln_objs_repaired_phase2);
 533         des->ln_objs_failed_phase2 = le64_to_cpu(src->ln_objs_failed_phase2);
 534         des->ln_objs_nlink_repaired = le64_to_cpu(src->ln_objs_nlink_repaired);
 535         des->ln_objs_lost_found = le64_to_cpu(src->ln_objs_lost_found);
 536         fid_le_to_cpu(&des->ln_fid_latest_scanned_phase2,
 537                       &src->ln_fid_latest_scanned_phase2);
 538 }
 539
 540 static void inline mdd_lfsck_namespace_to_le(struct lfsck_namespace *des,
 541                                              struct lfsck_namespace *src)
 542 {
 543         des->ln_magic = cpu_to_le32(src->ln_magic);
 544         des->ln_status = cpu_to_le32(src->ln_status);
 545         des->ln_flags = cpu_to_le32(src->ln_flags);
 546         des->ln_success_count = cpu_to_le32(src->ln_success_count);
 547         des->ln_run_time_phase1 = cpu_to_le32(src->ln_run_time_phase1);
 548         des->ln_run_time_phase2 = cpu_to_le32(src->ln_run_time_phase2);
 549         des->ln_time_last_complete = cpu_to_le64(src->ln_time_last_complete);
 550         des->ln_time_latest_start = cpu_to_le64(src->ln_time_latest_start);
 551         des->ln_time_last_checkpoint =
 552                                 cpu_to_le64(src->ln_time_last_checkpoint);
 553         mdd_lfsck_position_to_le(&des->ln_pos_latest_start,
 554                                  &src->ln_pos_latest_start);
 555         mdd_lfsck_position_to_le(&des->ln_pos_last_checkpoint,
 556                                  &src->ln_pos_last_checkpoint);
 557         mdd_lfsck_position_to_le(&des->ln_pos_first_inconsistent,
 558                                  &src->ln_pos_first_inconsistent);
 559         des->ln_items_checked = cpu_to_le64(src->ln_items_checked);
 560         des->ln_items_repaired = cpu_to_le64(src->ln_items_repaired);
 561         des->ln_items_failed = cpu_to_le64(src->ln_items_failed);
 562         des->ln_dirs_checked = cpu_to_le64(src->ln_dirs_checked);
 563         des->ln_mlinked_checked = cpu_to_le64(src->ln_mlinked_checked);
 564         des->ln_objs_checked_phase2 = cpu_to_le64(src->ln_objs_checked_phase2);
 565         des->ln_objs_repaired_phase2 =
 566                                 cpu_to_le64(src->ln_objs_repaired_phase2);
 567         des->ln_objs_failed_phase2 = cpu_to_le64(src->ln_objs_failed_phase2);
 568         des->ln_objs_nlink_repaired = cpu_to_le64(src->ln_objs_nlink_repaired);
 569         des->ln_objs_lost_found = cpu_to_le64(src->ln_objs_lost_found);
 570         fid_cpu_to_le(&des->ln_fid_latest_scanned_phase2,
 571                       &src->ln_fid_latest_scanned_phase2);
 572 }
 573
 574 /**
 575  * \retval +ve: the lfsck_namespace is broken, the caller should reset it.
 576  * \retval 0: succeed.
 577  * \retval -ve: failed cases.
 578  */
 579 static int mdd_lfsck_namespace_load(const struct lu_env *env,
 580                                     struct lfsck_component *com)
 581 {
 582         int len = com->lc_file_size;
 583         int rc;
 584
 585         rc = dt_xattr_get(env, com->lc_obj,
 586                           mdd_buf_get(env, com->lc_file_disk, len),
 587                           XATTR_NAME_LFSCK_NAMESPACE, BYPASS_CAPA);
 588         if (rc == len) {
 589                 struct lfsck_namespace *ns = com->lc_file_ram;
 590
 591                 mdd_lfsck_namespace_to_cpu(ns,
 592                                 (struct lfsck_namespace *)com->lc_file_disk);
 593                 if (ns->ln_magic != LFSCK_NAMESPACE_MAGIC) {
 594                         CWARN("%.16s: invalid lfsck_namespace magic "
 595                               "0x%x != 0x%x\n",
 596                               mdd_lfsck2name(com->lc_lfsck),
 597                               ns->ln_magic, LFSCK_NAMESPACE_MAGIC);
 598                         rc = 1;
 599                 } else {
 600                         rc = 0;
 601                 }
 602         } else if (rc != -ENODATA) {
 603                 CERROR("%.16s: fail to load lfsck_namespace, expected = %d, "
 604                        "rc = %d\n", mdd_lfsck2name(com->lc_lfsck), len, rc);
 605                 if (rc >= 0)
 606                         rc = 1;
 607         }
 608         return rc;
 609 }
 610
 611 static int mdd_lfsck_namespace_store(const struct lu_env *env,
 612                                      struct lfsck_component *com, bool init)
 613 {
 614         struct dt_object  *obj    = com->lc_obj;
 615         struct md_lfsck   *lfsck  = com->lc_lfsck;
 616         struct mdd_device *mdd    = mdd_lfsck2mdd(lfsck);
 617         struct thandle    *handle;
 618         int                len    = com->lc_file_size;
 619         int                rc;
 620         ENTRY;
 621
 622         mdd_lfsck_namespace_to_le((struct lfsck_namespace *)com->lc_file_disk,
 623                                   (struct lfsck_namespace *)com->lc_file_ram);
 624         handle = dt_trans_create(env, mdd->mdd_bottom);
 625         if (IS_ERR(handle)) {
 626                 rc = PTR_ERR(handle);
 627                 CERROR("%.16s: fail to create trans for storing "
 628                        "lfsck_namespace: %d\n,", mdd_lfsck2name(lfsck), rc);
 629                 RETURN(rc);
 630         }
 631
 632         rc = dt_declare_xattr_set(env, obj,
 633                                   mdd_buf_get(env, com->lc_file_disk, len),
 634                                   XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
 635         if (rc != 0) {
 636                 CERROR("%.16s: fail to declare trans for storing "
 637                        "lfsck_namespace: %d\n,", mdd_lfsck2name(lfsck), rc);
 638                 GOTO(out, rc);
 639         }
 640
 641         rc = dt_trans_start_local(env, mdd->mdd_bottom, handle);
 642         if (rc != 0) {
 643                 CERROR("%.16s: fail to start trans for storing "
 644                        "lfsck_namespace: %d\n,", mdd_lfsck2name(lfsck), rc);
 645                 GOTO(out, rc);
 646         }
 647
 648         rc = dt_xattr_set(env, obj,
 649                           mdd_buf_get(env, com->lc_file_disk, len),
 650                           XATTR_NAME_LFSCK_NAMESPACE,
 651                           init ? LU_XATTR_CREATE : LU_XATTR_REPLACE,
 652                           handle, BYPASS_CAPA);
 653         if (rc != 0)
 654                 CERROR("%.16s: fail to store lfsck_namespace, len = %d, "
 655                        "rc = %d\n", mdd_lfsck2name(lfsck), len, rc);
 656
 657         GOTO(out, rc);
 658
 659 out:
 660         dt_trans_stop(env, mdd->mdd_bottom, handle);
 661         return rc;
 662 }
 663
 664 static int mdd_lfsck_namespace_init(const struct lu_env *env,
 665                                     struct lfsck_component *com)
 666 {
 667         struct lfsck_namespace *ns = (struct lfsck_namespace *)com->lc_file_ram;
 668         int rc;
 669
 670         memset(ns, 0, sizeof(*ns));
 671         ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
 672         ns->ln_status = LS_INIT;
 673         down_write(&com->lc_sem);
 674         rc = mdd_lfsck_namespace_store(env, com, true);
 675         up_write(&com->lc_sem);
 676         return rc;
 677 }
 678
 679 static int mdd_lfsck_namespace_lookup(const struct lu_env *env,
 680                                       struct lfsck_component *com,
 681                                       const struct lu_fid *fid,
 682                                       __u8 *flags)
 683 {
 684         struct lu_fid *key = &mdd_env_info(env)->mti_fid;
 685         int            rc;
 686
 687         fid_cpu_to_be(key, fid);
 688         rc = dt_lookup(env, com->lc_obj, (struct dt_rec *)flags,
 689                        (const struct dt_key *)key, BYPASS_CAPA);
 690         return rc;
 691 }
 692
 693 static int mdd_lfsck_namespace_delete(const struct lu_env *env,
 694                                       struct lfsck_component *com,
 695                                       const struct lu_fid *fid)
 696 {
 697         struct mdd_device *mdd    = mdd_lfsck2mdd(com->lc_lfsck);
 698         struct lu_fid     *key    = &mdd_env_info(env)->mti_fid;
 699         struct thandle    *handle;
 700         struct dt_object *obj     = com->lc_obj;
 701         int               rc;
 702         ENTRY;
 703
 704         handle = dt_trans_create(env, mdd->mdd_bottom);
 705         if (IS_ERR(handle))
 706                 RETURN(PTR_ERR(handle));
 707
 708         rc = dt_declare_delete(env, obj, (const struct dt_key *)fid, handle);
 709         if (rc != 0)
 710                 GOTO(out, rc);
 711
 712         rc = dt_trans_start_local(env, mdd->mdd_bottom, handle);
 713         if (rc != 0)
 714                 GOTO(out, rc);
 715
 716         fid_cpu_to_be(key, fid);
 717         rc = dt_delete(env, obj, (const struct dt_key *)key, handle,
 718                        BYPASS_CAPA);
 719
 720         GOTO(out, rc);
 721
 722 out:
 723         dt_trans_stop(env, mdd->mdd_bottom, handle);
 724         return rc;
 725 }
 726
 727 static int mdd_lfsck_namespace_update(const struct lu_env *env,
 728                                       struct lfsck_component *com,
 729                                       const struct lu_fid *fid,
 730                                       __u8 flags, bool force)
 731 {
 732         struct mdd_device *mdd    = mdd_lfsck2mdd(com->lc_lfsck);
 733         struct lu_fid     *key    = &mdd_env_info(env)->mti_fid;
 734         struct thandle    *handle;
 735         struct dt_object *obj     = com->lc_obj;
 736         int               rc;
 737         bool              exist   = false;
 738         __u8              tf;
 739         ENTRY;
 740
 741         rc = mdd_lfsck_namespace_lookup(env, com, fid, &tf);
 742         if (rc != 0 && rc != -ENOENT)
 743                 RETURN(rc);
 744
 745         if (rc == 0) {
 746                 if (!force || flags == tf)
 747                         RETURN(0);
 748
 749                 exist = true;
 750                 handle = dt_trans_create(env, mdd->mdd_bottom);
 751                 if (IS_ERR(handle))
 752                         RETURN(PTR_ERR(handle));
 753
 754                 rc = dt_declare_delete(env, obj, (const struct dt_key *)fid,
 755                                        handle);
 756                 if (rc != 0)
 757                         GOTO(out, rc);
 758         } else {
 759                 handle = dt_trans_create(env, mdd->mdd_bottom);
 760                 if (IS_ERR(handle))
 761                         RETURN(PTR_ERR(handle));
 762         }
 763
 764         rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
 765                                (const struct dt_key *)fid, handle);
 766         if (rc != 0)
 767                 GOTO(out, rc);
 768
 769         rc = dt_trans_start_local(env, mdd->mdd_bottom, handle);
 770         if (rc != 0)
 771                 GOTO(out, rc);
 772
 773         fid_cpu_to_be(key, fid);
 774         if (exist) {
 775                 rc = dt_delete(env, obj, (const struct dt_key *)key, handle,
 776                                BYPASS_CAPA);
 777                 if (rc != 0) {
 778                         CERROR("%s: fail to insert "DFID", rc = %d\n",
 779                                mdd_lfsck2name(com->lc_lfsck), PFID(fid), rc);
 780                         GOTO(out, rc);
 781                 }
 782         }
 783
 784         rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
 785                        (const struct dt_key *)key, handle, BYPASS_CAPA, 1);
 786
 787         GOTO(out, rc);
 788
 789 out:
 790         dt_trans_stop(env, mdd->mdd_bottom, handle);
 791         return rc;
 792 }
 793
 794 /**
 795  * \retval +ve  repaired
 796  * \retval 0    no need to repair
 797  * \retval -ve  error cases
 798  */
 799 static int mdd_lfsck_namespace_double_scan_one(const struct lu_env *env,
 800                                                struct lfsck_component *com,
 801                                                struct mdd_object *child,
 802                                                __u8 flags)
 803 {
 804         struct mdd_thread_info  *info     = mdd_env_info(env);
 805         struct lu_attr          *la       = &info->mti_la;
 806         struct lu_name          *cname    = &info->mti_name;
 807         struct lu_fid           *pfid     = &info->mti_fid;
 808         struct lu_fid           *cfid     = &info->mti_fid2;
 809         struct md_lfsck         *lfsck    = com->lc_lfsck;
 810         struct mdd_device       *mdd      = mdd_lfsck2mdd(lfsck);
 811         struct lfsck_bookmark   *bk       = &lfsck->ml_bookmark_ram;
 812         struct lfsck_namespace  *ns       =
 813                                 (struct lfsck_namespace *)com->lc_file_ram;
 814         struct linkea_data       ldata    = { 0 };
 815         struct thandle          *handle   = NULL;
 816         bool                     locked   = false;
 817         bool                     update   = false;
 818         int                      count;
 819         int                      rc;
 820         ENTRY;
 821
 822         if (com->lc_journal) {
 823
 824 again:
 825                 LASSERT(!locked);
 826
 827                 com->lc_journal = 1;
 828                 handle = mdd_trans_create(env, mdd);
