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 <lu_object.h>
  37 #include <dt_object.h>
  38 #include <md_object.h>
  39 #include <lustre_fld.h>
  40 #include <lustre_lib.h>
  41 #include <lustre_net.h>
  42 #include <lustre_lfsck.h>
  43
  44 #include "lfsck_internal.h"
  45
  46 #define LFSCK_CHECKPOINT_SKIP   1
  47
  48 /* define lfsck thread key */
  49 LU_KEY_INIT(lfsck, struct lfsck_thread_info);
  50
  51 static void lfsck_key_fini(const struct lu_context *ctx,
  52                            struct lu_context_key *key, void *data)
  53 {
  54         struct lfsck_thread_info *info = data;
  55
  56         lu_buf_free(&info->lti_linkea_buf);
  57         lu_buf_free(&info->lti_linkea_buf2);
  58         lu_buf_free(&info->lti_big_buf);
  59         OBD_FREE_PTR(info);
  60 }
  61
  62 LU_CONTEXT_KEY_DEFINE(lfsck, LCT_MD_THREAD | LCT_DT_THREAD);
  63 LU_KEY_INIT_GENERIC(lfsck);
  64
  65 static struct list_head lfsck_instance_list;
  66 static struct list_head lfsck_ost_orphan_list;
  67 static struct list_head lfsck_mdt_orphan_list;
  68 static DEFINE_SPINLOCK(lfsck_instance_lock);
  69
  70 const char *lfsck_flags_names[] = {
  71         "scanned-once",
  72         "inconsistent",
  73         "upgrade",
  74         "incomplete",
  75         "crashed_lastid",
  76         NULL
  77 };
  78
  79 const char *lfsck_param_names[] = {
  80         NULL,
  81         "failout",
  82         "dryrun",
  83         "all_targets",
  84         "broadcast",
  85         "orphan",
  86         "create_ostobj",
  87         "create_mdtobj",
  88         NULL,
  89         "delay_create_ostobj",
  90         NULL
  91 };
  92
  93 enum lfsck_verify_lpf_types {
  94         LVLT_BY_BOOKMARK        = 0,
  95         LVLT_BY_NAMEENTRY       = 1,
  96 };
  97
  98 static inline void
  99 lfsck_reset_ltd_status(struct lfsck_tgt_desc *ltd, enum lfsck_type type)
 100 {
 101         if (type == LFSCK_TYPE_LAYOUT) {
 102                 ltd->ltd_layout_status = LS_MAX;
 103                 ltd->ltd_layout_repaired = 0;
 104         } else {
 105                 ltd->ltd_namespace_status = LS_MAX;
 106                 ltd->ltd_namespace_repaired = 0;
 107         }
 108 }
 109
 110 static int lfsck_tgt_descs_init(struct lfsck_tgt_descs *ltds)
 111 {
 112         spin_lock_init(&ltds->ltd_lock);
 113         init_rwsem(&ltds->ltd_rw_sem);
 114         INIT_LIST_HEAD(&ltds->ltd_orphan);
 115         ltds->ltd_tgts_bitmap = CFS_ALLOCATE_BITMAP(BITS_PER_LONG);
 116         if (ltds->ltd_tgts_bitmap == NULL)
 117                 return -ENOMEM;
 118
 119         return 0;
 120 }
 121
 122 static void lfsck_tgt_descs_fini(struct lfsck_tgt_descs *ltds)
 123 {
 124         struct lfsck_tgt_desc   *ltd;
 125         struct lfsck_tgt_desc   *next;
 126         int                      idx;
 127
 128         down_write(&ltds->ltd_rw_sem);
 129
 130         list_for_each_entry_safe(ltd, next, &ltds->ltd_orphan,
 131                                  ltd_orphan_list) {
 132                 list_del_init(&ltd->ltd_orphan_list);
 133                 lfsck_tgt_put(ltd);
 134         }
 135
 136         if (unlikely(ltds->ltd_tgts_bitmap == NULL)) {
 137                 up_write(&ltds->ltd_rw_sem);
 138
 139                 return;
 140         }
 141
 142         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
 143                 ltd = lfsck_ltd2tgt(ltds, idx);
 144                 if (likely(ltd != NULL)) {
 145                         LASSERT(list_empty(&ltd->ltd_layout_list));
 146                         LASSERT(list_empty(&ltd->ltd_layout_phase_list));
 147                         LASSERT(list_empty(&ltd->ltd_namespace_list));
 148                         LASSERT(list_empty(&ltd->ltd_namespace_phase_list));
 149
 150                         ltds->ltd_tgtnr--;
 151                         cfs_bitmap_clear(ltds->ltd_tgts_bitmap, idx);
 152                         lfsck_assign_tgt(ltds, NULL, idx);
 153                         lfsck_tgt_put(ltd);
 154                 }
 155         }
 156
 157         LASSERTF(ltds->ltd_tgtnr == 0, "tgt count unmatched: %d\n",
 158                  ltds->ltd_tgtnr);
 159
 160         for (idx = 0; idx < TGT_PTRS; idx++) {
 161                 if (ltds->ltd_tgts_idx[idx] != NULL) {
 162                         OBD_FREE_PTR(ltds->ltd_tgts_idx[idx]);
 163                         ltds->ltd_tgts_idx[idx] = NULL;
 164                 }
 165         }
 166
 167         CFS_FREE_BITMAP(ltds->ltd_tgts_bitmap);
 168         ltds->ltd_tgts_bitmap = NULL;
 169         up_write(&ltds->ltd_rw_sem);
 170 }
 171
 172 static int __lfsck_add_target(const struct lu_env *env,
 173                               struct lfsck_instance *lfsck,
 174                               struct lfsck_tgt_desc *ltd,
 175                               bool for_ost, bool locked)
 176 {
 177         struct lfsck_tgt_descs *ltds;
 178         __u32                   index = ltd->ltd_index;
 179         int                     rc    = 0;
 180         ENTRY;
 181
 182         if (for_ost)
 183                 ltds = &lfsck->li_ost_descs;
 184         else
 185                 ltds = &lfsck->li_mdt_descs;
 186
 187         if (!locked)
 188                 down_write(&ltds->ltd_rw_sem);
 189
 190         LASSERT(ltds->ltd_tgts_bitmap != NULL);
 191
 192         if (index >= ltds->ltd_tgts_bitmap->size) {
 193                 __u32 newsize = max((__u32)ltds->ltd_tgts_bitmap->size,
 194                                     (__u32)BITS_PER_LONG);
 195                 struct cfs_bitmap *old_bitmap = ltds->ltd_tgts_bitmap;
 196                 struct cfs_bitmap *new_bitmap;
 197
 198                 while (newsize < index + 1)
 199                         newsize <<= 1;
 200
 201                 new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
 202                 if (new_bitmap == NULL)
 203                         GOTO(unlock, rc = -ENOMEM);
 204
 205                 if (ltds->ltd_tgtnr > 0)
 206                         cfs_bitmap_copy(new_bitmap, old_bitmap);
 207                 ltds->ltd_tgts_bitmap = new_bitmap;
 208                 CFS_FREE_BITMAP(old_bitmap);
 209         }
 210
 211         if (cfs_bitmap_check(ltds->ltd_tgts_bitmap, index)) {
 212                 CERROR("%s: the device %s (%u) is registered already\n",
 213                        lfsck_lfsck2name(lfsck),
 214                        ltd->ltd_tgt->dd_lu_dev.ld_obd->obd_name, index);
 215                 GOTO(unlock, rc = -EEXIST);
 216         }
 217
 218         if (ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 219                 OBD_ALLOC_PTR(ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK]);
 220                 if (ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK] == NULL)
 221                         GOTO(unlock, rc = -ENOMEM);
 222         }
 223
 224         lfsck_assign_tgt(ltds, ltd, index);
 225         cfs_bitmap_set(ltds->ltd_tgts_bitmap, index);
 226         ltds->ltd_tgtnr++;
 227
 228         GOTO(unlock, rc = 0);
 229
 230 unlock:
 231         if (!locked)
 232                 up_write(&ltds->ltd_rw_sem);
 233
 234         return rc;
 235 }
 236
 237 static int lfsck_add_target_from_orphan(const struct lu_env *env,
 238                                         struct lfsck_instance *lfsck)
 239 {
 240         struct lfsck_tgt_descs  *ltds    = &lfsck->li_ost_descs;
 241         struct lfsck_tgt_desc   *ltd;
 242         struct lfsck_tgt_desc   *next;
 243         struct list_head        *head    = &lfsck_ost_orphan_list;
 244         int                      rc;
 245         bool                     for_ost = true;
 246
 247 again:
 248         spin_lock(&lfsck_instance_lock);
 249         list_for_each_entry_safe(ltd, next, head, ltd_orphan_list) {
 250                 if (ltd->ltd_key == lfsck->li_bottom)
 251                         list_move_tail(&ltd->ltd_orphan_list,
 252                                        &ltds->ltd_orphan);
 253         }
 254         spin_unlock(&lfsck_instance_lock);
 255
 256         down_write(&ltds->ltd_rw_sem);
 257         while (!list_empty(&ltds->ltd_orphan)) {
 258                 ltd = list_entry(ltds->ltd_orphan.next,
 259                                  struct lfsck_tgt_desc,
 260                                  ltd_orphan_list);
 261                 list_del_init(&ltd->ltd_orphan_list);
 262                 rc = __lfsck_add_target(env, lfsck, ltd, for_ost, true);
 263                 /* Do not hold the semaphore for too long time. */
 264                 up_write(&ltds->ltd_rw_sem);
 265                 if (rc != 0)
 266                         return rc;
 267
 268                 down_write(&ltds->ltd_rw_sem);
 269         }
 270         up_write(&ltds->ltd_rw_sem);
 271
 272         if (for_ost) {
 273                 ltds = &lfsck->li_mdt_descs;
 274                 head = &lfsck_mdt_orphan_list;
 275                 for_ost = false;
 276                 goto again;
 277         }
 278
 279         return 0;
 280 }
 281
 282 static inline struct lfsck_component *
 283 __lfsck_component_find(struct lfsck_instance *lfsck, __u16 type,
 284                        struct list_head *list)
 285 {
 286         struct lfsck_component *com;
 287
 288         list_for_each_entry(com, list, lc_link) {
 289                 if (com->lc_type == type)
 290                         return com;
 291         }
 292         return NULL;
 293 }
 294
 295 struct lfsck_component *
 296 lfsck_component_find(struct lfsck_instance *lfsck, __u16 type)
 297 {
 298         struct lfsck_component *com;
 299
 300         spin_lock(&lfsck->li_lock);
 301         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_scan);
 302         if (com != NULL)
 303                 goto unlock;
 304
 305         com = __lfsck_component_find(lfsck, type,
 306                                      &lfsck->li_list_double_scan);
 307         if (com != NULL)
 308                 goto unlock;
 309
 310         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_idle);
 311
 312 unlock:
 313         if (com != NULL)
 314                 lfsck_component_get(com);
 315         spin_unlock(&lfsck->li_lock);
 316         return com;
 317 }
 318
 319 void lfsck_component_cleanup(const struct lu_env *env,
 320                              struct lfsck_component *com)
 321 {
 322         if (!list_empty(&com->lc_link))
 323                 list_del_init(&com->lc_link);
 324         if (!list_empty(&com->lc_link_dir))
 325                 list_del_init(&com->lc_link_dir);
 326
 327         lfsck_component_put(env, com);
 328 }
 329
 330 int lfsck_fid_alloc(const struct lu_env *env, struct lfsck_instance *lfsck,
 331                     struct lu_fid *fid, bool locked)
 332 {
 333         struct lfsck_bookmark   *bk = &lfsck->li_bookmark_ram;
 334         int                      rc = 0;
 335         ENTRY;
 336
 337         if (!locked)
 338                 mutex_lock(&lfsck->li_mutex);
 339
 340         rc = seq_client_alloc_fid(env, lfsck->li_seq, fid);
 341         if (rc >= 0) {
 342                 bk->lb_last_fid = *fid;
 343                 /* We do not care about whether the subsequent sub-operations
 344                  * failed or not. The worst case is that one FID is lost that
 345                  * is not a big issue for the LFSCK since it is relative rare
 346                  * for LFSCK create. */
 347                 rc = lfsck_bookmark_store(env, lfsck);
 348         }
 349
 350         if (!locked)
 351                 mutex_unlock(&lfsck->li_mutex);
 352
 353         RETURN(rc);
 354 }
 355
 356 static int __lfsck_ibits_lock(const struct lu_env *env,
 357                               struct lfsck_instance *lfsck,
 358                               struct dt_object *obj, struct ldlm_res_id *resid,
 359                               struct lustre_handle *lh, __u64 bits,
 360                               enum ldlm_mode mode)
 361 {
 362         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 363         union ldlm_policy_data          *policy = &info->lti_policy;
 364         __u64                            flags  = LDLM_FL_ATOMIC_CB;
 365         int                              rc;
 366
 367         LASSERT(lfsck->li_namespace != NULL);
 368
 369         memset(policy, 0, sizeof(*policy));
 370         policy->l_inodebits.bits = bits;
 371         if (dt_object_remote(obj)) {
 372                 struct ldlm_enqueue_info *einfo = &info->lti_einfo;
 373
 374                 memset(einfo, 0, sizeof(*einfo));
 375                 einfo->ei_type = LDLM_IBITS;
 376                 einfo->ei_mode = mode;
 377                 einfo->ei_cb_bl = ldlm_blocking_ast;
 378                 einfo->ei_cb_cp = ldlm_completion_ast;
 379                 einfo->ei_res_id = resid;
 380
 381                 rc = dt_object_lock(env, obj, lh, einfo, policy);
 382         } else {
 383                 rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid,
 384                                             LDLM_IBITS, policy, mode,
 385                                             &flags, ldlm_blocking_ast,
 386                                             ldlm_completion_ast, NULL, NULL,
 387                                             0, LVB_T_NONE, NULL, lh);
 388         }
 389
 390         if (rc == ELDLM_OK) {
 391                 rc = 0;
 392         } else {
 393                 memset(lh, 0, sizeof(*lh));
 394                 rc = -EIO;
 395         }
 396
 397         return rc;
 398 }
 399
 400 /**
 401  * Request the specified ibits lock for the given object.
 402  *
 403  * Before the LFSCK modifying on the namespace visible object,
 404  * it needs to acquire related ibits ldlm lock.
 405  *
 406  * \param[in] env       pointer to the thread context
 407  * \param[in] lfsck     pointer to the lfsck instance
 408  * \param[in] obj       pointer to the dt_object to be locked
 409  * \param[out] lh       pointer to the lock handle
 410  * \param[in] bits      the bits for the ldlm lock to be acquired
 411  * \param[in] mode      the mode for the ldlm lock to be acquired
 412  *
 413  * \retval              0 for success
 414  * \retval              negative error number on failure
 415  */
 416 int lfsck_ibits_lock(const struct lu_env *env, struct lfsck_instance *lfsck,
 417                      struct dt_object *obj, struct lustre_handle *lh,
 418                      __u64 bits, enum ldlm_mode mode)
 419 {
 420         struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid;
 421
 422         LASSERT(!lustre_handle_is_used(lh));
 423
 424         fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
 425         return __lfsck_ibits_lock(env, lfsck, obj, resid, lh, bits, mode);
 426 }
 427
 428 /**
 429  * Release the the specified ibits lock.
 430  *
 431  * If the lock has been acquired before, release it
 432  * and cleanup the handle. Otherwise, do nothing.
 433  *
 434  * \param[in] lh        pointer to the lock handle
 435  * \param[in] mode      the mode for the ldlm lock to be released
 436  */
 437 void lfsck_ibits_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
 438 {
 439         if (lustre_handle_is_used(lh)) {
 440                 ldlm_lock_decref(lh, mode);
 441                 memset(lh, 0, sizeof(*lh));
 442         }
 443 }
 444
 445 /**
 446  * Request compound ibits locks for the given <obj, name> pairs.
 447  *
 448  * Before the LFSCK modifying on the namespace visible object, it needs to
 449  * acquire related ibits ldlm lock. Usually, we can use lfsck_ibits_lock for
 450  * the lock purpose. But the simple lfsck_ibits_lock for directory-based
 451  * modificationis (such as insert name entry to the directory) may be too
 452  * coarse-grained and not efficient.
 453  *
 454  * The lfsck_lock() will request compound ibits locks on the specified
 455  * <obj, name> pairs: the PDO (Parallel Directory Operations) ibits (UPDATE)
 456  * lock on the directory object, and the regular ibits lock on the name hash.
 457  *
 458  * \param[in] env       pointer to the thread context
 459  * \param[in] lfsck     pointer to the lfsck instance
 460  * \param[in] obj       pointer to the dt_object to be locked
 461  * \param[in] name      used for building the PDO lock resource
 462  * \param[out] llh      pointer to the lfsck_lock_handle
 463  * \param[in] bits      the bits for the ldlm lock to be acquired
 464  * \param[in] mode      the mode for the ldlm lock to be acquired
 465  *
 466  * \retval              0 for success
 467  * \retval              negative error number on failure
 468  */
 469 int lfsck_lock(const struct lu_env *env, struct lfsck_instance *lfsck,
 470                struct dt_object *obj, const char *name,
 471                struct lfsck_lock_handle *llh, __u64 bits, enum ldlm_mode mode)
 472 {
 473         struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid;
 474         int                 rc;
 475
 476         LASSERT(S_ISDIR(lfsck_object_type(obj)));
 477         LASSERT(name != NULL);
 478         LASSERT(name[0] != 0);
 479         LASSERT(!lustre_handle_is_used(&llh->llh_pdo_lh));
 480         LASSERT(!lustre_handle_is_used(&llh->llh_reg_lh));
 481
 482         switch (mode) {
 483         case LCK_EX:
 484                 llh->llh_pdo_mode = LCK_EX;
 485                 break;
 486         case LCK_PW:
 487                 llh->llh_pdo_mode = LCK_CW;
 488                 break;
 489         case LCK_PR:
 490                 llh->llh_pdo_mode = LCK_CR;
 491                 break;
 492         default:
 493                 CDEBUG(D_LFSCK, "%s: unexpected PDO lock mode %u on the obj "
 494                        DFID"\n", lfsck_lfsck2name(lfsck), mode,
 495                        PFID(lfsck_dto2fid(obj)));
 496                 LBUG();
 497         }
 498
 499         fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
 500         rc = __lfsck_ibits_lock(env, lfsck, obj, resid, &llh->llh_pdo_lh,
 501                                 MDS_INODELOCK_UPDATE, llh->llh_pdo_mode);
 502         if (rc != 0)
 503                 return rc;
 504
 505         llh->llh_reg_mode = mode;
 506         resid->name[LUSTRE_RES_ID_HSH_OFF] = ll_full_name_hash(NULL, name,
 507                                                                strlen(name));
 508         LASSERT(resid->name[LUSTRE_RES_ID_HSH_OFF] != 0);
 509         rc = __lfsck_ibits_lock(env, lfsck, obj, resid, &llh->llh_reg_lh,
 510                                 bits, llh->llh_reg_mode);
 511         if (rc != 0)
 512                 lfsck_ibits_unlock(&llh->llh_pdo_lh, llh->llh_pdo_mode);
 513
 514         return rc;
 515 }
 516
 517 /**
 518  * Release the the compound ibits locks.
