lustre/osd-zfs/osd_scrub.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) 2017, Intel Corporation.
  24  */
  25 /*
  26  * lustre/osd-zfs/osd_scrub.c
  27  *
  28  * Top-level entry points into osd module
  29  *
  30  * The OI scrub is used for rebuilding Object Index files when restores MDT from
  31  * file-level backup.
  32  *
  33  * The otable based iterator scans ZFS objects to feed up layer LFSCK.
  34  *
  35  * Author: Fan Yong <fan.yong@intel.com>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_LFSCK
  39
  40 #include <linux/kthread.h>
  41 #include <uapi/linux/lustre/lustre_idl.h>
  42 #include <lustre_disk.h>
  43 #include <dt_object.h>
  44 #include <linux/xattr.h>
  45 #include <lustre_scrub.h>
  46 #include <obd_class.h>
  47 #include <lustre_nodemap.h>
  48 #include <sys/dsl_dataset.h>
  49 #include <sys/zap_impl.h>
  50 #include <sys/zap.h>
  51 #include <sys/zap_leaf.h>
  52
  53 #include "osd_internal.h"
  54
  55 #define OSD_OTABLE_MAX_HASH             ((1ULL << 48) - 1)
  56 #define OTABLE_PREFETCH                 256
  57
  58 #define DTO_INDEX_INSERT                1
  59 #define DTO_INDEX_DELETE                2
  60 #define DTO_INDEX_UPDATE                3
  61
  62 static inline bool osd_scrub_has_window(struct osd_otable_it *it)
  63 {
  64         return it->ooi_prefetched < OTABLE_PREFETCH;
  65 }
  66
  67 /**
  68  * update/insert/delete the specified OI mapping (@fid @id) according to the ops
  69  *
  70  * \retval   1, changed nothing
  71  * \retval   0, changed successfully
  72  * \retval -ve, on error
  73  */
  74 static int osd_scrub_refresh_mapping(const struct lu_env *env,
  75                                      struct osd_device *dev,
  76                                      const struct lu_fid *fid,
  77                                      uint64_t oid, int ops,
  78                                      bool force, const char *name)
  79 {
  80         struct osd_thread_info *info = osd_oti_get(env);
  81         struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
  82         char *buf = info->oti_str;
  83         dmu_tx_t *tx = NULL;
  84         dnode_t *dn = NULL;
  85         uint64_t zapid;
  86         int rc;
  87         ENTRY;
  88
  89         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
  90                 GOTO(log, rc = 0);
  91
  92         tx = dmu_tx_create(dev->od_os);
  93         if (!tx)
  94                 GOTO(log, rc = -ENOMEM);
  95
  96         zapid = osd_get_name_n_idx(env, dev, fid, buf,
  97                                    sizeof(info->oti_str), &dn);
  98         osd_tx_hold_zap(tx, zapid, dn,
  99                         ops == DTO_INDEX_INSERT ? TRUE : FALSE, NULL);
 100         rc = -dmu_tx_assign(tx, TXG_WAIT);
 101         if (rc) {
 102                 dmu_tx_abort(tx);
 103                 GOTO(log, rc);
 104         }
 105
 106         switch (ops) {
 107         case DTO_INDEX_UPDATE:
 108                 zde->zde_pad = 0;
 109                 zde->zde_dnode = oid;
 110                 zde->zde_type = 0; /* The type in OI mapping is useless. */
 111                 rc = -zap_update(dev->od_os, zapid, buf, 8, sizeof(*zde) / 8,
 112                                  zde, tx);
 113                 if (unlikely(rc == -ENOENT)) {
 114                         /* Some unlink thread may removed the OI mapping. */
 115                         rc = 1;
 116                 }
 117                 break;
 118         case DTO_INDEX_INSERT:
 119                 zde->zde_pad = 0;
 120                 zde->zde_dnode = oid;
 121                 zde->zde_type = 0; /* The type in OI mapping is useless. */
 122                 rc = osd_zap_add(dev, zapid, dn, buf, 8, sizeof(*zde) / 8,
 123                                  zde, tx);
 124                 if (unlikely(rc == -EEXIST))
 125                         rc = 1;
 126                 break;
 127         case DTO_INDEX_DELETE:
 128                 rc = osd_zap_remove(dev, zapid, dn, buf, tx);
 129                 if (rc == -ENOENT) {
 130                         /* It is normal that the unlink thread has removed the