 829                 if (IS_ERR(handle))
 830                         RETURN(rc = PTR_ERR(handle));
 831
 832                 rc = mdd_declare_links_add(env, child, handle, NULL);
 833                 if (rc != 0)
 834                         GOTO(stop, rc);
 835
 836                 rc = mdd_trans_start(env, mdd, handle);
 837                 if (rc != 0)
 838                         GOTO(stop, rc);
 839
 840                 mdd_write_lock(env, child, MOR_TGT_CHILD);
 841                 locked = true;
 842         }
 843
 844         if (unlikely(mdd_is_dead_obj(child)))
 845                 GOTO(stop, rc = 0);
 846
 847         rc = mdd_links_read(env, child, &ldata);
 848         if (rc != 0) {
 849                 if ((bk->lb_param & LPF_DRYRUN) &&
 850                     (rc == -EINVAL || rc == -ENODATA))
 851                         rc = 1;
 852
 853                 GOTO(stop, rc);
 854         }
 855
 856         rc = mdd_la_get(env, child, la, BYPASS_CAPA);
 857         if (rc != 0)
 858                 GOTO(stop, rc);
 859
 860         ldata.ld_lee = LINKEA_FIRST_ENTRY(ldata);
 861         count = ldata.ld_leh->leh_reccount;
 862         while (count-- > 0) {
 863                 struct mdd_object *parent = NULL;
 864                 struct dt_object *dir;
 865
 866                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname,
 867                                     pfid);
 868                 if (!fid_is_sane(pfid))
 869                         goto shrink;
 870
 871                 parent = mdd_object_find(env, mdd, pfid);
 872                 if (parent == NULL)
 873                         goto shrink;
 874                 else if (IS_ERR(parent))
 875                         GOTO(stop, rc = PTR_ERR(parent));
 876
 877                 if (!mdd_object_exists(parent))
 878                         goto shrink;
 879
 880                 /* XXX: Currently, skip remote object, the consistency for
 881                  *      remote object will be processed in LFSCK phase III. */
 882                 if (mdd_object_remote(parent)) {
 883                         mdd_object_put(env, parent);
 884                         ldata.ld_lee = LINKEA_NEXT_ENTRY(ldata);
 885                         continue;
 886                 }
 887
 888                 dir = mdd_object_child(parent);
 889                 if (unlikely(!dt_try_as_dir(env, dir)))
 890                         goto shrink;
 891
 892                 /* To guarantee the 'name' is terminated with '0'. */
 893                 memcpy(info->mti_key, cname->ln_name, cname->ln_namelen);
 894                 info->mti_key[cname->ln_namelen] = 0;
 895                 cname->ln_name = info->mti_key;
 896                 rc = dt_lookup(env, dir, (struct dt_rec *)cfid,
 897                                (const struct dt_key *)cname->ln_name,
 898                                BYPASS_CAPA);
 899                 if (rc != 0 && rc != -ENOENT) {
 900                         mdd_object_put(env, parent);
 901                         GOTO(stop, rc);
 902                 }
 903
 904                 if (rc == 0) {
 905                         if (lu_fid_eq(cfid, mdo2fid(child))) {
 906                                 mdd_object_put(env, parent);
 907                                 ldata.ld_lee = LINKEA_NEXT_ENTRY(ldata);
 908                                 continue;
 909                         }
 910
 911                         goto shrink;
 912                 }
 913
 914                 if (ldata.ld_leh->leh_reccount > la->la_nlink)
 915                         goto shrink;
 916
 917                 /* XXX: For the case of there is linkea entry, but without name
 918                  *      entry pointing to the object, and the object link count
 919                  *      isn't less than the count of name entries, then add the
 920                  *      name entry back to namespace.
 921                  *
 922                  *      It is out of LFSCK 1.5 scope, will implement it in the
 923                  *      future. Keep the linkEA entry. */
 924                 mdd_object_put(env, parent);
 925                 ldata.ld_lee = LINKEA_NEXT_ENTRY(ldata);
 926                 continue;
 927
 928 shrink:
 929                 if (parent != NULL)
 930                         mdd_object_put(env, parent);
 931                 if (bk->lb_param & LPF_DRYRUN)
 932                         RETURN(1);
 933
 934                 CDEBUG(D_LFSCK, "Remove linkEA: "DFID"[%.*s], "DFID"\n",
 935                        PFID(mdo2fid(child)), cname->ln_namelen, cname->ln_name,
 936                        PFID(pfid));
 937                 linkea_del_buf(&ldata, cname);
 938                 update = true;
 939         }
 940
 941         if (update) {
 942                 if (!com->lc_journal) {
 943                         com->lc_journal = 1;
 944                         goto again;
 945                 }
 946
 947                 rc = mdd_links_write(env, child, &ldata, handle);
 948         }
 949
 950         GOTO(stop, rc);
 951
 952 stop:
 953         if (locked)
 954                 mdd_write_unlock(env, child);
 955
 956         if (handle != NULL)
 957                 mdd_trans_stop(env, mdd, rc, handle);
 958
 959         if (rc == 0 && update) {
 960                 ns->ln_objs_nlink_repaired++;
 961                 rc = 1;
 962         }
 963         return rc;
 964 }
 965
 966 /* namespace APIs */
 967
 968 static int mdd_lfsck_namespace_reset(const struct lu_env *env,
 969                                      struct lfsck_component *com, bool init)
 970 {
 971         struct lfsck_namespace  *ns   = (struct lfsck_namespace *)com->lc_file_ram;
 972         struct mdd_device       *mdd  = mdd_lfsck2mdd(com->lc_lfsck);
 973         struct dt_object        *dto, *root;
 974         int                      rc;
 975         ENTRY;
 976
 977         down_write(&com->lc_sem);
 978         if (init) {
 979                 memset(ns, 0, sizeof(*ns));
 980         } else {
 981                 __u32 count = ns->ln_success_count;
 982                 __u64 last_time = ns->ln_time_last_complete;
 983
 984                 memset(ns, 0, sizeof(*ns));
 985                 ns->ln_success_count = count;
 986                 ns->ln_time_last_complete = last_time;
 987         }
 988         ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
 989         ns->ln_status = LS_INIT;
 990
 991         root = dt_locate(env, mdd->mdd_bottom, &mdd->mdd_local_root_fid);
 992         if (unlikely(IS_ERR(root)))
 993                 GOTO(out, rc = PTR_ERR(root));
 994
 995         rc = local_object_unlink(env, mdd->mdd_bottom, root,
 996                                  lfsck_namespace_name);
 997         if (rc != 0)
 998                 GOTO(out, rc);
 999
1000         lu_object_put(env, &com->lc_obj->do_lu);
1001         com->lc_obj = NULL;
1002         dto = local_index_find_or_create(env, mdd->mdd_los, root,
1003                                          lfsck_namespace_name,
1004                                          S_IFREG | S_IRUGO | S_IWUSR,
1005                                          &dt_lfsck_features);
1006         if (IS_ERR(dto))
1007                 GOTO(out, rc = PTR_ERR(dto));
1008
1009         rc = dto->do_ops->do_index_try(env, dto, &dt_lfsck_features);
1010         if (rc != 0)
1011                 GOTO(out, rc);
1012         com->lc_obj = dto;
1013
1014         rc = mdd_lfsck_namespace_store(env, com, true);
1015
1016         GOTO(out, rc);
1017 out:
1018         lu_object_put(env, &root->do_lu);
1019         up_write(&com->lc_sem);
1020         return rc;
1021 }
1022
1023 static void
1024 mdd_lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
1025                          bool new_checked)
1026 {
1027         struct lfsck_namespace *ns = (struct lfsck_namespace *)com->lc_file_ram;
1028
1029         down_write(&com->lc_sem);
1030         if (new_checked)
1031                 com->lc_new_checked++;
1032         ns->ln_items_failed++;
1033         if (mdd_lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
1034                 mdd_lfsck_pos_fill(env, com->lc_lfsck,
1035                                    &ns->ln_pos_first_inconsistent, false);
1036         up_write(&com->lc_sem);
1037 }
1038
1039 static int mdd_lfsck_namespace_checkpoint(const struct lu_env *env,
1040                                           struct lfsck_component *com,
1041                                           bool init)
1042 {
1043         struct md_lfsck         *lfsck = com->lc_lfsck;
1044         struct lfsck_namespace  *ns    =
1045                                 (struct lfsck_namespace *)com->lc_file_ram;
1046         int                      rc;
1047
1048         if (com->lc_new_checked == 0 && !init)
1049                 return 0;
1050
1051         down_write(&com->lc_sem);
1052
1053         if (init) {
1054                 ns->ln_pos_latest_start = lfsck->ml_pos_current;
1055         } else {
1056                 ns->ln_pos_last_checkpoint = lfsck->ml_pos_current;
1057                 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
1058                                 HALF_SEC - lfsck->ml_time_last_checkpoint);
1059                 ns->ln_time_last_checkpoint = cfs_time_current_sec();
1060                 ns->ln_items_checked += com->lc_new_checked;
1061                 com->lc_new_checked = 0;
1062         }
1063
1064         rc = mdd_lfsck_namespace_store(env, com, false);
1065
1066         up_write(&com->lc_sem);
1067         return rc;
1068 }
1069
1070 static int mdd_lfsck_namespace_prep(const struct lu_env *env,
1071                                     struct lfsck_component *com)
1072 {
1073         struct md_lfsck         *lfsck  = com->lc_lfsck;
1074         struct lfsck_namespace  *ns     =
1075                                 (struct lfsck_namespace *)com->lc_file_ram;
1076         struct lfsck_position   *pos    = &com->lc_pos_start;
1077
1078         if (ns->ln_status == LS_COMPLETED) {
1079                 int rc;
1080
1081                 rc = mdd_lfsck_namespace_reset(env, com, false);
1082                 if (rc != 0)
1083                         return rc;
1084         }
1085
1086         down_write(&com->lc_sem);
1087
1088         ns->ln_time_latest_start = cfs_time_current_sec();
1089
1090         spin_lock(&lfsck->ml_lock);
1091         if (ns->ln_flags & LF_SCANNED_ONCE) {
1092                 if (!lfsck->ml_drop_dryrun ||
1093                     mdd_lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
1094                         ns->ln_status = LS_SCANNING_PHASE2;
1095                         cfs_list_del_init(&com->lc_link);
1096                         cfs_list_add_tail(&com->lc_link,
1097                                           &lfsck->ml_list_double_scan);
1098                         if (!cfs_list_empty(&com->lc_link_dir))
1099                                 cfs_list_del_init(&com->lc_link_dir);
1100                         mdd_lfsck_pos_set_zero(pos);
1101                 } else {
1102                         ns->ln_status = LS_SCANNING_PHASE1;
1103                         ns->ln_run_time_phase1 = 0;
1104                         ns->ln_run_time_phase2 = 0;
1105                         ns->ln_items_checked = 0;
1106                         ns->ln_items_repaired = 0;
1107                         ns->ln_items_failed = 0;
1108                         ns->ln_dirs_checked = 0;
1109                         ns->ln_mlinked_checked = 0;
1110                         ns->ln_objs_checked_phase2 = 0;
1111                         ns->ln_objs_repaired_phase2 = 0;
1112                         ns->ln_objs_failed_phase2 = 0;
1113                         ns->ln_objs_nlink_repaired = 0;
1114                         ns->ln_objs_lost_found = 0;
1115                         fid_zero(&ns->ln_fid_latest_scanned_phase2);
1116                         if (cfs_list_empty(&com->lc_link_dir))
1117                                 cfs_list_add_tail(&com->lc_link_dir,
1118                                                   &lfsck->ml_list_dir);
1119                         *pos = ns->ln_pos_first_inconsistent;
1120                 }
1121         } else {
1122                 ns->ln_status = LS_SCANNING_PHASE1;
1123                 if (cfs_list_empty(&com->lc_link_dir))
1124                         cfs_list_add_tail(&com->lc_link_dir,
1125                                           &lfsck->ml_list_dir);
1126                 if (!lfsck->ml_drop_dryrun ||
1127                     mdd_lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
1128                         *pos = ns->ln_pos_last_checkpoint;
1129                         pos->lp_oit_cookie++;
1130                 } else {
1131                         *pos = ns->ln_pos_first_inconsistent;
1132                 }
1133         }
1134         spin_unlock(&lfsck->ml_lock);
1135
1136         up_write(&com->lc_sem);
1137         return 0;
1138 }
1139
1140 static int mdd_lfsck_namespace_exec_oit(const struct lu_env *env,
1141                                         struct lfsck_component *com,
1142                                         struct mdd_object *obj)
1143 {
1144         down_write(&com->lc_sem);
1145         com->lc_new_checked++;