 519  *
 520  * \param[in] llh       pointer to the lfsck_lock_handle to be released
 521  */
 522 void lfsck_unlock(struct lfsck_lock_handle *llh)
 523 {
 524         lfsck_ibits_unlock(&llh->llh_reg_lh, llh->llh_reg_mode);
 525         lfsck_ibits_unlock(&llh->llh_pdo_lh, llh->llh_pdo_mode);
 526 }
 527
 528 int lfsck_find_mdt_idx_by_fid(const struct lu_env *env,
 529                               struct lfsck_instance *lfsck,
 530                               const struct lu_fid *fid)
 531 {
 532         struct seq_server_site  *ss     = lfsck_dev_site(lfsck);
 533         struct lu_seq_range     *range  = &lfsck_env_info(env)->lti_range;
 534         int                      rc;
 535
 536         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
 537                 /* "ROOT" is always on the MDT0. */
 538                 if (lu_fid_eq(fid, &lfsck->li_global_root_fid))
 539                         return 0;
 540
 541                 return lfsck_dev_idx(lfsck);
 542         }
 543
 544         fld_range_set_mdt(range);
 545         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
 546         if (rc == 0)
 547                 rc = range->lsr_index;
 548
 549         return rc;
 550 }
 551
 552 const char dot[] = ".";
 553 const char dotdot[] = "..";
 554 static const char dotlustre[] = ".lustre";
 555 static const char lostfound[] = "lost+found";
 556
 557 /**
 558  * Remove the name entry from the .lustre/lost+found directory.
 559  *
 560  * No need to care about the object referenced by the name entry,
 561  * either the name entry is invalid or redundant, or the referenced
 562  * object has been processed or will be handled by others.
 563  *
 564  * \param[in] env       pointer to the thread context
 565  * \param[in] lfsck     pointer to the lfsck instance
 566  * \param[in] name      the name for the name entry to be removed
 567  *
 568  * \retval              0 for success
 569  * \retval              negative error number on failure
 570  */
 571 static int lfsck_lpf_remove_name_entry(const struct lu_env *env,
 572                                        struct lfsck_instance *lfsck,
 573                                        const char *name)
 574 {
 575         struct dt_object        *parent = lfsck->li_lpf_root_obj;
 576         struct dt_device        *dev    = lfsck_obj2dev(parent);
 577         struct thandle          *th;
 578         struct lfsck_lock_handle *llh   = &lfsck_env_info(env)->lti_llh;
 579         int                      rc;
 580         ENTRY;
 581
 582         rc = lfsck_lock(env, lfsck, parent, name, llh,
 583                         MDS_INODELOCK_UPDATE, LCK_PW);
 584         if (rc != 0)
 585                 RETURN(rc);
 586
 587         th = dt_trans_create(env, dev);
 588         if (IS_ERR(th))
 589                 GOTO(unlock, rc = PTR_ERR(th));
 590
 591         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
 592         if (rc != 0)
 593                 GOTO(stop, rc);
 594
 595         rc = dt_declare_ref_del(env, parent, th);
 596         if (rc != 0)
 597                 GOTO(stop, rc);
 598
 599         rc = dt_trans_start_local(env, dev, th);
 600         if (rc != 0)
 601                 GOTO(stop, rc);
 602
 603         rc = dt_delete(env, parent, (const struct dt_key *)name, th);
 604         if (rc != 0)
 605                 GOTO(stop, rc);
 606
 607         dt_write_lock(env, parent, 0);
 608         rc = dt_ref_del(env, parent, th);
 609         dt_write_unlock(env, parent);
 610
 611         GOTO(stop, rc);
 612
 613 stop:
 614         dt_trans_stop(env, dev, th);
 615
 616 unlock:
 617         lfsck_unlock(llh);
 618
 619         CDEBUG(D_LFSCK, "%s: remove name entry "DFID"/%s: rc = %d\n",
 620                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(parent)), name, rc);
 621
 622         return rc;
 623 }
 624
 625 static int lfsck_create_lpf_local(const struct lu_env *env,
 626                                   struct lfsck_instance *lfsck,
 627                                   struct dt_object *child,
 628                                   struct lu_attr *la,
 629                                   struct dt_object_format *dof,
 630                                   const char *name)
 631 {
 632         struct dt_insert_rec    *rec    = &lfsck_env_info(env)->lti_dt_rec;
 633         struct dt_object        *parent = lfsck->li_lpf_root_obj;
 634         struct dt_device        *dev    = lfsck_obj2dev(child);
 635         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
 636         struct dt_object        *bk_obj = lfsck->li_bookmark_obj;
 637         const struct lu_fid     *cfid   = lfsck_dto2fid(child);
 638         struct thandle          *th     = NULL;
 639         struct linkea_data       ldata  = { NULL };
 640         struct lu_buf            linkea_buf;
 641         const struct lu_name    *cname;
 642         loff_t                   pos    = 0;
 643         int                      len    = sizeof(struct lfsck_bookmark);
 644         int                      rc;
 645         ENTRY;
 646
 647         cname = lfsck_name_get_const(env, name, strlen(name));
 648         rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf2,
 649                               cname, lfsck_dto2fid(parent));
 650         if (rc != 0)
 651                 RETURN(rc);
 652
 653         th = dt_trans_create(env, dev);
 654         if (IS_ERR(th))
 655                 RETURN(PTR_ERR(th));
 656
 657         /* 1a. create child */
 658         rc = dt_declare_create(env, child, la, NULL, dof, th);
 659         if (rc != 0)
 660                 GOTO(stop, rc);
 661
 662         if (!dt_try_as_dir(env, child))
 663                 GOTO(stop, rc = -ENOTDIR);
 664
 665         /* 2a. increase child nlink */
 666         rc = dt_declare_ref_add(env, child, th);
 667         if (rc != 0)
 668                 GOTO(stop, rc);
 669
 670         /* 3a. insert dot into child dir */
 671         rec->rec_type = S_IFDIR;
 672         rec->rec_fid = cfid;
 673         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
 674                                (const struct dt_key *)dot, th);
 675         if (rc != 0)
 676                 GOTO(stop, rc);
 677
 678         /* 4a. insert dotdot into child dir */
 679         rec->rec_fid = &LU_LPF_FID;
 680         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
 681                                (const struct dt_key *)dotdot, th);
 682         if (rc != 0)
 683                 GOTO(stop, rc);
 684
 685         /* 5a. insert linkEA for child */
 686         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
 687                        ldata.ld_leh->leh_len);
 688         rc = dt_declare_xattr_set(env, child, &linkea_buf,
 689                                   XATTR_NAME_LINK, 0, th);
 690         if (rc != 0)
 691                 GOTO(stop, rc);
 692
 693         /* 6a. insert name into parent dir */
 694         rec->rec_type = S_IFDIR;
 695         rec->rec_fid = cfid;
 696         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
 697                                (const struct dt_key *)name, th);
 698         if (rc != 0)
 699                 GOTO(stop, rc);
 700
 701         /* 7a. increase parent nlink */
 702         rc = dt_declare_ref_add(env, parent, th);
 703         if (rc != 0)
 704                 GOTO(stop, rc);
 705
 706         /* 8a. update bookmark */
 707         rc = dt_declare_record_write(env, bk_obj,
 708                                      lfsck_buf_get(env, bk, len), 0, th);
 709         if (rc != 0)
 710                 GOTO(stop, rc);
 711
 712         rc = dt_trans_start_local(env, dev, th);
 713         if (rc != 0)
 714                 GOTO(stop, rc);
 715
 716         dt_write_lock(env, child, 0);
 717         /* 1b. create child */
 718         rc = dt_create(env, child, la, NULL, dof, th);
 719         if (rc != 0)
 720                 GOTO(unlock, rc);
 721
 722         /* 2b. increase child nlink */
 723         rc = dt_ref_add(env, child, th);
 724         if (rc != 0)
 725                 GOTO(unlock, rc);
 726
 727         /* 3b. insert dot into child dir */
 728         rec->rec_fid = cfid;
 729         rc = dt_insert(env, child, (const struct dt_rec *)rec,
 730                        (const struct dt_key *)dot, th, 1);
 731         if (rc != 0)
 732                 GOTO(unlock, rc);
 733
 734         /* 4b. insert dotdot into child dir */
 735         rec->rec_fid = &LU_LPF_FID;
 736         rc = dt_insert(env, child, (const struct dt_rec *)rec,
 737                        (const struct dt_key *)dotdot, th, 1);
 738         if (rc != 0)
 739                 GOTO(unlock, rc);
 740
 741         /* 5b. insert linkEA for child. */
 742         rc = dt_xattr_set(env, child, &linkea_buf,
 743                           XATTR_NAME_LINK, 0, th);
 744         dt_write_unlock(env, child);
 745         if (rc != 0)
 746                 GOTO(stop, rc);
 747
 748         /* 6b. insert name into parent dir */
 749         rec->rec_fid = cfid;
 750         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
 751                        (const struct dt_key *)name, th, 1);
 752         if (rc != 0)
 753                 GOTO(stop, rc);
 754
 755         dt_write_lock(env, parent, 0);
 756         /* 7b. increase parent nlink */
 757         rc = dt_ref_add(env, parent, th);
 758         dt_write_unlock(env, parent);
 759         if (rc != 0)
 760                 GOTO(stop, rc);
 761
 762         bk->lb_lpf_fid = *cfid;
 763         lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk);
 764
 765         /* 8b. update bookmark */
 766         rc = dt_record_write(env, bk_obj,
 767                              lfsck_buf_get(env, bk, len), &pos, th);
 768
 769         GOTO(stop, rc);
 770
 771 unlock:
 772         dt_write_unlock(env, child);
 773
 774 stop:
 775         dt_trans_stop(env, dev, th);
 776
 777         return rc;
 778 }
 779
 780 static int lfsck_create_lpf_remote(const struct lu_env *env,
 781                                    struct lfsck_instance *lfsck,
 782                                    struct dt_object *child,
 783                                    struct lu_attr *la,
 784                                    struct dt_object_format *dof,
 785                                    const char *name)
 786 {
 787         struct dt_insert_rec    *rec    = &lfsck_env_info(env)->lti_dt_rec;
 788         struct dt_object        *parent = lfsck->li_lpf_root_obj;
 789         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
 790         struct dt_object        *bk_obj = lfsck->li_bookmark_obj;
 791         const struct lu_fid     *cfid   = lfsck_dto2fid(child);
 792         struct thandle          *th     = NULL;
 793         struct linkea_data       ldata  = { NULL };
 794         struct lu_buf            linkea_buf;
 795         const struct lu_name    *cname;
 796         struct dt_device        *dev;
 797         loff_t                   pos    = 0;
 798         int                      len    = sizeof(struct lfsck_bookmark);
 799         int                      rc;
 800         ENTRY;
 801
 802         cname = lfsck_name_get_const(env, name, strlen(name));
 803         rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf2,
 804                               cname, lfsck_dto2fid(parent));
 805         if (rc != 0)
 806                 RETURN(rc);
 807
 808         /* Create .lustre/lost+found/MDTxxxx. */
 809
 810         /* XXX: Currently, cross-MDT create operation needs to create the child
 811          *      object firstly, then insert name into the parent directory. For
 812          *      this case, the child object resides on current MDT (local), but
 813          *      the parent ".lustre/lost+found" may be on remote MDT. It is not
 814          *      easy to contain all the sub-modifications orderly within single
 815          *      transaction.
 816          *
 817          *      To avoid more inconsistency, we split the create operation into
 818          *      two transactions:
 819          *
 820          *      1) create the child and update the lfsck_bookmark::lb_lpf_fid
 821          *         locally.
 822          *      2) insert the name "MDTXXXX" in the parent ".lustre/lost+found"
 823          *         remotely.
 824          *
 825          *      If 1) done, but 2) failed, then go ahead, the LFSCK will try to
 826          *      repair such inconsistency when LFSCK run next time. */
 827
 828         /* Transaction I: locally */
 829
 830         dev = lfsck_obj2dev(child);
 831         th = dt_trans_create(env, dev);
 832         if (IS_ERR(th))
 833                 RETURN(PTR_ERR(th));
 834
 835         /* 1a. create child */
 836         rc = dt_declare_create(env, child, la, NULL, dof, th);
 837         if (rc != 0)
 838                 GOTO(stop, rc);
 839
 840         if (!dt_try_as_dir(env, child))
 841                 GOTO(stop, rc = -ENOTDIR);
 842
 843         /* 2a. increase child nlink */
 844         rc = dt_declare_ref_add(env, child, th);
 845         if (rc != 0)
 846                 GOTO(stop, rc);
 847
 848         /* 3a. insert dot into child dir */
 849         rec->rec_type = S_IFDIR;
 850         rec->rec_fid = cfid;
 851         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
 852                                (const struct dt_key *)dot, th);
 853         if (rc != 0)
 854                 GOTO(stop, rc);
 855
 856         /* 4a. insert dotdot into child dir */
 857         rec->rec_fid = &LU_LPF_FID;
 858         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
 859                                (const struct dt_key *)dotdot, th);
 860         if (rc != 0)
 861                 GOTO(stop, rc);
 862
 863         /* 5a. insert linkEA for child */
 864         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
 865                        ldata.ld_leh->leh_len);
 866         rc = dt_declare_xattr_set(env, child, &linkea_buf,
 867                                   XATTR_NAME_LINK, 0, th);
 868         if (rc != 0)
 869                 GOTO(stop, rc);
 870
 871         /* 6a. update bookmark */
 872         rc = dt_declare_record_write(env, bk_obj,
 873                                      lfsck_buf_get(env, bk, len), 0, th);
 874         if (rc != 0)
 875                 GOTO(stop, rc);
 876
 877         rc = dt_trans_start_local(env, dev, th);
 878         if (rc != 0)
 879                 GOTO(stop, rc);
 880
 881         dt_write_lock(env, child, 0);
 882         /* 1b. create child */
 883         rc = dt_create(env, child, la, NULL, dof, th);
 884         if (rc != 0)
 885                 GOTO(unlock, rc);
 886
 887         /* 2b. increase child nlink */
 888         rc = dt_ref_add(env, child, th);
 889         if (rc != 0)
 890                 GOTO(unlock, rc);
 891
 892         /* 3b. insert dot into child dir */
 893         rec->rec_type = S_IFDIR;
 894         rec->rec_fid = cfid;
 895         rc = dt_insert(env, child, (const struct dt_rec *)rec,
 896                        (const struct dt_key *)dot, th, 1);
 897         if (rc != 0)
 898                 GOTO(unlock, rc);
 899
 900         /* 4b. insert dotdot into child dir */
 901         rec->rec_fid = &LU_LPF_FID;
 902         rc = dt_insert(env, child, (const struct dt_rec *)rec,
 903                        (const struct dt_key *)dotdot, th, 1);
 904         if (rc != 0)
 905                 GOTO(unlock, rc);
 906
 907         /* 5b. insert linkEA for child */
 908         rc = dt_xattr_set(env, child, &linkea_buf,
 909                           XATTR_NAME_LINK, 0, th);
 910         if (rc != 0)
 911                 GOTO(unlock, rc);
 912
 913         bk->lb_lpf_fid = *cfid;
 914         lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk);
 915
 916         /* 6b. update bookmark */
 917         rc = dt_record_write(env, bk_obj,
 918                              lfsck_buf_get(env, bk, len), &pos, th);
 919
 920         dt_write_unlock(env, child);
 921         dt_trans_stop(env, dev, th);
 922         if (rc != 0)
 923                 RETURN(rc);
 924
 925         /* Transaction II: remotely */
 926
 927         dev = lfsck_obj2dev(parent);
 928         th = dt_trans_create(env, dev);
 929         if (IS_ERR(th))
 930                 RETURN(PTR_ERR(th));
 931
 932         th->th_sync = 1;
 933         /* 5a. insert name into parent dir */
 934         rec->rec_fid = cfid;
 935         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
 936                                (const struct dt_key *)name, th);
 937         if (rc != 0)
 938                 GOTO(stop, rc);
 939
 940         /* 6a. increase parent nlink */
 941         rc = dt_declare_ref_add(env, parent, th);
 942         if (rc != 0)
 943                 GOTO(stop, rc);
 944
 945         rc = dt_trans_start_local(env, dev, th);
 946         if (rc != 0)
 947                 GOTO(stop, rc);
 948
 949         /* 5b. insert name into parent dir */
 950         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
 951                        (const struct dt_key *)name, th, 1);
 952         if (rc != 0)
 953                 GOTO(stop, rc);
 954
 955         dt_write_lock(env, parent, 0);
 956         /* 6b. increase parent nlink */
 957         rc = dt_ref_add(env, parent, th);
 958         dt_write_unlock(env, parent);
 959
 960         GOTO(stop, rc);
 961
 962 unlock:
 963         dt_write_unlock(env, child);
 964 stop:
 965         dt_trans_stop(env, dev, th);
 966
 967         if (rc != 0 && dev == lfsck_obj2dev(parent))
 968                 CDEBUG(D_LFSCK, "%s: partially created the object "DFID
 969                        "for orphans, but failed to insert the name %s "
 970                        "to the .lustre/lost+found/. Such inconsistency "
 971                        "will be repaired when LFSCK run next time: rc = %d\n",
 972                        lfsck_lfsck2name(lfsck), PFID(cfid), name, rc);
 973
 974         return rc;
 975 }
 976
 977 /**
 978  * Create the MDTxxxx directory under /ROOT/.lustre/lost+found/
 979  *
 980  * The /ROOT/.lustre/lost+found/MDTxxxx/ directory is used for holding
 981  * orphans and other uncertain inconsistent objects found during the
 982  * LFSCK. Such directory will be created by the LFSCK engine on the
 983  * local MDT before the LFSCK scanning.
 984  *
 985  * \param[in] env       pointer to the thread context
 986  * \param[in] lfsck     pointer to the lfsck instance
 987  *
 988  * \retval              0 for success
 989  * \retval              negative error number on failure
 990  */
 991 static int lfsck_create_lpf(const struct lu_env *env,
 992                             struct lfsck_instance *lfsck)
 993 {
 994         struct lfsck_bookmark    *bk    = &lfsck->li_bookmark_ram;
 995         struct lfsck_thread_info *info  = lfsck_env_info(env);
 996         struct lu_fid            *cfid  = &info->lti_fid2;
 997         struct lu_attr           *la    = &info->lti_la;
 998         struct dt_object_format  *dof   = &info->lti_dof;
 999         struct dt_object         *parent = lfsck->li_lpf_root_obj;
1000         struct dt_object         *child = NULL;
1001         struct lfsck_lock_handle *llh   = &info->lti_llh;
1002         char                      name[8];
1003         int                       node  = lfsck_dev_idx(lfsck);
1004         int                       rc    = 0;
1005         ENTRY;
1006
1007         LASSERT(lfsck->li_master);
1008         LASSERT(parent != NULL);
1009         LASSERT(lfsck->li_lpf_obj == NULL);
1010
1011         snprintf(name, 8, "MDT%04x", node);
1012         rc = lfsck_lock(env, lfsck, parent, name, llh,
1013                         MDS_INODELOCK_UPDATE, LCK_PW);
1014         if (rc != 0)
1015                 RETURN(rc);
1016
1017         if (fid_is_zero(&bk->lb_lpf_fid)) {
1018                 /* There is corner case that: in former LFSCK scanning we have
1019                  * created the .lustre/lost+found/MDTxxxx but failed to update
1020                  * the lfsck_bookmark::lb_lpf_fid successfully. So need lookup
1021                  * it from MDT0 firstly. */
1022                 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1023                                (const struct dt_key *)name);
1024                 if (rc != 0 && rc != -ENOENT)
1025                         GOTO(unlock, rc);
1026
1027                 if (rc == 0) {
1028                         bk->lb_lpf_fid = *cfid;
1029                         rc = lfsck_bookmark_store(env, lfsck);
1030                 } else {
1031                         rc = lfsck_fid_alloc(env, lfsck, cfid, true);
1032                 }
1033                 if (rc != 0)
1034                         GOTO(unlock, rc);
1035         } else {
1036                 *cfid = bk->lb_lpf_fid;
1037         }
1038
1039         child = lfsck_object_find_bottom_new(env, lfsck, cfid);
1040         if (IS_ERR(child))
1041                 GOTO(unlock, rc = PTR_ERR(child));
1042
1043         if (dt_object_exists(child) != 0) {
1044                 if (unlikely(!dt_try_as_dir(env, child)))
1045                         rc = -ENOTDIR;
1046                 else
1047                         lfsck->li_lpf_obj = child;
1048
1049                 GOTO(unlock, rc);
1050         }
1051
1052         memset(la, 0, sizeof(*la));
1053         la->la_atime = la->la_mtime = la->la_ctime = ktime_get_real_seconds();
1054         la->la_mode = S_IFDIR | S_IRWXU;
1055         la->la_valid = LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1056                        LA_UID | LA_GID;
1057         memset(dof, 0, sizeof(*dof));
1058         dof->dof_type = dt_mode_to_dft(S_IFDIR);
1059
1060         if (node == 0)
1061                 rc = lfsck_create_lpf_local(env, lfsck, child, la, dof, name);
1062         else
1063                 rc = lfsck_create_lpf_remote(env, lfsck, child, la, dof, name);
1064         if (rc == 0)
1065                 lfsck->li_lpf_obj = child;
1066
1067         GOTO(unlock, rc);
1068
1069 unlock:
1070         lfsck_unlock(llh);
1071         if (rc != 0 && child != NULL && !IS_ERR(child))
1072                 lfsck_object_put(env, child);
1073
1074         return rc;
1075 }
1076
1077 /**
1078  * Scan .lustre/lost+found for bad name entries and remove them.
