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, 2017, 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 <linux/delay.h>
  37 #include <lu_object.h>
  38 #include <dt_object.h>
  39 #include <md_object.h>
  40 #include <lustre_fld.h>
  41 #include <lustre_lib.h>
  42 #include <lustre_net.h>
  43 #include <lustre_lfsck.h>
  44
  45 #include "lfsck_internal.h"
  46
  47 #define LFSCK_CHECKPOINT_SKIP   1
  48
  49 /* define lfsck thread key */
  50 LU_KEY_INIT(lfsck, struct lfsck_thread_info);
  51
  52 static void lfsck_key_fini(const struct lu_context *ctx,
  53                            struct lu_context_key *key, void *data)
  54 {
  55         struct lfsck_thread_info *info = data;
  56
  57         lu_buf_free(&info->lti_linkea_buf);
  58         lu_buf_free(&info->lti_linkea_buf2);
  59         lu_buf_free(&info->lti_big_buf);
  60         OBD_FREE_PTR(info);
  61 }
  62
  63 LU_CONTEXT_KEY_DEFINE(lfsck, LCT_MD_THREAD | LCT_DT_THREAD);
  64 LU_KEY_INIT_GENERIC(lfsck);
  65
  66 static LIST_HEAD(lfsck_instance_list);
  67 static LIST_HEAD(lfsck_ost_orphan_list);
  68 static LIST_HEAD(lfsck_mdt_orphan_list);
  69 static DEFINE_SPINLOCK(lfsck_instance_lock);
  70
  71 const char *const lfsck_flags_names[] = {
  72         "scanned-once",
  73         "inconsistent",
  74         "upgrade",
  75         "incomplete",
  76         "crashed_lastid",
  77         NULL
  78 };
  79
  80 const char *const lfsck_param_names[] = {
  81         NULL,
  82         "failout",
  83         "dryrun",
  84         "all_targets",
  85         "broadcast",
  86         "orphan",
  87         "create_ostobj",
  88         "create_mdtobj",
  89         NULL,
  90         "delay_create_ostobj",
  91         NULL
  92 };
  93
  94 enum lfsck_verify_lpf_types {
  95         LVLT_BY_BOOKMARK        = 0,
  96         LVLT_BY_NAMEENTRY       = 1,
  97 };
  98
  99 static inline void
 100 lfsck_reset_ltd_status(struct lfsck_tgt_desc *ltd, enum lfsck_type type)
 101 {
 102         if (type == LFSCK_TYPE_LAYOUT) {
 103                 ltd->ltd_layout_status = LS_MAX;
 104                 ltd->ltd_layout_repaired = 0;
 105         } else {
 106                 ltd->ltd_namespace_status = LS_MAX;
 107                 ltd->ltd_namespace_repaired = 0;
 108         }
 109 }
 110
 111 static int lfsck_tgt_descs_init(struct lfsck_tgt_descs *ltds)
 112 {
 113         spin_lock_init(&ltds->ltd_lock);
 114         init_rwsem(&ltds->ltd_rw_sem);
 115         INIT_LIST_HEAD(&ltds->ltd_orphan);
 116         ltds->ltd_tgts_bitmap = CFS_ALLOCATE_BITMAP(BITS_PER_LONG);
 117         if (ltds->ltd_tgts_bitmap == NULL)
 118                 return -ENOMEM;
 119
 120         return 0;
 121 }
 122
 123 static void lfsck_tgt_descs_fini(struct lfsck_tgt_descs *ltds)
 124 {
 125         struct lfsck_tgt_desc   *ltd;
 126         struct lfsck_tgt_desc   *next;
 127         int                      idx;
 128
 129         down_write(&ltds->ltd_rw_sem);
 130
 131         list_for_each_entry_safe(ltd, next, &ltds->ltd_orphan,
 132                                  ltd_orphan_list) {
 133                 list_del_init(&ltd->ltd_orphan_list);
 134                 lfsck_tgt_put(ltd);
 135         }
 136
 137         if (unlikely(ltds->ltd_tgts_bitmap == NULL)) {
 138                 up_write(&ltds->ltd_rw_sem);
 139
 140                 return;
 141         }
 142
 143         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
 144                 ltd = lfsck_ltd2tgt(ltds, idx);
 145                 if (likely(ltd != NULL)) {
 146                         LASSERT(list_empty(&ltd->ltd_layout_list));
 147                         LASSERT(list_empty(&ltd->ltd_layout_phase_list));
 148                         LASSERT(list_empty(&ltd->ltd_namespace_list));
 149                         LASSERT(list_empty(&ltd->ltd_namespace_phase_list));
 150
 151                         ltds->ltd_tgtnr--;
 152                         cfs_bitmap_clear(ltds->ltd_tgts_bitmap, idx);
 153                         lfsck_assign_tgt(ltds, NULL, idx);
 154                         lfsck_tgt_put(ltd);
 155                 }
 156         }
 157
 158         LASSERTF(ltds->ltd_tgtnr == 0, "tgt count unmatched: %d\n",
 159                  ltds->ltd_tgtnr);
 160
 161         for (idx = 0; idx < ARRAY_SIZE(ltds->ltd_tgts_idx); idx++) {
 162                 if (ltds->ltd_tgts_idx[idx] != NULL) {
 163                         OBD_FREE_PTR(ltds->ltd_tgts_idx[idx]);
 164                         ltds->ltd_tgts_idx[idx] = NULL;
 165                 }
 166         }
 167
 168         CFS_FREE_BITMAP(ltds->ltd_tgts_bitmap);
 169         ltds->ltd_tgts_bitmap = NULL;
 170         up_write(&ltds->ltd_rw_sem);
 171 }
 172
 173 static int __lfsck_add_target(const struct lu_env *env,
 174                               struct lfsck_instance *lfsck,
 175                               struct lfsck_tgt_desc *ltd,
 176                               bool for_ost, bool locked)
 177 {
 178         struct lfsck_tgt_descs *ltds;
 179         __u32                   index = ltd->ltd_index;
 180         int                     rc    = 0;
 181         ENTRY;
 182
 183         if (for_ost)
 184                 ltds = &lfsck->li_ost_descs;
 185         else
 186                 ltds = &lfsck->li_mdt_descs;
 187
 188         if (!locked)
 189                 down_write(&ltds->ltd_rw_sem);
 190
 191         LASSERT(ltds->ltd_tgts_bitmap != NULL);
 192
 193         if (index >= ltds->ltd_tgts_bitmap->size) {
 194                 __u32 newsize = max((__u32)ltds->ltd_tgts_bitmap->size,
 195                                     (__u32)BITS_PER_LONG);
 196                 struct cfs_bitmap *old_bitmap = ltds->ltd_tgts_bitmap;
 197                 struct cfs_bitmap *new_bitmap;
 198
 199                 while (newsize < index + 1)
 200                         newsize <<= 1;
 201
 202                 new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
 203                 if (new_bitmap == NULL)
 204                         GOTO(unlock, rc = -ENOMEM);
 205
 206                 if (ltds->ltd_tgtnr > 0)
 207                         cfs_bitmap_copy(new_bitmap, old_bitmap);
 208                 ltds->ltd_tgts_bitmap = new_bitmap;
 209                 CFS_FREE_BITMAP(old_bitmap);
 210         }
 211
 212         if (cfs_bitmap_check(ltds->ltd_tgts_bitmap, index)) {
 213                 CERROR("%s: the device %s (%u) is registered already\n",
 214                        lfsck_lfsck2name(lfsck),
 215                        ltd->ltd_tgt->dd_lu_dev.ld_obd->obd_name, index);
 216                 GOTO(unlock, rc = -EEXIST);
 217         }
 218
 219         if (ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 220                 OBD_ALLOC_PTR(ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK]);
 221                 if (ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK] == NULL)
 222                         GOTO(unlock, rc = -ENOMEM);
 223         }
 224
 225         lfsck_assign_tgt(ltds, ltd, index);
 226         cfs_bitmap_set(ltds->ltd_tgts_bitmap, index);
 227         ltds->ltd_tgtnr++;
 228
 229         GOTO(unlock, rc = 0);
 230
 231 unlock:
 232         if (!locked)
 233                 up_write(&ltds->ltd_rw_sem);
 234
 235         return rc;
 236 }
 237
 238 static int lfsck_add_target_from_orphan(const struct lu_env *env,
 239                                         struct lfsck_instance *lfsck)
 240 {
 241         struct lfsck_tgt_descs  *ltds    = &lfsck->li_ost_descs;
 242         struct lfsck_tgt_desc   *ltd;
 243         struct lfsck_tgt_desc   *next;
 244         struct list_head        *head    = &lfsck_ost_orphan_list;
 245         int                      rc;
 246         bool                     for_ost = true;
 247
 248 again:
 249         spin_lock(&lfsck_instance_lock);
 250         list_for_each_entry_safe(ltd, next, head, ltd_orphan_list) {
 251                 if (ltd->ltd_key == lfsck->li_bottom)
 252                         list_move_tail(&ltd->ltd_orphan_list,
 253                                        &ltds->ltd_orphan);
 254         }
 255         spin_unlock(&lfsck_instance_lock);
 256
 257         down_write(&ltds->ltd_rw_sem);
 258         while (!list_empty(&ltds->ltd_orphan)) {
 259                 ltd = list_entry(ltds->ltd_orphan.next,
 260                                  struct lfsck_tgt_desc,
 261                                  ltd_orphan_list);
 262                 list_del_init(&ltd->ltd_orphan_list);
 263                 rc = __lfsck_add_target(env, lfsck, ltd, for_ost, true);
 264                 /* Do not hold the semaphore for too long time. */
 265                 up_write(&ltds->ltd_rw_sem);
 266                 if (rc != 0)
 267                         return rc;
 268
 269                 down_write(&ltds->ltd_rw_sem);
 270         }
 271         up_write(&ltds->ltd_rw_sem);
 272
 273         if (for_ost) {
 274                 ltds = &lfsck->li_mdt_descs;
 275                 head = &lfsck_mdt_orphan_list;
 276                 for_ost = false;
 277                 goto again;
 278         }
 279
 280         return 0;
 281 }
 282
 283 static inline struct lfsck_component *
 284 __lfsck_component_find(struct lfsck_instance *lfsck, __u16 type,
 285                        struct list_head *list)
 286 {
 287         struct lfsck_component *com;
 288
 289         list_for_each_entry(com, list, lc_link) {
 290                 if (com->lc_type == type)
 291                         return com;
 292         }
 293         return NULL;
 294 }
 295
 296 struct lfsck_component *
 297 lfsck_component_find(struct lfsck_instance *lfsck, __u16 type)
 298 {
 299         struct lfsck_component *com;
 300
 301         spin_lock(&lfsck->li_lock);
 302         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_scan);
 303         if (com != NULL)
 304                 goto unlock;
 305
 306         com = __lfsck_component_find(lfsck, type,
 307                                      &lfsck->li_list_double_scan);
 308         if (com != NULL)
 309                 goto unlock;
 310
 311         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_idle);
 312
 313 unlock:
 314         if (com != NULL)
 315                 lfsck_component_get(com);
 316         spin_unlock(&lfsck->li_lock);
 317         return com;
 318 }
 319
 320 void lfsck_component_cleanup(const struct lu_env *env,
 321                              struct lfsck_component *com)
 322 {
 323         if (!list_empty(&com->lc_link))
 324                 list_del_init(&com->lc_link);
 325         if (!list_empty(&com->lc_link_dir))
 326                 list_del_init(&com->lc_link_dir);
 327
 328         lfsck_component_put(env, com);
 329 }
 330
 331 int lfsck_fid_alloc(const struct lu_env *env, struct lfsck_instance *lfsck,
 332                     struct lu_fid *fid, bool locked)
 333 {
 334         struct lfsck_bookmark   *bk = &lfsck->li_bookmark_ram;
 335         int                      rc = 0;
 336         ENTRY;
 337
 338         if (!locked)
 339                 mutex_lock(&lfsck->li_mutex);
 340
 341         rc = seq_client_alloc_fid(env, lfsck->li_seq, fid);
 342         if (rc >= 0) {
 343                 bk->lb_last_fid = *fid;
 344                 /* We do not care about whether the subsequent sub-operations
 345                  * failed or not. The worst case is that one FID is lost that
 346                  * is not a big issue for the LFSCK since it is relative rare
 347                  * for LFSCK create. */
 348                 rc = lfsck_bookmark_store(env, lfsck);
 349         }
 350
 351         if (!locked)
 352                 mutex_unlock(&lfsck->li_mutex);
 353
 354         RETURN(rc);
 355 }
 356
 357 static int __lfsck_ibits_lock(const struct lu_env *env,
 358                               struct lfsck_instance *lfsck,
 359                               struct dt_object *obj, struct ldlm_res_id *resid,
 360                               struct lustre_handle *lh, __u64 bits,
 361                               enum ldlm_mode mode)
 362 {
 363         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 364         union ldlm_policy_data          *policy = &info->lti_policy;
 365         __u64                            flags  = LDLM_FL_ATOMIC_CB;
 366         int                              rc;
 367
 368         LASSERT(lfsck->li_namespace != NULL);
 369
 370         memset(policy, 0, sizeof(*policy));
 371         policy->l_inodebits.bits = bits;
 372         if (dt_object_remote(obj)) {
 373                 struct ldlm_enqueue_info *einfo = &info->lti_einfo;
 374
 375                 memset(einfo, 0, sizeof(*einfo));
 376                 einfo->ei_type = LDLM_IBITS;
 377                 einfo->ei_mode = mode;
 378                 einfo->ei_cb_bl = ldlm_blocking_ast;
 379                 einfo->ei_cb_cp = ldlm_completion_ast;
 380                 einfo->ei_res_id = resid;
 381
 382                 rc = dt_object_lock(env, obj, lh, einfo, policy);
 383                 /* for regular checks LFSCK doesn't use LDLM locking,
 384                  * so the state isn't coherent. here we just took LDLM
 385                  * lock for coherency and it's time to invalidate
 386                  * previous state */
 387                 if (rc == ELDLM_OK)
 388                         dt_invalidate(env, obj);
 389         } else {
 390                 rc = ldlm_cli_enqueue_local(env, lfsck->li_namespace, resid,
 391                                             LDLM_IBITS, policy, mode,
 392                                             &flags, ldlm_blocking_ast,
 393                                             ldlm_completion_ast, NULL, NULL,
 394                                             0, LVB_T_NONE, NULL, lh);
 395         }
 396
 397         if (rc == ELDLM_OK) {
 398                 rc = 0;
 399         } else {
 400                 memset(lh, 0, sizeof(*lh));
 401                 rc = -EIO;
 402         }
 403
 404         return rc;
 405 }
 406
 407 /**
 408  * Request the specified ibits lock for the given object.
 409  *
 410  * Before the LFSCK modifying on the namespace visible object,
 411  * it needs to acquire related ibits ldlm lock.
 412  *
 413  * \param[in] env       pointer to the thread context
 414  * \param[in] lfsck     pointer to the lfsck instance
 415  * \param[in] obj       pointer to the dt_object to be locked
 416  * \param[out] lh       pointer to the lock handle
 417  * \param[in] bits      the bits for the ldlm lock to be acquired
 418  * \param[in] mode      the mode for the ldlm lock to be acquired
 419  *
 420  * \retval              0 for success
 421  * \retval              negative error number on failure
 422  */
 423 int lfsck_ibits_lock(const struct lu_env *env, struct lfsck_instance *lfsck,
 424                      struct dt_object *obj, struct lustre_handle *lh,
 425                      __u64 bits, enum ldlm_mode mode)
 426 {
 427         struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid;
 428
 429         LASSERT(!lustre_handle_is_used(lh));
 430
 431         fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
 432         return __lfsck_ibits_lock(env, lfsck, obj, resid, lh, bits, mode);
 433 }
 434
 435 /**
 436  * Request the remote LOOKUP lock for the given object.
 437  *
 438  * If \a pobj is remote, the LOOKUP lock of \a obj is on the MDT where
 439  * \a pobj is, acquire LOOKUP lock there.
 440  *
 441  * \param[in] env       pointer to the thread context
 442  * \param[in] lfsck     pointer to the lfsck instance
 443  * \param[in] pobj      pointer to parent dt_object
 444  * \param[in] obj       pointer to the dt_object to be locked
 445  * \param[out] lh       pointer to the lock handle
 446  * \param[in] mode      the mode for the ldlm lock to be acquired
 447  *
 448  * \retval              0 for success
 449  * \retval              negative error number on failure
 450  */
 451 int lfsck_remote_lookup_lock(const struct lu_env *env,
 452                              struct lfsck_instance *lfsck,
 453                              struct dt_object *pobj, struct dt_object *obj,
 454                              struct lustre_handle *lh, enum ldlm_mode mode)
 455 {
 456         struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid;
 457
 458         LASSERT(!lustre_handle_is_used(lh));
 459
 460         fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
 461         return __lfsck_ibits_lock(env, lfsck, pobj, resid, lh,
 462                                   MDS_INODELOCK_LOOKUP, mode);
 463 }
 464
 465 /**
 466  * Release the the specified ibits lock.
 467  *
 468  * If the lock has been acquired before, release it
 469  * and cleanup the handle. Otherwise, do nothing.
 470  *
 471  * \param[in] lh        pointer to the lock handle
 472  * \param[in] mode      the mode for the ldlm lock to be released
 473  */
 474 void lfsck_ibits_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
 475 {
 476         if (lustre_handle_is_used(lh)) {
 477                 ldlm_lock_decref(lh, mode);
 478                 memset(lh, 0, sizeof(*lh));
 479         }
 480 }
 481
 482 /**
 483  * Request compound ibits locks for the given <obj, name> pairs.
 484  *
 485  * Before the LFSCK modifying on the namespace visible object, it needs to
 486  * acquire related ibits ldlm lock. Usually, we can use lfsck_ibits_lock for
 487  * the lock purpose. But the simple lfsck_ibits_lock for directory-based
 488  * modificationis (such as insert name entry to the directory) may be too
 489  * coarse-grained and not efficient.
 490  *
 491  * The lfsck_lock() will request compound ibits locks on the specified
 492  * <obj, name> pairs: the PDO (Parallel Directory Operations) ibits (UPDATE)
 493  * lock on the directory object, and the regular ibits lock on the name hash.