1146         if (S_ISDIR(mdd_object_type(obj)))
1147                 ((struct lfsck_namespace *)com->lc_file_ram)->ln_dirs_checked++;
1148         up_write(&com->lc_sem);
1149         return 0;
1150 }
1151
1152 static int mdd_declare_lfsck_namespace_exec_dir(const struct lu_env *env,
1153                                                 struct mdd_object *obj,
1154                                                 struct thandle *handle)
1155 {
1156         int rc;
1157
1158         /* For destroying all invalid linkEA entries. */
1159         rc = mdo_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
1160         if (rc != 0)
1161                 return rc;
1162
1163         /* For insert new linkEA entry. */
1164         rc = mdd_declare_links_add(env, obj, handle, NULL);
1165         return rc;
1166 }
1167
1168 static int mdd_lfsck_namespace_check_exist(const struct lu_env *env,
1169                                            struct md_lfsck *lfsck,
1170                                            struct mdd_object *obj,
1171                                            const char *name)
1172 {
1173         struct dt_object *dir = lfsck->ml_obj_dir;
1174         struct lu_fid    *fid = &mdd_env_info(env)->mti_fid;
1175         int               rc;
1176         ENTRY;
1177
1178         if (unlikely(mdd_is_dead_obj(obj)))
1179                 RETURN(LFSCK_NAMEENTRY_DEAD);
1180
1181         rc = dt_lookup(env, dir, (struct dt_rec *)fid,
1182                        (const struct dt_key *)name, BYPASS_CAPA);
1183         if (rc == -ENOENT)
1184                 RETURN(LFSCK_NAMEENTRY_REMOVED);
1185
1186         if (rc < 0)
1187                 RETURN(rc);
1188
1189         if (!lu_fid_eq(fid, mdo2fid(obj)))
1190                 RETURN(LFSCK_NAMEENTRY_RECREATED);
1191
1192         RETURN(0);
1193 }
1194
1195 static int mdd_lfsck_namespace_exec_dir(const struct lu_env *env,
1196                                         struct lfsck_component *com,
1197                                         struct mdd_object *obj,
1198                                         struct lu_dirent *ent)
1199 {
1200         struct mdd_thread_info     *info     = mdd_env_info(env);
1201         struct lu_attr             *la       = &info->mti_la;
1202         struct md_lfsck            *lfsck    = com->lc_lfsck;
1203         struct lfsck_bookmark      *bk       = &lfsck->ml_bookmark_ram;
1204         struct lfsck_namespace     *ns       =
1205                                 (struct lfsck_namespace *)com->lc_file_ram;
1206         struct mdd_device          *mdd      = mdd_lfsck2mdd(lfsck);
1207         struct linkea_data          ldata    = { 0 };
1208         const struct lu_fid        *pfid     =
1209                                 lu_object_fid(&lfsck->ml_obj_dir->do_lu);
1210         const struct lu_fid        *cfid     = mdo2fid(obj);
1211         const struct lu_name       *cname;
1212         struct thandle             *handle   = NULL;
1213         bool                        repaired = false;
1214         bool                        locked   = false;
1215         int                         count    = 0;
1216         int                         rc;
1217         ENTRY;
1218
1219         cname = mdd_name_get_const(env, ent->lde_name, ent->lde_namelen);
1220         down_write(&com->lc_sem);
1221         com->lc_new_checked++;
1222
1223         if (ent->lde_attrs & LUDA_UPGRADE) {
1224                 ns->ln_flags |= LF_UPGRADE;
1225                 repaired = true;
1226         } else if (ent->lde_attrs & LUDA_REPAIR) {
1227                 ns->ln_flags |= LF_INCONSISTENT;
1228                 repaired = true;
1229         }
1230
1231         if (ent->lde_name[0] == '.' &&
1232             (ent->lde_namelen == 1 ||
1233              (ent->lde_namelen == 2 && ent->lde_name[1] == '.') ||
1234              fid_is_dot_lustre(&ent->lde_fid)))
1235                 GOTO(out, rc = 0);
1236
1237         if (!(bk->lb_param & LPF_DRYRUN) &&
1238             (com->lc_journal || repaired)) {
1239
1240 again:
1241                 LASSERT(!locked);
1242
1243                 com->lc_journal = 1;
1244                 handle = mdd_trans_create(env, mdd);
1245                 if (IS_ERR(handle))
1246                         GOTO(out, rc = PTR_ERR(handle));
1247
1248                 rc = mdd_declare_lfsck_namespace_exec_dir(env, obj, handle);
1249                 if (rc != 0)
1250                         GOTO(stop, rc);
1251
1252                 rc = mdd_trans_start(env, mdd, handle);
1253                 if (rc != 0)
1254                         GOTO(stop, rc);
1255
1256                 mdd_write_lock(env, obj, MOR_TGT_CHILD);
1257                 locked = true;
1258         }
1259
1260         rc = mdd_lfsck_namespace_check_exist(env, lfsck, obj, ent->lde_name);
1261         if (rc != 0)
1262                 GOTO(stop, rc);
1263
1264         rc = mdd_links_read(env, obj, &ldata);
1265         if (rc == 0) {
1266                 count = ldata.ld_leh->leh_reccount;
1267                 rc = linkea_links_find(&ldata, cname, pfid);
1268                 if (rc == 0) {
1269                         /* For dir, if there are more than one linkea entries,
1270                          * then remove all the other redundant linkea entries.*/
1271                         if (unlikely(count > 1 &&
1272                                      S_ISDIR(mdd_object_type(obj))))
1273                                 goto unmatch;
1274
1275                         goto record;
1276                 } else {
1277
1278 unmatch:
1279                         ns->ln_flags |= LF_INCONSISTENT;
1280                         if (bk->lb_param & LPF_DRYRUN) {
1281                                 repaired = true;
1282                                 goto record;
1283                         }
1284
1285                         /*For dir, remove the unmatched linkea entry directly.*/
1286                         if (S_ISDIR(mdd_object_type(obj))) {
1287                                 if (!com->lc_journal)
1288                                         goto again;
1289
1290                                 rc = mdo_xattr_del(env, obj, XATTR_NAME_LINK,
1291                                                    handle, BYPASS_CAPA);
1292                                 if (rc != 0)
1293                                         GOTO(stop, rc);
1294
1295                                 goto nodata;
1296                         } else {
1297                                 goto add;
1298                         }
1299                 }
1300         } else if (unlikely(rc == -EINVAL)) {
1301                 ns->ln_flags |= LF_INCONSISTENT;
1302                 if (bk->lb_param & LPF_DRYRUN) {
1303                         count = 1;
1304                         repaired = true;
1305                         goto record;
1306                 }
1307
1308                 if (!com->lc_journal)
1309                         goto again;
1310
1311                 /* The magic crashed, we are not sure whether there are more
1312                  * corrupt data in the linkea, so remove all linkea entries. */
1313                 rc = mdo_xattr_del(env, obj, XATTR_NAME_LINK, handle,
1314                                    BYPASS_CAPA);
1315                 if (rc != 0)
1316                         GOTO(stop, rc);
1317
1318                 goto nodata;
1319         } else if (rc == -ENODATA) {
1320                 ns->ln_flags |= LF_UPGRADE;
1321                 if (bk->lb_param & LPF_DRYRUN) {
1322                         count = 1;
1323                         repaired = true;
1324                         goto record;
1325                 }
1326
1327 nodata:
1328                 rc = linkea_data_new(&ldata, &mdd_env_info(env)->mti_link_buf);
1329                 if (rc != 0)
1330                         GOTO(stop, rc);
1331
1332 add:
1333                 if (!com->lc_journal)
1334                         goto again;
1335
1336                 rc = linkea_add_buf(&ldata, cname, pfid);
1337                 if (rc != 0)
1338                         GOTO(stop, rc);
1339
1340                 rc = mdd_links_write(env, obj, &ldata, handle);
1341                 if (rc != 0)
1342                         GOTO(stop, rc);
1343
1344                 count = ldata.ld_leh->leh_reccount;
1345                 repaired = true;
1346         } else {
1347                 GOTO(stop, rc);
1348         }
1349
1350 record:
1351         LASSERT(count > 0);
1352
1353         rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
1354         if (rc != 0)
1355                 GOTO(stop, rc);
1356
1357         if ((count == 1) &&
1358             (la->la_nlink == 1 || S_ISDIR(mdd_object_type(obj))))
1359                 /* Usually, it is for single linked object or dir, do nothing.*/
1360                 GOTO(stop, rc);
1361
1362         /* Following modification will be in another transaction.  */
1363         if (handle != NULL) {
1364                 LASSERT(mdd_write_locked(env, obj));
1365
1366                 mdd_write_unlock(env, obj);
1367                 locked = false;
1368
1369                 mdd_trans_stop(env, mdd, 0, handle);
1370                 handle = NULL;
1371         }
1372
1373         ns->ln_mlinked_checked++;
1374         rc = mdd_lfsck_namespace_update(env, com, cfid,
1375                         count != la->la_nlink ? LLF_UNMATCH_NLINKS : 0, false);
1376
1377         GOTO(out, rc);
1378
1379 stop:
1380         if (locked)
1381                 mdd_write_unlock(env, obj);
1382
1383         if (handle != NULL)
1384                 mdd_trans_stop(env, mdd, rc, handle);
1385
1386 out:
1387         if (rc < 0) {
1388                 ns->ln_items_failed++;
1389                 if (mdd_lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
1390                         mdd_lfsck_pos_fill(env, lfsck,
1391                                            &ns->ln_pos_first_inconsistent,
1392                                            false);
1393                 if (!(bk->lb_param & LPF_FAILOUT))
1394                         rc = 0;
1395         } else {
1396                 if (repaired)
1397                         ns->ln_items_repaired++;
1398                 else
1399                         com->lc_journal = 0;
1400                 rc = 0;
1401         }
1402         up_write(&com->lc_sem);
1403         return rc;
1404 }
1405
1406 static int mdd_lfsck_namespace_post(const struct lu_env *env,
1407                                     struct lfsck_component *com,
1408                                     int result, bool init)
1409 {
1410         struct md_lfsck         *lfsck = com->lc_lfsck;
1411         struct lfsck_namespace  *ns    =
1412                                 (struct lfsck_namespace *)com->lc_file_ram;
1413         int                      rc;
1414
1415         down_write(&com->lc_sem);
1416
1417         spin_lock(&lfsck->ml_lock);
1418         if (!init)
1419                 ns->ln_pos_last_checkpoint = lfsck->ml_pos_current;
1420         if (result > 0) {
1421                 ns->ln_status = LS_SCANNING_PHASE2;
1422                 ns->ln_flags |= LF_SCANNED_ONCE;
1423                 ns->ln_flags &= ~LF_UPGRADE;
1424                 cfs_list_del_init(&com->lc_link);
1425                 cfs_list_del_init(&com->lc_link_dir);
1426                 cfs_list_add_tail(&com->lc_link, &lfsck->ml_list_double_scan);
1427         } else if (result == 0) {
1428                 if (lfsck->ml_paused) {
1429                         ns->ln_status = LS_PAUSED;
1430                 } else {
1431                         ns->ln_status = LS_STOPPED;
1432                         cfs_list_del_init(&com->lc_link);
1433                         cfs_list_del_init(&com->lc_link_dir);
1434                         cfs_list_add_tail(&com->lc_link, &lfsck->ml_list_idle);
1435                 }
1436         } else {
1437                 ns->ln_status = LS_FAILED;
1438                 cfs_list_del_init(&com->lc_link);
1439                 cfs_list_del_init(&com->lc_link_dir);
1440                 cfs_list_add_tail(&com->lc_link, &lfsck->ml_list_idle);
1441         }
1442         spin_unlock(&lfsck->ml_lock);
1443
1444         if (!init) {
1445                 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
1446                                 HALF_SEC - lfsck->ml_time_last_checkpoint);
1447                 ns->ln_time_last_checkpoint = cfs_time_current_sec();
1448                 ns->ln_items_checked += com->lc_new_checked;
1449                 com->lc_new_checked = 0;
1450         }
1451
1452         rc = mdd_lfsck_namespace_store(env, com, false);
1453
1454         up_write(&com->lc_sem);
1455         return rc;
1456 }
1457
1458 static int
1459 mdd_lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
1460                          char *buf, int len)
1461 {
1462         struct md_lfsck         *lfsck = com->lc_lfsck;
1463         struct lfsck_bookmark   *bk    = &lfsck->ml_bookmark_ram;
1464         struct lfsck_namespace  *ns    =
1465                                 (struct lfsck_namespace *)com->lc_file_ram;
1466         int                      save  = len;
1467         int                      ret   = -ENOSPC;
1468         int                      rc;
1469
