lustre/lfsck/lfsck_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2013, 2016, Intel Corporation.
  24  */
  25 /*
  26  * lustre/lfsck/lfsck_lib.c
  27  *
  28  * Author: Fan, Yong <fan.yong@intel.com>
  29  */
  30
  31 #define DEBUG_SUBSYSTEM S_LFSCK
  32
  33 #include <linux/kthread.h>
  34 #include <linux/sched.h>
  35 #include <linux/list.h>
  36 #include <lu_object.h>
  37 #include <dt_object.h>
  38 #include <md_object.h>
  39 #include <lustre_fld.h>
  40 #include <lustre_lib.h>
  41 #include <lustre_net.h>
  42 #include <lustre_lfsck.h>
  43
  44 #include "lfsck_internal.h"
  45
  46 #define LFSCK_CHECKPOINT_SKIP   1
  47
  48 /* define lfsck thread key */
  49 LU_KEY_INIT(lfsck, struct lfsck_thread_info);
  50
  51 static void lfsck_key_fini(const struct lu_context *ctx,
  52                            struct lu_context_key *key, void *data)
  53 {
  54         struct lfsck_thread_info *info = data;
  55
  56         lu_buf_free(&info->lti_linkea_buf);
  57         lu_buf_free(&info->lti_linkea_buf2);
  58         lu_buf_free(&info->lti_big_buf);
  59         OBD_FREE_PTR(info);
  60 }
  61
  62 LU_CONTEXT_KEY_DEFINE(lfsck, LCT_MD_THREAD | LCT_DT_THREAD);
  63 LU_KEY_INIT_GENERIC(lfsck);
  64
  65 static struct list_head lfsck_instance_list;
  66 static struct list_head lfsck_ost_orphan_list;
  67 static struct list_head lfsck_mdt_orphan_list;
  68 static DEFINE_SPINLOCK(lfsck_instance_lock);
  69
  70 const char *lfsck_flags_names[] = {
  71         "scanned-once",
  72         "inconsistent",
  73         "upgrade",
  74         "incomplete",
  75         "crashed_lastid",
  76         NULL
  77 };
  78
  79 const char *lfsck_param_names[] = {
  80         NULL,
  81         "failout",
  82         "dryrun",
  83         "all_targets",
  84         "broadcast",
  85         "orphan",
  86         "create_ostobj",
  87         "create_mdtobj",
  88         NULL,
  89         "delay_create_ostobj",
  90         NULL
  91 };
  92
  93 enum lfsck_verify_lpf_types {
  94         LVLT_BY_BOOKMARK        = 0,
  95         LVLT_BY_NAMEENTRY       = 1,
  96 };
  97
  98 static inline void
  99 lfsck_reset_ltd_status(struct lfsck_tgt_desc *ltd, enum lfsck_type type)
 100 {
 101         if (type == LFSCK_TYPE_LAYOUT) {
 102                 ltd->ltd_layout_status = LS_MAX;
 103                 ltd->ltd_layout_repaired = 0;
 104         } else {
 105                 ltd->ltd_namespace_status = LS_MAX;
 106                 ltd->ltd_namespace_repaired = 0;
 107         }
 108 }
 109
 110 static int lfsck_tgt_descs_init(struct lfsck_tgt_descs *ltds)
 111 {
 112         spin_lock_init(&ltds->ltd_lock);
 113         init_rwsem(&ltds->ltd_rw_sem);
 114         INIT_LIST_HEAD(&ltds->ltd_orphan);
 115         ltds->ltd_tgts_bitmap = CFS_ALLOCATE_BITMAP(BITS_PER_LONG);
 116         if (ltds->ltd_tgts_bitmap == NULL)
 117                 return -ENOMEM;
 118
 119         return 0;
 120 }
 121
 122 static void lfsck_tgt_descs_fini(struct lfsck_tgt_descs *ltds)
 123 {
 124         struct lfsck_tgt_desc   *ltd;
 125         struct lfsck_tgt_desc   *next;
 126         int                      idx;
 127
 128         down_write(&ltds->ltd_rw_sem);
 129
 130         list_for_each_entry_safe(ltd, next, &ltds->ltd_orphan,
 131                                  ltd_orphan_list) {
 132                 list_del_init(&ltd->ltd_orphan_list);
 133                 lfsck_tgt_put(ltd);
 134         }
 135
 136         if (unlikely(ltds->ltd_tgts_bitmap == NULL)) {
 137                 up_write(&ltds->ltd_rw_sem);
 138
 139                 return;
 140         }
 141
 142         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
 143                 ltd = lfsck_ltd2tgt(ltds, idx);
 144                 if (likely(ltd != NULL)) {
 145                         LASSERT(list_empty(&ltd->ltd_layout_list));
 146                         LASSERT(list_empty(&ltd->ltd_layout_phase_list));
 147                         LASSERT(list_empty(&ltd->ltd_namespace_list));
 148                         LASSERT(list_empty(&ltd->ltd_namespace_phase_list));
 149
 150                         ltds->ltd_tgtnr--;
 151                         cfs_bitmap_clear(ltds->ltd_tgts_bitmap, idx);
 152                         lfsck_assign_tgt(ltds, NULL, idx);
 153                         lfsck_tgt_put(ltd);
 154                 }
 155         }
 156
 157         LASSERTF(ltds->ltd_tgtnr == 0, "tgt count unmatched: %d\n",
 158                  ltds->ltd_tgtnr);
 159
 160         for (idx = 0; idx < TGT_PTRS; idx++) {
 161                 if (ltds->ltd_tgts_idx[idx] != NULL) {
 162                         OBD_FREE_PTR(ltds->ltd_tgts_idx[idx]);
 163                         ltds->ltd_tgts_idx[idx] = NULL;
 164                 }
 165         }
 166
 167         CFS_FREE_BITMAP(ltds->ltd_tgts_bitmap);
 168         ltds->ltd_tgts_bitmap = NULL;
 169         up_write(&ltds->ltd_rw_sem);
 170 }
 171
 172 static int __lfsck_add_target(const struct lu_env *env,
 173                               struct lfsck_instance *lfsck,
 174                               struct lfsck_tgt_desc *ltd,
 175                               bool for_ost, bool locked)
 176 {
 177         struct lfsck_tgt_descs *ltds;
 178         __u32                   index = ltd->ltd_index;
 179         int                     rc    = 0;
 180         ENTRY;
 181
 182         if (for_ost)
 183                 ltds = &lfsck->li_ost_descs;
 184         else
 185                 ltds = &lfsck->li_mdt_descs;
 186
 187         if (!locked)
 188                 down_write(&ltds->ltd_rw_sem);
 189
 190         LASSERT(ltds->ltd_tgts_bitmap != NULL);
 191
 192         if (index >= ltds->ltd_tgts_bitmap->size) {
 193                 __u32 newsize = max((__u32)ltds->ltd_tgts_bitmap->size,
 194                                     (__u32)BITS_PER_LONG);
 195                 struct cfs_bitmap *old_bitmap = ltds->ltd_tgts_bitmap;
 196                 struct cfs_bitmap *new_bitmap;
 197
 198                 while (newsize < index + 1)
 199                         newsize <<= 1;
 200
 201                 new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
 202                 if (new_bitmap == NULL)
 203                         GOTO(unlock, rc = -ENOMEM);
 204
 205                 if (ltds->ltd_tgtnr > 0)
 206                         cfs_bitmap_copy(new_bitmap, old_bitmap);
 207                 ltds->ltd_tgts_bitmap = new_bitmap;
 208                 CFS_FREE_BITMAP(old_bitmap);
 209         }
 210
 211         if (cfs_bitmap_check(ltds->ltd_tgts_bitmap, index)) {
 212                 CERROR("%s: the device %s (%u) is registered already\n",
 213                        lfsck_lfsck2name(lfsck),
 214                        ltd->ltd_tgt->dd_lu_dev.ld_obd->obd_name, index);
 215                 GOTO(unlock, rc = -EEXIST);
 216         }
 217
 218         if (ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 219                 OBD_ALLOC_PTR(ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK]);
 220                 if (ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK] == NULL)
 221                         GOTO(unlock, rc = -ENOMEM);
 222         }
 223
 224         lfsck_assign_tgt(ltds, ltd, index);
 225         cfs_bitmap_set(ltds->ltd_tgts_bitmap, index);
 226         ltds->ltd_tgtnr++;
 227
 228         GOTO(unlock, rc = 0);
 229
 230 unlock:
 231         if (!locked)
 232                 up_write(&ltds->ltd_rw_sem);
 233
 234         return rc;
 235 }
 236
 237 static int lfsck_add_target_from_orphan(const struct lu_env *env,
 238                                         struct lfsck_instance *lfsck)
 239 {
 240         struct lfsck_tgt_descs  *ltds    = &lfsck->li_ost_descs;
 241         struct lfsck_tgt_desc   *ltd;
 242         struct lfsck_tgt_desc   *next;
 243         struct list_head        *head    = &lfsck_ost_orphan_list;
 244         int                      rc;
 245         bool                     for_ost = true;
 246
 247 again:
 248         spin_lock(&lfsck_instance_lock);
 249         list_for_each_entry_safe(ltd, next, head, ltd_orphan_list) {
 250                 if (ltd->ltd_key == lfsck->li_bottom)
 251                         list_move_tail(&ltd->ltd_orphan_list,
 252                                        &ltds->ltd_orphan);
 253         }
 254         spin_unlock(&lfsck_instance_lock);
 255
 256         down_write(&ltds->ltd_rw_sem);
 257         while (!list_empty(&ltds->ltd_orphan)) {
 258                 ltd = list_entry(ltds->ltd_orphan.next,
 259                                  struct lfsck_tgt_desc,
 260                                  ltd_orphan_list);
 261                 list_del_init(&ltd->ltd_orphan_list);
 262                 rc = __lfsck_add_target(env, lfsck, ltd, for_ost, true);
 263                 /* Do not hold the semaphore for too long time. */
 264                 up_write(&ltds->ltd_rw_sem);
 265                 if (rc != 0)
 266                         return rc;
 267
 268                 down_write(&ltds->ltd_rw_sem);
 269         }
 270         up_write(&ltds->ltd_rw_sem);
 271
 272         if (for_ost) {
 273                 ltds = &lfsck->li_mdt_descs;
 274                 head = &lfsck_mdt_orphan_list;
 275                 for_ost = false;
 276                 goto again;
 277         }
 278
 279         return 0;
 280 }
 281
 282 static inline struct lfsck_component *
 283 __lfsck_component_find(struct lfsck_instance *lfsck, __u16 type,
 284                        struct list_head *list)
 285 {
 286         struct lfsck_component *com;
 287
 288         list_for_each_entry(com, list, lc_link) {
 289                 if (com->lc_type == type)
 290                         return com;
 291         }
 292         return NULL;
 293 }
 294
 295 struct lfsck_component *
 296 lfsck_component_find(struct lfsck_instance *lfsck, __u16 type)
 297 {
 298         struct lfsck_component *com;
 299
 300         spin_lock(&lfsck->li_lock);
 301         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_scan);
 302         if (com != NULL)
 303                 goto unlock;
 304
 305         com = __lfsck_component_find(lfsck, type,
 306                                      &lfsck->li_list_double_scan);
 307         if (com != NULL)
 308                 goto unlock;
 309
 310         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_idle);
 311
 312 unlock:
 313         if (com != NULL)
 314                 lfsck_component_get(com);
 315         spin_unlock(&lfsck->li_lock);
 316         return com;
 317 }
 318
 319 void lfsck_component_cleanup(const struct lu_env *env,
 320                              struct lfsck_component *com)
 321 {
 322         if (!list_empty(&com->lc_link))
 323                 list_del_init(&com->lc_link);
 324         if (!list_empty(&com->lc_link_dir))
 325                 list_del_init(&com->lc_link_dir);
 326
 327         lfsck_component_put(env, com);
 328 }
 329
 330 int lfsck_fid_alloc(const struct lu_env *env, struct lfsck_instance *lfsck,
 331                     struct lu_fid *fid, bool locked)
 332 {
 333         struct lfsck_bookmark   *bk = &lfsck->li_bookmark_ram;
 334         int                      rc = 0;
 335         ENTRY;
 336
 337         if (!locked)
 338                 mutex_lock(&lfsck->li_mutex);
 339
 340         rc = seq_client_alloc_fid(env, lfsck->li_seq, fid);
 341         if (rc >= 0) {
 342                 bk->lb_last_fid = *fid;
 343                 /* We do not care about whether the subsequent sub-operations
 344                  * failed or not. The worst case is that one FID is lost that
 345                  * is not a big issue for the LFSCK since it is relative rare
 346                  * for LFSCK create. */
 347                 rc = lfsck_bookmark_store(env, lfsck);
 348         }
 349
 350         if (!locked)
 351                 mutex_unlock(&lfsck->li_mutex);
 352
 353         RETURN(rc);
 354 }
 355
 356 static int __lfsck_ibits_lock(const struct lu_env *env,
 357                               struct lfsck_instance *lfsck,
 358                               struct dt_object *obj, struct ldlm_res_id *resid,
 359                               struct lustre_handle *lh, __u64 bits,
 360                               enum ldlm_mode mode)
 361 {
 362         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 363         union ldlm_policy_data          *policy = &info->lti_policy;
 364         __u64                            flags  = LDLM_FL_ATOMIC_CB;
 365         int                              rc;
 366
 367         LASSERT(lfsck->li_namespace != NULL);
 368
 369         memset(policy, 0, sizeof(*policy));
 370         policy->l_inodebits.bits = bits;
 371         if (dt_object_remote(obj)) {
 372                 struct ldlm_enqueue_info *einfo = &info->lti_einfo;
 373
 374                 memset(einfo, 0, sizeof(*einfo));
 375                 einfo->ei_type = LDLM_IBITS;
 376                 einfo->ei_mode = mode;
 377                 einfo->ei_cb_bl = ldlm_blocking_ast;
 378                 einfo->ei_cb_cp = ldlm_completion_ast;
 379                 einfo->ei_res_id = resid;
 380
 381                 rc = dt_object_lock(env, obj, lh, einfo, policy);
 382         } else {
 383                 rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid,
 384                                             LDLM_IBITS, policy, mode,
 385                                             &flags, ldlm_blocking_ast,
 386                                             ldlm_completion_ast, NULL, NULL,
 387                                             0, LVB_T_NONE, NULL, lh);
 388         }
 389
 390         if (rc == ELDLM_OK) {
 391                 rc = 0;
 392         } else {
 393                 memset(lh, 0, sizeof(*lh));
 394                 rc = -EIO;
 395         }
 396
 397         return rc;
 398 }
 399
 400 /**
 401  * Request the specified ibits lock for the given object.
 402  *
 403  * Before the LFSCK modifying on the namespace visible object,
 404  * it needs to acquire related ibits ldlm lock.
 405  *
 406  * \param[in] env       pointer to the thread context
 407  * \param[in] lfsck     pointer to the lfsck instance
 408  * \param[in] obj       pointer to the dt_object to be locked
 409  * \param[out] lh       pointer to the lock handle
 410  * \param[in] bits      the bits for the ldlm lock to be acquired
 411  * \param[in] mode      the mode for the ldlm lock to be acquired
 412  *
 413  * \retval              0 for success
 414  * \retval              negative error number on failure
 415  */
 416 int lfsck_ibits_lock(const struct lu_env *env, struct lfsck_instance *lfsck,
 417                      struct dt_object *obj, struct lustre_handle *lh,
 418                      __u64 bits, enum ldlm_mode mode)
 419 {
 420         struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid;
 421
 422         LASSERT(!lustre_handle_is_used(lh));
 423
 424         fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
 425         return __lfsck_ibits_lock(env, lfsck, obj, resid, lh, bits, mode);
 426 }
 427
 428 /**
 429  * Release the the specified ibits lock.
 430  *
 431  * If the lock has been acquired before, release it
 432  * and cleanup the handle. Otherwise, do nothing.
 433  *
 434  * \param[in] lh        pointer to the lock handle
 435  * \param[in] mode      the mode for the ldlm lock to be released
 436  */
 437 void lfsck_ibits_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
 438 {
 439         if (lustre_handle_is_used(lh)) {
 440                 ldlm_lock_decref(lh, mode);
 441                 memset(lh, 0, sizeof(*lh));
 442         }
 443 }
 444
 445 /**
 446  * Request compound ibits locks for the given <obj, name> pairs.
 447  *
 448  * Before the LFSCK modifying on the namespace visible object, it needs to
 449  * acquire related ibits ldlm lock. Usually, we can use lfsck_ibits_lock for
 450  * the lock purpose. But the simple lfsck_ibits_lock for directory-based
 451  * modificationis (such as insert name entry to the directory) may be too
 452  * coarse-grained and not efficient.
 453  *
 454  * The lfsck_lock() will request compound ibits locks on the specified
 455  * <obj, name> pairs: the PDO (Parallel Directory Operations) ibits (UPDATE)
 456  * lock on the directory object, and the regular ibits lock on the name hash.
 457  *
 458  * \param[in] env       pointer to the thread context
 459  * \param[in] lfsck     pointer to the lfsck instance
 460  * \param[in] obj       pointer to the dt_object to be locked
 461  * \param[in] name      used for building the PDO lock resource
 462  * \param[out] llh      pointer to the lfsck_lock_handle
 463  * \param[in] bits      the bits for the ldlm lock to be acquired
 464  * \param[in] mode      the mode for the ldlm lock to be acquired
 465  *
 466  * \retval              0 for success
 467  * \retval              negative error number on failure
 468  */
 469 int lfsck_lock(const struct lu_env *env, struct lfsck_instance *lfsck,
 470                struct dt_object *obj, const char *name,
 471                struct lfsck_lock_handle *llh, __u64 bits, enum ldlm_mode mode)
 472 {
 473         struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid;
 474         int                 rc;
 475
 476         LASSERT(S_ISDIR(lfsck_object_type(obj)));
 477         LASSERT(name != NULL);
 478         LASSERT(name[0] != 0);
 479         LASSERT(!lustre_handle_is_used(&llh->llh_pdo_lh));
 480         LASSERT(!lustre_handle_is_used(&llh->llh_reg_lh));
 481
 482         switch (mode) {
 483         case LCK_EX:
 484                 llh->llh_pdo_mode = LCK_EX;
 485                 break;
 486         case LCK_PW:
 487                 llh->llh_pdo_mode = LCK_CW;
 488                 break;
 489         case LCK_PR:
 490                 llh->llh_pdo_mode = LCK_CR;
 491                 break;
 492         default:
 493                 CDEBUG(D_LFSCK, "%s: unexpected PDO lock mode %u on the obj "
 494                        DFID"\n", lfsck_lfsck2name(lfsck), mode,
 495                        PFID(lfsck_dto2fid(obj)));
 496                 LBUG();
 497         }
 498
 499         fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
 500         rc = __lfsck_ibits_lock(env, lfsck, obj, resid, &llh->llh_pdo_lh,
 501                                 MDS_INODELOCK_UPDATE, llh->llh_pdo_mode);
 502         if (rc != 0)
 503                 return rc;
 504
 505         llh->llh_reg_mode = mode;
 506         resid->name[LUSTRE_RES_ID_HSH_OFF] = ll_full_name_hash(NULL, name,
 507                                                                strlen(name));
 508         LASSERT(resid->name[LUSTRE_RES_ID_HSH_OFF] != 0);
 509         rc = __lfsck_ibits_lock(env, lfsck, obj, resid, &llh->llh_reg_lh,
 510                                 bits, llh->llh_reg_mode);
 511         if (rc != 0)
 512                 lfsck_ibits_unlock(&llh->llh_pdo_lh, llh->llh_pdo_mode);
 513
 514         return rc;
 515 }
 516
 517 /**
 518  * Release the the compound ibits locks.