 494  *
 495  * \param[in] env       pointer to the thread context
 496  * \param[in] lfsck     pointer to the lfsck instance
 497  * \param[in] obj       pointer to the dt_object to be locked
 498  * \param[in] name      used for building the PDO lock resource
 499  * \param[out] llh      pointer to the lfsck_lock_handle
 500  * \param[in] bits      the bits for the ldlm lock to be acquired
 501  * \param[in] mode      the mode for the ldlm lock to be acquired
 502  *
 503  * \retval              0 for success
 504  * \retval              negative error number on failure
 505  */
 506 int lfsck_lock(const struct lu_env *env, struct lfsck_instance *lfsck,
 507                struct dt_object *obj, const char *name,
 508                struct lfsck_lock_handle *llh, __u64 bits, enum ldlm_mode mode)
 509 {
 510         struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid;
 511         int                 rc;
 512
 513         LASSERT(S_ISDIR(lfsck_object_type(obj)));
 514         LASSERT(name != NULL);
 515         LASSERT(name[0] != 0);
 516         LASSERT(!lustre_handle_is_used(&llh->llh_pdo_lh));
 517         LASSERT(!lustre_handle_is_used(&llh->llh_reg_lh));
 518
 519         switch (mode) {
 520         case LCK_EX:
 521                 llh->llh_pdo_mode = LCK_EX;
 522                 break;
 523         case LCK_PW:
 524                 llh->llh_pdo_mode = LCK_CW;
 525                 break;
 526         case LCK_PR:
 527                 llh->llh_pdo_mode = LCK_CR;
 528                 break;
 529         default:
 530                 CDEBUG(D_LFSCK, "%s: unexpected PDO lock mode %u on the obj "
 531                        DFID"\n", lfsck_lfsck2name(lfsck), mode,
 532                        PFID(lfsck_dto2fid(obj)));
 533                 LBUG();
 534         }
 535
 536         fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
 537         rc = __lfsck_ibits_lock(env, lfsck, obj, resid, &llh->llh_pdo_lh,
 538                                 MDS_INODELOCK_UPDATE, llh->llh_pdo_mode);
 539         if (rc != 0)
 540                 return rc;
 541
 542         llh->llh_reg_mode = mode;
 543         resid->name[LUSTRE_RES_ID_HSH_OFF] = ll_full_name_hash(NULL, name,
 544                                                                strlen(name));
 545         LASSERT(resid->name[LUSTRE_RES_ID_HSH_OFF] != 0);
 546         rc = __lfsck_ibits_lock(env, lfsck, obj, resid, &llh->llh_reg_lh,
 547                                 bits, llh->llh_reg_mode);
 548         if (rc != 0)
 549                 lfsck_ibits_unlock(&llh->llh_pdo_lh, llh->llh_pdo_mode);
 550
 551         return rc;
 552 }
 553
 554 /**
 555  * Release the the compound ibits locks.
 556  *
 557  * \param[in] llh       pointer to the lfsck_lock_handle to be released
 558  */
 559 void lfsck_unlock(struct lfsck_lock_handle *llh)
 560 {
 561         lfsck_ibits_unlock(&llh->llh_reg_lh, llh->llh_reg_mode);
 562         lfsck_ibits_unlock(&llh->llh_pdo_lh, llh->llh_pdo_mode);
 563 }
 564
 565 int lfsck_find_mdt_idx_by_fid(const struct lu_env *env,
 566                               struct lfsck_instance *lfsck,
 567                               const struct lu_fid *fid)
 568 {
 569         struct seq_server_site  *ss     = lfsck_dev_site(lfsck);
 570         struct lu_seq_range     *range  = &lfsck_env_info(env)->lti_range;
 571         int                      rc;
 572
 573         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
 574                 /* "ROOT" is always on the MDT0. */
 575                 if (lu_fid_eq(fid, &lfsck->li_global_root_fid))
 576                         return 0;
 577
 578                 return lfsck_dev_idx(lfsck);
 579         }
 580
 581         fld_range_set_mdt(range);
 582         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
 583         if (rc == 0)
 584                 rc = range->lsr_index;
 585
 586         return rc;
 587 }
 588
 589 const char dot[] = ".";
 590 const char dotdot[] = "..";
 591 static const char dotlustre[] = ".lustre";
 592 static const char lostfound[] = "lost+found";
 593
 594 /**
 595  * Remove the name entry from the .lustre/lost+found directory.
 596  *
 597  * No need to care about the object referenced by the name entry,
 598  * either the name entry is invalid or redundant, or the referenced
 599  * object has been processed or will be handled by others.
 600  *
 601  * \param[in] env       pointer to the thread context
 602  * \param[in] lfsck     pointer to the lfsck instance
 603  * \param[in] name      the name for the name entry to be removed
 604  *
 605  * \retval              0 for success
 606  * \retval              negative error number on failure
 607  */
 608 static int lfsck_lpf_remove_name_entry(const struct lu_env *env,
 609                                        struct lfsck_instance *lfsck,
 610                                        const char *name)
 611 {
 612         struct dt_object        *parent = lfsck->li_lpf_root_obj;
 613         struct dt_device        *dev    = lfsck_obj2dev(parent);
 614         struct thandle          *th;
 615         struct lfsck_lock_handle *llh   = &lfsck_env_info(env)->lti_llh;
 616         int                      rc;
 617         ENTRY;
 618
 619         rc = lfsck_lock(env, lfsck, parent, name, llh,
 620                         MDS_INODELOCK_UPDATE, LCK_PW);
 621         if (rc != 0)
 622                 RETURN(rc);
 623
 624         th = dt_trans_create(env, dev);
 625         if (IS_ERR(th))
 626                 GOTO(unlock, rc = PTR_ERR(th));
 627
 628         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
 629         if (rc != 0)
 630                 GOTO(stop, rc);
 631
 632         rc = dt_declare_ref_del(env, parent, th);
 633         if (rc != 0)
 634                 GOTO(stop, rc);
 635
 636         rc = dt_trans_start_local(env, dev, th);
 637         if (rc != 0)
 638                 GOTO(stop, rc);
 639
 640         rc = dt_delete(env, parent, (const struct dt_key *)name, th);
 641         if (rc != 0)
 642                 GOTO(stop, rc);
 643
 644         dt_write_lock(env, parent, 0);
 645         rc = dt_ref_del(env, parent, th);
 646         dt_write_unlock(env, parent);
 647
 648         GOTO(stop, rc);
 649
 650 stop:
 651         dt_trans_stop(env, dev, th);
 652
 653 unlock:
 654         lfsck_unlock(llh);
 655
 656         CDEBUG(D_LFSCK, "%s: remove name entry "DFID"/%s: rc = %d\n",
 657                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(parent)), name, rc);
 658
 659         return rc;
 660 }
 661
 662 static int lfsck_create_lpf_local(const struct lu_env *env,
 663                                   struct lfsck_instance *lfsck,
 664                                   struct dt_object *child,
 665                                   struct lu_attr *la,
 666                                   struct dt_object_format *dof,
 667                                   const char *name)
 668 {
 669         struct dt_insert_rec    *rec    = &lfsck_env_info(env)->lti_dt_rec;
 670         struct dt_object        *parent = lfsck->li_lpf_root_obj;
 671         struct dt_device        *dev    = lfsck_obj2dev(child);
 672         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
 673         struct dt_object        *bk_obj = lfsck->li_bookmark_obj;
 674         const struct lu_fid     *cfid   = lfsck_dto2fid(child);
 675         struct thandle          *th     = NULL;
 676         struct linkea_data       ldata  = { NULL };
 677         struct lu_buf            linkea_buf;
 678         const struct lu_name    *cname;
 679         loff_t                   pos    = 0;
 680         int                      len    = sizeof(struct lfsck_bookmark);
 681         int                      rc;
 682         ENTRY;
 683
 684         cname = lfsck_name_get_const(env, name, strlen(name));
 685         rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf2,
 686                               cname, lfsck_dto2fid(parent));
 687         if (rc != 0)
 688                 RETURN(rc);
 689
 690         th = dt_trans_create(env, dev);
 691         if (IS_ERR(th))
 692                 RETURN(PTR_ERR(th));
 693
 694         /* 1a. create child */
 695         rc = dt_declare_create(env, child, la, NULL, dof, th);
 696         if (rc != 0)
 697                 GOTO(stop, rc);
 698
 699         if (!dt_try_as_dir(env, child))
 700                 GOTO(stop, rc = -ENOTDIR);
 701
 702         /* 2a. increase child nlink */
 703         rc = dt_declare_ref_add(env, child, th);
 704         if (rc != 0)
 705                 GOTO(stop, rc);
 706
 707         /* 3a. insert dot into child dir */
 708         rec->rec_type = S_IFDIR;
 709         rec->rec_fid = cfid;
 710         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
 711                                (const struct dt_key *)dot, th);
 712         if (rc != 0)
 713                 GOTO(stop, rc);
 714
 715         /* 4a. insert dotdot into child dir */
 716         rec->rec_fid = &LU_LPF_FID;
 717         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
 718                                (const struct dt_key *)dotdot, th);
 719         if (rc != 0)
 720                 GOTO(stop, rc);
 721
 722         /* 5a. insert linkEA for child */
 723         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
 724                        ldata.ld_leh->leh_len);
 725         rc = dt_declare_xattr_set(env, child, &linkea_buf,
 726                                   XATTR_NAME_LINK, 0, th);
 727         if (rc != 0)
 728                 GOTO(stop, rc);
 729
 730         /* 6a. insert name into parent dir */
 731         rec->rec_type = S_IFDIR;
 732         rec->rec_fid = cfid;
 733         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
 734                                (const struct dt_key *)name, th);
 735         if (rc != 0)
 736                 GOTO(stop, rc);
 737
 738         /* 7a. increase parent nlink */
 739         rc = dt_declare_ref_add(env, parent, th);
 740         if (rc != 0)
 741                 GOTO(stop, rc);
 742
 743         /* 8a. update bookmark */
 744         rc = dt_declare_record_write(env, bk_obj,
 745                                      lfsck_buf_get(env, bk, len), 0, th);
 746         if (rc != 0)
 747                 GOTO(stop, rc);
 748
 749         rc = dt_trans_start_local(env, dev, th);
 750         if (rc != 0)
 751                 GOTO(stop, rc);
 752
 753         dt_write_lock(env, child, 0);
 754         /* 1b. create child */
 755         rc = dt_create(env, child, la, NULL, dof, th);
 756         if (rc != 0)
 757                 GOTO(unlock, rc);
 758
 759         /* 2b. increase child nlink */
 760         rc = dt_ref_add(env, child, th);
 761         if (rc != 0)
 762                 GOTO(unlock, rc);
 763
 764         /* 3b. insert dot into child dir */
 765         rec->rec_fid = cfid;
 766         rc = dt_insert(env, child, (const struct dt_rec *)rec,
 767                        (const struct dt_key *)dot, th);
 768         if (rc != 0)
 769                 GOTO(unlock, rc);
 770
 771         /* 4b. insert dotdot into child dir */
 772         rec->rec_fid = &LU_LPF_FID;
 773         rc = dt_insert(env, child, (const struct dt_rec *)rec,
 774                        (const struct dt_key *)dotdot, th);
 775         if (rc != 0)
 776                 GOTO(unlock, rc);
 777
 778         /* 5b. insert linkEA for child. */
 779         rc = dt_xattr_set(env, child, &linkea_buf,
 780                           XATTR_NAME_LINK, 0, th);
 781         dt_write_unlock(env, child);
 782         if (rc != 0)
 783                 GOTO(stop, rc);
 784
 785         /* 6b. insert name into parent dir */
 786         rec->rec_fid = cfid;
 787         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
 788                        (const struct dt_key *)name, th);
 789         if (rc != 0)
 790                 GOTO(stop, rc);
 791
 792         dt_write_lock(env, parent, 0);
 793         /* 7b. increase parent nlink */
 794         rc = dt_ref_add(env, parent, th);
 795         dt_write_unlock(env, parent);
 796         if (rc != 0)
 797                 GOTO(stop, rc);
 798
 799         bk->lb_lpf_fid = *cfid;
 800         lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk);
 801
 802         /* 8b. update bookmark */
 803         rc = dt_record_write(env, bk_obj,
 804                              lfsck_buf_get(env, bk, len), &pos, th);
 805
 806         GOTO(stop, rc);
 807
 808 unlock:
 809         dt_write_unlock(env, child);
 810
 811 stop:
 812         dt_trans_stop(env, dev, th);
 813
 814         return rc;
 815 }
 816
 817 static int lfsck_create_lpf_remote(const struct lu_env *env,
 818                                    struct lfsck_instance *lfsck,
 819                                    struct dt_object *child,
 820                                    struct lu_attr *la,
 821                                    struct dt_object_format *dof,
 822                                    const char *name)
 823 {
 824         struct dt_insert_rec    *rec    = &lfsck_env_info(env)->lti_dt_rec;
 825         struct dt_object        *parent = lfsck->li_lpf_root_obj;
 826         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
 827         struct dt_object        *bk_obj = lfsck->li_bookmark_obj;
 828         const struct lu_fid     *cfid   = lfsck_dto2fid(child);
 829         struct thandle          *th     = NULL;
 830         struct linkea_data       ldata  = { NULL };
 831         struct lu_buf            linkea_buf;
 832         const struct lu_name    *cname;
 833         struct dt_device        *dev;
 834         loff_t                   pos    = 0;
 835         int                      len    = sizeof(struct lfsck_bookmark);
 836         int                      rc;
 837         ENTRY;
 838
 839         cname = lfsck_name_get_const(env, name, strlen(name));
 840         rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf2,
 841                               cname, lfsck_dto2fid(parent));
 842         if (rc != 0)
 843                 RETURN(rc);
 844
 845         /* Create .lustre/lost+found/MDTxxxx. */
 846
 847         /* XXX: Currently, cross-MDT create operation needs to create the child
 848          *      object firstly, then insert name into the parent directory. For
 849          *      this case, the child object resides on current MDT (local), but
 850          *      the parent ".lustre/lost+found" may be on remote MDT. It is not
 851          *      easy to contain all the sub-modifications orderly within single
 852          *      transaction.
 853          *
 854          *      To avoid more inconsistency, we split the create operation into
 855          *      two transactions:
 856          *
 857          *      1) create the child and update the lfsck_bookmark::lb_lpf_fid
 858          *         locally.
 859          *      2) insert the name "MDTXXXX" in the parent ".lustre/lost+found"
 860          *         remotely.
 861          *
 862          *      If 1) done, but 2) failed, then go ahead, the LFSCK will try to
 863          *      repair such inconsistency when LFSCK run next time. */
 864
 865         /* Transaction I: locally */
 866
 867         dev = lfsck_obj2dev(child);
 868         th = dt_trans_create(env, dev);
 869         if (IS_ERR(th))
 870                 RETURN(PTR_ERR(th));
 871
 872         /* 1a. create child */
 873         rc = dt_declare_create(env, child, la, NULL, dof, th);
 874         if (rc != 0)
 875                 GOTO(stop, rc);
 876
 877         if (!dt_try_as_dir(env, child))
 878                 GOTO(stop, rc = -ENOTDIR);
 879
 880         /* 2a. increase child nlink */
 881         rc = dt_declare_ref_add(env, child, th);
 882         if (rc != 0)
 883                 GOTO(stop, rc);
 884
 885         /* 3a. insert dot into child dir */
 886         rec->rec_type = S_IFDIR;
 887         rec->rec_fid = cfid;
 888         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
 889                                (const struct dt_key *)dot, th);
 890         if (rc != 0)
 891                 GOTO(stop, rc);
 892
 893         /* 4a. insert dotdot into child dir */
 894         rec->rec_fid = &LU_LPF_FID;
 895         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
 896                                (const struct dt_key *)dotdot, th);
 897         if (rc != 0)
 898                 GOTO(stop, rc);
 899
 900         /* 5a. insert linkEA for child */
 901         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
 902                        ldata.ld_leh->leh_len);
 903         rc = dt_declare_xattr_set(env, child, &linkea_buf,
 904                                   XATTR_NAME_LINK, 0, th);
 905         if (rc != 0)
 906                 GOTO(stop, rc);
 907
 908         /* 6a. update bookmark */
 909         rc = dt_declare_record_write(env, bk_obj,
 910                                      lfsck_buf_get(env, bk, len), 0, th);
 911         if (rc != 0)
 912                 GOTO(stop, rc);
 913
 914         rc = dt_trans_start_local(env, dev, th);
 915         if (rc != 0)
 916                 GOTO(stop, rc);
 917
 918         dt_write_lock(env, child, 0);
 919         /* 1b. create child */
 920         rc = dt_create(env, child, la, NULL, dof, th);
 921         if (rc != 0)
 922                 GOTO(unlock, rc);
 923
 924         /* 2b. increase child nlink */
 925         rc = dt_ref_add(env, child, th);
 926         if (rc != 0)
 927                 GOTO(unlock, rc);
 928
 929         /* 3b. insert dot into child dir */
 930         rec->rec_type = S_IFDIR;
 931         rec->rec_fid = cfid;
 932         rc = dt_insert(env, child, (const struct dt_rec *)rec,
 933                        (const struct dt_key *)dot, th);
 934         if (rc != 0)
 935                 GOTO(unlock, rc);
 936
 937         /* 4b. insert dotdot into child dir */
 938         rec->rec_fid = &LU_LPF_FID;
 939         rc = dt_insert(env, child, (const struct dt_rec *)rec,
 940                        (const struct dt_key *)dotdot, th);
 941         if (rc != 0)
 942                 GOTO(unlock, rc);
 943
 944         /* 5b. insert linkEA for child */
 945         rc = dt_xattr_set(env, child, &linkea_buf,
 946                           XATTR_NAME_LINK, 0, th);
 947         if (rc != 0)
 948                 GOTO(unlock, rc);
 949
 950         bk->lb_lpf_fid = *cfid;
 951         lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk);
 952
 953         /* 6b. update bookmark */
 954         rc = dt_record_write(env, bk_obj,
 955                              lfsck_buf_get(env, bk, len), &pos, th);
 956
 957         dt_write_unlock(env, child);
 958         dt_trans_stop(env, dev, th);
 959         if (rc != 0)
 960                 RETURN(rc);
 961
 962         /* Transaction II: remotely */
 963
 964         dev = lfsck_obj2dev(parent);
 965         th = dt_trans_create(env, dev);
 966         if (IS_ERR(th))
 967                 RETURN(PTR_ERR(th));
 968
 969         th->th_sync = 1;
 970         /* 5a. insert name into parent dir */
 971         rec->rec_fid = cfid;
 972         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
 973                                (const struct dt_key *)name, th);
 974         if (rc != 0)
 975                 GOTO(stop, rc);
 976
 977         /* 6a. increase parent nlink */
 978         rc = dt_declare_ref_add(env, parent, th);
 979         if (rc != 0)
 980                 GOTO(stop, rc);
 981
 982         rc = dt_trans_start_local(env, dev, th);
 983         if (rc != 0)
 984                 GOTO(stop, rc);
 985
 986         /* 5b. insert name into parent dir */
 987         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
 988                        (const struct dt_key *)name, th);
 989         if (rc != 0)
 990                 GOTO(stop, rc);
 991
 992         dt_write_lock(env, parent, 0);
 993         /* 6b. increase parent nlink */
 994         rc = dt_ref_add(env, parent, th);
 995         dt_write_unlock(env, parent);
 996
 997         GOTO(stop, rc);
 998
 999 unlock:
1000         dt_write_unlock(env, child);
1001 stop:
1002         dt_trans_stop(env, dev, th);
1003
1004         if (rc != 0 && dev == lfsck_obj2dev(parent))
1005                 CDEBUG(D_LFSCK, "%s: partially created the object "DFID
1006                        "for orphans, but failed to insert the name %s "
1007                        "to the .lustre/lost+found/. Such inconsistency "
1008                        "will be repaired when LFSCK run next time: rc = %d\n",
1009                        lfsck_lfsck2name(lfsck), PFID(cfid), name, rc);
1010
1011         return rc;
1012 }
1013
1014 /**
1015  * Create the MDTxxxx directory under /ROOT/.lustre/lost+found/
1016  *
1017  * The /ROOT/.lustre/lost+found/MDTxxxx/ directory is used for holding
1018  * orphans and other uncertain inconsistent objects found during the
1019  * LFSCK. Such directory will be created by the LFSCK engine on the
1020  * local MDT before the LFSCK scanning.