1079  *
1080  * The valid name entry should be "MDTxxxx", the "xxxx" is the MDT device
1081  * index in the system. Any other formatted name is invalid and should be
1082  * removed.
1083  *
1084  * \param[in] env       pointer to the thread context
1085  * \param[in] lfsck     pointer to the lfsck instance
1086  *
1087  * \retval              0 for success
1088  * \retval              negative error number on failure
1089  */
1090 static int lfsck_scan_lpf_bad_entries(const struct lu_env *env,
1091                                       struct lfsck_instance *lfsck)
1092 {
1093         struct dt_object        *parent = lfsck->li_lpf_root_obj;
1094         struct lu_dirent        *ent    =
1095                         (struct lu_dirent *)lfsck_env_info(env)->lti_key;
1096         const struct dt_it_ops  *iops   = &parent->do_index_ops->dio_it;
1097         struct dt_it            *it;
1098         int                      rc;
1099         ENTRY;
1100
1101         it = iops->init(env, parent, LUDA_64BITHASH);
1102         if (IS_ERR(it))
1103                 RETURN(PTR_ERR(it));
1104
1105         rc = iops->load(env, it, 0);
1106         if (rc == 0)
1107                 rc = iops->next(env, it);
1108         else if (rc > 0)
1109                 rc = 0;
1110
1111         while (rc == 0) {
1112                 int off = 3;
1113
1114                 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
1115                 if (rc != 0)
1116                         break;
1117
1118                 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
1119                 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
1120                         goto next;
1121
1122                 /* name length must be strlen("MDTxxxx") */
1123                 if (ent->lde_namelen != 7)
1124                         goto remove;
1125
1126                 if (memcmp(ent->lde_name, "MDT", off) != 0)
1127                         goto remove;
1128
1129                 while (off < 7 && isxdigit(ent->lde_name[off]))
1130                         off++;
1131
1132                 if (off != 7) {
1133
1134 remove:
1135                         rc = lfsck_lpf_remove_name_entry(env, lfsck,
1136                                                          ent->lde_name);
1137                         if (rc != 0)
1138                                 break;
1139                 }
1140
1141 next:
1142                 rc = iops->next(env, it);
1143         }
1144
1145         iops->put(env, it);
1146         iops->fini(env, it);
1147
1148         RETURN(rc > 0 ? 0 : rc);
1149 }
1150
1151 static int lfsck_update_lpf_entry(const struct lu_env *env,
1152                                   struct lfsck_instance *lfsck,
1153                                   struct dt_object *parent,
1154                                   struct dt_object *child,
1155                                   const char *name,
1156                                   enum lfsck_verify_lpf_types type)
1157 {
1158         int rc;
1159
1160         if (type == LVLT_BY_BOOKMARK) {
1161                 rc = lfsck_update_name_entry(env, lfsck, parent, name,
1162                                              lfsck_dto2fid(child), S_IFDIR);
1163         } else /* if (type == LVLT_BY_NAMEENTRY) */ {
1164                 lfsck->li_bookmark_ram.lb_lpf_fid = *lfsck_dto2fid(child);
1165                 rc = lfsck_bookmark_store(env, lfsck);
1166
1167                 CDEBUG(D_LFSCK, "%s: update LPF fid "DFID
1168                        " in the bookmark file: rc = %d\n",
1169                        lfsck_lfsck2name(lfsck),
1170                        PFID(lfsck_dto2fid(child)), rc);
1171         }
1172
1173         return rc;
1174 }
1175
1176 /**
1177  * Check whether the @child back references the @parent.
1178  *
1179  * Two cases:
1180  * 1) The child's FID is stored in the bookmark file. If the child back
1181  *    references the parent (LU_LPF_FID object) via its ".." entry, then
1182  *    insert the name (MDTxxxx) to the .lustre/lost+found; otherwise, if
1183  *    the child back references another parent2, then:
1184  * 1.1) If the parent2 recognizes the child, then update the bookmark file;
1185  * 1.2) Otherwise, the LFSCK cannot know whether there will be parent3 that
1186  *      references the child. So keep them there. As the LFSCK processing,
1187  *      the parent3 may be found, then when the LFSCK run next time, the
1188  *      inconsistency can be repaired.
1189  *
1190  * 2) The child's FID is stored in the .lustre/lost+found/ sub-directory name
1191  *    entry (MDTxxxx). If the child back references the parent (LU_LPF_FID obj)
1192  *    via its ".." entry, then update the bookmark file, otherwise, if the child
1193  *    back references another parent2, then:
1194  * 2.1) If the parent2 recognizes the child, then remove the sub-directory
1195  *      from .lustre/lost+found/;
1196  * 2.2) Otherwise, if the parent2 does not recognizes the child, trust the
1197  *      sub-directory name entry and update the child;
1198  * 2.3) Otherwise, if we do not know whether the parent2 recognizes the child
1199  *      or not, then keep them there.
1200  *
1201  * \param[in] env       pointer to the thread context
1202  * \param[in] lfsck     pointer to the lfsck instance
1203  * \param[in] child     pointer to the lost+found sub-directory object
1204  * \param[in] name      the name for lost+found sub-directory object
1205  * \param[out] fid      pointer to the buffer to hold the FID of the object
1206  *                      (called it as parent2) that is referenced via the
1207  *                      child's dotdot entry; it also can be the FID that
1208  *                      is referenced by the name entry under the parent2.
1209  * \param[in] type      to indicate where the child's FID is stored in
1210  *
1211  * \retval              positive number for uncertain inconsistency
1212  * \retval              0 for success
1213  * \retval              negative error number on failure
1214  */
1215 static int lfsck_verify_lpf_pairs(const struct lu_env *env,
1216                                   struct lfsck_instance *lfsck,
1217                                   struct dt_object *child, const char *name,
1218                                   struct lu_fid *fid,
1219                                   enum lfsck_verify_lpf_types type)
1220 {
1221         struct dt_object         *parent  = lfsck->li_lpf_root_obj;
1222         struct lfsck_thread_info *info    = lfsck_env_info(env);
1223         char                     *name2   = info->lti_key;
1224         struct lu_fid            *fid2    = &info->lti_fid3;
1225         struct dt_object         *parent2 = NULL;
1226         struct lustre_handle      lh      = { 0 };
1227         int                       rc;
1228         ENTRY;
1229
1230         fid_zero(fid);
1231         rc = dt_lookup(env, child, (struct dt_rec *)fid,
1232                        (const struct dt_key *)dotdot);
1233         if (rc != 0)
1234                 GOTO(linkea, rc);
1235
1236         if (!fid_is_sane(fid))
1237                 GOTO(linkea, rc = -EINVAL);
1238
1239         if (lu_fid_eq(fid, &LU_LPF_FID)) {
1240                 const struct lu_name *cname;
1241
1242                 if (lfsck->li_lpf_obj == NULL) {
1243                         lu_object_get(&child->do_lu);
1244                         lfsck->li_lpf_obj = child;
1245                 }
1246
1247                 cname = lfsck_name_get_const(env, name, strlen(name));
1248                 rc = lfsck_verify_linkea(env, child, cname, &LU_LPF_FID);
1249                 if (rc == 0)
1250                         rc = lfsck_update_lpf_entry(env, lfsck, parent, child,
1251                                                     name, type);
1252
1253                 GOTO(out_done, rc);
1254         }
1255
1256         parent2 = lfsck_object_find_bottom(env, lfsck, fid);
1257         if (IS_ERR(parent2))
1258                 GOTO(linkea, parent2);
1259
1260         if (!dt_object_exists(parent2)) {
1261                 lfsck_object_put(env, parent2);
1262
1263                 GOTO(linkea, parent2 = ERR_PTR(-ENOENT));
1264         }
1265
1266         if (!dt_try_as_dir(env, parent2)) {
1267                 lfsck_object_put(env, parent2);
1268
1269                 GOTO(linkea, parent2 = ERR_PTR(-ENOTDIR));
1270         }
1271
1272 linkea:
1273         /* To prevent rename/unlink race */
1274         rc = lfsck_ibits_lock(env, lfsck, child, &lh,
1275                               MDS_INODELOCK_UPDATE, LCK_PR);
1276         if (rc != 0)
1277                 GOTO(out_put, rc);
1278
1279         dt_read_lock(env, child, 0);
1280         rc = lfsck_links_get_first(env, child, name2, fid2);
1281         if (rc != 0) {
1282                 dt_read_unlock(env, child);
1283                 lfsck_ibits_unlock(&lh, LCK_PR);
1284
1285                 GOTO(out_put, rc = 1);
1286         }
1287
1288         /* It is almost impossible that the bookmark file (or the name entry)
1289          * and the linkEA hit the same data corruption. Trust the linkEA. */
1290         if (lu_fid_eq(fid2, &LU_LPF_FID) && strcmp(name, name2) == 0) {
1291                 dt_read_unlock(env, child);
1292                 lfsck_ibits_unlock(&lh, LCK_PR);
1293
1294                 *fid = *fid2;
1295                 if (lfsck->li_lpf_obj == NULL) {
1296                         lu_object_get(&child->do_lu);
1297                         lfsck->li_lpf_obj = child;
1298                 }
1299
1300                 /* Update the child's dotdot entry */
1301                 rc = lfsck_update_name_entry(env, lfsck, child, dotdot,
1302                                              &LU_LPF_FID, S_IFDIR);
1303                 if (rc == 0)
1304                         rc = lfsck_update_lpf_entry(env, lfsck, parent, child,
1305                                                     name, type);
1306
1307                 GOTO(out_put, rc);
1308         }
1309
1310         if (parent2 == NULL || IS_ERR(parent2)) {
1311                 dt_read_unlock(env, child);
1312                 lfsck_ibits_unlock(&lh, LCK_PR);
1313
1314                 GOTO(out_done, rc = 1);
1315         }
1316
1317         rc = dt_lookup(env, parent2, (struct dt_rec *)fid,
1318                        (const struct dt_key *)name2);
1319         dt_read_unlock(env, child);
1320         lfsck_ibits_unlock(&lh, LCK_PR);
1321         if (rc != 0 && rc != -ENOENT)
1322                 GOTO(out_put, rc);
1323
1324         if (rc == -ENOENT || !lu_fid_eq(fid, lfsck_dto2fid(child))) {
1325                 if (type == LVLT_BY_BOOKMARK)
1326                         GOTO(out_put, rc = 1);
1327
1328                 /* Trust the name entry, update the child's dotdot entry. */
1329                 rc = lfsck_update_name_entry(env, lfsck, child, dotdot,
1330                                              &LU_LPF_FID, S_IFDIR);
1331
1332                 GOTO(out_put, rc);
1333         }
1334
1335         if (type == LVLT_BY_BOOKMARK) {
1336                 /* Invalid FID record in the bookmark file, reset it. */
1337                 fid_zero(&lfsck->li_bookmark_ram.lb_lpf_fid);
1338                 rc = lfsck_bookmark_store(env, lfsck);
1339
1340                 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1341                        " in the bookmark file: rc = %d\n",
1342                        lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(child)), rc);
1343         } else /* if (type == LVLT_BY_NAMEENTRY) */ {
1344                 /* The name entry is wrong, remove it. */
1345                 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1346         }
1347
1348         GOTO(out_put, rc);
1349
1350 out_put:
1351         if (parent2 != NULL && !IS_ERR(parent2))
1352                 lfsck_object_put(env, parent2);
1353
1354 out_done:
1355         return rc;
1356 }
1357
1358 /**
1359  * Verify the /ROOT/.lustre/lost+found/ directory.
1360  *
1361  * /ROOT/.lustre/lost+found/ is a special directory to hold the objects that
1362  * the LFSCK does not exactly know how to handle, such as orphans. So before
1363  * the LFSCK scanning the system, the consistency of such directory needs to
1364  * be verified firstly to allow the users to use it during the LFSCK.
