lustre/lfsck/lfsck_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2013, 2017, Intel Corporation.
  24  */
  25 /*
  26  * lustre/lfsck/lfsck_lib.c
  27  *
  28  * Author: Fan, Yong <fan.yong@intel.com>
  29  */
  30
  31 #define DEBUG_SUBSYSTEM S_LFSCK
  32
  33 #include <linux/kthread.h>
  34 #include <linux/sched.h>
  35 #include <linux/list.h>
  36 #include <linux/delay.h>
  37 #include <lu_object.h>
  38 #include <dt_object.h>
  39 #include <md_object.h>
  40 #include <lustre_fld.h>
  41 #include <lustre_lib.h>
  42 #include <lustre_net.h>
  43 #include <lustre_lfsck.h>
  44
  45 #include "lfsck_internal.h"
  46
  47 #define LFSCK_CHECKPOINT_SKIP   1
  48
  49 /* define lfsck thread key */
  50 LU_KEY_INIT(lfsck, struct lfsck_thread_info);
  51
  52 static void lfsck_key_fini(const struct lu_context *ctx,
  53                            struct lu_context_key *key, void *data)
  54 {
  55         struct lfsck_thread_info *info = data;
  56
  57         lu_buf_free(&info->lti_linkea_buf);
  58         lu_buf_free(&info->lti_linkea_buf2);
  59         lu_buf_free(&info->lti_big_buf);
  60         OBD_FREE_PTR(info);
  61 }
  62
  63 LU_CONTEXT_KEY_DEFINE(lfsck, LCT_MD_THREAD | LCT_DT_THREAD);
  64 LU_KEY_INIT_GENERIC(lfsck);
  65
  66 static struct list_head lfsck_instance_list;
  67 static struct list_head lfsck_ost_orphan_list;
  68 static struct list_head lfsck_mdt_orphan_list;
  69 static DEFINE_SPINLOCK(lfsck_instance_lock);
  70
  71 const char *lfsck_flags_names[] = {
  72         "scanned-once",
  73         "inconsistent",
  74         "upgrade",
  75         "incomplete",
  76         "crashed_lastid",
  77         NULL
  78 };
  79
  80 const char *lfsck_param_names[] = {
  81         NULL,
  82         "failout",
  83         "dryrun",
  84         "all_targets",
  85         "broadcast",
  86         "orphan",
  87         "create_ostobj",
  88         "create_mdtobj",
  89         NULL,
  90         "delay_create_ostobj",
  91         NULL
  92 };
  93
  94 enum lfsck_verify_lpf_types {
  95         LVLT_BY_BOOKMARK        = 0,
  96         LVLT_BY_NAMEENTRY       = 1,
  97 };
  98
  99 static inline void
 100 lfsck_reset_ltd_status(struct lfsck_tgt_desc *ltd, enum lfsck_type type)
 101 {
 102         if (type == LFSCK_TYPE_LAYOUT) {
 103                 ltd->ltd_layout_status = LS_MAX;
 104                 ltd->ltd_layout_repaired = 0;
 105         } else {
 106                 ltd->ltd_namespace_status = LS_MAX;
 107                 ltd->ltd_namespace_repaired = 0;
 108         }
 109 }
 110
 111 static int lfsck_tgt_descs_init(struct lfsck_tgt_descs *ltds)
 112 {
 113         spin_lock_init(&ltds->ltd_lock);
 114         init_rwsem(&ltds->ltd_rw_sem);
 115         INIT_LIST_HEAD(&ltds->ltd_orphan);
 116         ltds->ltd_tgts_bitmap = CFS_ALLOCATE_BITMAP(BITS_PER_LONG);
 117         if (ltds->ltd_tgts_bitmap == NULL)
 118                 return -ENOMEM;
 119
 120         return 0;
 121 }
 122
 123 static void lfsck_tgt_descs_fini(struct lfsck_tgt_descs *ltds)
 124 {
 125         struct lfsck_tgt_desc   *ltd;
 126         struct lfsck_tgt_desc   *next;
 127         int                      idx;
 128
 129         down_write(&ltds->ltd_rw_sem);
 130
 131         list_for_each_entry_safe(ltd, next, &ltds->ltd_orphan,
 132                                  ltd_orphan_list) {
 133                 list_del_init(&ltd->ltd_orphan_list);
 134                 lfsck_tgt_put(ltd);
 135         }
 136
 137         if (unlikely(ltds->ltd_tgts_bitmap == NULL)) {
 138                 up_write(&ltds->ltd_rw_sem);
 139
 140                 return;
 141         }
 142
 143         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
 144                 ltd = lfsck_ltd2tgt(ltds, idx);
 145                 if (likely(ltd != NULL)) {
 146                         LASSERT(list_empty(&ltd->ltd_layout_list));
 147                         LASSERT(list_empty(&ltd->ltd_layout_phase_list));
 148                         LASSERT(list_empty(&ltd->ltd_namespace_list));
 149                         LASSERT(list_empty(&ltd->ltd_namespace_phase_list));
 150
 151                         ltds->ltd_tgtnr--;
 152                         cfs_bitmap_clear(ltds->ltd_tgts_bitmap, idx);
 153                         lfsck_assign_tgt(ltds, NULL, idx);
 154                         lfsck_tgt_put(ltd);
 155                 }
 156         }
 157
 158         LASSERTF(ltds->ltd_tgtnr == 0, "tgt count unmatched: %d\n",
 159                  ltds->ltd_tgtnr);
 160
 161         for (idx = 0; idx < TGT_PTRS; idx++) {
 162                 if (ltds->ltd_tgts_idx[idx] != NULL) {
 163                         OBD_FREE_PTR(ltds->ltd_tgts_idx[idx]);
 164                         ltds->ltd_tgts_idx[idx] = NULL;
 165                 }
 166         }
 167
 168         CFS_FREE_BITMAP(ltds->ltd_tgts_bitmap);
 169         ltds->ltd_tgts_bitmap = NULL;
 170         up_write(&ltds->ltd_rw_sem);
 171 }
 172
 173 static int __lfsck_add_target(const struct lu_env *env,
 174                               struct lfsck_instance *lfsck,
 175                               struct lfsck_tgt_desc *ltd,
 176                               bool for_ost, bool locked)
 177 {
 178         struct lfsck_tgt_descs *ltds;
 179         __u32                   index = ltd->ltd_index;
 180         int                     rc    = 0;
 181         ENTRY;
 182
 183         if (for_ost)
 184                 ltds = &lfsck->li_ost_descs;
 185         else
 186                 ltds = &lfsck->li_mdt_descs;
 187
 188         if (!locked)
 189                 down_write(&ltds->ltd_rw_sem);
 190
 191         LASSERT(ltds->ltd_tgts_bitmap != NULL);
 192
 193         if (index >= ltds->ltd_tgts_bitmap->size) {
 194                 __u32 newsize = max((__u32)ltds->ltd_tgts_bitmap->size,
 195                                     (__u32)BITS_PER_LONG);
 196                 struct cfs_bitmap *old_bitmap = ltds->ltd_tgts_bitmap;
 197                 struct cfs_bitmap *new_bitmap;
 198
 199                 while (newsize < index + 1)
 200                         newsize <<= 1;
 201
 202                 new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
 203                 if (new_bitmap == NULL)
 204                         GOTO(unlock, rc = -ENOMEM);
 205
 206                 if (ltds->ltd_tgtnr > 0)
 207                         cfs_bitmap_copy(new_bitmap, old_bitmap);
 208                 ltds->ltd_tgts_bitmap = new_bitmap;
 209                 CFS_FREE_BITMAP(old_bitmap);
 210         }
 211
 212         if (cfs_bitmap_check(ltds->ltd_tgts_bitmap, index)) {
 213                 CERROR("%s: the device %s (%u) is registered already\n",
 214                        lfsck_lfsck2name(lfsck),
 215                        ltd->ltd_tgt->dd_lu_dev.ld_obd->obd_name, index);
 216                 GOTO(unlock, rc = -EEXIST);
 217         }
 218
 219         if (ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
 220                 OBD_ALLOC_PTR(ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK]);
 221                 if (ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK] == NULL)
 222                         GOTO(unlock, rc = -ENOMEM);
 223         }
 224
 225         lfsck_assign_tgt(ltds, ltd, index);
 226         cfs_bitmap_set(ltds->ltd_tgts_bitmap, index);
 227         ltds->ltd_tgtnr++;
 228
 229         GOTO(unlock, rc = 0);
 230
 231 unlock:
 232         if (!locked)
 233                 up_write(&ltds->ltd_rw_sem);
 234
 235         return rc;
 236 }
 237
 238 static int lfsck_add_target_from_orphan(const struct lu_env *env,
 239                                         struct lfsck_instance *lfsck)
 240 {
 241         struct lfsck_tgt_descs  *ltds    = &lfsck->li_ost_descs;
 242         struct lfsck_tgt_desc   *ltd;
 243         struct lfsck_tgt_desc   *next;
 244         struct list_head        *head    = &lfsck_ost_orphan_list;
 245         int                      rc;
 246         bool                     for_ost = true;
 247
 248 again:
 249         spin_lock(&lfsck_instance_lock);
 250         list_for_each_entry_safe(ltd, next, head, ltd_orphan_list) {
 251                 if (ltd->ltd_key == lfsck->li_bottom)
 252                         list_move_tail(&ltd->ltd_orphan_list,
 253                                        &ltds->ltd_orphan);
 254         }
 255         spin_unlock(&lfsck_instance_lock);
 256
 257         down_write(&ltds->ltd_rw_sem);
 258         while (!list_empty(&ltds->ltd_orphan)) {
 259                 ltd = list_entry(ltds->ltd_orphan.next,
 260                                  struct lfsck_tgt_desc,
 261                                  ltd_orphan_list);
 262                 list_del_init(&ltd->ltd_orphan_list);
 263                 rc = __lfsck_add_target(env, lfsck, ltd, for_ost, true);
 264                 /* Do not hold the semaphore for too long time. */
 265                 up_write(&ltds->ltd_rw_sem);
 266                 if (rc != 0)
 267                         return rc;
 268
 269                 down_write(&ltds->ltd_rw_sem);
 270         }
 271         up_write(&ltds->ltd_rw_sem);
 272
 273         if (for_ost) {
 274                 ltds = &lfsck->li_mdt_descs;
 275                 head = &lfsck_mdt_orphan_list;
 276                 for_ost = false;
 277                 goto again;
 278         }
 279
 280         return 0;
 281 }
 282
 283 static inline struct lfsck_component *
 284 __lfsck_component_find(struct lfsck_instance *lfsck, __u16 type,
 285                        struct list_head *list)
 286 {
 287         struct lfsck_component *com;
 288
 289         list_for_each_entry(com, list, lc_link) {
 290                 if (com->lc_type == type)
 291                         return com;
 292         }
 293         return NULL;
 294 }
 295
 296 struct lfsck_component *
 297 lfsck_component_find(struct lfsck_instance *lfsck, __u16 type)
 298 {
 299         struct lfsck_component *com;
 300
 301         spin_lock(&lfsck->li_lock);
 302         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_scan);
 303         if (com != NULL)
 304                 goto unlock;
 305
 306         com = __lfsck_component_find(lfsck, type,
 307                                      &lfsck->li_list_double_scan);
 308         if (com != NULL)
 309                 goto unlock;
 310
 311         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_idle);
 312
 313 unlock:
 314         if (com != NULL)
 315                 lfsck_component_get(com);
 316         spin_unlock(&lfsck->li_lock);
 317         return com;
 318 }
 319
 320 void lfsck_component_cleanup(const struct lu_env *env,
 321                              struct lfsck_component *com)
 322 {
 323         if (!list_empty(&com->lc_link))
 324                 list_del_init(&com->lc_link);
 325         if (!list_empty(&com->lc_link_dir))
 326                 list_del_init(&com->lc_link_dir);
 327
 328         lfsck_component_put(env, com);
 329 }
 330
 331 int lfsck_fid_alloc(const struct lu_env *env, struct lfsck_instance *lfsck,
 332                     struct lu_fid *fid, bool locked)
 333 {
 334         struct lfsck_bookmark   *bk = &lfsck->li_bookmark_ram;
 335         int                      rc = 0;
 336         ENTRY;
 337
 338         if (!locked)
 339                 mutex_lock(&lfsck->li_mutex);
 340
 341         rc = seq_client_alloc_fid(env, lfsck->li_seq, fid);
 342         if (rc >= 0) {
 343                 bk->lb_last_fid = *fid;
 344                 /* We do not care about whether the subsequent sub-operations
 345                  * failed or not. The worst case is that one FID is lost that
 346                  * is not a big issue for the LFSCK since it is relative rare
 347                  * for LFSCK create. */
 348                 rc = lfsck_bookmark_store(env, lfsck);
 349         }
 350
 351         if (!locked)
 352                 mutex_unlock(&lfsck->li_mutex);
 353
 354         RETURN(rc);
 355 }
 356
 357 static int __lfsck_ibits_lock(const struct lu_env *env,
 358                               struct lfsck_instance *lfsck,
 359                               struct dt_object *obj, struct ldlm_res_id *resid,
 360                               struct lustre_handle *lh, __u64 bits,
 361                               enum ldlm_mode mode)
 362 {
 363         struct lfsck_thread_info        *info   = lfsck_env_info(env);
 364         union ldlm_policy_data          *policy = &info->lti_policy;
 365         __u64                            flags  = LDLM_FL_ATOMIC_CB;
 366         int                              rc;
 367
 368         LASSERT(lfsck->li_namespace != NULL);
 369
 370         memset(policy, 0, sizeof(*policy));
 371         policy->l_inodebits.bits = bits;
 372         if (dt_object_remote(obj)) {
 373                 struct ldlm_enqueue_info *einfo = &info->lti_einfo;
 374
 375                 memset(einfo, 0, sizeof(*einfo));
 376                 einfo->ei_type = LDLM_IBITS;
 377                 einfo->ei_mode = mode;
 378                 einfo->ei_cb_bl = ldlm_blocking_ast;
 379                 einfo->ei_cb_cp = ldlm_completion_ast;
 380                 einfo->ei_res_id = resid;
 381
 382                 rc = dt_object_lock(env, obj, lh, einfo, policy);
 383                 /* for regular checks LFSCK doesn't use LDLM locking,
 384                  * so the state isn't coherent. here we just took LDLM
 385                  * lock for coherency and it's time to invalidate
 386                  * previous state */
 387                 if (rc == ELDLM_OK)
 388                         dt_invalidate(env, obj);
 389         } else {
 390                 rc = ldlm_cli_enqueue_local(env, lfsck->li_namespace, resid,
 391                                             LDLM_IBITS, policy, mode,
 392                                             &flags, ldlm_blocking_ast,
 393                                             ldlm_completion_ast, NULL, NULL,
 394                                             0, LVB_T_NONE, NULL, lh);
 395         }
 396
 397         if (rc == ELDLM_OK) {
 398                 rc = 0;
 399         } else {
 400                 memset(lh, 0, sizeof(*lh));
 401                 rc = -EIO;
 402         }
 403
 404         return rc;
 405 }
 406
 407 /**
 408  * Request the specified ibits lock for the given object.
 409  *
 410  * Before the LFSCK modifying on the namespace visible object,
 411  * it needs to acquire related ibits ldlm lock.
 412  *
 413  * \param[in] env       pointer to the thread context
 414  * \param[in] lfsck     pointer to the lfsck instance
 415  * \param[in] obj       pointer to the dt_object to be locked
 416  * \param[out] lh       pointer to the lock handle
 417  * \param[in] bits      the bits for the ldlm lock to be acquired
 418  * \param[in] mode      the mode for the ldlm lock to be acquired
 419  *
 420  * \retval              0 for success
 421  * \retval              negative error number on failure
 422  */
 423 int lfsck_ibits_lock(const struct lu_env *env, struct lfsck_instance *lfsck,
 424                      struct dt_object *obj, struct lustre_handle *lh,
 425                      __u64 bits, enum ldlm_mode mode)
 426 {
 427         struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid;
 428
 429         LASSERT(!lustre_handle_is_used(lh));
 430
 431         fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
 432         return __lfsck_ibits_lock(env, lfsck, obj, resid, lh, bits, mode);
 433 }
 434
 435 /**
 436  * Request the remote LOOKUP lock for the given object.
 437  *
 438  * If \a pobj is remote, the LOOKUP lock of \a obj is on the MDT where
 439  * \a pobj is, acquire LOOKUP lock there.
 440  *
 441  * \param[in] env       pointer to the thread context
 442  * \param[in] lfsck     pointer to the lfsck instance
 443  * \param[in] pobj      pointer to parent dt_object
 444  * \param[in] obj       pointer to the dt_object to be locked
 445  * \param[out] lh       pointer to the lock handle
 446  * \param[in] mode      the mode for the ldlm lock to be acquired
 447  *
 448  * \retval              0 for success
 449  * \retval              negative error number on failure
 450  */
 451 int lfsck_remote_lookup_lock(const struct lu_env *env,
 452                              struct lfsck_instance *lfsck,
 453                              struct dt_object *pobj, struct dt_object *obj,
 454                              struct lustre_handle *lh, enum ldlm_mode mode)
 455 {
 456         struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid;
 457
 458         LASSERT(!lustre_handle_is_used(lh));
 459
 460         fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
 461         return __lfsck_ibits_lock(env, lfsck, pobj, resid, lh,
 462                                   MDS_INODELOCK_LOOKUP, mode);
 463 }
 464
 465 /**
 466  * Release the the specified ibits lock.
 467  *
 468  * If the lock has been acquired before, release it
 469  * and cleanup the handle. Otherwise, do nothing.
 470  *
 471  * \param[in] lh        pointer to the lock handle
 472  * \param[in] mode      the mode for the ldlm lock to be released
 473  */
 474 void lfsck_ibits_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
 475 {
 476         if (lustre_handle_is_used(lh)) {
 477                 ldlm_lock_decref(lh, mode);
 478                 memset(lh, 0, sizeof(*lh));
 479         }
 480 }
 481
 482 /**
 483  * Request compound ibits locks for the given <obj, name> pairs.
 484  *
 485  * Before the LFSCK modifying on the namespace visible object, it needs to
 486  * acquire related ibits ldlm lock. Usually, we can use lfsck_ibits_lock for
 487  * the lock purpose. But the simple lfsck_ibits_lock for directory-based
 488  * modificationis (such as insert name entry to the directory) may be too
 489  * coarse-grained and not efficient.
 490  *
 491  * The lfsck_lock() will request compound ibits locks on the specified
 492  * <obj, name> pairs: the PDO (Parallel Directory Operations) ibits (UPDATE)
 493  * lock on the directory object, and the regular ibits lock on the name hash.