 519  *
 520  * \param[in] llh       pointer to the lfsck_lock_handle to be released
 521  */
 522 void lfsck_unlock(struct lfsck_lock_handle *llh)
 523 {
 524         lfsck_ibits_unlock(&llh->llh_reg_lh, llh->llh_reg_mode);
 525         lfsck_ibits_unlock(&llh->llh_pdo_lh, llh->llh_pdo_mode);
 526 }
 527
 528 int lfsck_find_mdt_idx_by_fid(const struct lu_env *env,
 529                               struct lfsck_instance *lfsck,
 530                               const struct lu_fid *fid)
 531 {
 532         struct seq_server_site  *ss     = lfsck_dev_site(lfsck);
 533         struct lu_seq_range     *range  = &lfsck_env_info(env)->lti_range;
 534         int                      rc;
 535
 536         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
 537                 /* "ROOT" is always on the MDT0. */
 538                 if (lu_fid_eq(fid, &lfsck->li_global_root_fid))
 539                         return 0;
 540
 541                 return lfsck_dev_idx(lfsck);
 542         }
 543
 544         fld_range_set_mdt(range);
 545         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
 546         if (rc == 0)
 547                 rc = range->lsr_index;
 548
 549         return rc;
 550 }
 551
 552 const char dot[] = ".";
 553 const char dotdot[] = "..";
 554 static const char dotlustre[] = ".lustre";
 555 static const char lostfound[] = "lost+found";
 556
 557 /**
 558  * Remove the name entry from the .lustre/lost+found directory.
 559  *
 560  * No need to care about the object referenced by the name entry,
 561  * either the name entry is invalid or redundant, or the referenced
 562  * object has been processed or will be handled by others.
 563  *
 564  * \param[in] env       pointer to the thread context
 565  * \param[in] lfsck     pointer to the lfsck instance
 566  * \param[in] name      the name for the name entry to be removed
 567  *
 568  * \retval              0 for success
 569  * \retval              negative error number on failure
 570  */
 571 static int lfsck_lpf_remove_name_entry(const struct lu_env *env,
 572                                        struct lfsck_instance *lfsck,
 573                                        const char *name)
 574 {
 575         struct dt_object        *parent = lfsck->li_lpf_root_obj;
 576         struct dt_device        *dev    = lfsck_obj2dev(parent);
 577         struct thandle          *th;
 578         struct lfsck_lock_handle *llh   = &lfsck_env_info(env)->lti_llh;
 579         int                      rc;
 580         ENTRY;
 581
 582         rc = lfsck_lock(env, lfsck, parent, name, llh,
 583                         MDS_INODELOCK_UPDATE, LCK_PW);
 584         if (rc != 0)
 585                 RETURN(rc);
 586
 587         th = dt_trans_create(env, dev);
 588         if (IS_ERR(th))
 589                 GOTO(unlock, rc = PTR_ERR(th));
 590
 591         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
 592         if (rc != 0)
 593                 GOTO(stop, rc);
 594
 595         rc = dt_declare_ref_del(env, parent, th);
 596         if (rc != 0)
 597                 GOTO(stop, rc);
 598
 599         rc = dt_trans_start_local(env, dev, th);
 600         if (rc != 0)
 601                 GOTO(stop, rc);
 602
 603         rc = dt_delete(env, parent, (const struct dt_key *)name, th);
 604         if (rc != 0)
 605                 GOTO(stop, rc);
 606
 607         dt_write_lock(env, parent, 0);
 608         rc = dt_ref_del(env, parent, th);
 609         dt_write_unlock(env, parent);
 610
 611         GOTO(stop, rc);
 612
 613 stop:
 614         dt_trans_stop(env, dev, th);
 615
 616 unlock:
 617         lfsck_unlock(llh);
 618
 619         CDEBUG(D_LFSCK, "%s: remove name entry "DFID"/%s: rc = %d\n",
 620                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(parent)), name, rc);
 621
 622         return rc;
 623 }
 624
 625 static int lfsck_create_lpf_local(const struct lu_env *env,
 626                                   struct lfsck_instance *lfsck,
 627                                   struct dt_object *child,
 628                                   struct lu_attr *la,
 629                                   struct dt_object_format *dof,
 630                                   const char *name)
 631 {
 632         struct dt_insert_rec    *rec    = &lfsck_env_info(env)->lti_dt_rec;
 633         struct dt_object        *parent = lfsck->li_lpf_root_obj;
 634         struct dt_device        *dev    = lfsck_obj2dev(child);
 635         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
 636         struct dt_object        *bk_obj = lfsck->li_bookmark_obj;
 637         const struct lu_fid     *cfid   = lfsck_dto2fid(child);
 638         struct thandle          *th     = NULL;
 639         struct linkea_data       ldata  = { NULL };
 640         struct lu_buf            linkea_buf;
 641         const struct lu_name    *cname;
 642         loff_t                   pos    = 0;
 643         int                      len    = sizeof(struct lfsck_bookmark);
 644         int                      rc;
 645         ENTRY;
 646
 647         cname = lfsck_name_get_const(env, name, strlen(name));
 648         rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf2,
 649                               cname, lfsck_dto2fid(parent));
 650         if (rc != 0)
 651                 RETURN(rc);
 652
 653         th = dt_trans_create(env, dev);
 654         if (IS_ERR(th))
 655                 RETURN(PTR_ERR(th));
 656
 657         /* 1a. create child */
 658         rc = dt_declare_create(env, child, la, NULL, dof, th);
 659         if (rc != 0)
 660                 GOTO(stop, rc);
 661
 662         if (!dt_try_as_dir(env, child))
 663                 GOTO(stop, rc = -ENOTDIR);
 664
 665         /* 2a. increase child nlink */
 666         rc = dt_declare_ref_add(env, child, th);
 667         if (rc != 0)
 668                 GOTO(stop, rc);
 669
 670         /* 3a. insert dot into child dir */
 671         rec->rec_type = S_IFDIR;
 672         rec->rec_fid = cfid;
 673         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
 674                                (const struct dt_key *)dot, th);
 675         if (rc != 0)
 676                 GOTO(stop, rc);
 677
 678         /* 4a. insert dotdot into child dir */
 679         rec->rec_fid = &LU_LPF_FID;
 680         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
 681                                (const struct dt_key *)dotdot, th);
 682         if (rc != 0)
 683                 GOTO(stop, rc);
 684
 685         /* 5a. insert linkEA for child */
 686         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
 687                        ldata.ld_leh->leh_len);
 688         rc = dt_declare_xattr_set(env, child, &linkea_buf,
 689                                   XATTR_NAME_LINK, 0, th);
 690         if (rc != 0)
 691                 GOTO(stop, rc);
 692
 693         /* 6a. insert name into parent dir */
 694         rec->rec_type = S_IFDIR;
 695         rec->rec_fid = cfid;
 696         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
 697                                (const struct dt_key *)name, th);
 698         if (rc != 0)
 699                 GOTO(stop, rc);
 700
 701         /* 7a. increase parent nlink */
 702         rc = dt_declare_ref_add(env, parent, th);
 703         if (rc != 0)
 704                 GOTO(stop, rc);
 705
 706         /* 8a. update bookmark */
 707         rc = dt_declare_record_write(env, bk_obj,
 708                                      lfsck_buf_get(env, bk, len), 0, th);
 709         if (rc != 0)
 710                 GOTO(stop, rc);
 711
 712         rc = dt_trans_start_local(env, dev, th);
 713         if (rc != 0)
 714                 GOTO(stop, rc);
 715
 716         dt_write_lock(env, child, 0);
 717         /* 1b. create child */
 718         rc = dt_create(env, child, la, NULL, dof, th);
 719         if (rc != 0)
 720                 GOTO(unlock, rc);
 721
 722         /* 2b. increase child nlink */
 723         rc = dt_ref_add(env, child, th);
 724         if (rc != 0)
 725                 GOTO(unlock, rc);
 726
 727         /* 3b. insert dot into child dir */
 728         rec->rec_fid = cfid;
 729         rc = dt_insert(env, child, (const struct dt_rec *)rec,
 730                        (const struct dt_key *)dot, th, 1);
 731         if (rc != 0)
 732                 GOTO(unlock, rc);
 733
 734         /* 4b. insert dotdot into child dir */
 735         rec->rec_fid = &LU_LPF_FID;
 736         rc = dt_insert(env, child, (const struct dt_rec *)rec,
 737                        (const struct dt_key *)dotdot, th, 1);
 738         if (rc != 0)
 739                 GOTO(unlock, rc);
 740
 741         /* 5b. insert linkEA for child. */
 742         rc = dt_xattr_set(env, child, &linkea_buf,
 743                           XATTR_NAME_LINK, 0, th);
 744         dt_write_unlock(env, child);
 745         if (rc != 0)
 746                 GOTO(stop, rc);
 747
 748         /* 6b. insert name into parent dir */
 749         rec->rec_fid = cfid;
 750         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
 751                        (const struct dt_key *)name, th, 1);
 752         if (rc != 0)
 753                 GOTO(stop, rc);
 754
 755         dt_write_lock(env, parent, 0);
 756         /* 7b. increase parent nlink */
 757         rc = dt_ref_add(env, parent, th);
 758         dt_write_unlock(env, parent);
 759         if (rc != 0)
 760                 GOTO(stop, rc);
 761
 762         bk->lb_lpf_fid = *cfid;
 763         lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk);
 764
 765         /* 8b. update bookmark */
 766         rc = dt_record_write(env, bk_obj,
 767                              lfsck_buf_get(env, bk, len), &pos, th);
 768
 769         GOTO(stop, rc);
 770
 771 unlock:
 772         dt_write_unlock(env, child);
 773
 774 stop:
 775         dt_trans_stop(env, dev, th);
 776
 777         return rc;
 778 }
 779
 780 static int lfsck_create_lpf_remote(const struct lu_env *env,
 781                                    struct lfsck_instance *lfsck,
 782                                    struct dt_object *child,
 783                                    struct lu_attr *la,
 784                                    struct dt_object_format *dof,
 785                                    const char *name)
 786 {
 787         struct dt_insert_rec    *rec    = &lfsck_env_info(env)->lti_dt_rec;
 788         struct dt_object        *parent = lfsck->li_lpf_root_obj;
 789         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
 790         struct dt_object        *bk_obj = lfsck->li_bookmark_obj;
 791         const struct lu_fid     *cfid   = lfsck_dto2fid(child);
 792         struct thandle          *th     = NULL;
 793         struct linkea_data       ldata  = { NULL };
 794         struct lu_buf            linkea_buf;
 795         const struct lu_name    *cname;
 796         struct dt_device        *dev;
 797         loff_t                   pos    = 0;
 798         int                      len    = sizeof(struct lfsck_bookmark);
 799         int                      rc;
 800         ENTRY;
 801
 802         cname = lfsck_name_get_const(env, name, strlen(name));
 803         rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf2,
 804                               cname, lfsck_dto2fid(parent));
 805         if (rc != 0)
 806                 RETURN(rc);
 807
 808         /* Create .lustre/lost+found/MDTxxxx. */
 809
 810         /* XXX: Currently, cross-MDT create operation needs to create the child
 811          *      object firstly, then insert name into the parent directory. For
 812          *      this case, the child object resides on current MDT (local), but
 813          *      the parent ".lustre/lost+found" may be on remote MDT. It is not
 814          *      easy to contain all the sub-modifications orderly within single
 815          *      transaction.
 816          *
 817          *      To avoid more inconsistency, we split the create operation into
 818          *      two transactions:
 819          *
 820          *      1) create the child and update the lfsck_bookmark::lb_lpf_fid
 821          *         locally.
 822          *      2) insert the name "MDTXXXX" in the parent ".lustre/lost+found"
 823          *         remotely.
 824          *
 825          *      If 1) done, but 2) failed, then go ahead, the LFSCK will try to
 826          *      repair such inconsistency when LFSCK run next time. */
 827
 828         /* Transaction I: locally */
 829
 830         dev = lfsck_obj2dev(child);
 831         th = dt_trans_create(env, dev);
 832         if (IS_ERR(th))
 833                 RETURN(PTR_ERR(th));
 834
 835         /* 1a. create child */
 836         rc = dt_declare_create(env, child, la, NULL, dof, th);
 837         if (rc != 0)
 838                 GOTO(stop, rc);
 839
 840         if (!dt_try_as_dir(env, child))
 841                 GOTO(stop, rc = -ENOTDIR);
 842
 843         /* 2a. increase child nlink */
 844         rc = dt_declare_ref_add(env, child, th);
 845         if (rc != 0)
 846                 GOTO(stop, rc);
 847
 848         /* 3a. insert dot into child dir */
 849         rec->rec_type = S_IFDIR;
 850         rec->rec_fid = cfid;
 851         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
 852                                (const struct dt_key *)dot, th);
 853         if (rc != 0)
 854                 GOTO(stop, rc);
 855
 856         /* 4a. insert dotdot into child dir */
 857         rec->rec_fid = &LU_LPF_FID;
 858         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
 859                                (const struct dt_key *)dotdot, th);
 860         if (rc != 0)
 861                 GOTO(stop, rc);
 862
 863         /* 5a. insert linkEA for child */
 864         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
 865                        ldata.ld_leh->leh_len);
 866         rc = dt_declare_xattr_set(env, child, &linkea_buf,
 867                                   XATTR_NAME_LINK, 0, th);
 868         if (rc != 0)
 869                 GOTO(stop, rc);
 870
 871         /* 6a. update bookmark */
 872         rc = dt_declare_record_write(env, bk_obj,
 873                                      lfsck_buf_get(env, bk, len), 0, th);
 874         if (rc != 0)
 875                 GOTO(stop, rc);
 876
 877         rc = dt_trans_start_local(env, dev, th);
 878         if (rc != 0)
 879                 GOTO(stop, rc);
 880
 881         dt_write_lock(env, child, 0);
 882         /* 1b. create child */
 883         rc = dt_create(env, child, la, NULL, dof, th);
 884         if (rc != 0)
 885                 GOTO(unlock, rc);
 886
 887         /* 2b. increase child nlink */
 888         rc = dt_ref_add(env, child, th);
 889         if (rc != 0)
 890                 GOTO(unlock, rc);
 891
 892         /* 3b. insert dot into child dir */
 893         rec->rec_type = S_IFDIR;
 894         rec->rec_fid = cfid;
 895         rc = dt_insert(env, child, (const struct dt_rec *)rec,
 896                        (const struct dt_key *)dot, th, 1);
 897         if (rc != 0)
 898                 GOTO(unlock, rc);
 899
 900         /* 4b. insert dotdot into child dir */
 901         rec->rec_fid = &LU_LPF_FID;
 902         rc = dt_insert(env, child, (const struct dt_rec *)rec,
 903                        (const struct dt_key *)dotdot, th, 1);
 904         if (rc != 0)
 905                 GOTO(unlock, rc);
 906
 907         /* 5b. insert linkEA for child */
 908         rc = dt_xattr_set(env, child, &linkea_buf,
 909                           XATTR_NAME_LINK, 0, th);
 910         if (rc != 0)
 911                 GOTO(unlock, rc);
 912
 913         bk->lb_lpf_fid = *cfid;
 914         lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk);
 915
 916         /* 6b. update bookmark */
 917         rc = dt_record_write(env, bk_obj,
 918                              lfsck_buf_get(env, bk, len), &pos, th);
 919
 920         dt_write_unlock(env, child);
 921         dt_trans_stop(env, dev, th);
 922         if (rc != 0)
 923                 RETURN(rc);
 924
 925         /* Transaction II: remotely */
 926
 927         dev = lfsck_obj2dev(parent);
 928         th = dt_trans_create(env, dev);
 929         if (IS_ERR(th))
 930                 RETURN(PTR_ERR(th));
 931
 932         th->th_sync = 1;
 933         /* 5a. insert name into parent dir */
 934         rec->rec_fid = cfid;
 935         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
 936                                (const struct dt_key *)name, th);
 937         if (rc != 0)
 938                 GOTO(stop, rc);
 939
 940         /* 6a. increase parent nlink */
 941         rc = dt_declare_ref_add(env, parent, th);
 942         if (rc != 0)
 943                 GOTO(stop, rc);
 944
 945         rc = dt_trans_start_local(env, dev, th);
 946         if (rc != 0)
 947                 GOTO(stop, rc);
 948
 949         /* 5b. insert name into parent dir */
 950         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
 951                        (const struct dt_key *)name, th, 1);
 952         if (rc != 0)
 953                 GOTO(stop, rc);
 954
 955         dt_write_lock(env, parent, 0);
 956         /* 6b. increase parent nlink */
 957         rc = dt_ref_add(env, parent, th);
 958         dt_write_unlock(env, parent);
 959
 960         GOTO(stop, rc);
 961
 962 unlock:
 963         dt_write_unlock(env, child);
 964 stop:
 965         dt_trans_stop(env, dev, th);
 966
 967         if (rc != 0 && dev == lfsck_obj2dev(parent))
 968                 CDEBUG(D_LFSCK, "%s: partially created the object "DFID
 969                        "for orphans, but failed to insert the name %s "
 970                        "to the .lustre/lost+found/. Such inconsistency "
 971                        "will be repaired when LFSCK run next time: rc = %d\n",
 972                        lfsck_lfsck2name(lfsck), PFID(cfid), name, rc);
 973
 974         return rc;
 975 }
 976
 977 /**
 978  * Create the MDTxxxx directory under /ROOT/.lustre/lost+found/
 979  *
 980  * The /ROOT/.lustre/lost+found/MDTxxxx/ directory is used for holding
 981  * orphans and other uncertain inconsistent objects found during the
 982  * LFSCK. Such directory will be created by the LFSCK engine on the
 983  * local MDT before the LFSCK scanning.