1365  *
1366  * \param[in] env       pointer to the thread context
1367  * \param[in] lfsck     pointer to the lfsck instance
1368  *
1369  * \retval              positive number for uncertain inconsistency
1370  * \retval              0 for success
1371  * \retval              negative error number on failure
1372  */
1373 int lfsck_verify_lpf(const struct lu_env *env, struct lfsck_instance *lfsck)
1374 {
1375         struct lfsck_thread_info *info   = lfsck_env_info(env);
1376         struct lu_fid            *pfid   = &info->lti_fid;
1377         struct lu_fid            *cfid   = &info->lti_fid2;
1378         struct lfsck_bookmark    *bk     = &lfsck->li_bookmark_ram;
1379         struct dt_object         *parent;
1380         /* child1's FID is in the bookmark file. */
1381         struct dt_object         *child1 = NULL;
1382         /* child2's FID is in the name entry MDTxxxx. */
1383         struct dt_object         *child2 = NULL;
1384         const struct lu_name     *cname;
1385         char                      name[8];
1386         int                       node   = lfsck_dev_idx(lfsck);
1387         int                       rc     = 0;
1388         ENTRY;
1389
1390         LASSERT(lfsck->li_master);
1391
1392         if (lfsck->li_lpf_root_obj != NULL)
1393                 RETURN(0);
1394
1395         if (node == 0) {
1396                 parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
1397                                                   &LU_LPF_FID);
1398         } else {
1399                 struct lfsck_tgt_desc *ltd;
1400
1401                 ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, 0);
1402                 if (unlikely(ltd == NULL))
1403                         RETURN(-ENXIO);
1404
1405                 parent = lfsck_object_find_by_dev(env, ltd->ltd_tgt,
1406                                                   &LU_LPF_FID);
1407                 lfsck_tgt_put(ltd);
1408         }
1409
1410         if (IS_ERR(parent))
1411                 RETURN(PTR_ERR(parent));
1412
1413         LASSERT(dt_object_exists(parent));
1414
1415         if (unlikely(!dt_try_as_dir(env, parent))) {
1416                 lfsck_object_put(env, parent);
1417
1418                 GOTO(put, rc = -ENOTDIR);
1419         }
1420
1421         lfsck->li_lpf_root_obj = parent;
1422         if (node == 0) {
1423                 rc = lfsck_scan_lpf_bad_entries(env, lfsck);
1424                 if (rc != 0)
1425                         CDEBUG(D_LFSCK, "%s: scan .lustre/lost+found/ "
1426                                "for bad sub-directories: rc = %d\n",
1427                                lfsck_lfsck2name(lfsck), rc);
1428         }
1429
1430         /* child2 */
1431         snprintf(name, 8, "MDT%04x", node);
1432         rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1433                        (const struct dt_key *)name);
1434         if (rc == -ENOENT) {
1435                 rc = 0;
1436                 goto find_child1;
1437         }
1438
1439         if (rc != 0)
1440                 GOTO(put, rc);
1441
1442         /* Invalid FID in the name entry, remove the name entry. */
1443         if (!fid_is_norm(cfid)) {
1444                 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1445                 if (rc != 0)
1446                         GOTO(put, rc);
1447
1448                 goto find_child1;
1449         }
1450
1451         child2 = lfsck_object_find_bottom(env, lfsck, cfid);
1452         if (IS_ERR(child2))
1453                 GOTO(put, rc = PTR_ERR(child2));
1454
1455         if (unlikely(!dt_object_exists(child2) ||
1456                      dt_object_remote(child2)) ||
1457                      !S_ISDIR(lfsck_object_type(child2))) {
1458                 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1459                 if (rc != 0)
1460                         GOTO(put, rc);
1461
1462                 goto find_child1;
1463         }
1464
1465         if (unlikely(!dt_try_as_dir(env, child2))) {
1466                 lfsck_object_put(env, child2);
1467                 child2 = NULL;
1468                 rc = -ENOTDIR;
1469         }
1470
1471 find_child1:
1472         if (fid_is_zero(&bk->lb_lpf_fid))
1473                 goto check_child2;
1474
1475         if (likely(lu_fid_eq(cfid, &bk->lb_lpf_fid))) {
1476                 if (lfsck->li_lpf_obj == NULL) {
1477                         lu_object_get(&child2->do_lu);
1478                         lfsck->li_lpf_obj = child2;
1479                 }
1480
1481                 cname = lfsck_name_get_const(env, name, strlen(name));
1482                 rc = lfsck_verify_linkea(env, child2, cname, &LU_LPF_FID);
1483
1484                 GOTO(put, rc);
1485         }
1486
1487         if (unlikely(!fid_is_norm(&bk->lb_lpf_fid))) {
1488                 struct lu_fid tfid = bk->lb_lpf_fid;
1489
1490                 /* Invalid FID record in the bookmark file, reset it. */
1491                 fid_zero(&bk->lb_lpf_fid);
1492                 rc = lfsck_bookmark_store(env, lfsck);
1493
1494                 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1495                        " in the bookmark file: rc = %d\n",
1496                        lfsck_lfsck2name(lfsck), PFID(&tfid), rc);
1497
1498                 if (rc != 0)
1499                         GOTO(put, rc);
1500
1501                 goto check_child2;
1502         }
1503
1504         child1 = lfsck_object_find_bottom(env, lfsck, &bk->lb_lpf_fid);
1505         if (IS_ERR(child1)) {
1506                 child1 = NULL;
1507                 goto check_child2;
1508         }
1509
1510         if (unlikely(!dt_object_exists(child1) ||
1511                      dt_object_remote(child1)) ||
1512                      !S_ISDIR(lfsck_object_type(child1))) {
1513                 /* Invalid FID record in the bookmark file, reset it. */
1514                 fid_zero(&bk->lb_lpf_fid);
1515                 rc = lfsck_bookmark_store(env, lfsck);
1516
1517                 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1518                        " in the bookmark file: rc = %d\n",
1519                        lfsck_lfsck2name(lfsck),
1520                        PFID(lfsck_dto2fid(child1)), rc);
1521
1522                 if (rc != 0)
1523                         GOTO(put, rc);
1524
1525                 lfsck_object_put(env, child1);
1526                 child1 = NULL;
1527                 goto check_child2;
1528         }
1529
1530         if (unlikely(!dt_try_as_dir(env, child1))) {
1531                 lfsck_object_put(env, child1);
1532                 child1 = NULL;
1533                 rc = -ENOTDIR;
1534                 goto check_child2;
1535         }
1536
1537         rc = lfsck_verify_lpf_pairs(env, lfsck, child1, name, pfid,
1538                                     LVLT_BY_BOOKMARK);
1539         if (lu_fid_eq(pfid, &LU_LPF_FID))
1540                 GOTO(put, rc);
1541
1542 check_child2:
1543         if (child2 != NULL)
1544                 rc = lfsck_verify_lpf_pairs(env, lfsck, child2, name,
1545                                             pfid, LVLT_BY_NAMEENTRY);
1546
1547         GOTO(put, rc);
1548
1549 put:
1550         if (lfsck->li_lpf_obj != NULL) {
1551                 if (unlikely(!dt_try_as_dir(env, lfsck->li_lpf_obj))) {
1552                         lfsck_object_put(env, lfsck->li_lpf_obj);
1553                         lfsck->li_lpf_obj = NULL;
1554                         rc = -ENOTDIR;
1555                 }
1556         } else if (rc == 0) {
1557                 rc = lfsck_create_lpf(env, lfsck);
1558         }
1559
1560         if (child2 != NULL && !IS_ERR(child2))
1561                 lfsck_object_put(env, child2);
1562         if (child1 != NULL && !IS_ERR(child1))
1563                 lfsck_object_put(env, child1);
1564
1565         return rc;
1566 }
1567
1568 static int lfsck_fid_init(struct lfsck_instance *lfsck)
1569 {
1570         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
1571         struct seq_server_site  *ss     = lfsck_dev_site(lfsck);
1572         char                    *prefix;
1573         int                      rc     = 0;
1574         ENTRY;
1575
1576         if (unlikely(ss == NULL))
1577                 RETURN(-ENXIO);
1578
1579         OBD_ALLOC_PTR(lfsck->li_seq);
1580         if (lfsck->li_seq == NULL)
1581                 RETURN(-ENOMEM);
1582
1583         OBD_ALLOC(prefix, MAX_OBD_NAME + 7);
1584         if (prefix == NULL)
1585                 GOTO(out, rc = -ENOMEM);
1586
1587         snprintf(prefix, MAX_OBD_NAME + 7, "lfsck-%s", lfsck_lfsck2name(lfsck));
1588         rc = seq_client_init(lfsck->li_seq, NULL, LUSTRE_SEQ_METADATA, prefix,
1589                              ss->ss_server_seq);
1590         OBD_FREE(prefix, MAX_OBD_NAME + 7);
1591         if (rc != 0)
1592                 GOTO(out, rc);
1593
1594         if (fid_is_sane(&bk->lb_last_fid))
1595                 lfsck->li_seq->lcs_fid = bk->lb_last_fid;
1596
1597         RETURN(0);
1598
1599 out:
1600         OBD_FREE_PTR(lfsck->li_seq);
1601         lfsck->li_seq = NULL;
1602
1603         return rc;
1604 }
1605
1606 static void lfsck_fid_fini(struct lfsck_instance *lfsck)
1607 {
1608         if (lfsck->li_seq != NULL) {
1609                 seq_client_fini(lfsck->li_seq);
1610                 OBD_FREE_PTR(lfsck->li_seq);
1611                 lfsck->li_seq = NULL;
1612         }
1613 }
1614
1615 void lfsck_instance_cleanup(const struct lu_env *env,
1616                             struct lfsck_instance *lfsck)
1617 {
1618         struct ptlrpc_thread    *thread = &lfsck->li_thread;
1619         struct lfsck_component  *com;
1620         struct lfsck_component  *next;
1621         struct lfsck_lmv_unit   *llu;
1622         struct lfsck_lmv_unit   *llu_next;
1623         struct lfsck_lmv        *llmv;
1624         ENTRY;
1625
1626         LASSERT(list_empty(&lfsck->li_link));
1627         LASSERT(thread_is_init(thread) || thread_is_stopped(thread));
1628
1629         if (lfsck->li_obj_oit != NULL) {
1630                 lfsck_object_put(env, lfsck->li_obj_oit);
1631                 lfsck->li_obj_oit = NULL;
1632         }
1633
1634         LASSERT(lfsck->li_obj_dir == NULL);
1635         LASSERT(lfsck->li_lmv == NULL);
1636
1637         list_for_each_entry_safe(llu, llu_next, &lfsck->li_list_lmv, llu_link) {
1638                 llmv = &llu->llu_lmv;
1639
1640                 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
1641                          "still in using: %u\n",
1642                          atomic_read(&llmv->ll_ref));
1643
1644                 lfsck_lmv_put(env, llmv);
1645         }
1646
1647         list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
1648                 lfsck_component_cleanup(env, com);
1649         }
1650
1651         LASSERT(list_empty(&lfsck->li_list_dir));
1652
1653         list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
1654                                  lc_link) {
1655                 lfsck_component_cleanup(env, com);
1656         }
1657
1658         list_for_each_entry_safe(com, next, &lfsck->li_list_idle, lc_link) {
1659                 lfsck_component_cleanup(env, com);
1660         }
1661
1662         lfsck_tgt_descs_fini(&lfsck->li_ost_descs);
1663         lfsck_tgt_descs_fini(&lfsck->li_mdt_descs);
1664
1665         if (lfsck->li_lfsck_dir != NULL) {
1666                 lfsck_object_put(env, lfsck->li_lfsck_dir);
1667                 lfsck->li_lfsck_dir = NULL;
1668         }
1669
1670         if (lfsck->li_bookmark_obj != NULL) {
1671                 lfsck_object_put(env, lfsck->li_bookmark_obj);
1672                 lfsck->li_bookmark_obj = NULL;
1673         }
1674
1675         if (lfsck->li_lpf_obj != NULL) {
1676                 lfsck_object_put(env, lfsck->li_lpf_obj);
1677                 lfsck->li_lpf_obj = NULL;
1678         }
1679
1680         if (lfsck->li_lpf_root_obj != NULL) {
1681                 lfsck_object_put(env, lfsck->li_lpf_root_obj);
1682                 lfsck->li_lpf_root_obj = NULL;
1683         }
1684
1685         if (lfsck->li_los != NULL) {
1686                 local_oid_storage_fini(env, lfsck->li_los);
1687                 lfsck->li_los = NULL;
1688         }
1689
1690         lfsck_fid_fini(lfsck);
1691
1692         OBD_FREE_PTR(lfsck);
1693 }
1694
1695 static inline struct lfsck_instance *
1696 __lfsck_instance_find(struct dt_device *key, bool ref, bool unlink)
1697 {
1698         struct lfsck_instance *lfsck;
1699
1700         list_for_each_entry(lfsck, &lfsck_instance_list, li_link) {
1701                 if (lfsck->li_bottom == key) {
1702                         if (ref)
1703                                 lfsck_instance_get(lfsck);
1704                         if (unlink)
1705                                 list_del_init(&lfsck->li_link);
1706
1707                         return lfsck;
1708                 }
1709         }
1710
1711         return NULL;
1712 }
1713
1714 struct lfsck_instance *lfsck_instance_find(struct dt_device *key, bool ref,
1715                                            bool unlink)
1716 {
1717         struct lfsck_instance *lfsck;
1718
1719         spin_lock(&lfsck_instance_lock);
1720         lfsck = __lfsck_instance_find(key, ref, unlink);
1721         spin_unlock(&lfsck_instance_lock);
1722
1723         return lfsck;
1724 }
1725
1726 static inline int lfsck_instance_add(struct lfsck_instance *lfsck)
1727 {
1728         struct lfsck_instance *tmp;
1729
1730         spin_lock(&lfsck_instance_lock);
1731         list_for_each_entry(tmp, &lfsck_instance_list, li_link) {
1732                 if (lfsck->li_bottom == tmp->li_bottom) {
1733                         spin_unlock(&lfsck_instance_lock);
1734                         return -EEXIST;
1735                 }
1736         }
1737
1738         list_add_tail(&lfsck->li_link, &lfsck_instance_list);
1739         spin_unlock(&lfsck_instance_lock);
1740         return 0;
1741 }
1742
1743 void lfsck_bits_dump(struct seq_file *m, int bits, const char *names[],
1744                      const char *prefix)
1745 {
1746         int flag;
1747         int i;
1748         bool newline = (bits != 0 ? false : true);
1749
1750         seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
1751
1752         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
1753                 if (flag & bits) {
1754                         bits &= ~flag;
1755                         if (names[i] != NULL) {
1756                                 if (bits == 0)
1757                                         newline = true;
1758
1759                                 seq_printf(m, "%s%c", names[i],
1760                                            newline ? '\n' : ',');
1761                         }
1762                 }
1763         }
1764
1765         if (!newline)
1766                 seq_putc(m, '\n');
1767 }
1768
1769 void lfsck_time_dump(struct seq_file *m, time64_t time, const char *name)
1770 {
1771         if (time == 0) {
1772                 seq_printf(m, "%s_time: N/A\n", name);
1773                 seq_printf(m, "time_since_%s: N/A\n", name);
1774         } else {
1775                 seq_printf(m, "%s_time: %lld\n", name, time);
1776                 seq_printf(m, "time_since_%s: %lld seconds\n",
1777                            name, ktime_get_real_seconds() - time);
1778         }
1779 }
1780
1781 void lfsck_pos_dump(struct seq_file *m, struct lfsck_position *pos,
1782                     const char *prefix)
1783 {
1784         if (fid_is_zero(&pos->lp_dir_parent)) {
1785                 if (pos->lp_oit_cookie == 0) {
1786                         seq_printf(m, "%s: N/A, N/A, N/A\n", prefix);
1787                         return;
1788                 }
1789                 seq_printf(m, "%s: %llu, N/A, N/A\n",
1790                            prefix, pos->lp_oit_cookie);
1791         } else {
1792                 seq_printf(m, "%s: %llu, "DFID", %#llx\n",
1793                            prefix, pos->lp_oit_cookie,
1794                            PFID(&pos->lp_dir_parent), pos->lp_dir_cookie);
1795         }
1796 }
1797
1798 void lfsck_pos_fill(const struct lu_env *env, struct lfsck_instance *lfsck,
1799                     struct lfsck_position *pos, bool init)
1800 {
1801         const struct dt_it_ops *iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
1802
1803         if (unlikely(lfsck->li_di_oit == NULL)) {
1804                 memset(pos, 0, sizeof(*pos));
1805                 return;
1806         }
1807
1808         pos->lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
1809         if (!lfsck->li_current_oit_processed && !init)
1810                 pos->lp_oit_cookie--;
1811
1812         if (unlikely(pos->lp_oit_cookie == 0))
1813                 pos->lp_oit_cookie = 1;
1814
1815         if (lfsck->li_di_dir != NULL) {
1816                 struct dt_object *dto = lfsck->li_obj_dir;
1817
1818                 pos->lp_dir_cookie = dto->do_index_ops->dio_it.store(env,
1819                                                         lfsck->li_di_dir);
1820
1821                 if (pos->lp_dir_cookie >= MDS_DIR_END_OFF) {
1822                         fid_zero(&pos->lp_dir_parent);
1823                         pos->lp_dir_cookie = 0;
1824                 } else {
1825                         pos->lp_dir_parent = *lfsck_dto2fid(dto);
1826                 }
1827         } else {
1828                 fid_zero(&pos->lp_dir_parent);
1829                 pos->lp_dir_cookie = 0;
1830         }
1831 }
1832
1833 bool __lfsck_set_speed(struct lfsck_instance *lfsck, __u32 limit)
1834 {
1835         bool dirty = false;
1836
1837         if (limit != LFSCK_SPEED_NO_LIMIT) {
1838                 if (limit > msecs_to_jiffies(MSEC_PER_SEC)) {
1839                         lfsck->li_sleep_rate = limit /
1840                                                msecs_to_jiffies(MSEC_PER_SEC);
1841                         lfsck->li_sleep_jif = 1;
1842                 } else {
1843                         lfsck->li_sleep_rate = 1;
1844                         lfsck->li_sleep_jif = msecs_to_jiffies(MSEC_PER_SEC) /
1845                                               limit;
1846                 }
1847         } else {
1848                 lfsck->li_sleep_jif = 0;
1849                 lfsck->li_sleep_rate = 0;
1850         }
1851
1852         if (lfsck->li_bookmark_ram.lb_speed_limit != limit) {
1853                 lfsck->li_bookmark_ram.lb_speed_limit = limit;
1854                 dirty = true;
1855         }
1856
1857         return dirty;
1858 }
1859
1860 void lfsck_control_speed(struct lfsck_instance *lfsck)
1861 {
1862         struct ptlrpc_thread *thread = &lfsck->li_thread;
1863         struct l_wait_info    lwi;
1864
1865         if (lfsck->li_sleep_jif > 0 &&
1866             lfsck->li_new_scanned >= lfsck->li_sleep_rate) {
1867                 lwi = LWI_TIMEOUT_INTR(lfsck->li_sleep_jif, NULL,
1868                                        LWI_ON_SIGNAL_NOOP, NULL);
1869
1870                 l_wait_event(thread->t_ctl_waitq,
1871                              !thread_is_running(thread),
1872                              &lwi);
1873                 lfsck->li_new_scanned = 0;
1874         }
1875 }
1876
1877 void lfsck_control_speed_by_self(struct lfsck_component *com)
1878 {
1879         struct lfsck_instance   *lfsck  = com->lc_lfsck;
1880         struct ptlrpc_thread    *thread = &lfsck->li_thread;
1881         struct l_wait_info       lwi;
1882
1883         if (lfsck->li_sleep_jif > 0 &&
1884             com->lc_new_scanned >= lfsck->li_sleep_rate) {
1885                 lwi = LWI_TIMEOUT_INTR(lfsck->li_sleep_jif, NULL,
1886                                        LWI_ON_SIGNAL_NOOP, NULL);
1887
1888                 l_wait_event(thread->t_ctl_waitq,
1889                              !thread_is_running(thread),
1890                              &lwi);
1891                 com->lc_new_scanned = 0;
1892         }
1893 }
1894
1895 static struct lfsck_thread_args *
1896 lfsck_thread_args_init(struct lfsck_instance *lfsck,
1897                        struct lfsck_component *com,
1898                        struct lfsck_start_param *lsp)
1899 {
1900         struct lfsck_thread_args *lta;
1901         int                       rc;
1902
1903         OBD_ALLOC_PTR(lta);
1904         if (lta == NULL)
1905                 return ERR_PTR(-ENOMEM);
1906
1907         rc = lu_env_init(&lta->lta_env, LCT_MD_THREAD | LCT_DT_THREAD);
1908         if (rc != 0) {
1909                 OBD_FREE_PTR(lta);
1910                 return ERR_PTR(rc);
1911         }
1912
1913         lta->lta_lfsck = lfsck_instance_get(lfsck);
1914         if (com != NULL)
1915                 lta->lta_com = lfsck_component_get(com);
1916
1917         lta->lta_lsp = lsp;
1918
1919         return lta;
1920 }
1921
1922 void lfsck_thread_args_fini(struct lfsck_thread_args *lta)
1923 {
1924         if (lta->lta_com != NULL)
1925                 lfsck_component_put(&lta->lta_env, lta->lta_com);
1926         lfsck_instance_put(&lta->lta_env, lta->lta_lfsck);
1927         lu_env_fini(&lta->lta_env);
1928         OBD_FREE_PTR(lta);
1929 }
1930
1931 struct lfsck_assistant_data *
1932 lfsck_assistant_data_init(struct lfsck_assistant_operations *lao,
1933                           const char *name)
1934 {
1935         struct lfsck_assistant_data *lad;
1936
1937         OBD_ALLOC_PTR(lad);
1938         if (lad != NULL) {
1939                 lad->lad_bitmap = CFS_ALLOCATE_BITMAP(BITS_PER_LONG);
1940                 if (lad->lad_bitmap == NULL) {
1941                         OBD_FREE_PTR(lad);
1942                         return NULL;
1943                 }
1944
1945                 INIT_LIST_HEAD(&lad->lad_req_list);
1946                 spin_lock_init(&lad->lad_lock);
1947                 INIT_LIST_HEAD(&lad->lad_ost_list);
1948                 INIT_LIST_HEAD(&lad->lad_ost_phase1_list);
1949                 INIT_LIST_HEAD(&lad->lad_ost_phase2_list);
1950                 INIT_LIST_HEAD(&lad->lad_mdt_list);
1951                 INIT_LIST_HEAD(&lad->lad_mdt_phase1_list);
1952                 INIT_LIST_HEAD(&lad->lad_mdt_phase2_list);
1953                 init_waitqueue_head(&lad->lad_thread.t_ctl_waitq);
1954                 lad->lad_ops = lao;
1955                 lad->lad_name = name;
1956         }
1957
1958         return lad;
1959 }
1960
1961 struct lfsck_assistant_object *
1962 lfsck_assistant_object_init(const struct lu_env *env, const struct lu_fid *fid,
1963                             const struct lu_attr *attr, __u64 cookie,
1964                             bool is_dir)
1965 {
1966         struct lfsck_assistant_object   *lso;
1967
1968         OBD_ALLOC_PTR(lso);
1969         if (lso == NULL)
1970                 return ERR_PTR(-ENOMEM);
1971
1972         lso->lso_fid = *fid;
1973         if (attr != NULL)
1974                 lso->lso_attr = *attr;
1975
1976         atomic_set(&lso->lso_ref, 1);
1977         lso->lso_oit_cookie = cookie;
1978         if (is_dir)
1979                 lso->lso_is_dir = 1;
1980
1981         return lso;
1982 }
1983
1984 struct dt_object *
1985 lfsck_assistant_object_load(const struct lu_env *env,
1986                             struct lfsck_instance *lfsck,
1987                             struct lfsck_assistant_object *lso)
1988 {
1989         struct dt_object *obj;
1990
1991         obj = lfsck_object_find_bottom(env, lfsck, &lso->lso_fid);
1992         if (IS_ERR(obj))
1993                 return obj;
1994
1995         if (unlikely(!dt_object_exists(obj) || lfsck_is_dead_obj(obj))) {
1996                 lso->lso_dead = 1;
1997                 lfsck_object_put(env, obj);
1998
1999                 return ERR_PTR(-ENOENT);
2000         }
2001
2002         if (lso->lso_is_dir && unlikely(!dt_try_as_dir(env, obj))) {
2003                 lfsck_object_put(env, obj);
2004
2005                 return ERR_PTR(-ENOTDIR);
2006         }
2007
2008         return obj;
2009 }
2010
2011 /**
2012  * Generic LFSCK asynchronous communication interpretor function.
2013  * The LFSCK RPC reply for both the event notification and status
2014  * querying will be handled here.