 494  *
 495  * \param[in] env       pointer to the thread context
 496  * \param[in] lfsck     pointer to the lfsck instance
 497  * \param[in] obj       pointer to the dt_object to be locked
 498  * \param[in] name      used for building the PDO lock resource
 499  * \param[out] llh      pointer to the lfsck_lock_handle
 500  * \param[in] bits      the bits for the ldlm lock to be acquired
 501  * \param[in] mode      the mode for the ldlm lock to be acquired
 502  *
 503  * \retval              0 for success
 504  * \retval              negative error number on failure
 505  */
 506 int lfsck_lock(const struct lu_env *env, struct lfsck_instance *lfsck,
 507                struct dt_object *obj, const char *name,
 508                struct lfsck_lock_handle *llh, __u64 bits, enum ldlm_mode mode)
 509 {
 510         struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid;
 511         int                 rc;
 512
 513         LASSERT(S_ISDIR(lfsck_object_type(obj)));
 514         LASSERT(name != NULL);
 515         LASSERT(name[0] != 0);
 516         LASSERT(!lustre_handle_is_used(&llh->llh_pdo_lh));
 517         LASSERT(!lustre_handle_is_used(&llh->llh_reg_lh));
 518
 519         switch (mode) {
 520         case LCK_EX:
 521                 llh->llh_pdo_mode = LCK_EX;
 522                 break;
 523         case LCK_PW:
 524                 llh->llh_pdo_mode = LCK_CW;
 525                 break;
 526         case LCK_PR:
 527                 llh->llh_pdo_mode = LCK_CR;
 528                 break;
 529         default:
 530                 CDEBUG(D_LFSCK, "%s: unexpected PDO lock mode %u on the obj "
 531                        DFID"\n", lfsck_lfsck2name(lfsck), mode,
 532                        PFID(lfsck_dto2fid(obj)));
 533                 LBUG();
 534         }
 535
 536         fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
 537         rc = __lfsck_ibits_lock(env, lfsck, obj, resid, &llh->llh_pdo_lh,
 538                                 MDS_INODELOCK_UPDATE, llh->llh_pdo_mode);
 539         if (rc != 0)
 540                 return rc;
 541
 542         llh->llh_reg_mode = mode;
 543         resid->name[LUSTRE_RES_ID_HSH_OFF] = ll_full_name_hash(NULL, name,
 544                                                                strlen(name));
 545         LASSERT(resid->name[LUSTRE_RES_ID_HSH_OFF] != 0);
 546         rc = __lfsck_ibits_lock(env, lfsck, obj, resid, &llh->llh_reg_lh,
 547                                 bits, llh->llh_reg_mode);
 548         if (rc != 0)
 549                 lfsck_ibits_unlock(&llh->llh_pdo_lh, llh->llh_pdo_mode);
 550
 551         return rc;
 552 }
 553
 554 /**
 555  * Release the the compound ibits locks.
 556  *
 557  * \param[in] llh       pointer to the lfsck_lock_handle to be released
 558  */
 559 void lfsck_unlock(struct lfsck_lock_handle *llh)
 560 {
 561         lfsck_ibits_unlock(&llh->llh_reg_lh, llh->llh_reg_mode);
 562         lfsck_ibits_unlock(&llh->llh_pdo_lh, llh->llh_pdo_mode);
 563 }
 564
 565 int lfsck_find_mdt_idx_by_fid(const struct lu_env *env,
 566                               struct lfsck_instance *lfsck,
 567                               const struct lu_fid *fid)
 568 {
 569         struct seq_server_site  *ss     = lfsck_dev_site(lfsck);
 570         struct lu_seq_range     *range  = &lfsck_env_info(env)->lti_range;
 571         int                      rc;
 572
 573         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
 574                 /* "ROOT" is always on the MDT0. */
 575                 if (lu_fid_eq(fid, &lfsck->li_global_root_fid))
 576                         return 0;
 577
 578                 return lfsck_dev_idx(lfsck);
 579         }
 580
 581         fld_range_set_mdt(range);
 582         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
 583         if (rc == 0)
 584                 rc = range->lsr_index;
 585
 586         return rc;
 587 }
 588
 589 const char dot[] = ".";
 590 const char dotdot[] = "..";
 591 static const char dotlustre[] = ".lustre";
 592 static const char lostfound[] = "lost+found";
 593
 594 /**
 595  * Remove the name entry from the .lustre/lost+found directory.
 596  *
 597  * No need to care about the object referenced by the name entry,
 598  * either the name entry is invalid or redundant, or the referenced
 599  * object has been processed or will be handled by others.
 600  *
 601  * \param[in] env       pointer to the thread context
 602  * \param[in] lfsck     pointer to the lfsck instance
 603  * \param[in] name      the name for the name entry to be removed
 604  *
 605  * \retval              0 for success
 606  * \retval              negative error number on failure
 607  */
 608 static int lfsck_lpf_remove_name_entry(const struct lu_env *env,
 609                                        struct lfsck_instance *lfsck,
 610                                        const char *name)
 611 {
 612         struct dt_object        *parent = lfsck->li_lpf_root_obj;
 613         struct dt_device        *dev    = lfsck_obj2dev(parent);
 614         struct thandle          *th;
 615         struct lfsck_lock_handle *llh   = &lfsck_env_info(env)->lti_llh;
 616         int                      rc;
 617         ENTRY;
 618
 619         rc = lfsck_lock(env, lfsck, parent, name, llh,
 620                         MDS_INODELOCK_UPDATE, LCK_PW);
 621         if (rc != 0)
 622                 RETURN(rc);
 623
 624         th = dt_trans_create(env, dev);
 625         if (IS_ERR(th))
 626                 GOTO(unlock, rc = PTR_ERR(th));
 627
 628         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
 629         if (rc != 0)
 630                 GOTO(stop, rc);
 631
 632         rc = dt_declare_ref_del(env, parent, th);
 633         if (rc != 0)
 634                 GOTO(stop, rc);
 635
 636         rc = dt_trans_start_local(env, dev, th);
 637         if (rc != 0)
 638                 GOTO(stop, rc);
 639
 640         rc = dt_delete(env, parent, (const struct dt_key *)name, th);
 641         if (rc != 0)
 642                 GOTO(stop, rc);
 643
 644         dt_write_lock(env, parent, 0);
 645         rc = dt_ref_del(env, parent, th);
 646         dt_write_unlock(env, parent);
 647
 648         GOTO(stop, rc);
 649
 650 stop:
 651         dt_trans_stop(env, dev, th);
 652
 653 unlock:
 654         lfsck_unlock(llh);
 655
 656         CDEBUG(D_LFSCK, "%s: remove name entry "DFID"/%s: rc = %d\n",
 657                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(parent)), name, rc);
 658
 659         return rc;
 660 }
 661
 662 static int lfsck_create_lpf_local(const struct lu_env *env,
 663                                   struct lfsck_instance *lfsck,
 664                                   struct dt_object *child,
 665                                   struct lu_attr *la,
 666                                   struct dt_object_format *dof,
 667                                   const char *name)
 668 {
 669         struct dt_insert_rec    *rec    = &lfsck_env_info(env)->lti_dt_rec;
 670         struct dt_object        *parent = lfsck->li_lpf_root_obj;
 671         struct dt_device        *dev    = lfsck_obj2dev(child);
 672         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
 673         struct dt_object        *bk_obj = lfsck->li_bookmark_obj;
 674         const struct lu_fid     *cfid   = lfsck_dto2fid(child);
 675         struct thandle          *th     = NULL;
 676         struct linkea_data       ldata  = { NULL };
 677         struct lu_buf            linkea_buf;
 678         const struct lu_name    *cname;
 679         loff_t                   pos    = 0;
 680         int                      len    = sizeof(struct lfsck_bookmark);
 681         int                      rc;
 682         ENTRY;
 683
 684         cname = lfsck_name_get_const(env, name, strlen(name));
 685         rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf2,
 686                               cname, lfsck_dto2fid(parent));
 687         if (rc != 0)
 688                 RETURN(rc);
 689
 690         th = dt_trans_create(env, dev);
 691         if (IS_ERR(th))
 692                 RETURN(PTR_ERR(th));
 693
 694         /* 1a. create child */
 695         rc = dt_declare_create(env, child, la, NULL, dof, th);
 696         if (rc != 0)
 697                 GOTO(stop, rc);
 698
 699         if (!dt_try_as_dir(env, child))
 700                 GOTO(stop, rc = -ENOTDIR);
 701
 702         /* 2a. increase child nlink */
 703         rc = dt_declare_ref_add(env, child, th);
 704         if (rc != 0)
 705                 GOTO(stop, rc);
 706
 707         /* 3a. insert dot into child dir */
 708         rec->rec_type = S_IFDIR;
 709         rec->rec_fid = cfid;
 710         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
 711                                (const struct dt_key *)dot, th);
 712         if (rc != 0)
 713                 GOTO(stop, rc);
 714
 715         /* 4a. insert dotdot into child dir */
 716         rec->rec_fid = &LU_LPF_FID;
 717         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
 718                                (const struct dt_key *)dotdot, th);
 719         if (rc != 0)
 720                 GOTO(stop, rc);
 721
 722         /* 5a. insert linkEA for child */
 723         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
 724                        ldata.ld_leh->leh_len);
 725         rc = dt_declare_xattr_set(env, child, &linkea_buf,
 726                                   XATTR_NAME_LINK, 0, th);
 727         if (rc != 0)
 728                 GOTO(stop, rc);
 729
 730         /* 6a. insert name into parent dir */
 731         rec->rec_type = S_IFDIR;
 732         rec->rec_fid = cfid;
 733         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
 734                                (const struct dt_key *)name, th);
 735         if (rc != 0)
 736                 GOTO(stop, rc);
 737
 738         /* 7a. increase parent nlink */
 739         rc = dt_declare_ref_add(env, parent, th);
 740         if (rc != 0)
 741                 GOTO(stop, rc);
 742
 743         /* 8a. update bookmark */
 744         rc = dt_declare_record_write(env, bk_obj,
 745                                      lfsck_buf_get(env, bk, len), 0, th);
 746         if (rc != 0)
 747                 GOTO(stop, rc);
 748
 749         rc = dt_trans_start_local(env, dev, th);
 750         if (rc != 0)
 751                 GOTO(stop, rc);
 752
 753         dt_write_lock(env, child, 0);
 754         /* 1b. create child */
 755         rc = dt_create(env, child, la, NULL, dof, th);
 756         if (rc != 0)
 757                 GOTO(unlock, rc);
 758
 759         /* 2b. increase child nlink */
 760         rc = dt_ref_add(env, child, th);
 761         if (rc != 0)
 762                 GOTO(unlock, rc);
 763
 764         /* 3b. insert dot into child dir */
 765         rec->rec_fid = cfid;
 766         rc = dt_insert(env, child, (const struct dt_rec *)rec,
 767                        (const struct dt_key *)dot, th);
 768         if (rc != 0)
 769                 GOTO(unlock, rc);
 770
 771         /* 4b. insert dotdot into child dir */
 772         rec->rec_fid = &LU_LPF_FID;
 773         rc = dt_insert(env, child, (const struct dt_rec *)rec,
 774                        (const struct dt_key *)dotdot, th);
 775         if (rc != 0)
 776                 GOTO(unlock, rc);
 777
 778         /* 5b. insert linkEA for child. */
 779         rc = dt_xattr_set(env, child, &linkea_buf,
 780                           XATTR_NAME_LINK, 0, th);
 781         dt_write_unlock(env, child);
 782         if (rc != 0)
 783                 GOTO(stop, rc);
 784
 785         /* 6b. insert name into parent dir */
 786         rec->rec_fid = cfid;
 787         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
 788                        (const struct dt_key *)name, th);
 789         if (rc != 0)
 790                 GOTO(stop, rc);
 791
 792         dt_write_lock(env, parent, 0);
 793         /* 7b. increase parent nlink */
 794         rc = dt_ref_add(env, parent, th);
 795         dt_write_unlock(env, parent);
 796         if (rc != 0)
 797                 GOTO(stop, rc);
 798
 799         bk->lb_lpf_fid = *cfid;
 800         lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk);
 801
 802         /* 8b. update bookmark */
 803         rc = dt_record_write(env, bk_obj,
 804                              lfsck_buf_get(env, bk, len), &pos, th);
 805
 806         GOTO(stop, rc);
 807
 808 unlock:
 809         dt_write_unlock(env, child);
 810
 811 stop:
 812         dt_trans_stop(env, dev, th);
 813
 814         return rc;
 815 }
 816
 817 static int lfsck_create_lpf_remote(const struct lu_env *env,
 818                                    struct lfsck_instance *lfsck,
 819                                    struct dt_object *child,
 820                                    struct lu_attr *la,
 821                                    struct dt_object_format *dof,
 822                                    const char *name)
 823 {
 824         struct dt_insert_rec    *rec    = &lfsck_env_info(env)->lti_dt_rec;
 825         struct dt_object        *parent = lfsck->li_lpf_root_obj;
 826         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
 827         struct dt_object        *bk_obj = lfsck->li_bookmark_obj;
 828         const struct lu_fid     *cfid   = lfsck_dto2fid(child);
 829         struct thandle          *th     = NULL;
 830         struct linkea_data       ldata  = { NULL };
 831         struct lu_buf            linkea_buf;
 832         const struct lu_name    *cname;
 833         struct dt_device        *dev;
 834         loff_t                   pos    = 0;
 835         int                      len    = sizeof(struct lfsck_bookmark);
 836         int                      rc;
 837         ENTRY;
 838
 839         cname = lfsck_name_get_const(env, name, strlen(name));
 840         rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf2,
 841                               cname, lfsck_dto2fid(parent));
 842         if (rc != 0)
 843                 RETURN(rc);
 844
 845         /* Create .lustre/lost+found/MDTxxxx. */
 846
 847         /* XXX: Currently, cross-MDT create operation needs to create the child
 848          *      object firstly, then insert name into the parent directory. For
 849          *      this case, the child object resides on current MDT (local), but
 850          *      the parent ".lustre/lost+found" may be on remote MDT. It is not
 851          *      easy to contain all the sub-modifications orderly within single
 852          *      transaction.
 853          *
 854          *      To avoid more inconsistency, we split the create operation into
 855          *      two transactions:
 856          *
 857          *      1) create the child and update the lfsck_bookmark::lb_lpf_fid
 858          *         locally.
 859          *      2) insert the name "MDTXXXX" in the parent ".lustre/lost+found"
 860          *         remotely.
 861          *
 862          *      If 1) done, but 2) failed, then go ahead, the LFSCK will try to
 863          *      repair such inconsistency when LFSCK run next time. */
 864
 865         /* Transaction I: locally */
 866
 867         dev = lfsck_obj2dev(child);
 868         th = dt_trans_create(env, dev);
 869         if (IS_ERR(th))
 870                 RETURN(PTR_ERR(th));
 871
 872         /* 1a. create child */
 873         rc = dt_declare_create(env, child, la, NULL, dof, th);
 874         if (rc != 0)
 875                 GOTO(stop, rc);
 876
 877         if (!dt_try_as_dir(env, child))
 878                 GOTO(stop, rc = -ENOTDIR);
 879
 880         /* 2a. increase child nlink */
 881         rc = dt_declare_ref_add(env, child, th);
 882         if (rc != 0)
 883                 GOTO(stop, rc);
 884
 885         /* 3a. insert dot into child dir */
 886         rec->rec_type = S_IFDIR;
 887         rec->rec_fid = cfid;
 888         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
 889                                (const struct dt_key *)dot, th);
 890         if (rc != 0)
 891                 GOTO(stop, rc);
 892
 893         /* 4a. insert dotdot into child dir */
 894         rec->rec_fid = &LU_LPF_FID;
 895         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
 896                                (const struct dt_key *)dotdot, th);
 897         if (rc != 0)
 898                 GOTO(stop, rc);
 899
 900         /* 5a. insert linkEA for child */
 901         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
 902                        ldata.ld_leh->leh_len);
 903         rc = dt_declare_xattr_set(env, child, &linkea_buf,
 904                                   XATTR_NAME_LINK, 0, th);
 905         if (rc != 0)
 906                 GOTO(stop, rc);
 907
 908         /* 6a. update bookmark */
 909         rc = dt_declare_record_write(env, bk_obj,
 910                                      lfsck_buf_get(env, bk, len), 0, th);
 911         if (rc != 0)
 912                 GOTO(stop, rc);
 913
 914         rc = dt_trans_start_local(env, dev, th);
 915         if (rc != 0)
 916                 GOTO(stop, rc);
 917
 918         dt_write_lock(env, child, 0);
 919         /* 1b. create child */
 920         rc = dt_create(env, child, la, NULL, dof, th);
 921         if (rc != 0)
 922                 GOTO(unlock, rc);
 923
 924         /* 2b. increase child nlink */
 925         rc = dt_ref_add(env, child, th);
 926         if (rc != 0)
 927                 GOTO(unlock, rc);
 928
 929         /* 3b. insert dot into child dir */
 930         rec->rec_type = S_IFDIR;
 931         rec->rec_fid = cfid;
 932         rc = dt_insert(env, child, (const struct dt_rec *)rec,
 933                        (const struct dt_key *)dot, th);
 934         if (rc != 0)
 935                 GOTO(unlock, rc);
 936
 937         /* 4b. insert dotdot into child dir */
 938         rec->rec_fid = &LU_LPF_FID;
 939         rc = dt_insert(env, child, (const struct dt_rec *)rec,
 940                        (const struct dt_key *)dotdot, th);
 941         if (rc != 0)
 942                 GOTO(unlock, rc);
 943
 944         /* 5b. insert linkEA for child */
 945         rc = dt_xattr_set(env, child, &linkea_buf,
 946                           XATTR_NAME_LINK, 0, th);
 947         if (rc != 0)
 948                 GOTO(unlock, rc);
 949
 950         bk->lb_lpf_fid = *cfid;
 951         lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk);
 952
 953         /* 6b. update bookmark */
 954         rc = dt_record_write(env, bk_obj,
 955                              lfsck_buf_get(env, bk, len), &pos, th);
 956
 957         dt_write_unlock(env, child);
 958         dt_trans_stop(env, dev, th);
 959         if (rc != 0)
 960                 RETURN(rc);
 961
 962         /* Transaction II: remotely */
 963
 964         dev = lfsck_obj2dev(parent);
 965         th = dt_trans_create(env, dev);
 966         if (IS_ERR(th))
 967                 RETURN(PTR_ERR(th));
 968
 969         th->th_sync = 1;
 970         /* 5a. insert name into parent dir */
 971         rec->rec_fid = cfid;
 972         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
 973                                (const struct dt_key *)name, th);
 974         if (rc != 0)
 975                 GOTO(stop, rc);
 976
 977         /* 6a. increase parent nlink */
 978         rc = dt_declare_ref_add(env, parent, th);
 979         if (rc != 0)
 980                 GOTO(stop, rc);
 981
 982         rc = dt_trans_start_local(env, dev, th);
 983         if (rc != 0)
 984                 GOTO(stop, rc);
 985
 986         /* 5b. insert name into parent dir */
 987         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
 988                        (const struct dt_key *)name, th);
 989         if (rc != 0)
 990                 GOTO(stop, rc);
 991
 992         dt_write_lock(env, parent, 0);
 993         /* 6b. increase parent nlink */
 994         rc = dt_ref_add(env, parent, th);
 995         dt_write_unlock(env, parent);
 996
 997         GOTO(stop, rc);
 998
 999 unlock:
1000         dt_write_unlock(env, child);
1001 stop:
1002         dt_trans_stop(env, dev, th);
1003
1004         if (rc != 0 && dev == lfsck_obj2dev(parent))
1005                 CDEBUG(D_LFSCK, "%s: partially created the object "DFID
1006                        "for orphans, but failed to insert the name %s "
1007                        "to the .lustre/lost+found/. Such inconsistency "
1008                        "will be repaired when LFSCK run next time: rc = %d\n",
1009                        lfsck_lfsck2name(lfsck), PFID(cfid), name, rc);
1010
1011         return rc;
1012 }
1013
1014 /**
1015  * Create the MDTxxxx directory under /ROOT/.lustre/lost+found/
1016  *
1017  * The /ROOT/.lustre/lost+found/MDTxxxx/ directory is used for holding
1018  * orphans and other uncertain inconsistent objects found during the
1019  * LFSCK. Such directory will be created by the LFSCK engine on the
1020  * local MDT before the LFSCK scanning.