 984  *
 985  * \param[in] env       pointer to the thread context
 986  * \param[in] lfsck     pointer to the lfsck instance
 987  *
 988  * \retval              0 for success
 989  * \retval              negative error number on failure
 990  */
 991 static int lfsck_create_lpf(const struct lu_env *env,
 992                             struct lfsck_instance *lfsck)
 993 {
 994         struct lfsck_bookmark    *bk    = &lfsck->li_bookmark_ram;
 995         struct lfsck_thread_info *info  = lfsck_env_info(env);
 996         struct lu_fid            *cfid  = &info->lti_fid2;
 997         struct lu_attr           *la    = &info->lti_la;
 998         struct dt_object_format  *dof   = &info->lti_dof;
 999         struct dt_object         *parent = lfsck->li_lpf_root_obj;
1000         struct dt_object         *child = NULL;
1001         struct lfsck_lock_handle *llh   = &info->lti_llh;
1002         char                      name[8];
1003         int                       node  = lfsck_dev_idx(lfsck);
1004         int                       rc    = 0;
1005         ENTRY;
1006
1007         LASSERT(lfsck->li_master);
1008         LASSERT(parent != NULL);
1009         LASSERT(lfsck->li_lpf_obj == NULL);
1010
1011         snprintf(name, 8, "MDT%04x", node);
1012         rc = lfsck_lock(env, lfsck, parent, name, llh,
1013                         MDS_INODELOCK_UPDATE, LCK_PW);
1014         if (rc != 0)
1015                 RETURN(rc);
1016
1017         if (fid_is_zero(&bk->lb_lpf_fid)) {
1018                 /* There is corner case that: in former LFSCK scanning we have
1019                  * created the .lustre/lost+found/MDTxxxx but failed to update
1020                  * the lfsck_bookmark::lb_lpf_fid successfully. So need lookup
1021                  * it from MDT0 firstly. */
1022                 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1023                                (const struct dt_key *)name);
1024                 if (rc != 0 && rc != -ENOENT)
1025                         GOTO(unlock, rc);
1026
1027                 if (rc == 0) {
1028                         bk->lb_lpf_fid = *cfid;
1029                         rc = lfsck_bookmark_store(env, lfsck);
1030                 } else {
1031                         rc = lfsck_fid_alloc(env, lfsck, cfid, true);
1032                 }
1033                 if (rc != 0)
1034                         GOTO(unlock, rc);
1035         } else {
1036                 *cfid = bk->lb_lpf_fid;
1037         }
1038
1039         child = lfsck_object_find_bottom_new(env, lfsck, cfid);
1040         if (IS_ERR(child))
1041                 GOTO(unlock, rc = PTR_ERR(child));
1042
1043         if (dt_object_exists(child) != 0) {
1044                 if (unlikely(!dt_try_as_dir(env, child)))
1045                         rc = -ENOTDIR;
1046                 else
1047                         lfsck->li_lpf_obj = child;
1048
1049                 GOTO(unlock, rc);
1050         }
1051
1052         memset(la, 0, sizeof(*la));
1053         la->la_atime = la->la_mtime = la->la_ctime = ktime_get_real_seconds();
1054         la->la_mode = S_IFDIR | S_IRWXU;
1055         la->la_valid = LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1056                        LA_UID | LA_GID;
1057         memset(dof, 0, sizeof(*dof));
1058         dof->dof_type = dt_mode_to_dft(S_IFDIR);
1059
1060         if (node == 0)
1061                 rc = lfsck_create_lpf_local(env, lfsck, child, la, dof, name);
1062         else
1063                 rc = lfsck_create_lpf_remote(env, lfsck, child, la, dof, name);
1064         if (rc == 0)
1065                 lfsck->li_lpf_obj = child;
1066
1067         GOTO(unlock, rc);
1068
1069 unlock:
1070         lfsck_unlock(llh);
1071         if (rc != 0 && child != NULL && !IS_ERR(child))
1072                 lfsck_object_put(env, child);
1073
1074         return rc;
1075 }
1076
1077 /**
1078  * Scan .lustre/lost+found for bad name entries and remove them.
1079  *
1080  * The valid name entry should be "MDTxxxx", the "xxxx" is the MDT device
1081  * index in the system. Any other formatted name is invalid and should be
1082  * removed.
1083  *
1084  * \param[in] env       pointer to the thread context
1085  * \param[in] lfsck     pointer to the lfsck instance
1086  *
1087  * \retval              0 for success
1088  * \retval              negative error number on failure
1089  */
1090 static int lfsck_scan_lpf_bad_entries(const struct lu_env *env,
1091                                       struct lfsck_instance *lfsck)
1092 {
1093         struct dt_object        *parent = lfsck->li_lpf_root_obj;
1094         struct lu_dirent        *ent    =
1095                         (struct lu_dirent *)lfsck_env_info(env)->lti_key;
1096         const struct dt_it_ops  *iops   = &parent->do_index_ops->dio_it;
1097         struct dt_it            *it;
1098         int                      rc;
1099         ENTRY;
1100
1101         it = iops->init(env, parent, LUDA_64BITHASH);
1102         if (IS_ERR(it))
1103                 RETURN(PTR_ERR(it));
1104
1105         rc = iops->load(env, it, 0);
1106         if (rc == 0)
1107                 rc = iops->next(env, it);
1108         else if (rc > 0)
1109                 rc = 0;
1110
1111         while (rc == 0) {
1112                 int off = 3;
1113
1114                 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
1115                 if (rc != 0)
1116                         break;
1117
1118                 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
1119                 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
1120                         goto next;
1121
1122                 /* name length must be strlen("MDTxxxx") */
1123                 if (ent->lde_namelen != 7)
1124                         goto remove;
1125
1126                 if (memcmp(ent->lde_name, "MDT", off) != 0)
1127                         goto remove;
1128
1129                 while (off < 7 && isxdigit(ent->lde_name[off]))
1130                         off++;
1131
1132                 if (off != 7) {
1133
1134 remove:
1135                         rc = lfsck_lpf_remove_name_entry(env, lfsck,
1136                                                          ent->lde_name);
1137                         if (rc != 0)
1138                                 break;
1139                 }
1140
1141 next:
1142                 rc = iops->next(env, it);
1143         }
1144
1145         iops->put(env, it);
1146         iops->fini(env, it);
1147
1148         RETURN(rc > 0 ? 0 : rc);
1149 }
1150
1151 static int lfsck_update_lpf_entry(const struct lu_env *env,
1152                                   struct lfsck_instance *lfsck,
1153                                   struct dt_object *parent,
1154                                   struct dt_object *child,
1155                                   const char *name,
1156                                   enum lfsck_verify_lpf_types type)
1157 {
1158         int rc;
1159
1160         if (type == LVLT_BY_BOOKMARK) {
1161                 rc = lfsck_update_name_entry(env, lfsck, parent, name,
1162                                              lfsck_dto2fid(child), S_IFDIR);
1163         } else /* if (type == LVLT_BY_NAMEENTRY) */ {
1164                 lfsck->li_bookmark_ram.lb_lpf_fid = *lfsck_dto2fid(child);
1165                 rc = lfsck_bookmark_store(env, lfsck);
1166
1167                 CDEBUG(D_LFSCK, "%s: update LPF fid "DFID
1168                        " in the bookmark file: rc = %d\n",
1169                        lfsck_lfsck2name(lfsck),
1170                        PFID(lfsck_dto2fid(child)), rc);
1171         }
1172
1173         return rc;
1174 }
1175
1176 /**
1177  * Check whether the @child back references the @parent.
1178  *
1179  * Two cases:
1180  * 1) The child's FID is stored in the bookmark file. If the child back
1181  *    references the parent (LU_LPF_FID object) via its ".." entry, then
1182  *    insert the name (MDTxxxx) to the .lustre/lost+found; otherwise, if
1183  *    the child back references another parent2, then:
1184  * 1.1) If the parent2 recognizes the child, then update the bookmark file;
1185  * 1.2) Otherwise, the LFSCK cannot know whether there will be parent3 that
1186  *      references the child. So keep them there. As the LFSCK processing,
1187  *      the parent3 may be found, then when the LFSCK run next time, the
1188  *      inconsistency can be repaired.
1189  *
1190  * 2) The child's FID is stored in the .lustre/lost+found/ sub-directory name
1191  *    entry (MDTxxxx). If the child back references the parent (LU_LPF_FID obj)
1192  *    via its ".." entry, then update the bookmark file, otherwise, if the child
1193  *    back references another parent2, then:
1194  * 2.1) If the parent2 recognizes the child, then remove the sub-directory
1195  *      from .lustre/lost+found/;
1196  * 2.2) Otherwise, if the parent2 does not recognizes the child, trust the
1197  *      sub-directory name entry and update the child;
1198  * 2.3) Otherwise, if we do not know whether the parent2 recognizes the child
1199  *      or not, then keep them there.
1200  *
1201  * \param[in] env       pointer to the thread context
1202  * \param[in] lfsck     pointer to the lfsck instance
1203  * \param[in] child     pointer to the lost+found sub-directory object
1204  * \param[in] name      the name for lost+found sub-directory object
1205  * \param[out] fid      pointer to the buffer to hold the FID of the object
1206  *                      (called it as parent2) that is referenced via the
1207  *                      child's dotdot entry; it also can be the FID that
1208  *                      is referenced by the name entry under the parent2.
1209  * \param[in] type      to indicate where the child's FID is stored in
1210  *
1211  * \retval              positive number for uncertain inconsistency
1212  * \retval              0 for success
1213  * \retval              negative error number on failure
1214  */
1215 static int lfsck_verify_lpf_pairs(const struct lu_env *env,
1216                                   struct lfsck_instance *lfsck,
1217                                   struct dt_object *child, const char *name,
1218                                   struct lu_fid *fid,
1219                                   enum lfsck_verify_lpf_types type)
1220 {
1221         struct dt_object         *parent  = lfsck->li_lpf_root_obj;
1222         struct lfsck_thread_info *info    = lfsck_env_info(env);
1223         char                     *name2   = info->lti_key;
1224         struct lu_fid            *fid2    = &info->lti_fid3;
1225         struct dt_object         *parent2 = NULL;
1226         struct lustre_handle      lh      = { 0 };
1227         int                       rc;
1228         ENTRY;
1229
1230         fid_zero(fid);
1231         rc = dt_lookup(env, child, (struct dt_rec *)fid,
1232                        (const struct dt_key *)dotdot);
1233         if (rc != 0)
1234                 GOTO(linkea, rc);
1235
1236         if (!fid_is_sane(fid))
1237                 GOTO(linkea, rc = -EINVAL);
1238
1239         if (lu_fid_eq(fid, &LU_LPF_FID)) {
1240                 const struct lu_name *cname;
1241
1242                 if (lfsck->li_lpf_obj == NULL) {
1243                         lu_object_get(&child->do_lu);
1244                         lfsck->li_lpf_obj = child;
1245                 }
1246
1247                 cname = lfsck_name_get_const(env, name, strlen(name));
1248                 rc = lfsck_verify_linkea(env, child, cname, &LU_LPF_FID);
1249                 if (rc == 0)
1250                         rc = lfsck_update_lpf_entry(env, lfsck, parent, child,
1251                                                     name, type);
1252
1253                 GOTO(out_done, rc);
1254         }
1255
1256         parent2 = lfsck_object_find_bottom(env, lfsck, fid);
1257         if (IS_ERR(parent2))
1258                 GOTO(linkea, parent2);
1259
1260         if (!dt_object_exists(parent2)) {
1261                 lfsck_object_put(env, parent2);
1262
1263                 GOTO(linkea, parent2 = ERR_PTR(-ENOENT));
1264         }
1265
1266         if (!dt_try_as_dir(env, parent2)) {
1267                 lfsck_object_put(env, parent2);
1268
1269                 GOTO(linkea, parent2 = ERR_PTR(-ENOTDIR));
1270         }
1271
1272 linkea:
1273         /* To prevent rename/unlink race */
1274         rc = lfsck_ibits_lock(env, lfsck, child, &lh,
1275                               MDS_INODELOCK_UPDATE, LCK_PR);
1276         if (rc != 0)
1277                 GOTO(out_put, rc);
1278
1279         dt_read_lock(env, child, 0);
1280         rc = lfsck_links_get_first(env, child, name2, fid2);
1281         if (rc != 0) {
1282                 dt_read_unlock(env, child);
1283                 lfsck_ibits_unlock(&lh, LCK_PR);
1284
1285                 GOTO(out_put, rc = 1);
1286         }
1287
1288         /* It is almost impossible that the bookmark file (or the name entry)
1289          * and the linkEA hit the same data corruption. Trust the linkEA. */
1290         if (lu_fid_eq(fid2, &LU_LPF_FID) && strcmp(name, name2) == 0) {
1291                 dt_read_unlock(env, child);
1292                 lfsck_ibits_unlock(&lh, LCK_PR);
1293
1294                 *fid = *fid2;
1295                 if (lfsck->li_lpf_obj == NULL) {
1296                         lu_object_get(&child->do_lu);
1297                         lfsck->li_lpf_obj = child;
1298                 }
1299
1300                 /* Update the child's dotdot entry */
1301                 rc = lfsck_update_name_entry(env, lfsck, child, dotdot,
1302                                              &LU_LPF_FID, S_IFDIR);
1303                 if (rc == 0)
1304                         rc = lfsck_update_lpf_entry(env, lfsck, parent, child,
1305                                                     name, type);
1306
1307                 GOTO(out_put, rc);
1308         }
1309
1310         if (parent2 == NULL || IS_ERR(parent2)) {
1311                 dt_read_unlock(env, child);
1312                 lfsck_ibits_unlock(&lh, LCK_PR);
1313
1314                 GOTO(out_done, rc = 1);
1315         }
1316
1317         rc = dt_lookup(env, parent2, (struct dt_rec *)fid,
1318                        (const struct dt_key *)name2);
1319         dt_read_unlock(env, child);
1320         lfsck_ibits_unlock(&lh, LCK_PR);
1321         if (rc != 0 && rc != -ENOENT)
1322                 GOTO(out_put, rc);
1323
1324         if (rc == -ENOENT || !lu_fid_eq(fid, lfsck_dto2fid(child))) {
1325                 if (type == LVLT_BY_BOOKMARK)
1326                         GOTO(out_put, rc = 1);
1327
1328                 /* Trust the name entry, update the child's dotdot entry. */
1329                 rc = lfsck_update_name_entry(env, lfsck, child, dotdot,
1330                                              &LU_LPF_FID, S_IFDIR);
1331
1332                 GOTO(out_put, rc);
1333         }
1334
1335         if (type == LVLT_BY_BOOKMARK) {
1336                 /* Invalid FID record in the bookmark file, reset it. */
1337                 fid_zero(&lfsck->li_bookmark_ram.lb_lpf_fid);
1338                 rc = lfsck_bookmark_store(env, lfsck);
1339
1340                 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1341                        " in the bookmark file: rc = %d\n",
1342                        lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(child)), rc);
1343         } else /* if (type == LVLT_BY_NAMEENTRY) */ {
1344                 /* The name entry is wrong, remove it. */
1345                 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1346         }
1347
1348         GOTO(out_put, rc);
1349
1350 out_put:
1351         if (parent2 != NULL && !IS_ERR(parent2))
1352                 lfsck_object_put(env, parent2);
1353
1354 out_done:
1355         return rc;
1356 }
1357
1358 /**
1359  * Verify the /ROOT/.lustre/lost+found/ directory.
1360  *
1361  * /ROOT/.lustre/lost+found/ is a special directory to hold the objects that
1362  * the LFSCK does not exactly know how to handle, such as orphans. So before
1363  * the LFSCK scanning the system, the consistency of such directory needs to
1364  * be verified firstly to allow the users to use it during the LFSCK.