1021  *
1022  * \param[in] env       pointer to the thread context
1023  * \param[in] lfsck     pointer to the lfsck instance
1024  *
1025  * \retval              0 for success
1026  * \retval              negative error number on failure
1027  */
1028 static int lfsck_create_lpf(const struct lu_env *env,
1029                             struct lfsck_instance *lfsck)
1030 {
1031         struct lfsck_bookmark    *bk    = &lfsck->li_bookmark_ram;
1032         struct lfsck_thread_info *info  = lfsck_env_info(env);
1033         struct lu_fid            *cfid  = &info->lti_fid2;
1034         struct lu_attr           *la    = &info->lti_la;
1035         struct dt_object_format  *dof   = &info->lti_dof;
1036         struct dt_object         *parent = lfsck->li_lpf_root_obj;
1037         struct dt_object         *child = NULL;
1038         struct lfsck_lock_handle *llh   = &info->lti_llh;
1039         char                      name[8];
1040         int                       node  = lfsck_dev_idx(lfsck);
1041         int                       rc    = 0;
1042         ENTRY;
1043
1044         LASSERT(lfsck->li_master);
1045         LASSERT(parent != NULL);
1046         LASSERT(lfsck->li_lpf_obj == NULL);
1047
1048         snprintf(name, 8, "MDT%04x", node);
1049         rc = lfsck_lock(env, lfsck, parent, name, llh,
1050                         MDS_INODELOCK_UPDATE, LCK_PW);
1051         if (rc != 0)
1052                 RETURN(rc);
1053
1054         if (fid_is_zero(&bk->lb_lpf_fid)) {
1055                 /* There is corner case that: in former LFSCK scanning we have
1056                  * created the .lustre/lost+found/MDTxxxx but failed to update
1057                  * the lfsck_bookmark::lb_lpf_fid successfully. So need lookup
1058                  * it from MDT0 firstly. */
1059                 rc = dt_lookup_dir(env, parent, name, cfid);
1060                 if (rc != 0 && rc != -ENOENT)
1061                         GOTO(unlock, rc);
1062
1063                 if (rc == 0) {
1064                         bk->lb_lpf_fid = *cfid;
1065                         rc = lfsck_bookmark_store(env, lfsck);
1066                 } else {
1067                         rc = lfsck_fid_alloc(env, lfsck, cfid, true);
1068                 }
1069                 if (rc != 0)
1070                         GOTO(unlock, rc);
1071         } else {
1072                 *cfid = bk->lb_lpf_fid;
1073         }
1074
1075         child = lfsck_object_find_bottom_new(env, lfsck, cfid);
1076         if (IS_ERR(child))
1077                 GOTO(unlock, rc = PTR_ERR(child));
1078
1079         if (dt_object_exists(child) != 0) {
1080                 if (unlikely(!dt_try_as_dir(env, child)))
1081                         rc = -ENOTDIR;
1082                 else
1083                         lfsck->li_lpf_obj = child;
1084
1085                 GOTO(unlock, rc);
1086         }
1087
1088         memset(la, 0, sizeof(*la));
1089         la->la_atime = la->la_mtime = la->la_ctime = ktime_get_real_seconds();
1090         la->la_mode = S_IFDIR | S_IRWXU;
1091         la->la_valid = LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1092                        LA_UID | LA_GID | LA_TYPE;
1093         memset(dof, 0, sizeof(*dof));
1094         dof->dof_type = dt_mode_to_dft(S_IFDIR);
1095
1096         if (node == 0)
1097                 rc = lfsck_create_lpf_local(env, lfsck, child, la, dof, name);
1098         else
1099                 rc = lfsck_create_lpf_remote(env, lfsck, child, la, dof, name);
1100         if (rc == 0)
1101                 lfsck->li_lpf_obj = child;
1102
1103         GOTO(unlock, rc);
1104
1105 unlock:
1106         lfsck_unlock(llh);
1107         if (rc != 0 && child != NULL && !IS_ERR(child))
1108                 lfsck_object_put(env, child);
1109
1110         return rc;
1111 }
1112
1113 /**
1114  * Scan .lustre/lost+found for bad name entries and remove them.
1115  *
1116  * The valid name entry should be "MDTxxxx", the "xxxx" is the MDT device
1117  * index in the system. Any other formatted name is invalid and should be
1118  * removed.
1119  *
1120  * \param[in] env       pointer to the thread context
1121  * \param[in] lfsck     pointer to the lfsck instance
1122  *
1123  * \retval              0 for success
1124  * \retval              negative error number on failure
1125  */
1126 static int lfsck_scan_lpf_bad_entries(const struct lu_env *env,
1127                                       struct lfsck_instance *lfsck)
1128 {
1129         struct dt_object        *parent = lfsck->li_lpf_root_obj;
1130         struct lu_dirent        *ent    =
1131                         (struct lu_dirent *)lfsck_env_info(env)->lti_key;
1132         const struct dt_it_ops  *iops   = &parent->do_index_ops->dio_it;
1133         struct dt_it            *it;
1134         int                      rc;
1135         ENTRY;
1136
1137         it = iops->init(env, parent, LUDA_64BITHASH);
1138         if (IS_ERR(it))
1139                 RETURN(PTR_ERR(it));
1140
1141         rc = iops->load(env, it, 0);
1142         if (rc == 0)
1143                 rc = iops->next(env, it);
1144         else if (rc > 0)
1145                 rc = 0;
1146
1147         while (rc == 0) {
1148                 int off = 3;
1149
1150                 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
1151                 if (rc != 0)
1152                         break;
1153
1154                 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
1155                 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
1156                         goto next;
1157
1158                 /* name length must be strlen("MDTxxxx") */
1159                 if (ent->lde_namelen != 7)
1160                         goto remove;
1161
1162                 if (memcmp(ent->lde_name, "MDT", off) != 0)
1163                         goto remove;
1164
1165                 while (off < 7 && isxdigit(ent->lde_name[off]))
1166                         off++;
1167
1168                 if (off != 7) {
1169
1170 remove:
1171                         rc = lfsck_lpf_remove_name_entry(env, lfsck,
1172                                                          ent->lde_name);
1173                         if (rc != 0)
1174                                 break;
1175                 }
1176
1177 next:
1178                 rc = iops->next(env, it);
1179         }
1180
1181         iops->put(env, it);
1182         iops->fini(env, it);
1183
1184         RETURN(rc > 0 ? 0 : rc);
1185 }
1186
1187 static int lfsck_update_lpf_entry(const struct lu_env *env,
1188                                   struct lfsck_instance *lfsck,
1189                                   struct dt_object *parent,
1190                                   struct dt_object *child,
1191                                   const char *name,
1192                                   enum lfsck_verify_lpf_types type)
1193 {
1194         int rc;
1195
1196         if (type == LVLT_BY_BOOKMARK) {
1197                 rc = lfsck_update_name_entry(env, lfsck, parent, name,
1198                                              lfsck_dto2fid(child), S_IFDIR);
1199         } else /* if (type == LVLT_BY_NAMEENTRY) */ {
1200                 lfsck->li_bookmark_ram.lb_lpf_fid = *lfsck_dto2fid(child);
1201                 rc = lfsck_bookmark_store(env, lfsck);
1202
1203                 CDEBUG(D_LFSCK, "%s: update LPF fid "DFID
1204                        " in the bookmark file: rc = %d\n",
1205                        lfsck_lfsck2name(lfsck),
1206                        PFID(lfsck_dto2fid(child)), rc);
1207         }
1208
1209         return rc;
1210 }
1211
1212 /**
1213  * Check whether the @child back references the @parent.
1214  *
1215  * Two cases:
1216  * 1) The child's FID is stored in the bookmark file. If the child back
1217  *    references the parent (LU_LPF_FID object) via its ".." entry, then
1218  *    insert the name (MDTxxxx) to the .lustre/lost+found; otherwise, if
1219  *    the child back references another parent2, then:
1220  * 1.1) If the parent2 recognizes the child, then update the bookmark file;
1221  * 1.2) Otherwise, the LFSCK cannot know whether there will be parent3 that
1222  *      references the child. So keep them there. As the LFSCK processing,
1223  *      the parent3 may be found, then when the LFSCK run next time, the
1224  *      inconsistency can be repaired.
1225  *
1226  * 2) The child's FID is stored in the .lustre/lost+found/ sub-directory name
1227  *    entry (MDTxxxx). If the child back references the parent (LU_LPF_FID obj)
1228  *    via its ".." entry, then update the bookmark file, otherwise, if the child
1229  *    back references another parent2, then:
1230  * 2.1) If the parent2 recognizes the child, then remove the sub-directory
1231  *      from .lustre/lost+found/;
1232  * 2.2) Otherwise, if the parent2 does not recognizes the child, trust the
1233  *      sub-directory name entry and update the child;
1234  * 2.3) Otherwise, if we do not know whether the parent2 recognizes the child
1235  *      or not, then keep them there.
1236  *
1237  * \param[in] env       pointer to the thread context
1238  * \param[in] lfsck     pointer to the lfsck instance
1239  * \param[in] child     pointer to the lost+found sub-directory object
1240  * \param[in] name      the name for lost+found sub-directory object
1241  * \param[out] fid      pointer to the buffer to hold the FID of the object
1242  *                      (called it as parent2) that is referenced via the
1243  *                      child's dotdot entry; it also can be the FID that
1244  *                      is referenced by the name entry under the parent2.
1245  * \param[in] type      to indicate where the child's FID is stored in
1246  *
1247  * \retval              positive number for uncertain inconsistency
1248  * \retval              0 for success
1249  * \retval              negative error number on failure
1250  */
1251 static int lfsck_verify_lpf_pairs(const struct lu_env *env,
1252                                   struct lfsck_instance *lfsck,
1253                                   struct dt_object *child, const char *name,
1254                                   struct lu_fid *fid,
1255                                   enum lfsck_verify_lpf_types type)
1256 {
1257         struct dt_object         *parent  = lfsck->li_lpf_root_obj;
1258         struct lfsck_thread_info *info    = lfsck_env_info(env);
1259         char                     *name2   = info->lti_key;
1260         struct lu_fid            *fid2    = &info->lti_fid3;
1261         struct dt_object         *parent2 = NULL;
1262         struct lustre_handle      lh      = { 0 };
1263         int                       rc;
1264         ENTRY;
1265
1266         fid_zero(fid);
1267         rc = dt_lookup_dir(env, child, dotdot, fid);
1268         if (rc != 0)
1269                 GOTO(linkea, rc);
1270
1271         if (!fid_is_sane(fid))
1272                 GOTO(linkea, rc = -EINVAL);
1273
1274         if (lu_fid_eq(fid, &LU_LPF_FID)) {
1275                 const struct lu_name *cname;
1276
1277                 if (lfsck->li_lpf_obj == NULL) {
1278                         lu_object_get(&child->do_lu);
1279                         lfsck->li_lpf_obj = child;
1280                 }
1281
1282                 cname = lfsck_name_get_const(env, name, strlen(name));
1283                 rc = lfsck_verify_linkea(env, child, cname, &LU_LPF_FID);
1284                 if (rc == 0)
1285                         rc = lfsck_update_lpf_entry(env, lfsck, parent, child,
1286                                                     name, type);
1287
1288                 GOTO(out_done, rc);
1289         }
1290
1291         parent2 = lfsck_object_find_bottom(env, lfsck, fid);
1292         if (IS_ERR(parent2))
1293                 GOTO(linkea, parent2);
1294
1295         if (!dt_object_exists(parent2)) {
1296                 lfsck_object_put(env, parent2);
1297
1298                 GOTO(linkea, parent2 = ERR_PTR(-ENOENT));
1299         }
1300
1301         if (!dt_try_as_dir(env, parent2)) {
1302                 lfsck_object_put(env, parent2);
1303
1304                 GOTO(linkea, parent2 = ERR_PTR(-ENOTDIR));
1305         }
1306
1307 linkea:
1308         /* To prevent rename/unlink race */
1309         rc = lfsck_ibits_lock(env, lfsck, child, &lh,
1310                               MDS_INODELOCK_UPDATE, LCK_PR);
1311         if (rc != 0)
1312                 GOTO(out_put, rc);
1313
1314         dt_read_lock(env, child, 0);
1315         rc = lfsck_links_get_first(env, child, name2, fid2);
1316         if (rc != 0) {
1317                 dt_read_unlock(env, child);
1318                 lfsck_ibits_unlock(&lh, LCK_PR);
1319
1320                 GOTO(out_put, rc = 1);
1321         }
1322
1323         /* It is almost impossible that the bookmark file (or the name entry)
1324          * and the linkEA hit the same data corruption. Trust the linkEA. */
1325         if (lu_fid_eq(fid2, &LU_LPF_FID) && strcmp(name, name2) == 0) {
1326                 dt_read_unlock(env, child);
1327                 lfsck_ibits_unlock(&lh, LCK_PR);
1328
1329                 *fid = *fid2;
1330                 if (lfsck->li_lpf_obj == NULL) {
1331                         lu_object_get(&child->do_lu);
1332                         lfsck->li_lpf_obj = child;
1333                 }
1334
1335                 /* Update the child's dotdot entry */
1336                 rc = lfsck_update_name_entry(env, lfsck, child, dotdot,
1337                                              &LU_LPF_FID, S_IFDIR);
1338                 if (rc == 0)
1339                         rc = lfsck_update_lpf_entry(env, lfsck, parent, child,
1340                                                     name, type);
1341
1342                 GOTO(out_put, rc);
1343         }
1344
1345         if (parent2 == NULL || IS_ERR(parent2)) {
1346                 dt_read_unlock(env, child);
1347                 lfsck_ibits_unlock(&lh, LCK_PR);
1348
1349                 GOTO(out_done, rc = 1);
1350         }
1351
1352         rc = dt_lookup_dir(env, parent2, name2, fid);
1353         dt_read_unlock(env, child);
1354         lfsck_ibits_unlock(&lh, LCK_PR);
1355         if (rc != 0 && rc != -ENOENT)
1356                 GOTO(out_put, rc);
1357
1358         if (rc == -ENOENT || !lu_fid_eq(fid, lfsck_dto2fid(child))) {
1359                 if (type == LVLT_BY_BOOKMARK)
1360                         GOTO(out_put, rc = 1);
1361
1362                 /* Trust the name entry, update the child's dotdot entry. */
1363                 rc = lfsck_update_name_entry(env, lfsck, child, dotdot,
1364                                              &LU_LPF_FID, S_IFDIR);
1365
1366                 GOTO(out_put, rc);
1367         }
1368
1369         if (type == LVLT_BY_BOOKMARK) {
1370                 /* Invalid FID record in the bookmark file, reset it. */
1371                 fid_zero(&lfsck->li_bookmark_ram.lb_lpf_fid);
1372                 rc = lfsck_bookmark_store(env, lfsck);
1373
1374                 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1375                        " in the bookmark file: rc = %d\n",
1376                        lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(child)), rc);
1377         } else /* if (type == LVLT_BY_NAMEENTRY) */ {
1378                 /* The name entry is wrong, remove it. */
1379                 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1380         }
1381
1382         GOTO(out_put, rc);
1383
1384 out_put:
1385         if (parent2 != NULL && !IS_ERR(parent2))
1386                 lfsck_object_put(env, parent2);
1387
1388 out_done:
1389         return rc;
1390 }
1391
1392 /**
1393  * Verify the /ROOT/.lustre/lost+found/ directory.
1394  *
1395  * /ROOT/.lustre/lost+found/ is a special directory to hold the objects that
1396  * the LFSCK does not exactly know how to handle, such as orphans. So before
1397  * the LFSCK scanning the system, the consistency of such directory needs to
1398  * be verified firstly to allow the users to use it during the LFSCK.