1021  *
1022  * \param[in] env       pointer to the thread context
1023  * \param[in] lfsck     pointer to the lfsck instance
1024  *
1025  * \retval              0 for success
1026  * \retval              negative error number on failure
1027  */
1028 static int lfsck_create_lpf(const struct lu_env *env,
1029                             struct lfsck_instance *lfsck)
1030 {
1031         struct lfsck_bookmark    *bk    = &lfsck->li_bookmark_ram;
1032         struct lfsck_thread_info *info  = lfsck_env_info(env);
1033         struct lu_fid            *cfid  = &info->lti_fid2;
1034         struct lu_attr           *la    = &info->lti_la;
1035         struct dt_object_format  *dof   = &info->lti_dof;
1036         struct dt_object         *parent = lfsck->li_lpf_root_obj;
1037         struct dt_object         *child = NULL;
1038         struct lfsck_lock_handle *llh   = &info->lti_llh;
1039         char                      name[8];
1040         int                       node  = lfsck_dev_idx(lfsck);
1041         int                       rc    = 0;
1042         ENTRY;
1043
1044         LASSERT(lfsck->li_master);
1045         LASSERT(parent != NULL);
1046         LASSERT(lfsck->li_lpf_obj == NULL);
1047
1048         snprintf(name, 8, "MDT%04x", node);
1049         rc = lfsck_lock(env, lfsck, parent, name, llh,
1050                         MDS_INODELOCK_UPDATE, LCK_PW);
1051         if (rc != 0)
1052                 RETURN(rc);
1053
1054         if (fid_is_zero(&bk->lb_lpf_fid)) {
1055                 /* There is corner case that: in former LFSCK scanning we have
1056                  * created the .lustre/lost+found/MDTxxxx but failed to update
1057                  * the lfsck_bookmark::lb_lpf_fid successfully. So need lookup
1058                  * it from MDT0 firstly. */
1059                 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1060                                (const struct dt_key *)name);
1061                 if (rc != 0 && rc != -ENOENT)
1062                         GOTO(unlock, rc);
1063
1064                 if (rc == 0) {
1065                         bk->lb_lpf_fid = *cfid;
1066                         rc = lfsck_bookmark_store(env, lfsck);
1067                 } else {
1068                         rc = lfsck_fid_alloc(env, lfsck, cfid, true);
1069                 }
1070                 if (rc != 0)
1071                         GOTO(unlock, rc);
1072         } else {
1073                 *cfid = bk->lb_lpf_fid;
1074         }
1075
1076         child = lfsck_object_find_bottom_new(env, lfsck, cfid);
1077         if (IS_ERR(child))
1078                 GOTO(unlock, rc = PTR_ERR(child));
1079
1080         if (dt_object_exists(child) != 0) {
1081                 if (unlikely(!dt_try_as_dir(env, child)))
1082                         rc = -ENOTDIR;
1083                 else
1084                         lfsck->li_lpf_obj = child;
1085
1086                 GOTO(unlock, rc);
1087         }
1088
1089         memset(la, 0, sizeof(*la));
1090         la->la_atime = la->la_mtime = la->la_ctime = ktime_get_real_seconds();
1091         la->la_mode = S_IFDIR | S_IRWXU;
1092         la->la_valid = LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1093                        LA_UID | LA_GID;
1094         memset(dof, 0, sizeof(*dof));
1095         dof->dof_type = dt_mode_to_dft(S_IFDIR);
1096
1097         if (node == 0)
1098                 rc = lfsck_create_lpf_local(env, lfsck, child, la, dof, name);
1099         else
1100                 rc = lfsck_create_lpf_remote(env, lfsck, child, la, dof, name);
1101         if (rc == 0)
1102                 lfsck->li_lpf_obj = child;
1103
1104         GOTO(unlock, rc);
1105
1106 unlock:
1107         lfsck_unlock(llh);
1108         if (rc != 0 && child != NULL && !IS_ERR(child))
1109                 lfsck_object_put(env, child);
1110
1111         return rc;
1112 }
1113
1114 /**
1115  * Scan .lustre/lost+found for bad name entries and remove them.
1116  *
1117  * The valid name entry should be "MDTxxxx", the "xxxx" is the MDT device
1118  * index in the system. Any other formatted name is invalid and should be
1119  * removed.
1120  *
1121  * \param[in] env       pointer to the thread context
1122  * \param[in] lfsck     pointer to the lfsck instance
1123  *
1124  * \retval              0 for success
1125  * \retval              negative error number on failure
1126  */
1127 static int lfsck_scan_lpf_bad_entries(const struct lu_env *env,
1128                                       struct lfsck_instance *lfsck)
1129 {
1130         struct dt_object        *parent = lfsck->li_lpf_root_obj;
1131         struct lu_dirent        *ent    =
1132                         (struct lu_dirent *)lfsck_env_info(env)->lti_key;
1133         const struct dt_it_ops  *iops   = &parent->do_index_ops->dio_it;
1134         struct dt_it            *it;
1135         int                      rc;
1136         ENTRY;
1137
1138         it = iops->init(env, parent, LUDA_64BITHASH);
1139         if (IS_ERR(it))
1140                 RETURN(PTR_ERR(it));
1141
1142         rc = iops->load(env, it, 0);
1143         if (rc == 0)
1144                 rc = iops->next(env, it);
1145         else if (rc > 0)
1146                 rc = 0;
1147
1148         while (rc == 0) {
1149                 int off = 3;
1150
1151                 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
1152                 if (rc != 0)
1153                         break;
1154
1155                 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
1156                 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
1157                         goto next;
1158
1159                 /* name length must be strlen("MDTxxxx") */
1160                 if (ent->lde_namelen != 7)
1161                         goto remove;
1162
1163                 if (memcmp(ent->lde_name, "MDT", off) != 0)
1164                         goto remove;
1165
1166                 while (off < 7 && isxdigit(ent->lde_name[off]))
1167                         off++;
1168
1169                 if (off != 7) {
1170
1171 remove:
1172                         rc = lfsck_lpf_remove_name_entry(env, lfsck,
1173                                                          ent->lde_name);
1174                         if (rc != 0)
1175                                 break;
1176                 }
1177
1178 next:
1179                 rc = iops->next(env, it);
1180         }
1181
1182         iops->put(env, it);
1183         iops->fini(env, it);
1184
1185         RETURN(rc > 0 ? 0 : rc);
1186 }
1187
1188 static int lfsck_update_lpf_entry(const struct lu_env *env,
1189                                   struct lfsck_instance *lfsck,
1190                                   struct dt_object *parent,
1191                                   struct dt_object *child,
1192                                   const char *name,
1193                                   enum lfsck_verify_lpf_types type)
1194 {
1195         int rc;
1196
1197         if (type == LVLT_BY_BOOKMARK) {
1198                 rc = lfsck_update_name_entry(env, lfsck, parent, name,
1199                                              lfsck_dto2fid(child), S_IFDIR);
1200         } else /* if (type == LVLT_BY_NAMEENTRY) */ {
1201                 lfsck->li_bookmark_ram.lb_lpf_fid = *lfsck_dto2fid(child);
1202                 rc = lfsck_bookmark_store(env, lfsck);
1203
1204                 CDEBUG(D_LFSCK, "%s: update LPF fid "DFID
1205                        " in the bookmark file: rc = %d\n",
1206                        lfsck_lfsck2name(lfsck),
1207                        PFID(lfsck_dto2fid(child)), rc);
1208         }
1209
1210         return rc;
1211 }
1212
1213 /**
1214  * Check whether the @child back references the @parent.
1215  *
1216  * Two cases:
1217  * 1) The child's FID is stored in the bookmark file. If the child back
1218  *    references the parent (LU_LPF_FID object) via its ".." entry, then
1219  *    insert the name (MDTxxxx) to the .lustre/lost+found; otherwise, if
1220  *    the child back references another parent2, then:
1221  * 1.1) If the parent2 recognizes the child, then update the bookmark file;
1222  * 1.2) Otherwise, the LFSCK cannot know whether there will be parent3 that
1223  *      references the child. So keep them there. As the LFSCK processing,
1224  *      the parent3 may be found, then when the LFSCK run next time, the
1225  *      inconsistency can be repaired.
1226  *
1227  * 2) The child's FID is stored in the .lustre/lost+found/ sub-directory name
1228  *    entry (MDTxxxx). If the child back references the parent (LU_LPF_FID obj)
1229  *    via its ".." entry, then update the bookmark file, otherwise, if the child
1230  *    back references another parent2, then:
1231  * 2.1) If the parent2 recognizes the child, then remove the sub-directory
1232  *      from .lustre/lost+found/;
1233  * 2.2) Otherwise, if the parent2 does not recognizes the child, trust the
1234  *      sub-directory name entry and update the child;
1235  * 2.3) Otherwise, if we do not know whether the parent2 recognizes the child
1236  *      or not, then keep them there.
1237  *
1238  * \param[in] env       pointer to the thread context
1239  * \param[in] lfsck     pointer to the lfsck instance
1240  * \param[in] child     pointer to the lost+found sub-directory object
1241  * \param[in] name      the name for lost+found sub-directory object
1242  * \param[out] fid      pointer to the buffer to hold the FID of the object
1243  *                      (called it as parent2) that is referenced via the
1244  *                      child's dotdot entry; it also can be the FID that
1245  *                      is referenced by the name entry under the parent2.
1246  * \param[in] type      to indicate where the child's FID is stored in
1247  *
1248  * \retval              positive number for uncertain inconsistency
1249  * \retval              0 for success
1250  * \retval              negative error number on failure
1251  */
1252 static int lfsck_verify_lpf_pairs(const struct lu_env *env,
1253                                   struct lfsck_instance *lfsck,
1254                                   struct dt_object *child, const char *name,
1255                                   struct lu_fid *fid,
1256                                   enum lfsck_verify_lpf_types type)
1257 {
1258         struct dt_object         *parent  = lfsck->li_lpf_root_obj;
1259         struct lfsck_thread_info *info    = lfsck_env_info(env);
1260         char                     *name2   = info->lti_key;
1261         struct lu_fid            *fid2    = &info->lti_fid3;
1262         struct dt_object         *parent2 = NULL;
1263         struct lustre_handle      lh      = { 0 };
1264         int                       rc;
1265         ENTRY;
1266
1267         fid_zero(fid);
1268         rc = dt_lookup(env, child, (struct dt_rec *)fid,
1269                        (const struct dt_key *)dotdot);
1270         if (rc != 0)
1271                 GOTO(linkea, rc);
1272
1273         if (!fid_is_sane(fid))
1274                 GOTO(linkea, rc = -EINVAL);
1275
1276         if (lu_fid_eq(fid, &LU_LPF_FID)) {
1277                 const struct lu_name *cname;
1278
1279                 if (lfsck->li_lpf_obj == NULL) {
1280                         lu_object_get(&child->do_lu);
1281                         lfsck->li_lpf_obj = child;
1282                 }
1283
1284                 cname = lfsck_name_get_const(env, name, strlen(name));
1285                 rc = lfsck_verify_linkea(env, child, cname, &LU_LPF_FID);
1286                 if (rc == 0)
1287                         rc = lfsck_update_lpf_entry(env, lfsck, parent, child,
1288                                                     name, type);
1289
1290                 GOTO(out_done, rc);
1291         }
1292
1293         parent2 = lfsck_object_find_bottom(env, lfsck, fid);
1294         if (IS_ERR(parent2))
1295                 GOTO(linkea, parent2);
1296
1297         if (!dt_object_exists(parent2)) {
1298                 lfsck_object_put(env, parent2);
1299
1300                 GOTO(linkea, parent2 = ERR_PTR(-ENOENT));
1301         }
1302
1303         if (!dt_try_as_dir(env, parent2)) {
1304                 lfsck_object_put(env, parent2);
1305
1306                 GOTO(linkea, parent2 = ERR_PTR(-ENOTDIR));
1307         }
1308
1309 linkea:
1310         /* To prevent rename/unlink race */
1311         rc = lfsck_ibits_lock(env, lfsck, child, &lh,
1312                               MDS_INODELOCK_UPDATE, LCK_PR);
1313         if (rc != 0)
1314                 GOTO(out_put, rc);
1315
1316         dt_read_lock(env, child, 0);
1317         rc = lfsck_links_get_first(env, child, name2, fid2);
1318         if (rc != 0) {
1319                 dt_read_unlock(env, child);
1320                 lfsck_ibits_unlock(&lh, LCK_PR);
1321
1322                 GOTO(out_put, rc = 1);
1323         }
1324
1325         /* It is almost impossible that the bookmark file (or the name entry)
1326          * and the linkEA hit the same data corruption. Trust the linkEA. */
1327         if (lu_fid_eq(fid2, &LU_LPF_FID) && strcmp(name, name2) == 0) {
1328                 dt_read_unlock(env, child);
1329                 lfsck_ibits_unlock(&lh, LCK_PR);
1330
1331                 *fid = *fid2;
1332                 if (lfsck->li_lpf_obj == NULL) {
1333                         lu_object_get(&child->do_lu);
1334                         lfsck->li_lpf_obj = child;
1335                 }
1336
1337                 /* Update the child's dotdot entry */
1338                 rc = lfsck_update_name_entry(env, lfsck, child, dotdot,
1339                                              &LU_LPF_FID, S_IFDIR);
1340                 if (rc == 0)
1341                         rc = lfsck_update_lpf_entry(env, lfsck, parent, child,
1342                                                     name, type);
1343
1344                 GOTO(out_put, rc);
1345         }
1346
1347         if (parent2 == NULL || IS_ERR(parent2)) {
1348                 dt_read_unlock(env, child);
1349                 lfsck_ibits_unlock(&lh, LCK_PR);
1350
1351                 GOTO(out_done, rc = 1);
1352         }
1353
1354         rc = dt_lookup(env, parent2, (struct dt_rec *)fid,
1355                        (const struct dt_key *)name2);
1356         dt_read_unlock(env, child);
1357         lfsck_ibits_unlock(&lh, LCK_PR);
1358         if (rc != 0 && rc != -ENOENT)
1359                 GOTO(out_put, rc);
1360
1361         if (rc == -ENOENT || !lu_fid_eq(fid, lfsck_dto2fid(child))) {
1362                 if (type == LVLT_BY_BOOKMARK)
1363                         GOTO(out_put, rc = 1);
1364
1365                 /* Trust the name entry, update the child's dotdot entry. */
1366                 rc = lfsck_update_name_entry(env, lfsck, child, dotdot,
1367                                              &LU_LPF_FID, S_IFDIR);
1368
1369                 GOTO(out_put, rc);
1370         }
1371
1372         if (type == LVLT_BY_BOOKMARK) {
1373                 /* Invalid FID record in the bookmark file, reset it. */
1374                 fid_zero(&lfsck->li_bookmark_ram.lb_lpf_fid);
1375                 rc = lfsck_bookmark_store(env, lfsck);
1376
1377                 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1378                        " in the bookmark file: rc = %d\n",
1379                        lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(child)), rc);
1380         } else /* if (type == LVLT_BY_NAMEENTRY) */ {
1381                 /* The name entry is wrong, remove it. */
1382                 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1383         }
1384
1385         GOTO(out_put, rc);
1386
1387 out_put:
1388         if (parent2 != NULL && !IS_ERR(parent2))
1389                 lfsck_object_put(env, parent2);
1390
1391 out_done:
1392         return rc;
1393 }
1394
1395 /**
1396  * Verify the /ROOT/.lustre/lost+found/ directory.
1397  *
1398  * /ROOT/.lustre/lost+found/ is a special directory to hold the objects that
1399  * the LFSCK does not exactly know how to handle, such as orphans. So before
1400  * the LFSCK scanning the system, the consistency of such directory needs to
1401  * be verified firstly to allow the users to use it during the LFSCK.