1365  *
1366  * \param[in] env       pointer to the thread context
1367  * \param[in] lfsck     pointer to the lfsck instance
1368  *
1369  * \retval              positive number for uncertain inconsistency
1370  * \retval              0 for success
1371  * \retval              negative error number on failure
1372  */
1373 int lfsck_verify_lpf(const struct lu_env *env, struct lfsck_instance *lfsck)
1374 {
1375         struct lfsck_thread_info *info   = lfsck_env_info(env);
1376         struct lu_fid            *pfid   = &info->lti_fid;
1377         struct lu_fid            *cfid   = &info->lti_fid2;
1378         struct lfsck_bookmark    *bk     = &lfsck->li_bookmark_ram;
1379         struct dt_object         *parent;
1380         /* child1's FID is in the bookmark file. */
1381         struct dt_object         *child1 = NULL;
1382         /* child2's FID is in the name entry MDTxxxx. */
1383         struct dt_object         *child2 = NULL;
1384         const struct lu_name     *cname;
1385         char                      name[8];
1386         int                       node   = lfsck_dev_idx(lfsck);
1387         int                       rc     = 0;
1388         ENTRY;
1389
1390         LASSERT(lfsck->li_master);
1391
1392         if (lfsck->li_lpf_root_obj != NULL)
1393                 RETURN(0);
1394
1395         if (node == 0) {
1396                 parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
1397                                                   &LU_LPF_FID);
1398         } else {
1399                 struct lfsck_tgt_desc *ltd;
1400
1401                 ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, 0);
1402                 if (unlikely(ltd == NULL))
1403                         RETURN(-ENXIO);
1404
1405                 parent = lfsck_object_find_by_dev(env, ltd->ltd_tgt,
1406                                                   &LU_LPF_FID);
1407                 lfsck_tgt_put(ltd);
1408         }
1409
1410         if (IS_ERR(parent))
1411                 RETURN(PTR_ERR(parent));
1412
1413         LASSERT(dt_object_exists(parent));
1414
1415         if (unlikely(!dt_try_as_dir(env, parent))) {
1416                 lfsck_object_put(env, parent);
1417
1418                 GOTO(put, rc = -ENOTDIR);
1419         }
1420
1421         lfsck->li_lpf_root_obj = parent;
1422         if (node == 0) {
1423                 rc = lfsck_scan_lpf_bad_entries(env, lfsck);
1424                 if (rc != 0)
1425                         CDEBUG(D_LFSCK, "%s: scan .lustre/lost+found/ "
1426                                "for bad sub-directories: rc = %d\n",
1427                                lfsck_lfsck2name(lfsck), rc);
1428         }
1429
1430         /* child2 */
1431         snprintf(name, 8, "MDT%04x", node);
1432         rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1433                        (const struct dt_key *)name);
1434         if (rc == -ENOENT) {
1435                 rc = 0;
1436                 goto find_child1;
1437         }
1438
1439         if (rc != 0)
1440                 GOTO(put, rc);
1441
1442         /* Invalid FID in the name entry, remove the name entry. */
1443         if (!fid_is_norm(cfid)) {
1444                 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1445                 if (rc != 0)
1446                         GOTO(put, rc);
1447
1448                 goto find_child1;
1449         }
1450
1451         child2 = lfsck_object_find_bottom(env, lfsck, cfid);
1452         if (IS_ERR(child2))
1453                 GOTO(put, rc = PTR_ERR(child2));
1454
1455         if (unlikely(!dt_object_exists(child2) ||
1456                      dt_object_remote(child2)) ||
1457                      !S_ISDIR(lfsck_object_type(child2))) {
1458                 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1459                 if (rc != 0)
1460                         GOTO(put, rc);
1461
1462                 goto find_child1;
1463         }
1464
1465         if (unlikely(!dt_try_as_dir(env, child2))) {
1466                 lfsck_object_put(env, child2);
1467                 child2 = NULL;
1468                 rc = -ENOTDIR;
1469         }
1470
1471 find_child1:
1472         if (fid_is_zero(&bk->lb_lpf_fid))
1473                 goto check_child2;
1474
1475         if (likely(lu_fid_eq(cfid, &bk->lb_lpf_fid))) {
1476                 if (lfsck->li_lpf_obj == NULL) {
1477                         lu_object_get(&child2->do_lu);
1478                         lfsck->li_lpf_obj = child2;
1479                 }
1480
1481                 cname = lfsck_name_get_const(env, name, strlen(name));
1482                 rc = lfsck_verify_linkea(env, child2, cname, &LU_LPF_FID);
1483
1484                 GOTO(put, rc);
1485         }
1486
1487         if (unlikely(!fid_is_norm(&bk->lb_lpf_fid))) {
1488                 struct lu_fid tfid = bk->lb_lpf_fid;
1489
1490                 /* Invalid FID record in the bookmark file, reset it. */
1491                 fid_zero(&bk->lb_lpf_fid);
1492                 rc = lfsck_bookmark_store(env, lfsck);
1493
1494                 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1495                        " in the bookmark file: rc = %d\n",
1496                        lfsck_lfsck2name(lfsck), PFID(&tfid), rc);
1497
1498                 if (rc != 0)
1499                         GOTO(put, rc);
1500
1501                 goto check_child2;
1502         }
1503
1504         child1 = lfsck_object_find_bottom(env, lfsck, &bk->lb_lpf_fid);
1505         if (IS_ERR(child1)) {
1506                 child1 = NULL;
1507                 goto check_child2;
1508         }
1509
1510         if (unlikely(!dt_object_exists(child1) ||
1511                      dt_object_remote(child1)) ||
1512                      !S_ISDIR(lfsck_object_type(child1))) {
1513                 /* Invalid FID record in the bookmark file, reset it. */
1514                 fid_zero(&bk->lb_lpf_fid);
1515                 rc = lfsck_bookmark_store(env, lfsck);
1516
1517                 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1518                        " in the bookmark file: rc = %d\n",
1519                        lfsck_lfsck2name(lfsck),
1520                        PFID(lfsck_dto2fid(child1)), rc);
1521
1522                 if (rc != 0)
1523                         GOTO(put, rc);
1524
1525                 lfsck_object_put(env, child1);
1526                 child1 = NULL;
1527                 goto check_child2;
1528         }
1529
1530         if (unlikely(!dt_try_as_dir(env, child1))) {
1531                 lfsck_object_put(env, child1);
1532                 child1 = NULL;
1533                 rc = -ENOTDIR;
1534                 goto check_child2;
1535         }
1536
1537         rc = lfsck_verify_lpf_pairs(env, lfsck, child1, name, pfid,
1538                                     LVLT_BY_BOOKMARK);
1539         if (lu_fid_eq(pfid, &LU_LPF_FID))
1540                 GOTO(put, rc);
1541
1542 check_child2:
1543         if (child2 != NULL)
1544                 rc = lfsck_verify_lpf_pairs(env, lfsck, child2, name,
1545                                             pfid, LVLT_BY_NAMEENTRY);
1546
1547         GOTO(put, rc);
1548
1549 put:
1550         if (lfsck->li_lpf_obj != NULL) {
1551                 if (unlikely(!dt_try_as_dir(env, lfsck->li_lpf_obj))) {
1552                         lfsck_object_put(env, lfsck->li_lpf_obj);
1553                         lfsck->li_lpf_obj = NULL;
1554                         rc = -ENOTDIR;
1555                 }
1556         } else if (rc == 0) {
1557                 rc = lfsck_create_lpf(env, lfsck);
1558         }
1559
1560         if (child2 != NULL && !IS_ERR(child2))
1561                 lfsck_object_put(env, child2);
1562         if (child1 != NULL && !IS_ERR(child1))
1563                 lfsck_object_put(env, child1);
1564
1565         return rc;
1566 }
1567
1568 static int lfsck_fid_init(struct lfsck_instance *lfsck)
1569 {
1570         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
1571         struct seq_server_site  *ss     = lfsck_dev_site(lfsck);
1572         char                    *prefix;
1573         int                      rc     = 0;
1574         ENTRY;
1575
1576         if (unlikely(ss == NULL))
1577                 RETURN(-ENXIO);
1578
1579         OBD_ALLOC_PTR(lfsck->li_seq);
1580         if (lfsck->li_seq == NULL)
1581                 RETURN(-ENOMEM);
1582
1583         OBD_ALLOC(prefix, MAX_OBD_NAME + 7);
1584         if (prefix == NULL)
1585                 GOTO(out, rc = -ENOMEM);
1586
1587         snprintf(prefix, MAX_OBD_NAME + 7, "lfsck-%s", lfsck_lfsck2name(lfsck));
1588         rc = seq_client_init(lfsck->li_seq, NULL, LUSTRE_SEQ_METADATA, prefix,
1589                              ss->ss_server_seq);
1590         OBD_FREE(prefix, MAX_OBD_NAME + 7);
1591         if (rc != 0)
1592                 GOTO(out, rc);
1593
1594         if (fid_is_sane(&bk->lb_last_fid))
1595                 lfsck->li_seq->lcs_fid = bk->lb_last_fid;
1596
1597         RETURN(0);
1598
1599 out:
1600         OBD_FREE_PTR(lfsck->li_seq);
1601         lfsck->li_seq = NULL;
1602
1603         return rc;
1604 }
1605
1606 static void lfsck_fid_fini(struct lfsck_instance *lfsck)
1607 {
1608         if (lfsck->li_seq != NULL) {
1609                 seq_client_fini(lfsck->li_seq);
1610                 OBD_FREE_PTR(lfsck->li_seq);
1611                 lfsck->li_seq = NULL;
1612         }
1613 }
1614
1615 void lfsck_instance_cleanup(const struct lu_env *env,
1616                             struct lfsck_instance *lfsck)
1617 {
1618         struct ptlrpc_thread    *thread = &lfsck->li_thread;
1619         struct lfsck_component  *com;
1620         struct lfsck_component  *next;
1621         struct lfsck_lmv_unit   *llu;
1622         struct lfsck_lmv_unit   *llu_next;
1623         struct lfsck_lmv        *llmv;
1624         ENTRY;
1625
1626         LASSERT(list_empty(&lfsck->li_link));
1627         LASSERT(thread_is_init(thread) || thread_is_stopped(thread));
1628
1629         if (lfsck->li_obj_oit != NULL) {
1630                 lfsck_object_put(env, lfsck->li_obj_oit);
1631                 lfsck->li_obj_oit = NULL;
1632         }
1633
1634         LASSERT(lfsck->li_obj_dir == NULL);
1635         LASSERT(lfsck->li_lmv == NULL);
1636
1637         list_for_each_entry_safe(llu, llu_next, &lfsck->li_list_lmv, llu_link) {
1638                 llmv = &llu->llu_lmv;
1639
1640                 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
1641                          "still in using: %u\n",
1642                          atomic_read(&llmv->ll_ref));
1643
1644                 lfsck_lmv_put(env, llmv);
1645         }
1646
1647         list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
1648                 lfsck_component_cleanup(env, com);
1649         }
1650
1651         LASSERT(list_empty(&lfsck->li_list_dir));
1652
1653         list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
1654                                  lc_link) {
1655                 lfsck_component_cleanup(env, com);
1656         }
1657
1658         list_for_each_entry_safe(com, next, &lfsck->li_list_idle, lc_link) {
1659                 lfsck_component_cleanup(env, com);
1660         }
1661
1662         lfsck_tgt_descs_fini(&lfsck->li_ost_descs);
1663         lfsck_tgt_descs_fini(&lfsck->li_mdt_descs);
1664
1665         if (lfsck->li_lfsck_dir != NULL) {
1666                 lfsck_object_put(env, lfsck->li_lfsck_dir);
1667                 lfsck->li_lfsck_dir = NULL;
1668         }
1669
1670         if (lfsck->li_bookmark_obj != NULL) {
1671                 lfsck_object_put(env, lfsck->li_bookmark_obj);
1672                 lfsck->li_bookmark_obj = NULL;
1673         }
1674
1675         if (lfsck->li_lpf_obj != NULL) {
1676                 lfsck_object_put(env, lfsck->li_lpf_obj);
1677                 lfsck->li_lpf_obj = NULL;
1678         }
1679
1680         if (lfsck->li_lpf_root_obj != NULL) {
1681                 lfsck_object_put(env, lfsck->li_lpf_root_obj);
1682                 lfsck->li_lpf_root_obj = NULL;
1683         }
1684
1685         if (lfsck->li_los != NULL) {
1686                 local_oid_storage_fini(env, lfsck->li_los);
1687                 lfsck->li_los = NULL;
1688         }
1689
1690         lfsck_fid_fini(lfsck);
1691
1692         OBD_FREE_PTR(lfsck);
1693 }
1694
1695 static inline struct lfsck_instance *
1696 __lfsck_instance_find(struct dt_device *key, bool ref, bool unlink)
1697 {
1698         struct lfsck_instance *lfsck;
1699
1700         list_for_each_entry(lfsck, &lfsck_instance_list, li_link) {
1701                 if (lfsck->li_bottom == key) {
1702                         if (ref)
1703                                 lfsck_instance_get(lfsck);
1704                         if (unlink)
1705                                 list_del_init(&lfsck->li_link);
1706
1707                         return lfsck;
1708                 }
1709         }
1710
1711         return NULL;
1712 }
1713
1714 struct lfsck_instance *lfsck_instance_find(struct dt_device *key, bool ref,
1715                                            bool unlink)
1716 {
1717         struct lfsck_instance *lfsck;
1718
1719         spin_lock(&lfsck_instance_lock);
1720         lfsck = __lfsck_instance_find(key, ref, unlink);
1721         spin_unlock(&lfsck_instance_lock);
1722
1723         return lfsck;
1724 }
1725
1726 static inline int lfsck_instance_add(struct lfsck_instance *lfsck)
1727 {
1728         struct lfsck_instance *tmp;
1729
1730         spin_lock(&lfsck_instance_lock);
1731         list_for_each_entry(tmp, &lfsck_instance_list, li_link) {
1732                 if (lfsck->li_bottom == tmp->li_bottom) {
1733                         spin_unlock(&lfsck_instance_lock);
1734                         return -EEXIST;
1735                 }
1736         }
1737
1738         list_add_tail(&lfsck->li_link, &lfsck_instance_list);
1739         spin_unlock(&lfsck_instance_lock);
1740         return 0;
1741 }
1742
1743 void lfsck_bits_dump(struct seq_file *m, int bits, const char *names[],
1744                      const char *prefix)
1745 {
1746         int flag;
1747         int i;
1748         bool newline = (bits != 0 ? false : true);
1749
1750         seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
1751
1752         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
1753                 if (flag & bits) {
1754                         bits &= ~flag;
1755                         if (names[i] != NULL) {
1756                                 if (bits == 0)
1757                                         newline = true;
1758
1759                                 seq_printf(m, "%s%c", names[i],
1760                                            newline ? '\n' : ',');
1761                         }
1762                 }
1763         }
1764
1765         if (!newline)
1766                 seq_putc(m, '\n');
1767 }
1768
1769 void lfsck_time_dump(struct seq_file *m, time64_t time, const char *name)
1770 {
1771         if (time == 0) {
1772                 seq_printf(m, "%s_time: N/A\n", name);
1773                 seq_printf(m, "time_since_%s: N/A\n", name);
1774         } else {
1775                 seq_printf(m, "%s_time: %lld\n", name, time);
1776                 seq_printf(m, "time_since_%s: %lld seconds\n",
1777                            name, ktime_get_real_seconds() - time);
1778         }
1779 }
1780
1781 void lfsck_pos_dump(struct seq_file *m, struct lfsck_position *pos,
1782                     const char *prefix)
1783 {
1784         if (fid_is_zero(&pos->lp_dir_parent)) {
1785                 if (pos->lp_oit_cookie == 0) {
1786                         seq_printf(m, "%s: N/A, N/A, N/A\n", prefix);
1787                         return;
1788                 }
1789                 seq_printf(m, "%s: %llu, N/A, N/A\n",
1790                            prefix, pos->lp_oit_cookie);
1791         } else {
1792                 seq_printf(m, "%s: %llu, "DFID", %#llx\n",
1793                            prefix, pos->lp_oit_cookie,
1794                            PFID(&pos->lp_dir_parent), pos->lp_dir_cookie);
1795         }
1796 }
1797
1798 void lfsck_pos_fill(const struct lu_env *env, struct lfsck_instance *lfsck,
1799                     struct lfsck_position *pos, bool init)
1800 {
1801         const struct dt_it_ops *iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
1802
1803         if (unlikely(lfsck->li_di_oit == NULL)) {
1804                 memset(pos, 0, sizeof(*pos));
1805                 return;
1806         }
1807
1808         pos->lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
1809         if (!lfsck->li_current_oit_processed && !init)
1810                 pos->lp_oit_cookie--;
1811
1812         LASSERT(pos->lp_oit_cookie > 0);
1813
1814         if (lfsck->li_di_dir != NULL) {
1815                 struct dt_object *dto = lfsck->li_obj_dir;
1816
1817                 pos->lp_dir_cookie = dto->do_index_ops->dio_it.store(env,
1818                                                         lfsck->li_di_dir);
1819
1820                 if (pos->lp_dir_cookie >= MDS_DIR_END_OFF) {
1821                         fid_zero(&pos->lp_dir_parent);
1822                         pos->lp_dir_cookie = 0;
1823                 } else {
1824                         pos->lp_dir_parent = *lfsck_dto2fid(dto);
1825                 }
1826         } else {
1827                 fid_zero(&pos->lp_dir_parent);
1828                 pos->lp_dir_cookie = 0;
1829         }
1830 }
1831
1832 bool __lfsck_set_speed(struct lfsck_instance *lfsck, __u32 limit)
1833 {
1834         bool dirty = false;
1835
1836         if (limit != LFSCK_SPEED_NO_LIMIT) {
1837                 if (limit > msecs_to_jiffies(MSEC_PER_SEC)) {
1838                         lfsck->li_sleep_rate = limit /
1839                                                msecs_to_jiffies(MSEC_PER_SEC);
1840                         lfsck->li_sleep_jif = 1;
1841                 } else {
1842                         lfsck->li_sleep_rate = 1;
1843                         lfsck->li_sleep_jif = msecs_to_jiffies(MSEC_PER_SEC) /
1844                                               limit;
1845                 }
1846         } else {
1847                 lfsck->li_sleep_jif = 0;
1848                 lfsck->li_sleep_rate = 0;
1849         }
1850
1851         if (lfsck->li_bookmark_ram.lb_speed_limit != limit) {
1852                 lfsck->li_bookmark_ram.lb_speed_limit = limit;
1853                 dirty = true;
1854         }
1855
1856         return dirty;
1857 }
1858
1859 void lfsck_control_speed(struct lfsck_instance *lfsck)
1860 {
1861         struct ptlrpc_thread *thread = &lfsck->li_thread;
1862         struct l_wait_info    lwi;
1863
1864         if (lfsck->li_sleep_jif > 0 &&
1865             lfsck->li_new_scanned >= lfsck->li_sleep_rate) {
1866                 lwi = LWI_TIMEOUT_INTR(lfsck->li_sleep_jif, NULL,
1867                                        LWI_ON_SIGNAL_NOOP, NULL);
1868
1869                 l_wait_event(thread->t_ctl_waitq,
1870                              !thread_is_running(thread),
1871                              &lwi);
1872                 lfsck->li_new_scanned = 0;
1873         }
1874 }
1875
1876 void lfsck_control_speed_by_self(struct lfsck_component *com)
1877 {
1878         struct lfsck_instance   *lfsck  = com->lc_lfsck;
1879         struct ptlrpc_thread    *thread = &lfsck->li_thread;
1880         struct l_wait_info       lwi;
1881
1882         if (lfsck->li_sleep_jif > 0 &&
1883             com->lc_new_scanned >= lfsck->li_sleep_rate) {
1884                 lwi = LWI_TIMEOUT_INTR(lfsck->li_sleep_jif, NULL,
1885                                        LWI_ON_SIGNAL_NOOP, NULL);
1886
1887                 l_wait_event(thread->t_ctl_waitq,
1888                              !thread_is_running(thread),
1889                              &lwi);
1890                 com->lc_new_scanned = 0;
1891         }
1892 }
1893
1894 static struct lfsck_thread_args *
1895 lfsck_thread_args_init(struct lfsck_instance *lfsck,
1896                        struct lfsck_component *com,
1897                        struct lfsck_start_param *lsp)
1898 {
1899         struct lfsck_thread_args *lta;
1900         int                       rc;
1901
1902         OBD_ALLOC_PTR(lta);
1903         if (lta == NULL)
1904                 return ERR_PTR(-ENOMEM);
1905
1906         rc = lu_env_init(&lta->lta_env, LCT_MD_THREAD | LCT_DT_THREAD);
1907         if (rc != 0) {
1908                 OBD_FREE_PTR(lta);
1909                 return ERR_PTR(rc);
1910         }
1911
1912         lta->lta_lfsck = lfsck_instance_get(lfsck);
1913         if (com != NULL)
1914                 lta->lta_com = lfsck_component_get(com);
1915
1916         lta->lta_lsp = lsp;
1917
1918         return lta;
1919 }
1920
1921 void lfsck_thread_args_fini(struct lfsck_thread_args *lta)
1922 {
1923         if (lta->lta_com != NULL)
1924                 lfsck_component_put(&lta->lta_env, lta->lta_com);
1925         lfsck_instance_put(&lta->lta_env, lta->lta_lfsck);
1926         lu_env_fini(&lta->lta_env);
1927         OBD_FREE_PTR(lta);
1928 }
1929
1930 struct lfsck_assistant_data *
1931 lfsck_assistant_data_init(struct lfsck_assistant_operations *lao,
1932                           const char *name)
1933 {
1934         struct lfsck_assistant_data *lad;
1935
1936         OBD_ALLOC_PTR(lad);
1937         if (lad != NULL) {
1938                 lad->lad_bitmap = CFS_ALLOCATE_BITMAP(BITS_PER_LONG);
1939                 if (lad->lad_bitmap == NULL) {
1940                         OBD_FREE_PTR(lad);
1941                         return NULL;
1942                 }
1943
1944                 INIT_LIST_HEAD(&lad->lad_req_list);
1945                 spin_lock_init(&lad->lad_lock);
1946                 INIT_LIST_HEAD(&lad->lad_ost_list);
1947                 INIT_LIST_HEAD(&lad->lad_ost_phase1_list);
1948                 INIT_LIST_HEAD(&lad->lad_ost_phase2_list);
1949                 INIT_LIST_HEAD(&lad->lad_mdt_list);
1950                 INIT_LIST_HEAD(&lad->lad_mdt_phase1_list);
1951                 INIT_LIST_HEAD(&lad->lad_mdt_phase2_list);
1952                 init_waitqueue_head(&lad->lad_thread.t_ctl_waitq);
1953                 lad->lad_ops = lao;
1954                 lad->lad_name = name;
1955         }
1956
1957         return lad;
1958 }
1959
1960 struct lfsck_assistant_object *
1961 lfsck_assistant_object_init(const struct lu_env *env, const struct lu_fid *fid,
1962                             const struct lu_attr *attr, __u64 cookie,
1963                             bool is_dir)
1964 {
1965         struct lfsck_assistant_object   *lso;
1966
1967         OBD_ALLOC_PTR(lso);
1968         if (lso == NULL)
1969                 return ERR_PTR(-ENOMEM);
1970
1971         lso->lso_fid = *fid;
1972         if (attr != NULL)
1973                 lso->lso_attr = *attr;
1974
1975         atomic_set(&lso->lso_ref, 1);
1976         lso->lso_oit_cookie = cookie;
1977         if (is_dir)
1978                 lso->lso_is_dir = 1;
1979
1980         return lso;
1981 }
1982
1983 struct dt_object *
1984 lfsck_assistant_object_load(const struct lu_env *env,
1985                             struct lfsck_instance *lfsck,
1986                             struct lfsck_assistant_object *lso)
1987 {
1988         struct dt_object *obj;
1989
1990         obj = lfsck_object_find_bottom(env, lfsck, &lso->lso_fid);
1991         if (IS_ERR(obj))
1992                 return obj;
1993
1994         if (unlikely(!dt_object_exists(obj) || lfsck_is_dead_obj(obj))) {
1995                 lso->lso_dead = 1;
1996                 lfsck_object_put(env, obj);
1997
1998                 return ERR_PTR(-ENOENT);
1999         }
2000
2001         if (lso->lso_is_dir && unlikely(!dt_try_as_dir(env, obj))) {
2002                 lfsck_object_put(env, obj);
2003
2004                 return ERR_PTR(-ENOTDIR);
2005         }
2006
2007         return obj;
2008 }
2009
2010 /**
2011  * Generic LFSCK asynchronous communication interpretor function.