 131                          * OI mapping already. */
 132                         rc = 1;
 133                 }
 134                 break;
 135         default:
 136                 LASSERTF(0, "Unexpected ops %d\n", ops);
 137                 rc = -EINVAL;
 138                 break;
 139         }
 140
 141         dmu_tx_commit(tx);
 142         GOTO(log, rc);
 143
 144 log:
 145         CDEBUG(D_LFSCK, "%s: refresh OI map for scrub, op %d, force %s, "
 146                DFID" => %llu (%s): rc = %d\n", osd_name(dev), ops,
 147                force ? "yes" : "no", PFID(fid), oid, name ? name : "null", rc);
 148
 149         return rc;
 150 }
 151
 152 static int
 153 osd_scrub_check_update(const struct lu_env *env, struct osd_device *dev,
 154                        const struct lu_fid *fid, uint64_t oid, int val)
 155 {
 156         struct lustre_scrub *scrub = &dev->od_scrub;
 157         struct scrub_file *sf = &scrub->os_file;
 158         struct osd_inconsistent_item *oii = NULL;
 159         nvlist_t *nvbuf = NULL;
 160         dnode_t *dn = NULL;
 161         uint64_t oid2;
 162         int ops = DTO_INDEX_UPDATE;
 163         int rc;
 164         ENTRY;
 165
 166         down_write(&scrub->os_rwsem);
 167         scrub->os_new_checked++;
 168         if (val < 0)
 169                 GOTO(out, rc = val);
 170
 171         if (scrub->os_in_prior)
 172                 oii = list_entry(scrub->os_inconsistent_items.next,
 173                                  struct osd_inconsistent_item, oii_list);
 174
 175         if (oid < sf->sf_pos_latest_start && !oii)
 176                 GOTO(out, rc = 0);
 177
 178         if (oii && oii->oii_insert) {
 179                 ops = DTO_INDEX_INSERT;
 180                 goto zget;
 181         }
 182
 183         rc = osd_fid_lookup(env, dev, fid, &oid2);
 184         if (rc) {
 185                 if (rc != -ENOENT)
 186                         GOTO(out, rc);
 187
 188                 ops = DTO_INDEX_INSERT;
 189
 190 zget:
 191                 rc = __osd_obj2dnode(dev->od_os, oid, &dn);
 192                 if (rc) {
 193                         /* Someone removed the object by race. */
 194                         if (rc == -ENOENT || rc == -EEXIST)
 195                                 rc = 0;
 196                         GOTO(out, rc);
 197                 }
 198
 199                 scrub->os_full_speed = 1;
 200                 sf->sf_flags |= SF_INCONSISTENT;
 201         } else if (oid == oid2) {
 202                 GOTO(out, rc = 0);
 203         } else {
 204                 struct lustre_mdt_attrs *lma = NULL;
 205                 int size;
 206
 207                 rc = __osd_xattr_load_by_oid(dev, oid2, &nvbuf);
 208                 if (rc == -ENOENT || rc == -EEXIST || rc == -ENODATA)
 209                         goto update;
 210                 if (rc)
 211                         GOTO(out, rc);
 212
 213                 rc = -nvlist_lookup_byte_array(nvbuf, XATTR_NAME_LMA,
 214                                                (uchar_t **)&lma, &size);
 215                 if (rc == -ENOENT || rc == -EEXIST || rc == -ENODATA)
 216                         goto update;
 217                 if (rc)
 218                         GOTO(out, rc);
 219
 220                 lustre_lma_swab(lma);
 221                 if (unlikely(lu_fid_eq(&lma->lma_self_fid, fid))) {
 222                         CDEBUG(D_LFSCK, "%s: the FID "DFID" is used by "
 223                                "two objects: %llu and %llu (in OI)\n",
 224                                osd_name(dev), PFID(fid), oid, oid2);
 225
 226                         GOTO(out, rc = -EEXIST);
 227                 }
 228
 229 update:
 230                 scrub->os_full_speed = 1;
 231                 sf->sf_flags |= SF_INCONSISTENT;
 232         }
 233
 234         rc = osd_scrub_refresh_mapping(env, dev, fid, oid, ops, false, NULL);
 235         if (!rc) {
 236                 if (scrub->os_in_prior)
 237                         sf->sf_items_updated_prior++;
 238                 else
 239                         sf->sf_items_updated++;
 240         }
 241
 242         GOTO(out, rc);
 243