1402  *
1403  * \param[in] env       pointer to the thread context
1404  * \param[in] lfsck     pointer to the lfsck instance
1405  *
1406  * \retval              positive number for uncertain inconsistency
1407  * \retval              0 for success
1408  * \retval              negative error number on failure
1409  */
1410 int lfsck_verify_lpf(const struct lu_env *env, struct lfsck_instance *lfsck)
1411 {
1412         struct lfsck_thread_info *info   = lfsck_env_info(env);
1413         struct lu_fid            *pfid   = &info->lti_fid;
1414         struct lu_fid            *cfid   = &info->lti_fid2;
1415         struct lfsck_bookmark    *bk     = &lfsck->li_bookmark_ram;
1416         struct dt_object         *parent;
1417         /* child1's FID is in the bookmark file. */
1418         struct dt_object         *child1 = NULL;
1419         /* child2's FID is in the name entry MDTxxxx. */
1420         struct dt_object         *child2 = NULL;
1421         const struct lu_name     *cname;
1422         char                      name[8];
1423         int                       node   = lfsck_dev_idx(lfsck);
1424         int                       rc     = 0;
1425         ENTRY;
1426
1427         LASSERT(lfsck->li_master);
1428
1429         if (lfsck->li_lpf_root_obj != NULL)
1430                 RETURN(0);
1431
1432         if (node == 0) {
1433                 parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
1434                                                   &LU_LPF_FID);
1435         } else {
1436                 struct lfsck_tgt_desc *ltd;
1437
1438                 ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, 0);
1439                 if (unlikely(ltd == NULL))
1440                         RETURN(-ENXIO);
1441
1442                 parent = lfsck_object_find_by_dev(env, ltd->ltd_tgt,
1443                                                   &LU_LPF_FID);
1444                 lfsck_tgt_put(ltd);
1445         }
1446
1447         if (IS_ERR(parent))
1448                 RETURN(PTR_ERR(parent));
1449
1450         LASSERT(dt_object_exists(parent));
1451
1452         if (unlikely(!dt_try_as_dir(env, parent))) {
1453                 lfsck_object_put(env, parent);
1454
1455                 GOTO(put, rc = -ENOTDIR);
1456         }
1457
1458         lfsck->li_lpf_root_obj = parent;
1459         if (node == 0) {
1460                 rc = lfsck_scan_lpf_bad_entries(env, lfsck);
1461                 if (rc != 0)
1462                         CDEBUG(D_LFSCK, "%s: scan .lustre/lost+found/ "
1463                                "for bad sub-directories: rc = %d\n",
1464                                lfsck_lfsck2name(lfsck), rc);
1465         }
1466
1467         /* child2 */
1468         snprintf(name, 8, "MDT%04x", node);
1469         rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1470                        (const struct dt_key *)name);
1471         if (rc == -ENOENT) {
1472                 rc = 0;
1473                 goto find_child1;
1474         }
1475
1476         if (rc != 0)
1477                 GOTO(put, rc);
1478
1479         /* Invalid FID in the name entry, remove the name entry. */
1480         if (!fid_is_norm(cfid)) {
1481                 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1482                 if (rc != 0)
1483                         GOTO(put, rc);
1484
1485                 goto find_child1;
1486         }
1487
1488         child2 = lfsck_object_find_bottom(env, lfsck, cfid);
1489         if (IS_ERR(child2))
1490                 GOTO(put, rc = PTR_ERR(child2));
1491
1492         if (unlikely(!dt_object_exists(child2) ||
1493                      dt_object_remote(child2)) ||
1494                      !S_ISDIR(lfsck_object_type(child2))) {
1495                 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1496                 if (rc != 0)
1497                         GOTO(put, rc);
1498
1499                 goto find_child1;
1500         }
1501
1502         if (unlikely(!dt_try_as_dir(env, child2))) {
1503                 lfsck_object_put(env, child2);
1504                 child2 = NULL;
1505                 rc = -ENOTDIR;
1506         }
1507
1508 find_child1:
1509         if (fid_is_zero(&bk->lb_lpf_fid))
1510                 goto check_child2;
1511
1512         if (likely(lu_fid_eq(cfid, &bk->lb_lpf_fid))) {
1513                 if (lfsck->li_lpf_obj == NULL) {
1514                         lu_object_get(&child2->do_lu);
1515                         lfsck->li_lpf_obj = child2;
1516                 }
1517
1518                 cname = lfsck_name_get_const(env, name, strlen(name));
1519                 rc = lfsck_verify_linkea(env, child2, cname, &LU_LPF_FID);
1520
1521                 GOTO(put, rc);
1522         }
1523
1524         if (unlikely(!fid_is_norm(&bk->lb_lpf_fid))) {
1525                 struct lu_fid tfid = bk->lb_lpf_fid;
1526
1527                 /* Invalid FID record in the bookmark file, reset it. */
1528                 fid_zero(&bk->lb_lpf_fid);
1529                 rc = lfsck_bookmark_store(env, lfsck);
1530
1531                 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1532                        " in the bookmark file: rc = %d\n",
1533                        lfsck_lfsck2name(lfsck), PFID(&tfid), rc);
1534
1535                 if (rc != 0)
1536                         GOTO(put, rc);
1537
1538                 goto check_child2;
1539         }
1540
1541         child1 = lfsck_object_find_bottom(env, lfsck, &bk->lb_lpf_fid);
1542         if (IS_ERR(child1)) {
1543                 child1 = NULL;
1544                 goto check_child2;
1545         }
1546
1547         if (unlikely(!dt_object_exists(child1) ||
1548                      dt_object_remote(child1)) ||
1549                      !S_ISDIR(lfsck_object_type(child1))) {
1550                 /* Invalid FID record in the bookmark file, reset it. */
1551                 fid_zero(&bk->lb_lpf_fid);
1552                 rc = lfsck_bookmark_store(env, lfsck);
1553
1554                 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1555                        " in the bookmark file: rc = %d\n",
1556                        lfsck_lfsck2name(lfsck),
1557                        PFID(lfsck_dto2fid(child1)), rc);
1558
1559                 if (rc != 0)
1560                         GOTO(put, rc);
1561
1562                 lfsck_object_put(env, child1);
1563                 child1 = NULL;
1564                 goto check_child2;
1565         }
1566
1567         if (unlikely(!dt_try_as_dir(env, child1))) {
1568                 lfsck_object_put(env, child1);
1569                 child1 = NULL;
1570                 rc = -ENOTDIR;
1571                 goto check_child2;
1572         }
1573
1574         rc = lfsck_verify_lpf_pairs(env, lfsck, child1, name, pfid,
1575                                     LVLT_BY_BOOKMARK);
1576         if (lu_fid_eq(pfid, &LU_LPF_FID))
1577                 GOTO(put, rc);
1578
1579 check_child2:
1580         if (child2 != NULL)
1581                 rc = lfsck_verify_lpf_pairs(env, lfsck, child2, name,
1582                                             pfid, LVLT_BY_NAMEENTRY);
1583
1584         GOTO(put, rc);
1585
1586 put:
1587         if (lfsck->li_lpf_obj != NULL) {
1588                 if (unlikely(!dt_try_as_dir(env, lfsck->li_lpf_obj))) {
1589                         lfsck_object_put(env, lfsck->li_lpf_obj);
1590                         lfsck->li_lpf_obj = NULL;
1591                         rc = -ENOTDIR;
1592                 }
1593         } else if (rc == 0) {
1594                 rc = lfsck_create_lpf(env, lfsck);
1595         }
1596
1597         if (child2 != NULL && !IS_ERR(child2))
1598                 lfsck_object_put(env, child2);
1599         if (child1 != NULL && !IS_ERR(child1))
1600                 lfsck_object_put(env, child1);
1601
1602         return rc;
1603 }
1604
1605 static int lfsck_fid_init(struct lfsck_instance *lfsck)
1606 {
1607         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
1608         struct seq_server_site  *ss     = lfsck_dev_site(lfsck);
1609         char                    *prefix;
1610         int                      rc     = 0;
1611         ENTRY;
1612
1613         if (unlikely(ss == NULL))
1614                 RETURN(-ENXIO);
1615
1616         OBD_ALLOC_PTR(lfsck->li_seq);
1617         if (lfsck->li_seq == NULL)
1618                 RETURN(-ENOMEM);
1619
1620         OBD_ALLOC(prefix, MAX_OBD_NAME + 7);
1621         if (prefix == NULL)
1622                 GOTO(out, rc = -ENOMEM);
1623
1624         snprintf(prefix, MAX_OBD_NAME + 7, "lfsck-%s", lfsck_lfsck2name(lfsck));
1625         rc = seq_client_init(lfsck->li_seq, NULL, LUSTRE_SEQ_METADATA, prefix,
1626                              ss->ss_server_seq);
1627         OBD_FREE(prefix, MAX_OBD_NAME + 7);
1628         if (rc != 0)
1629                 GOTO(out, rc);
1630
1631         if (fid_is_sane(&bk->lb_last_fid))
1632                 lfsck->li_seq->lcs_fid = bk->lb_last_fid;
1633
1634         RETURN(0);
1635
1636 out:
1637         OBD_FREE_PTR(lfsck->li_seq);
1638         lfsck->li_seq = NULL;
1639
1640         return rc;
1641 }
1642
1643 static void lfsck_fid_fini(struct lfsck_instance *lfsck)
1644 {
1645         if (lfsck->li_seq != NULL) {
1646                 seq_client_fini(lfsck->li_seq);
1647                 OBD_FREE_PTR(lfsck->li_seq);
1648                 lfsck->li_seq = NULL;
1649         }
1650 }
1651
1652 void lfsck_instance_cleanup(const struct lu_env *env,
1653                             struct lfsck_instance *lfsck)
1654 {
1655         struct ptlrpc_thread    *thread = &lfsck->li_thread;
1656         struct lfsck_component  *com;
1657         struct lfsck_component  *next;
1658         struct lfsck_lmv_unit   *llu;
1659         struct lfsck_lmv_unit   *llu_next;
1660         struct lfsck_lmv        *llmv;
1661         ENTRY;
1662
1663         LASSERT(list_empty(&lfsck->li_link));
1664         LASSERT(thread_is_init(thread) || thread_is_stopped(thread));
1665
1666         if (lfsck->li_obj_oit != NULL) {
1667                 lfsck_object_put(env, lfsck->li_obj_oit);
1668                 lfsck->li_obj_oit = NULL;
1669         }
1670
1671         LASSERT(lfsck->li_obj_dir == NULL);
1672         LASSERT(lfsck->li_lmv == NULL);
1673
1674         list_for_each_entry_safe(llu, llu_next, &lfsck->li_list_lmv, llu_link) {
1675                 llmv = &llu->llu_lmv;
1676
1677                 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
1678                          "still in using: %u\n",
1679                          atomic_read(&llmv->ll_ref));
1680
1681                 lfsck_lmv_put(env, llmv);
1682         }
1683
1684         list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
1685                 lfsck_component_cleanup(env, com);
1686         }
1687
1688         LASSERT(list_empty(&lfsck->li_list_dir));
1689
1690         list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
1691                                  lc_link) {
1692                 lfsck_component_cleanup(env, com);
1693         }
1694
1695         list_for_each_entry_safe(com, next, &lfsck->li_list_idle, lc_link) {
1696                 lfsck_component_cleanup(env, com);
1697         }
1698
1699         lfsck_tgt_descs_fini(&lfsck->li_ost_descs);
1700         lfsck_tgt_descs_fini(&lfsck->li_mdt_descs);
1701
1702         if (lfsck->li_lfsck_dir != NULL) {
1703                 lfsck_object_put(env, lfsck->li_lfsck_dir);
1704                 lfsck->li_lfsck_dir = NULL;
1705         }
1706
1707         if (lfsck->li_bookmark_obj != NULL) {
1708                 lfsck_object_put(env, lfsck->li_bookmark_obj);
1709                 lfsck->li_bookmark_obj = NULL;
1710         }
1711
1712         if (lfsck->li_lpf_obj != NULL) {
1713                 lfsck_object_put(env, lfsck->li_lpf_obj);
1714                 lfsck->li_lpf_obj = NULL;
1715         }
1716
1717         if (lfsck->li_lpf_root_obj != NULL) {
1718                 lfsck_object_put(env, lfsck->li_lpf_root_obj);
1719                 lfsck->li_lpf_root_obj = NULL;
1720         }
1721
1722         if (lfsck->li_los != NULL) {
1723                 local_oid_storage_fini(env, lfsck->li_los);
1724                 lfsck->li_los = NULL;
1725         }
1726
1727         lfsck_fid_fini(lfsck);
1728
1729         OBD_FREE_PTR(lfsck);
1730 }
1731
1732 static inline struct lfsck_instance *
1733 __lfsck_instance_find(struct dt_device *key, bool ref, bool unlink)
1734 {
1735         struct lfsck_instance *lfsck;
1736
1737         list_for_each_entry(lfsck, &lfsck_instance_list, li_link) {
1738                 if (lfsck->li_bottom == key) {
1739                         if (ref)
1740                                 lfsck_instance_get(lfsck);
1741                         if (unlink)
1742                                 list_del_init(&lfsck->li_link);
1743
1744                         return lfsck;
1745                 }
1746         }
1747
1748         return NULL;
1749 }
1750
1751 struct lfsck_instance *lfsck_instance_find(struct dt_device *key, bool ref,
1752                                            bool unlink)
1753 {
1754         struct lfsck_instance *lfsck;
1755
1756         spin_lock(&lfsck_instance_lock);
1757         lfsck = __lfsck_instance_find(key, ref, unlink);
1758         spin_unlock(&lfsck_instance_lock);
1759
1760         return lfsck;
1761 }
1762
1763 static inline int lfsck_instance_add(struct lfsck_instance *lfsck)
1764 {
1765         struct lfsck_instance *tmp;
1766
1767         spin_lock(&lfsck_instance_lock);
1768         list_for_each_entry(tmp, &lfsck_instance_list, li_link) {
1769                 if (lfsck->li_bottom == tmp->li_bottom) {
1770                         spin_unlock(&lfsck_instance_lock);
1771                         return -EEXIST;
1772                 }
1773         }
1774
1775         list_add_tail(&lfsck->li_link, &lfsck_instance_list);
1776         spin_unlock(&lfsck_instance_lock);
1777         return 0;
1778 }
1779
1780 void lfsck_bits_dump(struct seq_file *m, int bits, const char *names[],
1781                      const char *prefix)
1782 {
1783         int flag;
1784         int i;
1785         bool newline = (bits != 0 ? false : true);
1786
1787         seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
1788
1789         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
1790                 if (flag & bits) {
1791                         bits &= ~flag;
1792                         if (names[i] != NULL) {
1793                                 if (bits == 0)
1794                                         newline = true;
1795
1796                                 seq_printf(m, "%s%c", names[i],
1797                                            newline ? '\n' : ',');
1798                         }
1799                 }
1800         }
1801
1802         if (!newline)
1803                 seq_putc(m, '\n');
1804 }
1805
1806 void lfsck_time_dump(struct seq_file *m, time64_t time, const char *name)
1807 {
1808         if (time == 0) {
1809                 seq_printf(m, "%s_time: N/A\n", name);
1810                 seq_printf(m, "time_since_%s: N/A\n", name);
1811         } else {
1812                 seq_printf(m, "%s_time: %lld\n", name, time);
1813                 seq_printf(m, "time_since_%s: %lld seconds\n",
1814                            name, ktime_get_real_seconds() - time);
1815         }
1816 }
1817
1818 void lfsck_pos_dump(struct seq_file *m, struct lfsck_position *pos,
1819                     const char *prefix)
1820 {
1821         if (fid_is_zero(&pos->lp_dir_parent)) {
1822                 if (pos->lp_oit_cookie == 0) {
1823                         seq_printf(m, "%s: N/A, N/A, N/A\n", prefix);
1824                         return;
1825                 }
1826                 seq_printf(m, "%s: %llu, N/A, N/A\n",
1827                            prefix, pos->lp_oit_cookie);
1828         } else {
1829                 seq_printf(m, "%s: %llu, "DFID", %#llx\n",
1830                            prefix, pos->lp_oit_cookie,
1831                            PFID(&pos->lp_dir_parent), pos->lp_dir_cookie);
1832         }
1833 }
1834
1835 void lfsck_pos_fill(const struct lu_env *env, struct lfsck_instance *lfsck,
1836                     struct lfsck_position *pos, bool init)
1837 {
1838         const struct dt_it_ops *iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
1839
1840         if (unlikely(lfsck->li_di_oit == NULL)) {
1841                 memset(pos, 0, sizeof(*pos));
1842                 return;
1843         }
1844
1845         pos->lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
1846         if (!lfsck->li_current_oit_processed && !init)
1847                 pos->lp_oit_cookie--;
1848
1849         if (unlikely(pos->lp_oit_cookie == 0))
1850                 pos->lp_oit_cookie = 1;
1851
1852         if (lfsck->li_di_dir != NULL) {
1853                 struct dt_object *dto = lfsck->li_obj_dir;
1854
1855                 pos->lp_dir_cookie = dto->do_index_ops->dio_it.store(env,
1856                                                         lfsck->li_di_dir);
1857
1858                 if (pos->lp_dir_cookie >= MDS_DIR_END_OFF) {
1859                         fid_zero(&pos->lp_dir_parent);
1860                         pos->lp_dir_cookie = 0;
1861                 } else {
1862                         pos->lp_dir_parent = *lfsck_dto2fid(dto);
1863                 }
1864         } else {
1865                 fid_zero(&pos->lp_dir_parent);
1866                 pos->lp_dir_cookie = 0;
1867         }
1868 }
1869
1870 bool __lfsck_set_speed(struct lfsck_instance *lfsck, __u32 limit)
1871 {
1872         bool dirty = false;
1873
1874         if (limit != LFSCK_SPEED_NO_LIMIT) {
1875                 if (limit > cfs_time_seconds(1)) {
1876                         lfsck->li_sleep_rate = limit / cfs_time_seconds(1);
1877                         lfsck->li_sleep_jif = 1;
1878                 } else {
1879                         lfsck->li_sleep_rate = 1;
1880                         lfsck->li_sleep_jif = cfs_time_seconds(1) / limit;
1881                 }
1882         } else {
1883                 lfsck->li_sleep_jif = 0;
1884                 lfsck->li_sleep_rate = 0;
1885         }
1886
1887         if (lfsck->li_bookmark_ram.lb_speed_limit != limit) {
1888                 lfsck->li_bookmark_ram.lb_speed_limit = limit;
1889                 dirty = true;
1890         }
1891
1892         return dirty;
1893 }
1894
1895 void lfsck_control_speed(struct lfsck_instance *lfsck)
1896 {
1897         struct ptlrpc_thread *thread = &lfsck->li_thread;
1898         struct l_wait_info    lwi;
1899
1900         if (lfsck->li_sleep_jif > 0 &&
1901             lfsck->li_new_scanned >= lfsck->li_sleep_rate) {
1902                 lwi = LWI_TIMEOUT_INTR(lfsck->li_sleep_jif, NULL,
1903                                        LWI_ON_SIGNAL_NOOP, NULL);
1904
1905                 l_wait_event(thread->t_ctl_waitq,
1906                              !thread_is_running(thread),
1907                              &lwi);
1908                 lfsck->li_new_scanned = 0;
1909         }
1910 }
1911
1912 void lfsck_control_speed_by_self(struct lfsck_component *com)
1913 {
1914         struct lfsck_instance   *lfsck  = com->lc_lfsck;
1915         struct ptlrpc_thread    *thread = &lfsck->li_thread;
1916         struct l_wait_info       lwi;
1917
1918         if (lfsck->li_sleep_jif > 0 &&
1919             com->lc_new_scanned >= lfsck->li_sleep_rate) {
1920                 lwi = LWI_TIMEOUT_INTR(lfsck->li_sleep_jif, NULL,
1921                                        LWI_ON_SIGNAL_NOOP, NULL);
1922
1923                 l_wait_event(thread->t_ctl_waitq,
1924                              !thread_is_running(thread),
1925                              &lwi);
1926                 com->lc_new_scanned = 0;
1927         }
1928 }
1929
1930 static struct lfsck_thread_args *
1931 lfsck_thread_args_init(struct lfsck_instance *lfsck,
1932                        struct lfsck_component *com,
1933                        struct lfsck_start_param *lsp)
1934 {
1935         struct lfsck_thread_args *lta;
1936         int                       rc;
1937
1938         OBD_ALLOC_PTR(lta);
1939         if (lta == NULL)
1940                 return ERR_PTR(-ENOMEM);
1941
1942         rc = lu_env_init(&lta->lta_env, LCT_MD_THREAD | LCT_DT_THREAD);
1943         if (rc != 0) {
1944                 OBD_FREE_PTR(lta);
1945                 return ERR_PTR(rc);
1946         }
1947
1948         lta->lta_lfsck = lfsck_instance_get(lfsck);
1949         if (com != NULL)
1950                 lta->lta_com = lfsck_component_get(com);
1951
1952         lta->lta_lsp = lsp;
1953
1954         return lta;
1955 }
1956
1957 void lfsck_thread_args_fini(struct lfsck_thread_args *lta)
1958 {
1959         if (lta->lta_com != NULL)
1960                 lfsck_component_put(&lta->lta_env, lta->lta_com);
1961         lfsck_instance_put(&lta->lta_env, lta->lta_lfsck);
1962         lu_env_fini(&lta->lta_env);
1963         OBD_FREE_PTR(lta);
1964 }
1965
1966 struct lfsck_assistant_data *
1967 lfsck_assistant_data_init(struct lfsck_assistant_operations *lao,
1968                           const char *name)
1969 {
1970         struct lfsck_assistant_data *lad;
1971
1972         OBD_ALLOC_PTR(lad);
1973         if (lad != NULL) {
1974                 lad->lad_bitmap = CFS_ALLOCATE_BITMAP(BITS_PER_LONG);
1975                 if (lad->lad_bitmap == NULL) {
1976                         OBD_FREE_PTR(lad);
1977                         return NULL;
1978                 }
1979
1980                 INIT_LIST_HEAD(&lad->lad_req_list);
1981                 spin_lock_init(&lad->lad_lock);
1982                 INIT_LIST_HEAD(&lad->lad_ost_list);
1983                 INIT_LIST_HEAD(&lad->lad_ost_phase1_list);
1984                 INIT_LIST_HEAD(&lad->lad_ost_phase2_list);
1985                 INIT_LIST_HEAD(&lad->lad_mdt_list);
1986                 INIT_LIST_HEAD(&lad->lad_mdt_phase1_list);
1987                 INIT_LIST_HEAD(&lad->lad_mdt_phase2_list);
1988                 init_waitqueue_head(&lad->lad_thread.t_ctl_waitq);
1989                 lad->lad_ops = lao;
1990                 lad->lad_name = name;
1991         }
1992
1993         return lad;
1994 }
1995
1996 struct lfsck_assistant_object *
1997 lfsck_assistant_object_init(const struct lu_env *env, const struct lu_fid *fid,
1998                             const struct lu_attr *attr, __u64 cookie,
1999                             bool is_dir)
2000 {
2001         struct lfsck_assistant_object   *lso;
2002
2003         OBD_ALLOC_PTR(lso);
2004         if (lso == NULL)
2005                 return ERR_PTR(-ENOMEM);
2006
2007         lso->lso_fid = *fid;
2008         if (attr != NULL)
2009                 lso->lso_attr = *attr;
2010
2011         atomic_set(&lso->lso_ref, 1);
2012         lso->lso_oit_cookie = cookie;
2013         if (is_dir)
2014                 lso->lso_is_dir = 1;
2015
2016         return lso;
2017 }
2018
2019 struct dt_object *
2020 lfsck_assistant_object_load(const struct lu_env *env,
2021                             struct lfsck_instance *lfsck,
2022                             struct lfsck_assistant_object *lso)
2023 {
2024         struct dt_object *obj;
2025
2026         obj = lfsck_object_find_bottom(env, lfsck, &lso->lso_fid);
2027         if (IS_ERR(obj))
2028                 return obj;
2029
2030         if (unlikely(!dt_object_exists(obj) || lfsck_is_dead_obj(obj))) {
2031                 lso->lso_dead = 1;
2032                 lfsck_object_put(env, obj);
2033
2034                 return ERR_PTR(-ENOENT);
2035         }
2036
2037         if (lso->lso_is_dir && unlikely(!dt_try_as_dir(env, obj))) {
2038                 lfsck_object_put(env, obj);
2039
2040                 return ERR_PTR(-ENOTDIR);
2041         }
2042
2043         return obj;
2044 }
2045
2046 /**
2047  * Generic LFSCK asynchronous communication interpretor function.