2012  * The LFSCK RPC reply for both the event notification and status
2013  * querying will be handled here.
2014  *
2015  * \param[in] env       pointer to the thread context
2016  * \param[in] req       pointer to the LFSCK request
2017  * \param[in] args      pointer to the lfsck_async_interpret_args
2018  * \param[in] rc        the result for handling the LFSCK request
2019  *
2020  * \retval              0 for success
2021  * \retval              negative error number on failure
2022  */
2023 int lfsck_async_interpret_common(const struct lu_env *env,
2024                                  struct ptlrpc_request *req,
2025                                  void *args, int rc)
2026 {
2027         struct lfsck_async_interpret_args *laia = args;
2028         struct lfsck_component            *com  = laia->laia_com;
2029         struct lfsck_assistant_data       *lad  = com->lc_data;
2030         struct lfsck_tgt_descs            *ltds = laia->laia_ltds;
2031         struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
2032         struct lfsck_request              *lr   = laia->laia_lr;
2033
2034         LASSERT(com->lc_lfsck->li_master);
2035
2036         switch (lr->lr_event) {
2037         case LE_START:
2038                 if (rc != 0) {
2039                         CDEBUG(D_LFSCK, "%s: fail to notify %s %x for %s "
2040                                "start: rc = %d\n",
2041                                lfsck_lfsck2name(com->lc_lfsck),
2042                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2043                                ltd->ltd_index, lad->lad_name, rc);
2044
2045                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2046                                 struct lfsck_layout *lo = com->lc_file_ram;
2047
2048                                 if (lr->lr_flags & LEF_TO_OST)
2049                                         lfsck_lad_set_bitmap(env, com,
2050                                                              ltd->ltd_index);
2051                                 else
2052                                         lo->ll_flags |= LF_INCOMPLETE;
2053                         } else {
2054                                 struct lfsck_namespace *ns = com->lc_file_ram;
2055
2056                                 /* If some MDT does not join the namespace
2057                                  * LFSCK, then we cannot know whether there
2058                                  * is some name entry on such MDT that with
2059                                  * the referenced MDT-object on this MDT or
2060                                  * not. So the namespace LFSCK on this MDT
2061                                  * cannot handle orphan MDT-objects properly.
2062                                  * So we mark the LFSCK as LF_INCOMPLETE and
2063                                  * skip orphan MDT-objects handling. */
2064                                 ns->ln_flags |= LF_INCOMPLETE;
2065                         }
2066                         break;
2067                 }
2068
2069                 spin_lock(&ltds->ltd_lock);
2070                 if (ltd->ltd_dead) {
2071                         spin_unlock(&ltds->ltd_lock);
2072                         break;
2073                 }
2074
2075                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2076                         struct list_head *list;
2077                         struct list_head *phase_list;
2078
2079                         if (ltd->ltd_layout_done) {
2080                                 spin_unlock(&ltds->ltd_lock);
2081                                 break;
2082                         }
2083
2084                         if (lr->lr_flags & LEF_TO_OST) {
2085                                 list = &lad->lad_ost_list;
2086                                 phase_list = &lad->lad_ost_phase1_list;
2087                         } else {
2088                                 list = &lad->lad_mdt_list;
2089                                 phase_list = &lad->lad_mdt_phase1_list;
2090                         }
2091
2092                         if (list_empty(&ltd->ltd_layout_list))
2093                                 list_add_tail(&ltd->ltd_layout_list, list);
2094                         if (list_empty(&ltd->ltd_layout_phase_list))
2095                                 list_add_tail(&ltd->ltd_layout_phase_list,
2096                                               phase_list);
2097                 } else {
2098                         if (ltd->ltd_namespace_done) {
2099                                 spin_unlock(&ltds->ltd_lock);
2100                                 break;
2101                         }
2102
2103                         if (list_empty(&ltd->ltd_namespace_list))
2104                                 list_add_tail(&ltd->ltd_namespace_list,
2105                                               &lad->lad_mdt_list);
2106                         if (list_empty(&ltd->ltd_namespace_phase_list))
2107                                 list_add_tail(&ltd->ltd_namespace_phase_list,
2108                                               &lad->lad_mdt_phase1_list);
2109                 }
2110                 spin_unlock(&ltds->ltd_lock);
2111                 break;
2112         case LE_STOP:
2113         case LE_PHASE1_DONE:
2114         case LE_PHASE2_DONE:
2115         case LE_PEER_EXIT:
2116                 if (rc != 0 && rc != -EALREADY)
2117                         CDEBUG(D_LFSCK, "%s: fail to notify %s %x for %s: "
2118                               "event = %d, rc = %d\n",
2119                               lfsck_lfsck2name(com->lc_lfsck),
2120                               (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2121                               ltd->ltd_index, lad->lad_name, lr->lr_event, rc);
2122                 break;
2123         case LE_QUERY: {
2124                 struct lfsck_reply *reply;
2125                 struct list_head *list;
2126                 struct list_head *phase_list;
2127
2128                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2129                         list = &ltd->ltd_layout_list;
2130                         phase_list = &ltd->ltd_layout_phase_list;
2131                 } else {
2132                         list = &ltd->ltd_namespace_list;
2133                         phase_list = &ltd->ltd_namespace_phase_list;
2134                 }
2135
2136                 if (rc != 0) {
2137                         if (lr->lr_flags & LEF_QUERY_ALL) {
2138                                 lfsck_reset_ltd_status(ltd, com->lc_type);
2139                                 break;
2140                         }
2141
2142                         spin_lock(&ltds->ltd_lock);
2143                         list_del_init(phase_list);
2144                         list_del_init(list);
2145                         spin_unlock(&ltds->ltd_lock);
2146                         break;
2147                 }
2148
2149                 reply = req_capsule_server_get(&req->rq_pill,
2150                                                &RMF_LFSCK_REPLY);
2151                 if (reply == NULL) {
2152                         rc = -EPROTO;
2153                         CDEBUG(D_LFSCK, "%s: invalid query reply for %s: "
2154                                "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
2155                                lad->lad_name, rc);
2156
2157                         if (lr->lr_flags & LEF_QUERY_ALL) {
2158                                 lfsck_reset_ltd_status(ltd, com->lc_type);
2159                                 break;
2160                         }
2161
2162                         spin_lock(&ltds->ltd_lock);
2163                         list_del_init(phase_list);
2164                         list_del_init(list);
2165                         spin_unlock(&ltds->ltd_lock);
2166                         break;
2167                 }
2168
2169                 if (lr->lr_flags & LEF_QUERY_ALL) {
2170                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2171                                 ltd->ltd_layout_status = reply->lr_status;
2172                                 ltd->ltd_layout_repaired = reply->lr_repaired;
2173                         } else {
2174                                 ltd->ltd_namespace_status = reply->lr_status;
2175                                 ltd->ltd_namespace_repaired =
2176                                                         reply->lr_repaired;
2177                         }
2178                         break;
2179                 }
2180
2181                 switch (reply->lr_status) {
2182                 case LS_SCANNING_PHASE1:
2183                         break;
2184                 case LS_SCANNING_PHASE2:
2185                         spin_lock(&ltds->ltd_lock);
2186                         list_del_init(phase_list);
2187                         if (ltd->ltd_dead) {
2188                                 spin_unlock(&ltds->ltd_lock);
2189                                 break;
2190                         }
2191
2192                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2193                                 if (ltd->ltd_layout_done) {
2194                                         spin_unlock(&ltds->ltd_lock);
2195                                         break;
2196                                 }
2197
2198                                 if (lr->lr_flags & LEF_TO_OST)
2199                                         list_add_tail(phase_list,
2200                                                 &lad->lad_ost_phase2_list);
2201                                 else
2202                                         list_add_tail(phase_list,
2203                                                 &lad->lad_mdt_phase2_list);
2204                         } else {
2205                                 if (ltd->ltd_namespace_done) {
2206                                         spin_unlock(&ltds->ltd_lock);
2207                                         break;
2208                                 }
2209
2210                                 list_add_tail(phase_list,
2211                                               &lad->lad_mdt_phase2_list);
2212                         }
2213                         spin_unlock(&ltds->ltd_lock);
2214                         break;
2215                 default:
2216                         spin_lock(&ltds->ltd_lock);
2217                         list_del_init(phase_list);
2218                         list_del_init(list);
2219                         spin_unlock(&ltds->ltd_lock);
2220                         break;
2221                 }
2222                 break;
2223         }
2224         default:
2225                 CDEBUG(D_LFSCK, "%s: unexpected event: rc = %d\n",
2226                        lfsck_lfsck2name(com->lc_lfsck), lr->lr_event);
2227                 break;
2228         }
2229
2230         if (!laia->laia_shared) {
2231                 lfsck_tgt_put(ltd);
2232                 lfsck_component_put(env, com);
2233         }
2234
2235         return 0;
2236 }
2237
2238 static void lfsck_interpret(const struct lu_env *env,
2239                             struct lfsck_instance *lfsck,
2240                             struct ptlrpc_request *req, void *args, int result)
2241 {
2242         struct lfsck_async_interpret_args *laia = args;
2243         struct lfsck_component            *com;
2244
2245         LASSERT(laia->laia_com == NULL);
2246         LASSERT(laia->laia_shared);
2247
2248         spin_lock(&lfsck->li_lock);
2249         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
2250                 laia->laia_com = com;
2251                 lfsck_async_interpret_common(env, req, laia, result);
2252         }
2253
2254         list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
2255                 laia->laia_com = com;
2256                 lfsck_async_interpret_common(env, req, laia, result);
2257         }
2258         spin_unlock(&lfsck->li_lock);
2259 }
2260
2261 static int lfsck_stop_notify(const struct lu_env *env,
2262                              struct lfsck_instance *lfsck,
2263                              struct lfsck_tgt_descs *ltds,
2264                              struct lfsck_tgt_desc *ltd, __u16 type)
2265 {
2266         struct lfsck_component *com;
2267         int                     rc = 0;
2268         ENTRY;
2269
2270         LASSERT(lfsck->li_master);
2271
2272         spin_lock(&lfsck->li_lock);
2273         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_scan);
2274         if (com == NULL)
2275                 com = __lfsck_component_find(lfsck, type,
2276                                              &lfsck->li_list_double_scan);
2277         if (com != NULL)
2278                 lfsck_component_get(com);
2279         spin_unlock(&lfsck->li_lock);
2280
2281         if (com != NULL) {
2282                 struct lfsck_thread_info          *info  = lfsck_env_info(env);
2283                 struct lfsck_async_interpret_args *laia  = &info->lti_laia;
2284                 struct lfsck_request              *lr    = &info->lti_lr;
2285                 struct lfsck_assistant_data       *lad   = com->lc_data;
2286                 struct list_head                  *list;
2287                 struct list_head                  *phase_list;
2288                 struct ptlrpc_request_set         *set;
2289
2290                 set = ptlrpc_prep_set();
2291                 if (set == NULL) {
2292                         lfsck_component_put(env, com);
2293
2294                         RETURN(-ENOMEM);
2295                 }
2296
2297                 if (type == LFSCK_TYPE_LAYOUT) {
2298                         list = &ltd->ltd_layout_list;
2299                         phase_list = &ltd->ltd_layout_phase_list;
2300                 } else {
2301                         list = &ltd->ltd_namespace_list;
2302                         phase_list = &ltd->ltd_namespace_phase_list;
2303                 }
2304
2305                 spin_lock(&ltds->ltd_lock);
2306                 if (list_empty(list)) {
2307                         LASSERT(list_empty(phase_list));
2308                         spin_unlock(&ltds->ltd_lock);
2309                         ptlrpc_set_destroy(set);
2310
2311                         RETURN(0);
2312                 }
2313
2314                 list_del_init(phase_list);
2315                 list_del_init(list);
2316                 spin_unlock(&ltds->ltd_lock);
2317
2318                 memset(lr, 0, sizeof(*lr));
2319                 lr->lr_index = lfsck_dev_idx(lfsck);
2320                 lr->lr_event = LE_PEER_EXIT;
2321                 lr->lr_active = type;
2322                 lr->lr_status = LS_CO_PAUSED;
2323                 if (ltds == &lfsck->li_ost_descs)
2324                         lr->lr_flags = LEF_TO_OST;
2325
2326                 memset(laia, 0, sizeof(*laia));
2327                 laia->laia_com = com;
2328                 laia->laia_ltds = ltds;
2329                 atomic_inc(&ltd->ltd_ref);
2330                 laia->laia_ltd = ltd;
2331                 laia->laia_lr = lr;
2332
2333                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2334                                          lfsck_async_interpret_common,
2335                                          laia, LFSCK_NOTIFY);
2336                 if (rc != 0) {
2337                         CDEBUG(D_LFSCK, "%s: fail to notify %s %x for "
2338                                "co-stop for %s: rc = %d\n",
2339                                lfsck_lfsck2name(lfsck),
2340                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2341                                ltd->ltd_index, lad->lad_name, rc);
2342                         lfsck_tgt_put(ltd);
2343                 } else {
2344                         rc = ptlrpc_set_wait(set);
2345                 }
2346
2347                 ptlrpc_set_destroy(set);
2348                 lfsck_component_put(env, com);
2349         }
2350
2351         RETURN(rc);
2352 }
2353
2354 static int lfsck_async_interpret(const struct lu_env *env,
2355                                  struct ptlrpc_request *req,
2356                                  void *args, int rc)
2357 {
2358         struct lfsck_async_interpret_args *laia = args;
2359         struct lfsck_instance             *lfsck;
2360
2361         lfsck = container_of0(laia->laia_ltds, struct lfsck_instance,
2362                               li_mdt_descs);
2363         lfsck_interpret(env, lfsck, req, laia, rc);
2364         lfsck_tgt_put(laia->laia_ltd);
2365         if (rc != 0 && laia->laia_result != -EALREADY)
2366                 laia->laia_result = rc;
2367
2368         return 0;
2369 }
2370
2371 int lfsck_async_request(const struct lu_env *env, struct obd_export *exp,
2372                         struct lfsck_request *lr,
2373                         struct ptlrpc_request_set *set,
2374                         ptlrpc_interpterer_t interpreter,
2375                         void *args, int request)
2376 {
2377         struct lfsck_async_interpret_args *laia;
2378         struct ptlrpc_request             *req;
2379         struct lfsck_request              *tmp;
2380         struct req_format                 *format;
2381         int                                rc;
2382
2383         switch (request) {
2384         case LFSCK_NOTIFY:
2385                 format = &RQF_LFSCK_NOTIFY;
2386                 break;
2387         case LFSCK_QUERY:
2388                 format = &RQF_LFSCK_QUERY;
2389                 break;
2390         default:
2391                 CDEBUG(D_LFSCK, "%s: unknown async request %d: rc = %d\n",
2392                        exp->exp_obd->obd_name, request, -EINVAL);
2393                 return -EINVAL;
2394         }
2395
2396         req = ptlrpc_request_alloc(class_exp2cliimp(exp), format);
2397         if (req == NULL)
2398                 return -ENOMEM;
2399
2400         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, request);
2401         if (rc != 0) {
2402                 ptlrpc_request_free(req);
2403
2404                 return rc;
2405         }
2406
2407         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2408         *tmp = *lr;
2409         ptlrpc_request_set_replen(req);
2410
2411         laia = ptlrpc_req_async_args(req);
2412         *laia = *(struct lfsck_async_interpret_args *)args;
2413         if (laia->laia_com != NULL)
2414                 lfsck_component_get(laia->laia_com);
2415         req->rq_interpret_reply = interpreter;
2416         req->rq_allow_intr = 1;
2417         ptlrpc_set_add_req(set, req);
2418
2419         return 0;
2420 }
2421
2422 int lfsck_query_all(const struct lu_env *env, struct lfsck_component *com)
2423 {
2424         struct lfsck_thread_info          *info  = lfsck_env_info(env);
2425         struct lfsck_request              *lr    = &info->lti_lr;
2426         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
2427         struct lfsck_instance             *lfsck = com->lc_lfsck;
2428         struct lfsck_tgt_descs            *ltds  = &lfsck->li_mdt_descs;
2429         struct lfsck_tgt_desc             *ltd;
2430         struct ptlrpc_request_set         *set;
2431         int                                idx;
2432         int                                rc;
2433         ENTRY;
2434
2435         memset(lr, 0, sizeof(*lr));
2436         lr->lr_event = LE_QUERY;
2437         lr->lr_active = com->lc_type;
2438         lr->lr_flags = LEF_QUERY_ALL;
2439
2440         memset(laia, 0, sizeof(*laia));
2441         laia->laia_com = com;
2442         laia->laia_lr = lr;
2443
2444         set = ptlrpc_prep_set();
2445         if (set == NULL)
2446                 RETURN(-ENOMEM);
2447
2448 again:
2449         laia->laia_ltds = ltds;
2450         down_read(&ltds->ltd_rw_sem);
2451         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
2452                 ltd = lfsck_tgt_get(ltds, idx);
2453                 LASSERT(ltd != NULL);
2454
2455                 laia->laia_ltd = ltd;