2015  *
2016  * \param[in] env       pointer to the thread context
2017  * \param[in] req       pointer to the LFSCK request
2018  * \param[in] args      pointer to the lfsck_async_interpret_args
2019  * \param[in] rc        the result for handling the LFSCK request
2020  *
2021  * \retval              0 for success
2022  * \retval              negative error number on failure
2023  */
2024 int lfsck_async_interpret_common(const struct lu_env *env,
2025                                  struct ptlrpc_request *req,
2026                                  void *args, int rc)
2027 {
2028         struct lfsck_async_interpret_args *laia = args;
2029         struct lfsck_component            *com  = laia->laia_com;
2030         struct lfsck_assistant_data       *lad  = com->lc_data;
2031         struct lfsck_tgt_descs            *ltds = laia->laia_ltds;
2032         struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
2033         struct lfsck_request              *lr   = laia->laia_lr;
2034
2035         LASSERT(com->lc_lfsck->li_master);
2036
2037         switch (lr->lr_event) {
2038         case LE_START:
2039                 if (unlikely(rc == -EINPROGRESS)) {
2040                         ltd->ltd_retry_start = 1;
2041                         break;
2042                 }
2043
2044                 if (rc != 0) {
2045                         CDEBUG(D_LFSCK, "%s: fail to notify %s %x for %s "
2046                                "start: rc = %d\n",
2047                                lfsck_lfsck2name(com->lc_lfsck),
2048                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2049                                ltd->ltd_index, lad->lad_name, rc);
2050
2051                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2052                                 struct lfsck_layout *lo = com->lc_file_ram;
2053
2054                                 if (lr->lr_flags & LEF_TO_OST)
2055                                         lfsck_lad_set_bitmap(env, com,
2056                                                              ltd->ltd_index);
2057                                 else
2058                                         lo->ll_flags |= LF_INCOMPLETE;
2059                         } else {
2060                                 struct lfsck_namespace *ns = com->lc_file_ram;
2061
2062                                 /* If some MDT does not join the namespace
2063                                  * LFSCK, then we cannot know whether there
2064                                  * is some name entry on such MDT that with
2065                                  * the referenced MDT-object on this MDT or
2066                                  * not. So the namespace LFSCK on this MDT
2067                                  * cannot handle orphan MDT-objects properly.
2068                                  * So we mark the LFSCK as LF_INCOMPLETE and
2069                                  * skip orphan MDT-objects handling. */
2070                                 ns->ln_flags |= LF_INCOMPLETE;
2071                         }
2072                         break;
2073                 }
2074
2075                 spin_lock(&ltds->ltd_lock);
2076                 if (ltd->ltd_dead) {
2077                         spin_unlock(&ltds->ltd_lock);
2078                         break;
2079                 }
2080
2081                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2082                         struct list_head *list;
2083                         struct list_head *phase_list;
2084
2085                         if (ltd->ltd_layout_done) {
2086                                 spin_unlock(&ltds->ltd_lock);
2087                                 break;
2088                         }
2089
2090                         if (lr->lr_flags & LEF_TO_OST) {
2091                                 list = &lad->lad_ost_list;
2092                                 phase_list = &lad->lad_ost_phase1_list;
2093                         } else {
2094                                 list = &lad->lad_mdt_list;
2095                                 phase_list = &lad->lad_mdt_phase1_list;
2096                         }
2097
2098                         if (list_empty(&ltd->ltd_layout_list))
2099                                 list_add_tail(&ltd->ltd_layout_list, list);
2100                         if (list_empty(&ltd->ltd_layout_phase_list))
2101                                 list_add_tail(&ltd->ltd_layout_phase_list,
2102                                               phase_list);
2103                 } else {
2104                         if (ltd->ltd_namespace_done) {
2105                                 spin_unlock(&ltds->ltd_lock);
2106                                 break;
2107                         }
2108
2109                         if (list_empty(&ltd->ltd_namespace_list))
2110                                 list_add_tail(&ltd->ltd_namespace_list,
2111                                               &lad->lad_mdt_list);
2112                         if (list_empty(&ltd->ltd_namespace_phase_list))
2113                                 list_add_tail(&ltd->ltd_namespace_phase_list,
2114                                               &lad->lad_mdt_phase1_list);
2115                 }
2116                 spin_unlock(&ltds->ltd_lock);
2117                 break;
2118         case LE_STOP:
2119         case LE_PHASE1_DONE:
2120         case LE_PHASE2_DONE:
2121         case LE_PEER_EXIT:
2122                 if (rc != 0 && rc != -EALREADY)
2123                         CDEBUG(D_LFSCK, "%s: fail to notify %s %x for %s: "
2124                               "event = %d, rc = %d\n",
2125                               lfsck_lfsck2name(com->lc_lfsck),
2126                               (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2127                               ltd->ltd_index, lad->lad_name, lr->lr_event, rc);
2128                 break;
2129         case LE_QUERY: {
2130                 struct lfsck_reply *reply;
2131                 struct list_head *list;
2132                 struct list_head *phase_list;
2133
2134                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2135                         list = &ltd->ltd_layout_list;
2136                         phase_list = &ltd->ltd_layout_phase_list;
2137                 } else {
2138                         list = &ltd->ltd_namespace_list;
2139                         phase_list = &ltd->ltd_namespace_phase_list;
2140                 }
2141
2142                 if (rc != 0) {
2143                         if (lr->lr_flags & LEF_QUERY_ALL) {
2144                                 lfsck_reset_ltd_status(ltd, com->lc_type);
2145                                 break;
2146                         }
2147
2148                         spin_lock(&ltds->ltd_lock);
2149                         list_del_init(phase_list);
2150                         list_del_init(list);
2151                         spin_unlock(&ltds->ltd_lock);
2152                         break;
2153                 }
2154
2155                 reply = req_capsule_server_get(&req->rq_pill,
2156                                                &RMF_LFSCK_REPLY);
2157                 if (reply == NULL) {
2158                         rc = -EPROTO;
2159                         CDEBUG(D_LFSCK, "%s: invalid query reply for %s: "
2160                                "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
2161                                lad->lad_name, rc);
2162
2163                         if (lr->lr_flags & LEF_QUERY_ALL) {
2164                                 lfsck_reset_ltd_status(ltd, com->lc_type);
2165                                 break;
2166                         }
2167
2168                         spin_lock(&ltds->ltd_lock);
2169                         list_del_init(phase_list);
2170                         list_del_init(list);
2171                         spin_unlock(&ltds->ltd_lock);
2172                         break;
2173                 }
2174
2175                 if (lr->lr_flags & LEF_QUERY_ALL) {
2176                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2177                                 ltd->ltd_layout_status = reply->lr_status;
2178                                 ltd->ltd_layout_repaired = reply->lr_repaired;
2179                         } else {
2180                                 ltd->ltd_namespace_status = reply->lr_status;
2181                                 ltd->ltd_namespace_repaired =
2182                                                         reply->lr_repaired;
2183                         }
2184                         break;
2185                 }
2186
2187                 switch (reply->lr_status) {
2188                 case LS_SCANNING_PHASE1:
2189                         break;
2190                 case LS_SCANNING_PHASE2:
2191                         spin_lock(&ltds->ltd_lock);
2192                         list_del_init(phase_list);
2193                         if (ltd->ltd_dead) {
2194                                 spin_unlock(&ltds->ltd_lock);
2195                                 break;
2196                         }
2197
2198                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2199                                 if (ltd->ltd_layout_done) {
2200                                         spin_unlock(&ltds->ltd_lock);
2201                                         break;
2202                                 }
2203
2204                                 if (lr->lr_flags & LEF_TO_OST)
2205                                         list_add_tail(phase_list,
2206                                                 &lad->lad_ost_phase2_list);
2207                                 else
2208                                         list_add_tail(phase_list,
2209                                                 &lad->lad_mdt_phase2_list);
2210                         } else {
2211                                 if (ltd->ltd_namespace_done) {
2212                                         spin_unlock(&ltds->ltd_lock);
2213                                         break;
2214                                 }
2215
2216                                 list_add_tail(phase_list,
2217                                               &lad->lad_mdt_phase2_list);
2218                         }
2219                         spin_unlock(&ltds->ltd_lock);
2220                         break;
2221                 default:
2222                         spin_lock(&ltds->ltd_lock);
2223                         list_del_init(phase_list);
2224                         list_del_init(list);
2225                         spin_unlock(&ltds->ltd_lock);
2226                         break;
2227                 }
2228                 break;
2229         }
2230         default:
2231                 CDEBUG(D_LFSCK, "%s: unexpected event: rc = %d\n",
2232                        lfsck_lfsck2name(com->lc_lfsck), lr->lr_event);
2233                 break;
2234         }
2235
2236         if (!laia->laia_shared) {
2237                 lfsck_tgt_put(ltd);
2238                 lfsck_component_put(env, com);
2239         }
2240
2241         return 0;
2242 }
2243
2244 static void lfsck_interpret(const struct lu_env *env,
2245                             struct lfsck_instance *lfsck,
2246                             struct ptlrpc_request *req, void *args, int result)
2247 {
2248         struct lfsck_async_interpret_args *laia = args;
2249         struct lfsck_component            *com;
2250
2251         LASSERT(laia->laia_com == NULL);
2252         LASSERT(laia->laia_shared);
2253
2254         spin_lock(&lfsck->li_lock);
2255         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
2256                 laia->laia_com = com;
2257                 lfsck_async_interpret_common(env, req, laia, result);
2258         }
2259
2260         list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
2261                 laia->laia_com = com;
2262                 lfsck_async_interpret_common(env, req, laia, result);
2263         }
2264         spin_unlock(&lfsck->li_lock);
2265 }
2266
2267 static int lfsck_stop_notify(const struct lu_env *env,
2268                              struct lfsck_instance *lfsck,
2269                              struct lfsck_tgt_descs *ltds,
2270                              struct lfsck_tgt_desc *ltd, __u16 type)
2271 {
2272         struct lfsck_component *com;
2273         int                     rc = 0;
2274         ENTRY;
2275
2276         LASSERT(lfsck->li_master);
2277
2278         spin_lock(&lfsck->li_lock);
2279         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_scan);
2280         if (com == NULL)
2281                 com = __lfsck_component_find(lfsck, type,
2282                                              &lfsck->li_list_double_scan);
2283         if (com != NULL)
2284                 lfsck_component_get(com);
2285         spin_unlock(&lfsck->li_lock);
2286
2287         if (com != NULL) {
2288                 struct lfsck_thread_info          *info  = lfsck_env_info(env);
2289                 struct lfsck_async_interpret_args *laia  = &info->lti_laia;
2290                 struct lfsck_request              *lr    = &info->lti_lr;
2291                 struct lfsck_assistant_data       *lad   = com->lc_data;
2292                 struct list_head                  *list;
2293                 struct list_head                  *phase_list;
2294                 struct ptlrpc_request_set         *set;
2295
2296                 set = ptlrpc_prep_set();
2297                 if (set == NULL) {
2298                         lfsck_component_put(env, com);
2299
2300                         RETURN(-ENOMEM);
2301                 }
2302
2303                 if (type == LFSCK_TYPE_LAYOUT) {
2304                         list = &ltd->ltd_layout_list;
2305                         phase_list = &ltd->ltd_layout_phase_list;
2306                 } else {
2307                         list = &ltd->ltd_namespace_list;
2308                         phase_list = &ltd->ltd_namespace_phase_list;
2309                 }
2310
2311                 spin_lock(&ltds->ltd_lock);
2312                 if (list_empty(list)) {
2313                         LASSERT(list_empty(phase_list));
2314                         spin_unlock(&ltds->ltd_lock);
2315                         ptlrpc_set_destroy(set);
2316
2317                         RETURN(0);
2318                 }
2319
2320                 list_del_init(phase_list);
2321                 list_del_init(list);
2322                 spin_unlock(&ltds->ltd_lock);
2323
2324                 memset(lr, 0, sizeof(*lr));
2325                 lr->lr_index = lfsck_dev_idx(lfsck);
2326                 lr->lr_event = LE_PEER_EXIT;
2327                 lr->lr_active = type;
2328                 lr->lr_status = LS_CO_PAUSED;
2329                 if (ltds == &lfsck->li_ost_descs)
2330                         lr->lr_flags = LEF_TO_OST;
2331
2332                 memset(laia, 0, sizeof(*laia));
2333                 laia->laia_com = com;
2334                 laia->laia_ltds = ltds;
2335                 atomic_inc(&ltd->ltd_ref);
2336                 laia->laia_ltd = ltd;
2337                 laia->laia_lr = lr;
2338
2339                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2340                                          lfsck_async_interpret_common,
2341                                          laia, LFSCK_NOTIFY);
2342                 if (rc != 0) {
2343                         CDEBUG(D_LFSCK, "%s: fail to notify %s %x for "
2344                                "co-stop for %s: rc = %d\n",
2345                                lfsck_lfsck2name(lfsck),
2346                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2347                                ltd->ltd_index, lad->lad_name, rc);
2348                         lfsck_tgt_put(ltd);
2349                 } else {
2350                         rc = ptlrpc_set_wait(set);
2351                 }
2352
2353                 ptlrpc_set_destroy(set);
2354                 lfsck_component_put(env, com);
2355         }
2356
2357         RETURN(rc);
2358 }
2359
2360 static int lfsck_async_interpret(const struct lu_env *env,
2361                                  struct ptlrpc_request *req,
2362                                  void *args, int rc)
2363 {
2364         struct lfsck_async_interpret_args *laia = args;
2365         struct lfsck_instance             *lfsck;
2366
2367         lfsck = container_of0(laia->laia_ltds, struct lfsck_instance,
2368                               li_mdt_descs);
2369         lfsck_interpret(env, lfsck, req, laia, rc);
2370         lfsck_tgt_put(laia->laia_ltd);
2371         if (rc != 0 && laia->laia_result != -EALREADY)
2372                 laia->laia_result = rc;
2373
2374         return 0;
2375 }
2376
2377 int lfsck_async_request(const struct lu_env *env, struct obd_export *exp,
2378                         struct lfsck_request *lr,
2379                         struct ptlrpc_request_set *set,
2380                         ptlrpc_interpterer_t interpreter,
2381                         void *args, int request)
2382 {
2383         struct lfsck_async_interpret_args *laia;
2384         struct ptlrpc_request             *req;
2385         struct lfsck_request              *tmp;
2386         struct req_format                 *format;
2387         int                                rc;
2388
2389         switch (request) {
2390         case LFSCK_NOTIFY:
2391                 format = &RQF_LFSCK_NOTIFY;
2392                 break;
2393         case LFSCK_QUERY:
2394                 format = &RQF_LFSCK_QUERY;
2395                 break;
2396         default:
2397                 CDEBUG(D_LFSCK, "%s: unknown async request %d: rc = %d\n",
2398                        exp->exp_obd->obd_name, request, -EINVAL);
2399                 return -EINVAL;
2400         }
2401
2402         req = ptlrpc_request_alloc(class_exp2cliimp(exp), format);
2403         if (req == NULL)
2404                 return -ENOMEM;
2405
2406         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, request);
2407         if (rc != 0) {
2408                 ptlrpc_request_free(req);
2409
2410                 return rc;
2411         }
2412
2413         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2414         *tmp = *lr;
2415         ptlrpc_request_set_replen(req);
2416
2417         laia = ptlrpc_req_async_args(req);
2418         *laia = *(struct lfsck_async_interpret_args *)args;
2419         if (laia->laia_com != NULL)
2420                 lfsck_component_get(laia->laia_com);
2421         req->rq_interpret_reply = interpreter;
2422         req->rq_allow_intr = 1;
2423         req->rq_no_delay = 1;
2424         ptlrpc_set_add_req(set, req);
2425
2426         return 0;
2427 }
2428
2429 int lfsck_query_all(const struct lu_env *env, struct lfsck_component *com)
2430 {
2431         struct lfsck_thread_info          *info  = lfsck_env_info(env);
2432         struct lfsck_request              *lr    = &info->lti_lr;
2433         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
2434         struct lfsck_instance             *lfsck = com->lc_lfsck;
2435         struct lfsck_tgt_descs            *ltds  = &lfsck->li_mdt_descs;
2436         struct lfsck_tgt_desc             *ltd;
2437         struct ptlrpc_request_set         *set;
2438         int                                idx;
2439         int                                rc;
2440         ENTRY;
2441
2442         memset(lr, 0, sizeof(*lr));
2443         lr->lr_event = LE_QUERY;
2444         lr->lr_active = com->lc_type;
2445         lr->lr_flags = LEF_QUERY_ALL;
2446
2447         memset(laia, 0, sizeof(*laia));
2448         laia->laia_com = com;
2449         laia->laia_lr = lr;
2450
2451         set = ptlrpc_prep_set();
2452         if (set == NULL)
2453                 RETURN(-ENOMEM);
2454
2455 again:
2456         laia->laia_ltds = ltds;
2457         down_read(&ltds->ltd_rw_sem);
2458         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
2459                 ltd = lfsck_tgt_get(ltds, idx);
2460                 LASSERT(ltd != NULL);
2461
2462                 laia->laia_ltd = ltd;
2463                 up_read(&ltds->ltd_rw_sem);
2464                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2465                                          lfsck_async_interpret_common,
2466                                          laia, LFSCK_QUERY);
2467                 if (rc != 0) {
2468                         struct lfsck_assistant_data *lad = com->lc_data;
2469