2048  * The LFSCK RPC reply for both the event notification and status
2049  * querying will be handled here.
2050  *
2051  * \param[in] env       pointer to the thread context
2052  * \param[in] req       pointer to the LFSCK request
2053  * \param[in] args      pointer to the lfsck_async_interpret_args
2054  * \param[in] rc        the result for handling the LFSCK request
2055  *
2056  * \retval              0 for success
2057  * \retval              negative error number on failure
2058  */
2059 int lfsck_async_interpret_common(const struct lu_env *env,
2060                                  struct ptlrpc_request *req,
2061                                  void *args, int rc)
2062 {
2063         struct lfsck_async_interpret_args *laia = args;
2064         struct lfsck_component            *com  = laia->laia_com;
2065         struct lfsck_assistant_data       *lad  = com->lc_data;
2066         struct lfsck_tgt_descs            *ltds = laia->laia_ltds;
2067         struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
2068         struct lfsck_request              *lr   = laia->laia_lr;
2069
2070         LASSERT(com->lc_lfsck->li_master);
2071
2072         switch (lr->lr_event) {
2073         case LE_START:
2074                 if (unlikely(rc == -EINPROGRESS)) {
2075                         ltd->ltd_retry_start = 1;
2076                         break;
2077                 }
2078
2079                 if (rc != 0) {
2080                         CDEBUG(D_LFSCK, "%s: fail to notify %s %x for %s "
2081                                "start: rc = %d\n",
2082                                lfsck_lfsck2name(com->lc_lfsck),
2083                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2084                                ltd->ltd_index, lad->lad_name, rc);
2085
2086                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2087                                 struct lfsck_layout *lo = com->lc_file_ram;
2088
2089                                 if (lr->lr_flags & LEF_TO_OST)
2090                                         lfsck_lad_set_bitmap(env, com,
2091                                                              ltd->ltd_index);
2092                                 else
2093                                         lo->ll_flags |= LF_INCOMPLETE;
2094                         } else {
2095                                 struct lfsck_namespace *ns = com->lc_file_ram;
2096
2097                                 /* If some MDT does not join the namespace
2098                                  * LFSCK, then we cannot know whether there
2099                                  * is some name entry on such MDT that with
2100                                  * the referenced MDT-object on this MDT or
2101                                  * not. So the namespace LFSCK on this MDT
2102                                  * cannot handle orphan MDT-objects properly.
2103                                  * So we mark the LFSCK as LF_INCOMPLETE and
2104                                  * skip orphan MDT-objects handling. */
2105                                 ns->ln_flags |= LF_INCOMPLETE;
2106                         }
2107                         break;
2108                 }
2109
2110                 spin_lock(&ltds->ltd_lock);
2111                 if (ltd->ltd_dead) {
2112                         spin_unlock(&ltds->ltd_lock);
2113                         break;
2114                 }
2115
2116                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2117                         struct list_head *list;
2118                         struct list_head *phase_list;
2119
2120                         if (ltd->ltd_layout_done) {
2121                                 spin_unlock(&ltds->ltd_lock);
2122                                 break;
2123                         }
2124
2125                         if (lr->lr_flags & LEF_TO_OST) {
2126                                 list = &lad->lad_ost_list;
2127                                 phase_list = &lad->lad_ost_phase1_list;
2128                         } else {
2129                                 list = &lad->lad_mdt_list;
2130                                 phase_list = &lad->lad_mdt_phase1_list;
2131                         }
2132
2133                         if (list_empty(&ltd->ltd_layout_list))
2134                                 list_add_tail(&ltd->ltd_layout_list, list);
2135                         if (list_empty(&ltd->ltd_layout_phase_list))
2136                                 list_add_tail(&ltd->ltd_layout_phase_list,
2137                                               phase_list);
2138                 } else {
2139                         if (ltd->ltd_namespace_done) {
2140                                 spin_unlock(&ltds->ltd_lock);
2141                                 break;
2142                         }
2143
2144                         if (list_empty(&ltd->ltd_namespace_list))
2145                                 list_add_tail(&ltd->ltd_namespace_list,
2146                                               &lad->lad_mdt_list);
2147                         if (list_empty(&ltd->ltd_namespace_phase_list))
2148                                 list_add_tail(&ltd->ltd_namespace_phase_list,
2149                                               &lad->lad_mdt_phase1_list);
2150                 }
2151                 spin_unlock(&ltds->ltd_lock);
2152                 break;
2153         case LE_STOP:
2154         case LE_PHASE1_DONE:
2155         case LE_PHASE2_DONE:
2156         case LE_PEER_EXIT:
2157                 if (rc != 0 && rc != -EALREADY)
2158                         CDEBUG(D_LFSCK, "%s: fail to notify %s %x for %s: "
2159                               "event = %d, rc = %d\n",
2160                               lfsck_lfsck2name(com->lc_lfsck),
2161                               (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2162                               ltd->ltd_index, lad->lad_name, lr->lr_event, rc);
2163                 break;
2164         case LE_QUERY: {
2165                 struct lfsck_reply *reply;
2166                 struct list_head *list;
2167                 struct list_head *phase_list;
2168
2169                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2170                         list = &ltd->ltd_layout_list;
2171                         phase_list = &ltd->ltd_layout_phase_list;
2172                 } else {
2173                         list = &ltd->ltd_namespace_list;
2174                         phase_list = &ltd->ltd_namespace_phase_list;
2175                 }
2176
2177                 if (rc != 0) {
2178                         if (lr->lr_flags & LEF_QUERY_ALL) {
2179                                 lfsck_reset_ltd_status(ltd, com->lc_type);
2180                                 break;
2181                         }
2182
2183                         spin_lock(&ltds->ltd_lock);
2184                         list_del_init(phase_list);
2185                         list_del_init(list);
2186                         spin_unlock(&ltds->ltd_lock);
2187                         break;
2188                 }
2189
2190                 reply = req_capsule_server_get(&req->rq_pill,
2191                                                &RMF_LFSCK_REPLY);
2192                 if (reply == NULL) {
2193                         rc = -EPROTO;
2194                         CDEBUG(D_LFSCK, "%s: invalid query reply for %s: "
2195                                "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
2196                                lad->lad_name, rc);
2197
2198                         if (lr->lr_flags & LEF_QUERY_ALL) {
2199                                 lfsck_reset_ltd_status(ltd, com->lc_type);
2200                                 break;
2201                         }
2202
2203                         spin_lock(&ltds->ltd_lock);
2204                         list_del_init(phase_list);
2205                         list_del_init(list);
2206                         spin_unlock(&ltds->ltd_lock);
2207                         break;
2208                 }
2209
2210                 if (lr->lr_flags & LEF_QUERY_ALL) {
2211                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2212                                 ltd->ltd_layout_status = reply->lr_status;
2213                                 ltd->ltd_layout_repaired = reply->lr_repaired;
2214                         } else {
2215                                 ltd->ltd_namespace_status = reply->lr_status;
2216                                 ltd->ltd_namespace_repaired =
2217                                                         reply->lr_repaired;
2218                         }
2219                         break;
2220                 }
2221
2222                 switch (reply->lr_status) {
2223                 case LS_SCANNING_PHASE1:
2224                         break;
2225                 case LS_SCANNING_PHASE2:
2226                         spin_lock(&ltds->ltd_lock);
2227                         list_del_init(phase_list);
2228                         if (ltd->ltd_dead) {
2229                                 spin_unlock(&ltds->ltd_lock);
2230                                 break;
2231                         }
2232
2233                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2234                                 if (ltd->ltd_layout_done) {
2235                                         spin_unlock(&ltds->ltd_lock);
2236                                         break;
2237                                 }
2238
2239                                 if (lr->lr_flags & LEF_TO_OST)
2240                                         list_add_tail(phase_list,
2241                                                 &lad->lad_ost_phase2_list);
2242                                 else
2243                                         list_add_tail(phase_list,
2244                                                 &lad->lad_mdt_phase2_list);
2245                         } else {
2246                                 if (ltd->ltd_namespace_done) {
2247                                         spin_unlock(&ltds->ltd_lock);
2248                                         break;
2249                                 }
2250
2251                                 list_add_tail(phase_list,
2252                                               &lad->lad_mdt_phase2_list);
2253                         }
2254                         spin_unlock(&ltds->ltd_lock);
2255                         break;
2256                 default:
2257                         spin_lock(&ltds->ltd_lock);
2258                         list_del_init(phase_list);
2259                         list_del_init(list);
2260                         spin_unlock(&ltds->ltd_lock);
2261                         break;
2262                 }
2263                 break;
2264         }
2265         default:
2266                 CDEBUG(D_LFSCK, "%s: unexpected event: rc = %d\n",
2267                        lfsck_lfsck2name(com->lc_lfsck), lr->lr_event);
2268                 break;
2269         }
2270
2271         if (!laia->laia_shared) {
2272                 lfsck_tgt_put(ltd);
2273                 lfsck_component_put(env, com);
2274         }
2275
2276         return 0;
2277 }
2278
2279 static void lfsck_interpret(const struct lu_env *env,
2280                             struct lfsck_instance *lfsck,
2281                             struct ptlrpc_request *req, void *args, int result)
2282 {
2283         struct lfsck_async_interpret_args *laia = args;
2284         struct lfsck_component            *com;
2285
2286         LASSERT(laia->laia_com == NULL);
2287         LASSERT(laia->laia_shared);
2288
2289         spin_lock(&lfsck->li_lock);
2290         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
2291                 laia->laia_com = com;
2292                 lfsck_async_interpret_common(env, req, laia, result);
2293         }
2294
2295         list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
2296                 laia->laia_com = com;
2297                 lfsck_async_interpret_common(env, req, laia, result);
2298         }
2299         spin_unlock(&lfsck->li_lock);
2300 }
2301
2302 static int lfsck_stop_notify(const struct lu_env *env,
2303                              struct lfsck_instance *lfsck,
2304                              struct lfsck_tgt_descs *ltds,
2305                              struct lfsck_tgt_desc *ltd, __u16 type)
2306 {
2307         struct lfsck_component *com;
2308         int                     rc = 0;
2309         ENTRY;
2310
2311         LASSERT(lfsck->li_master);
2312
2313         spin_lock(&lfsck->li_lock);
2314         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_scan);
2315         if (com == NULL)
2316                 com = __lfsck_component_find(lfsck, type,
2317                                              &lfsck->li_list_double_scan);
2318         if (com != NULL)
2319                 lfsck_component_get(com);
2320         spin_unlock(&lfsck->li_lock);
2321
2322         if (com != NULL) {
2323                 struct lfsck_thread_info          *info  = lfsck_env_info(env);
2324                 struct lfsck_async_interpret_args *laia  = &info->lti_laia;
2325                 struct lfsck_request              *lr    = &info->lti_lr;
2326                 struct lfsck_assistant_data       *lad   = com->lc_data;
2327                 struct list_head                  *list;
2328                 struct list_head                  *phase_list;
2329                 struct ptlrpc_request_set         *set;
2330
2331                 set = ptlrpc_prep_set();
2332                 if (set == NULL) {
2333                         lfsck_component_put(env, com);
2334
2335                         RETURN(-ENOMEM);
2336                 }
2337
2338                 if (type == LFSCK_TYPE_LAYOUT) {
2339                         list = &ltd->ltd_layout_list;
2340                         phase_list = &ltd->ltd_layout_phase_list;
2341                 } else {
2342                         list = &ltd->ltd_namespace_list;
2343                         phase_list = &ltd->ltd_namespace_phase_list;
2344                 }
2345
2346                 spin_lock(&ltds->ltd_lock);
2347                 if (list_empty(list)) {
2348                         LASSERT(list_empty(phase_list));
2349                         spin_unlock(&ltds->ltd_lock);
2350                         ptlrpc_set_destroy(set);
2351
2352                         RETURN(0);
2353                 }
2354
2355                 list_del_init(phase_list);
2356                 list_del_init(list);
2357                 spin_unlock(&ltds->ltd_lock);
2358
2359                 memset(lr, 0, sizeof(*lr));
2360                 lr->lr_index = lfsck_dev_idx(lfsck);
2361                 lr->lr_event = LE_PEER_EXIT;
2362                 lr->lr_active = type;
2363                 lr->lr_status = LS_CO_PAUSED;
2364                 if (ltds == &lfsck->li_ost_descs)
2365                         lr->lr_flags = LEF_TO_OST;
2366
2367                 memset(laia, 0, sizeof(*laia));
2368                 laia->laia_com = com;
2369                 laia->laia_ltds = ltds;
2370                 atomic_inc(&ltd->ltd_ref);
2371                 laia->laia_ltd = ltd;
2372                 laia->laia_lr = lr;
2373
2374                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2375                                          lfsck_async_interpret_common,
2376                                          laia, LFSCK_NOTIFY);
2377                 if (rc != 0) {
2378                         CDEBUG(D_LFSCK, "%s: fail to notify %s %x for "
2379                                "co-stop for %s: rc = %d\n",
2380                                lfsck_lfsck2name(lfsck),
2381                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2382                                ltd->ltd_index, lad->lad_name, rc);
2383                         lfsck_tgt_put(ltd);
2384                 } else {
2385                         rc = ptlrpc_set_wait(env, set);
2386                 }
2387
2388                 ptlrpc_set_destroy(set);
2389                 lfsck_component_put(env, com);
2390         }
2391
2392         RETURN(rc);
2393 }
2394
2395 static int lfsck_async_interpret(const struct lu_env *env,
2396                                  struct ptlrpc_request *req,
2397                                  void *args, int rc)
2398 {
2399         struct lfsck_async_interpret_args *laia = args;
2400         struct lfsck_instance             *lfsck;
2401
2402         lfsck = container_of0(laia->laia_ltds, struct lfsck_instance,
2403                               li_mdt_descs);
2404         lfsck_interpret(env, lfsck, req, laia, rc);
2405         lfsck_tgt_put(laia->laia_ltd);
2406         if (rc != 0 && laia->laia_result != -EALREADY)
2407                 laia->laia_result = rc;
2408
2409         return 0;
2410 }
2411
2412 int lfsck_async_request(const struct lu_env *env, struct obd_export *exp,
2413                         struct lfsck_request *lr,
2414                         struct ptlrpc_request_set *set,
2415                         ptlrpc_interpterer_t interpreter,
2416                         void *args, int request)
2417 {
2418         struct lfsck_async_interpret_args *laia;
2419         struct ptlrpc_request             *req;
2420         struct lfsck_request              *tmp;
2421         struct req_format                 *format;
2422         int                                rc;
2423
2424         switch (request) {
2425         case LFSCK_NOTIFY:
2426                 format = &RQF_LFSCK_NOTIFY;
2427                 break;
2428         case LFSCK_QUERY:
2429                 format = &RQF_LFSCK_QUERY;
2430                 break;
2431         default:
2432                 CDEBUG(D_LFSCK, "%s: unknown async request %d: rc = %d\n",
2433                        exp->exp_obd->obd_name, request, -EINVAL);
2434                 return -EINVAL;
2435         }
2436
2437         req = ptlrpc_request_alloc(class_exp2cliimp(exp), format);
2438         if (req == NULL)
2439                 return -ENOMEM;
2440
2441         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, request);
2442         if (rc != 0) {
2443                 ptlrpc_request_free(req);
2444
2445                 return rc;
2446         }
2447
2448         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2449         *tmp = *lr;
2450         ptlrpc_request_set_replen(req);
2451
2452         laia = ptlrpc_req_async_args(laia, req);
2453         *laia = *(struct lfsck_async_interpret_args *)args;
2454         if (laia->laia_com != NULL)
2455                 lfsck_component_get(laia->laia_com);
2456         req->rq_interpret_reply = interpreter;
2457         req->rq_allow_intr = 1;
2458         req->rq_no_delay = 1;
2459         ptlrpc_set_add_req(set, req);
2460
2461         return 0;
2462 }
2463
2464 int lfsck_query_all(const struct lu_env *env, struct lfsck_component *com)
2465 {
2466         struct lfsck_thread_info          *info  = lfsck_env_info(env);
2467         struct lfsck_request              *lr    = &info->lti_lr;
2468         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
2469         struct lfsck_instance             *lfsck = com->lc_lfsck;
2470         struct lfsck_tgt_descs            *ltds  = &lfsck->li_mdt_descs;
2471         struct lfsck_tgt_desc             *ltd;
2472         struct ptlrpc_request_set         *set;
2473         int                                idx;
2474         int                                rc;
2475         ENTRY;
2476
2477         memset(lr, 0, sizeof(*lr));
2478         lr->lr_event = LE_QUERY;
2479         lr->lr_active = com->lc_type;
2480         lr->lr_flags = LEF_QUERY_ALL;
2481
2482         memset(laia, 0, sizeof(*laia));
2483         laia->laia_com = com;
2484         laia->laia_lr = lr;
2485
2486         set = ptlrpc_prep_set();
2487         if (set == NULL)
2488                 RETURN(-ENOMEM);
2489
2490 again:
2491         laia->laia_ltds = ltds;
2492         down_read(&ltds->ltd_rw_sem);
2493         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
2494                 ltd = lfsck_tgt_get(ltds, idx);
2495                 LASSERT(ltd != NULL);
2496
2497                 laia->laia_ltd = ltd;
2498                 up_read(&ltds->ltd_rw_sem);
2499                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2500                                          lfsck_async_interpret_common,
2501                                          laia, LFSCK_QUERY);
2502                 if (rc != 0) {
2503                         struct lfsck_assistant_data *lad = com->lc_data;
2504
2505                         CDEBUG(D_LFSCK, "%s: Fail to query %s %x for stat %s: "
2506                                "rc = %d\n", lfsck_lfsck2name(lfsck),
2507                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2508                                ltd->ltd_index, lad->lad_name, rc);
2509                         lfsck_reset_ltd_status(ltd, com->lc_type);
2510                         lfsck_tgt_put(ltd);
2511                 }
2512                 down_read(&ltds->ltd_rw_sem);
2513         }
2514         up_read(&ltds->ltd_rw_sem);
2515
2516         if (com->lc_type == LFSCK_TYPE_LAYOUT && !(lr->lr_flags & LEF_TO_OST)) {
2517                 ltds = &lfsck->li_ost_descs;
2518                 lr->lr_flags |= LEF_TO_OST;
2519                 goto again;
2520         }
2521
2522         rc = ptlrpc_set_wait(env, set);
2523         ptlrpc_set_destroy(set);
2524
2525         RETURN(rc);
2526 }
2527
2528 int lfsck_start_assistant(const struct lu_env *env, struct lfsck_component *com,
2529                           struct lfsck_start_param *lsp)
2530 {
2531         struct lfsck_instance           *lfsck   = com->lc_lfsck;
2532         struct lfsck_assistant_data     *lad     = com->lc_data;
2533         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
2534         struct ptlrpc_thread            *athread = &lad->lad_thread;