2456                 up_read(&ltds->ltd_rw_sem);
2457                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2458                                          lfsck_async_interpret_common,
2459                                          laia, LFSCK_QUERY);
2460                 if (rc != 0) {
2461                         struct lfsck_assistant_data *lad = com->lc_data;
2462
2463                         CDEBUG(D_LFSCK, "%s: Fail to query %s %x for stat %s: "
2464                                "rc = %d\n", lfsck_lfsck2name(lfsck),
2465                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2466                                ltd->ltd_index, lad->lad_name, rc);
2467                         lfsck_reset_ltd_status(ltd, com->lc_type);
2468                         lfsck_tgt_put(ltd);
2469                 }
2470                 down_read(&ltds->ltd_rw_sem);
2471         }
2472         up_read(&ltds->ltd_rw_sem);
2473
2474         if (com->lc_type == LFSCK_TYPE_LAYOUT && !(lr->lr_flags & LEF_TO_OST)) {
2475                 ltds = &lfsck->li_ost_descs;
2476                 lr->lr_flags |= LEF_TO_OST;
2477                 goto again;
2478         }
2479
2480         rc = ptlrpc_set_wait(set);
2481         ptlrpc_set_destroy(set);
2482
2483         RETURN(rc);
2484 }
2485
2486 int lfsck_start_assistant(const struct lu_env *env, struct lfsck_component *com,
2487                           struct lfsck_start_param *lsp)
2488 {
2489         struct lfsck_instance           *lfsck   = com->lc_lfsck;
2490         struct lfsck_assistant_data     *lad     = com->lc_data;
2491         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
2492         struct ptlrpc_thread            *athread = &lad->lad_thread;
2493         struct lfsck_thread_args        *lta;
2494         struct task_struct              *task;
2495         int                              rc;
2496         ENTRY;
2497
2498         lad->lad_assistant_status = 0;
2499         lad->lad_post_result = 0;
2500         lad->lad_to_post = 0;
2501         lad->lad_to_double_scan = 0;
2502         lad->lad_in_double_scan = 0;
2503         lad->lad_exit = 0;
2504         lad->lad_advance_lock = false;
2505         thread_set_flags(athread, 0);
2506
2507         lta = lfsck_thread_args_init(lfsck, com, lsp);
2508         if (IS_ERR(lta))
2509                 RETURN(PTR_ERR(lta));
2510
2511         task = kthread_run(lfsck_assistant_engine, lta, lad->lad_name);
2512         if (IS_ERR(task)) {
2513                 rc = PTR_ERR(task);
2514                 CERROR("%s: cannot start LFSCK assistant thread for %s: "
2515                        "rc = %d\n", lfsck_lfsck2name(lfsck), lad->lad_name, rc);
2516                 lfsck_thread_args_fini(lta);
2517         } else {
2518                 struct l_wait_info lwi = { 0 };
2519
2520                 l_wait_event(mthread->t_ctl_waitq,
2521                              thread_is_running(athread) ||
2522                              thread_is_stopped(athread),
2523                              &lwi);
2524                 if (unlikely(!thread_is_running(athread)))
2525                         rc = lad->lad_assistant_status;
2526                 else
2527                         rc = 0;
2528         }
2529
2530         RETURN(rc);
2531 }
2532
2533 int lfsck_checkpoint_generic(const struct lu_env *env,
2534                              struct lfsck_component *com)
2535 {
2536         struct lfsck_assistant_data     *lad     = com->lc_data;
2537         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2538         struct ptlrpc_thread            *athread = &lad->lad_thread;
2539         struct l_wait_info               lwi     = { 0 };
2540
2541         l_wait_event(mthread->t_ctl_waitq,
2542                      list_empty(&lad->lad_req_list) ||
2543                      !thread_is_running(mthread) ||
2544                      thread_is_stopped(athread),
2545                      &lwi);
2546
2547         if (!thread_is_running(mthread) || thread_is_stopped(athread))
2548                 return LFSCK_CHECKPOINT_SKIP;
2549
2550         return 0;
2551 }
2552
2553 void lfsck_post_generic(const struct lu_env *env,
2554                         struct lfsck_component *com, int *result)
2555 {
2556         struct lfsck_assistant_data     *lad     = com->lc_data;
2557         struct ptlrpc_thread            *athread = &lad->lad_thread;
2558         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2559         struct l_wait_info               lwi     = { 0 };
2560
2561         lad->lad_post_result = *result;
2562         if (*result <= 0)
2563                 lad->lad_exit = 1;
2564         lad->lad_to_post = 1;
2565
2566         CDEBUG(D_LFSCK, "%s: waiting for assistant to do %s post, rc = %d\n",
2567                lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, *result);
2568
2569         wake_up_all(&athread->t_ctl_waitq);
2570         l_wait_event(mthread->t_ctl_waitq,
2571                      (*result > 0 && list_empty(&lad->lad_req_list)) ||
2572                      thread_is_stopped(athread),
2573                      &lwi);
2574
2575         if (lad->lad_assistant_status < 0)
2576                 *result = lad->lad_assistant_status;
2577
2578         CDEBUG(D_LFSCK, "%s: the assistant has done %s post, rc = %d\n",
2579                lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, *result);
2580 }
2581
2582 int lfsck_double_scan_generic(const struct lu_env *env,
2583                               struct lfsck_component *com, int status)
2584 {
2585         struct lfsck_assistant_data     *lad     = com->lc_data;
2586         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2587         struct ptlrpc_thread            *athread = &lad->lad_thread;
2588         struct l_wait_info               lwi     = { 0 };
2589
2590         if (status != LS_SCANNING_PHASE2)
2591                 lad->lad_exit = 1;
2592         else
2593                 lad->lad_to_double_scan = 1;
2594
2595         CDEBUG(D_LFSCK, "%s: waiting for assistant to do %s double_scan, "
2596                "status %d\n",
2597                lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, status);
2598
2599         wake_up_all(&athread->t_ctl_waitq);
2600         l_wait_event(mthread->t_ctl_waitq,
2601                      lad->lad_in_double_scan ||
2602                      thread_is_stopped(athread),
2603                      &lwi);
2604
2605         CDEBUG(D_LFSCK, "%s: the assistant has done %s double_scan, "
2606                "status %d\n", lfsck_lfsck2name(com->lc_lfsck), lad->lad_name,
2607                lad->lad_assistant_status);
2608
2609         if (lad->lad_assistant_status < 0)
2610                 return lad->lad_assistant_status;
2611
2612         return 0;
2613 }
2614
2615 void lfsck_quit_generic(const struct lu_env *env,
2616                         struct lfsck_component *com)
2617 {
2618         struct lfsck_assistant_data     *lad     = com->lc_data;
2619         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2620         struct ptlrpc_thread            *athread = &lad->lad_thread;
2621         struct l_wait_info               lwi     = { 0 };
2622
2623         lad->lad_exit = 1;
2624         wake_up_all(&athread->t_ctl_waitq);
2625         l_wait_event(mthread->t_ctl_waitq,
2626                      thread_is_init(athread) ||
2627                      thread_is_stopped(athread),
2628                      &lwi);
2629 }
2630
2631 int lfsck_load_one_trace_file(const struct lu_env *env,
2632                               struct lfsck_component *com,
2633                               struct dt_object *parent,
2634                               struct dt_object **child,
2635                               const struct dt_index_features *ft,
2636                               const char *name, bool reset)
2637 {
2638         struct lfsck_instance *lfsck = com->lc_lfsck;
2639         struct dt_object *obj;
2640         int rc;
2641         ENTRY;
2642
2643         if (*child != NULL) {
2644                 struct dt_it *it;
2645                 const struct dt_it_ops *iops;
2646                 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid3;
2647
2648                 if (!reset)
2649                         RETURN(0);
2650
2651                 obj = *child;
2652                 rc = obj->do_ops->do_index_try(env, obj, ft);
2653                 if (rc)
2654                         /* unlink by force */
2655                         goto unlink;
2656
2657                 iops = &obj->do_index_ops->dio_it;
2658                 it = iops->init(env, obj, 0);
2659                 if (IS_ERR(it))
2660                         /* unlink by force */
2661                         goto unlink;
2662
2663                 fid_zero(fid);
2664                 rc = iops->get(env, it, (const struct dt_key *)fid);
2665                 if (rc >= 0) {
2666                         rc = iops->next(env, it);
2667                         iops->put(env, it);
2668                 }
2669                 iops->fini(env, it);
2670                 if (rc > 0)
2671                         /* "rc > 0" means the index file is empty. */
2672                         RETURN(0);
2673
2674 unlink:
2675                 /* The old index is not empty, remove it firstly. */
2676                 rc = local_object_unlink(env, lfsck->li_bottom, parent, name);
2677                 CDEBUG_LIMIT(rc ? D_ERROR : D_LFSCK,
2678                              "%s: unlink lfsck sub trace file %s: rc = %d\n",
2679                              lfsck_lfsck2name(com->lc_lfsck), name, rc);
2680                 if (rc)
2681                         RETURN(rc);
2682
2683                 if (*child) {
2684                         lfsck_object_put(env, *child);
2685                         *child = NULL;
2686                 }
2687         } else if (reset) {
2688                 goto unlink;
2689         }
2690
2691         obj = local_index_find_or_create(env, lfsck->li_los, parent, name,
2692                                          S_IFREG | S_IRUGO | S_IWUSR, ft);
2693         if (IS_ERR(obj))
2694                 RETURN(PTR_ERR(obj));
2695
2696         rc = obj->do_ops->do_index_try(env, obj, ft);
2697         if (rc)
2698                 lfsck_object_put(env, obj);
2699         else
2700                 *child = obj;
2701
2702         RETURN(rc);
2703 }
2704
2705 int lfsck_load_sub_trace_files(const struct lu_env *env,
2706                                struct lfsck_component *com,
2707                                const struct dt_index_features *ft,
2708                                const char *prefix, bool reset)
2709 {
2710         char *name = lfsck_env_info(env)->lti_key;
2711         struct lfsck_sub_trace_obj *lsto;
2712         int rc;
2713         int i;
2714
2715         for (i = 0, rc = 0, lsto = &com->lc_sub_trace_objs[0];
2716              i < LFSCK_STF_COUNT && rc == 0; i++, lsto++) {
2717                 snprintf(name, NAME_MAX, "%s_%02d", prefix, i);
2718                 rc = lfsck_load_one_trace_file(env, com,
2719                                 com->lc_lfsck->li_lfsck_dir,
2720                                 &lsto->lsto_obj, ft, name, reset);
2721         }
2722
2723         return rc;
2724 }
2725
2726 /* external interfaces */
2727
2728 int lfsck_get_speed(struct seq_file *m, struct dt_device *key)
2729 {
2730         struct lu_env           env;
2731         struct lfsck_instance  *lfsck;
2732         int                     rc;
2733         ENTRY;
2734
2735         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2736         if (rc != 0)
2737                 RETURN(rc);
2738
2739         lfsck = lfsck_instance_find(key, true, false);
2740         if (likely(lfsck != NULL)) {
2741                 seq_printf(m, "%u\n", lfsck->li_bookmark_ram.lb_speed_limit);
2742                 lfsck_instance_put(&env, lfsck);
2743         } else {
2744                 rc = -ENXIO;
2745         }
2746
2747         lu_env_fini(&env);
2748
2749         RETURN(rc);
2750 }
2751 EXPORT_SYMBOL(lfsck_get_speed);
2752
2753 int lfsck_set_speed(struct dt_device *key, __u32 val)
2754 {
2755         struct lu_env           env;
2756         struct lfsck_instance  *lfsck;
2757         int                     rc;
2758         ENTRY;
2759
2760         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2761         if (rc != 0)
2762                 RETURN(rc);
2763
2764         lfsck = lfsck_instance_find(key, true, false);
2765         if (likely(lfsck != NULL)) {
2766                 mutex_lock(&lfsck->li_mutex);
2767                 if (__lfsck_set_speed(lfsck, val))
2768                         rc = lfsck_bookmark_store(&env, lfsck);
2769                 mutex_unlock(&lfsck->li_mutex);
2770                 lfsck_instance_put(&env, lfsck);
2771         } else {
2772                 rc = -ENXIO;
2773         }
2774
2775         lu_env_fini(&env);
2776
2777         RETURN(rc);
2778 }
2779 EXPORT_SYMBOL(lfsck_set_speed);
2780
2781 int lfsck_get_windows(struct seq_file *m, struct dt_device *key)
2782 {
2783         struct lu_env           env;
2784         struct lfsck_instance  *lfsck;
2785         int                     rc;
2786         ENTRY;
2787
2788         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2789         if (rc != 0)
2790                 RETURN(rc);
2791
2792         lfsck = lfsck_instance_find(key, true, false);
2793         if (likely(lfsck != NULL)) {
2794                 seq_printf(m, "%u\n", lfsck->li_bookmark_ram.lb_async_windows);
2795                 lfsck_instance_put(&env, lfsck);
2796         } else {
2797                 rc = -ENXIO;
2798         }
2799
2800         lu_env_fini(&env);
2801
2802         RETURN(rc);
2803 }
2804 EXPORT_SYMBOL(lfsck_get_windows);
2805
2806 int lfsck_set_windows(struct dt_device *key, int val)
2807 {
2808         struct lu_env           env;
2809         struct lfsck_instance  *lfsck;
2810         int                     rc;
2811         ENTRY;
2812
2813         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2814         if (rc != 0)
2815                 RETURN(rc);
2816
2817         lfsck = lfsck_instance_find(key, true, false);
2818         if (likely(lfsck != NULL)) {
2819                 if (val < 1 || val > LFSCK_ASYNC_WIN_MAX) {
2820                         CWARN("%s: invalid async windows size that may "
2821                               "cause memory issues. The valid range is "
2822                               "[1 - %u].\n",
2823                               lfsck_lfsck2name(lfsck), LFSCK_ASYNC_WIN_MAX);
2824                         rc = -EINVAL;
2825                 } else if (lfsck->li_bookmark_ram.lb_async_windows != val) {
2826                         mutex_lock(&lfsck->li_mutex);
2827                         lfsck->li_bookmark_ram.lb_async_windows = val;
2828                         rc = lfsck_bookmark_store(&env, lfsck);
2829                         mutex_unlock(&lfsck->li_mutex);
2830                 }
2831                 lfsck_instance_put(&env, lfsck);
2832         } else {
2833                 rc = -ENXIO;
2834         }
2835
2836         lu_env_fini(&env);
2837
2838         RETURN(rc);
2839 }
2840 EXPORT_SYMBOL(lfsck_set_windows);
2841
2842 int lfsck_dump(struct seq_file *m, struct dt_device *key, enum lfsck_type type)
2843 {
2844         struct lu_env           env;
2845         struct lfsck_instance  *lfsck;
2846         struct lfsck_component *com;
2847         int                     rc;
2848         ENTRY;
2849
2850         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2851         if (rc != 0)
2852                 RETURN(rc);
2853
2854         lfsck = lfsck_instance_find(key, true, false);
2855         if (likely(lfsck != NULL)) {
2856                 com = lfsck_component_find(lfsck, type);
2857                 if (likely(com != NULL)) {
2858                         com->lc_ops->lfsck_dump(&env, com, m);
2859                         lfsck_component_put(&env, com);
2860                 } else {
2861                         rc = -ENOTSUPP;
2862                 }
2863
2864                 lfsck_instance_put(&env, lfsck);
2865         } else {
2866                 rc = -ENXIO;
2867         }
2868
2869         lu_env_fini(&env);
2870
2871         RETURN(rc);
2872 }
2873 EXPORT_SYMBOL(lfsck_dump);
2874
2875 static int lfsck_stop_all(const struct lu_env *env,
2876                           struct lfsck_instance *lfsck,
2877                           struct lfsck_stop *stop)
2878 {
2879         struct lfsck_thread_info          *info   = lfsck_env_info(env);
2880         struct lfsck_request              *lr     = &info->lti_lr;
2881         struct lfsck_async_interpret_args *laia   = &info->lti_laia;
2882         struct ptlrpc_request_set         *set;
2883         struct lfsck_tgt_descs            *ltds   = &lfsck->li_mdt_descs;
2884         struct lfsck_tgt_desc             *ltd;
2885         struct lfsck_bookmark             *bk     = &lfsck->li_bookmark_ram;
2886         __u32                              idx;
2887         int                                rc     = 0;
2888         int                                rc1    = 0;
2889         ENTRY;
2890
2891         LASSERT(stop->ls_flags & LPF_BROADCAST);
2892
2893         set = ptlrpc_prep_set();
2894         if (unlikely(set == NULL))
2895                 RETURN(-ENOMEM);
2896
2897         memset(lr, 0, sizeof(*lr));
2898         lr->lr_event = LE_STOP;
2899         lr->lr_index = lfsck_dev_idx(lfsck);
2900         lr->lr_status = stop->ls_status;
2901         lr->lr_version = bk->lb_version;
2902         lr->lr_active = LFSCK_TYPES_ALL;
2903         lr->lr_param = stop->ls_flags;
2904
2905         memset(laia, 0, sizeof(*laia));
2906         laia->laia_ltds = ltds;
2907         laia->laia_lr = lr;
2908         laia->laia_shared = 1;
2909
2910         down_read(&ltds->ltd_rw_sem);
2911         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
2912                 ltd = lfsck_tgt_get(ltds, idx);
2913                 LASSERT(ltd != NULL);
2914
2915                 laia->laia_ltd = ltd;
2916                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2917                                          lfsck_async_interpret, laia,
2918                                          LFSCK_NOTIFY);
2919                 if (rc != 0) {
2920                         lfsck_interpret(env, lfsck, NULL, laia, rc);
2921                         lfsck_tgt_put(ltd);
2922                         CERROR("%s: cannot notify MDT %x for LFSCK stop: "
2923                                "rc = %d\n", lfsck_lfsck2name(lfsck), idx, rc);
2924                         rc1 = rc;
2925                 }
2926         }
2927         up_read(&ltds->ltd_rw_sem);
2928
2929         rc = ptlrpc_set_wait(set);
2930         ptlrpc_set_destroy(set);
2931
2932         if (rc == 0)
2933                 rc = laia->laia_result;
2934
2935         if (rc == -EALREADY)
2936                 rc = 0;
2937
2938         if (rc != 0)
2939                 CERROR("%s: fail to stop LFSCK on some MDTs: rc = %d\n",
2940                        lfsck_lfsck2name(lfsck), rc);
2941
2942         RETURN(rc != 0 ? rc : rc1);
2943 }
2944
2945 static int lfsck_start_all(const struct lu_env *env,
2946                            struct lfsck_instance *lfsck,
2947                            struct lfsck_start *start)
2948 {
2949         struct lfsck_thread_info          *info   = lfsck_env_info(env);
2950         struct lfsck_request              *lr     = &info->lti_lr;
2951         struct lfsck_async_interpret_args *laia   = &info->lti_laia;
2952         struct ptlrpc_request_set         *set;
2953         struct lfsck_tgt_descs            *ltds   = &lfsck->li_mdt_descs;
2954         struct lfsck_tgt_desc             *ltd;
2955         struct lfsck_bookmark             *bk     = &lfsck->li_bookmark_ram;
2956         __u32                              idx;
2957         int                                rc     = 0;
2958         ENTRY;
2959
2960         LASSERT(start->ls_flags & LPF_BROADCAST);