1399  *
1400  * \param[in] env       pointer to the thread context
1401  * \param[in] lfsck     pointer to the lfsck instance
1402  *
1403  * \retval              positive number for uncertain inconsistency
1404  * \retval              0 for success
1405  * \retval              negative error number on failure
1406  */
1407 int lfsck_verify_lpf(const struct lu_env *env, struct lfsck_instance *lfsck)
1408 {
1409         struct lfsck_thread_info *info   = lfsck_env_info(env);
1410         struct lu_fid            *pfid   = &info->lti_fid;
1411         struct lu_fid            *cfid   = &info->lti_fid2;
1412         struct lfsck_bookmark    *bk     = &lfsck->li_bookmark_ram;
1413         struct dt_object         *parent;
1414         /* child1's FID is in the bookmark file. */
1415         struct dt_object         *child1 = NULL;
1416         /* child2's FID is in the name entry MDTxxxx. */
1417         struct dt_object         *child2 = NULL;
1418         const struct lu_name     *cname;
1419         char                      name[8];
1420         int                       node   = lfsck_dev_idx(lfsck);
1421         int                       rc     = 0;
1422         ENTRY;
1423
1424         LASSERT(lfsck->li_master);
1425
1426         if (lfsck_is_dryrun(lfsck))
1427                 RETURN(0);
1428
1429         if (lfsck->li_lpf_root_obj != NULL)
1430                 RETURN(0);
1431
1432         if (node == 0) {
1433                 parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
1434                                                   &LU_LPF_FID);
1435         } else {
1436                 struct lfsck_tgt_desc *ltd;
1437
1438                 ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, 0);
1439                 if (unlikely(ltd == NULL))
1440                         RETURN(-ENXIO);
1441
1442                 parent = lfsck_object_find_by_dev(env, ltd->ltd_tgt,
1443                                                   &LU_LPF_FID);
1444                 lfsck_tgt_put(ltd);
1445         }
1446
1447         if (IS_ERR(parent))
1448                 RETURN(PTR_ERR(parent));
1449
1450         LASSERT(dt_object_exists(parent));
1451
1452         if (unlikely(!dt_try_as_dir(env, parent))) {
1453                 lfsck_object_put(env, parent);
1454
1455                 GOTO(put, rc = -ENOTDIR);
1456         }
1457
1458         lfsck->li_lpf_root_obj = parent;
1459         if (node == 0) {
1460                 rc = lfsck_scan_lpf_bad_entries(env, lfsck);
1461                 if (rc != 0)
1462                         CDEBUG(D_LFSCK, "%s: scan .lustre/lost+found/ "
1463                                "for bad sub-directories: rc = %d\n",
1464                                lfsck_lfsck2name(lfsck), rc);
1465         }
1466
1467         /* child2 */
1468         snprintf(name, 8, "MDT%04x", node);
1469         rc = dt_lookup_dir(env, parent, name, cfid);
1470         if (rc == -ENOENT) {
1471                 rc = 0;
1472                 goto find_child1;
1473         }
1474
1475         if (rc != 0)
1476                 GOTO(put, rc);
1477
1478         /* Invalid FID in the name entry, remove the name entry. */
1479         if (!fid_is_norm(cfid)) {
1480                 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1481                 if (rc != 0)
1482                         GOTO(put, rc);
1483
1484                 goto find_child1;
1485         }
1486
1487         child2 = lfsck_object_find_bottom(env, lfsck, cfid);
1488         if (IS_ERR(child2))
1489                 GOTO(put, rc = PTR_ERR(child2));
1490
1491         if (unlikely(!dt_object_exists(child2) ||
1492                      dt_object_remote(child2)) ||
1493                      !S_ISDIR(lfsck_object_type(child2))) {
1494                 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1495                 if (rc != 0)
1496                         GOTO(put, rc);
1497
1498                 goto find_child1;
1499         }
1500
1501         if (unlikely(!dt_try_as_dir(env, child2))) {
1502                 lfsck_object_put(env, child2);
1503                 child2 = NULL;
1504                 rc = -ENOTDIR;
1505         }
1506
1507 find_child1:
1508         if (fid_is_zero(&bk->lb_lpf_fid))
1509                 goto check_child2;
1510
1511         if (likely(lu_fid_eq(cfid, &bk->lb_lpf_fid))) {
1512                 if (lfsck->li_lpf_obj == NULL) {
1513                         lu_object_get(&child2->do_lu);
1514                         lfsck->li_lpf_obj = child2;
1515                 }
1516
1517                 cname = lfsck_name_get_const(env, name, strlen(name));
1518                 rc = lfsck_verify_linkea(env, child2, cname, &LU_LPF_FID);
1519
1520                 GOTO(put, rc);
1521         }
1522
1523         if (unlikely(!fid_is_norm(&bk->lb_lpf_fid))) {
1524                 struct lu_fid tfid = bk->lb_lpf_fid;
1525
1526                 /* Invalid FID record in the bookmark file, reset it. */
1527                 fid_zero(&bk->lb_lpf_fid);
1528                 rc = lfsck_bookmark_store(env, lfsck);
1529
1530                 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1531                        " in the bookmark file: rc = %d\n",
1532                        lfsck_lfsck2name(lfsck), PFID(&tfid), rc);
1533
1534                 if (rc != 0)
1535                         GOTO(put, rc);
1536
1537                 goto check_child2;
1538         }
1539
1540         child1 = lfsck_object_find_bottom(env, lfsck, &bk->lb_lpf_fid);
1541         if (IS_ERR(child1)) {
1542                 child1 = NULL;
1543                 goto check_child2;
1544         }
1545
1546         if (unlikely(!dt_object_exists(child1) ||
1547                      dt_object_remote(child1)) ||
1548                      !S_ISDIR(lfsck_object_type(child1))) {
1549                 /* Invalid FID record in the bookmark file, reset it. */
1550                 fid_zero(&bk->lb_lpf_fid);
1551                 rc = lfsck_bookmark_store(env, lfsck);
1552
1553                 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1554                        " in the bookmark file: rc = %d\n",
1555                        lfsck_lfsck2name(lfsck),
1556                        PFID(lfsck_dto2fid(child1)), rc);
1557
1558                 if (rc != 0)
1559                         GOTO(put, rc);
1560
1561                 lfsck_object_put(env, child1);
1562                 child1 = NULL;
1563                 goto check_child2;
1564         }
1565
1566         if (unlikely(!dt_try_as_dir(env, child1))) {
1567                 lfsck_object_put(env, child1);
1568                 child1 = NULL;
1569                 rc = -ENOTDIR;
1570                 goto check_child2;
1571         }
1572
1573         rc = lfsck_verify_lpf_pairs(env, lfsck, child1, name, pfid,
1574                                     LVLT_BY_BOOKMARK);
1575         if (lu_fid_eq(pfid, &LU_LPF_FID))
1576                 GOTO(put, rc);
1577
1578 check_child2:
1579         if (child2 != NULL)
1580                 rc = lfsck_verify_lpf_pairs(env, lfsck, child2, name,
1581                                             pfid, LVLT_BY_NAMEENTRY);
1582
1583         GOTO(put, rc);
1584
1585 put:
1586         if (lfsck->li_lpf_obj != NULL) {
1587                 if (unlikely(!dt_try_as_dir(env, lfsck->li_lpf_obj))) {
1588                         lfsck_object_put(env, lfsck->li_lpf_obj);
1589                         lfsck->li_lpf_obj = NULL;
1590                         rc = -ENOTDIR;
1591                 }
1592         } else if (rc == 0) {
1593                 rc = lfsck_create_lpf(env, lfsck);
1594         }
1595
1596         if (child2 != NULL && !IS_ERR(child2))
1597                 lfsck_object_put(env, child2);
1598         if (child1 != NULL && !IS_ERR(child1))
1599                 lfsck_object_put(env, child1);
1600
1601         return rc;
1602 }
1603
1604 static int lfsck_fid_init(struct lfsck_instance *lfsck)
1605 {
1606         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
1607         struct seq_server_site  *ss     = lfsck_dev_site(lfsck);
1608         char                    *prefix;
1609         int                      rc     = 0;
1610         ENTRY;
1611
1612         if (unlikely(ss == NULL))
1613                 RETURN(-ENXIO);
1614
1615         OBD_ALLOC_PTR(lfsck->li_seq);
1616         if (lfsck->li_seq == NULL)
1617                 RETURN(-ENOMEM);
1618
1619         OBD_ALLOC(prefix, MAX_OBD_NAME + 7);
1620         if (prefix == NULL)
1621                 GOTO(out, rc = -ENOMEM);
1622
1623         snprintf(prefix, MAX_OBD_NAME + 7, "lfsck-%s", lfsck_lfsck2name(lfsck));
1624         seq_client_init(lfsck->li_seq, NULL, LUSTRE_SEQ_METADATA, prefix,
1625                              ss->ss_server_seq);
1626         OBD_FREE(prefix, MAX_OBD_NAME + 7);
1627
1628         if (fid_is_sane(&bk->lb_last_fid))
1629                 lfsck->li_seq->lcs_fid = bk->lb_last_fid;
1630
1631         RETURN(0);
1632
1633 out:
1634         OBD_FREE_PTR(lfsck->li_seq);
1635         lfsck->li_seq = NULL;
1636
1637         return rc;
1638 }
1639
1640 static void lfsck_fid_fini(struct lfsck_instance *lfsck)
1641 {
1642         if (lfsck->li_seq != NULL) {
1643                 seq_client_fini(lfsck->li_seq);
1644                 OBD_FREE_PTR(lfsck->li_seq);
1645                 lfsck->li_seq = NULL;
1646         }
1647 }
1648
1649 void lfsck_instance_cleanup(const struct lu_env *env,
1650                             struct lfsck_instance *lfsck)
1651 {
1652         struct ptlrpc_thread    *thread = &lfsck->li_thread;
1653         struct lfsck_component  *com;
1654         struct lfsck_component  *next;
1655         struct lfsck_lmv_unit   *llu;
1656         struct lfsck_lmv_unit   *llu_next;
1657         struct lfsck_lmv        *llmv;
1658         ENTRY;
1659
1660         LASSERT(list_empty(&lfsck->li_link));
1661         LASSERT(thread_is_init(thread) || thread_is_stopped(thread));
1662
1663         if (lfsck->li_obj_oit != NULL) {
1664                 lfsck_object_put(env, lfsck->li_obj_oit);
1665                 lfsck->li_obj_oit = NULL;
1666         }
1667
1668         LASSERT(lfsck->li_obj_dir == NULL);
1669         LASSERT(lfsck->li_lmv == NULL);
1670
1671         list_for_each_entry_safe(llu, llu_next, &lfsck->li_list_lmv, llu_link) {
1672                 llmv = &llu->llu_lmv;
1673
1674                 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
1675                          "still in using: %u\n",
1676                          atomic_read(&llmv->ll_ref));
1677
1678                 lfsck_lmv_put(env, llmv);
1679         }
1680
1681         list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
1682                 lfsck_component_cleanup(env, com);
1683         }
1684
1685         LASSERT(list_empty(&lfsck->li_list_dir));
1686
1687         list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
1688                                  lc_link) {
1689                 lfsck_component_cleanup(env, com);
1690         }
1691
1692         list_for_each_entry_safe(com, next, &lfsck->li_list_idle, lc_link) {
1693                 lfsck_component_cleanup(env, com);
1694         }
1695
1696         lfsck_tgt_descs_fini(&lfsck->li_ost_descs);
1697         lfsck_tgt_descs_fini(&lfsck->li_mdt_descs);
1698
1699         if (lfsck->li_lfsck_dir != NULL) {
1700                 lfsck_object_put(env, lfsck->li_lfsck_dir);
1701                 lfsck->li_lfsck_dir = NULL;
1702         }
1703
1704         if (lfsck->li_bookmark_obj != NULL) {
1705                 lfsck_object_put(env, lfsck->li_bookmark_obj);
1706                 lfsck->li_bookmark_obj = NULL;
1707         }
1708
1709         if (lfsck->li_lpf_obj != NULL) {
1710                 lfsck_object_put(env, lfsck->li_lpf_obj);
1711                 lfsck->li_lpf_obj = NULL;
1712         }
1713
1714         if (lfsck->li_lpf_root_obj != NULL) {
1715                 lfsck_object_put(env, lfsck->li_lpf_root_obj);
1716                 lfsck->li_lpf_root_obj = NULL;
1717         }
1718
1719         if (lfsck->li_los != NULL) {
1720                 local_oid_storage_fini(env, lfsck->li_los);
1721                 lfsck->li_los = NULL;
1722         }
1723
1724         lfsck_fid_fini(lfsck);
1725
1726         OBD_FREE_PTR(lfsck);
1727 }
1728
1729 static inline struct lfsck_instance *
1730 __lfsck_instance_find(struct dt_device *key, bool ref, bool unlink)
1731 {
1732         struct lfsck_instance *lfsck;
1733
1734         list_for_each_entry(lfsck, &lfsck_instance_list, li_link) {
1735                 if (lfsck->li_bottom == key) {
1736                         if (ref)
1737                                 lfsck_instance_get(lfsck);
1738                         if (unlink)
1739                                 list_del_init(&lfsck->li_link);
1740
1741                         return lfsck;
1742                 }
1743         }
1744
1745         return NULL;
1746 }
1747
1748 struct lfsck_instance *lfsck_instance_find(struct dt_device *key, bool ref,
1749                                            bool unlink)
1750 {
1751         struct lfsck_instance *lfsck;
1752
1753         spin_lock(&lfsck_instance_lock);
1754         lfsck = __lfsck_instance_find(key, ref, unlink);
1755         spin_unlock(&lfsck_instance_lock);
1756
1757         return lfsck;
1758 }
1759
1760 static inline int lfsck_instance_add(struct lfsck_instance *lfsck)
1761 {
1762         struct lfsck_instance *tmp;
1763
1764         spin_lock(&lfsck_instance_lock);
1765         list_for_each_entry(tmp, &lfsck_instance_list, li_link) {
1766                 if (lfsck->li_bottom == tmp->li_bottom) {
1767                         spin_unlock(&lfsck_instance_lock);
1768                         return -EEXIST;
1769                 }
1770         }
1771
1772         list_add_tail(&lfsck->li_link, &lfsck_instance_list);
1773         spin_unlock(&lfsck_instance_lock);
1774         return 0;
1775 }
1776
1777 void lfsck_bits_dump(struct seq_file *m, int bits, const char *const names[],
1778                      const char *prefix)
1779 {
1780         int flag;
1781         int i;
1782         bool newline = (bits != 0 ? false : true);
1783
1784         seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
1785
1786         for (i = 0, flag = 1; bits != 0; i++, flag = BIT(i)) {
1787                 if (flag & bits) {
1788                         bits &= ~flag;
1789                         if (names[i] != NULL) {
1790                                 if (bits == 0)
1791                                         newline = true;
1792
1793                                 seq_printf(m, "%s%c", names[i],
1794                                            newline ? '\n' : ',');
1795                         }
1796                 }
1797         }
1798
1799         if (!newline)
1800                 seq_putc(m, '\n');
1801 }
1802
1803 void lfsck_time_dump(struct seq_file *m, time64_t time, const char *name)
1804 {
1805         if (time == 0) {
1806                 seq_printf(m, "%s_time: N/A\n", name);
1807                 seq_printf(m, "time_since_%s: N/A\n", name);
1808         } else {
1809                 seq_printf(m, "%s_time: %lld\n", name, time);
1810                 seq_printf(m, "time_since_%s: %lld seconds\n",
1811                            name, ktime_get_real_seconds() - time);
1812         }
1813 }
1814
1815 void lfsck_pos_dump(struct seq_file *m, struct lfsck_position *pos,
1816                     const char *prefix)
1817 {
1818         if (fid_is_zero(&pos->lp_dir_parent)) {
1819                 if (pos->lp_oit_cookie == 0) {
1820                         seq_printf(m, "%s: N/A, N/A, N/A\n", prefix);
1821                         return;
1822                 }
1823                 seq_printf(m, "%s: %llu, N/A, N/A\n",
1824                            prefix, pos->lp_oit_cookie);
1825         } else {
1826                 seq_printf(m, "%s: %llu, "DFID", %#llx\n",
1827                            prefix, pos->lp_oit_cookie,
1828                            PFID(&pos->lp_dir_parent), pos->lp_dir_cookie);
1829         }
1830 }
1831
1832 void lfsck_pos_fill(const struct lu_env *env, struct lfsck_instance *lfsck,
1833                     struct lfsck_position *pos, bool init)
1834 {
1835         const struct dt_it_ops *iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
1836
1837         if (unlikely(lfsck->li_di_oit == NULL)) {
1838                 memset(pos, 0, sizeof(*pos));
1839                 return;
1840         }
1841
1842         pos->lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
1843         if (!lfsck->li_current_oit_processed && !init)
1844                 pos->lp_oit_cookie--;
1845
1846         if (unlikely(pos->lp_oit_cookie == 0))
1847                 pos->lp_oit_cookie = 1;
1848
1849         if (lfsck->li_di_dir != NULL) {
1850                 struct dt_object *dto = lfsck->li_obj_dir;
1851
1852                 pos->lp_dir_cookie = dto->do_index_ops->dio_it.store(env,
1853                                                         lfsck->li_di_dir);
1854
1855                 if (pos->lp_dir_cookie >= MDS_DIR_END_OFF) {
1856                         fid_zero(&pos->lp_dir_parent);
1857                         pos->lp_dir_cookie = 0;
1858                 } else {
1859                         pos->lp_dir_parent = *lfsck_dto2fid(dto);
1860                 }
1861         } else {
1862                 fid_zero(&pos->lp_dir_parent);
1863                 pos->lp_dir_cookie = 0;
1864         }
1865 }
1866
1867 bool __lfsck_set_speed(struct lfsck_instance *lfsck, __u32 limit)
1868 {
1869         bool dirty = false;
1870
1871         if (limit != LFSCK_SPEED_NO_LIMIT) {
1872                 if (limit > cfs_time_seconds(1)) {
1873                         lfsck->li_sleep_rate = limit / cfs_time_seconds(1);
1874                         lfsck->li_sleep_jif = 1;
1875                 } else {
1876                         lfsck->li_sleep_rate = 1;
1877                         lfsck->li_sleep_jif = cfs_time_seconds(1) / limit;
1878                 }
1879         } else {
1880                 lfsck->li_sleep_jif = 0;
1881                 lfsck->li_sleep_rate = 0;
1882         }
1883
1884         if (lfsck->li_bookmark_ram.lb_speed_limit != limit) {
1885                 lfsck->li_bookmark_ram.lb_speed_limit = limit;
1886                 dirty = true;
1887         }
1888
1889         return dirty;
1890 }
1891
1892 void lfsck_control_speed(struct lfsck_instance *lfsck)
1893 {
1894         struct ptlrpc_thread *thread = &lfsck->li_thread;
1895
1896         if (lfsck->li_sleep_jif > 0 &&
1897             lfsck->li_new_scanned >= lfsck->li_sleep_rate) {
1898                 wait_event_idle_timeout(thread->t_ctl_waitq,
1899                                         !thread_is_running(thread),
1900                                         lfsck->li_sleep_jif);
1901                 lfsck->li_new_scanned = 0;
1902         }
1903 }
1904
1905 void lfsck_control_speed_by_self(struct lfsck_component *com)
1906 {
1907         struct lfsck_instance   *lfsck  = com->lc_lfsck;
1908         struct ptlrpc_thread    *thread = &lfsck->li_thread;
1909
1910         if (lfsck->li_sleep_jif > 0 &&
1911             com->lc_new_scanned >= lfsck->li_sleep_rate) {
1912                 wait_event_idle_timeout(thread->t_ctl_waitq,
1913                                         !thread_is_running(thread),
1914                                         lfsck->li_sleep_jif);
1915                 com->lc_new_scanned = 0;
1916         }
1917 }
1918
1919 static struct lfsck_thread_args *
1920 lfsck_thread_args_init(struct lfsck_instance *lfsck,
1921                        struct lfsck_component *com,
1922                        struct lfsck_start_param *lsp)
1923 {
1924         struct lfsck_thread_args *lta;
1925         int                       rc;
1926
1927         OBD_ALLOC_PTR(lta);
1928         if (lta == NULL)
1929                 return ERR_PTR(-ENOMEM);
1930
1931         rc = lu_env_init(&lta->lta_env, LCT_MD_THREAD | LCT_DT_THREAD);
1932         if (rc != 0) {
1933                 OBD_FREE_PTR(lta);
1934                 return ERR_PTR(rc);
1935         }
1936
1937         lta->lta_lfsck = lfsck_instance_get(lfsck);
1938         if (com != NULL)
1939                 lta->lta_com = lfsck_component_get(com);
1940
1941         lta->lta_lsp = lsp;
1942
1943         return lta;
1944 }
1945
1946 void lfsck_thread_args_fini(struct lfsck_thread_args *lta)
1947 {
1948         if (lta->lta_com != NULL)
1949                 lfsck_component_put(&lta->lta_env, lta->lta_com);
1950         lfsck_instance_put(&lta->lta_env, lta->lta_lfsck);
1951         lu_env_fini(&lta->lta_env);
1952         OBD_FREE_PTR(lta);
1953 }
1954
1955 struct lfsck_assistant_data *
1956 lfsck_assistant_data_init(const struct lfsck_assistant_operations *lao,
1957                           const char *name)
1958 {
1959         struct lfsck_assistant_data *lad;
1960
1961         OBD_ALLOC_PTR(lad);
1962         if (lad != NULL) {
1963                 lad->lad_bitmap = CFS_ALLOCATE_BITMAP(BITS_PER_LONG);
1964                 if (lad->lad_bitmap == NULL) {
1965                         OBD_FREE_PTR(lad);
1966                         return NULL;
1967                 }
1968
1969                 INIT_LIST_HEAD(&lad->lad_req_list);
1970                 spin_lock_init(&lad->lad_lock);
1971                 INIT_LIST_HEAD(&lad->lad_ost_list);
1972                 INIT_LIST_HEAD(&lad->lad_ost_phase1_list);
1973                 INIT_LIST_HEAD(&lad->lad_ost_phase2_list);
1974                 INIT_LIST_HEAD(&lad->lad_mdt_list);
1975                 INIT_LIST_HEAD(&lad->lad_mdt_phase1_list);
1976                 INIT_LIST_HEAD(&lad->lad_mdt_phase2_list);
1977                 init_waitqueue_head(&lad->lad_thread.t_ctl_waitq);
1978                 lad->lad_ops = lao;
1979                 lad->lad_name = name;
1980         }
1981
1982         return lad;
1983 }
1984
1985 struct lfsck_assistant_object *
1986 lfsck_assistant_object_init(const struct lu_env *env, const struct lu_fid *fid,
1987                             const struct lu_attr *attr, __u64 cookie,
1988                             bool is_dir)
1989 {
1990         struct lfsck_assistant_object   *lso;
1991
1992         OBD_ALLOC_PTR(lso);
1993         if (lso == NULL)
1994                 return ERR_PTR(-ENOMEM);
1995
1996         lso->lso_fid = *fid;
1997         if (attr != NULL)
1998                 lso->lso_attr = *attr;
1999
2000         atomic_set(&lso->lso_ref, 1);
2001         lso->lso_oit_cookie = cookie;
2002         if (is_dir)
2003                 lso->lso_is_dir = 1;
2004
2005         return lso;
2006 }
2007
2008 struct dt_object *
2009 lfsck_assistant_object_load(const struct lu_env *env,
2010                             struct lfsck_instance *lfsck,
2011                             struct lfsck_assistant_object *lso)
2012 {
2013         struct dt_object *obj;
2014
2015         obj = lfsck_object_find_bottom(env, lfsck, &lso->lso_fid);
2016         if (IS_ERR(obj))
2017                 return obj;
2018
2019         if (unlikely(!dt_object_exists(obj) || lfsck_is_dead_obj(obj))) {
2020                 lso->lso_dead = 1;
2021                 lfsck_object_put(env, obj);
2022
2023                 return ERR_PTR(-ENOENT);
2024         }
2025
2026         if (lso->lso_is_dir && unlikely(!dt_try_as_dir(env, obj))) {
2027                 lfsck_object_put(env, obj);
2028
2029                 return ERR_PTR(-ENOTDIR);
2030         }
2031
2032         return obj;
2033 }
2034
2035 /**
2036  * Generic LFSCK asynchronous communication interpretor function.