2535         struct lfsck_thread_args        *lta;
2536         struct task_struct              *task;
2537         int                              rc;
2538         ENTRY;
2539
2540         lad->lad_assistant_status = 0;
2541         lad->lad_post_result = 0;
2542         lad->lad_flags = 0;
2543         lad->lad_advance_lock = false;
2544         thread_set_flags(athread, 0);
2545
2546         lta = lfsck_thread_args_init(lfsck, com, lsp);
2547         if (IS_ERR(lta))
2548                 RETURN(PTR_ERR(lta));
2549
2550         task = kthread_run(lfsck_assistant_engine, lta, lad->lad_name);
2551         if (IS_ERR(task)) {
2552                 rc = PTR_ERR(task);
2553                 CERROR("%s: cannot start LFSCK assistant thread for %s: "
2554                        "rc = %d\n", lfsck_lfsck2name(lfsck), lad->lad_name, rc);
2555                 lfsck_thread_args_fini(lta);
2556         } else {
2557                 wait_event_idle(mthread->t_ctl_waitq,
2558                                 thread_is_running(athread) ||
2559                                 thread_is_stopped(athread) ||
2560                                 !thread_is_starting(mthread));
2561                 if (unlikely(!thread_is_starting(mthread)))
2562                         /* stopped by race */
2563                         rc = -ESRCH;
2564                 else if (unlikely(!thread_is_running(athread)))
2565                         rc = lad->lad_assistant_status;
2566                 else
2567                         rc = 0;
2568         }
2569
2570         RETURN(rc);
2571 }
2572
2573 int lfsck_checkpoint_generic(const struct lu_env *env,
2574                              struct lfsck_component *com)
2575 {
2576         struct lfsck_assistant_data     *lad     = com->lc_data;
2577         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2578         struct ptlrpc_thread            *athread = &lad->lad_thread;
2579
2580         wait_event_idle(mthread->t_ctl_waitq,
2581                         list_empty(&lad->lad_req_list) ||
2582                         !thread_is_running(mthread) ||
2583                         thread_is_stopped(athread));
2584
2585         if (!thread_is_running(mthread) || thread_is_stopped(athread))
2586                 return LFSCK_CHECKPOINT_SKIP;
2587
2588         return 0;
2589 }
2590
2591 void lfsck_post_generic(const struct lu_env *env,
2592                         struct lfsck_component *com, int *result)
2593 {
2594         struct lfsck_assistant_data     *lad     = com->lc_data;
2595         struct ptlrpc_thread            *athread = &lad->lad_thread;
2596         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2597
2598         lad->lad_post_result = *result;
2599         if (*result <= 0)
2600                 set_bit(LAD_EXIT, &lad->lad_flags);
2601         set_bit(LAD_TO_POST, &lad->lad_flags);
2602
2603         CDEBUG(D_LFSCK, "%s: waiting for assistant to do %s post, rc = %d\n",
2604                lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, *result);
2605
2606         wake_up_all(&athread->t_ctl_waitq);
2607         wait_event_idle(mthread->t_ctl_waitq,
2608                         (*result > 0 && list_empty(&lad->lad_req_list)) ||
2609                         thread_is_stopped(athread));
2610
2611         if (lad->lad_assistant_status < 0)
2612                 *result = lad->lad_assistant_status;
2613
2614         CDEBUG(D_LFSCK, "%s: the assistant has done %s post, rc = %d\n",
2615                lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, *result);
2616 }
2617
2618 int lfsck_double_scan_generic(const struct lu_env *env,
2619                               struct lfsck_component *com, int status)
2620 {
2621         struct lfsck_assistant_data     *lad     = com->lc_data;
2622         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2623         struct ptlrpc_thread            *athread = &lad->lad_thread;
2624
2625         if (status != LS_SCANNING_PHASE2)
2626                 set_bit(LAD_EXIT, &lad->lad_flags);
2627         else
2628                 set_bit(LAD_TO_DOUBLE_SCAN, &lad->lad_flags);
2629
2630         CDEBUG(D_LFSCK, "%s: waiting for assistant to do %s double_scan, "
2631                "status %d\n",
2632                lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, status);
2633
2634         wake_up_all(&athread->t_ctl_waitq);
2635         wait_event_idle(mthread->t_ctl_waitq,
2636                         test_bit(LAD_IN_DOUBLE_SCAN, &lad->lad_flags) ||
2637                         thread_is_stopped(athread));
2638
2639         CDEBUG(D_LFSCK, "%s: the assistant has done %s double_scan, "
2640                "status %d\n", lfsck_lfsck2name(com->lc_lfsck), lad->lad_name,
2641                lad->lad_assistant_status);
2642
2643         if (lad->lad_assistant_status < 0)
2644                 return lad->lad_assistant_status;
2645
2646         return 0;
2647 }
2648
2649 void lfsck_quit_generic(const struct lu_env *env,
2650                         struct lfsck_component *com)
2651 {
2652         struct lfsck_assistant_data     *lad     = com->lc_data;
2653         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2654         struct ptlrpc_thread            *athread = &lad->lad_thread;
2655
2656         set_bit(LAD_EXIT, &lad->lad_flags);
2657         wake_up_all(&athread->t_ctl_waitq);
2658         wait_event_idle(mthread->t_ctl_waitq,
2659                         thread_is_init(athread) ||
2660                         thread_is_stopped(athread));
2661 }
2662
2663 int lfsck_load_one_trace_file(const struct lu_env *env,
2664                               struct lfsck_component *com,
2665                               struct dt_object *parent,
2666                               struct dt_object **child,
2667                               const struct dt_index_features *ft,
2668                               const char *name, bool reset)
2669 {
2670         struct lfsck_instance *lfsck = com->lc_lfsck;
2671         struct dt_object *obj;
2672         int rc;
2673         ENTRY;
2674
2675         if (*child != NULL) {
2676                 struct dt_it *it;
2677                 const struct dt_it_ops *iops;
2678                 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid3;
2679
2680                 if (!reset)
2681                         RETURN(0);
2682
2683                 obj = *child;
2684                 rc = obj->do_ops->do_index_try(env, obj, ft);
2685                 if (rc)
2686                         /* unlink by force */
2687                         goto unlink;
2688
2689                 iops = &obj->do_index_ops->dio_it;
2690                 it = iops->init(env, obj, 0);
2691                 if (IS_ERR(it))
2692                         /* unlink by force */
2693                         goto unlink;
2694
2695                 fid_zero(fid);
2696                 rc = iops->get(env, it, (const struct dt_key *)fid);
2697                 if (rc >= 0) {
2698                         rc = iops->next(env, it);
2699                         iops->put(env, it);
2700                 }
2701                 iops->fini(env, it);
2702                 if (rc > 0)
2703                         /* "rc > 0" means the index file is empty. */
2704                         RETURN(0);
2705
2706 unlink:
2707                 /* The old index is not empty, remove it firstly. */
2708                 rc = local_object_unlink(env, lfsck->li_bottom, parent, name);
2709                 CDEBUG_LIMIT(rc ? D_ERROR : D_LFSCK,
2710                              "%s: unlink lfsck sub trace file %s: rc = %d\n",
2711                              lfsck_lfsck2name(com->lc_lfsck), name, rc);
2712                 if (rc)
2713                         RETURN(rc);
2714
2715                 if (*child) {
2716                         lfsck_object_put(env, *child);
2717                         *child = NULL;
2718                 }
2719         } else if (reset) {
2720                 goto unlink;
2721         }
2722
2723         obj = local_index_find_or_create(env, lfsck->li_los, parent, name,
2724                                          S_IFREG | S_IRUGO | S_IWUSR, ft);
2725         if (IS_ERR(obj))
2726                 RETURN(PTR_ERR(obj));
2727
2728         rc = obj->do_ops->do_index_try(env, obj, ft);
2729         if (rc) {
2730                 lfsck_object_put(env, obj);
2731                 CDEBUG(D_LFSCK, "%s: LFSCK fail to load "
2732                        "sub trace file %s: rc = %d\n",
2733                        lfsck_lfsck2name(com->lc_lfsck), name, rc);
2734         } else {
2735                 *child = obj;
2736         }
2737
2738         RETURN(rc);
2739 }
2740
2741 int lfsck_load_sub_trace_files(const struct lu_env *env,
2742                                struct lfsck_component *com,
2743                                const struct dt_index_features *ft,
2744                                const char *prefix, bool reset)
2745 {
2746         char *name = lfsck_env_info(env)->lti_key;
2747         struct lfsck_sub_trace_obj *lsto;
2748         int rc;
2749         int i;
2750
2751         for (i = 0, rc = 0, lsto = &com->lc_sub_trace_objs[0];
2752              i < LFSCK_STF_COUNT && rc == 0; i++, lsto++) {
2753                 snprintf(name, NAME_MAX, "%s_%02d", prefix, i);
2754                 rc = lfsck_load_one_trace_file(env, com,
2755                                 com->lc_lfsck->li_lfsck_dir,
2756                                 &lsto->lsto_obj, ft, name, reset);
2757         }
2758
2759         return rc;
2760 }
2761
2762 /* external interfaces */
2763 int lfsck_get_speed(char *buf, struct dt_device *key)
2764 {
2765         struct lu_env           env;
2766         struct lfsck_instance  *lfsck;
2767         int                     rc;
2768         ENTRY;
2769
2770         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2771         if (rc != 0)
2772                 RETURN(rc);
2773
2774         lfsck = lfsck_instance_find(key, true, false);
2775         if (lfsck && buf) {
2776                 rc = sprintf(buf, "%u\n",
2777                              lfsck->li_bookmark_ram.lb_speed_limit);
2778                 lfsck_instance_put(&env, lfsck);
2779         } else {
2780                 rc = -ENXIO;
2781         }
2782
2783         lu_env_fini(&env);
2784
2785         RETURN(rc);
2786 }
2787 EXPORT_SYMBOL(lfsck_get_speed);
2788
2789 int lfsck_set_speed(struct dt_device *key, __u32 val)
2790 {
2791         struct lu_env           env;
2792         struct lfsck_instance  *lfsck;
2793         int                     rc;
2794         ENTRY;
2795
2796         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2797         if (rc != 0)
2798                 RETURN(rc);
2799
2800         lfsck = lfsck_instance_find(key, true, false);
2801         if (likely(lfsck != NULL)) {
2802                 mutex_lock(&lfsck->li_mutex);
2803                 if (__lfsck_set_speed(lfsck, val))
2804                         rc = lfsck_bookmark_store(&env, lfsck);
2805                 mutex_unlock(&lfsck->li_mutex);
2806                 lfsck_instance_put(&env, lfsck);
2807         } else {
2808                 rc = -ENXIO;
2809         }
2810
2811         lu_env_fini(&env);
2812
2813         RETURN(rc);
2814 }
2815 EXPORT_SYMBOL(lfsck_set_speed);
2816
2817 int lfsck_get_windows(char *buf, struct dt_device *key)
2818 {
2819         struct lu_env           env;
2820         struct lfsck_instance  *lfsck;
2821         int                     rc;
2822         ENTRY;
2823
2824         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2825         if (rc != 0)
2826                 RETURN(rc);
2827
2828         lfsck = lfsck_instance_find(key, true, false);
2829         if (likely(lfsck != NULL)) {
2830                 rc = sprintf(buf, "%u\n",
2831                              lfsck->li_bookmark_ram.lb_async_windows);
2832                 lfsck_instance_put(&env, lfsck);
2833         } else {
2834                 rc = -ENXIO;
2835         }
2836
2837         lu_env_fini(&env);
2838
2839         RETURN(rc);
2840 }
2841 EXPORT_SYMBOL(lfsck_get_windows);
2842
2843 int lfsck_set_windows(struct dt_device *key, unsigned int val)
2844 {
2845         struct lu_env           env;
2846         struct lfsck_instance  *lfsck;
2847         int                     rc;
2848         ENTRY;
2849
2850         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2851         if (rc != 0)
2852                 RETURN(rc);
2853
2854         lfsck = lfsck_instance_find(key, true, false);
2855         if (likely(lfsck != NULL)) {
2856                 if (val < 1 || val > LFSCK_ASYNC_WIN_MAX) {
2857                         CWARN("%s: invalid async windows size that may "
2858                               "cause memory issues. The valid range is "
2859                               "[1 - %u].\n",
2860                               lfsck_lfsck2name(lfsck), LFSCK_ASYNC_WIN_MAX);
2861                         rc = -EINVAL;
2862                 } else if (lfsck->li_bookmark_ram.lb_async_windows != val) {
2863                         mutex_lock(&lfsck->li_mutex);
2864                         lfsck->li_bookmark_ram.lb_async_windows = val;
2865                         rc = lfsck_bookmark_store(&env, lfsck);
2866                         mutex_unlock(&lfsck->li_mutex);
2867                 }
2868                 lfsck_instance_put(&env, lfsck);
2869         } else {
2870                 rc = -ENXIO;
2871         }
2872
2873         lu_env_fini(&env);
2874
2875         RETURN(rc);
2876 }
2877 EXPORT_SYMBOL(lfsck_set_windows);
2878
2879 int lfsck_dump(struct seq_file *m, struct dt_device *key, enum lfsck_type type)
2880 {
2881         struct lu_env           env;
2882         struct lfsck_instance  *lfsck;
2883         struct lfsck_component *com;
2884         int                     rc;
2885         ENTRY;
2886
2887         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2888         if (rc != 0)
2889                 RETURN(rc);
2890
2891         lfsck = lfsck_instance_find(key, true, false);
2892         if (likely(lfsck != NULL)) {
2893                 com = lfsck_component_find(lfsck, type);
2894                 if (likely(com != NULL)) {
2895                         com->lc_ops->lfsck_dump(&env, com, m);
2896                         lfsck_component_put(&env, com);
2897                 } else {
2898                         rc = -ENOTSUPP;
2899                 }
2900
2901                 lfsck_instance_put(&env, lfsck);
2902         } else {
2903                 rc = -ENXIO;
2904         }
2905
2906         lu_env_fini(&env);
2907
2908         RETURN(rc);
2909 }
2910 EXPORT_SYMBOL(lfsck_dump);
2911
2912 static int lfsck_stop_all(const struct lu_env *env,
2913                           struct lfsck_instance *lfsck,
2914                           struct lfsck_stop *stop)
2915 {
2916         struct lfsck_thread_info          *info   = lfsck_env_info(env);
2917         struct lfsck_request              *lr     = &info->lti_lr;
2918         struct lfsck_async_interpret_args *laia   = &info->lti_laia;
2919         struct ptlrpc_request_set         *set;
2920         struct lfsck_tgt_descs            *ltds   = &lfsck->li_mdt_descs;
2921         struct lfsck_tgt_desc             *ltd;
2922         struct lfsck_bookmark             *bk     = &lfsck->li_bookmark_ram;
2923         __u32                              idx;
2924         int                                rc     = 0;
2925         int                                rc1    = 0;
2926         ENTRY;
2927
2928         LASSERT(stop->ls_flags & LPF_BROADCAST);
2929
2930         set = ptlrpc_prep_set();
2931         if (unlikely(set == NULL))
2932                 RETURN(-ENOMEM);
2933
2934         memset(lr, 0, sizeof(*lr));
2935         lr->lr_event = LE_STOP;
2936         lr->lr_index = lfsck_dev_idx(lfsck);
2937         lr->lr_status = stop->ls_status;
2938         lr->lr_version = bk->lb_version;
2939         lr->lr_active = LFSCK_TYPES_ALL;
2940         lr->lr_param = stop->ls_flags;
2941
2942         memset(laia, 0, sizeof(*laia));
2943         laia->laia_ltds = ltds;
2944         laia->laia_lr = lr;
2945         laia->laia_shared = 1;
2946
2947         down_read(&ltds->ltd_rw_sem);
2948         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
2949                 ltd = lfsck_tgt_get(ltds, idx);
2950                 LASSERT(ltd != NULL);
2951
2952                 laia->laia_ltd = ltd;
2953                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2954                                          lfsck_async_interpret, laia,
2955                                          LFSCK_NOTIFY);
2956                 if (rc != 0) {
2957                         lfsck_interpret(env, lfsck, NULL, laia, rc);
2958                         lfsck_tgt_put(ltd);
2959                         CERROR("%s: cannot notify MDT %x for LFSCK stop: "
2960                                "rc = %d\n", lfsck_lfsck2name(lfsck), idx, rc);
2961                         rc1 = rc;
2962                 }
2963         }
2964         up_read(&ltds->ltd_rw_sem);
2965
2966         rc = ptlrpc_set_wait(env, set);
2967         ptlrpc_set_destroy(set);
2968
2969         if (rc == 0)
2970                 rc = laia->laia_result;
2971
2972         if (rc == -EALREADY)
2973                 rc = 0;
2974
2975         if (rc != 0)
2976                 CERROR("%s: fail to stop LFSCK on some MDTs: rc = %d\n",
2977                        lfsck_lfsck2name(lfsck), rc);
2978
2979         RETURN(rc != 0 ? rc : rc1);
2980 }
2981
2982 static int lfsck_start_all(const struct lu_env *env,
2983                            struct lfsck_instance *lfsck,
2984                            struct lfsck_start *start)
2985 {
2986         struct lfsck_thread_info          *info   = lfsck_env_info(env);
2987         struct lfsck_request              *lr     = &info->lti_lr;
2988         struct lfsck_async_interpret_args *laia   = &info->lti_laia;
2989         struct ptlrpc_request_set         *set;
2990         struct lfsck_tgt_descs            *ltds   = &lfsck->li_mdt_descs;
2991         struct lfsck_tgt_desc             *ltd;
2992         struct lfsck_bookmark             *bk     = &lfsck->li_bookmark_ram;
2993         __u32                              idx;
2994         int                                rc     = 0;
2995         bool retry = false;
2996         ENTRY;
2997
2998         LASSERT(start->ls_flags & LPF_BROADCAST);
2999
3000         memset(lr, 0, sizeof(*lr));
3001         lr->lr_event = LE_START;
3002         lr->lr_index = lfsck_dev_idx(lfsck);
3003         lr->lr_speed = bk->lb_speed_limit;
3004         lr->lr_version = bk->lb_version;
3005         lr->lr_active = start->ls_active;
3006         lr->lr_param = start->ls_flags;
3007         lr->lr_async_windows = bk->lb_async_windows;
3008         lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN |
3009                        LSV_ASYNC_WINDOWS | LSV_CREATE_OSTOBJ |
3010                        LSV_CREATE_MDTOBJ;
3011
3012         memset(laia, 0, sizeof(*laia));
3013         laia->laia_ltds = ltds;
3014         laia->laia_lr = lr;
3015         laia->laia_shared = 1;
3016
3017 again:
3018         set = ptlrpc_prep_set();
3019         if (unlikely(!set))
3020                 RETURN(-ENOMEM);
3021
3022         down_read(&ltds->ltd_rw_sem);
3023         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
3024                 ltd = lfsck_tgt_get(ltds, idx);
3025                 LASSERT(ltd != NULL);
3026
3027                 if (retry && !ltd->ltd_retry_start) {
3028                         lfsck_tgt_put(ltd);
3029                         continue;
3030                 }
3031
3032                 laia->laia_ltd = ltd;
3033                 ltd->ltd_retry_start = 0;
3034                 ltd->ltd_layout_done = 0;
3035                 ltd->ltd_namespace_done = 0;
3036                 ltd->ltd_synced_failures = 0;
3037                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
3038                                          lfsck_async_interpret, laia,
3039                                          LFSCK_NOTIFY);
3040                 if (rc != 0) {
3041                         lfsck_interpret(env, lfsck, NULL, laia, rc);
3042                         lfsck_tgt_put(ltd);
3043                         CERROR("%s: cannot notify MDT %x for LFSCK "
3044                                "start, failout: rc = %d\n",
3045                                lfsck_lfsck2name(lfsck), idx, rc);
3046                         break;
3047                 }
3048         }
3049         up_read(&ltds->ltd_rw_sem);
3050
3051         if (rc != 0) {
3052                 ptlrpc_set_destroy(set);
3053
3054                 RETURN(rc);
3055         }
3056
3057         rc = ptlrpc_set_wait(env, set);
3058         ptlrpc_set_destroy(set);
3059
3060         if (rc == 0)
3061                 rc = laia->laia_result;
3062
3063         if (unlikely(rc == -EINPROGRESS)) {
3064                 retry = true;
3065                 set_current_state(TASK_INTERRUPTIBLE);
3066                 schedule_timeout(cfs_time_seconds(1));
3067                 set_current_state(TASK_RUNNING);
3068                 if (!signal_pending(current) &&
3069                     thread_is_running(&lfsck->li_thread))
3070                         goto again;
3071
3072                 rc = -EINTR;
3073         }
3074
3075         if (rc != 0) {
3076                 struct lfsck_stop *stop = &info->lti_stop;
3077
3078                 CERROR("%s: cannot start LFSCK on some MDTs, "
3079                        "stop all: rc = %d\n",
3080                        lfsck_lfsck2name(lfsck), rc);
3081                 if (rc != -EALREADY) {
3082                         stop->ls_status = LS_FAILED;
3083                         stop->ls_flags = LPF_ALL_TGT | LPF_BROADCAST;
3084                         lfsck_stop_all(env, lfsck, stop);
3085                 }
3086         }
3087
3088         RETURN(rc);
3089 }
3090
3091 int lfsck_start(const struct lu_env *env, struct dt_device *key,