2470                         CDEBUG(D_LFSCK, "%s: Fail to query %s %x for stat %s: "
2471                                "rc = %d\n", lfsck_lfsck2name(lfsck),
2472                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2473                                ltd->ltd_index, lad->lad_name, rc);
2474                         lfsck_reset_ltd_status(ltd, com->lc_type);
2475                         lfsck_tgt_put(ltd);
2476                 }
2477                 down_read(&ltds->ltd_rw_sem);
2478         }
2479         up_read(&ltds->ltd_rw_sem);
2480
2481         if (com->lc_type == LFSCK_TYPE_LAYOUT && !(lr->lr_flags & LEF_TO_OST)) {
2482                 ltds = &lfsck->li_ost_descs;
2483                 lr->lr_flags |= LEF_TO_OST;
2484                 goto again;
2485         }
2486
2487         rc = ptlrpc_set_wait(set);
2488         ptlrpc_set_destroy(set);
2489
2490         RETURN(rc);
2491 }
2492
2493 int lfsck_start_assistant(const struct lu_env *env, struct lfsck_component *com,
2494                           struct lfsck_start_param *lsp)
2495 {
2496         struct lfsck_instance           *lfsck   = com->lc_lfsck;
2497         struct lfsck_assistant_data     *lad     = com->lc_data;
2498         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
2499         struct ptlrpc_thread            *athread = &lad->lad_thread;
2500         struct lfsck_thread_args        *lta;
2501         struct task_struct              *task;
2502         int                              rc;
2503         ENTRY;
2504
2505         lad->lad_assistant_status = 0;
2506         lad->lad_post_result = 0;
2507         lad->lad_to_post = 0;
2508         lad->lad_to_double_scan = 0;
2509         lad->lad_in_double_scan = 0;
2510         lad->lad_exit = 0;
2511         lad->lad_advance_lock = false;
2512         thread_set_flags(athread, 0);
2513
2514         lta = lfsck_thread_args_init(lfsck, com, lsp);
2515         if (IS_ERR(lta))
2516                 RETURN(PTR_ERR(lta));
2517
2518         task = kthread_run(lfsck_assistant_engine, lta, lad->lad_name);
2519         if (IS_ERR(task)) {
2520                 rc = PTR_ERR(task);
2521                 CERROR("%s: cannot start LFSCK assistant thread for %s: "
2522                        "rc = %d\n", lfsck_lfsck2name(lfsck), lad->lad_name, rc);
2523                 lfsck_thread_args_fini(lta);
2524         } else {
2525                 struct l_wait_info lwi = { 0 };
2526
2527                 l_wait_event(mthread->t_ctl_waitq,
2528                              thread_is_running(athread) ||
2529                              thread_is_stopped(athread) ||
2530                              !thread_is_starting(mthread),
2531                              &lwi);
2532                 if (unlikely(!thread_is_starting(mthread)))
2533                         /* stopped by race */
2534                         rc = -ESRCH;
2535                 else if (unlikely(!thread_is_running(athread)))
2536                         rc = lad->lad_assistant_status;
2537                 else
2538                         rc = 0;
2539         }
2540
2541         RETURN(rc);
2542 }
2543
2544 int lfsck_checkpoint_generic(const struct lu_env *env,
2545                              struct lfsck_component *com)
2546 {
2547         struct lfsck_assistant_data     *lad     = com->lc_data;
2548         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2549         struct ptlrpc_thread            *athread = &lad->lad_thread;
2550         struct l_wait_info               lwi     = { 0 };
2551
2552         l_wait_event(mthread->t_ctl_waitq,
2553                      list_empty(&lad->lad_req_list) ||
2554                      !thread_is_running(mthread) ||
2555                      thread_is_stopped(athread),
2556                      &lwi);
2557
2558         if (!thread_is_running(mthread) || thread_is_stopped(athread))
2559                 return LFSCK_CHECKPOINT_SKIP;
2560
2561         return 0;
2562 }
2563
2564 void lfsck_post_generic(const struct lu_env *env,
2565                         struct lfsck_component *com, int *result)
2566 {
2567         struct lfsck_assistant_data     *lad     = com->lc_data;
2568         struct ptlrpc_thread            *athread = &lad->lad_thread;
2569         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2570         struct l_wait_info               lwi     = { 0 };
2571
2572         lad->lad_post_result = *result;
2573         if (*result <= 0)
2574                 lad->lad_exit = 1;
2575         lad->lad_to_post = 1;
2576
2577         CDEBUG(D_LFSCK, "%s: waiting for assistant to do %s post, rc = %d\n",
2578                lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, *result);
2579
2580         wake_up_all(&athread->t_ctl_waitq);
2581         l_wait_event(mthread->t_ctl_waitq,
2582                      (*result > 0 && list_empty(&lad->lad_req_list)) ||
2583                      thread_is_stopped(athread),
2584                      &lwi);
2585
2586         if (lad->lad_assistant_status < 0)
2587                 *result = lad->lad_assistant_status;
2588
2589         CDEBUG(D_LFSCK, "%s: the assistant has done %s post, rc = %d\n",
2590                lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, *result);
2591 }
2592
2593 int lfsck_double_scan_generic(const struct lu_env *env,
2594                               struct lfsck_component *com, int status)
2595 {
2596         struct lfsck_assistant_data     *lad     = com->lc_data;
2597         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2598         struct ptlrpc_thread            *athread = &lad->lad_thread;
2599         struct l_wait_info               lwi     = { 0 };
2600
2601         if (status != LS_SCANNING_PHASE2)
2602                 lad->lad_exit = 1;
2603         else
2604                 lad->lad_to_double_scan = 1;
2605
2606         CDEBUG(D_LFSCK, "%s: waiting for assistant to do %s double_scan, "
2607                "status %d\n",
2608                lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, status);
2609
2610         wake_up_all(&athread->t_ctl_waitq);
2611         l_wait_event(mthread->t_ctl_waitq,
2612                      lad->lad_in_double_scan ||
2613                      thread_is_stopped(athread),
2614                      &lwi);
2615
2616         CDEBUG(D_LFSCK, "%s: the assistant has done %s double_scan, "
2617                "status %d\n", lfsck_lfsck2name(com->lc_lfsck), lad->lad_name,
2618                lad->lad_assistant_status);
2619
2620         if (lad->lad_assistant_status < 0)
2621                 return lad->lad_assistant_status;
2622
2623         return 0;
2624 }
2625
2626 void lfsck_quit_generic(const struct lu_env *env,
2627                         struct lfsck_component *com)
2628 {
2629         struct lfsck_assistant_data     *lad     = com->lc_data;
2630         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2631         struct ptlrpc_thread            *athread = &lad->lad_thread;
2632         struct l_wait_info               lwi     = { 0 };
2633
2634         lad->lad_exit = 1;
2635         wake_up_all(&athread->t_ctl_waitq);
2636         l_wait_event(mthread->t_ctl_waitq,
2637                      thread_is_init(athread) ||
2638                      thread_is_stopped(athread),
2639                      &lwi);
2640 }
2641
2642 int lfsck_load_one_trace_file(const struct lu_env *env,
2643                               struct lfsck_component *com,
2644                               struct dt_object *parent,
2645                               struct dt_object **child,
2646                               const struct dt_index_features *ft,
2647                               const char *name, bool reset)
2648 {
2649         struct lfsck_instance *lfsck = com->lc_lfsck;
2650         struct dt_object *obj;
2651         int rc;
2652         ENTRY;
2653
2654         if (*child != NULL) {
2655                 struct dt_it *it;
2656                 const struct dt_it_ops *iops;
2657                 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid3;
2658
2659                 if (!reset)
2660                         RETURN(0);
2661
2662                 obj = *child;
2663                 rc = obj->do_ops->do_index_try(env, obj, ft);
2664                 if (rc)
2665                         /* unlink by force */
2666                         goto unlink;
2667
2668                 iops = &obj->do_index_ops->dio_it;
2669                 it = iops->init(env, obj, 0);
2670                 if (IS_ERR(it))
2671                         /* unlink by force */
2672                         goto unlink;
2673
2674                 fid_zero(fid);
2675                 rc = iops->get(env, it, (const struct dt_key *)fid);
2676                 if (rc >= 0) {
2677                         rc = iops->next(env, it);
2678                         iops->put(env, it);
2679                 }
2680                 iops->fini(env, it);
2681                 if (rc > 0)
2682                         /* "rc > 0" means the index file is empty. */
2683                         RETURN(0);
2684
2685 unlink:
2686                 /* The old index is not empty, remove it firstly. */
2687                 rc = local_object_unlink(env, lfsck->li_bottom, parent, name);
2688                 CDEBUG_LIMIT(rc ? D_ERROR : D_LFSCK,
2689                              "%s: unlink lfsck sub trace file %s: rc = %d\n",
2690                              lfsck_lfsck2name(com->lc_lfsck), name, rc);
2691                 if (rc)
2692                         RETURN(rc);
2693
2694                 if (*child) {
2695                         lfsck_object_put(env, *child);
2696                         *child = NULL;
2697                 }
2698         } else if (reset) {
2699                 goto unlink;
2700         }
2701
2702         obj = local_index_find_or_create(env, lfsck->li_los, parent, name,
2703                                          S_IFREG | S_IRUGO | S_IWUSR, ft);
2704         if (IS_ERR(obj))
2705                 RETURN(PTR_ERR(obj));
2706
2707         rc = obj->do_ops->do_index_try(env, obj, ft);
2708         if (rc) {
2709                 lfsck_object_put(env, obj);
2710                 CDEBUG(D_LFSCK, "%s: LFSCK fail to load "
2711                        "sub trace file %s: rc = %d\n",
2712                        lfsck_lfsck2name(com->lc_lfsck), name, rc);
2713         } else {
2714                 *child = obj;
2715         }
2716
2717         RETURN(rc);
2718 }
2719
2720 int lfsck_load_sub_trace_files(const struct lu_env *env,
2721                                struct lfsck_component *com,
2722                                const struct dt_index_features *ft,
2723                                const char *prefix, bool reset)
2724 {
2725         char *name = lfsck_env_info(env)->lti_key;
2726         struct lfsck_sub_trace_obj *lsto;
2727         int rc;
2728         int i;
2729
2730         for (i = 0, rc = 0, lsto = &com->lc_sub_trace_objs[0];
2731              i < LFSCK_STF_COUNT && rc == 0; i++, lsto++) {
2732                 snprintf(name, NAME_MAX, "%s_%02d", prefix, i);
2733                 rc = lfsck_load_one_trace_file(env, com,
2734                                 com->lc_lfsck->li_lfsck_dir,
2735                                 &lsto->lsto_obj, ft, name, reset);
2736         }
2737
2738         return rc;
2739 }
2740
2741 /* external interfaces */
2742
2743 int lfsck_get_speed(struct seq_file *m, struct dt_device *key)
2744 {
2745         struct lu_env           env;
2746         struct lfsck_instance  *lfsck;
2747         int                     rc;
2748         ENTRY;
2749
2750         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2751         if (rc != 0)
2752                 RETURN(rc);
2753
2754         lfsck = lfsck_instance_find(key, true, false);
2755         if (likely(lfsck != NULL)) {
2756                 seq_printf(m, "%u\n", lfsck->li_bookmark_ram.lb_speed_limit);
2757                 lfsck_instance_put(&env, lfsck);
2758         } else {
2759                 rc = -ENXIO;
2760         }
2761
2762         lu_env_fini(&env);
2763
2764         RETURN(rc);
2765 }
2766 EXPORT_SYMBOL(lfsck_get_speed);
2767
2768 int lfsck_set_speed(struct dt_device *key, __u32 val)
2769 {
2770         struct lu_env           env;
2771         struct lfsck_instance  *lfsck;
2772         int                     rc;
2773         ENTRY;
2774
2775         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2776         if (rc != 0)
2777                 RETURN(rc);
2778
2779         lfsck = lfsck_instance_find(key, true, false);
2780         if (likely(lfsck != NULL)) {
2781                 mutex_lock(&lfsck->li_mutex);
2782                 if (__lfsck_set_speed(lfsck, val))
2783                         rc = lfsck_bookmark_store(&env, lfsck);
2784                 mutex_unlock(&lfsck->li_mutex);
2785                 lfsck_instance_put(&env, lfsck);
2786         } else {
2787                 rc = -ENXIO;
2788         }
2789
2790         lu_env_fini(&env);
2791
2792         RETURN(rc);
2793 }
2794 EXPORT_SYMBOL(lfsck_set_speed);
2795
2796 int lfsck_get_windows(struct seq_file *m, struct dt_device *key)
2797 {
2798         struct lu_env           env;
2799         struct lfsck_instance  *lfsck;
2800         int                     rc;
2801         ENTRY;
2802
2803         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2804         if (rc != 0)
2805                 RETURN(rc);
2806
2807         lfsck = lfsck_instance_find(key, true, false);
2808         if (likely(lfsck != NULL)) {
2809                 seq_printf(m, "%u\n", lfsck->li_bookmark_ram.lb_async_windows);
2810                 lfsck_instance_put(&env, lfsck);
2811         } else {
2812                 rc = -ENXIO;
2813         }
2814
2815         lu_env_fini(&env);
2816
2817         RETURN(rc);
2818 }
2819 EXPORT_SYMBOL(lfsck_get_windows);
2820
2821 int lfsck_set_windows(struct dt_device *key, unsigned int val)
2822 {
2823         struct lu_env           env;
2824         struct lfsck_instance  *lfsck;
2825         int                     rc;
2826         ENTRY;
2827
2828         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2829         if (rc != 0)
2830                 RETURN(rc);
2831
2832         lfsck = lfsck_instance_find(key, true, false);
2833         if (likely(lfsck != NULL)) {
2834                 if (val < 1 || val > LFSCK_ASYNC_WIN_MAX) {
2835                         CWARN("%s: invalid async windows size that may "
2836                               "cause memory issues. The valid range is "
2837                               "[1 - %u].\n",
2838                               lfsck_lfsck2name(lfsck), LFSCK_ASYNC_WIN_MAX);
2839                         rc = -EINVAL;
2840                 } else if (lfsck->li_bookmark_ram.lb_async_windows != val) {
2841                         mutex_lock(&lfsck->li_mutex);
2842                         lfsck->li_bookmark_ram.lb_async_windows = val;
2843                         rc = lfsck_bookmark_store(&env, lfsck);
2844                         mutex_unlock(&lfsck->li_mutex);
2845                 }
2846                 lfsck_instance_put(&env, lfsck);
2847         } else {
2848                 rc = -ENXIO;
2849         }
2850
2851         lu_env_fini(&env);
2852
2853         RETURN(rc);
2854 }
2855 EXPORT_SYMBOL(lfsck_set_windows);
2856
2857 int lfsck_dump(struct seq_file *m, struct dt_device *key, enum lfsck_type type)
2858 {
2859         struct lu_env           env;
2860         struct lfsck_instance  *lfsck;
2861         struct lfsck_component *com;
2862         int                     rc;
2863         ENTRY;
2864
2865         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2866         if (rc != 0)
2867                 RETURN(rc);
2868
2869         lfsck = lfsck_instance_find(key, true, false);
2870         if (likely(lfsck != NULL)) {
2871                 com = lfsck_component_find(lfsck, type);
2872                 if (likely(com != NULL)) {
2873                         com->lc_ops->lfsck_dump(&env, com, m);
2874                         lfsck_component_put(&env, com);
2875                 } else {
2876                         rc = -ENOTSUPP;
2877                 }
2878
2879                 lfsck_instance_put(&env, lfsck);
2880         } else {
2881                 rc = -ENXIO;
2882         }
2883
2884         lu_env_fini(&env);
2885
2886         RETURN(rc);
2887 }
2888 EXPORT_SYMBOL(lfsck_dump);
2889
2890 static int lfsck_stop_all(const struct lu_env *env,
2891                           struct lfsck_instance *lfsck,
2892                           struct lfsck_stop *stop)
2893 {
2894         struct lfsck_thread_info          *info   = lfsck_env_info(env);
2895         struct lfsck_request              *lr     = &info->lti_lr;
2896         struct lfsck_async_interpret_args *laia   = &info->lti_laia;
2897         struct ptlrpc_request_set         *set;
2898         struct lfsck_tgt_descs            *ltds   = &lfsck->li_mdt_descs;
2899         struct lfsck_tgt_desc             *ltd;
2900         struct lfsck_bookmark             *bk     = &lfsck->li_bookmark_ram;
2901         __u32                              idx;
2902         int                                rc     = 0;
2903         int                                rc1    = 0;
2904         ENTRY;
2905
2906         LASSERT(stop->ls_flags & LPF_BROADCAST);
2907
2908         set = ptlrpc_prep_set();
2909         if (unlikely(set == NULL))
2910                 RETURN(-ENOMEM);
2911
2912         memset(lr, 0, sizeof(*lr));
2913         lr->lr_event = LE_STOP;
2914         lr->lr_index = lfsck_dev_idx(lfsck);
2915         lr->lr_status = stop->ls_status;
2916         lr->lr_version = bk->lb_version;
2917         lr->lr_active = LFSCK_TYPES_ALL;
2918         lr->lr_param = stop->ls_flags;
2919
2920         memset(laia, 0, sizeof(*laia));
2921         laia->laia_ltds = ltds;
2922         laia->laia_lr = lr;
2923         laia->laia_shared = 1;
2924
2925         down_read(&ltds->ltd_rw_sem);
2926         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
2927                 ltd = lfsck_tgt_get(ltds, idx);
2928                 LASSERT(ltd != NULL);
2929
2930                 laia->laia_ltd = ltd;
2931                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2932                                          lfsck_async_interpret, laia,
2933                                          LFSCK_NOTIFY);
2934                 if (rc != 0) {
2935                         lfsck_interpret(env, lfsck, NULL, laia, rc);
2936                         lfsck_tgt_put(ltd);
2937                         CERROR("%s: cannot notify MDT %x for LFSCK stop: "
2938                                "rc = %d\n", lfsck_lfsck2name(lfsck), idx, rc);
2939                         rc1 = rc;
2940                 }
2941         }
2942         up_read(&ltds->ltd_rw_sem);
2943
2944         rc = ptlrpc_set_wait(set);
2945         ptlrpc_set_destroy(set);
2946
2947         if (rc == 0)
2948                 rc = laia->laia_result;
2949
2950         if (rc == -EALREADY)
2951                 rc = 0;
2952
2953         if (rc != 0)
2954                 CERROR("%s: fail to stop LFSCK on some MDTs: rc = %d\n",
2955                        lfsck_lfsck2name(lfsck), rc);
2956
2957         RETURN(rc != 0 ? rc : rc1);
2958 }
2959
2960 static int lfsck_start_all(const struct lu_env *env,
2961                            struct lfsck_instance *lfsck,
2962                            struct lfsck_start *start)
2963 {
2964         struct lfsck_thread_info          *info   = lfsck_env_info(env);
2965         struct lfsck_request              *lr     = &info->lti_lr;
2966         struct lfsck_async_interpret_args *laia   = &info->lti_laia;
2967         struct ptlrpc_request_set         *set;
2968         struct lfsck_tgt_descs            *ltds   = &lfsck->li_mdt_descs;
2969         struct lfsck_tgt_desc             *ltd;
2970         struct lfsck_bookmark             *bk     = &lfsck->li_bookmark_ram;
2971         __u32                              idx;
2972         int                                rc     = 0;
2973         bool retry = false;
2974         ENTRY;
2975
2976         LASSERT(start->ls_flags & LPF_BROADCAST);
2977
2978         memset(lr, 0, sizeof(*lr));
2979         lr->lr_event = LE_START;
2980         lr->lr_index = lfsck_dev_idx(lfsck);
2981         lr->lr_speed = bk->lb_speed_limit;
2982         lr->lr_version = bk->lb_version;
2983         lr->lr_active = start->ls_active;
2984         lr->lr_param = start->ls_flags;
2985         lr->lr_async_windows = bk->lb_async_windows;