1470         down_read(&com->lc_sem);
1471         rc = snprintf(buf, len,
1472                       "name: lfsck_namespace\n"
1473                       "magic: 0x%x\n"
1474                       "version: %d\n"
1475                       "status: %s\n",
1476                       ns->ln_magic,
1477                       bk->lb_version,
1478                       lfsck_status_names[ns->ln_status]);
1479         if (rc <= 0)
1480                 goto out;
1481
1482         buf += rc;
1483         len -= rc;
1484         rc = lfsck_bits_dump(&buf, &len, ns->ln_flags, lfsck_flags_names,
1485                              "flags");
1486         if (rc < 0)
1487                 goto out;
1488
1489         rc = lfsck_bits_dump(&buf, &len, bk->lb_param, lfsck_param_names,
1490                              "param");
1491         if (rc < 0)
1492                 goto out;
1493
1494         rc = lfsck_time_dump(&buf, &len, ns->ln_time_last_complete,
1495                              "time_since_last_completed");
1496         if (rc < 0)
1497                 goto out;
1498
1499         rc = lfsck_time_dump(&buf, &len, ns->ln_time_latest_start,
1500                              "time_since_latest_start");
1501         if (rc < 0)
1502                 goto out;
1503
1504         rc = lfsck_time_dump(&buf, &len, ns->ln_time_last_checkpoint,
1505                              "time_since_last_checkpoint");
1506         if (rc < 0)
1507                 goto out;
1508
1509         rc = lfsck_pos_dump(&buf, &len, &ns->ln_pos_latest_start,
1510                             "latest_start_position");
1511         if (rc < 0)
1512                 goto out;
1513
1514         rc = lfsck_pos_dump(&buf, &len, &ns->ln_pos_last_checkpoint,
1515                             "last_checkpoint_position");
1516         if (rc < 0)
1517                 goto out;
1518
1519         rc = lfsck_pos_dump(&buf, &len, &ns->ln_pos_first_inconsistent,
1520                             "first_failure_position");
1521         if (rc < 0)
1522                 goto out;
1523
1524         if (ns->ln_status == LS_SCANNING_PHASE1) {
1525                 struct lfsck_position pos;
1526                 cfs_duration_t duration = cfs_time_current() -
1527                                           lfsck->ml_time_last_checkpoint;
1528                 __u64 checked = ns->ln_items_checked + com->lc_new_checked;
1529                 __u64 speed = checked;
1530                 __u64 new_checked = com->lc_new_checked * CFS_HZ;
1531                 __u32 rtime = ns->ln_run_time_phase1 +
1532                               cfs_duration_sec(duration + HALF_SEC);
1533
1534                 if (duration != 0)
1535                         do_div(new_checked, duration);
1536                 if (rtime != 0)
1537                         do_div(speed, rtime);
1538                 rc = snprintf(buf, len,
1539                               "checked_phase1: "LPU64"\n"
1540                               "checked_phase2: "LPU64"\n"
1541                               "updated_phase1: "LPU64"\n"
1542                               "updated_phase2: "LPU64"\n"
1543                               "failed_phase1: "LPU64"\n"
1544                               "failed_phase2: "LPU64"\n"
1545                               "dirs: "LPU64"\n"
1546                               "M-linked: "LPU64"\n"
1547                               "nlinks_repaired: "LPU64"\n"
1548                               "lost_found: "LPU64"\n"
1549                               "success_count: %u\n"
1550                               "run_time_phase1: %u seconds\n"
1551                               "run_time_phase2: %u seconds\n"
1552                               "average_speed_phase1: "LPU64" items/sec\n"
1553                               "average_speed_phase2: N/A\n"
1554                               "real-time_speed_phase1: "LPU64" items/sec\n"
1555                               "real-time_speed_phase2: N/A\n",
1556                               checked,
1557                               ns->ln_objs_checked_phase2,
1558                               ns->ln_items_repaired,
1559                               ns->ln_objs_repaired_phase2,
1560                               ns->ln_items_failed,
1561                               ns->ln_objs_failed_phase2,
1562                               ns->ln_dirs_checked,
1563                               ns->ln_mlinked_checked,
1564                               ns->ln_objs_nlink_repaired,
1565                               ns->ln_objs_lost_found,
1566                               ns->ln_success_count,
1567                               rtime,
1568                               ns->ln_run_time_phase2,
1569                               speed,
1570                               new_checked);
1571                 if (rc <= 0)
1572                         goto out;
1573
1574                 buf += rc;
1575                 len -= rc;
1576                 mdd_lfsck_pos_fill(env, lfsck, &pos, false);
1577                 rc = lfsck_pos_dump(&buf, &len, &pos, "current_position");
1578                 if (rc <= 0)
1579                         goto out;
1580         } else if (ns->ln_status == LS_SCANNING_PHASE2) {
1581                 cfs_duration_t duration = cfs_time_current() -
1582                                           lfsck->ml_time_last_checkpoint;
1583                 __u64 checked = ns->ln_objs_checked_phase2 +
1584                                 com->lc_new_checked;
1585                 __u64 speed1 = ns->ln_items_checked;
1586                 __u64 speed2 = checked;
1587                 __u64 new_checked = com->lc_new_checked * CFS_HZ;
1588                 __u32 rtime = ns->ln_run_time_phase2 +
1589                               cfs_duration_sec(duration + HALF_SEC);
1590
1591                 if (duration != 0)
1592                         do_div(new_checked, duration);
1593                 if (ns->ln_run_time_phase1 != 0)
1594                         do_div(speed1, ns->ln_run_time_phase1);
1595                 if (rtime != 0)
1596                         do_div(speed2, rtime);
1597                 rc = snprintf(buf, len,
1598                               "checked_phase1: "LPU64"\n"
1599                               "checked_phase2: "LPU64"\n"
1600                               "updated_phase1: "LPU64"\n"
1601                               "updated_phase2: "LPU64"\n"
1602                               "failed_phase1: "LPU64"\n"
1603                               "failed_phase2: "LPU64"\n"
1604                               "dirs: "LPU64"\n"
1605                               "M-linked: "LPU64"\n"
1606                               "nlinks_repaired: "LPU64"\n"
1607                               "lost_found: "LPU64"\n"
1608                               "success_count: %u\n"
1609                               "run_time_phase1: %u seconds\n"
1610                               "run_time_phase2: %u seconds\n"
1611                               "average_speed_phase1: "LPU64" items/sec\n"
1612                               "average_speed_phase2: "LPU64" objs/sec\n"
1613                               "real-time_speed_phase1: N/A\n"
1614                               "real-time_speed_phase2: "LPU64" objs/sec\n"
1615                               "current_position: "DFID"\n",
1616                               ns->ln_items_checked,
1617                               checked,
1618                               ns->ln_items_repaired,
1619                               ns->ln_objs_repaired_phase2,
1620                               ns->ln_items_failed,
1621                               ns->ln_objs_failed_phase2,
1622                               ns->ln_dirs_checked,
1623                               ns->ln_mlinked_checked,
1624                               ns->ln_objs_nlink_repaired,
1625                               ns->ln_objs_lost_found,
1626                               ns->ln_success_count,
1627                               ns->ln_run_time_phase1,
1628                               rtime,
1629                               speed1,
1630                               speed2,
1631                               new_checked,
1632                               PFID(&ns->ln_fid_latest_scanned_phase2));
1633                 if (rc <= 0)
1634                         goto out;
1635
1636                 buf += rc;
1637                 len -= rc;
1638         } else {
1639                 __u64 speed1 = ns->ln_items_checked;
1640                 __u64 speed2 = ns->ln_objs_checked_phase2;
1641
1642                 if (ns->ln_run_time_phase1 != 0)
1643                         do_div(speed1, ns->ln_run_time_phase1);
1644                 if (ns->ln_run_time_phase2 != 0)
1645                         do_div(speed2, ns->ln_run_time_phase2);
1646                 rc = snprintf(buf, len,
1647                               "checked_phase1: "LPU64"\n"
1648                               "checked_phase2: "LPU64"\n"
1649                               "updated_phase1: "LPU64"\n"
1650                               "updated_phase2: "LPU64"\n"
1651                               "failed_phase1: "LPU64"\n"
1652                               "failed_phase2: "LPU64"\n"
1653                               "dirs: "LPU64"\n"
1654                               "M-linked: "LPU64"\n"
1655                               "nlinks_repaired: "LPU64"\n"
1656                               "lost_found: "LPU64"\n"
1657                               "success_count: %u\n"
1658                               "run_time_phase1: %u seconds\n"
1659                               "run_time_phase2: %u seconds\n"
1660                               "average_speed_phase1: "LPU64" items/sec\n"
1661                               "average_speed_phase2: "LPU64" objs/sec\n"
1662                               "real-time_speed_phase1: N/A\n"
1663                               "real-time_speed_phase2: N/A\n"
1664                               "current_position: N/A\n",
1665                               ns->ln_items_checked,
1666                               ns->ln_objs_checked_phase2,
1667                               ns->ln_items_repaired,
1668                               ns->ln_objs_repaired_phase2,
1669                               ns->ln_items_failed,
1670                               ns->ln_objs_failed_phase2,
1671                               ns->ln_dirs_checked,
1672                               ns->ln_mlinked_checked,
1673                               ns->ln_objs_nlink_repaired,
1674                               ns->ln_objs_lost_found,
1675                               ns->ln_success_count,
1676                               ns->ln_run_time_phase1,
1677                               ns->ln_run_time_phase2,
1678                               speed1,
1679                               speed2);
1680                 if (rc <= 0)
1681                         goto out;
1682
1683                 buf += rc;
1684                 len -= rc;
1685         }
1686         ret = save - len;
1687
1688 out:
1689         up_read(&com->lc_sem);
1690         return ret;
1691 }
1692
1693 static int mdd_lfsck_namespace_double_scan(const struct lu_env *env,
1694                                            struct lfsck_component *com)
1695 {
1696         struct md_lfsck         *lfsck  = com->lc_lfsck;
1697         struct ptlrpc_thread    *thread = &lfsck->ml_thread;
1698         struct mdd_device       *mdd    = mdd_lfsck2mdd(lfsck);
1699         struct lfsck_bookmark   *bk     = &lfsck->ml_bookmark_ram;
1700         struct lfsck_namespace  *ns     =
1701                                 (struct lfsck_namespace *)com->lc_file_ram;
1702         struct dt_object        *obj    = com->lc_obj;
1703         const struct dt_it_ops  *iops   = &obj->do_index_ops->dio_it;
1704         struct mdd_object       *target;
1705         struct dt_it            *di;
1706         struct dt_key           *key;
1707         struct lu_fid            fid;
1708         int                      rc;
1709         __u8                     flags;
1710         ENTRY;
1711
1712         lfsck->ml_new_scanned = 0;
1713         lfsck->ml_time_last_checkpoint = cfs_time_current();
1714         lfsck->ml_time_next_checkpoint = lfsck->ml_time_last_checkpoint +
1715                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
1716
1717         di = iops->init(env, obj, 0, BYPASS_CAPA);
1718         if (IS_ERR(di))
1719                 RETURN(PTR_ERR(di));
1720
1721         fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
1722         rc = iops->get(env, di, (const struct dt_key *)&fid);
1723         if (rc < 0)
1724                 GOTO(fini, rc);
1725
1726         /* Skip the start one, which either has been processed or non-exist. */
1727         rc = iops->next(env, di);
1728         if (rc != 0)
1729                 GOTO(put, rc);
1730
1731         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_DOUBLESCAN))
1732                 GOTO(put, rc = 0);
1733
1734         do {
1735                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY3) &&
1736                     cfs_fail_val > 0) {
1737                         struct l_wait_info lwi;
1738
1739                         lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