2037  * The LFSCK RPC reply for both the event notification and status
2038  * querying will be handled here.
2039  *
2040  * \param[in] env       pointer to the thread context
2041  * \param[in] req       pointer to the LFSCK request
2042  * \param[in] args      pointer to the lfsck_async_interpret_args
2043  * \param[in] rc        the result for handling the LFSCK request
2044  *
2045  * \retval              0 for success
2046  * \retval              negative error number on failure
2047  */
2048 int lfsck_async_interpret_common(const struct lu_env *env,
2049                                  struct ptlrpc_request *req,
2050                                  void *args, int rc)
2051 {
2052         struct lfsck_async_interpret_args *laia = args;
2053         struct lfsck_component            *com  = laia->laia_com;
2054         struct lfsck_assistant_data       *lad  = com->lc_data;
2055         struct lfsck_tgt_descs            *ltds = laia->laia_ltds;
2056         struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
2057         struct lfsck_request              *lr   = laia->laia_lr;
2058
2059         LASSERT(com->lc_lfsck->li_master);
2060
2061         switch (lr->lr_event) {
2062         case LE_START:
2063                 if (unlikely(rc == -EINPROGRESS)) {
2064                         ltd->ltd_retry_start = 1;
2065                         break;
2066                 }
2067
2068                 if (rc != 0) {
2069                         CDEBUG(D_LFSCK, "%s: fail to notify %s %x for %s "
2070                                "start: rc = %d\n",
2071                                lfsck_lfsck2name(com->lc_lfsck),
2072                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2073                                ltd->ltd_index, lad->lad_name, rc);
2074
2075                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2076                                 struct lfsck_layout *lo = com->lc_file_ram;
2077
2078                                 if (lr->lr_flags & LEF_TO_OST)
2079                                         lfsck_lad_set_bitmap(env, com,
2080                                                              ltd->ltd_index);
2081                                 else
2082                                         lo->ll_flags |= LF_INCOMPLETE;
2083                         } else {
2084                                 struct lfsck_namespace *ns = com->lc_file_ram;
2085
2086                                 /* If some MDT does not join the namespace
2087                                  * LFSCK, then we cannot know whether there
2088                                  * is some name entry on such MDT that with
2089                                  * the referenced MDT-object on this MDT or
2090                                  * not. So the namespace LFSCK on this MDT
2091                                  * cannot handle orphan MDT-objects properly.
2092                                  * So we mark the LFSCK as LF_INCOMPLETE and
2093                                  * skip orphan MDT-objects handling. */
2094                                 ns->ln_flags |= LF_INCOMPLETE;
2095                         }
2096                         break;
2097                 }
2098
2099                 spin_lock(&ltds->ltd_lock);
2100                 if (ltd->ltd_dead) {
2101                         spin_unlock(&ltds->ltd_lock);
2102                         break;
2103                 }
2104
2105                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2106                         struct list_head *list;
2107                         struct list_head *phase_list;
2108
2109                         if (ltd->ltd_layout_done) {
2110                                 spin_unlock(&ltds->ltd_lock);
2111                                 break;
2112                         }
2113
2114                         if (lr->lr_flags & LEF_TO_OST) {
2115                                 list = &lad->lad_ost_list;
2116                                 phase_list = &lad->lad_ost_phase1_list;
2117                         } else {
2118                                 list = &lad->lad_mdt_list;
2119                                 phase_list = &lad->lad_mdt_phase1_list;
2120                         }
2121
2122                         if (list_empty(&ltd->ltd_layout_list))
2123                                 list_add_tail(&ltd->ltd_layout_list, list);
2124                         if (list_empty(&ltd->ltd_layout_phase_list))
2125                                 list_add_tail(&ltd->ltd_layout_phase_list,
2126                                               phase_list);
2127                 } else {
2128                         if (ltd->ltd_namespace_done) {
2129                                 spin_unlock(&ltds->ltd_lock);
2130                                 break;
2131                         }
2132
2133                         if (list_empty(&ltd->ltd_namespace_list))
2134                                 list_add_tail(&ltd->ltd_namespace_list,
2135                                               &lad->lad_mdt_list);
2136                         if (list_empty(&ltd->ltd_namespace_phase_list))
2137                                 list_add_tail(&ltd->ltd_namespace_phase_list,
2138                                               &lad->lad_mdt_phase1_list);
2139                 }
2140                 spin_unlock(&ltds->ltd_lock);
2141                 break;
2142         case LE_STOP:
2143         case LE_PHASE1_DONE:
2144         case LE_PHASE2_DONE:
2145         case LE_PEER_EXIT:
2146                 if (rc != 0 && rc != -EALREADY)
2147                         CDEBUG(D_LFSCK, "%s: fail to notify %s %x for %s: "
2148                               "event = %d, rc = %d\n",
2149                               lfsck_lfsck2name(com->lc_lfsck),
2150                               (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2151                               ltd->ltd_index, lad->lad_name, lr->lr_event, rc);
2152                 break;
2153         case LE_QUERY: {
2154                 struct lfsck_reply *reply;
2155                 struct list_head *list;
2156                 struct list_head *phase_list;
2157
2158                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2159                         list = &ltd->ltd_layout_list;
2160                         phase_list = &ltd->ltd_layout_phase_list;
2161                 } else {
2162                         list = &ltd->ltd_namespace_list;
2163                         phase_list = &ltd->ltd_namespace_phase_list;
2164                 }
2165
2166                 if (rc != 0) {
2167                         if (lr->lr_flags & LEF_QUERY_ALL) {
2168                                 lfsck_reset_ltd_status(ltd, com->lc_type);
2169                                 break;
2170                         }
2171
2172                         spin_lock(&ltds->ltd_lock);
2173                         list_del_init(phase_list);
2174                         list_del_init(list);
2175                         spin_unlock(&ltds->ltd_lock);
2176                         break;
2177                 }
2178
2179                 reply = req_capsule_server_get(&req->rq_pill,
2180                                                &RMF_LFSCK_REPLY);
2181                 if (reply == NULL) {
2182                         rc = -EPROTO;
2183                         CDEBUG(D_LFSCK, "%s: invalid query reply for %s: "
2184                                "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
2185                                lad->lad_name, rc);
2186
2187                         if (lr->lr_flags & LEF_QUERY_ALL) {
2188                                 lfsck_reset_ltd_status(ltd, com->lc_type);
2189                                 break;
2190                         }
2191
2192                         spin_lock(&ltds->ltd_lock);
2193                         list_del_init(phase_list);
2194                         list_del_init(list);
2195                         spin_unlock(&ltds->ltd_lock);
2196                         break;
2197                 }
2198
2199                 if (lr->lr_flags & LEF_QUERY_ALL) {
2200                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2201                                 ltd->ltd_layout_status = reply->lr_status;
2202                                 ltd->ltd_layout_repaired = reply->lr_repaired;
2203                         } else {
2204                                 ltd->ltd_namespace_status = reply->lr_status;
2205                                 ltd->ltd_namespace_repaired =
2206                                                         reply->lr_repaired;
2207                         }
2208                         break;
2209                 }
2210
2211                 switch (reply->lr_status) {
2212                 case LS_SCANNING_PHASE1:
2213                         break;
2214                 case LS_SCANNING_PHASE2:
2215                         spin_lock(&ltds->ltd_lock);
2216                         list_del_init(phase_list);
2217                         if (ltd->ltd_dead) {
2218                                 spin_unlock(&ltds->ltd_lock);
2219                                 break;
2220                         }
2221
2222                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2223                                 if (ltd->ltd_layout_done) {
2224                                         spin_unlock(&ltds->ltd_lock);
2225                                         break;
2226                                 }
2227
2228                                 if (lr->lr_flags & LEF_TO_OST)
2229                                         list_add_tail(phase_list,
2230                                                 &lad->lad_ost_phase2_list);
2231                                 else
2232                                         list_add_tail(phase_list,
2233                                                 &lad->lad_mdt_phase2_list);
2234                         } else {
2235                                 if (ltd->ltd_namespace_done) {
2236                                         spin_unlock(&ltds->ltd_lock);
2237                                         break;
2238                                 }
2239
2240                                 list_add_tail(phase_list,
2241                                               &lad->lad_mdt_phase2_list);
2242                         }
2243                         spin_unlock(&ltds->ltd_lock);
2244                         break;
2245                 default:
2246                         spin_lock(&ltds->ltd_lock);
2247                         list_del_init(phase_list);
2248                         list_del_init(list);
2249                         spin_unlock(&ltds->ltd_lock);
2250                         break;
2251                 }
2252                 break;
2253         }
2254         default:
2255                 CDEBUG(D_LFSCK, "%s: unexpected event: rc = %d\n",
2256                        lfsck_lfsck2name(com->lc_lfsck), lr->lr_event);
2257                 break;
2258         }
2259
2260         if (!laia->laia_shared) {
2261                 lfsck_tgt_put(ltd);
2262                 lfsck_component_put(env, com);
2263         }
2264
2265         return 0;
2266 }
2267
2268 static void lfsck_interpret(const struct lu_env *env,
2269                             struct lfsck_instance *lfsck,
2270                             struct ptlrpc_request *req, void *args, int result)
2271 {
2272         struct lfsck_async_interpret_args *laia = args;
2273         struct lfsck_component            *com;
2274
2275         LASSERT(laia->laia_com == NULL);
2276         LASSERT(laia->laia_shared);
2277
2278         spin_lock(&lfsck->li_lock);
2279         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
2280                 laia->laia_com = com;
2281                 lfsck_async_interpret_common(env, req, laia, result);
2282         }
2283
2284         list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
2285                 laia->laia_com = com;
2286                 lfsck_async_interpret_common(env, req, laia, result);
2287         }
2288         spin_unlock(&lfsck->li_lock);
2289 }
2290
2291 static int lfsck_stop_notify(const struct lu_env *env,
2292                              struct lfsck_instance *lfsck,
2293                              struct lfsck_tgt_descs *ltds,
2294                              struct lfsck_tgt_desc *ltd, __u16 type)
2295 {
2296         struct lfsck_component *com;
2297         int                     rc = 0;
2298         ENTRY;
2299
2300         LASSERT(lfsck->li_master);
2301
2302         spin_lock(&lfsck->li_lock);
2303         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_scan);
2304         if (com == NULL)
2305                 com = __lfsck_component_find(lfsck, type,
2306                                              &lfsck->li_list_double_scan);
2307         if (com != NULL)
2308                 lfsck_component_get(com);
2309         spin_unlock(&lfsck->li_lock);
2310
2311         if (com != NULL) {
2312                 struct lfsck_thread_info          *info  = lfsck_env_info(env);
2313                 struct lfsck_async_interpret_args *laia  = &info->lti_laia;
2314                 struct lfsck_request              *lr    = &info->lti_lr;
2315                 struct lfsck_assistant_data       *lad   = com->lc_data;
2316                 struct list_head                  *list;
2317                 struct list_head                  *phase_list;
2318                 struct ptlrpc_request_set         *set;
2319
2320                 set = ptlrpc_prep_set();
2321                 if (set == NULL) {
2322                         lfsck_component_put(env, com);
2323
2324                         RETURN(-ENOMEM);
2325                 }
2326
2327                 if (type == LFSCK_TYPE_LAYOUT) {
2328                         list = &ltd->ltd_layout_list;
2329                         phase_list = &ltd->ltd_layout_phase_list;
2330                 } else {
2331                         list = &ltd->ltd_namespace_list;
2332                         phase_list = &ltd->ltd_namespace_phase_list;
2333                 }
2334
2335                 spin_lock(&ltds->ltd_lock);
2336                 if (list_empty(list)) {
2337                         LASSERT(list_empty(phase_list));
2338                         spin_unlock(&ltds->ltd_lock);
2339                         ptlrpc_set_destroy(set);
2340
2341                         RETURN(0);
2342                 }
2343
2344                 list_del_init(phase_list);
2345                 list_del_init(list);
2346                 spin_unlock(&ltds->ltd_lock);
2347
2348                 memset(lr, 0, sizeof(*lr));
2349                 lr->lr_index = lfsck_dev_idx(lfsck);
2350                 lr->lr_event = LE_PEER_EXIT;
2351                 lr->lr_active = type;
2352                 lr->lr_status = LS_CO_PAUSED;
2353                 if (ltds == &lfsck->li_ost_descs)
2354                         lr->lr_flags = LEF_TO_OST;
2355
2356                 memset(laia, 0, sizeof(*laia));
2357                 laia->laia_com = com;
2358                 laia->laia_ltds = ltds;
2359                 atomic_inc(&ltd->ltd_ref);
2360                 laia->laia_ltd = ltd;
2361                 laia->laia_lr = lr;
2362
2363                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2364                                          lfsck_async_interpret_common,
2365                                          laia, LFSCK_NOTIFY);
2366                 if (rc != 0) {
2367                         CDEBUG(D_LFSCK, "%s: fail to notify %s %x for "
2368                                "co-stop for %s: rc = %d\n",
2369                                lfsck_lfsck2name(lfsck),
2370                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2371                                ltd->ltd_index, lad->lad_name, rc);
2372                         lfsck_tgt_put(ltd);
2373                 } else {
2374                         rc = ptlrpc_set_wait(env, set);
2375                 }
2376
2377                 ptlrpc_set_destroy(set);
2378                 lfsck_component_put(env, com);
2379         }
2380
2381         RETURN(rc);
2382 }
2383
2384 static int lfsck_async_interpret(const struct lu_env *env,
2385                                  struct ptlrpc_request *req,
2386                                  void *args, int rc)
2387 {
2388         struct lfsck_async_interpret_args *laia = args;
2389         struct lfsck_instance             *lfsck;
2390
2391         lfsck = container_of(laia->laia_ltds, struct lfsck_instance,
2392                              li_mdt_descs);
2393         lfsck_interpret(env, lfsck, req, laia, rc);
2394         lfsck_tgt_put(laia->laia_ltd);
2395         if (rc != 0 && laia->laia_result != -EALREADY)
2396                 laia->laia_result = rc;
2397
2398         return 0;
2399 }
2400
2401 int lfsck_async_request(const struct lu_env *env, struct obd_export *exp,
2402                         struct lfsck_request *lr,
2403                         struct ptlrpc_request_set *set,
2404                         ptlrpc_interpterer_t interpreter,
2405                         void *args, int request)
2406 {
2407         struct lfsck_async_interpret_args *laia;
2408         struct ptlrpc_request             *req;
2409         struct lfsck_request              *tmp;
2410         struct req_format                 *format;
2411         int                                rc;
2412
2413         switch (request) {
2414         case LFSCK_NOTIFY:
2415                 format = &RQF_LFSCK_NOTIFY;
2416                 break;
2417         case LFSCK_QUERY:
2418                 format = &RQF_LFSCK_QUERY;
2419                 break;
2420         default:
2421                 CDEBUG(D_LFSCK, "%s: unknown async request %d: rc = %d\n",
2422                        exp->exp_obd->obd_name, request, -EINVAL);
2423                 return -EINVAL;
2424         }
2425
2426         req = ptlrpc_request_alloc(class_exp2cliimp(exp), format);
2427         if (req == NULL)
2428                 return -ENOMEM;
2429
2430         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, request);
2431         if (rc != 0) {
2432                 ptlrpc_request_free(req);
2433
2434                 return rc;
2435         }
2436
2437         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2438         *tmp = *lr;
2439         ptlrpc_request_set_replen(req);
2440
2441         laia = ptlrpc_req_async_args(laia, req);
2442         *laia = *(struct lfsck_async_interpret_args *)args;
2443         if (laia->laia_com != NULL)
2444                 lfsck_component_get(laia->laia_com);
2445         req->rq_interpret_reply = interpreter;
2446         req->rq_allow_intr = 1;
2447         req->rq_no_delay = 1;
2448         ptlrpc_set_add_req(set, req);
2449
2450         return 0;
2451 }
2452
2453 int lfsck_query_all(const struct lu_env *env, struct lfsck_component *com)
2454 {
2455         struct lfsck_thread_info          *info  = lfsck_env_info(env);
2456         struct lfsck_request              *lr    = &info->lti_lr;
2457         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
2458         struct lfsck_instance             *lfsck = com->lc_lfsck;
2459         struct lfsck_tgt_descs            *ltds  = &lfsck->li_mdt_descs;
2460         struct lfsck_tgt_desc             *ltd;
2461         struct ptlrpc_request_set         *set;
2462         int                                idx;
2463         int                                rc;
2464         ENTRY;
2465
2466         memset(lr, 0, sizeof(*lr));
2467         lr->lr_event = LE_QUERY;
2468         lr->lr_active = com->lc_type;
2469         lr->lr_flags = LEF_QUERY_ALL;
2470
2471         memset(laia, 0, sizeof(*laia));
2472         laia->laia_com = com;
2473         laia->laia_lr = lr;
2474
2475         set = ptlrpc_prep_set();
2476         if (set == NULL)
2477                 RETURN(-ENOMEM);
2478
2479 again:
2480         laia->laia_ltds = ltds;
2481         down_read(&ltds->ltd_rw_sem);
2482         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
2483                 ltd = lfsck_tgt_get(ltds, idx);
2484                 LASSERT(ltd != NULL);
2485
2486                 laia->laia_ltd = ltd;
2487                 up_read(&ltds->ltd_rw_sem);
2488                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2489                                          lfsck_async_interpret_common,
2490                                          laia, LFSCK_QUERY);
2491                 if (rc != 0) {
2492                         struct lfsck_assistant_data *lad = com->lc_data;
2493
2494                         CDEBUG(D_LFSCK, "%s: Fail to query %s %x for stat %s: "
2495                                "rc = %d\n", lfsck_lfsck2name(lfsck),
2496                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2497                                ltd->ltd_index, lad->lad_name, rc);
2498                         lfsck_reset_ltd_status(ltd, com->lc_type);
2499                         lfsck_tgt_put(ltd);
2500                 }
2501                 down_read(&ltds->ltd_rw_sem);
2502         }
2503         up_read(&ltds->ltd_rw_sem);
2504
2505         if (com->lc_type == LFSCK_TYPE_LAYOUT && !(lr->lr_flags & LEF_TO_OST)) {
2506                 ltds = &lfsck->li_ost_descs;
2507                 lr->lr_flags |= LEF_TO_OST;
2508                 goto again;
2509         }
2510
2511         rc = ptlrpc_set_wait(env, set);
2512         ptlrpc_set_destroy(set);
2513
2514         RETURN(rc);
2515 }
2516
2517 int lfsck_start_assistant(const struct lu_env *env, struct lfsck_component *com,
2518                           struct lfsck_start_param *lsp)
2519 {
2520         struct lfsck_instance           *lfsck   = com->lc_lfsck;