2986         lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN |
2987                        LSV_ASYNC_WINDOWS | LSV_CREATE_OSTOBJ |
2988                        LSV_CREATE_MDTOBJ;
2989
2990         memset(laia, 0, sizeof(*laia));
2991         laia->laia_ltds = ltds;
2992         laia->laia_lr = lr;
2993         laia->laia_shared = 1;
2994
2995 again:
2996         set = ptlrpc_prep_set();
2997         if (unlikely(!set))
2998                 RETURN(-ENOMEM);
2999
3000         down_read(&ltds->ltd_rw_sem);
3001         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
3002                 ltd = lfsck_tgt_get(ltds, idx);
3003                 LASSERT(ltd != NULL);
3004
3005                 if (retry && !ltd->ltd_retry_start) {
3006                         lfsck_tgt_put(ltd);
3007                         continue;
3008                 }
3009
3010                 laia->laia_ltd = ltd;
3011                 ltd->ltd_retry_start = 0;
3012                 ltd->ltd_layout_done = 0;
3013                 ltd->ltd_namespace_done = 0;
3014                 ltd->ltd_synced_failures = 0;
3015                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
3016                                          lfsck_async_interpret, laia,
3017                                          LFSCK_NOTIFY);
3018                 if (rc != 0) {
3019                         lfsck_interpret(env, lfsck, NULL, laia, rc);
3020                         lfsck_tgt_put(ltd);
3021                         CERROR("%s: cannot notify MDT %x for LFSCK "
3022                                "start, failout: rc = %d\n",
3023                                lfsck_lfsck2name(lfsck), idx, rc);
3024                         break;
3025                 }
3026         }
3027         up_read(&ltds->ltd_rw_sem);
3028
3029         if (rc != 0) {
3030                 ptlrpc_set_destroy(set);
3031
3032                 RETURN(rc);
3033         }
3034
3035         rc = ptlrpc_set_wait(set);
3036         ptlrpc_set_destroy(set);
3037
3038         if (rc == 0)
3039                 rc = laia->laia_result;
3040
3041         if (unlikely(rc == -EINPROGRESS)) {
3042                 retry = true;
3043                 set_current_state(TASK_INTERRUPTIBLE);
3044                 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC));
3045                 set_current_state(TASK_RUNNING);
3046                 if (!signal_pending(current) &&
3047                     thread_is_running(&lfsck->li_thread))
3048                         goto again;
3049
3050                 rc = -EINTR;
3051         }
3052
3053         if (rc != 0) {
3054                 struct lfsck_stop *stop = &info->lti_stop;
3055
3056                 CERROR("%s: cannot start LFSCK on some MDTs, "
3057                        "stop all: rc = %d\n",
3058                        lfsck_lfsck2name(lfsck), rc);
3059                 if (rc != -EALREADY) {
3060                         stop->ls_status = LS_FAILED;
3061                         stop->ls_flags = LPF_ALL_TGT | LPF_BROADCAST;
3062                         lfsck_stop_all(env, lfsck, stop);
3063                 }
3064         }
3065
3066         RETURN(rc);
3067 }
3068
3069 int lfsck_start(const struct lu_env *env, struct dt_device *key,
3070                 struct lfsck_start_param *lsp)
3071 {
3072         struct lfsck_start              *start  = lsp->lsp_start;
3073         struct lfsck_instance           *lfsck;
3074         struct lfsck_bookmark           *bk;
3075         struct ptlrpc_thread            *thread;
3076         struct lfsck_component          *com;
3077         struct l_wait_info               lwi    = { 0 };
3078         struct lfsck_thread_args        *lta;
3079         struct task_struct              *task;
3080         struct lfsck_tgt_descs          *ltds;
3081         struct lfsck_tgt_desc           *ltd;
3082         __u32                            idx;
3083         int                              rc     = 0;
3084         __u16                            valid  = 0;
3085         __u16                            flags  = 0;
3086         __u16                            type   = 1;
3087         ENTRY;
3088
3089         if (key->dd_rdonly)
3090                 RETURN(-EROFS);
3091
3092         lfsck = lfsck_instance_find(key, true, false);
3093         if (unlikely(lfsck == NULL))
3094                 RETURN(-ENXIO);
3095
3096         if (unlikely(lfsck->li_stopping))
3097                 GOTO(put, rc = -ENXIO);
3098
3099         /* System is not ready, try again later. */
3100         if (unlikely(lfsck->li_namespace == NULL ||
3101                      lfsck_dev_site(lfsck)->ss_server_fld == NULL))
3102                 GOTO(put, rc = -EINPROGRESS);
3103
3104         /* start == NULL means auto trigger paused LFSCK. */
3105         if (!start) {
3106                 if (list_empty(&lfsck->li_list_scan) ||
3107                     OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_AUTO))
3108                         GOTO(put, rc = 0);
3109         } else if (start->ls_flags & LPF_BROADCAST && !lfsck->li_master) {
3110                 CERROR("%s: only allow to specify '-A | -o' via MDS\n",
3111                        lfsck_lfsck2name(lfsck));
3112
3113                 GOTO(put, rc = -EPERM);
3114         }
3115
3116         bk = &lfsck->li_bookmark_ram;
3117         thread = &lfsck->li_thread;
3118         mutex_lock(&lfsck->li_mutex);
3119         spin_lock(&lfsck->li_lock);
3120         if (unlikely(thread_is_stopping(thread))) {
3121                 /* Someone is stopping the LFSCK. */
3122                 spin_unlock(&lfsck->li_lock);
3123                 GOTO(out, rc = -EBUSY);
3124         }
3125
3126         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
3127                 rc = -EALREADY;
3128                 if (unlikely(start == NULL)) {
3129                         spin_unlock(&lfsck->li_lock);
3130                         GOTO(out, rc);
3131                 }
3132
3133                 while (start->ls_active != 0) {
3134                         if (!(type & start->ls_active)) {
3135                                 type <<= 1;
3136                                 continue;
3137                         }
3138
3139                         com = __lfsck_component_find(lfsck, type,
3140                                                      &lfsck->li_list_scan);
3141                         if (com == NULL)
3142                                 com = __lfsck_component_find(lfsck, type,
3143                                                 &lfsck->li_list_double_scan);
3144                         if (com == NULL) {
3145                                 rc = -EOPNOTSUPP;
3146                                 break;
3147                         }
3148
3149                         if (com->lc_ops->lfsck_join != NULL) {
3150                                 rc = com->lc_ops->lfsck_join( env, com, lsp);
3151                                 if (rc != 0 && rc != -EALREADY)
3152                                         break;
3153                         }
3154                         start->ls_active &= ~type;
3155                         type <<= 1;
3156                 }
3157                 spin_unlock(&lfsck->li_lock);
3158                 GOTO(out, rc);
3159         }
3160         spin_unlock(&lfsck->li_lock);
3161
3162         lfsck->li_status = 0;
3163         lfsck->li_oit_over = 0;
3164         lfsck->li_start_unplug = 0;
3165         lfsck->li_drop_dryrun = 0;
3166         lfsck->li_new_scanned = 0;
3167
3168         /* For auto trigger. */
3169         if (start == NULL)
3170                 goto trigger;
3171
3172         start->ls_version = bk->lb_version;
3173
3174         if (start->ls_active != 0) {
3175                 struct lfsck_component *next;
3176
3177                 if (start->ls_active == LFSCK_TYPES_ALL)
3178                         start->ls_active = LFSCK_TYPES_SUPPORTED;
3179
3180                 if (start->ls_active & ~LFSCK_TYPES_SUPPORTED) {
3181                         start->ls_active &= ~LFSCK_TYPES_SUPPORTED;
3182                         GOTO(out, rc = -ENOTSUPP);
3183                 }
3184
3185                 list_for_each_entry_safe(com, next,
3186                                          &lfsck->li_list_scan, lc_link) {
3187                         if (!(com->lc_type & start->ls_active)) {
3188                                 rc = com->lc_ops->lfsck_post(env, com, 0,
3189                                                              false);
3190                                 if (rc != 0)
3191                                         GOTO(out, rc);
3192                         }
3193                 }
3194
3195                 while (start->ls_active != 0) {
3196                         if (type & start->ls_active) {
3197                                 com = __lfsck_component_find(lfsck, type,
3198                                                         &lfsck->li_list_idle);
3199                                 if (com != NULL)
3200                                         /* The component status will be updated
3201                                          * when its prep() is called later by
3202                                          * the LFSCK main engine. */
3203                                         list_move_tail(&com->lc_link,
3204                                                        &lfsck->li_list_scan);
3205                                 start->ls_active &= ~type;
3206                         }
3207                         type <<= 1;
3208                 }
3209         }
3210
3211         if (list_empty(&lfsck->li_list_scan)) {
3212                 /* The speed limit will be used to control both the LFSCK and
3213                  * low layer scrub (if applied), need to be handled firstly. */
3214                 if (start->ls_valid & LSV_SPEED_LIMIT) {
3215                         if (__lfsck_set_speed(lfsck, start->ls_speed_limit)) {
3216                                 rc = lfsck_bookmark_store(env, lfsck);
3217                                 if (rc != 0)
3218                                         GOTO(out, rc);
3219                         }
3220                 }
3221
3222                 goto trigger;
3223         }
3224
3225         if (start->ls_flags & LPF_RESET)
3226                 flags |= DOIF_RESET;
3227
3228         rc = lfsck_set_param(env, lfsck, start, !!(flags & DOIF_RESET));
3229         if (rc != 0)
3230                 GOTO(out, rc);
3231
3232         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
3233                 start->ls_active |= com->lc_type;
3234                 if (flags & DOIF_RESET) {
3235                         rc = com->lc_ops->lfsck_reset(env, com, false);
3236                         if (rc != 0)
3237                                 GOTO(out, rc);
3238                 }
3239         }
3240
3241         ltds = &lfsck->li_mdt_descs;
3242         down_read(&ltds->ltd_rw_sem);
3243         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
3244                 ltd = lfsck_ltd2tgt(ltds, idx);
3245                 LASSERT(ltd != NULL);
3246
3247                 ltd->ltd_layout_done = 0;
3248                 ltd->ltd_namespace_done = 0;
3249                 ltd->ltd_synced_failures = 0;
3250                 lfsck_reset_ltd_status(ltd, LFSCK_TYPE_NAMESPACE);
3251                 lfsck_reset_ltd_status(ltd, LFSCK_TYPE_LAYOUT);
3252                 list_del_init(&ltd->ltd_layout_phase_list);
3253                 list_del_init(&ltd->ltd_layout_list);
3254                 list_del_init(&ltd->ltd_namespace_phase_list);
3255                 list_del_init(&ltd->ltd_namespace_list);
3256         }
3257         up_read(&ltds->ltd_rw_sem);
3258
3259         ltds = &lfsck->li_ost_descs;
3260         down_read(&ltds->ltd_rw_sem);
3261         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
3262                 ltd = lfsck_ltd2tgt(ltds, idx);
3263                 LASSERT(ltd != NULL);
3264
3265                 ltd->ltd_layout_done = 0;
3266                 ltd->ltd_synced_failures = 0;
3267                 lfsck_reset_ltd_status(ltd, LFSCK_TYPE_LAYOUT);
3268                 list_del_init(&ltd->ltd_layout_phase_list);
3269                 list_del_init(&ltd->ltd_layout_list);
3270         }
3271         up_read(&ltds->ltd_rw_sem);
3272
3273 trigger:
3274         lfsck->li_args_dir = LUDA_64BITHASH | LUDA_VERIFY | LUDA_TYPE;
3275         if (bk->lb_param & LPF_DRYRUN)
3276                 lfsck->li_args_dir |= LUDA_VERIFY_DRYRUN;
3277
3278         if (start != NULL && start->ls_valid & LSV_ERROR_HANDLE) {
3279                 valid |= DOIV_ERROR_HANDLE;
3280                 if (start->ls_flags & LPF_FAILOUT)
3281                         flags |= DOIF_FAILOUT;
3282         }
3283
3284         if (start != NULL && start->ls_valid & LSV_DRYRUN) {
3285                 valid |= DOIV_DRYRUN;
3286                 if (start->ls_flags & LPF_DRYRUN)
3287                         flags |= DOIF_DRYRUN;
3288         }
3289
3290         if (!list_empty(&lfsck->li_list_scan))
3291                 flags |= DOIF_OUTUSED;
3292
3293         lfsck->li_args_oit = (flags << DT_OTABLE_IT_FLAGS_SHIFT) | valid;
3294         lta = lfsck_thread_args_init(lfsck, NULL, lsp);
3295         if (IS_ERR(lta))
3296                 GOTO(out, rc = PTR_ERR(lta));
3297
3298         __lfsck_set_speed(lfsck, bk->lb_speed_limit);
3299         spin_lock(&lfsck->li_lock);
3300         thread_set_flags(thread, SVC_STARTING);
3301         spin_unlock(&lfsck->li_lock);
3302         task = kthread_run(lfsck_master_engine, lta, "lfsck");
3303         if (IS_ERR(task)) {
3304                 rc = PTR_ERR(task);
3305                 CERROR("%s: cannot start LFSCK thread: rc = %d\n",
3306                        lfsck_lfsck2name(lfsck), rc);
3307                 lfsck_thread_args_fini(lta);
3308
3309                 GOTO(out, rc);
3310         }
3311
3312         l_wait_event(thread->t_ctl_waitq,
3313                      thread_is_running(thread) ||
3314                      thread_is_stopped(thread),
3315                      &lwi);
3316         if (start == NULL || !(start->ls_flags & LPF_BROADCAST)) {
3317                 lfsck->li_start_unplug = 1;
3318                 wake_up_all(&thread->t_ctl_waitq);
3319
3320                 GOTO(out, rc = 0);
3321         }
3322
3323         /* release lfsck::li_mutex to avoid deadlock. */
3324         mutex_unlock(&lfsck->li_mutex);
3325         rc = lfsck_start_all(env, lfsck, start);
3326         if (rc != 0) {
3327                 spin_lock(&lfsck->li_lock);
3328                 if (thread_is_stopped(thread)) {
3329                         spin_unlock(&lfsck->li_lock);
3330                 } else {
3331                         lfsck->li_status = LS_FAILED;
3332                         lfsck->li_flags = 0;
3333                         thread_set_flags(thread, SVC_STOPPING);
3334                         spin_unlock(&lfsck->li_lock);
3335
3336                         lfsck->li_start_unplug = 1;
3337                         wake_up_all(&thread->t_ctl_waitq);
3338                         l_wait_event(thread->t_ctl_waitq,
3339                                      thread_is_stopped(thread),
3340                                      &lwi);
3341                 }
3342         } else {
3343                 lfsck->li_start_unplug = 1;
3344                 wake_up_all(&thread->t_ctl_waitq);
3345         }
3346
3347         GOTO(put, rc);
3348
3349 out:
3350         mutex_unlock(&lfsck->li_mutex);
3351
3352 put:
3353         lfsck_instance_put(env, lfsck);
3354
3355         return rc < 0 ? rc : 0;
3356 }
3357 EXPORT_SYMBOL(lfsck_start);
3358
3359 int lfsck_stop(const struct lu_env *env, struct dt_device *key,
3360                struct lfsck_stop *stop)
3361 {
3362         struct lfsck_instance   *lfsck;
3363         struct ptlrpc_thread    *thread;
3364         struct l_wait_info       lwi    = { 0 };
3365         int                      rc     = 0;
3366         int                      rc1    = 0;
3367         ENTRY;
3368
3369         lfsck = lfsck_instance_find(key, true, false);
3370         if (unlikely(lfsck == NULL))
3371                 RETURN(-ENXIO);
3372
3373         thread = &lfsck->li_thread;
3374         if (stop && stop->ls_flags & LPF_BROADCAST && !lfsck->li_master) {
3375                 CERROR("%s: only allow to specify '-A' via MDS\n",
3376                        lfsck_lfsck2name(lfsck));
3377                 GOTO(put, rc = -EPERM);
3378         }
3379
3380         spin_lock(&lfsck->li_lock);
3381         /* The target is umounted */
3382         if (stop && stop->ls_status == LS_PAUSED)
3383                 lfsck->li_stopping = 1;
3384
3385         if (thread_is_init(thread) || thread_is_stopped(thread))
3386                 /* no error if LFSCK stopped already, or not started */
3387                 GOTO(unlock, rc = 0);
3388
3389         if (thread_is_stopping(thread))
3390                 /* Someone is stopping LFSCK. */
3391                 GOTO(unlock, rc = -EINPROGRESS);
3392
3393         if (stop) {
3394                 lfsck->li_status = stop->ls_status;
3395                 lfsck->li_flags = stop->ls_flags;
3396         } else {
3397                 lfsck->li_status = LS_STOPPED;
3398                 lfsck->li_flags = 0;
3399         }
3400
3401         thread_set_flags(thread, SVC_STOPPING);
3402
3403         LASSERT(lfsck->li_task != NULL);
3404         force_sig(SIGINT, lfsck->li_task);
3405
3406         if (lfsck->li_master) {
3407                 struct lfsck_component *com;
3408                 struct lfsck_assistant_data *lad;
3409
3410                 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
3411                         lad = com->lc_data;
3412                         spin_lock(&lad->lad_lock);
3413                         if (lad->lad_task != NULL)
3414                                 force_sig(SIGINT, lad->lad_task);
3415                         spin_unlock(&lad->lad_lock);
3416                 }
3417
3418                 list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
3419                         lad = com->lc_data;
3420                         spin_lock(&lad->lad_lock);
3421                         if (lad->lad_task != NULL)
3422                                 force_sig(SIGINT, lad->lad_task);
3423                         spin_unlock(&lad->lad_lock);
3424                 }
3425         }
3426
3427         wake_up_all(&thread->t_ctl_waitq);
3428         spin_unlock(&lfsck->li_lock);
3429         if (stop && stop->ls_flags & LPF_BROADCAST)
3430                 rc1 = lfsck_stop_all(env, lfsck, stop);
3431
3432         /* It was me set the status as 'stopping' just now, if it is not
3433          * 'stopping' now, then either stopped, or re-started by race. */
3434         l_wait_event(thread->t_ctl_waitq,
3435                      !thread_is_stopping(thread),
3436                      &lwi);
3437
3438         GOTO(put, rc = 0);
3439
3440 unlock:
3441         spin_unlock(&lfsck->li_lock);
3442 put:
3443         lfsck_instance_put(env, lfsck);
3444
3445         return rc != 0 ? rc : rc1;
3446 }
3447 EXPORT_SYMBOL(lfsck_stop);
3448
3449 int lfsck_in_notify_local(const struct lu_env *env, struct dt_device *key,
3450                           struct lfsck_req_local *lrl, struct thandle *th)
3451 {
3452         struct lfsck_instance *lfsck;
3453         struct lfsck_component *com;
3454         int rc = -EOPNOTSUPP;
3455         ENTRY;
3456
3457         lfsck = lfsck_instance_find(key, true, false);
3458         if (unlikely(!lfsck))
3459                 RETURN(-ENXIO);
3460
3461         com = lfsck_component_find(lfsck, lrl->lrl_active);
3462         if (likely(com && com->lc_ops->lfsck_in_notify_local)) {
3463                 rc = com->lc_ops->lfsck_in_notify_local(env, com, lrl, th);
3464                 lfsck_component_put(env, com);
3465         }
3466
3467         lfsck_instance_put(env, lfsck);
3468
3469         RETURN(rc);
3470 }
3471 EXPORT_SYMBOL(lfsck_in_notify_local);
3472
3473 int lfsck_in_notify(const struct lu_env *env, struct dt_device *key,
3474                     struct lfsck_request *lr)
3475 {
3476         int rc = -EOPNOTSUPP;
3477         ENTRY;
3478
3479         switch (lr->lr_event) {
3480         case LE_START: {
3481                 struct lfsck_start       *start = &lfsck_env_info(env)->lti_start;
3482                 struct lfsck_start_param  lsp;
3483
3484                 memset(start, 0, sizeof(*start));