1740                                           NULL, NULL);
1741                         l_wait_event(thread->t_ctl_waitq,
1742                                      !thread_is_running(thread),
1743                                      &lwi);
1744                 }
1745
1746                 key = iops->key(env, di);
1747                 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
1748                 target = mdd_object_find(env, mdd, &fid);
1749                 down_write(&com->lc_sem);
1750                 if (target == NULL) {
1751                         rc = 0;
1752                         goto checkpoint;
1753                 } else if (IS_ERR(target)) {
1754                         rc = PTR_ERR(target);
1755                         goto checkpoint;
1756                 }
1757
1758                 /* XXX: Currently, skip remote object, the consistency for
1759                  *      remote object will be processed in LFSCK phase III. */
1760                 if (!mdd_object_exists(target) || mdd_object_remote(target))
1761                         goto obj_put;
1762
1763                 rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
1764                 if (rc == 0)
1765                         rc = mdd_lfsck_namespace_double_scan_one(env, com,
1766                                                                  target, flags);
1767
1768 obj_put:
1769                 mdd_object_put(env, target);
1770
1771 checkpoint:
1772                 lfsck->ml_new_scanned++;
1773                 com->lc_new_checked++;
1774                 ns->ln_fid_latest_scanned_phase2 = fid;
1775                 if (rc > 0)
1776                         ns->ln_objs_repaired_phase2++;
1777                 else if (rc < 0)
1778                         ns->ln_objs_failed_phase2++;
1779                 up_write(&com->lc_sem);
1780
1781                 if ((rc == 0) || ((rc > 0) && !(bk->lb_param & LPF_DRYRUN))) {
1782                         mdd_lfsck_namespace_delete(env, com, &fid);
1783                 } else if (rc < 0) {
1784                         flags |= LLF_REPAIR_FAILED;
1785                         mdd_lfsck_namespace_update(env, com, &fid, flags, true);
1786                 }
1787
1788                 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
1789                         GOTO(put, rc);
1790
1791                 if (likely(cfs_time_beforeq(cfs_time_current(),
1792                                             lfsck->ml_time_next_checkpoint)) ||
1793                     com->lc_new_checked == 0)
1794                         goto speed;
1795
1796                 down_write(&com->lc_sem);
1797                 ns->ln_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
1798                                 HALF_SEC - lfsck->ml_time_last_checkpoint);
1799                 ns->ln_time_last_checkpoint = cfs_time_current_sec();
1800                 ns->ln_objs_checked_phase2 += com->lc_new_checked;
1801                 com->lc_new_checked = 0;
1802                 rc = mdd_lfsck_namespace_store(env, com, false);
1803                 up_write(&com->lc_sem);
1804                 if (rc != 0)
1805                         GOTO(put, rc);
1806
1807                 lfsck->ml_time_last_checkpoint = cfs_time_current();
1808                 lfsck->ml_time_next_checkpoint = lfsck->ml_time_last_checkpoint +
1809                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
1810
1811 speed:
1812                 mdd_lfsck_control_speed(lfsck);
1813                 if (unlikely(!thread_is_running(thread)))
1814                         GOTO(put, rc = 0);
1815
1816                 rc = iops->next(env, di);
1817         } while (rc == 0);
1818
1819         GOTO(put, rc);
1820
1821 put:
1822         iops->put(env, di);
1823
1824 fini:
1825         iops->fini(env, di);
1826         down_write(&com->lc_sem);
1827
1828         ns->ln_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
1829                                 HALF_SEC - lfsck->ml_time_last_checkpoint);
1830         ns->ln_time_last_checkpoint = cfs_time_current_sec();
1831         ns->ln_objs_checked_phase2 += com->lc_new_checked;
1832         com->lc_new_checked = 0;
1833
1834         if (rc > 0) {
1835                 com->lc_journal = 0;
1836                 ns->ln_status = LS_COMPLETED;
1837                 if (!(bk->lb_param & LPF_DRYRUN))
1838                         ns->ln_flags &=
1839                         ~(LF_SCANNED_ONCE | LF_INCONSISTENT | LF_UPGRADE);
1840                 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
1841                 ns->ln_success_count++;
1842         } else if (rc == 0) {
1843                 if (lfsck->ml_paused)
1844                         ns->ln_status = LS_PAUSED;
1845                 else
1846                         ns->ln_status = LS_STOPPED;
1847         } else {
1848                 ns->ln_status = LS_FAILED;
1849         }
1850
1851         if (ns->ln_status != LS_PAUSED) {
1852                 spin_lock(&lfsck->ml_lock);
1853                 cfs_list_del_init(&com->lc_link);
1854                 cfs_list_add_tail(&com->lc_link, &lfsck->ml_list_idle);
1855                 spin_unlock(&lfsck->ml_lock);
1856         }
1857
1858         rc = mdd_lfsck_namespace_store(env, com, false);
1859
1860         up_write(&com->lc_sem);
1861         return rc;
1862 }
1863
1864 static struct lfsck_operations mdd_lfsck_namespace_ops = {
1865         .lfsck_reset            = mdd_lfsck_namespace_reset,
1866         .lfsck_fail             = mdd_lfsck_namespace_fail,
1867         .lfsck_checkpoint       = mdd_lfsck_namespace_checkpoint,
1868         .lfsck_prep             = mdd_lfsck_namespace_prep,
1869         .lfsck_exec_oit         = mdd_lfsck_namespace_exec_oit,
1870         .lfsck_exec_dir         = mdd_lfsck_namespace_exec_dir,
1871         .lfsck_post             = mdd_lfsck_namespace_post,
1872         .lfsck_dump             = mdd_lfsck_namespace_dump,
1873         .lfsck_double_scan      = mdd_lfsck_namespace_double_scan,
1874 };
1875
1876 /* LFSCK component setup/cleanup functions */
1877
1878 static int mdd_lfsck_namespace_setup(const struct lu_env *env,
1879                                      struct md_lfsck *lfsck)
1880 {
1881         struct mdd_device       *mdd = mdd_lfsck2mdd(lfsck);
1882         struct lfsck_component  *com;
1883         struct lfsck_namespace  *ns;
1884         struct dt_object        *obj, *root;
1885         int                      rc;
1886         ENTRY;
1887
1888         OBD_ALLOC_PTR(com);
1889         if (com == NULL)
1890                 RETURN(-ENOMEM);
1891
1892         CFS_INIT_LIST_HEAD(&com->lc_link);
1893         CFS_INIT_LIST_HEAD(&com->lc_link_dir);
1894         init_rwsem(&com->lc_sem);
1895         atomic_set(&com->lc_ref, 1);
1896         com->lc_lfsck = lfsck;
1897         com->lc_type = LT_NAMESPACE;
1898         com->lc_ops = &mdd_lfsck_namespace_ops;
1899         com->lc_file_size = sizeof(struct lfsck_namespace);
1900         OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
1901         if (com->lc_file_ram == NULL)
1902                 GOTO(out, rc = -ENOMEM);
1903
1904         OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
1905         if (com->lc_file_disk == NULL)
1906                 GOTO(out, rc = -ENOMEM);
1907
1908         root = dt_locate(env, mdd->mdd_bottom, &mdd->mdd_local_root_fid);
1909         if (unlikely(IS_ERR(root)))
1910                 GOTO(out, rc = PTR_ERR(root));
1911
1912         obj = local_index_find_or_create(env, mdd->mdd_los, root,
1913                                          lfsck_namespace_name,
1914                                          S_IFREG | S_IRUGO | S_IWUSR,
1915                                          &dt_lfsck_features);
1916         lu_object_put(env, &root->do_lu);
1917         if (IS_ERR(obj))
1918                 GOTO(out, rc = PTR_ERR(obj));
1919
1920         com->lc_obj = obj;
1921         rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_features);
1922         if (rc != 0)
1923                 GOTO(out, rc);
1924
1925         rc = mdd_lfsck_namespace_load(env, com);
1926         if (rc > 0)
1927                 rc = mdd_lfsck_namespace_reset(env, com, true);
1928         else if (rc == -ENODATA)
1929                 rc = mdd_lfsck_namespace_init(env, com);
1930         if (rc != 0)
1931                 GOTO(out, rc);
1932
1933         ns = (struct lfsck_namespace *)com->lc_file_ram;
1934         switch (ns->ln_status) {
1935         case LS_INIT:
1936         case LS_COMPLETED:
1937         case LS_FAILED:
1938         case LS_STOPPED:
1939                 cfs_list_add_tail(&com->lc_link, &lfsck->ml_list_idle);
1940                 break;
1941         default:
1942                 CERROR("%s: unknown status: %u\n",
1943                        mdd_lfsck2name(lfsck), ns->ln_status);
1944                 /* fall through */
1945         case LS_SCANNING_PHASE1:
1946         case LS_SCANNING_PHASE2:
1947                 /* No need to store the status to disk right now.
1948                  * If the system crashed before the status stored,
1949                  * it will be loaded back when next time. */
1950                 ns->ln_status = LS_CRASHED;
1951                 /* fall through */
1952         case LS_PAUSED:
1953         case LS_CRASHED:
1954                 cfs_list_add_tail(&com->lc_link, &lfsck->ml_list_scan);
1955                 cfs_list_add_tail(&com->lc_link_dir, &lfsck->ml_list_dir);
1956                 break;
1957         }
1958
1959         GOTO(out, rc = 0);
1960
1961 out:
1962         if (rc != 0)
1963                 mdd_lfsck_component_cleanup(env, com);
1964         return rc;
1965 }
1966
1967 /* helper functions for framework */
1968
1969 static int object_needs_lfsck(const struct lu_env *env, struct mdd_device *mdd,
1970                               struct mdd_object *obj)
1971 {
1972         struct lu_fid *fid   = &mdd_env_info(env)->mti_fid;
1973         int            depth = 0;
1974         int            rc;
1975
1976         LASSERT(S_ISDIR(mdd_object_type(obj)));
1977
1978         while (1) {
1979                 if (mdd_is_root(mdd, mdo2fid(obj))) {
1980                         if (depth > 0)
1981                                 mdd_object_put(env, obj);
1982                         return 1;
1983                 }
1984
1985                 /* .lustre doesn't contain "real" user objects, no need lfsck */
1986                 if (fid_is_dot_lustre(mdo2fid(obj))) {
1987                         if (depth > 0)
1988                                 mdd_object_put(env, obj);
1989                         return 0;
1990                 }
1991
1992                 mdd_read_lock(env, obj, MOR_TGT_CHILD);
1993                 if (unlikely(mdd_is_dead_obj(obj))) {
1994                         mdd_read_unlock(env, obj);
1995                         if (depth > 0)
1996                                 mdd_object_put(env, obj);
1997                         return 0;
1998                 }
1999
2000                 rc = dt_xattr_get(env, mdd_object_child(obj),
2001                                   mdd_buf_get(env, NULL, 0), XATTR_NAME_LINK,
2002                                   BYPASS_CAPA);
2003                 mdd_read_unlock(env, obj);
2004                 if (rc >= 0) {
2005                         if (depth > 0)
2006                                 mdd_object_put(env, obj);
2007                         return 1;
2008                 }
2009
2010                 if (rc < 0 && rc != -ENODATA) {
2011                         if (depth > 0)
2012                                 mdd_object_put(env, obj);
2013                         return rc;
2014                 }
2015
2016                 rc = mdd_parent_fid(env, obj, fid);
2017                 if (depth > 0)
2018                         mdd_object_put(env, obj);
2019                 if (rc != 0)
2020                         return rc;
2021
2022                 if (unlikely(lu_fid_eq(fid, &mdd->mdd_local_root_fid)))
2023                         return 0;
2024
2025                 obj = mdd_object_find(env, mdd, fid);
2026                 if (obj == NULL)
2027                         return 0;
2028                 else if (IS_ERR(obj))
2029                         return PTR_ERR(obj);
2030
2031                 if (!mdd_object_exists(obj)) {
2032                         mdd_object_put(env, obj);
2033                         return 0;
2034                 }
2035
2036                 /* Currently, only client visible directory can be remote. */
2037                 if (mdd_object_remote(obj)) {
2038                         mdd_object_put(env, obj);
2039                         return 1;
2040                 }
2041
2042                 depth++;
2043         }
2044         return 0;
2045 }
2046
2047 static void mdd_lfsck_unpack_ent(struct lu_dirent *ent)
2048 {
2049         fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
2050         ent->lde_hash = le64_to_cpu(ent->lde_hash);
2051         ent->lde_reclen = le16_to_cpu(ent->lde_reclen);
2052         ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
2053         ent->lde_attrs = le32_to_cpu(ent->lde_attrs);
2054
2055         /* Make sure the name is terminated with '0'.