2521         struct lfsck_assistant_data     *lad     = com->lc_data;
2522         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
2523         struct ptlrpc_thread            *athread = &lad->lad_thread;
2524         struct lfsck_thread_args        *lta;
2525         struct task_struct              *task;
2526         int                              rc;
2527         ENTRY;
2528
2529         lad->lad_assistant_status = 0;
2530         lad->lad_post_result = 0;
2531         lad->lad_flags = 0;
2532         lad->lad_advance_lock = false;
2533         thread_set_flags(athread, 0);
2534
2535         lta = lfsck_thread_args_init(lfsck, com, lsp);
2536         if (IS_ERR(lta))
2537                 RETURN(PTR_ERR(lta));
2538
2539         task = kthread_run(lfsck_assistant_engine, lta, lad->lad_name);
2540         if (IS_ERR(task)) {
2541                 rc = PTR_ERR(task);
2542                 CERROR("%s: cannot start LFSCK assistant thread for %s: "
2543                        "rc = %d\n", lfsck_lfsck2name(lfsck), lad->lad_name, rc);
2544                 lfsck_thread_args_fini(lta);
2545         } else {
2546                 wait_event_idle(mthread->t_ctl_waitq,
2547                                 thread_is_running(athread) ||
2548                                 thread_is_stopped(athread) ||
2549                                 !thread_is_starting(mthread));
2550                 if (unlikely(!thread_is_starting(mthread)))
2551                         /* stopped by race */
2552                         rc = -ESRCH;
2553                 else if (unlikely(!thread_is_running(athread)))
2554                         rc = lad->lad_assistant_status;
2555                 else
2556                         rc = 0;
2557         }
2558
2559         RETURN(rc);
2560 }
2561
2562 int lfsck_checkpoint_generic(const struct lu_env *env,
2563                              struct lfsck_component *com)
2564 {
2565         struct lfsck_assistant_data     *lad     = com->lc_data;
2566         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2567         struct ptlrpc_thread            *athread = &lad->lad_thread;
2568
2569         wait_event_idle(mthread->t_ctl_waitq,
2570                         list_empty(&lad->lad_req_list) ||
2571                         !thread_is_running(mthread) ||
2572                         thread_is_stopped(athread));
2573
2574         if (!thread_is_running(mthread) || thread_is_stopped(athread))
2575                 return LFSCK_CHECKPOINT_SKIP;
2576
2577         return 0;
2578 }
2579
2580 void lfsck_post_generic(const struct lu_env *env,
2581                         struct lfsck_component *com, int *result)
2582 {
2583         struct lfsck_assistant_data     *lad     = com->lc_data;
2584         struct ptlrpc_thread            *athread = &lad->lad_thread;
2585         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2586
2587         lad->lad_post_result = *result;
2588         if (*result <= 0)
2589                 set_bit(LAD_EXIT, &lad->lad_flags);
2590         set_bit(LAD_TO_POST, &lad->lad_flags);
2591
2592         CDEBUG(D_LFSCK, "%s: waiting for assistant to do %s post, rc = %d\n",
2593                lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, *result);
2594
2595         wake_up_all(&athread->t_ctl_waitq);
2596         wait_event_idle(mthread->t_ctl_waitq,
2597                         (*result > 0 && list_empty(&lad->lad_req_list)) ||
2598                         thread_is_stopped(athread));
2599
2600         if (lad->lad_assistant_status < 0)
2601                 *result = lad->lad_assistant_status;
2602
2603         CDEBUG(D_LFSCK, "%s: the assistant has done %s post, rc = %d\n",
2604                lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, *result);
2605 }
2606
2607 int lfsck_double_scan_generic(const struct lu_env *env,
2608                               struct lfsck_component *com, int status)
2609 {
2610         struct lfsck_assistant_data     *lad     = com->lc_data;
2611         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2612         struct ptlrpc_thread            *athread = &lad->lad_thread;
2613
2614         if (status != LS_SCANNING_PHASE2)
2615                 set_bit(LAD_EXIT, &lad->lad_flags);
2616         else
2617                 set_bit(LAD_TO_DOUBLE_SCAN, &lad->lad_flags);
2618
2619         CDEBUG(D_LFSCK, "%s: waiting for assistant to do %s double_scan, "
2620                "status %d\n",
2621                lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, status);
2622
2623         wake_up_all(&athread->t_ctl_waitq);
2624         wait_event_idle(mthread->t_ctl_waitq,
2625                         test_bit(LAD_IN_DOUBLE_SCAN, &lad->lad_flags) ||
2626                         thread_is_stopped(athread));
2627
2628         CDEBUG(D_LFSCK, "%s: the assistant has done %s double_scan, "
2629                "status %d\n", lfsck_lfsck2name(com->lc_lfsck), lad->lad_name,
2630                lad->lad_assistant_status);
2631
2632         if (lad->lad_assistant_status < 0)
2633                 return lad->lad_assistant_status;
2634
2635         return 0;
2636 }
2637
2638 void lfsck_quit_generic(const struct lu_env *env,
2639                         struct lfsck_component *com)
2640 {
2641         struct lfsck_assistant_data     *lad     = com->lc_data;
2642         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2643         struct ptlrpc_thread            *athread = &lad->lad_thread;
2644
2645         set_bit(LAD_EXIT, &lad->lad_flags);
2646         wake_up_all(&athread->t_ctl_waitq);
2647         wait_event_idle(mthread->t_ctl_waitq,
2648                         thread_is_init(athread) ||
2649                         thread_is_stopped(athread));
2650 }
2651
2652 int lfsck_load_one_trace_file(const struct lu_env *env,
2653                               struct lfsck_component *com,
2654                               struct dt_object *parent,
2655                               struct dt_object **child,
2656                               const struct dt_index_features *ft,
2657                               const char *name, bool reset)
2658 {
2659         struct lfsck_instance *lfsck = com->lc_lfsck;
2660         struct dt_object *obj;
2661         int rc;
2662         ENTRY;
2663
2664         if (*child != NULL) {
2665                 struct dt_it *it;
2666                 const struct dt_it_ops *iops;
2667                 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid3;
2668
2669                 if (!reset)
2670                         RETURN(0);
2671
2672                 obj = *child;
2673                 rc = obj->do_ops->do_index_try(env, obj, ft);
2674                 if (rc)
2675                         /* unlink by force */
2676                         goto unlink;
2677
2678                 iops = &obj->do_index_ops->dio_it;
2679                 it = iops->init(env, obj, 0);
2680                 if (IS_ERR(it))
2681                         /* unlink by force */
2682                         goto unlink;
2683
2684                 fid_zero(fid);
2685                 rc = iops->get(env, it, (const struct dt_key *)fid);
2686                 if (rc >= 0) {
2687                         rc = iops->next(env, it);
2688                         iops->put(env, it);
2689                 }
2690                 iops->fini(env, it);
2691                 if (rc > 0)
2692                         /* "rc > 0" means the index file is empty. */
2693                         RETURN(0);
2694
2695 unlink:
2696                 /* The old index is not empty, remove it firstly. */
2697                 rc = local_object_unlink(env, lfsck->li_bottom, parent, name);
2698                 CDEBUG_LIMIT(rc ? D_ERROR : D_LFSCK,
2699                              "%s: unlink lfsck sub trace file %s: rc = %d\n",
2700                              lfsck_lfsck2name(com->lc_lfsck), name, rc);
2701                 if (rc)
2702                         RETURN(rc);
2703
2704                 if (*child) {
2705                         lfsck_object_put(env, *child);
2706                         *child = NULL;
2707                 }
2708         } else if (reset) {
2709                 goto unlink;
2710         }
2711
2712         obj = local_index_find_or_create(env, lfsck->li_los, parent, name,
2713                                          S_IFREG | S_IRUGO | S_IWUSR, ft);
2714         if (IS_ERR(obj))
2715                 RETURN(PTR_ERR(obj));
2716
2717         rc = obj->do_ops->do_index_try(env, obj, ft);
2718         if (rc) {
2719                 lfsck_object_put(env, obj);
2720                 CDEBUG(D_LFSCK, "%s: LFSCK fail to load "
2721                        "sub trace file %s: rc = %d\n",
2722                        lfsck_lfsck2name(com->lc_lfsck), name, rc);
2723         } else {
2724                 *child = obj;
2725         }
2726
2727         RETURN(rc);
2728 }
2729
2730 int lfsck_load_sub_trace_files(const struct lu_env *env,
2731                                struct lfsck_component *com,
2732                                const struct dt_index_features *ft,
2733                                const char *prefix, bool reset)
2734 {
2735         char *name = lfsck_env_info(env)->lti_key;
2736         struct lfsck_sub_trace_obj *lsto;
2737         int rc;
2738         int i;
2739
2740         for (i = 0, rc = 0, lsto = &com->lc_sub_trace_objs[0];
2741              i < LFSCK_STF_COUNT && rc == 0; i++, lsto++) {
2742                 snprintf(name, NAME_MAX, "%s_%02d", prefix, i);
2743                 rc = lfsck_load_one_trace_file(env, com,
2744                                 com->lc_lfsck->li_lfsck_dir,
2745                                 &lsto->lsto_obj, ft, name, reset);
2746         }
2747
2748         return rc;
2749 }
2750
2751 /* external interfaces */
2752 int lfsck_get_speed(char *buf, struct dt_device *key)
2753 {
2754         struct lu_env           env;
2755         struct lfsck_instance  *lfsck;
2756         int                     rc;
2757         ENTRY;
2758
2759         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2760         if (rc != 0)
2761                 RETURN(rc);
2762
2763         lfsck = lfsck_instance_find(key, true, false);
2764         if (lfsck && buf) {
2765                 rc = sprintf(buf, "%u\n",
2766                              lfsck->li_bookmark_ram.lb_speed_limit);
2767                 lfsck_instance_put(&env, lfsck);
2768         } else {
2769                 rc = -ENXIO;
2770         }
2771
2772         lu_env_fini(&env);
2773
2774         RETURN(rc);
2775 }
2776 EXPORT_SYMBOL(lfsck_get_speed);
2777
2778 int lfsck_set_speed(struct dt_device *key, __u32 val)
2779 {
2780         struct lu_env           env;
2781         struct lfsck_instance  *lfsck;
2782         int                     rc;
2783         ENTRY;
2784
2785         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2786         if (rc != 0)
2787                 RETURN(rc);
2788
2789         lfsck = lfsck_instance_find(key, true, false);
2790         if (likely(lfsck != NULL)) {
2791                 mutex_lock(&lfsck->li_mutex);
2792                 if (__lfsck_set_speed(lfsck, val))
2793                         rc = lfsck_bookmark_store(&env, lfsck);
2794                 mutex_unlock(&lfsck->li_mutex);
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_set_speed);
2805
2806 int lfsck_get_windows(char *buf, struct dt_device *key)
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                 rc = sprintf(buf, "%u\n",
2820                              lfsck->li_bookmark_ram.lb_async_windows);
2821                 lfsck_instance_put(&env, lfsck);
2822         } else {
2823                 rc = -ENXIO;
2824         }
2825
2826         lu_env_fini(&env);
2827
2828         RETURN(rc);
2829 }
2830 EXPORT_SYMBOL(lfsck_get_windows);
2831
2832 int lfsck_set_windows(struct dt_device *key, unsigned int val)
2833 {
2834         struct lu_env           env;
2835         struct lfsck_instance  *lfsck;
2836         int                     rc;
2837         ENTRY;
2838
2839         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2840         if (rc != 0)
2841                 RETURN(rc);
2842
2843         lfsck = lfsck_instance_find(key, true, false);
2844         if (likely(lfsck != NULL)) {
2845                 if (val < 1 || val > LFSCK_ASYNC_WIN_MAX) {
2846                         CWARN("%s: invalid async windows size that may "
2847                               "cause memory issues. The valid range is "
2848                               "[1 - %u].\n",
2849                               lfsck_lfsck2name(lfsck), LFSCK_ASYNC_WIN_MAX);
2850                         rc = -EINVAL;
2851                 } else if (lfsck->li_bookmark_ram.lb_async_windows != val) {
2852                         mutex_lock(&lfsck->li_mutex);
2853                         lfsck->li_bookmark_ram.lb_async_windows = val;
2854                         rc = lfsck_bookmark_store(&env, lfsck);
2855                         mutex_unlock(&lfsck->li_mutex);
2856                 }
2857                 lfsck_instance_put(&env, lfsck);
2858         } else {
2859                 rc = -ENXIO;
2860         }
2861
2862         lu_env_fini(&env);
2863
2864         RETURN(rc);
2865 }
2866 EXPORT_SYMBOL(lfsck_set_windows);
2867
2868 int lfsck_dump(struct seq_file *m, struct dt_device *key, enum lfsck_type type)
2869 {
2870         struct lu_env           env;
2871         struct lfsck_instance  *lfsck;
2872         struct lfsck_component *com;
2873         int                     rc;
2874         ENTRY;
2875
2876         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2877         if (rc != 0)
2878                 RETURN(rc);
2879
2880         lfsck = lfsck_instance_find(key, true, false);
2881         if (likely(lfsck != NULL)) {
2882                 com = lfsck_component_find(lfsck, type);
2883                 if (likely(com != NULL)) {
2884                         com->lc_ops->lfsck_dump(&env, com, m);
2885                         lfsck_component_put(&env, com);
2886                 } else {
2887                         rc = -ENOTSUPP;
2888                 }
2889
2890                 lfsck_instance_put(&env, lfsck);
2891         } else {
2892                 rc = -ENXIO;
2893         }
2894
2895         lu_env_fini(&env);
2896
2897         RETURN(rc);
2898 }
2899 EXPORT_SYMBOL(lfsck_dump);
2900
2901 static int lfsck_stop_all(const struct lu_env *env,
2902                           struct lfsck_instance *lfsck,
2903                           struct lfsck_stop *stop)
2904 {
2905         struct lfsck_thread_info          *info   = lfsck_env_info(env);
2906         struct lfsck_request              *lr     = &info->lti_lr;
2907         struct lfsck_async_interpret_args *laia   = &info->lti_laia;
2908         struct ptlrpc_request_set         *set;
2909         struct lfsck_tgt_descs            *ltds   = &lfsck->li_mdt_descs;
2910         struct lfsck_tgt_desc             *ltd;
2911         struct lfsck_bookmark             *bk     = &lfsck->li_bookmark_ram;
2912         __u32                              idx;
2913         int                                rc     = 0;
2914         int                                rc1    = 0;
2915         ENTRY;
2916
2917         LASSERT(stop->ls_flags & LPF_BROADCAST);
2918
2919         set = ptlrpc_prep_set();
2920         if (unlikely(set == NULL))
2921                 RETURN(-ENOMEM);
2922
2923         memset(lr, 0, sizeof(*lr));
2924         lr->lr_event = LE_STOP;
2925         lr->lr_index = lfsck_dev_idx(lfsck);
2926         lr->lr_status = stop->ls_status;
2927         lr->lr_version = bk->lb_version;
2928         lr->lr_active = LFSCK_TYPES_ALL;
2929         lr->lr_param = stop->ls_flags;
2930
2931         memset(laia, 0, sizeof(*laia));
2932         laia->laia_ltds = ltds;
2933         laia->laia_lr = lr;
2934         laia->laia_shared = 1;
2935
2936         down_read(&ltds->ltd_rw_sem);
2937         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
2938                 ltd = lfsck_tgt_get(ltds, idx);
2939                 LASSERT(ltd != NULL);
2940
2941                 laia->laia_ltd = ltd;
2942                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2943                                          lfsck_async_interpret, laia,
2944                                          LFSCK_NOTIFY);
2945                 if (rc != 0) {
2946                         lfsck_interpret(env, lfsck, NULL, laia, rc);
2947                         lfsck_tgt_put(ltd);
2948                         CERROR("%s: cannot notify MDT %x for LFSCK stop: "
2949                                "rc = %d\n", lfsck_lfsck2name(lfsck), idx, rc);
2950                         rc1 = rc;
2951                 }
2952         }
2953         up_read(&ltds->ltd_rw_sem);
2954
2955         rc = ptlrpc_set_wait(env, set);
2956         ptlrpc_set_destroy(set);
2957
2958         if (rc == 0)
2959                 rc = laia->laia_result;
2960
2961         if (rc == -EALREADY)
2962                 rc = 0;
2963
2964         if (rc != 0)
2965                 CERROR("%s: fail to stop LFSCK on some MDTs: rc = %d\n",
2966                        lfsck_lfsck2name(lfsck), rc);
2967
2968         RETURN(rc != 0 ? rc : rc1);
2969 }
2970
2971 static int lfsck_start_all(const struct lu_env *env,
2972                            struct lfsck_instance *lfsck,
2973                            struct lfsck_start *start)
2974 {
2975         struct lfsck_thread_info          *info   = lfsck_env_info(env);
2976         struct lfsck_request              *lr     = &info->lti_lr;
2977         struct lfsck_async_interpret_args *laia   = &info->lti_laia;
2978         struct ptlrpc_request_set         *set;
2979         struct lfsck_tgt_descs            *ltds   = &lfsck->li_mdt_descs;
2980         struct lfsck_tgt_desc             *ltd;
2981         struct lfsck_bookmark             *bk     = &lfsck->li_bookmark_ram;
2982         __u32                              idx;
2983         int                                rc     = 0;
2984         bool retry = false;
2985         ENTRY;
2986
2987         LASSERT(start->ls_flags & LPF_BROADCAST);
2988
2989         memset(lr, 0, sizeof(*lr));
2990         lr->lr_event = LE_START;
2991         lr->lr_index = lfsck_dev_idx(lfsck);
2992         lr->lr_speed = bk->lb_speed_limit;
2993         lr->lr_version = bk->lb_version;
2994         lr->lr_active = start->ls_active;
2995         lr->lr_param = start->ls_flags;
2996         lr->lr_async_windows = bk->lb_async_windows;
2997         lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN |
2998                        LSV_ASYNC_WINDOWS | LSV_CREATE_OSTOBJ |
2999                        LSV_CREATE_MDTOBJ;
3000
3001         memset(laia, 0, sizeof(*laia));
3002         laia->laia_ltds = ltds;
3003         laia->laia_lr = lr;
3004         laia->laia_shared = 1;
3005
3006 again:
3007         set = ptlrpc_prep_set();
3008         if (unlikely(!set))
3009                 RETURN(-ENOMEM);
3010
3011         down_read(&ltds->ltd_rw_sem);
3012         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
3013                 ltd = lfsck_tgt_get(ltds, idx);
3014                 LASSERT(ltd != NULL);
3015
3016                 if (retry && !ltd->ltd_retry_start) {
3017                         lfsck_tgt_put(ltd);
3018                         continue;
3019                 }
3020
3021                 laia->laia_ltd = ltd;
3022                 ltd->ltd_retry_start = 0;
3023                 ltd->ltd_layout_done = 0;
3024                 ltd->ltd_namespace_done = 0;
3025                 ltd->ltd_synced_failures = 0;
3026                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
3027                                          lfsck_async_interpret, laia,
3028                                          LFSCK_NOTIFY);
3029                 if (rc != 0) {
3030                         lfsck_interpret(env, lfsck, NULL, laia, rc);
3031                         lfsck_tgt_put(ltd);
3032                         CERROR("%s: cannot notify MDT %x for LFSCK "
3033                                "start, failout: rc = %d\n",
3034                                lfsck_lfsck2name(lfsck), idx, rc);
3035                         break;
3036                 }
3037         }
3038         up_read(&ltds->ltd_rw_sem);
3039
3040         if (rc != 0) {
3041                 ptlrpc_set_destroy(set);
3042
3043                 RETURN(rc);
3044         }
3045
3046         rc = ptlrpc_set_wait(env, set);
3047         ptlrpc_set_destroy(set);
3048
3049         if (rc == 0)
3050                 rc = laia->laia_result;
3051
3052         if (unlikely(rc == -EINPROGRESS)) {
3053                 retry = true;
3054                 schedule_timeout_interruptible(cfs_time_seconds(1));
3055                 set_current_state(TASK_RUNNING);
3056                 if (!signal_pending(current) &&
3057                     thread_is_running(&lfsck->li_thread))
3058                         goto again;
3059
3060                 rc = -EINTR;
3061         }
3062
3063         if (rc != 0) {
3064                 struct lfsck_stop *stop = &info->lti_stop;
3065
3066                 CERROR("%s: cannot start LFSCK on some MDTs, "
3067                        "stop all: rc = %d\n",
3068                        lfsck_lfsck2name(lfsck), rc);
3069                 if (rc != -EALREADY) {
3070                         stop->ls_status = LS_FAILED;
3071                         stop->ls_flags = LPF_ALL_TGT | LPF_BROADCAST;
3072                         lfsck_stop_all(env, lfsck, stop);
3073                 }
3074         }
3075
3076         RETURN(rc);
3077 }
3078
3079 int lfsck_start(const struct lu_env *env, struct dt_device *key,
3080                 struct lfsck_start_param *lsp)
3081 {
3082         struct lfsck_start              *start  = lsp->lsp_start;
3083         struct lfsck_instance           *lfsck;