3485                 start->ls_valid = lr->lr_valid;
3486                 start->ls_speed_limit = lr->lr_speed;
3487                 start->ls_version = lr->lr_version;
3488                 start->ls_active = lr->lr_active;
3489                 start->ls_flags = lr->lr_param & ~LPF_BROADCAST;
3490                 start->ls_async_windows = lr->lr_async_windows;
3491
3492                 lsp.lsp_start = start;
3493                 lsp.lsp_index = lr->lr_index;
3494                 lsp.lsp_index_valid = 1;
3495                 rc = lfsck_start(env, key, &lsp);
3496                 break;
3497         }
3498         case LE_STOP: {
3499                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
3500
3501                 memset(stop, 0, sizeof(*stop));
3502                 stop->ls_status = lr->lr_status;
3503                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
3504                 rc = lfsck_stop(env, key, stop);
3505                 break;
3506         }
3507         case LE_PHASE1_DONE:
3508         case LE_PHASE2_DONE:
3509         case LE_PEER_EXIT:
3510         case LE_CONDITIONAL_DESTROY:
3511         case LE_SET_LMV_MASTER:
3512         case LE_SET_LMV_SLAVE:
3513         case LE_PAIRS_VERIFY: {
3514                 struct lfsck_instance  *lfsck;
3515                 struct lfsck_component *com;
3516
3517                 lfsck = lfsck_instance_find(key, true, false);
3518                 if (unlikely(lfsck == NULL))
3519                         RETURN(-ENXIO);
3520
3521                 com = lfsck_component_find(lfsck, lr->lr_active);
3522                 if (likely(com)) {
3523                         rc = com->lc_ops->lfsck_in_notify(env, com, lr);
3524                         lfsck_component_put(env, com);
3525                 }
3526
3527                 lfsck_instance_put(env, lfsck);
3528                 break;
3529         }
3530         default:
3531                 break;
3532         }
3533
3534         RETURN(rc);
3535 }
3536 EXPORT_SYMBOL(lfsck_in_notify);
3537
3538 int lfsck_query(const struct lu_env *env, struct dt_device *key,
3539                 struct lfsck_request *req, struct lfsck_reply *rep,
3540                 struct lfsck_query *que)
3541 {
3542         struct lfsck_instance  *lfsck;
3543         struct lfsck_component *com;
3544         int                     i;
3545         int                     rc = 0;
3546         __u16                   type;
3547         ENTRY;
3548
3549         lfsck = lfsck_instance_find(key, true, false);
3550         if (unlikely(lfsck == NULL))
3551                 RETURN(-ENXIO);
3552
3553         if (que != NULL) {
3554                 if (que->lu_types == LFSCK_TYPES_ALL)
3555                         que->lu_types =
3556                                 LFSCK_TYPES_SUPPORTED & ~LFSCK_TYPE_SCRUB;
3557
3558                 if (que->lu_types & ~LFSCK_TYPES_SUPPORTED) {
3559                         que->lu_types &= ~LFSCK_TYPES_SUPPORTED;
3560
3561                         GOTO(out, rc = -ENOTSUPP);
3562                 }
3563
3564                 for (i = 0, type = 1 << i; i < LFSCK_TYPE_BITS;
3565                      i++, type = 1 << i) {
3566                         if (!(que->lu_types & type))
3567                                 continue;
3568
3569 again:
3570                         com = lfsck_component_find(lfsck, type);
3571                         if (unlikely(com == NULL))
3572                                 GOTO(out, rc = -ENOTSUPP);
3573
3574                         memset(que->lu_mdts_count[i], 0,
3575                                sizeof(__u32) * (LS_MAX + 1));
3576                         memset(que->lu_osts_count[i], 0,
3577                                sizeof(__u32) * (LS_MAX + 1));
3578                         que->lu_repaired[i] = 0;
3579                         rc = com->lc_ops->lfsck_query(env, com, req, rep,
3580                                                       que, i);
3581                         lfsck_component_put(env, com);
3582                         if  (rc < 0)
3583                                 GOTO(out, rc);
3584                 }
3585
3586                 if (!(que->lu_flags & LPF_WAIT))
3587                         GOTO(out, rc);
3588
3589                 for (i = 0, type = 1 << i; i < LFSCK_TYPE_BITS;
3590                      i++, type = 1 << i) {
3591                         if (!(que->lu_types & type))
3592                                 continue;
3593
3594                         if (que->lu_mdts_count[i][LS_SCANNING_PHASE1] != 0 ||
3595                             que->lu_mdts_count[i][LS_SCANNING_PHASE2] != 0 ||
3596                             que->lu_osts_count[i][LS_SCANNING_PHASE1] != 0 ||
3597                             que->lu_osts_count[i][LS_SCANNING_PHASE2] != 0) {
3598                                 struct l_wait_info lwi;
3599
3600                                 /* If it is required to wait, then sleep
3601                                  * 3 seconds and try to query again. */
3602                                 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(3),
3603                                                        NULL,
3604                                                        LWI_ON_SIGNAL_NOOP,
3605                                                        NULL);
3606                                 rc = l_wait_event(lfsck->li_thread.t_ctl_waitq,
3607                                                   0, &lwi);
3608                                 if (rc == -ETIMEDOUT)
3609                                         goto again;
3610                         }
3611                 }
3612         } else {
3613                 com = lfsck_component_find(lfsck, req->lr_active);
3614                 if (likely(com != NULL)) {
3615                         rc = com->lc_ops->lfsck_query(env, com, req, rep,
3616                                                       que, -1);
3617                         lfsck_component_put(env, com);
3618                 } else {
3619                         rc = -ENOTSUPP;
3620                 }
3621         }
3622
3623         GOTO(out, rc);
3624
3625 out:
3626         lfsck_instance_put(env, lfsck);
3627         return rc;
3628 }
3629 EXPORT_SYMBOL(lfsck_query);
3630
3631 int lfsck_register_namespace(const struct lu_env *env, struct dt_device *key,
3632                              struct ldlm_namespace *ns)
3633 {
3634         struct lfsck_instance  *lfsck;
3635         int                     rc      = -ENXIO;
3636
3637         lfsck = lfsck_instance_find(key, true, false);
3638         if (likely(lfsck != NULL)) {
3639                 lfsck->li_namespace = ns;
3640                 lfsck_instance_put(env, lfsck);
3641                 rc = 0;
3642         }
3643
3644         return rc;
3645 }
3646 EXPORT_SYMBOL(lfsck_register_namespace);
3647
3648 int lfsck_register(const struct lu_env *env, struct dt_device *key,
3649                    struct dt_device *next, struct obd_device *obd,
3650                    lfsck_out_notify notify, void *notify_data, bool master)
3651 {
3652         struct lfsck_instance   *lfsck;
3653         struct dt_object        *root  = NULL;
3654         struct dt_object        *obj   = NULL;
3655         struct lu_fid           *fid   = &lfsck_env_info(env)->lti_fid;
3656         int                      rc;
3657         ENTRY;
3658
3659         lfsck = lfsck_instance_find(key, false, false);
3660         if (unlikely(lfsck != NULL))
3661                 RETURN(-EEXIST);
3662
3663         OBD_ALLOC_PTR(lfsck);
3664         if (lfsck == NULL)
3665                 RETURN(-ENOMEM);
3666
3667         mutex_init(&lfsck->li_mutex);
3668         spin_lock_init(&lfsck->li_lock);
3669         INIT_LIST_HEAD(&lfsck->li_link);
3670         INIT_LIST_HEAD(&lfsck->li_list_scan);
3671         INIT_LIST_HEAD(&lfsck->li_list_dir);
3672         INIT_LIST_HEAD(&lfsck->li_list_double_scan);
3673         INIT_LIST_HEAD(&lfsck->li_list_idle);
3674         INIT_LIST_HEAD(&lfsck->li_list_lmv);
3675         atomic_set(&lfsck->li_ref, 1);
3676         atomic_set(&lfsck->li_double_scan_count, 0);
3677         init_waitqueue_head(&lfsck->li_thread.t_ctl_waitq);
3678         lfsck->li_out_notify = notify;
3679         lfsck->li_out_notify_data = notify_data;
3680         lfsck->li_next = next;
3681         lfsck->li_bottom = key;
3682         lfsck->li_obd = obd;
3683
3684         rc = lfsck_tgt_descs_init(&lfsck->li_ost_descs);
3685         if (rc != 0)
3686                 GOTO(out, rc);
3687
3688         rc = lfsck_tgt_descs_init(&lfsck->li_mdt_descs);
3689         if (rc != 0)
3690                 GOTO(out, rc);
3691
3692         fid->f_seq = FID_SEQ_LOCAL_NAME;
3693         fid->f_oid = 1;
3694         fid->f_ver = 0;
3695         rc = local_oid_storage_init(env, key, fid, &lfsck->li_los);
3696         if (rc != 0)
3697                 GOTO(out, rc);
3698
3699         rc = dt_root_get(env, key, fid);
3700         if (rc != 0)
3701                 GOTO(out, rc);
3702
3703         root = dt_locate(env, key, fid);
3704         if (IS_ERR(root))
3705                 GOTO(out, rc = PTR_ERR(root));
3706
3707         if (unlikely(!dt_try_as_dir(env, root)))
3708                 GOTO(out, rc = -ENOTDIR);
3709
3710         lfsck->li_local_root_fid = *fid;
3711         if (master) {
3712                 lfsck->li_master = 1;
3713                 if (lfsck_dev_idx(lfsck) == 0) {
3714                         struct lu_fid *pfid = &lfsck_env_info(env)->lti_fid2;
3715                         const struct lu_name *cname;
3716
3717                         rc = dt_lookup(env, root,
3718                                 (struct dt_rec *)(&lfsck->li_global_root_fid),
3719                                 (const struct dt_key *)"ROOT");
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                         if (unlikely(!dt_try_as_dir(env, obj)))
3728                                 GOTO(out, rc = -ENOTDIR);
3729
3730                         rc = dt_lookup(env, obj, (struct dt_rec *)fid,
3731                                 (const struct dt_key *)dotlustre);
3732                         if (rc != 0)
3733                                 GOTO(out, rc);
3734
3735                         lfsck_object_put(env, obj);
3736                         obj = dt_locate(env, key, fid);
3737                         if (IS_ERR(obj))
3738                                 GOTO(out, rc = PTR_ERR(obj));
3739
3740                         cname = lfsck_name_get_const(env, dotlustre,
3741                                                      strlen(dotlustre));
3742                         rc = lfsck_verify_linkea(env, obj, cname,
3743                                                  &lfsck->li_global_root_fid);
3744                         if (rc != 0)
3745                                 GOTO(out, rc);
3746
3747                         if (unlikely(!dt_try_as_dir(env, obj)))
3748                                 GOTO(out, rc = -ENOTDIR);
3749
3750                         *pfid = *fid;
3751                         rc = dt_lookup(env, obj, (struct dt_rec *)fid,
3752                                        (const struct dt_key *)lostfound);
3753                         if (rc != 0)
3754                                 GOTO(out, rc);
3755
3756                         lfsck_object_put(env, obj);
3757                         obj = dt_locate(env, key, fid);
3758                         if (IS_ERR(obj))
3759                                 GOTO(out, rc = PTR_ERR(obj));
3760
3761                         cname = lfsck_name_get_const(env, lostfound,
3762                                                      strlen(lostfound));
3763                         rc = lfsck_verify_linkea(env, obj, cname, pfid);
3764                         if (rc != 0)
3765                                 GOTO(out, rc);
3766
3767                         lfsck_object_put(env, obj);
3768                         obj = NULL;
3769                 }
3770         }
3771
3772         fid->f_seq = FID_SEQ_LOCAL_FILE;
3773         fid->f_oid = OTABLE_IT_OID;
3774         fid->f_ver = 0;
3775         obj = dt_locate(env, key, fid);
3776         if (IS_ERR(obj))
3777                 GOTO(out, rc = PTR_ERR(obj));
3778
3779         rc = obj->do_ops->do_index_try(env, obj, &dt_otable_features);
3780         if (rc != 0)
3781                 GOTO(out, rc);
3782
3783         lfsck->li_obj_oit = obj;
3784         obj = local_file_find_or_create(env, lfsck->li_los, root, LFSCK_DIR,
3785                                         S_IFDIR | S_IRUGO | S_IWUSR);
3786         if (IS_ERR(obj))
3787                 GOTO(out, rc = PTR_ERR(obj));
3788
3789         lu_object_get(&obj->do_lu);
3790         lfsck->li_lfsck_dir = obj;
3791         rc = lfsck_bookmark_setup(env, lfsck);
3792         if (rc != 0)
3793                 GOTO(out, rc);
3794
3795         if (master) {
3796                 rc = lfsck_fid_init(lfsck);
3797                 if (rc < 0)
3798                         GOTO(out, rc);
3799
3800                 rc = lfsck_namespace_setup(env, lfsck);
3801                 if (rc < 0)
3802                         GOTO(out, rc);
3803         }
3804
3805         rc = lfsck_layout_setup(env, lfsck);
3806         if (rc < 0)
3807                 GOTO(out, rc);
3808
3809         /* XXX: more LFSCK components initialization to be added here. */
3810
3811         rc = lfsck_instance_add(lfsck);
3812         if (rc == 0)
3813                 rc = lfsck_add_target_from_orphan(env, lfsck);
3814 out:
3815         if (obj != NULL && !IS_ERR(obj))
3816                 lfsck_object_put(env, obj);
3817         if (root != NULL && !IS_ERR(root))
3818                 lfsck_object_put(env, root);
3819         if (rc != 0)
3820                 lfsck_instance_cleanup(env, lfsck);
3821         return rc;
3822 }
3823 EXPORT_SYMBOL(lfsck_register);
3824
3825 void lfsck_degister(const struct lu_env *env, struct dt_device *key)
3826 {
3827         struct lfsck_instance *lfsck;
3828
3829         lfsck = lfsck_instance_find(key, false, true);
3830         if (lfsck != NULL)
3831                 lfsck_instance_put(env, lfsck);
3832 }
3833 EXPORT_SYMBOL(lfsck_degister);
3834
3835 int lfsck_add_target(const struct lu_env *env, struct dt_device *key,
3836                      struct dt_device *tgt, struct obd_export *exp,
3837                      __u32 index, bool for_ost)
3838 {
3839         struct lfsck_instance   *lfsck;
3840         struct lfsck_tgt_desc   *ltd;
3841         int                      rc;
3842         ENTRY;
3843
3844         OBD_ALLOC_PTR(ltd);
3845         if (ltd == NULL)
3846                 RETURN(-ENOMEM);
3847
3848         ltd->ltd_tgt = tgt;
3849         ltd->ltd_key = key;
3850         ltd->ltd_exp = exp;
3851         INIT_LIST_HEAD(&ltd->ltd_orphan_list);
3852         INIT_LIST_HEAD(&ltd->ltd_layout_list);
3853         INIT_LIST_HEAD(&ltd->ltd_layout_phase_list);
3854         INIT_LIST_HEAD(&ltd->ltd_namespace_list);
3855         INIT_LIST_HEAD(&ltd->ltd_namespace_phase_list);
3856         atomic_set(&ltd->ltd_ref, 1);
3857         ltd->ltd_index = index;
3858
3859         spin_lock(&lfsck_instance_lock);
3860         lfsck = __lfsck_instance_find(key, true, false);
3861         if (lfsck == NULL) {
3862                 if (for_ost)
3863                         list_add_tail(&ltd->ltd_orphan_list,
3864                                       &lfsck_ost_orphan_list);
3865                 else
3866                         list_add_tail(&ltd->ltd_orphan_list,
3867                                       &lfsck_mdt_orphan_list);
3868                 spin_unlock(&lfsck_instance_lock);
3869
3870                 RETURN(0);
3871         }
3872         spin_unlock(&lfsck_instance_lock);
3873
3874         rc = __lfsck_add_target(env, lfsck, ltd, for_ost, false);
3875         if (rc != 0)
3876                 lfsck_tgt_put(ltd);
3877
3878         lfsck_instance_put(env, lfsck);
3879
3880         RETURN(rc);
3881 }
3882 EXPORT_SYMBOL(lfsck_add_target);
3883
3884 void lfsck_del_target(const struct lu_env *env, struct dt_device *key,
3885                       struct dt_device *tgt, __u32 index, bool for_ost)
3886 {
3887         struct lfsck_instance   *lfsck;
3888         struct lfsck_tgt_descs  *ltds;
3889         struct lfsck_tgt_desc   *ltd;
3890         struct list_head        *head;
3891
3892         if (for_ost)
3893                 head = &lfsck_ost_orphan_list;
3894         else
3895                 head = &lfsck_mdt_orphan_list;
3896
3897         spin_lock(&lfsck_instance_lock);
3898         list_for_each_entry(ltd, head, ltd_orphan_list) {
3899                 if (ltd->ltd_tgt == tgt) {
3900                         list_del_init(&ltd->ltd_orphan_list);
3901                         spin_unlock(&lfsck_instance_lock);
3902                         lfsck_tgt_put(ltd);
3903
3904                         return;
3905                 }
3906         }
3907
3908         ltd = NULL;
3909         lfsck = __lfsck_instance_find(key, true, false);
3910         spin_unlock(&lfsck_instance_lock);
3911         if (unlikely(lfsck == NULL))
3912                 return;
3913
3914         if (for_ost)
3915                 ltds = &lfsck->li_ost_descs;
3916         else
3917                 ltds = &lfsck->li_mdt_descs;
3918
3919         down_write(&ltds->ltd_rw_sem);
3920         LASSERT(ltds->ltd_tgts_bitmap != NULL);
3921
3922         if (unlikely(index >= ltds->ltd_tgts_bitmap->size))
3923                 goto unlock;
3924
3925         ltd = lfsck_ltd2tgt(ltds, index);
3926         if (unlikely(ltd == NULL))
3927                 goto unlock;
3928
3929         LASSERT(ltds->ltd_tgtnr > 0);
3930
3931         ltds->ltd_tgtnr--;
3932         cfs_bitmap_clear(ltds->ltd_tgts_bitmap, index);
3933         lfsck_assign_tgt(ltds, NULL, index);
3934
3935 unlock:
3936         if (ltd == NULL) {
3937                 if (for_ost)
3938                         head = &lfsck->li_ost_descs.ltd_orphan;
3939                 else
3940                         head = &lfsck->li_mdt_descs.ltd_orphan;
3941
3942                 list_for_each_entry(ltd, head, ltd_orphan_list) {
3943                         if (ltd->ltd_tgt == tgt) {
3944                                 list_del_init(&ltd->ltd_orphan_list);
3945                                 break;
3946                         }
3947                 }
3948         }
3949
3950         up_write(&ltds->ltd_rw_sem);
3951         if (ltd != NULL) {
3952                 spin_lock(&ltds->ltd_lock);
3953                 ltd->ltd_dead = 1;
3954                 spin_unlock(&ltds->ltd_lock);
3955                 lfsck_stop_notify(env, lfsck, ltds, ltd, LFSCK_TYPE_NAMESPACE);
3956                 lfsck_stop_notify(env, lfsck, ltds, ltd, LFSCK_TYPE_LAYOUT);
3957                 lfsck_tgt_put(ltd);
3958         }
3959
3960         lfsck_instance_put(env, lfsck);
3961 }
3962 EXPORT_SYMBOL(lfsck_del_target);
3963
3964 static int __init lfsck_init(void)
3965 {
3966         int rc;
3967
3968         INIT_LIST_HEAD(&lfsck_instance_list);
3969         INIT_LIST_HEAD(&lfsck_ost_orphan_list);
3970         INIT_LIST_HEAD(&lfsck_mdt_orphan_list);
3971         lfsck_key_init_generic(&lfsck_thread_key, NULL);
3972         rc = lu_context_key_register(&lfsck_thread_key);
3973         if (!rc) {
3974                 tgt_register_lfsck_in_notify_local(lfsck_in_notify_local);
3975                 tgt_register_lfsck_in_notify(lfsck_in_notify);
3976                 tgt_register_lfsck_query(lfsck_query);
3977         }
3978
3979         return rc;
3980 }
3981
3982 static void __exit lfsck_exit(void)
3983 {
3984         struct lfsck_tgt_desc *ltd;
3985         struct lfsck_tgt_desc *next;
3986
3987         LASSERT(list_empty(&lfsck_instance_list));
3988
3989         list_for_each_entry_safe(ltd, next, &lfsck_ost_orphan_list,
3990                                  ltd_orphan_list) {
3991                 list_del_init(&ltd->ltd_orphan_list);
3992                 lfsck_tgt_put(ltd);
3993         }
3994
3995         list_for_each_entry_safe(ltd, next, &lfsck_mdt_orphan_list,
3996                                  ltd_orphan_list) {
3997                 list_del_init(&ltd->ltd_orphan_list);
3998                 lfsck_tgt_put(ltd);
3999         }
4000
4001         lu_context_key_degister(&lfsck_thread_key);
4002 }
4003
4004 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
4005 MODULE_DESCRIPTION("Lustre File System Checker");
4006 MODULE_VERSION(LUSTRE_VERSION_STRING);
4007 MODULE_LICENSE("GPL");
4008
4009 module_init(lfsck_init);
4010 module_exit(lfsck_exit);