 244 out:
 245         if (nvbuf)
 246                 nvlist_free(nvbuf);
 247
 248         if (rc < 0) {
 249                 sf->sf_items_failed++;
 250                 if (sf->sf_pos_first_inconsistent == 0 ||
 251                     sf->sf_pos_first_inconsistent > oid)
 252                         sf->sf_pos_first_inconsistent = oid;
 253         } else {
 254                 rc = 0;
 255         }
 256
 257         /* There may be conflict unlink during the OI scrub,
 258          * if happend, then remove the new added OI mapping. */
 259         if (ops == DTO_INDEX_INSERT && dn && dn->dn_free_txg)
 260                 osd_scrub_refresh_mapping(env, dev, fid, oid,
 261                                           DTO_INDEX_DELETE, false, NULL);
 262         up_write(&scrub->os_rwsem);
 263
 264         if (dn)
 265                 osd_dnode_rele(dn);
 266
 267         if (oii) {
 268                 spin_lock(&scrub->os_lock);
 269                 if (likely(!list_empty(&oii->oii_list)))
 270                         list_del(&oii->oii_list);
 271                 spin_unlock(&scrub->os_lock);
 272                 OBD_FREE_PTR(oii);
 273         }
 274
 275         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 276 }
 277
 278 static int osd_scrub_prep(const struct lu_env *env, struct osd_device *dev)
 279 {
 280         struct lustre_scrub *scrub = &dev->od_scrub;
 281         struct ptlrpc_thread *thread = &scrub->os_thread;
 282         struct scrub_file *sf = &scrub->os_file;
 283         __u32 flags = scrub->os_start_flags;
 284         int rc;
 285         bool drop_dryrun = false;
 286         ENTRY;
 287
 288         CDEBUG(D_LFSCK, "%s: OI scrub prep, flags = 0x%x\n",
 289                scrub->os_name, flags);
 290
 291         down_write(&scrub->os_rwsem);
 292         if (flags & SS_SET_FAILOUT)
 293                 sf->sf_param |= SP_FAILOUT;
 294         else if (flags & SS_CLEAR_FAILOUT)
 295                 sf->sf_param &= ~SP_FAILOUT;
 296
 297         if (flags & SS_SET_DRYRUN) {
 298                 sf->sf_param |= SP_DRYRUN;
 299         } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
 300                 sf->sf_param &= ~SP_DRYRUN;
 301                 drop_dryrun = true;
 302         }
 303
 304         if (flags & SS_RESET)
 305                 scrub_file_reset(scrub, dev->od_uuid, 0);
 306
 307         scrub->os_partial_scan = 0;
 308         if (flags & SS_AUTO_FULL) {
 309                 scrub->os_full_speed = 1;
 310                 sf->sf_flags |= SF_AUTO;
 311         } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
 312                                    SF_UPGRADE)) {
 313                 scrub->os_full_speed = 1;
 314         } else {
 315                 scrub->os_full_speed = 0;
 316         }
 317
 318         spin_lock(&scrub->os_lock);
 319         scrub->os_in_prior = 0;
 320         scrub->os_waiting = 0;
 321         scrub->os_paused = 0;
 322         scrub->os_in_join = 0;
 323         scrub->os_full_scrub = 0;
 324         spin_unlock(&scrub->os_lock);
 325         scrub->os_new_checked = 0;
 326         if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
 327                 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
 328         else if (sf->sf_pos_last_checkpoint != 0)
 329                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 330         else
 331                 sf->sf_pos_latest_start = 1;
 332
 333         scrub->os_pos_current = sf->sf_pos_latest_start;
 334         sf->sf_status = SS_SCANNING;
 335         sf->sf_time_latest_start = ktime_get_real_seconds();
 336         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 337         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
 338         rc = scrub_file_store(env, scrub);
 339         if (!rc) {
 340                 spin_lock(&scrub->os_lock);
 341                 thread_set_flags(thread, SVC_RUNNING);
 342                 spin_unlock(&scrub->os_lock);
 343                 wake_up_all(&thread->t_ctl_waitq);
 344         }
 345         up_write(&scrub->os_rwsem);
 346
 347         RETURN(rc);
 348 }
 349