3084         struct lfsck_bookmark           *bk;
3085         struct ptlrpc_thread            *thread;
3086         struct lfsck_component          *com;
3087         struct lfsck_thread_args        *lta;
3088         struct task_struct              *task;
3089         struct lfsck_tgt_descs          *ltds;
3090         struct lfsck_tgt_desc           *ltd;
3091         __u32                            idx;
3092         int                              rc     = 0;
3093         __u16                            valid  = 0;
3094         __u16                            flags  = 0;
3095         __u16                            type   = 1;
3096         ENTRY;
3097
3098         if (key->dd_rdonly)
3099                 RETURN(-EROFS);
3100
3101         lfsck = lfsck_instance_find(key, true, false);
3102         if (unlikely(lfsck == NULL))
3103                 RETURN(-ENXIO);
3104
3105         if (unlikely(lfsck->li_stopping))
3106                 GOTO(put, rc = -ENXIO);
3107
3108         /* System is not ready, try again later. */
3109         if (unlikely(lfsck->li_namespace == NULL ||
3110                      lfsck_dev_site(lfsck)->ss_server_fld == NULL))
3111                 GOTO(put, rc = -EINPROGRESS);
3112
3113         /* start == NULL means auto trigger paused LFSCK. */
3114         if (!start) {
3115                 if (list_empty(&lfsck->li_list_scan) ||
3116                     OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_AUTO))
3117                         GOTO(put, rc = 0);
3118         } else if (start->ls_flags & LPF_BROADCAST && !lfsck->li_master) {
3119                 CERROR("%s: only allow to specify '-A | -o' via MDS\n",
3120                        lfsck_lfsck2name(lfsck));
3121
3122                 GOTO(put, rc = -EPERM);
3123         }
3124
3125         bk = &lfsck->li_bookmark_ram;
3126         thread = &lfsck->li_thread;
3127         mutex_lock(&lfsck->li_mutex);
3128         spin_lock(&lfsck->li_lock);
3129         if (unlikely(thread_is_stopping(thread))) {
3130                 /* Someone is stopping the LFSCK. */
3131                 spin_unlock(&lfsck->li_lock);
3132                 GOTO(out, rc = -EBUSY);
3133         }
3134
3135         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
3136                 rc = -EALREADY;
3137                 if (unlikely(start == NULL)) {
3138                         spin_unlock(&lfsck->li_lock);
3139                         GOTO(out, rc);
3140                 }
3141
3142                 while (start->ls_active != 0) {
3143                         if (!(type & start->ls_active)) {
3144                                 type <<= 1;
3145                                 continue;
3146                         }
3147
3148                         com = __lfsck_component_find(lfsck, type,
3149                                                      &lfsck->li_list_scan);
3150                         if (com == NULL)
3151                                 com = __lfsck_component_find(lfsck, type,
3152                                                 &lfsck->li_list_double_scan);
3153                         if (com == NULL) {
3154                                 rc = -EOPNOTSUPP;
3155                                 break;
3156                         }
3157
3158                         if (com->lc_ops->lfsck_join != NULL) {
3159                                 rc = com->lc_ops->lfsck_join( env, com, lsp);
3160                                 if (rc != 0 && rc != -EALREADY)
3161                                         break;
3162                         }
3163                         start->ls_active &= ~type;
3164                         type <<= 1;
3165                 }
3166                 spin_unlock(&lfsck->li_lock);
3167                 GOTO(out, rc);
3168         }
3169         spin_unlock(&lfsck->li_lock);
3170
3171         lfsck->li_status = 0;
3172         lfsck->li_oit_over = 0;
3173         lfsck->li_start_unplug = 0;
3174         lfsck->li_drop_dryrun = 0;
3175         lfsck->li_new_scanned = 0;
3176
3177         /* For auto trigger. */
3178         if (start == NULL)
3179                 goto trigger;
3180
3181         start->ls_version = bk->lb_version;
3182
3183         if (start->ls_active != 0) {
3184                 struct lfsck_component *next;
3185
3186                 if (start->ls_active == LFSCK_TYPES_ALL)
3187                         start->ls_active = LFSCK_TYPES_SUPPORTED;
3188
3189                 if (start->ls_active & ~LFSCK_TYPES_SUPPORTED) {
3190                         start->ls_active &= ~LFSCK_TYPES_SUPPORTED;
3191                         GOTO(out, rc = -ENOTSUPP);
3192                 }
3193
3194                 list_for_each_entry_safe(com, next,
3195                                          &lfsck->li_list_scan, lc_link) {
3196                         if (!(com->lc_type & start->ls_active)) {
3197                                 rc = com->lc_ops->lfsck_post(env, com, 0,
3198                                                              false);
3199                                 if (rc != 0)
3200                                         GOTO(out, rc);
3201                         }
3202                 }
3203
3204                 while (start->ls_active != 0) {
3205                         if (type & start->ls_active) {
3206                                 com = __lfsck_component_find(lfsck, type,
3207                                                         &lfsck->li_list_idle);
3208                                 if (com != NULL)
3209                                         /* The component status will be updated
3210                                          * when its prep() is called later by
3211                                          * the LFSCK main engine. */
3212                                         list_move_tail(&com->lc_link,
3213                                                        &lfsck->li_list_scan);
3214                                 start->ls_active &= ~type;
3215                         }
3216                         type <<= 1;
3217                 }
3218         }
3219
3220         if (list_empty(&lfsck->li_list_scan)) {
3221                 /* The speed limit will be used to control both the LFSCK and
3222                  * low layer scrub (if applied), need to be handled firstly. */
3223                 if (start->ls_valid & LSV_SPEED_LIMIT) {
3224                         if (__lfsck_set_speed(lfsck, start->ls_speed_limit)) {
3225                                 rc = lfsck_bookmark_store(env, lfsck);
3226                                 if (rc != 0)
3227                                         GOTO(out, rc);
3228                         }
3229                 }
3230
3231                 goto trigger;
3232         }
3233
3234         if (start->ls_flags & LPF_RESET)
3235                 flags |= DOIF_RESET;
3236
3237         rc = lfsck_set_param(env, lfsck, start, !!(flags & DOIF_RESET));
3238         if (rc != 0)
3239                 GOTO(out, rc);
3240
3241         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
3242                 start->ls_active |= com->lc_type;
3243                 if (flags & DOIF_RESET) {
3244                         rc = com->lc_ops->lfsck_reset(env, com, false);
3245                         if (rc != 0)
3246                                 GOTO(out, rc);
3247                 }
3248         }
3249
3250         ltds = &lfsck->li_mdt_descs;
3251         down_read(&ltds->ltd_rw_sem);
3252         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
3253                 ltd = lfsck_ltd2tgt(ltds, idx);
3254                 LASSERT(ltd != NULL);
3255
3256                 ltd->ltd_layout_done = 0;
3257                 ltd->ltd_namespace_done = 0;
3258                 ltd->ltd_synced_failures = 0;
3259                 lfsck_reset_ltd_status(ltd, LFSCK_TYPE_NAMESPACE);
3260                 lfsck_reset_ltd_status(ltd, LFSCK_TYPE_LAYOUT);
3261                 list_del_init(&ltd->ltd_layout_phase_list);
3262                 list_del_init(&ltd->ltd_layout_list);
3263                 list_del_init(&ltd->ltd_namespace_phase_list);
3264                 list_del_init(&ltd->ltd_namespace_list);
3265         }
3266         up_read(&ltds->ltd_rw_sem);
3267
3268         ltds = &lfsck->li_ost_descs;
3269         down_read(&ltds->ltd_rw_sem);
3270         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
3271                 ltd = lfsck_ltd2tgt(ltds, idx);
3272                 LASSERT(ltd != NULL);
3273
3274                 ltd->ltd_layout_done = 0;
3275                 ltd->ltd_synced_failures = 0;
3276                 lfsck_reset_ltd_status(ltd, LFSCK_TYPE_LAYOUT);
3277                 list_del_init(&ltd->ltd_layout_phase_list);
3278                 list_del_init(&ltd->ltd_layout_list);
3279         }
3280         up_read(&ltds->ltd_rw_sem);
3281
3282 trigger:
3283         lfsck->li_args_dir = LUDA_64BITHASH | LUDA_VERIFY | LUDA_TYPE;
3284         if (bk->lb_param & LPF_DRYRUN)
3285                 lfsck->li_args_dir |= LUDA_VERIFY_DRYRUN;
3286
3287         if (start != NULL && start->ls_valid & LSV_ERROR_HANDLE) {
3288                 valid |= DOIV_ERROR_HANDLE;
3289                 if (start->ls_flags & LPF_FAILOUT)
3290                         flags |= DOIF_FAILOUT;
3291         }
3292
3293         if (start != NULL && start->ls_valid & LSV_DRYRUN) {
3294                 valid |= DOIV_DRYRUN;
3295                 if (start->ls_flags & LPF_DRYRUN)
3296                         flags |= DOIF_DRYRUN;
3297         }
3298
3299         if (!list_empty(&lfsck->li_list_scan))
3300                 flags |= DOIF_OUTUSED;
3301
3302         lfsck->li_args_oit = (flags << DT_OTABLE_IT_FLAGS_SHIFT) | valid;
3303         lta = lfsck_thread_args_init(lfsck, NULL, lsp);
3304         if (IS_ERR(lta))
3305                 GOTO(out, rc = PTR_ERR(lta));
3306
3307         __lfsck_set_speed(lfsck, bk->lb_speed_limit);
3308         spin_lock(&lfsck->li_lock);
3309         thread_set_flags(thread, SVC_STARTING);
3310         spin_unlock(&lfsck->li_lock);
3311         task = kthread_run(lfsck_master_engine, lta, "lfsck");
3312         if (IS_ERR(task)) {
3313                 rc = PTR_ERR(task);
3314                 CERROR("%s: cannot start LFSCK thread: rc = %d\n",
3315                        lfsck_lfsck2name(lfsck), rc);
3316                 lfsck_thread_args_fini(lta);
3317
3318                 GOTO(out, rc);
3319         }
3320
3321         wait_event_idle(thread->t_ctl_waitq,
3322                         thread_is_running(thread) ||
3323                         thread_is_stopped(thread));
3324         if (start == NULL || !(start->ls_flags & LPF_BROADCAST)) {
3325                 lfsck->li_start_unplug = 1;
3326                 wake_up_all(&thread->t_ctl_waitq);
3327
3328                 GOTO(out, rc = 0);
3329         }
3330
3331         /* release lfsck::li_mutex to avoid deadlock. */
3332         mutex_unlock(&lfsck->li_mutex);
3333         rc = lfsck_start_all(env, lfsck, start);
3334         if (rc != 0) {
3335                 spin_lock(&lfsck->li_lock);
3336                 if (thread_is_stopped(thread)) {
3337                         spin_unlock(&lfsck->li_lock);
3338                 } else {
3339                         lfsck->li_status = LS_FAILED;
3340                         lfsck->li_flags = 0;
3341                         thread_set_flags(thread, SVC_STOPPING);
3342                         spin_unlock(&lfsck->li_lock);
3343
3344                         lfsck->li_start_unplug = 1;
3345                         wake_up_all(&thread->t_ctl_waitq);
3346                         wait_event_idle(thread->t_ctl_waitq,
3347                                         thread_is_stopped(thread));
3348                 }
3349         } else {
3350                 lfsck->li_start_unplug = 1;
3351                 wake_up_all(&thread->t_ctl_waitq);
3352         }
3353
3354         GOTO(put, rc);
3355
3356 out:
3357         mutex_unlock(&lfsck->li_mutex);
3358
3359 put:
3360         lfsck_instance_put(env, lfsck);
3361
3362         return rc < 0 ? rc : 0;
3363 }
3364 EXPORT_SYMBOL(lfsck_start);
3365
3366 int lfsck_stop(const struct lu_env *env, struct dt_device *key,
3367                struct lfsck_stop *stop)
3368 {
3369         struct lfsck_instance   *lfsck;
3370         struct ptlrpc_thread    *thread;
3371         int                      rc     = 0;
3372         int                      rc1    = 0;
3373         ENTRY;
3374
3375         lfsck = lfsck_instance_find(key, true, false);
3376         if (unlikely(lfsck == NULL))
3377                 RETURN(-ENXIO);
3378
3379         thread = &lfsck->li_thread;
3380         if (stop && stop->ls_flags & LPF_BROADCAST && !lfsck->li_master) {
3381                 CERROR("%s: only allow to specify '-A' via MDS\n",
3382                        lfsck_lfsck2name(lfsck));
3383                 GOTO(put, rc = -EPERM);
3384         }
3385
3386         spin_lock(&lfsck->li_lock);
3387         /* The target is umounted */
3388         if (stop && stop->ls_status == LS_PAUSED)
3389                 lfsck->li_stopping = 1;
3390
3391         if (thread_is_init(thread) || thread_is_stopped(thread))
3392                 /* no error if LFSCK stopped already, or not started */
3393                 GOTO(unlock, rc = 0);
3394
3395         if (thread_is_stopping(thread))
3396                 /* Someone is stopping LFSCK. */
3397                 GOTO(unlock, rc = -EINPROGRESS);
3398
3399         if (stop) {
3400                 lfsck->li_status = stop->ls_status;
3401                 lfsck->li_flags = stop->ls_flags;
3402         } else {
3403                 lfsck->li_status = LS_STOPPED;
3404                 lfsck->li_flags = 0;
3405         }
3406
3407         thread_set_flags(thread, SVC_STOPPING);
3408
3409         LASSERT(lfsck->li_task != NULL);
3410         cfs_force_sig(SIGINT, lfsck->li_task);
3411
3412         if (lfsck->li_master) {
3413                 struct lfsck_component *com;
3414                 struct lfsck_assistant_data *lad;
3415
3416                 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
3417                         lad = com->lc_data;
3418                         spin_lock(&lad->lad_lock);
3419                         if (lad->lad_task != NULL)
3420                                 cfs_force_sig(SIGINT, lad->lad_task);
3421                         spin_unlock(&lad->lad_lock);
3422                 }
3423
3424                 list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
3425                         lad = com->lc_data;
3426                         spin_lock(&lad->lad_lock);
3427                         if (lad->lad_task != NULL)
3428                                 cfs_force_sig(SIGINT, lad->lad_task);
3429                         spin_unlock(&lad->lad_lock);
3430                 }
3431         }
3432
3433         wake_up_all(&thread->t_ctl_waitq);
3434         spin_unlock(&lfsck->li_lock);
3435         if (stop && stop->ls_flags & LPF_BROADCAST)
3436                 rc1 = lfsck_stop_all(env, lfsck, stop);
3437
3438         /* It was me set the status as 'stopping' just now, if it is not
3439          * 'stopping' now, then either stopped, or re-started by race. */
3440         wait_event_idle(thread->t_ctl_waitq,
3441                         !thread_is_stopping(thread));
3442
3443         GOTO(put, rc = 0);
3444
3445 unlock:
3446         spin_unlock(&lfsck->li_lock);
3447 put:
3448         lfsck_instance_put(env, lfsck);
3449
3450         return rc != 0 ? rc : rc1;
3451 }
3452 EXPORT_SYMBOL(lfsck_stop);
3453
3454 int lfsck_in_notify_local(const struct lu_env *env, struct dt_device *key,
3455                           struct lfsck_req_local *lrl, struct thandle *th)
3456 {
3457         struct lfsck_instance *lfsck;
3458         struct lfsck_component *com;
3459         int rc = -EOPNOTSUPP;
3460         ENTRY;
3461
3462         lfsck = lfsck_instance_find(key, true, false);
3463         if (unlikely(!lfsck))
3464                 RETURN(-ENXIO);
3465
3466         com = lfsck_component_find(lfsck, lrl->lrl_active);
3467         if (likely(com && com->lc_ops->lfsck_in_notify_local)) {
3468                 rc = com->lc_ops->lfsck_in_notify_local(env, com, lrl, th);
3469                 lfsck_component_put(env, com);
3470         }
3471
3472         lfsck_instance_put(env, lfsck);
3473
3474         RETURN(rc);
3475 }
3476 EXPORT_SYMBOL(lfsck_in_notify_local);
3477
3478 int lfsck_in_notify(const struct lu_env *env, struct dt_device *key,
3479                     struct lfsck_request *lr)
3480 {
3481         int rc = -EOPNOTSUPP;
3482         ENTRY;
3483
3484         switch (lr->lr_event) {
3485         case LE_START: {
3486                 struct lfsck_start       *start = &lfsck_env_info(env)->lti_start;
3487                 struct lfsck_start_param  lsp;
3488
3489                 memset(start, 0, sizeof(*start));
3490                 start->ls_valid = lr->lr_valid;
3491                 start->ls_speed_limit = lr->lr_speed;
3492                 start->ls_version = lr->lr_version;
3493                 start->ls_active = lr->lr_active;
3494                 start->ls_flags = lr->lr_param & ~LPF_BROADCAST;
3495                 start->ls_async_windows = lr->lr_async_windows;
3496
3497                 lsp.lsp_start = start;
3498                 lsp.lsp_index = lr->lr_index;
3499                 lsp.lsp_index_valid = 1;
3500                 rc = lfsck_start(env, key, &lsp);
3501                 break;
3502         }
3503         case LE_STOP: {
3504                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
3505
3506                 memset(stop, 0, sizeof(*stop));
3507                 stop->ls_status = lr->lr_status;
3508                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
3509                 rc = lfsck_stop(env, key, stop);
3510                 break;
3511         }
3512         case LE_PHASE1_DONE:
3513         case LE_PHASE2_DONE:
3514         case LE_PEER_EXIT:
3515         case LE_CONDITIONAL_DESTROY:
3516         case LE_SET_LMV_MASTER:
3517         case LE_SET_LMV_SLAVE:
3518         case LE_PAIRS_VERIFY: {
3519                 struct lfsck_instance  *lfsck;
3520                 struct lfsck_component *com;
3521
3522                 lfsck = lfsck_instance_find(key, true, false);
3523                 if (unlikely(lfsck == NULL))
3524                         RETURN(-ENXIO);
3525
3526                 com = lfsck_component_find(lfsck, lr->lr_active);
3527                 if (likely(com)) {
3528                         rc = com->lc_ops->lfsck_in_notify(env, com, lr);
3529                         lfsck_component_put(env, com);
3530                 }
3531
3532                 lfsck_instance_put(env, lfsck);
3533                 break;
3534         }
3535         default:
3536                 break;
3537         }
3538
3539         RETURN(rc);
3540 }
3541 EXPORT_SYMBOL(lfsck_in_notify);
3542
3543 int lfsck_query(const struct lu_env *env, struct dt_device *key,
3544                 struct lfsck_request *req, struct lfsck_reply *rep,
3545                 struct lfsck_query *que)
3546 {
3547         struct lfsck_instance  *lfsck;
3548         struct lfsck_component *com;
3549         int                     i;
3550         int                     rc = 0;
3551         __u16                   type;
3552         ENTRY;
3553
3554         lfsck = lfsck_instance_find(key, true, false);
3555         if (unlikely(lfsck == NULL))
3556                 RETURN(-ENXIO);
3557
3558         if (que != NULL) {
3559                 if (que->lu_types == LFSCK_TYPES_ALL)
3560                         que->lu_types =
3561                                 LFSCK_TYPES_SUPPORTED & ~LFSCK_TYPE_SCRUB;
3562
3563                 if (que->lu_types & ~LFSCK_TYPES_SUPPORTED) {
3564                         que->lu_types &= ~LFSCK_TYPES_SUPPORTED;
3565
3566                         GOTO(out, rc = -ENOTSUPP);
3567                 }
3568
3569                 for (i = 0, type = BIT(i); i < LFSCK_TYPE_BITS;
3570                      i++, type = BIT(i)) {
3571                         if (!(que->lu_types & type))
3572                                 continue;
3573
3574 again:
3575                         com = lfsck_component_find(lfsck, type);
3576                         if (unlikely(com == NULL))
3577                                 GOTO(out, rc = -ENOTSUPP);
3578
3579                         memset(que->lu_mdts_count[i], 0,
3580                                sizeof(__u32) * (LS_MAX + 1));
3581                         memset(que->lu_osts_count[i], 0,
3582                                sizeof(__u32) * (LS_MAX + 1));
3583                         que->lu_repaired[i] = 0;
3584                         rc = com->lc_ops->lfsck_query(env, com, req, rep,
3585                                                       que, i);
3586                         lfsck_component_put(env, com);
3587                         if  (rc < 0)
3588                                 GOTO(out, rc);
3589                 }
3590
3591                 if (!(que->lu_flags & LPF_WAIT))
3592                         GOTO(out, rc);
3593
3594                 for (i = 0, type = BIT(i); i < LFSCK_TYPE_BITS;
3595                      i++, type = BIT(i)) {
3596                         if (!(que->lu_types & type))
3597                                 continue;
3598
3599                         if (que->lu_mdts_count[i][LS_SCANNING_PHASE1] != 0 ||
3600                             que->lu_mdts_count[i][LS_SCANNING_PHASE2] != 0 ||
3601                             que->lu_osts_count[i][LS_SCANNING_PHASE1] != 0 ||
3602                             que->lu_osts_count[i][LS_SCANNING_PHASE2] != 0) {
3603                                 /* If it is required to wait, then sleep
3604                                  * 3 seconds and try to query again.