2056          * The data (type) after ent::lde_name maybe
2057          * broken, but we do not care. */
2058         ent->lde_name[ent->lde_namelen] = 0;
2059 }
2060
2061 /* LFSCK wrap functions */
2062
2063 static void mdd_lfsck_fail(const struct lu_env *env, struct md_lfsck *lfsck,
2064                            bool new_checked)
2065 {
2066         struct lfsck_component *com;
2067
2068         cfs_list_for_each_entry(com, &lfsck->ml_list_scan, lc_link) {
2069                 com->lc_ops->lfsck_fail(env, com, new_checked);
2070         }
2071 }
2072
2073 static int mdd_lfsck_checkpoint(const struct lu_env *env,
2074                                 struct md_lfsck *lfsck)
2075 {
2076         struct lfsck_component *com;
2077         int                     rc;
2078
2079         if (likely(cfs_time_beforeq(cfs_time_current(),
2080                                     lfsck->ml_time_next_checkpoint)))
2081                 return 0;
2082
2083         mdd_lfsck_pos_fill(env, lfsck, &lfsck->ml_pos_current, false);
2084         cfs_list_for_each_entry(com, &lfsck->ml_list_scan, lc_link) {
2085                 rc = com->lc_ops->lfsck_checkpoint(env, com, false);
2086                 if (rc != 0)
2087                         return rc;;
2088         }
2089
2090         lfsck->ml_time_last_checkpoint = cfs_time_current();
2091         lfsck->ml_time_next_checkpoint = lfsck->ml_time_last_checkpoint +
2092                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
2093         return 0;
2094 }
2095
2096 static int mdd_lfsck_prep(struct lu_env *env, struct md_lfsck *lfsck)
2097 {
2098         struct mdd_device      *mdd     = mdd_lfsck2mdd(lfsck);
2099         struct mdd_object      *obj     = NULL;
2100         struct dt_object       *dt_obj;
2101         struct lfsck_component *com;
2102         struct lfsck_component *next;
2103         struct lfsck_position  *pos     = NULL;
2104         const struct dt_it_ops *iops    =
2105                                 &lfsck->ml_obj_oit->do_index_ops->dio_it;
2106         struct dt_it           *di;
2107         int                     rc;
2108         ENTRY;
2109
2110         LASSERT(lfsck->ml_obj_dir == NULL);
2111         LASSERT(lfsck->ml_di_dir == NULL);
2112
2113         lfsck->ml_current_oit_processed = 0;
2114         cfs_list_for_each_entry_safe(com, next, &lfsck->ml_list_scan, lc_link) {
2115                 com->lc_new_checked = 0;
2116                 if (lfsck->ml_bookmark_ram.lb_param & LPF_DRYRUN)
2117                         com->lc_journal = 0;
2118
2119                 rc = com->lc_ops->lfsck_prep(env, com);
2120                 if (rc != 0)
2121                         RETURN(rc);
2122
2123                 if ((pos == NULL) ||
2124                     (!mdd_lfsck_pos_is_zero(&com->lc_pos_start) &&
2125                      mdd_lfsck_pos_is_eq(pos, &com->lc_pos_start) > 0))
2126                         pos = &com->lc_pos_start;
2127         }
2128
2129         /* Init otable-based iterator. */
2130         if (pos == NULL) {
2131                 rc = iops->load(env, lfsck->ml_di_oit, 0);
2132                 if (rc > 0) {
2133                         lfsck->ml_oit_over = 1;
2134                         rc = 0;
2135                 }
2136
2137                 GOTO(out, rc);
2138         }
2139
2140         rc = iops->load(env, lfsck->ml_di_oit, pos->lp_oit_cookie);
2141         if (rc < 0)
2142                 GOTO(out, rc);
2143         else if (rc > 0)
2144                 lfsck->ml_oit_over = 1;
2145
2146         if (fid_is_zero(&pos->lp_dir_parent))
2147                 GOTO(out, rc = 0);
2148
2149         /* Find the directory for namespace-based traverse. */
2150         obj = mdd_object_find(env, mdd, &pos->lp_dir_parent);
2151         if (obj == NULL)
2152                 GOTO(out, rc = 0);
2153         else if (IS_ERR(obj))
2154                 RETURN(PTR_ERR(obj));
2155
2156         /* XXX: Currently, skip remote object, the consistency for
2157          *      remote object will be processed in LFSCK phase III. */
2158         if (!mdd_object_exists(obj) || mdd_object_remote(obj) ||
2159             unlikely(!S_ISDIR(mdd_object_type(obj))))
2160                 GOTO(out, rc = 0);
2161
2162         if (unlikely(mdd_is_dead_obj(obj)))
2163                 GOTO(out, rc = 0);
2164
2165         dt_obj = mdd_object_child(obj);
2166         if (unlikely(!dt_try_as_dir(env, dt_obj)))
2167                 GOTO(out, rc = -ENOTDIR);
2168
2169         /* Init the namespace-based directory traverse. */
2170         iops = &dt_obj->do_index_ops->dio_it;
2171         di = iops->init(env, dt_obj, lfsck->ml_args_dir, BYPASS_CAPA);
2172         if (IS_ERR(di))
2173                 GOTO(out, rc = PTR_ERR(di));
2174
2175         LASSERT(pos->lp_dir_cookie < MDS_DIR_END_OFF);
2176
2177         rc = iops->load(env, di, pos->lp_dir_cookie);
2178         if ((rc == 0) || (rc > 0 && pos->lp_dir_cookie > 0))
2179                 rc = iops->next(env, di);
2180         else if (rc > 0)
2181                 rc = 0;
2182
2183         if (rc != 0) {
2184                 iops->put(env, di);
2185                 iops->fini(env, di);
2186                 GOTO(out, rc);
2187         }
2188
2189         lfsck->ml_obj_dir = dt_obj;
2190         spin_lock(&lfsck->ml_lock);
2191         lfsck->ml_di_dir = di;
2192         spin_unlock(&lfsck->ml_lock);
2193         obj = NULL;
2194
2195         GOTO(out, rc = 0);
2196
2197 out:
2198         if (obj != NULL)
2199                 mdd_object_put(env, obj);
2200
2201         if (rc < 0) {
2202                 cfs_list_for_each_entry_safe(com, next, &lfsck->ml_list_scan,
2203                                              lc_link)
2204                         com->lc_ops->lfsck_post(env, com, rc, true);
2205
2206                 return rc;
2207         }
2208
2209         rc = 0;
2210         mdd_lfsck_pos_fill(env, lfsck, &lfsck->ml_pos_current, true);
2211         cfs_list_for_each_entry(com, &lfsck->ml_list_scan, lc_link) {
2212                 rc = com->lc_ops->lfsck_checkpoint(env, com, true);
2213                 if (rc != 0)
2214                         break;
2215         }
2216
2217         lfsck->ml_time_last_checkpoint = cfs_time_current();
2218         lfsck->ml_time_next_checkpoint = lfsck->ml_time_last_checkpoint +
2219                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
2220         return rc;
2221 }
2222
2223 static int mdd_lfsck_exec_oit(const struct lu_env *env, struct md_lfsck *lfsck,
2224                               struct mdd_object *obj)
2225 {
2226         struct lfsck_component *com;
2227         struct dt_object       *dt_obj;
2228         const struct dt_it_ops *iops;
2229         struct dt_it           *di;
2230         int                     rc;
2231         ENTRY;
2232
2233         LASSERT(lfsck->ml_obj_dir == NULL);
2234
2235         cfs_list_for_each_entry(com, &lfsck->ml_list_scan, lc_link) {
2236                 rc = com->lc_ops->lfsck_exec_oit(env, com, obj);
2237                 if (rc != 0)
2238                         RETURN(rc);
2239         }
2240
2241         if (!S_ISDIR(mdd_object_type(obj)) ||
2242             cfs_list_empty(&lfsck->ml_list_dir))
2243                RETURN(0);
2244
2245         rc = object_needs_lfsck(env, mdd_lfsck2mdd(lfsck), obj);
2246         if (rc <= 0)
2247                 GOTO(out, rc);
2248
2249         if (unlikely(mdd_is_dead_obj(obj)))
2250                 GOTO(out, rc = 0);
2251
2252         dt_obj = mdd_object_child(obj);
2253         if (unlikely(!dt_try_as_dir(env, dt_obj)))
2254                 GOTO(out, rc = -ENOTDIR);
2255
2256         iops = &dt_obj->do_index_ops->dio_it;
2257         di = iops->init(env, dt_obj, lfsck->ml_args_dir, BYPASS_CAPA);
2258         if (IS_ERR(di))
2259                 GOTO(out, rc = PTR_ERR(di));
2260
2261         rc = iops->load(env, di, 0);
2262         if (rc == 0)
2263                 rc = iops->next(env, di);
2264         else if (rc > 0)
2265                 rc = 0;
2266
2267         if (rc != 0) {
2268                 iops->put(env, di);
2269                 iops->fini(env, di);
2270                 GOTO(out, rc);
2271         }
2272
2273         mdd_object_get(obj);
2274         lfsck->ml_obj_dir = dt_obj;
2275         spin_lock(&lfsck->ml_lock);
2276         lfsck->ml_di_dir = di;
2277         spin_unlock(&lfsck->ml_lock);
2278
2279         GOTO(out, rc = 0);
2280
2281 out:
2282         if (rc < 0)
2283                 mdd_lfsck_fail(env, lfsck, false);
2284         return (rc > 0 ? 0 : rc);
2285 }
2286
2287 static int mdd_lfsck_exec_dir(const struct lu_env *env, struct md_lfsck *lfsck,
2288                               struct mdd_object *obj, struct lu_dirent *ent)
2289 {
2290         struct lfsck_component *com;
2291         int                     rc;
2292
2293         cfs_list_for_each_entry(com, &lfsck->ml_list_scan, lc_link) {
2294                 rc = com->lc_ops->lfsck_exec_dir(env, com, obj, ent);
2295                 if (rc != 0)
2296                         return rc;
2297         }
2298         return 0;
2299 }
2300
2301 static int mdd_lfsck_post(const struct lu_env *env, struct md_lfsck *lfsck,
2302                           int result)
2303 {
2304         struct lfsck_component *com;
2305         struct lfsck_component *next;
2306         int                     rc;
2307
2308         mdd_lfsck_pos_fill(env, lfsck, &lfsck->ml_pos_current, false);
2309         cfs_list_for_each_entry_safe(com, next, &lfsck->ml_list_scan, lc_link) {
2310                 rc = com->lc_ops->lfsck_post(env, com, result, false);
2311                 if (rc != 0)
2312                         return rc;
2313         }
2314
2315         lfsck->ml_time_last_checkpoint = cfs_time_current();
2316         lfsck->ml_time_next_checkpoint = lfsck->ml_time_last_checkpoint +
2317                                 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
2318         return result;
2319 }
2320
2321 static int mdd_lfsck_double_scan(const struct lu_env *env,
2322                                  struct md_lfsck *lfsck)
2323 {
2324         struct lfsck_component *com;
2325         struct lfsck_component *next;
2326         int                     rc;
2327
2328         cfs_list_for_each_entry_safe(com, next, &lfsck->ml_list_double_scan,
2329                                      lc_link) {
2330                 if (lfsck->ml_bookmark_ram.lb_param & LPF_DRYRUN)
2331                         com->lc_journal = 0;
2332
2333                 rc = com->lc_ops->lfsck_double_scan(env, com);
2334                 if (rc != 0)
2335                         return rc;
2336         }
2337         return 0;
2338 }
2339
2340 /* LFSCK engines */
2341
2342 static int mdd_lfsck_dir_engine(const struct lu_env *env,
2343                                 struct md_lfsck *lfsck)
2344 {
2345         struct mdd_thread_info  *info   = mdd_env_info(env);
2346         struct mdd_device       *mdd    = mdd_lfsck2mdd(lfsck);
2347         const struct dt_it_ops  *iops   =
2348                         &lfsck->ml_obj_dir->do_index_ops->dio_it;
2349         struct dt_it            *di     = lfsck->ml_di_dir;
2350         struct lu_dirent        *ent    = &info->mti_ent;
2351         struct lu_fid           *fid    = &info->mti_fid;
2352         struct lfsck_bookmark   *bk     = &lfsck->ml_bookmark_ram;
2353         struct ptlrpc_thread    *thread = &lfsck->ml_thread;
2354         int                      rc;
2355         ENTRY;
2356
2357         do {
2358                 struct mdd_object *child;
2359
2360                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY2) &&
2361                     cfs_fail_val > 0) {
2362                         struct l_wait_info lwi;
2363
2364                         lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
2365                                           NULL, NULL);
2366                         l_wait_event(thread->t_ctl_waitq,
2367                                      !thread_is_running(thread),
2368                                      &lwi);
2369                 }
2370
2371                 lfsck->ml_new_scanned++;
2372                 rc = iops->rec(env, di, (struct dt_rec *)ent,
2373                                lfsck->ml_args_dir);
2374                 if (rc != 0) {
2375                         mdd_lfsck_fail(env, lfsck, true);
2376                         if (bk->lb_param & LPF_FAILOUT)