2961
2962         set = ptlrpc_prep_set();
2963         if (unlikely(set == NULL))
2964                 RETURN(-ENOMEM);
2965
2966         memset(lr, 0, sizeof(*lr));
2967         lr->lr_event = LE_START;
2968         lr->lr_index = lfsck_dev_idx(lfsck);
2969         lr->lr_speed = bk->lb_speed_limit;
2970         lr->lr_version = bk->lb_version;
2971         lr->lr_active = start->ls_active;
2972         lr->lr_param = start->ls_flags;
2973         lr->lr_async_windows = bk->lb_async_windows;
2974         lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN |
2975                        LSV_ASYNC_WINDOWS | LSV_CREATE_OSTOBJ |
2976                        LSV_CREATE_MDTOBJ;
2977
2978         memset(laia, 0, sizeof(*laia));
2979         laia->laia_ltds = ltds;
2980         laia->laia_lr = lr;
2981         laia->laia_shared = 1;
2982
2983         down_read(&ltds->ltd_rw_sem);
2984         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
2985                 ltd = lfsck_tgt_get(ltds, idx);
2986                 LASSERT(ltd != NULL);
2987
2988                 laia->laia_ltd = ltd;
2989                 ltd->ltd_layout_done = 0;
2990                 ltd->ltd_namespace_done = 0;
2991                 ltd->ltd_synced_failures = 0;
2992                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2993                                          lfsck_async_interpret, laia,
2994                                          LFSCK_NOTIFY);
2995                 if (rc != 0) {
2996                         lfsck_interpret(env, lfsck, NULL, laia, rc);
2997                         lfsck_tgt_put(ltd);
2998                         CERROR("%s: cannot notify MDT %x for LFSCK "
2999                                "start, failout: rc = %d\n",
3000                                lfsck_lfsck2name(lfsck), idx, rc);
3001                         break;
3002                 }
3003         }
3004         up_read(&ltds->ltd_rw_sem);
3005
3006         if (rc != 0) {
3007                 ptlrpc_set_destroy(set);
3008
3009                 RETURN(rc);
3010         }
3011
3012         rc = ptlrpc_set_wait(set);
3013         ptlrpc_set_destroy(set);
3014
3015         if (rc == 0)
3016                 rc = laia->laia_result;
3017
3018         if (rc != 0) {
3019                 struct lfsck_stop *stop = &info->lti_stop;
3020
3021                 CERROR("%s: cannot start LFSCK on some MDTs, "
3022                        "stop all: rc = %d\n",
3023                        lfsck_lfsck2name(lfsck), rc);
3024                 if (rc != -EALREADY) {
3025                         stop->ls_status = LS_FAILED;
3026                         stop->ls_flags = LPF_ALL_TGT | LPF_BROADCAST;
3027                         lfsck_stop_all(env, lfsck, stop);
3028                 }
3029         }
3030
3031         RETURN(rc);
3032 }
3033
3034 int lfsck_start(const struct lu_env *env, struct dt_device *key,
3035                 struct lfsck_start_param *lsp)
3036 {
3037         struct lfsck_start              *start  = lsp->lsp_start;
3038         struct lfsck_instance           *lfsck;
3039         struct lfsck_bookmark           *bk;
3040         struct ptlrpc_thread            *thread;
3041         struct lfsck_component          *com;
3042         struct l_wait_info               lwi    = { 0 };
3043         struct lfsck_thread_args        *lta;
3044         struct task_struct              *task;
3045         struct lfsck_tgt_descs          *ltds;
3046         struct lfsck_tgt_desc           *ltd;
3047         __u32                            idx;
3048         int                              rc     = 0;
3049         __u16                            valid  = 0;
3050         __u16                            flags  = 0;
3051         __u16                            type   = 1;
3052         ENTRY;
3053
3054         if (key->dd_rdonly)
3055                 RETURN(-EROFS);
3056
3057         lfsck = lfsck_instance_find(key, true, false);
3058         if (unlikely(lfsck == NULL))
3059                 RETURN(-ENXIO);
3060
3061         /* System is not ready, try again later. */
3062         if (unlikely(lfsck->li_namespace == NULL))
3063                 GOTO(put, rc = -EAGAIN);
3064
3065         /* start == NULL means auto trigger paused LFSCK. */
3066         if ((start == NULL) &&
3067             (list_empty(&lfsck->li_list_scan) ||
3068              OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_AUTO)))
3069                 GOTO(put, rc = 0);
3070
3071         bk = &lfsck->li_bookmark_ram;
3072         thread = &lfsck->li_thread;
3073         mutex_lock(&lfsck->li_mutex);
3074         spin_lock(&lfsck->li_lock);
3075         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
3076                 rc = -EALREADY;
3077                 if (unlikely(start == NULL)) {
3078                         spin_unlock(&lfsck->li_lock);
3079                         GOTO(out, rc);
3080                 }
3081
3082                 while (start->ls_active != 0) {
3083                         if (!(type & start->ls_active)) {
3084                                 type <<= 1;
3085                                 continue;
3086                         }
3087
3088                         com = __lfsck_component_find(lfsck, type,
3089                                                      &lfsck->li_list_scan);
3090                         if (com == NULL)
3091                                 com = __lfsck_component_find(lfsck, type,
3092                                                 &lfsck->li_list_double_scan);
3093                         if (com == NULL) {
3094                                 rc = -EOPNOTSUPP;
3095                                 break;
3096                         }
3097
3098                         if (com->lc_ops->lfsck_join != NULL) {
3099                                 rc = com->lc_ops->lfsck_join( env, com, lsp);
3100                                 if (rc != 0 && rc != -EALREADY)
3101                                         break;
3102                         }
3103                         start->ls_active &= ~type;
3104                         type <<= 1;
3105                 }
3106                 spin_unlock(&lfsck->li_lock);
3107                 GOTO(out, rc);
3108         }
3109         spin_unlock(&lfsck->li_lock);
3110
3111         lfsck->li_status = 0;
3112         lfsck->li_oit_over = 0;
3113         lfsck->li_start_unplug = 0;
3114         lfsck->li_drop_dryrun = 0;
3115         lfsck->li_new_scanned = 0;
3116
3117         /* For auto trigger. */
3118         if (start == NULL)
3119                 goto trigger;
3120
3121         if (start->ls_flags & LPF_BROADCAST && !lfsck->li_master) {
3122                 CERROR("%s: only allow to specify '-A | -o' via MDS\n",
3123                        lfsck_lfsck2name(lfsck));
3124
3125                 GOTO(out, rc = -EPERM);
3126         }
3127
3128         start->ls_version = bk->lb_version;
3129
3130         if (start->ls_active != 0) {
3131                 struct lfsck_component *next;
3132
3133                 if (start->ls_active == LFSCK_TYPES_ALL)
3134                         start->ls_active = LFSCK_TYPES_SUPPORTED;
3135
3136                 if (start->ls_active & ~LFSCK_TYPES_SUPPORTED) {
3137                         start->ls_active &= ~LFSCK_TYPES_SUPPORTED;
3138                         GOTO(out, rc = -ENOTSUPP);
3139                 }
3140
3141                 list_for_each_entry_safe(com, next,
3142                                          &lfsck->li_list_scan, lc_link) {
3143                         if (!(com->lc_type & start->ls_active)) {
3144                                 rc = com->lc_ops->lfsck_post(env, com, 0,
3145                                                              false);
3146                                 if (rc != 0)
3147                                         GOTO(out, rc);
3148                         }
3149                 }
3150
3151                 while (start->ls_active != 0) {
3152                         if (type & start->ls_active) {
3153                                 com = __lfsck_component_find(lfsck, type,
3154                                                         &lfsck->li_list_idle);
3155                                 if (com != NULL)
3156                                         /* The component status will be updated
3157                                          * when its prep() is called later by
3158                                          * the LFSCK main engine. */
3159                                         list_move_tail(&com->lc_link,
3160                                                        &lfsck->li_list_scan);
3161                                 start->ls_active &= ~type;
3162                         }
3163                         type <<= 1;
3164                 }
3165         }
3166
3167         if (list_empty(&lfsck->li_list_scan)) {
3168                 /* The speed limit will be used to control both the LFSCK and
3169                  * low layer scrub (if applied), need to be handled firstly. */
3170                 if (start->ls_valid & LSV_SPEED_LIMIT) {
3171                         if (__lfsck_set_speed(lfsck, start->ls_speed_limit)) {
3172                                 rc = lfsck_bookmark_store(env, lfsck);
3173                                 if (rc != 0)
3174                                         GOTO(out, rc);
3175                         }
3176                 }
3177
3178                 goto trigger;
3179         }
3180
3181         if (start->ls_flags & LPF_RESET)
3182                 flags |= DOIF_RESET;
3183
3184         rc = lfsck_set_param(env, lfsck, start, !!(flags & DOIF_RESET));
3185         if (rc != 0)
3186                 GOTO(out, rc);
3187
3188         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
3189                 start->ls_active |= com->lc_type;
3190                 if (flags & DOIF_RESET) {
3191                         rc = com->lc_ops->lfsck_reset(env, com, false);
3192                         if (rc != 0)
3193                                 GOTO(out, rc);
3194                 }
3195         }
3196
3197         ltds = &lfsck->li_mdt_descs;
3198         down_read(&ltds->ltd_rw_sem);
3199         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
3200                 ltd = lfsck_ltd2tgt(ltds, idx);
3201                 LASSERT(ltd != NULL);
3202
3203                 ltd->ltd_layout_done = 0;
3204                 ltd->ltd_namespace_done = 0;
3205                 ltd->ltd_synced_failures = 0;
3206                 lfsck_reset_ltd_status(ltd, LFSCK_TYPE_NAMESPACE);
3207                 lfsck_reset_ltd_status(ltd, LFSCK_TYPE_LAYOUT);
3208                 list_del_init(&ltd->ltd_layout_phase_list);
3209                 list_del_init(&ltd->ltd_layout_list);
3210                 list_del_init(&ltd->ltd_namespace_phase_list);
3211                 list_del_init(&ltd->ltd_namespace_list);
3212         }
3213         up_read(&ltds->ltd_rw_sem);
3214
3215         ltds = &lfsck->li_ost_descs;
3216         down_read(&ltds->ltd_rw_sem);
3217         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
3218                 ltd = lfsck_ltd2tgt(ltds, idx);
3219                 LASSERT(ltd != NULL);
3220
3221                 ltd->ltd_layout_done = 0;
3222                 ltd->ltd_synced_failures = 0;
3223                 lfsck_reset_ltd_status(ltd, LFSCK_TYPE_LAYOUT);
3224                 list_del_init(&ltd->ltd_layout_phase_list);
3225                 list_del_init(&ltd->ltd_layout_list);
3226         }
3227         up_read(&ltds->ltd_rw_sem);
3228
3229 trigger:
3230         lfsck->li_args_dir = LUDA_64BITHASH | LUDA_VERIFY | LUDA_TYPE;
3231         if (bk->lb_param & LPF_DRYRUN)
3232                 lfsck->li_args_dir |= LUDA_VERIFY_DRYRUN;
3233
3234         if (start != NULL && start->ls_valid & LSV_ERROR_HANDLE) {
3235                 valid |= DOIV_ERROR_HANDLE;
3236                 if (start->ls_flags & LPF_FAILOUT)
3237                         flags |= DOIF_FAILOUT;
3238         }
3239
3240         if (start != NULL && start->ls_valid & LSV_DRYRUN) {
3241                 valid |= DOIV_DRYRUN;
3242                 if (start->ls_flags & LPF_DRYRUN)
3243                         flags |= DOIF_DRYRUN;
3244         }
3245
3246         if (!list_empty(&lfsck->li_list_scan))
3247                 flags |= DOIF_OUTUSED;
3248
3249         lfsck->li_args_oit = (flags << DT_OTABLE_IT_FLAGS_SHIFT) | valid;
3250         thread_set_flags(thread, 0);
3251         lta = lfsck_thread_args_init(lfsck, NULL, lsp);
3252         if (IS_ERR(lta))
3253                 GOTO(out, rc = PTR_ERR(lta));
3254
3255         __lfsck_set_speed(lfsck, bk->lb_speed_limit);
3256         task = kthread_run(lfsck_master_engine, lta, "lfsck");
3257         if (IS_ERR(task)) {
3258                 rc = PTR_ERR(task);
3259                 CERROR("%s: cannot start LFSCK thread: rc = %d\n",
3260                        lfsck_lfsck2name(lfsck), rc);
3261                 lfsck_thread_args_fini(lta);
3262
3263                 GOTO(out, rc);
3264         }
3265
3266         l_wait_event(thread->t_ctl_waitq,
3267                      thread_is_running(thread) ||
3268                      thread_is_stopped(thread),
3269                      &lwi);
3270         if (start == NULL || !(start->ls_flags & LPF_BROADCAST)) {
3271                 lfsck->li_start_unplug = 1;
3272                 wake_up_all(&thread->t_ctl_waitq);
3273
3274                 GOTO(out, rc = 0);
3275         }
3276
3277         /* release lfsck::li_mutex to avoid deadlock. */
3278         mutex_unlock(&lfsck->li_mutex);
3279         rc = lfsck_start_all(env, lfsck, start);
3280         if (rc != 0) {
3281                 spin_lock(&lfsck->li_lock);
3282                 if (thread_is_stopped(thread)) {
3283                         spin_unlock(&lfsck->li_lock);
3284                 } else {
3285                         lfsck->li_status = LS_FAILED;
3286                         lfsck->li_flags = 0;
3287                         thread_set_flags(thread, SVC_STOPPING);
3288                         spin_unlock(&lfsck->li_lock);
3289
3290                         lfsck->li_start_unplug = 1;
3291                         wake_up_all(&thread->t_ctl_waitq);
3292                         l_wait_event(thread->t_ctl_waitq,
3293                                      thread_is_stopped(thread),
3294                                      &lwi);
3295                 }
3296         } else {
3297                 lfsck->li_start_unplug = 1;
3298                 wake_up_all(&thread->t_ctl_waitq);
3299         }
3300
3301         GOTO(put, rc);
3302
3303 out:
3304         mutex_unlock(&lfsck->li_mutex);
3305
3306 put:
3307         lfsck_instance_put(env, lfsck);
3308
3309         return rc < 0 ? rc : 0;
3310 }
3311 EXPORT_SYMBOL(lfsck_start);
3312
3313 int lfsck_stop(const struct lu_env *env, struct dt_device *key,
3314                struct lfsck_stop *stop)
3315 {
3316         struct lfsck_instance   *lfsck;
3317         struct ptlrpc_thread    *thread;
3318         struct l_wait_info       lwi    = { 0 };
3319         int                      rc     = 0;
3320         int                      rc1    = 0;
3321         ENTRY;
3322
3323         lfsck = lfsck_instance_find(key, true, false);
3324         if (unlikely(lfsck == NULL))
3325                 RETURN(-ENXIO);
3326
3327         thread = &lfsck->li_thread;
3328         /* release lfsck::li_mutex to avoid deadlock. */
3329         if (stop != NULL && stop->ls_flags & LPF_BROADCAST) {
3330                 if (!lfsck->li_master) {
3331                         CERROR("%s: only allow to specify '-A' via MDS\n",
3332                                lfsck_lfsck2name(lfsck));
3333
3334                         GOTO(out, rc = -EPERM);
3335                 }
3336
3337                 rc1 = lfsck_stop_all(env, lfsck, stop);
3338         }
3339
3340         mutex_lock(&lfsck->li_mutex);
3341         spin_lock(&lfsck->li_lock);
3342         /* no error if LFSCK is already stopped, or was never started */
3343         if (thread_is_init(thread) || thread_is_stopped(thread)) {
3344                 spin_unlock(&lfsck->li_lock);
3345                 GOTO(out, rc = 0);
3346         }
3347
3348         if (stop != NULL) {
3349                 lfsck->li_status = stop->ls_status;
3350                 lfsck->li_flags = stop->ls_flags;
3351         } else {
3352                 lfsck->li_status = LS_STOPPED;
3353                 lfsck->li_flags = 0;
3354         }
3355
3356         thread_set_flags(thread, SVC_STOPPING);
3357
3358         if (lfsck->li_master) {
3359                 struct lfsck_component *com;
3360                 struct lfsck_assistant_data *lad;
3361
3362                 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
3363                         lad = com->lc_data;
3364                         spin_lock(&lad->lad_lock);
3365                         if (lad->lad_task != NULL)
3366                                 force_sig(SIGINT, lad->lad_task);
3367                         spin_unlock(&lad->lad_lock);
3368                 }
3369
3370                 list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
3371                         lad = com->lc_data;
3372                         spin_lock(&lad->lad_lock);
3373                         if (lad->lad_task != NULL)
3374                                 force_sig(SIGINT, lad->lad_task);
3375                         spin_unlock(&lad->lad_lock);
3376                 }
3377         }
3378
3379         spin_unlock(&lfsck->li_lock);
3380
3381         wake_up_all(&thread->t_ctl_waitq);
3382         l_wait_event(thread->t_ctl_waitq,
3383                      thread_is_stopped(thread),
3384                      &lwi);
3385
3386         GOTO(out, rc = 0);
3387
3388 out:
3389         mutex_unlock(&lfsck->li_mutex);
3390         lfsck_instance_put(env, lfsck);
3391
3392         return rc != 0 ? rc : rc1;
3393 }
3394 EXPORT_SYMBOL(lfsck_stop);
3395
3396 int lfsck_in_notify_local(const struct lu_env *env, struct dt_device *key,
3397                           struct lfsck_req_local *lrl, struct thandle *th)
3398 {
3399         struct lfsck_instance *lfsck;
3400         struct lfsck_component *com;
3401         int rc = -EOPNOTSUPP;
3402         ENTRY;
3403
3404         lfsck = lfsck_instance_find(key, true, false);
3405         if (unlikely(!lfsck))
3406                 RETURN(-ENXIO);
3407
3408         com = lfsck_component_find(lfsck, lrl->lrl_active);
3409         if (likely(com && com->lc_ops->lfsck_in_notify_local)) {
3410                 rc = com->lc_ops->lfsck_in_notify_local(env, com, lrl, th);
3411                 lfsck_component_put(env, com);
3412         }
3413
3414         lfsck_instance_put(env, lfsck);
3415
3416         RETURN(rc);
3417 }
3418 EXPORT_SYMBOL(lfsck_in_notify_local);
3419
3420 int lfsck_in_notify(const struct lu_env *env, struct dt_device *key,
3421                     struct lfsck_request *lr)
3422 {
3423         int rc = -EOPNOTSUPP;
3424         ENTRY;
3425
3426         switch (lr->lr_event) {
3427         case LE_START: {
3428                 struct lfsck_start       *start = &lfsck_env_info(env)->lti_start;
3429                 struct lfsck_start_param  lsp;
3430
3431                 memset(start, 0, sizeof(*start));
3432                 start->ls_valid = lr->lr_valid;
3433                 start->ls_speed_limit = lr->lr_speed;
3434                 start->ls_version = lr->lr_version;
3435                 start->ls_active = lr->lr_active;
3436                 start->ls_flags = lr->lr_param & ~LPF_BROADCAST;
3437                 start->ls_async_windows = lr->lr_async_windows;
3438
3439                 lsp.lsp_start = start;
3440                 lsp.lsp_index = lr->lr_index;
3441                 lsp.lsp_index_valid = 1;
3442                 rc = lfsck_start(env, key, &lsp);
3443                 break;
3444         }
3445         case LE_STOP: {
3446                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
3447
3448                 memset(stop, 0, sizeof(*stop));