3605                                  */
3606                                 unsigned long timeout =
3607                                         msecs_to_jiffies(3000) + 1;
3608                                 while (timeout &&
3609                                        !fatal_signal_pending(current))
3610                                         timeout = schedule_timeout_killable(
3611                                                 timeout);
3612                                 if (timeout == 0)
3613                                         goto again;
3614                         }
3615                 }
3616         } else {
3617                 com = lfsck_component_find(lfsck, req->lr_active);
3618                 if (likely(com != NULL)) {
3619                         rc = com->lc_ops->lfsck_query(env, com, req, rep,
3620                                                       que, -1);
3621                         lfsck_component_put(env, com);
3622                 } else {
3623                         rc = -ENOTSUPP;
3624                 }
3625         }
3626
3627         GOTO(out, rc);
3628
3629 out:
3630         lfsck_instance_put(env, lfsck);
3631         return rc;
3632 }
3633 EXPORT_SYMBOL(lfsck_query);
3634
3635 int lfsck_register_namespace(const struct lu_env *env, struct dt_device *key,
3636                              struct ldlm_namespace *ns)
3637 {
3638         struct lfsck_instance  *lfsck;
3639         int                     rc      = -ENXIO;
3640
3641         lfsck = lfsck_instance_find(key, true, false);
3642         if (likely(lfsck != NULL)) {
3643                 lfsck->li_namespace = ns;
3644                 lfsck_instance_put(env, lfsck);
3645                 rc = 0;
3646         }
3647
3648         return rc;
3649 }
3650 EXPORT_SYMBOL(lfsck_register_namespace);
3651
3652 int lfsck_register(const struct lu_env *env, struct dt_device *key,
3653                    struct dt_device *next, struct obd_device *obd,
3654                    lfsck_out_notify notify, void *notify_data, bool master)
3655 {
3656         struct lfsck_instance   *lfsck;
3657         struct dt_object        *root  = NULL;
3658         struct dt_object        *obj   = NULL;
3659         struct lu_fid           *fid   = &lfsck_env_info(env)->lti_fid;
3660         int                      rc;
3661         ENTRY;
3662
3663         lfsck = lfsck_instance_find(key, false, false);
3664         if (unlikely(lfsck != NULL))
3665                 RETURN(-EEXIST);
3666
3667         OBD_ALLOC_PTR(lfsck);
3668         if (lfsck == NULL)
3669                 RETURN(-ENOMEM);
3670
3671         mutex_init(&lfsck->li_mutex);
3672         spin_lock_init(&lfsck->li_lock);
3673         INIT_LIST_HEAD(&lfsck->li_link);
3674         INIT_LIST_HEAD(&lfsck->li_list_scan);
3675         INIT_LIST_HEAD(&lfsck->li_list_dir);
3676         INIT_LIST_HEAD(&lfsck->li_list_double_scan);
3677         INIT_LIST_HEAD(&lfsck->li_list_idle);
3678         INIT_LIST_HEAD(&lfsck->li_list_lmv);
3679         atomic_set(&lfsck->li_ref, 1);
3680         atomic_set(&lfsck->li_double_scan_count, 0);
3681         init_waitqueue_head(&lfsck->li_thread.t_ctl_waitq);
3682         lfsck->li_out_notify = notify;
3683         lfsck->li_out_notify_data = notify_data;
3684         lfsck->li_next = next;
3685         lfsck->li_bottom = key;
3686         lfsck->li_obd = obd;
3687
3688         rc = lfsck_tgt_descs_init(&lfsck->li_ost_descs);
3689         if (rc != 0)
3690                 GOTO(out, rc);
3691
3692         rc = lfsck_tgt_descs_init(&lfsck->li_mdt_descs);
3693         if (rc != 0)
3694                 GOTO(out, rc);
3695
3696         fid->f_seq = FID_SEQ_LOCAL_NAME;
3697         fid->f_oid = 1;
3698         fid->f_ver = 0;
3699         rc = local_oid_storage_init(env, key, fid, &lfsck->li_los);
3700         if (rc != 0)
3701                 GOTO(out, rc);
3702
3703         rc = dt_root_get(env, key, fid);
3704         if (rc != 0)
3705                 GOTO(out, rc);
3706
3707         root = dt_locate(env, key, fid);
3708         if (IS_ERR(root))
3709                 GOTO(out, rc = PTR_ERR(root));
3710
3711         lfsck->li_local_root_fid = *fid;
3712         if (master) {
3713                 lfsck->li_master = 1;
3714                 if (lfsck_dev_idx(lfsck) == 0) {
3715                         struct lu_fid *pfid = &lfsck_env_info(env)->lti_fid2;
3716                         const struct lu_name *cname;
3717
3718                         rc = dt_lookup_dir(env, root, "ROOT",
3719                                            &lfsck->li_global_root_fid);
3720                         if (rc != 0)
3721                                 GOTO(out, rc);
3722
3723                         obj = dt_locate(env, key, &lfsck->li_global_root_fid);
3724                         if (IS_ERR(obj))
3725                                 GOTO(out, rc = PTR_ERR(obj));
3726
3727                         rc = dt_lookup_dir(env, obj, dotlustre, fid);
3728                         if (rc != 0)
3729                                 GOTO(out, rc);
3730
3731                         lfsck_object_put(env, obj);
3732                         obj = dt_locate(env, key, fid);
3733                         if (IS_ERR(obj))
3734                                 GOTO(out, rc = PTR_ERR(obj));
3735
3736                         cname = lfsck_name_get_const(env, dotlustre,
3737                                                      strlen(dotlustre));
3738                         rc = lfsck_verify_linkea(env, obj, cname,
3739                                                  &lfsck->li_global_root_fid);
3740                         if (rc != 0)
3741                                 GOTO(out, rc);
3742
3743                         if (unlikely(!dt_try_as_dir(env, obj)))
3744                                 GOTO(out, rc = -ENOTDIR);
3745
3746                         *pfid = *fid;
3747                         rc = dt_lookup_dir(env, obj, lostfound, fid);
3748                         if (rc != 0)
3749                                 GOTO(out, rc);
3750
3751                         lfsck_object_put(env, obj);
3752                         obj = dt_locate(env, key, fid);
3753                         if (IS_ERR(obj))
3754                                 GOTO(out, rc = PTR_ERR(obj));
3755
3756                         cname = lfsck_name_get_const(env, lostfound,
3757                                                      strlen(lostfound));
3758                         rc = lfsck_verify_linkea(env, obj, cname, pfid);
3759                         if (rc != 0)
3760                                 GOTO(out, rc);
3761
3762                         lfsck_object_put(env, obj);
3763                         obj = NULL;
3764                 }
3765         }
3766
3767         fid->f_seq = FID_SEQ_LOCAL_FILE;
3768         fid->f_oid = OTABLE_IT_OID;
3769         fid->f_ver = 0;
3770         obj = dt_locate(env, key, fid);
3771         if (IS_ERR(obj))
3772                 GOTO(out, rc = PTR_ERR(obj));
3773
3774         rc = obj->do_ops->do_index_try(env, obj, &dt_otable_features);
3775         if (rc != 0)
3776                 GOTO(out, rc);
3777
3778         lfsck->li_obj_oit = obj;
3779         obj = local_file_find_or_create(env, lfsck->li_los, root, LFSCK_DIR,
3780                                         S_IFDIR | S_IRUGO | S_IWUSR);
3781         if (IS_ERR(obj))
3782                 GOTO(out, rc = PTR_ERR(obj));
3783
3784         lu_object_get(&obj->do_lu);
3785         lfsck->li_lfsck_dir = obj;
3786         rc = lfsck_bookmark_setup(env, lfsck);
3787         if (rc != 0)
3788                 GOTO(out, rc);
3789
3790         if (master) {
3791                 rc = lfsck_fid_init(lfsck);
3792                 if (rc < 0)
3793                         GOTO(out, rc);
3794
3795                 rc = lfsck_namespace_setup(env, lfsck);
3796                 if (rc < 0)
3797                         GOTO(out, rc);
3798         }
3799
3800         rc = lfsck_layout_setup(env, lfsck);
3801         if (rc < 0)
3802                 GOTO(out, rc);
3803
3804         /* XXX: more LFSCK components initialization to be added here. */
3805
3806         rc = lfsck_instance_add(lfsck);
3807         if (rc == 0)
3808                 rc = lfsck_add_target_from_orphan(env, lfsck);
3809 out:
3810         if (obj != NULL && !IS_ERR(obj))
3811                 lfsck_object_put(env, obj);
3812         if (root != NULL && !IS_ERR(root))
3813                 lfsck_object_put(env, root);
3814         if (rc != 0)
3815                 lfsck_instance_cleanup(env, lfsck);
3816         return rc;
3817 }
3818 EXPORT_SYMBOL(lfsck_register);
3819
3820 void lfsck_degister(const struct lu_env *env, struct dt_device *key)
3821 {
3822         struct lfsck_instance *lfsck;
3823
3824         lfsck = lfsck_instance_find(key, false, true);
3825         if (lfsck != NULL)
3826                 lfsck_instance_put(env, lfsck);
3827 }
3828 EXPORT_SYMBOL(lfsck_degister);
3829
3830 int lfsck_add_target(const struct lu_env *env, struct dt_device *key,
3831                      struct dt_device *tgt, struct obd_export *exp,
3832                      __u32 index, bool for_ost)
3833 {
3834         struct lfsck_instance   *lfsck;
3835         struct lfsck_tgt_desc   *ltd;
3836         int                      rc;
3837         ENTRY;
3838
3839         OBD_ALLOC_PTR(ltd);
3840         if (ltd == NULL)
3841                 RETURN(-ENOMEM);
3842
3843         ltd->ltd_tgt = tgt;
3844         ltd->ltd_key = key;
3845         ltd->ltd_exp = exp;
3846         INIT_LIST_HEAD(&ltd->ltd_orphan_list);
3847         INIT_LIST_HEAD(&ltd->ltd_layout_list);
3848         INIT_LIST_HEAD(&ltd->ltd_layout_phase_list);
3849         INIT_LIST_HEAD(&ltd->ltd_namespace_list);
3850         INIT_LIST_HEAD(&ltd->ltd_namespace_phase_list);
3851         atomic_set(&ltd->ltd_ref, 1);
3852         ltd->ltd_index = index;
3853
3854         spin_lock(&lfsck_instance_lock);
3855         lfsck = __lfsck_instance_find(key, true, false);
3856         if (lfsck == NULL) {
3857                 if (for_ost)
3858                         list_add_tail(&ltd->ltd_orphan_list,
3859                                       &lfsck_ost_orphan_list);
3860                 else
3861                         list_add_tail(&ltd->ltd_orphan_list,
3862                                       &lfsck_mdt_orphan_list);
3863                 spin_unlock(&lfsck_instance_lock);
3864
3865                 RETURN(0);
3866         }
3867         spin_unlock(&lfsck_instance_lock);
3868
3869         rc = __lfsck_add_target(env, lfsck, ltd, for_ost, false);
3870         if (rc != 0)
3871                 lfsck_tgt_put(ltd);
3872
3873         lfsck_instance_put(env, lfsck);
3874
3875         RETURN(rc);
3876 }
3877 EXPORT_SYMBOL(lfsck_add_target);
3878
3879 void lfsck_del_target(const struct lu_env *env, struct dt_device *key,
3880                       struct dt_device *tgt, __u32 index, bool for_ost)
3881 {
3882         struct lfsck_instance   *lfsck;
3883         struct lfsck_tgt_descs  *ltds;
3884         struct lfsck_tgt_desc   *ltd;
3885         struct list_head        *head;
3886
3887         if (for_ost)
3888                 head = &lfsck_ost_orphan_list;
3889         else
3890                 head = &lfsck_mdt_orphan_list;
3891
3892         spin_lock(&lfsck_instance_lock);
3893         list_for_each_entry(ltd, head, ltd_orphan_list) {
3894                 if (ltd->ltd_tgt == tgt) {
3895                         list_del_init(&ltd->ltd_orphan_list);
3896                         spin_unlock(&lfsck_instance_lock);
3897                         lfsck_tgt_put(ltd);
3898
3899                         return;
3900                 }
3901         }
3902
3903         ltd = NULL;
3904         lfsck = __lfsck_instance_find(key, true, false);
3905         spin_unlock(&lfsck_instance_lock);
3906         if (unlikely(lfsck == NULL))
3907                 return;
3908
3909         if (for_ost)
3910                 ltds = &lfsck->li_ost_descs;
3911         else
3912                 ltds = &lfsck->li_mdt_descs;
3913
3914         down_write(&ltds->ltd_rw_sem);
3915         LASSERT(ltds->ltd_tgts_bitmap != NULL);
3916
3917         if (unlikely(index >= ltds->ltd_tgts_bitmap->size))
3918                 goto unlock;
3919
3920         ltd = lfsck_ltd2tgt(ltds, index);
3921         if (unlikely(ltd == NULL))
3922                 goto unlock;
3923
3924         LASSERT(ltds->ltd_tgtnr > 0);
3925
3926         ltds->ltd_tgtnr--;
3927         cfs_bitmap_clear(ltds->ltd_tgts_bitmap, index);
3928         lfsck_assign_tgt(ltds, NULL, index);
3929
3930 unlock:
3931         if (ltd == NULL) {
3932                 if (for_ost)
3933                         head = &lfsck->li_ost_descs.ltd_orphan;
3934                 else
3935                         head = &lfsck->li_mdt_descs.ltd_orphan;
3936
3937                 list_for_each_entry(ltd, head, ltd_orphan_list) {
3938                         if (ltd->ltd_tgt == tgt) {
3939                                 list_del_init(&ltd->ltd_orphan_list);
3940                                 break;
3941                         }
3942                 }
3943         }
3944
3945         up_write(&ltds->ltd_rw_sem);
3946         if (ltd != NULL) {
3947                 spin_lock(&ltds->ltd_lock);
3948                 ltd->ltd_dead = 1;
3949                 spin_unlock(&ltds->ltd_lock);
3950                 lfsck_stop_notify(env, lfsck, ltds, ltd, LFSCK_TYPE_NAMESPACE);
3951                 lfsck_stop_notify(env, lfsck, ltds, ltd, LFSCK_TYPE_LAYOUT);
3952                 lfsck_tgt_put(ltd);
3953         }
3954
3955         lfsck_instance_put(env, lfsck);
3956 }
3957 EXPORT_SYMBOL(lfsck_del_target);
3958
3959 static int __init lfsck_init(void)
3960 {
3961         int rc;
3962
3963         lfsck_key_init_generic(&lfsck_thread_key, NULL);
3964         rc = lu_context_key_register(&lfsck_thread_key);
3965         if (!rc) {
3966                 tgt_register_lfsck_in_notify_local(lfsck_in_notify_local);
3967                 tgt_register_lfsck_in_notify(lfsck_in_notify);
3968                 tgt_register_lfsck_query(lfsck_query);
3969         }
3970
3971         return rc;
3972 }
3973
3974 static void __exit lfsck_exit(void)
3975 {
3976         struct lfsck_tgt_desc *ltd;
3977         struct lfsck_tgt_desc *next;
3978
3979         LASSERT(list_empty(&lfsck_instance_list));
3980
3981         list_for_each_entry_safe(ltd, next, &lfsck_ost_orphan_list,
3982                                  ltd_orphan_list) {
3983                 list_del_init(&ltd->ltd_orphan_list);
3984                 lfsck_tgt_put(ltd);
3985         }
3986
3987         list_for_each_entry_safe(ltd, next, &lfsck_mdt_orphan_list,
3988                                  ltd_orphan_list) {
3989                 list_del_init(&ltd->ltd_orphan_list);
3990                 lfsck_tgt_put(ltd);
3991         }
3992
3993         lu_context_key_degister(&lfsck_thread_key);
3994 }
3995
3996 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
3997 MODULE_DESCRIPTION("Lustre File System Checker");
3998 MODULE_VERSION(LUSTRE_VERSION_STRING);
3999 MODULE_LICENSE("GPL");
4000
4001 module_init(lfsck_init);
4002 module_exit(lfsck_exit);