2377                                 RETURN(rc);
2378                         else
2379                                 goto checkpoint;
2380                 }
2381
2382                 mdd_lfsck_unpack_ent(ent);
2383                 if (ent->lde_attrs & LUDA_IGNORE)
2384                         goto checkpoint;
2385
2386                 *fid = ent->lde_fid;
2387                 child = mdd_object_find(env, mdd, fid);
2388                 if (child == NULL) {
2389                         goto checkpoint;
2390                 } else if (IS_ERR(child)) {
2391                         mdd_lfsck_fail(env, lfsck, true);
2392                         if (bk->lb_param & LPF_FAILOUT)
2393                                 RETURN(PTR_ERR(child));
2394                         else
2395                                 goto checkpoint;
2396                 }
2397
2398                 /* XXX: Currently, skip remote object, the consistency for
2399                  *      remote object will be processed in LFSCK phase III. */
2400                 if (mdd_object_exists(child) && !mdd_object_remote(child))
2401                         rc = mdd_lfsck_exec_dir(env, lfsck, child, ent);
2402                 mdd_object_put(env, child);
2403                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
2404                         RETURN(rc);
2405
2406 checkpoint:
2407                 rc = mdd_lfsck_checkpoint(env, lfsck);
2408                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
2409                         RETURN(rc);
2410
2411                 /* Rate control. */
2412                 mdd_lfsck_control_speed(lfsck);
2413                 if (unlikely(!thread_is_running(thread)))
2414                         RETURN(0);
2415
2416                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL2)) {
2417                         spin_lock(&lfsck->ml_lock);
2418                         thread_set_flags(thread, SVC_STOPPING);
2419                         spin_unlock(&lfsck->ml_lock);
2420                         RETURN(-EINVAL);
2421                 }
2422
2423                 rc = iops->next(env, di);
2424         } while (rc == 0);
2425
2426         if (rc > 0 && !lfsck->ml_oit_over)
2427                 mdd_lfsck_close_dir(env, lfsck);
2428
2429         RETURN(rc);
2430 }
2431
2432 static int mdd_lfsck_oit_engine(const struct lu_env *env,
2433                                 struct md_lfsck *lfsck)
2434 {
2435         struct mdd_thread_info  *info   = mdd_env_info(env);
2436         struct mdd_device       *mdd    = mdd_lfsck2mdd(lfsck);
2437         const struct dt_it_ops  *iops   =
2438                                 &lfsck->ml_obj_oit->do_index_ops->dio_it;
2439         struct dt_it            *di     = lfsck->ml_di_oit;
2440         struct lu_fid           *fid    = &info->mti_fid;
2441         struct lfsck_bookmark   *bk     = &lfsck->ml_bookmark_ram;
2442         struct ptlrpc_thread    *thread = &lfsck->ml_thread;
2443         int                      rc;
2444         ENTRY;
2445
2446         do {
2447                 struct mdd_object *target;
2448
2449                 if (lfsck->ml_di_dir != NULL) {
2450                         rc = mdd_lfsck_dir_engine(env, lfsck);
2451                         if (rc <= 0)
2452                                 RETURN(rc);
2453                 }
2454
2455                 if (unlikely(lfsck->ml_oit_over))
2456                         RETURN(1);
2457
2458                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY1) &&
2459                     cfs_fail_val > 0) {
2460                         struct l_wait_info lwi;
2461
2462                         lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
2463                                           NULL, NULL);
2464                         l_wait_event(thread->t_ctl_waitq,
2465                                      !thread_is_running(thread),
2466                                      &lwi);
2467                 }
2468
2469                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
2470                         RETURN(0);
2471
2472                 lfsck->ml_current_oit_processed = 1;
2473                 lfsck->ml_new_scanned++;
2474                 rc = iops->rec(env, di, (struct dt_rec *)fid, 0);
2475                 if (rc != 0) {
2476                         mdd_lfsck_fail(env, lfsck, true);
2477                         if (bk->lb_param & LPF_FAILOUT)
2478                                 RETURN(rc);
2479                         else
2480                                 goto checkpoint;
2481                 }
2482
2483                 target = mdd_object_find(env, mdd, fid);
2484                 if (target == NULL) {
2485                         goto checkpoint;
2486                 } else if (IS_ERR(target)) {
2487                         mdd_lfsck_fail(env, lfsck, true);
2488                         if (bk->lb_param & LPF_FAILOUT)
2489                                 RETURN(PTR_ERR(target));
2490                         else
2491                                 goto checkpoint;
2492                 }
2493
2494                 /* XXX: Currently, skip remote object, the consistency for
2495                  *      remote object will be processed in LFSCK phase III. */
2496                 if (mdd_object_exists(target) && !mdd_object_remote(target))
2497                         rc = mdd_lfsck_exec_oit(env, lfsck, target);
2498                 mdd_object_put(env, target);
2499                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
2500                         RETURN(rc);
2501
2502 checkpoint:
2503                 rc = mdd_lfsck_checkpoint(env, lfsck);
2504                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
2505                         RETURN(rc);
2506
2507                 /* Rate control. */
2508                 mdd_lfsck_control_speed(lfsck);
2509
2510                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL1)) {
2511                         spin_lock(&lfsck->ml_lock);
2512                         thread_set_flags(thread, SVC_STOPPING);
2513                         spin_unlock(&lfsck->ml_lock);
2514                         RETURN(-EINVAL);
2515                 }
2516
2517                 rc = iops->next(env, di);
2518                 if (unlikely(rc > 0))
2519                         lfsck->ml_oit_over = 1;
2520                 else if (likely(rc == 0))
2521                         lfsck->ml_current_oit_processed = 0;
2522
2523                 if (unlikely(!thread_is_running(thread)))
2524                         RETURN(0);
2525         } while (rc == 0 || lfsck->ml_di_dir != NULL);
2526
2527         RETURN(rc);
2528 }
2529
2530 static int mdd_lfsck_main(void *args)
2531 {
2532         struct lu_env            env;
2533         struct md_lfsck         *lfsck    = (struct md_lfsck *)args;
2534         struct ptlrpc_thread    *thread   = &lfsck->ml_thread;
2535         struct dt_object        *oit_obj  = lfsck->ml_obj_oit;
2536         const struct dt_it_ops  *oit_iops = &oit_obj->do_index_ops->dio_it;
2537         struct dt_it            *oit_di;
2538         int                      rc;
2539         ENTRY;
2540
2541         cfs_daemonize("lfsck");
2542         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2543         if (rc != 0) {
2544                 CERROR("%s: LFSCK, fail to init env, rc = %d\n",
2545                        mdd_lfsck2name(lfsck), rc);
2546                 GOTO(noenv, rc);
2547         }
2548
2549         oit_di = oit_iops->init(&env, oit_obj, lfsck->ml_args_oit, BYPASS_CAPA);
2550         if (IS_ERR(oit_di)) {
2551                 rc = PTR_ERR(oit_di);
2552                 CERROR("%s: LFSCK, fail to init iteration, rc = %d\n",
2553                        mdd_lfsck2name(lfsck), rc);
2554                 GOTO(fini_env, rc);
2555         }
2556
2557         spin_lock(&lfsck->ml_lock);
2558         lfsck->ml_di_oit = oit_di;
2559         spin_unlock(&lfsck->ml_lock);
2560         rc = mdd_lfsck_prep(&env, lfsck);
2561         if (rc != 0)
2562                 GOTO(fini_oit, rc);
2563
2564         CDEBUG(D_LFSCK, "LFSCK entry: oit_flags = 0x%x, dir_flags = 0x%x, "
2565                "oit_cookie = "LPU64", dir_cookie = "LPU64", parent = "DFID
2566                ", pid = %d\n", lfsck->ml_args_oit, lfsck->ml_args_dir,
2567                lfsck->ml_pos_current.lp_oit_cookie,
2568                lfsck->ml_pos_current.lp_dir_cookie,
2569                PFID(&lfsck->ml_pos_current.lp_dir_parent),
2570                cfs_curproc_pid());
2571
2572         spin_lock(&lfsck->ml_lock);
2573         thread_set_flags(thread, SVC_RUNNING);
2574         spin_unlock(&lfsck->ml_lock);
2575         cfs_waitq_broadcast(&thread->t_ctl_waitq);
2576
2577         if (!cfs_list_empty(&lfsck->ml_list_scan) ||
2578             cfs_list_empty(&lfsck->ml_list_double_scan))
2579                 rc = mdd_lfsck_oit_engine(&env, lfsck);
2580         else
2581                 rc = 1;
2582
2583         CDEBUG(D_LFSCK, "LFSCK exit: oit_flags = 0x%x, dir_flags = 0x%x, "
2584                "oit_cookie = "LPU64", dir_cookie = "LPU64", parent = "DFID
2585                ", pid = %d, rc = %d\n", lfsck->ml_args_oit, lfsck->ml_args_dir,
2586                lfsck->ml_pos_current.lp_oit_cookie,
2587                lfsck->ml_pos_current.lp_dir_cookie,
2588                PFID(&lfsck->ml_pos_current.lp_dir_parent),
2589                cfs_curproc_pid(), rc);
2590
2591         if (lfsck->ml_paused && cfs_list_empty(&lfsck->ml_list_scan))
2592                 oit_iops->put(&env, oit_di);
2593
2594         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
2595                 rc = mdd_lfsck_post(&env, lfsck, rc);
2596         if (lfsck->ml_di_dir != NULL)
2597                 mdd_lfsck_close_dir(&env, lfsck);
2598
2599 fini_oit:
2600         spin_lock(&lfsck->ml_lock);
2601         lfsck->ml_di_oit = NULL;
2602         spin_unlock(&lfsck->ml_lock);
2603
2604         oit_iops->fini(&env, oit_di);
2605         if (rc == 1) {
2606                 if (!cfs_list_empty(&lfsck->ml_list_double_scan))
2607                         rc = mdd_lfsck_double_scan(&env, lfsck);
2608                 else
2609                         rc = 0;
2610         }
2611
2612         /* XXX: Purge the pinned objects in the future. */
2613
2614 fini_env:
2615         lu_env_fini(&env);
2616
2617 noenv:
2618         spin_lock(&lfsck->ml_lock);
2619         thread_set_flags(thread, SVC_STOPPED);
2620         cfs_waitq_broadcast(&thread->t_ctl_waitq);
2621         spin_unlock(&lfsck->ml_lock);
2622         return rc;
2623 }
2624
2625 /* external interfaces */
2626
2627 int mdd_lfsck_set_speed(const struct lu_env *env, struct md_lfsck *lfsck,
2628                         __u32 limit)
2629 {
2630         int rc;
2631
2632         mutex_lock(&lfsck->ml_mutex);
2633         __mdd_lfsck_set_speed(lfsck, limit);
2634         rc = mdd_lfsck_bookmark_store(env, lfsck);
2635         mutex_unlock(&lfsck->ml_mutex);
2636         return rc;
2637 }
2638
2639 int mdd_lfsck_dump(const struct lu_env *env, struct md_lfsck *lfsck,
2640                    __u16 type, char *buf, int len)
2641 {
2642         struct lfsck_component *com;
2643         int                     rc;
2644
2645         if (!lfsck->ml_initialized)
2646                 return -ENODEV;
2647
2648         com = mdd_lfsck_component_find(lfsck, type);
2649         if (com == NULL)
2650                 return -ENOTSUPP;
2651
2652         rc = com->lc_ops->lfsck_dump(env, com, buf, len);
2653         mdd_lfsck_component_put(env, com);
2654         return rc;
2655 }
2656
2657 int mdd_lfsck_start(const struct lu_env *env, struct md_lfsck *lfsck,
2658                     struct lfsck_start *start)
2659 {
2660         struct lfsck_bookmark  *bk     = &lfsck->ml_bookmark_ram;
2661         struct ptlrpc_thread   *thread = &lfsck->ml_thread;
2662         struct lfsck_component *com;
2663         struct l_wait_info      lwi    = { 0 };
2664         bool                    dirty  = false;
2665         int                     rc     = 0;
2666         __u16                   valid  = 0;
2667         __u16                   flags  = 0;
2668         ENTRY;
2669
2670         if (lfsck->ml_obj_oit == NULL)
2671                 RETURN(-ENOTSUPP);
2672
2673         /* start == NULL means auto trigger paused LFSCK. */
2674         if ((start == NULL) &&