3092                 struct lfsck_start_param *lsp)
3093 {
3094         struct lfsck_start              *start  = lsp->lsp_start;
3095         struct lfsck_instance           *lfsck;
3096         struct lfsck_bookmark           *bk;
3097         struct ptlrpc_thread            *thread;
3098         struct lfsck_component          *com;
3099         struct lfsck_thread_args        *lta;
3100         struct task_struct              *task;
3101         struct lfsck_tgt_descs          *ltds;
3102         struct lfsck_tgt_desc           *ltd;
3103         __u32                            idx;
3104         int                              rc     = 0;
3105         __u16                            valid  = 0;
3106         __u16                            flags  = 0;
3107         __u16                            type   = 1;
3108         ENTRY;
3109
3110         if (key->dd_rdonly)
3111                 RETURN(-EROFS);
3112
3113         lfsck = lfsck_instance_find(key, true, false);
3114         if (unlikely(lfsck == NULL))
3115                 RETURN(-ENXIO);
3116
3117         if (unlikely(lfsck->li_stopping))
3118                 GOTO(put, rc = -ENXIO);
3119
3120         /* System is not ready, try again later. */
3121         if (unlikely(lfsck->li_namespace == NULL ||
3122                      lfsck_dev_site(lfsck)->ss_server_fld == NULL))
3123                 GOTO(put, rc = -EINPROGRESS);
3124
3125         /* start == NULL means auto trigger paused LFSCK. */
3126         if (!start) {
3127                 if (list_empty(&lfsck->li_list_scan) ||
3128                     OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_AUTO))
3129                         GOTO(put, rc = 0);
3130         } else if (start->ls_flags & LPF_BROADCAST && !lfsck->li_master) {
3131                 CERROR("%s: only allow to specify '-A | -o' via MDS\n",
3132                        lfsck_lfsck2name(lfsck));
3133
3134                 GOTO(put, rc = -EPERM);
3135         }
3136
3137         bk = &lfsck->li_bookmark_ram;
3138         thread = &lfsck->li_thread;
3139         mutex_lock(&lfsck->li_mutex);
3140         spin_lock(&lfsck->li_lock);
3141         if (unlikely(thread_is_stopping(thread))) {
3142                 /* Someone is stopping the LFSCK. */
3143                 spin_unlock(&lfsck->li_lock);
3144                 GOTO(out, rc = -EBUSY);
3145         }
3146
3147         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
3148                 rc = -EALREADY;
3149                 if (unlikely(start == NULL)) {
3150                         spin_unlock(&lfsck->li_lock);
3151                         GOTO(out, rc);
3152                 }
3153
3154                 while (start->ls_active != 0) {
3155                         if (!(type & start->ls_active)) {
3156                                 type <<= 1;
3157                                 continue;
3158                         }
3159
3160                         com = __lfsck_component_find(lfsck, type,
3161                                                      &lfsck->li_list_scan);
3162                         if (com == NULL)
3163                                 com = __lfsck_component_find(lfsck, type,
3164                                                 &lfsck->li_list_double_scan);
3165                         if (com == NULL) {
3166                                 rc = -EOPNOTSUPP;
3167                                 break;
3168                         }
3169
3170                         if (com->lc_ops->lfsck_join != NULL) {
3171                                 rc = com->lc_ops->lfsck_join( env, com, lsp);
3172                                 if (rc != 0 && rc != -EALREADY)
3173                                         break;
3174                         }
3175                         start->ls_active &= ~type;
3176                         type <<= 1;
3177                 }
3178                 spin_unlock(&lfsck->li_lock);
3179                 GOTO(out, rc);
3180         }
3181         spin_unlock(&lfsck->li_lock);
3182
3183         lfsck->li_status = 0;
3184         lfsck->li_oit_over = 0;
3185         lfsck->li_start_unplug = 0;
3186         lfsck->li_drop_dryrun = 0;
3187         lfsck->li_new_scanned = 0;
3188
3189         /* For auto trigger. */
3190         if (start == NULL)
3191                 goto trigger;
3192
3193         start->ls_version = bk->lb_version;
3194
3195         if (start->ls_active != 0) {
3196                 struct lfsck_component *next;
3197
3198                 if (start->ls_active == LFSCK_TYPES_ALL)
3199                         start->ls_active = LFSCK_TYPES_SUPPORTED;
3200
3201                 if (start->ls_active & ~LFSCK_TYPES_SUPPORTED) {
3202                         start->ls_active &= ~LFSCK_TYPES_SUPPORTED;
3203                         GOTO(out, rc = -ENOTSUPP);
3204                 }
3205
3206                 list_for_each_entry_safe(com, next,
3207                                          &lfsck->li_list_scan, lc_link) {
3208                         if (!(com->lc_type & start->ls_active)) {
3209                                 rc = com->lc_ops->lfsck_post(env, com, 0,
3210                                                              false);
3211                                 if (rc != 0)
3212                                         GOTO(out, rc);
3213                         }
3214                 }
3215
3216                 while (start->ls_active != 0) {
3217                         if (type & start->ls_active) {
3218                                 com = __lfsck_component_find(lfsck, type,
3219                                                         &lfsck->li_list_idle);
3220                                 if (com != NULL)
3221                                         /* The component status will be updated
3222                                          * when its prep() is called later by
3223                                          * the LFSCK main engine. */
3224                                         list_move_tail(&com->lc_link,
3225                                                        &lfsck->li_list_scan);
3226                                 start->ls_active &= ~type;
3227                         }
3228                         type <<= 1;
3229                 }
3230         }
3231
3232         if (list_empty(&lfsck->li_list_scan)) {
3233                 /* The speed limit will be used to control both the LFSCK and
3234                  * low layer scrub (if applied), need to be handled firstly. */
3235                 if (start->ls_valid & LSV_SPEED_LIMIT) {
3236                         if (__lfsck_set_speed(lfsck, start->ls_speed_limit)) {
3237                                 rc = lfsck_bookmark_store(env, lfsck);
3238                                 if (rc != 0)
3239                                         GOTO(out, rc);
3240                         }
3241                 }
3242
3243                 goto trigger;
3244         }
3245
3246         if (start->ls_flags & LPF_RESET)
3247                 flags |= DOIF_RESET;
3248
3249         rc = lfsck_set_param(env, lfsck, start, !!(flags & DOIF_RESET));
3250         if (rc != 0)
3251                 GOTO(out, rc);
3252
3253         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
3254                 start->ls_active |= com->lc_type;
3255                 if (flags & DOIF_RESET) {
3256                         rc = com->lc_ops->lfsck_reset(env, com, false);
3257                         if (rc != 0)
3258                                 GOTO(out, rc);
3259                 }
3260         }
3261
3262         ltds = &lfsck->li_mdt_descs;
3263         down_read(&ltds->ltd_rw_sem);
3264         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
3265                 ltd = lfsck_ltd2tgt(ltds, idx);
3266                 LASSERT(ltd != NULL);
3267
3268                 ltd->ltd_layout_done = 0;
3269                 ltd->ltd_namespace_done = 0;
3270                 ltd->ltd_synced_failures = 0;
3271                 lfsck_reset_ltd_status(ltd, LFSCK_TYPE_NAMESPACE);
3272                 lfsck_reset_ltd_status(ltd, LFSCK_TYPE_LAYOUT);
3273                 list_del_init(&ltd->ltd_layout_phase_list);
3274                 list_del_init(&ltd->ltd_layout_list);
3275                 list_del_init(&ltd->ltd_namespace_phase_list);
3276                 list_del_init(&ltd->ltd_namespace_list);
3277         }
3278         up_read(&ltds->ltd_rw_sem);
3279
3280         ltds = &lfsck->li_ost_descs;
3281         down_read(&ltds->ltd_rw_sem);
3282         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
3283                 ltd = lfsck_ltd2tgt(ltds, idx);
3284                 LASSERT(ltd != NULL);
3285
3286                 ltd->ltd_layout_done = 0;
3287                 ltd->ltd_synced_failures = 0;
3288                 lfsck_reset_ltd_status(ltd, LFSCK_TYPE_LAYOUT);
3289                 list_del_init(&ltd->ltd_layout_phase_list);
3290                 list_del_init(&ltd->ltd_layout_list);
3291         }
3292         up_read(&ltds->ltd_rw_sem);
3293
3294 trigger:
3295         lfsck->li_args_dir = LUDA_64BITHASH | LUDA_VERIFY | LUDA_TYPE;
3296         if (bk->lb_param & LPF_DRYRUN)
3297                 lfsck->li_args_dir |= LUDA_VERIFY_DRYRUN;
3298
3299         if (start != NULL && start->ls_valid & LSV_ERROR_HANDLE) {
3300                 valid |= DOIV_ERROR_HANDLE;
3301                 if (start->ls_flags & LPF_FAILOUT)
3302                         flags |= DOIF_FAILOUT;
3303         }
3304
3305         if (start != NULL && start->ls_valid & LSV_DRYRUN) {
3306                 valid |= DOIV_DRYRUN;
3307                 if (start->ls_flags & LPF_DRYRUN)
3308                         flags |= DOIF_DRYRUN;
3309         }
3310
3311         if (!list_empty(&lfsck->li_list_scan))
3312                 flags |= DOIF_OUTUSED;
3313
3314         lfsck->li_args_oit = (flags << DT_OTABLE_IT_FLAGS_SHIFT) | valid;
3315         lta = lfsck_thread_args_init(lfsck, NULL, lsp);
3316         if (IS_ERR(lta))
3317                 GOTO(out, rc = PTR_ERR(lta));
3318
3319         __lfsck_set_speed(lfsck, bk->lb_speed_limit);
3320         spin_lock(&lfsck->li_lock);
3321         thread_set_flags(thread, SVC_STARTING);
3322         spin_unlock(&lfsck->li_lock);
3323         task = kthread_run(lfsck_master_engine, lta, "lfsck");
3324         if (IS_ERR(task)) {
3325                 rc = PTR_ERR(task);
3326                 CERROR("%s: cannot start LFSCK thread: rc = %d\n",
3327                        lfsck_lfsck2name(lfsck), rc);
3328                 lfsck_thread_args_fini(lta);
3329
3330                 GOTO(out, rc);
3331         }
3332
3333         wait_event_idle(thread->t_ctl_waitq,
3334                         thread_is_running(thread) ||
3335                         thread_is_stopped(thread));
3336         if (start == NULL || !(start->ls_flags & LPF_BROADCAST)) {
3337                 lfsck->li_start_unplug = 1;
3338                 wake_up_all(&thread->t_ctl_waitq);
3339
3340                 GOTO(out, rc = 0);
3341         }
3342
3343         /* release lfsck::li_mutex to avoid deadlock. */
3344         mutex_unlock(&lfsck->li_mutex);
3345         rc = lfsck_start_all(env, lfsck, start);
3346         if (rc != 0) {
3347                 spin_lock(&lfsck->li_lock);
3348                 if (thread_is_stopped(thread)) {
3349                         spin_unlock(&lfsck->li_lock);
3350                 } else {
3351                         lfsck->li_status = LS_FAILED;
3352                         lfsck->li_flags = 0;
3353                         thread_set_flags(thread, SVC_STOPPING);
3354                         spin_unlock(&lfsck->li_lock);
3355
3356                         lfsck->li_start_unplug = 1;
3357                         wake_up_all(&thread->t_ctl_waitq);
3358                         wait_event_idle(thread->t_ctl_waitq,
3359                                         thread_is_stopped(thread));
3360                 }
3361         } else {
3362                 lfsck->li_start_unplug = 1;
3363                 wake_up_all(&thread->t_ctl_waitq);
3364         }
3365
3366         GOTO(put, rc);
3367
3368 out:
3369         mutex_unlock(&lfsck->li_mutex);
3370
3371 put:
3372         lfsck_instance_put(env, lfsck);
3373
3374         return rc < 0 ? rc : 0;
3375 }
3376 EXPORT_SYMBOL(lfsck_start);
3377
3378 int lfsck_stop(const struct lu_env *env, struct dt_device *key,
3379                struct lfsck_stop *stop)
3380 {
3381         struct lfsck_instance   *lfsck;
3382         struct ptlrpc_thread    *thread;
3383         int                      rc     = 0;
3384         int                      rc1    = 0;
3385         ENTRY;
3386
3387         lfsck = lfsck_instance_find(key, true, false);
3388         if (unlikely(lfsck == NULL))
3389                 RETURN(-ENXIO);
3390
3391         thread = &lfsck->li_thread;
3392         if (stop && stop->ls_flags & LPF_BROADCAST && !lfsck->li_master) {
3393                 CERROR("%s: only allow to specify '-A' via MDS\n",
3394                        lfsck_lfsck2name(lfsck));
3395                 GOTO(put, rc = -EPERM);
3396         }
3397
3398         spin_lock(&lfsck->li_lock);
3399         /* The target is umounted */
3400         if (stop && stop->ls_status == LS_PAUSED)
3401                 lfsck->li_stopping = 1;
3402
3403         if (thread_is_init(thread) || thread_is_stopped(thread))
3404                 /* no error if LFSCK stopped already, or not started */
3405                 GOTO(unlock, rc = 0);
3406
3407         if (thread_is_stopping(thread))
3408                 /* Someone is stopping LFSCK. */
3409                 GOTO(unlock, rc = -EINPROGRESS);
3410
3411         if (stop) {
3412                 lfsck->li_status = stop->ls_status;
3413                 lfsck->li_flags = stop->ls_flags;
3414         } else {
3415                 lfsck->li_status = LS_STOPPED;
3416                 lfsck->li_flags = 0;
3417         }
3418
3419         thread_set_flags(thread, SVC_STOPPING);
3420
3421         LASSERT(lfsck->li_task != NULL);
3422         force_sig(SIGINT, lfsck->li_task);
3423
3424         if (lfsck->li_master) {
3425                 struct lfsck_component *com;
3426                 struct lfsck_assistant_data *lad;
3427
3428                 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
3429                         lad = com->lc_data;
3430                         spin_lock(&lad->lad_lock);
3431                         if (lad->lad_task != NULL)
3432                                 force_sig(SIGINT, lad->lad_task);
3433                         spin_unlock(&lad->lad_lock);
3434                 }
3435
3436                 list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
3437                         lad = com->lc_data;
3438                         spin_lock(&lad->lad_lock);
3439                         if (lad->lad_task != NULL)
3440                                 force_sig(SIGINT, lad->lad_task);
3441                         spin_unlock(&lad->lad_lock);
3442                 }
3443         }
3444
3445         wake_up_all(&thread->t_ctl_waitq);
3446         spin_unlock(&lfsck->li_lock);
3447         if (stop && stop->ls_flags & LPF_BROADCAST)
3448                 rc1 = lfsck_stop_all(env, lfsck, stop);
3449
3450         /* It was me set the status as 'stopping' just now, if it is not
3451          * 'stopping' now, then either stopped, or re-started by race. */
3452         wait_event_idle(thread->t_ctl_waitq,
3453                         !thread_is_stopping(thread));
3454
3455         GOTO(put, rc = 0);
3456
3457 unlock:
3458         spin_unlock(&lfsck->li_lock);
3459 put:
3460         lfsck_instance_put(env, lfsck);
3461
3462         return rc != 0 ? rc : rc1;
3463 }
3464 EXPORT_SYMBOL(lfsck_stop);
3465
3466 int lfsck_in_notify_local(const struct lu_env *env, struct dt_device *key,
3467                           struct lfsck_req_local *lrl, struct thandle *th)
3468 {
3469         struct lfsck_instance *lfsck;
3470         struct lfsck_component *com;
3471         int rc = -EOPNOTSUPP;
3472         ENTRY;
3473
3474         lfsck = lfsck_instance_find(key, true, false);
3475         if (unlikely(!lfsck))
3476                 RETURN(-ENXIO);
3477
3478         com = lfsck_component_find(lfsck, lrl->lrl_active);
3479         if (likely(com && com->lc_ops->lfsck_in_notify_local)) {
3480                 rc = com->lc_ops->lfsck_in_notify_local(env, com, lrl, th);
3481                 lfsck_component_put(env, com);
3482         }
3483
3484         lfsck_instance_put(env, lfsck);
3485
3486         RETURN(rc);
3487 }
3488 EXPORT_SYMBOL(lfsck_in_notify_local);
3489
3490 int lfsck_in_notify(const struct lu_env *env, struct dt_device *key,
3491                     struct lfsck_request *lr)
3492 {
3493         int rc = -EOPNOTSUPP;
3494         ENTRY;
3495
3496         switch (lr->lr_event) {
3497         case LE_START: {
3498                 struct lfsck_start       *start = &lfsck_env_info(env)->lti_start;
3499                 struct lfsck_start_param  lsp;
3500
3501                 memset(start, 0, sizeof(*start));
3502                 start->ls_valid = lr->lr_valid;
3503                 start->ls_speed_limit = lr->lr_speed;
3504                 start->ls_version = lr->lr_version;
3505                 start->ls_active = lr->lr_active;
3506                 start->ls_flags = lr->lr_param & ~LPF_BROADCAST;
3507                 start->ls_async_windows = lr->lr_async_windows;
3508
3509                 lsp.lsp_start = start;
3510                 lsp.lsp_index = lr->lr_index;
3511                 lsp.lsp_index_valid = 1;
3512                 rc = lfsck_start(env, key, &lsp);
3513                 break;
3514         }
3515         case LE_STOP: {
3516                 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
3517
3518                 memset(stop, 0, sizeof(*stop));
3519                 stop->ls_status = lr->lr_status;
3520                 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
3521                 rc = lfsck_stop(env, key, stop);
3522                 break;
3523         }
3524         case LE_PHASE1_DONE:
3525         case LE_PHASE2_DONE:
3526         case LE_PEER_EXIT:
3527         case LE_CONDITIONAL_DESTROY:
3528         case LE_SET_LMV_MASTER:
3529         case LE_SET_LMV_SLAVE:
3530         case LE_PAIRS_VERIFY: {
3531                 struct lfsck_instance  *lfsck;
3532                 struct lfsck_component *com;
3533
3534                 lfsck = lfsck_instance_find(key, true, false);
3535                 if (unlikely(lfsck == NULL))
3536                         RETURN(-ENXIO);
3537
3538                 com = lfsck_component_find(lfsck, lr->lr_active);
3539                 if (likely(com)) {
3540                         rc = com->lc_ops->lfsck_in_notify(env, com, lr);
3541                         lfsck_component_put(env, com);
3542                 }
3543
3544                 lfsck_instance_put(env, lfsck);
3545                 break;
3546         }
3547         default:
3548                 break;
3549         }
3550
3551         RETURN(rc);
3552 }
3553 EXPORT_SYMBOL(lfsck_in_notify);
3554
3555 int lfsck_query(const struct lu_env *env, struct dt_device *key,
3556                 struct lfsck_request *req, struct lfsck_reply *rep,
3557                 struct lfsck_query *que)
3558 {
3559         struct lfsck_instance  *lfsck;
3560         struct lfsck_component *com;
3561         int                     i;
3562         int                     rc = 0;
3563         __u16                   type;
3564         ENTRY;
3565
3566         lfsck = lfsck_instance_find(key, true, false);
3567         if (unlikely(lfsck == NULL))
3568                 RETURN(-ENXIO);
3569
3570         if (que != NULL) {
3571                 if (que->lu_types == LFSCK_TYPES_ALL)
3572                         que->lu_types =
3573                                 LFSCK_TYPES_SUPPORTED & ~LFSCK_TYPE_SCRUB;
3574
3575                 if (que->lu_types & ~LFSCK_TYPES_SUPPORTED) {
3576                         que->lu_types &= ~LFSCK_TYPES_SUPPORTED;
3577
3578                         GOTO(out, rc = -ENOTSUPP);
3579                 }
3580
3581                 for (i = 0, type = 1 << i; i < LFSCK_TYPE_BITS;
3582                      i++, type = 1 << i) {
3583                         if (!(que->lu_types & type))
3584                                 continue;
3585
3586 again:
3587                         com = lfsck_component_find(lfsck, type);
3588                         if (unlikely(com == NULL))
3589                                 GOTO(out, rc = -ENOTSUPP);
3590
3591                         memset(que->lu_mdts_count[i], 0,
3592                                sizeof(__u32) * (LS_MAX + 1));
3593                         memset(que->lu_osts_count[i], 0,
3594                                sizeof(__u32) * (LS_MAX + 1));
3595                         que->lu_repaired[i] = 0;
3596                         rc = com->lc_ops->lfsck_query(env, com, req, rep,
3597                                                       que, i);
3598                         lfsck_component_put(env, com);
3599                         if  (rc < 0)
3600                                 GOTO(out, rc);
3601                 }
3602
3603                 if (!(que->lu_flags & LPF_WAIT))
3604                         GOTO(out, rc);
3605
3606                 for (i = 0, type = 1 << i; i < LFSCK_TYPE_BITS;
3607                      i++, type = 1 << i) {
3608                         if (!(que->lu_types & type))
3609                                 continue;
3610
3611                         if (que->lu_mdts_count[i][LS_SCANNING_PHASE1] != 0 ||
3612                             que->lu_mdts_count[i][LS_SCANNING_PHASE2] != 0 ||
3613                             que->lu_osts_count[i][LS_SCANNING_PHASE1] != 0 ||
3614                             que->lu_osts_count[i][LS_SCANNING_PHASE2] != 0) {
3615                                 /* If it is required to wait, then sleep
3616                                  * 3 seconds and try to query again.