3449                 stop->ls_status = lr->lr_status;
3450                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
3451                 rc = lfsck_stop(env, key, stop);
3452                 break;
3453         }
3454         case LE_PHASE1_DONE:
3455         case LE_PHASE2_DONE:
3456         case LE_PEER_EXIT:
3457         case LE_CONDITIONAL_DESTROY:
3458         case LE_SET_LMV_MASTER:
3459         case LE_SET_LMV_SLAVE:
3460         case LE_PAIRS_VERIFY: {
3461                 struct lfsck_instance  *lfsck;
3462                 struct lfsck_component *com;
3463
3464                 lfsck = lfsck_instance_find(key, true, false);
3465                 if (unlikely(lfsck == NULL))
3466                         RETURN(-ENXIO);
3467
3468                 com = lfsck_component_find(lfsck, lr->lr_active);
3469                 if (likely(com)) {
3470                         rc = com->lc_ops->lfsck_in_notify(env, com, lr);
3471                         lfsck_component_put(env, com);
3472                 }
3473
3474                 lfsck_instance_put(env, lfsck);
3475                 break;
3476         }
3477         default:
3478                 break;
3479         }
3480
3481         RETURN(rc);
3482 }
3483 EXPORT_SYMBOL(lfsck_in_notify);
3484
3485 int lfsck_query(const struct lu_env *env, struct dt_device *key,
3486                 struct lfsck_request *req, struct lfsck_reply *rep,
3487                 struct lfsck_query *que)
3488 {
3489         struct lfsck_instance  *lfsck;
3490         struct lfsck_component *com;
3491         int                     i;
3492         int                     rc = 0;
3493         __u16                   type;
3494         ENTRY;
3495
3496         lfsck = lfsck_instance_find(key, true, false);
3497         if (unlikely(lfsck == NULL))
3498                 RETURN(-ENXIO);
3499
3500         if (que != NULL) {
3501                 if (que->lu_types == LFSCK_TYPES_ALL)
3502                         que->lu_types =
3503                                 LFSCK_TYPES_SUPPORTED & ~LFSCK_TYPE_SCRUB;
3504
3505                 if (que->lu_types & ~LFSCK_TYPES_SUPPORTED) {
3506                         que->lu_types &= ~LFSCK_TYPES_SUPPORTED;
3507
3508                         GOTO(out, rc = -ENOTSUPP);
3509                 }
3510
3511                 for (i = 0, type = 1 << i; i < LFSCK_TYPE_BITS;
3512                      i++, type = 1 << i) {
3513                         if (!(que->lu_types & type))
3514                                 continue;
3515
3516 again:
3517                         com = lfsck_component_find(lfsck, type);
3518                         if (unlikely(com == NULL))
3519                                 GOTO(out, rc = -ENOTSUPP);
3520
3521                         memset(que->lu_mdts_count[i], 0,
3522                                sizeof(__u32) * (LS_MAX + 1));
3523                         memset(que->lu_osts_count[i], 0,
3524                                sizeof(__u32) * (LS_MAX + 1));
3525                         que->lu_repaired[i] = 0;
3526                         rc = com->lc_ops->lfsck_query(env, com, req, rep,
3527                                                       que, i);
3528                         lfsck_component_put(env, com);
3529                         if  (rc < 0)
3530                                 GOTO(out, rc);
3531                 }
3532
3533                 if (!(que->lu_flags & LPF_WAIT))
3534                         GOTO(out, rc);
3535
3536                 for (i = 0, type = 1 << i; i < LFSCK_TYPE_BITS;
3537                      i++, type = 1 << i) {
3538                         if (!(que->lu_types & type))
3539                                 continue;
3540
3541                         if (que->lu_mdts_count[i][LS_SCANNING_PHASE1] != 0 ||
3542                             que->lu_mdts_count[i][LS_SCANNING_PHASE2] != 0 ||
3543                             que->lu_osts_count[i][LS_SCANNING_PHASE1] != 0 ||
3544                             que->lu_osts_count[i][LS_SCANNING_PHASE2] != 0) {
3545                                 struct l_wait_info lwi;
3546
3547                                 /* If it is required to wait, then sleep
3548                                  * 3 seconds and try to query again. */
3549                                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(3),
3550                                                        NULL,
3551                                                        LWI_ON_SIGNAL_NOOP,
3552                                                        NULL);
3553                                 rc = l_wait_event(lfsck->li_thread.t_ctl_waitq,
3554                                                   0, &lwi);
3555                                 if (rc == -ETIMEDOUT)
3556                                         goto again;
3557                         }
3558                 }
3559         } else {
3560                 com = lfsck_component_find(lfsck, req->lr_active);
3561                 if (likely(com != NULL)) {
3562                         rc = com->lc_ops->lfsck_query(env, com, req, rep,
3563                                                       que, -1);
3564                         lfsck_component_put(env, com);
3565                 } else {
3566                         rc = -ENOTSUPP;
3567                 }
3568         }
3569
3570         GOTO(out, rc);
3571
3572 out:
3573         lfsck_instance_put(env, lfsck);
3574         return rc;
3575 }
3576 EXPORT_SYMBOL(lfsck_query);
3577
3578 int lfsck_register_namespace(const struct lu_env *env, struct dt_device *key,
3579                              struct ldlm_namespace *ns)
3580 {
3581         struct lfsck_instance  *lfsck;
3582         int                     rc      = -ENXIO;
3583
3584         lfsck = lfsck_instance_find(key, true, false);
3585         if (likely(lfsck != NULL)) {
3586                 lfsck->li_namespace = ns;
3587                 lfsck_instance_put(env, lfsck);
3588                 rc = 0;
3589         }
3590
3591         return rc;
3592 }
3593 EXPORT_SYMBOL(lfsck_register_namespace);
3594
3595 int lfsck_register(const struct lu_env *env, struct dt_device *key,
3596                    struct dt_device *next, struct obd_device *obd,
3597                    lfsck_out_notify notify, void *notify_data, bool master)
3598 {
3599         struct lfsck_instance   *lfsck;
3600         struct dt_object        *root  = NULL;
3601         struct dt_object        *obj   = NULL;
3602         struct lu_fid           *fid   = &lfsck_env_info(env)->lti_fid;
3603         int                      rc;
3604         ENTRY;
3605
3606         lfsck = lfsck_instance_find(key, false, false);
3607         if (unlikely(lfsck != NULL))
3608                 RETURN(-EEXIST);
3609
3610         OBD_ALLOC_PTR(lfsck);
3611         if (lfsck == NULL)
3612                 RETURN(-ENOMEM);
3613
3614         mutex_init(&lfsck->li_mutex);
3615         spin_lock_init(&lfsck->li_lock);
3616         INIT_LIST_HEAD(&lfsck->li_link);
3617         INIT_LIST_HEAD(&lfsck->li_list_scan);
3618         INIT_LIST_HEAD(&lfsck->li_list_dir);
3619         INIT_LIST_HEAD(&lfsck->li_list_double_scan);
3620         INIT_LIST_HEAD(&lfsck->li_list_idle);
3621         INIT_LIST_HEAD(&lfsck->li_list_lmv);
3622         atomic_set(&lfsck->li_ref, 1);
3623         atomic_set(&lfsck->li_double_scan_count, 0);
3624         init_waitqueue_head(&lfsck->li_thread.t_ctl_waitq);
3625         lfsck->li_out_notify = notify;
3626         lfsck->li_out_notify_data = notify_data;
3627         lfsck->li_next = next;
3628         lfsck->li_bottom = key;
3629         lfsck->li_obd = obd;
3630
3631         rc = lfsck_tgt_descs_init(&lfsck->li_ost_descs);
3632         if (rc != 0)
3633                 GOTO(out, rc);
3634
3635         rc = lfsck_tgt_descs_init(&lfsck->li_mdt_descs);
3636         if (rc != 0)
3637                 GOTO(out, rc);
3638
3639         fid->f_seq = FID_SEQ_LOCAL_NAME;
3640         fid->f_oid = 1;
3641         fid->f_ver = 0;
3642         rc = local_oid_storage_init(env, key, fid, &lfsck->li_los);
3643         if (rc != 0)
3644                 GOTO(out, rc);
3645
3646         rc = dt_root_get(env, key, fid);
3647         if (rc != 0)
3648                 GOTO(out, rc);
3649
3650         root = dt_locate(env, key, fid);
3651         if (IS_ERR(root))
3652                 GOTO(out, rc = PTR_ERR(root));
3653
3654         if (unlikely(!dt_try_as_dir(env, root)))
3655                 GOTO(out, rc = -ENOTDIR);
3656
3657         lfsck->li_local_root_fid = *fid;
3658         if (master) {
3659                 lfsck->li_master = 1;
3660                 if (lfsck_dev_idx(lfsck) == 0) {
3661                         struct lu_fid *pfid = &lfsck_env_info(env)->lti_fid2;
3662                         const struct lu_name *cname;
3663
3664                         rc = dt_lookup(env, root,
3665                                 (struct dt_rec *)(&lfsck->li_global_root_fid),
3666                                 (const struct dt_key *)"ROOT");
3667                         if (rc != 0)
3668                                 GOTO(out, rc);
3669
3670                         obj = dt_locate(env, key, &lfsck->li_global_root_fid);
3671                         if (IS_ERR(obj))
3672                                 GOTO(out, rc = PTR_ERR(obj));
3673
3674                         if (unlikely(!dt_try_as_dir(env, obj)))
3675                                 GOTO(out, rc = -ENOTDIR);
3676
3677                         rc = dt_lookup(env, obj, (struct dt_rec *)fid,
3678                                 (const struct dt_key *)dotlustre);
3679                         if (rc != 0)
3680                                 GOTO(out, rc);
3681
3682                         lfsck_object_put(env, obj);
3683                         obj = dt_locate(env, key, fid);
3684                         if (IS_ERR(obj))
3685                                 GOTO(out, rc = PTR_ERR(obj));
3686
3687                         cname = lfsck_name_get_const(env, dotlustre,
3688                                                      strlen(dotlustre));
3689                         rc = lfsck_verify_linkea(env, obj, cname,
3690                                                  &lfsck->li_global_root_fid);
3691                         if (rc != 0)
3692                                 GOTO(out, rc);
3693
3694                         if (unlikely(!dt_try_as_dir(env, obj)))
3695                                 GOTO(out, rc = -ENOTDIR);
3696
3697                         *pfid = *fid;
3698                         rc = dt_lookup(env, obj, (struct dt_rec *)fid,
3699                                        (const struct dt_key *)lostfound);
3700                         if (rc != 0)
3701                                 GOTO(out, rc);
3702
3703                         lfsck_object_put(env, obj);
3704                         obj = dt_locate(env, key, fid);
3705                         if (IS_ERR(obj))
3706                                 GOTO(out, rc = PTR_ERR(obj));
3707
3708                         cname = lfsck_name_get_const(env, lostfound,
3709                                                      strlen(lostfound));
3710                         rc = lfsck_verify_linkea(env, obj, cname, pfid);
3711                         if (rc != 0)
3712                                 GOTO(out, rc);
3713
3714                         lfsck_object_put(env, obj);
3715                         obj = NULL;
3716                 }
3717         }
3718
3719         fid->f_seq = FID_SEQ_LOCAL_FILE;
3720         fid->f_oid = OTABLE_IT_OID;
3721         fid->f_ver = 0;
3722         obj = dt_locate(env, key, fid);
3723         if (IS_ERR(obj))
3724                 GOTO(out, rc = PTR_ERR(obj));
3725
3726         rc = obj->do_ops->do_index_try(env, obj, &dt_otable_features);
3727         if (rc != 0)
3728                 GOTO(out, rc);
3729
3730         lfsck->li_obj_oit = obj;
3731         obj = local_file_find_or_create(env, lfsck->li_los, root, LFSCK_DIR,
3732                                         S_IFDIR | S_IRUGO | S_IWUSR);
3733         if (IS_ERR(obj))
3734                 GOTO(out, rc = PTR_ERR(obj));
3735
3736         lu_object_get(&obj->do_lu);
3737         lfsck->li_lfsck_dir = obj;
3738         rc = lfsck_bookmark_setup(env, lfsck);
3739         if (rc != 0)
3740                 GOTO(out, rc);
3741
3742         if (master) {
3743                 rc = lfsck_fid_init(lfsck);
3744                 if (rc < 0)
3745                         GOTO(out, rc);
3746
3747                 rc = lfsck_namespace_setup(env, lfsck);
3748                 if (rc < 0)
3749                         GOTO(out, rc);
3750         }
3751
3752         rc = lfsck_layout_setup(env, lfsck);
3753         if (rc < 0)
3754                 GOTO(out, rc);
3755
3756         /* XXX: more LFSCK components initialization to be added here. */
3757
3758         rc = lfsck_instance_add(lfsck);
3759         if (rc == 0)
3760                 rc = lfsck_add_target_from_orphan(env, lfsck);
3761 out:
3762         if (obj != NULL && !IS_ERR(obj))
3763                 lfsck_object_put(env, obj);
3764         if (root != NULL && !IS_ERR(root))
3765                 lfsck_object_put(env, root);
3766         if (rc != 0)
3767                 lfsck_instance_cleanup(env, lfsck);
3768         return rc;
3769 }
3770 EXPORT_SYMBOL(lfsck_register);
3771
3772 void lfsck_degister(const struct lu_env *env, struct dt_device *key)
3773 {
3774         struct lfsck_instance *lfsck;
3775
3776         lfsck = lfsck_instance_find(key, false, true);
3777         if (lfsck != NULL)
3778                 lfsck_instance_put(env, lfsck);
3779 }
3780 EXPORT_SYMBOL(lfsck_degister);
3781
3782 int lfsck_add_target(const struct lu_env *env, struct dt_device *key,
3783                      struct dt_device *tgt, struct obd_export *exp,
3784                      __u32 index, bool for_ost)
3785 {
3786         struct lfsck_instance   *lfsck;
3787         struct lfsck_tgt_desc   *ltd;
3788         int                      rc;
3789         ENTRY;
3790
3791         OBD_ALLOC_PTR(ltd);
3792         if (ltd == NULL)
3793                 RETURN(-ENOMEM);
3794
3795         ltd->ltd_tgt = tgt;
3796         ltd->ltd_key = key;
3797         ltd->ltd_exp = exp;
3798         INIT_LIST_HEAD(&ltd->ltd_orphan_list);
3799         INIT_LIST_HEAD(&ltd->ltd_layout_list);
3800         INIT_LIST_HEAD(&ltd->ltd_layout_phase_list);
3801         INIT_LIST_HEAD(&ltd->ltd_namespace_list);
3802         INIT_LIST_HEAD(&ltd->ltd_namespace_phase_list);
3803         atomic_set(&ltd->ltd_ref, 1);
3804         ltd->ltd_index = index;
3805
3806         spin_lock(&lfsck_instance_lock);
3807         lfsck = __lfsck_instance_find(key, true, false);
3808         if (lfsck == NULL) {
3809                 if (for_ost)
3810                         list_add_tail(&ltd->ltd_orphan_list,
3811                                       &lfsck_ost_orphan_list);
3812                 else
3813                         list_add_tail(&ltd->ltd_orphan_list,
3814                                       &lfsck_mdt_orphan_list);
3815                 spin_unlock(&lfsck_instance_lock);
3816
3817                 RETURN(0);
3818         }
3819         spin_unlock(&lfsck_instance_lock);
3820
3821         rc = __lfsck_add_target(env, lfsck, ltd, for_ost, false);
3822         if (rc != 0)
3823                 lfsck_tgt_put(ltd);
3824
3825         lfsck_instance_put(env, lfsck);
3826
3827         RETURN(rc);
3828 }
3829 EXPORT_SYMBOL(lfsck_add_target);
3830
3831 void lfsck_del_target(const struct lu_env *env, struct dt_device *key,
3832                       struct dt_device *tgt, __u32 index, bool for_ost)
3833 {
3834         struct lfsck_instance   *lfsck;
3835         struct lfsck_tgt_descs  *ltds;
3836         struct lfsck_tgt_desc   *ltd;
3837         struct list_head        *head;
3838
3839         if (for_ost)
3840                 head = &lfsck_ost_orphan_list;
3841         else
3842                 head = &lfsck_mdt_orphan_list;
3843
3844         spin_lock(&lfsck_instance_lock);
3845         list_for_each_entry(ltd, head, ltd_orphan_list) {
3846                 if (ltd->ltd_tgt == tgt) {
3847                         list_del_init(&ltd->ltd_orphan_list);
3848                         spin_unlock(&lfsck_instance_lock);
3849                         lfsck_tgt_put(ltd);
3850
3851                         return;
3852                 }
3853         }
3854
3855         ltd = NULL;
3856         lfsck = __lfsck_instance_find(key, true, false);
3857         spin_unlock(&lfsck_instance_lock);
3858         if (unlikely(lfsck == NULL))
3859                 return;
3860
3861         if (for_ost)
3862                 ltds = &lfsck->li_ost_descs;
3863         else
3864                 ltds = &lfsck->li_mdt_descs;
3865
3866         down_write(&ltds->ltd_rw_sem);
3867         LASSERT(ltds->ltd_tgts_bitmap != NULL);
3868
3869         if (unlikely(index >= ltds->ltd_tgts_bitmap->size))
3870                 goto unlock;
3871
3872         ltd = lfsck_ltd2tgt(ltds, index);
3873         if (unlikely(ltd == NULL))
3874                 goto unlock;
3875
3876         LASSERT(ltds->ltd_tgtnr > 0);
3877
3878         ltds->ltd_tgtnr--;
3879         cfs_bitmap_clear(ltds->ltd_tgts_bitmap, index);
3880         lfsck_assign_tgt(ltds, NULL, index);
3881
3882 unlock:
3883         if (ltd == NULL) {
3884                 if (for_ost)
3885                         head = &lfsck->li_ost_descs.ltd_orphan;
3886                 else
3887                         head = &lfsck->li_mdt_descs.ltd_orphan;
3888
3889                 list_for_each_entry(ltd, head, ltd_orphan_list) {
3890                         if (ltd->ltd_tgt == tgt) {
3891                                 list_del_init(&ltd->ltd_orphan_list);
3892                                 break;
3893                         }
3894                 }
3895         }
3896
3897         up_write(&ltds->ltd_rw_sem);
3898         if (ltd != NULL) {
3899                 spin_lock(&ltds->ltd_lock);
3900                 ltd->ltd_dead = 1;
3901                 spin_unlock(&ltds->ltd_lock);
3902                 lfsck_stop_notify(env, lfsck, ltds, ltd, LFSCK_TYPE_NAMESPACE);
3903                 lfsck_stop_notify(env, lfsck, ltds, ltd, LFSCK_TYPE_LAYOUT);
3904                 lfsck_tgt_put(ltd);
3905         }
3906
3907         lfsck_instance_put(env, lfsck);
3908 }
3909 EXPORT_SYMBOL(lfsck_del_target);
3910
3911 static int __init lfsck_init(void)
3912 {
3913         int rc;
3914
3915         INIT_LIST_HEAD(&lfsck_instance_list);
3916         INIT_LIST_HEAD(&lfsck_ost_orphan_list);
3917         INIT_LIST_HEAD(&lfsck_mdt_orphan_list);
3918         lfsck_key_init_generic(&lfsck_thread_key, NULL);
3919         rc = lu_context_key_register(&lfsck_thread_key);
3920         if (!rc) {
3921                 tgt_register_lfsck_in_notify_local(lfsck_in_notify_local);
3922                 tgt_register_lfsck_in_notify(lfsck_in_notify);
3923                 tgt_register_lfsck_query(lfsck_query);
3924         }
3925
3926         return rc;
3927 }
3928
3929 static void __exit lfsck_exit(void)
3930 {
3931         struct lfsck_tgt_desc *ltd;
3932         struct lfsck_tgt_desc *next;
3933
3934         LASSERT(list_empty(&lfsck_instance_list));
3935
3936         list_for_each_entry_safe(ltd, next, &lfsck_ost_orphan_list,
3937                                  ltd_orphan_list) {
3938                 list_del_init(&ltd->ltd_orphan_list);
3939                 lfsck_tgt_put(ltd);
3940         }
3941
3942         list_for_each_entry_safe(ltd, next, &lfsck_mdt_orphan_list,
3943                                  ltd_orphan_list) {
3944                 list_del_init(&ltd->ltd_orphan_list);
3945                 lfsck_tgt_put(ltd);
3946         }
3947
3948         lu_context_key_degister(&lfsck_thread_key);
3949 }
3950
3951 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3952 MODULE_DESCRIPTION("Lustre File System Checker");
3953 MODULE_VERSION(LUSTRE_VERSION_STRING);
3954 MODULE_LICENSE("GPL");
3955
3956 module_init(lfsck_init);
3957 module_exit(lfsck_exit);