2675             (cfs_list_empty(&lfsck->ml_list_scan) ||
2676              OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_AUTO)))
2677                 RETURN(0);
2678
2679         mutex_lock(&lfsck->ml_mutex);
2680         spin_lock(&lfsck->ml_lock);
2681         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2682                 spin_unlock(&lfsck->ml_lock);
2683                 mutex_unlock(&lfsck->ml_mutex);
2684                 RETURN(-EALREADY);
2685         }
2686
2687         spin_unlock(&lfsck->ml_lock);
2688
2689         lfsck->ml_paused = 0;
2690         lfsck->ml_oit_over = 0;
2691         lfsck->ml_drop_dryrun = 0;
2692         lfsck->ml_new_scanned = 0;
2693
2694         /* For auto trigger. */
2695         if (start == NULL)
2696                 goto trigger;
2697
2698         start->ls_version = bk->lb_version;
2699         if (start->ls_valid & LSV_SPEED_LIMIT) {
2700                 __mdd_lfsck_set_speed(lfsck, start->ls_speed_limit);
2701                 dirty = true;
2702         }
2703
2704         if (start->ls_valid & LSV_ERROR_HANDLE) {
2705                 valid |= DOIV_ERROR_HANDLE;
2706                 if (start->ls_flags & LPF_FAILOUT)
2707                         flags |= DOIF_FAILOUT;
2708
2709                 if ((start->ls_flags & LPF_FAILOUT) &&
2710                     !(bk->lb_param & LPF_FAILOUT)) {
2711                         bk->lb_param |= LPF_FAILOUT;
2712                         dirty = true;
2713                 } else if (!(start->ls_flags & LPF_FAILOUT) &&
2714                            (bk->lb_param & LPF_FAILOUT)) {
2715                         bk->lb_param &= ~LPF_FAILOUT;
2716                         dirty = true;
2717                 }
2718         }
2719
2720         if (start->ls_valid & LSV_DRYRUN) {
2721                 valid |= DOIV_DRYRUN;
2722                 if (start->ls_flags & LPF_DRYRUN)
2723                         flags |= DOIF_DRYRUN;
2724
2725                 if ((start->ls_flags & LPF_DRYRUN) &&
2726                     !(bk->lb_param & LPF_DRYRUN)) {
2727                         bk->lb_param |= LPF_DRYRUN;
2728                         dirty = true;
2729                 } else if (!(start->ls_flags & LPF_DRYRUN) &&
2730                            (bk->lb_param & LPF_DRYRUN)) {
2731                         bk->lb_param &= ~LPF_DRYRUN;
2732                         lfsck->ml_drop_dryrun = 1;
2733                         dirty = true;
2734                 }
2735         }
2736
2737         if (dirty) {
2738                 rc = mdd_lfsck_bookmark_store(env, lfsck);
2739                 if (rc != 0)
2740                         GOTO(out, rc);
2741         }
2742
2743         if (start->ls_flags & LPF_RESET)
2744                 flags |= DOIF_RESET;
2745
2746         if (start->ls_active != 0) {
2747                 struct lfsck_component *next;
2748                 __u16 type = 1;
2749
2750                 if (start->ls_active == LFSCK_TYPES_ALL)
2751                         start->ls_active = LFSCK_TYPES_SUPPORTED;
2752
2753                 if (start->ls_active & ~LFSCK_TYPES_SUPPORTED) {
2754                         start->ls_active &= ~LFSCK_TYPES_SUPPORTED;
2755                         GOTO(out, rc = -ENOTSUPP);
2756                 }
2757
2758                 cfs_list_for_each_entry_safe(com, next,
2759                                              &lfsck->ml_list_scan, lc_link) {
2760                         if (!(com->lc_type & start->ls_active)) {
2761                                 rc = com->lc_ops->lfsck_post(env, com, 0,
2762                                                              false);
2763                                 if (rc != 0)
2764                                         GOTO(out, rc);
2765                         }
2766                 }
2767
2768                 while (start->ls_active != 0) {
2769                         if (type & start->ls_active) {
2770                                 com = __mdd_lfsck_component_find(lfsck, type,
2771                                                         &lfsck->ml_list_idle);
2772                                 if (com != NULL) {
2773                                         /* The component status will be updated
2774                                          * when its prep() is called later by
2775                                          * the LFSCK main engine. */
2776                                         cfs_list_del_init(&com->lc_link);
2777                                         cfs_list_add_tail(&com->lc_link,
2778                                                           &lfsck->ml_list_scan);
2779                                 }
2780                                 start->ls_active &= ~type;
2781                         }
2782                         type <<= 1;
2783                 }
2784         }
2785
2786         cfs_list_for_each_entry(com, &lfsck->ml_list_scan, lc_link) {
2787                 start->ls_active |= com->lc_type;
2788                 if (flags & DOIF_RESET) {
2789                         rc = com->lc_ops->lfsck_reset(env, com, false);
2790                         if (rc != 0)
2791                                 GOTO(out, rc);
2792                 }
2793         }
2794
2795 trigger:
2796         lfsck->ml_args_dir = LUDA_64BITHASH | LUDA_VERIFY;
2797         if (bk->lb_param & LPF_DRYRUN) {
2798                 lfsck->ml_args_dir |= LUDA_VERIFY_DRYRUN;
2799                 valid |= DOIV_DRYRUN;
2800                 flags |= DOIF_DRYRUN;
2801         }
2802
2803         if (bk->lb_param & LPF_FAILOUT) {
2804                 valid |= DOIV_ERROR_HANDLE;
2805                 flags |= DOIF_FAILOUT;
2806         }
2807
2808         if (!cfs_list_empty(&lfsck->ml_list_scan))
2809                 flags |= DOIF_OUTUSED;
2810
2811         lfsck->ml_args_oit = (flags << DT_OTABLE_IT_FLAGS_SHIFT) | valid;
2812         thread_set_flags(thread, 0);
2813         rc = cfs_create_thread(mdd_lfsck_main, lfsck, 0);
2814         if (rc < 0)
2815                 CERROR("%s: cannot start LFSCK thread, rc = %d\n",
2816                        mdd_lfsck2name(lfsck), rc);
2817         else
2818                 l_wait_event(thread->t_ctl_waitq,
2819                              thread_is_running(thread) ||
2820                              thread_is_stopped(thread),
2821                              &lwi);
2822
2823         GOTO(out, rc = 0);
2824
2825 out:
2826         mutex_unlock(&lfsck->ml_mutex);
2827         return (rc < 0 ? rc : 0);
2828 }
2829
2830 int mdd_lfsck_stop(const struct lu_env *env, struct md_lfsck *lfsck,
2831                    bool pause)
2832 {
2833         struct ptlrpc_thread *thread = &lfsck->ml_thread;
2834         struct l_wait_info    lwi    = { 0 };
2835         ENTRY;
2836
2837         if (!lfsck->ml_initialized)
2838                 RETURN(0);
2839
2840         mutex_lock(&lfsck->ml_mutex);
2841         spin_lock(&lfsck->ml_lock);
2842         if (thread_is_init(thread) || thread_is_stopped(thread)) {
2843                 spin_unlock(&lfsck->ml_lock);
2844                 mutex_unlock(&lfsck->ml_mutex);
2845                 RETURN(-EALREADY);
2846         }
2847
2848         if (pause)
2849                 lfsck->ml_paused = 1;
2850         thread_set_flags(thread, SVC_STOPPING);
2851         /* The LFSCK thread may be sleeping on low layer wait queue,
2852          * wake it up. */
2853         if (likely(lfsck->ml_di_oit != NULL))
2854                 lfsck->ml_obj_oit->do_index_ops->dio_it.put(env,
2855                                                             lfsck->ml_di_oit);
2856         spin_unlock(&lfsck->ml_lock);
2857
2858         cfs_waitq_broadcast(&thread->t_ctl_waitq);
2859         l_wait_event(thread->t_ctl_waitq,
2860                      thread_is_stopped(thread),
2861                      &lwi);
2862         mutex_unlock(&lfsck->ml_mutex);
2863
2864         RETURN(0);
2865 }
2866
2867 static const struct lu_fid lfsck_it_fid = { .f_seq = FID_SEQ_LOCAL_FILE,
2868                                             .f_oid = OTABLE_IT_OID,
2869                                             .f_ver = 0 };
2870
2871 int mdd_lfsck_setup(const struct lu_env *env, struct mdd_device *mdd)
2872 {
2873         struct md_lfsck         *lfsck = &mdd->mdd_lfsck;
2874         struct dt_object        *obj;
2875         struct lu_fid            fid;
2876         int                      rc;
2877
2878         ENTRY;
2879
2880         LASSERT(!lfsck->ml_initialized);
2881
2882         lfsck->ml_initialized = 1;
2883         mutex_init(&lfsck->ml_mutex);
2884         spin_lock_init(&lfsck->ml_lock);
2885         CFS_INIT_LIST_HEAD(&lfsck->ml_list_scan);
2886         CFS_INIT_LIST_HEAD(&lfsck->ml_list_dir);
2887         CFS_INIT_LIST_HEAD(&lfsck->ml_list_double_scan);
2888         CFS_INIT_LIST_HEAD(&lfsck->ml_list_idle);
2889         cfs_waitq_init(&lfsck->ml_thread.t_ctl_waitq);
2890
2891         obj = dt_locate(env, mdd->mdd_bottom, &lfsck_it_fid);
2892         if (IS_ERR(obj))
2893                 RETURN(PTR_ERR(obj));
2894
2895         lfsck->ml_obj_oit = obj;
2896         rc = obj->do_ops->do_index_try(env, obj, &dt_otable_features);
2897         if (rc != 0) {
2898                 if (rc == -ENOTSUPP)
2899                         RETURN(0);
2900                 GOTO(out, rc);
2901         }
2902
2903         /* LFSCK bookmark */
2904         fid_zero(&fid);
2905         rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
2906                                    lfsck_bookmark_name,
2907                                    S_IFREG | S_IRUGO | S_IWUSR, &fid);
2908         if (rc < 0)
2909                 GOTO(out, rc);
2910
2911         obj = dt_locate(env, mdd->mdd_bottom, &fid);
2912         if (IS_ERR(obj))
2913                 GOTO(out, rc = PTR_ERR(obj));
2914
2915         LASSERT(lu_object_exists(&obj->do_lu));
2916         lfsck->ml_bookmark_obj = obj;
2917
2918         rc = mdd_lfsck_bookmark_load(env, lfsck);
2919         if (rc == -ENODATA)
2920                 rc = mdd_lfsck_bookmark_init(env, lfsck);
2921         if (rc != 0)
2922                 GOTO(out, rc);
2923
2924         rc = mdd_lfsck_namespace_setup(env, lfsck);
2925         if (rc < 0)
2926                 GOTO(out, rc);
2927         /* XXX: LFSCK components initialization to be added here. */
2928         RETURN(0);
2929 out:
2930         lu_object_put(env, &lfsck->ml_obj_oit->do_lu);
2931         lfsck->ml_obj_oit = NULL;
2932         return 0;
2933 }
2934
2935 void mdd_lfsck_cleanup(const struct lu_env *env, struct mdd_device *mdd)
2936 {
2937         struct md_lfsck         *lfsck  = &mdd->mdd_lfsck;
2938         struct ptlrpc_thread    *thread = &lfsck->ml_thread;
2939         struct lfsck_component  *com;
2940
2941         if (!lfsck->ml_initialized)
2942                 return;
2943
2944         LASSERT(thread_is_init(thread) || thread_is_stopped(thread));
2945
2946         if (lfsck->ml_obj_oit != NULL) {
2947                 lu_object_put(env, &lfsck->ml_obj_oit->do_lu);
2948                 lfsck->ml_obj_oit = NULL;
2949         }
2950
2951         LASSERT(lfsck->ml_obj_dir == NULL);
2952
2953         if (lfsck->ml_bookmark_obj != NULL) {
2954                 lu_object_put(env, &lfsck->ml_bookmark_obj->do_lu);
2955                 lfsck->ml_bookmark_obj = NULL;
2956         }
2957
2958         while (!cfs_list_empty(&lfsck->ml_list_scan)) {
2959                 com = cfs_list_entry(lfsck->ml_list_scan.next,
2960                                      struct lfsck_component,
2961                                      lc_link);
2962                 mdd_lfsck_component_cleanup(env, com);
2963         }
2964
2965         LASSERT(cfs_list_empty(&lfsck->ml_list_dir));
2966
2967         while (!cfs_list_empty(&lfsck->ml_list_double_scan)) {
2968                 com = cfs_list_entry(lfsck->ml_list_double_scan.next,
2969                                      struct lfsck_component,
2970                                      lc_link);
2971                 mdd_lfsck_component_cleanup(env, com);
2972         }
2973
2974         while (!cfs_list_empty(&lfsck->ml_list_idle)) {
2975                 com = cfs_list_entry(lfsck->ml_list_idle.next,
2976                                      struct lfsck_component,
2977                                      lc_link);
2978                 mdd_lfsck_component_cleanup(env, com);
2979         }
2980 }