3617                                  */
3618                                 unsigned long timeout =
3619                                         msecs_to_jiffies(3000) + 1;
3620                                 while (timeout &&
3621                                        !fatal_signal_pending(current))
3622                                         timeout = schedule_timeout_killable(
3623                                                 timeout);
3624                                 if (timeout == 0)
3625                                         goto again;
3626                         }
3627                 }
3628         } else {
3629                 com = lfsck_component_find(lfsck, req->lr_active);
3630                 if (likely(com != NULL)) {
3631                         rc = com->lc_ops->lfsck_query(env, com, req, rep,
3632                                                       que, -1);
3633                         lfsck_component_put(env, com);
3634                 } else {
3635                         rc = -ENOTSUPP;
3636                 }
3637         }
3638
3639         GOTO(out, rc);
3640
3641 out:
3642         lfsck_instance_put(env, lfsck);
3643         return rc;
3644 }
3645 EXPORT_SYMBOL(lfsck_query);
3646
3647 int lfsck_register_namespace(const struct lu_env *env, struct dt_device *key,
3648                              struct ldlm_namespace *ns)
3649 {
3650         struct lfsck_instance  *lfsck;
3651         int                     rc      = -ENXIO;
3652
3653         lfsck = lfsck_instance_find(key, true, false);
3654         if (likely(lfsck != NULL)) {
3655                 lfsck->li_namespace = ns;
3656                 lfsck_instance_put(env, lfsck);
3657                 rc = 0;
3658         }
3659
3660         return rc;
3661 }
3662 EXPORT_SYMBOL(lfsck_register_namespace);
3663
3664 int lfsck_register(const struct lu_env *env, struct dt_device *key,
3665                    struct dt_device *next, struct obd_device *obd,
3666                    lfsck_out_notify notify, void *notify_data, bool master)
3667 {
3668         struct lfsck_instance   *lfsck;
3669         struct dt_object        *root  = NULL;
3670         struct dt_object        *obj   = NULL;
3671         struct lu_fid           *fid   = &lfsck_env_info(env)->lti_fid;
3672         int                      rc;
3673         ENTRY;
3674
3675         lfsck = lfsck_instance_find(key, false, false);
3676         if (unlikely(lfsck != NULL))
3677                 RETURN(-EEXIST);
3678
3679         OBD_ALLOC_PTR(lfsck);
3680         if (lfsck == NULL)
3681                 RETURN(-ENOMEM);
3682
3683         mutex_init(&lfsck->li_mutex);
3684         spin_lock_init(&lfsck->li_lock);
3685         INIT_LIST_HEAD(&lfsck->li_link);
3686         INIT_LIST_HEAD(&lfsck->li_list_scan);
3687         INIT_LIST_HEAD(&lfsck->li_list_dir);
3688         INIT_LIST_HEAD(&lfsck->li_list_double_scan);
3689         INIT_LIST_HEAD(&lfsck->li_list_idle);
3690         INIT_LIST_HEAD(&lfsck->li_list_lmv);
3691         atomic_set(&lfsck->li_ref, 1);
3692         atomic_set(&lfsck->li_double_scan_count, 0);
3693         init_waitqueue_head(&lfsck->li_thread.t_ctl_waitq);
3694         lfsck->li_out_notify = notify;
3695         lfsck->li_out_notify_data = notify_data;
3696         lfsck->li_next = next;
3697         lfsck->li_bottom = key;
3698         lfsck->li_obd = obd;
3699
3700         rc = lfsck_tgt_descs_init(&lfsck->li_ost_descs);
3701         if (rc != 0)
3702                 GOTO(out, rc);
3703
3704         rc = lfsck_tgt_descs_init(&lfsck->li_mdt_descs);
3705         if (rc != 0)
3706                 GOTO(out, rc);
3707
3708         fid->f_seq = FID_SEQ_LOCAL_NAME;
3709         fid->f_oid = 1;
3710         fid->f_ver = 0;
3711         rc = local_oid_storage_init(env, key, fid, &lfsck->li_los);
3712         if (rc != 0)
3713                 GOTO(out, rc);
3714
3715         rc = dt_root_get(env, key, fid);
3716         if (rc != 0)
3717                 GOTO(out, rc);
3718
3719         root = dt_locate(env, key, fid);
3720         if (IS_ERR(root))
3721                 GOTO(out, rc = PTR_ERR(root));
3722
3723         if (unlikely(!dt_try_as_dir(env, root)))
3724                 GOTO(out, rc = -ENOTDIR);
3725
3726         lfsck->li_local_root_fid = *fid;
3727         if (master) {
3728                 lfsck->li_master = 1;
3729                 if (lfsck_dev_idx(lfsck) == 0) {
3730                         struct lu_fid *pfid = &lfsck_env_info(env)->lti_fid2;
3731                         const struct lu_name *cname;
3732
3733                         rc = dt_lookup(env, root,
3734                                 (struct dt_rec *)(&lfsck->li_global_root_fid),
3735                                 (const struct dt_key *)"ROOT");
3736                         if (rc != 0)
3737                                 GOTO(out, rc);
3738
3739                         obj = dt_locate(env, key, &lfsck->li_global_root_fid);
3740                         if (IS_ERR(obj))
3741                                 GOTO(out, rc = PTR_ERR(obj));
3742
3743                         if (unlikely(!dt_try_as_dir(env, obj)))
3744                                 GOTO(out, rc = -ENOTDIR);
3745
3746                         rc = dt_lookup(env, obj, (struct dt_rec *)fid,
3747                                 (const struct dt_key *)dotlustre);
3748                         if (rc != 0)
3749                                 GOTO(out, rc);
3750
3751                         lfsck_object_put(env, obj);
3752                         obj = dt_locate(env, key, fid);
3753                         if (IS_ERR(obj))
3754                                 GOTO(out, rc = PTR_ERR(obj));
3755
3756                         cname = lfsck_name_get_const(env, dotlustre,
3757                                                      strlen(dotlustre));
3758                         rc = lfsck_verify_linkea(env, obj, cname,
3759                                                  &lfsck->li_global_root_fid);
3760                         if (rc != 0)
3761                                 GOTO(out, rc);
3762
3763                         if (unlikely(!dt_try_as_dir(env, obj)))
3764                                 GOTO(out, rc = -ENOTDIR);
3765
3766                         *pfid = *fid;
3767                         rc = dt_lookup(env, obj, (struct dt_rec *)fid,
3768                                        (const struct dt_key *)lostfound);
3769                         if (rc != 0)
3770                                 GOTO(out, rc);
3771
3772                         lfsck_object_put(env, obj);
3773                         obj = dt_locate(env, key, fid);
3774                         if (IS_ERR(obj))
3775                                 GOTO(out, rc = PTR_ERR(obj));
3776
3777                         cname = lfsck_name_get_const(env, lostfound,
3778                                                      strlen(lostfound));
3779                         rc = lfsck_verify_linkea(env, obj, cname, pfid);
3780                         if (rc != 0)
3781                                 GOTO(out, rc);
3782
3783                         lfsck_object_put(env, obj);
3784                         obj = NULL;
3785                 }
3786         }
3787
3788         fid->f_seq = FID_SEQ_LOCAL_FILE;
3789         fid->f_oid = OTABLE_IT_OID;
3790         fid->f_ver = 0;
3791         obj = dt_locate(env, key, fid);
3792         if (IS_ERR(obj))
3793                 GOTO(out, rc = PTR_ERR(obj));
3794
3795         rc = obj->do_ops->do_index_try(env, obj, &dt_otable_features);
3796         if (rc != 0)
3797                 GOTO(out, rc);
3798
3799         lfsck->li_obj_oit = obj;
3800         obj = local_file_find_or_create(env, lfsck->li_los, root, LFSCK_DIR,
3801                                         S_IFDIR | S_IRUGO | S_IWUSR);
3802         if (IS_ERR(obj))
3803                 GOTO(out, rc = PTR_ERR(obj));
3804
3805         lu_object_get(&obj->do_lu);
3806         lfsck->li_lfsck_dir = obj;
3807         rc = lfsck_bookmark_setup(env, lfsck);
3808         if (rc != 0)
3809                 GOTO(out, rc);
3810
3811         if (master) {
3812                 rc = lfsck_fid_init(lfsck);
3813                 if (rc < 0)
3814                         GOTO(out, rc);
3815
3816                 rc = lfsck_namespace_setup(env, lfsck);
3817                 if (rc < 0)
3818                         GOTO(out, rc);
3819         }
3820
3821         rc = lfsck_layout_setup(env, lfsck);
3822         if (rc < 0)
3823                 GOTO(out, rc);
3824
3825         /* XXX: more LFSCK components initialization to be added here. */
3826
3827         rc = lfsck_instance_add(lfsck);
3828         if (rc == 0)
3829                 rc = lfsck_add_target_from_orphan(env, lfsck);
3830 out:
3831         if (obj != NULL && !IS_ERR(obj))
3832                 lfsck_object_put(env, obj);
3833         if (root != NULL && !IS_ERR(root))
3834                 lfsck_object_put(env, root);
3835         if (rc != 0)
3836                 lfsck_instance_cleanup(env, lfsck);
3837         return rc;
3838 }
3839 EXPORT_SYMBOL(lfsck_register);
3840
3841 void lfsck_degister(const struct lu_env *env, struct dt_device *key)
3842 {
3843         struct lfsck_instance *lfsck;
3844
3845         lfsck = lfsck_instance_find(key, false, true);
3846         if (lfsck != NULL)
3847                 lfsck_instance_put(env, lfsck);
3848 }
3849 EXPORT_SYMBOL(lfsck_degister);
3850
3851 int lfsck_add_target(const struct lu_env *env, struct dt_device *key,
3852                      struct dt_device *tgt, struct obd_export *exp,
3853                      __u32 index, bool for_ost)
3854 {
3855         struct lfsck_instance   *lfsck;
3856         struct lfsck_tgt_desc   *ltd;
3857         int                      rc;
3858         ENTRY;
3859
3860         OBD_ALLOC_PTR(ltd);
3861         if (ltd == NULL)
3862                 RETURN(-ENOMEM);
3863
3864         ltd->ltd_tgt = tgt;
3865         ltd->ltd_key = key;
3866         ltd->ltd_exp = exp;
3867         INIT_LIST_HEAD(&ltd->ltd_orphan_list);
3868         INIT_LIST_HEAD(&ltd->ltd_layout_list);
3869         INIT_LIST_HEAD(&ltd->ltd_layout_phase_list);
3870         INIT_LIST_HEAD(&ltd->ltd_namespace_list);
3871         INIT_LIST_HEAD(&ltd->ltd_namespace_phase_list);
3872         atomic_set(&ltd->ltd_ref, 1);
3873         ltd->ltd_index = index;
3874
3875         spin_lock(&lfsck_instance_lock);
3876         lfsck = __lfsck_instance_find(key, true, false);
3877         if (lfsck == NULL) {
3878                 if (for_ost)
3879                         list_add_tail(&ltd->ltd_orphan_list,
3880                                       &lfsck_ost_orphan_list);
3881                 else
3882                         list_add_tail(&ltd->ltd_orphan_list,
3883                                       &lfsck_mdt_orphan_list);
3884                 spin_unlock(&lfsck_instance_lock);
3885
3886                 RETURN(0);
3887         }
3888         spin_unlock(&lfsck_instance_lock);
3889
3890         rc = __lfsck_add_target(env, lfsck, ltd, for_ost, false);
3891         if (rc != 0)
3892                 lfsck_tgt_put(ltd);
3893
3894         lfsck_instance_put(env, lfsck);
3895
3896         RETURN(rc);
3897 }
3898 EXPORT_SYMBOL(lfsck_add_target);
3899
3900 void lfsck_del_target(const struct lu_env *env, struct dt_device *key,
3901                       struct dt_device *tgt, __u32 index, bool for_ost)
3902 {
3903         struct lfsck_instance   *lfsck;
3904         struct lfsck_tgt_descs  *ltds;
3905         struct lfsck_tgt_desc   *ltd;
3906         struct list_head        *head;
3907
3908         if (for_ost)
3909                 head = &lfsck_ost_orphan_list;
3910         else
3911                 head = &lfsck_mdt_orphan_list;
3912
3913         spin_lock(&lfsck_instance_lock);
3914         list_for_each_entry(ltd, head, ltd_orphan_list) {
3915                 if (ltd->ltd_tgt == tgt) {
3916                         list_del_init(&ltd->ltd_orphan_list);
3917                         spin_unlock(&lfsck_instance_lock);
3918                         lfsck_tgt_put(ltd);
3919
3920                         return;
3921                 }
3922         }
3923
3924         ltd = NULL;
3925         lfsck = __lfsck_instance_find(key, true, false);
3926         spin_unlock(&lfsck_instance_lock);
3927         if (unlikely(lfsck == NULL))
3928                 return;
3929
3930         if (for_ost)
3931                 ltds = &lfsck->li_ost_descs;
3932         else
3933                 ltds = &lfsck->li_mdt_descs;
3934
3935         down_write(&ltds->ltd_rw_sem);
3936         LASSERT(ltds->ltd_tgts_bitmap != NULL);
3937
3938         if (unlikely(index >= ltds->ltd_tgts_bitmap->size))
3939                 goto unlock;
3940
3941         ltd = lfsck_ltd2tgt(ltds, index);
3942         if (unlikely(ltd == NULL))
3943                 goto unlock;
3944
3945         LASSERT(ltds->ltd_tgtnr > 0);
3946
3947         ltds->ltd_tgtnr--;
3948         cfs_bitmap_clear(ltds->ltd_tgts_bitmap, index);
3949         lfsck_assign_tgt(ltds, NULL, index);
3950
3951 unlock:
3952         if (ltd == NULL) {
3953                 if (for_ost)
3954                         head = &lfsck->li_ost_descs.ltd_orphan;
3955                 else
3956                         head = &lfsck->li_mdt_descs.ltd_orphan;
3957
3958                 list_for_each_entry(ltd, head, ltd_orphan_list) {
3959                         if (ltd->ltd_tgt == tgt) {
3960                                 list_del_init(&ltd->ltd_orphan_list);
3961                                 break;
3962                         }
3963                 }
3964         }
3965
3966         up_write(&ltds->ltd_rw_sem);
3967         if (ltd != NULL) {
3968                 spin_lock(&ltds->ltd_lock);
3969                 ltd->ltd_dead = 1;
3970                 spin_unlock(&ltds->ltd_lock);
3971                 lfsck_stop_notify(env, lfsck, ltds, ltd, LFSCK_TYPE_NAMESPACE);
3972                 lfsck_stop_notify(env, lfsck, ltds, ltd, LFSCK_TYPE_LAYOUT);
3973                 lfsck_tgt_put(ltd);
3974         }
3975
3976         lfsck_instance_put(env, lfsck);
3977 }
3978 EXPORT_SYMBOL(lfsck_del_target);
3979
3980 static int __init lfsck_init(void)
3981 {
3982         int rc;
3983
3984         INIT_LIST_HEAD(&lfsck_instance_list);
3985         INIT_LIST_HEAD(&lfsck_ost_orphan_list);
3986         INIT_LIST_HEAD(&lfsck_mdt_orphan_list);
3987         lfsck_key_init_generic(&lfsck_thread_key, NULL);
3988         rc = lu_context_key_register(&lfsck_thread_key);
3989         if (!rc) {
3990                 tgt_register_lfsck_in_notify_local(lfsck_in_notify_local);
3991                 tgt_register_lfsck_in_notify(lfsck_in_notify);
3992                 tgt_register_lfsck_query(lfsck_query);
3993         }
3994
3995         return rc;
3996 }
3997
3998 static void __exit lfsck_exit(void)
3999 {
4000         struct lfsck_tgt_desc *ltd;
4001         struct lfsck_tgt_desc *next;
4002
4003         LASSERT(list_empty(&lfsck_instance_list));
4004
4005         list_for_each_entry_safe(ltd, next, &lfsck_ost_orphan_list,
4006                                  ltd_orphan_list) {
4007                 list_del_init(&ltd->ltd_orphan_list);
4008                 lfsck_tgt_put(ltd);
4009         }
4010
4011         list_for_each_entry_safe(ltd, next, &lfsck_mdt_orphan_list,
4012                                  ltd_orphan_list) {
4013                 list_del_init(&ltd->ltd_orphan_list);
4014                 lfsck_tgt_put(ltd);
4015         }
4016
4017         lu_context_key_degister(&lfsck_thread_key);
4018 }
4019
4020 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
4021 MODULE_DESCRIPTION("Lustre File System Checker");
4022 MODULE_VERSION(LUSTRE_VERSION_STRING);
4023 MODULE_LICENSE("GPL");
4024
4025 module_init(lfsck_init);
4026 module_exit(lfsck_exit);