 350 static int osd_scrub_post(const struct lu_env *env, struct osd_device *dev,
 351                           int result)
 352 {
 353         struct lustre_scrub *scrub = &dev->od_scrub;
 354         struct scrub_file *sf = &scrub->os_file;
 355         int rc;
 356         ENTRY;
 357
 358         CDEBUG(D_LFSCK, "%s: OI scrub post with result = %d\n",
 359                scrub->os_name, result);
 360
 361         down_write(&scrub->os_rwsem);
 362         spin_lock(&scrub->os_lock);
 363         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 364         spin_unlock(&scrub->os_lock);
 365         if (scrub->os_new_checked > 0) {
 366                 sf->sf_items_checked += scrub->os_new_checked;
 367                 scrub->os_new_checked = 0;
 368                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 369         }
 370         sf->sf_time_last_checkpoint = ktime_get_real_seconds();
 371         if (result > 0) {
 372                 sf->sf_status = SS_COMPLETED;
 373                 if (!(sf->sf_param & SP_DRYRUN)) {
 374                         memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 375                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 376                                           SF_UPGRADE | SF_AUTO);
 377                 }
 378                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 379                 sf->sf_success_count++;
 380         } else if (result == 0) {
 381                 if (scrub->os_paused)
 382                         sf->sf_status = SS_PAUSED;
 383                 else
 384                         sf->sf_status = SS_STOPPED;
 385         } else {
 386                 sf->sf_status = SS_FAILED;
 387         }
 388         sf->sf_run_time += ktime_get_seconds() -
 389                            scrub->os_time_last_checkpoint;
 390
 391         rc = scrub_file_store(env, scrub);
 392         up_write(&scrub->os_rwsem);
 393
 394         RETURN(rc < 0 ? rc : result);
 395 }
 396
 397 /* iteration engine */
 398
 399 static inline int
 400 osd_scrub_wakeup(struct lustre_scrub *scrub, struct osd_otable_it *it)
 401 {
 402         spin_lock(&scrub->os_lock);
 403         if (osd_scrub_has_window(it) ||
 404             !list_empty(&scrub->os_inconsistent_items) ||
 405             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
 406                 scrub->os_waiting = 0;
 407         else
 408                 scrub->os_waiting = 1;
 409         spin_unlock(&scrub->os_lock);
 410
 411         return !scrub->os_waiting;
 412 }
 413
 414 static int osd_scrub_next(const struct lu_env *env, struct osd_device *dev,
 415                           struct lu_fid *fid, uint64_t *oid)
 416 {
 417         struct l_wait_info lwi = { 0 };
 418         struct lustre_scrub *scrub = &dev->od_scrub;
 419         struct ptlrpc_thread *thread = &scrub->os_thread;
 420         struct osd_otable_it *it = dev->od_otable_it;
 421         struct lustre_mdt_attrs *lma = NULL;
 422         nvlist_t *nvbuf = NULL;
 423         int size = 0;
 424         int rc = 0;
 425         ENTRY;
 426
 427         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
 428                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
 429                 if (likely(lwi.lwi_timeout > 0)) {
 430                         l_wait_event(thread->t_ctl_waitq,
 431                                 !list_empty(&scrub->os_inconsistent_items) ||
 432                                 !thread_is_running(thread),
 433                                 &lwi);
 434                         if (unlikely(!thread_is_running(thread)))
 435                                 RETURN(SCRUB_NEXT_EXIT);
 436                 }
 437         }
 438
 439         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 440                 spin_lock(&scrub->os_lock);
 441                 thread_set_flags(thread, SVC_STOPPING);
 442                 spin_unlock(&scrub->os_lock);
 443                 RETURN(SCRUB_NEXT_CRASH);
 444         }
 445
 446         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
 447                 RETURN(SCRUB_NEXT_FATAL);
 448
 449 again:
 450         if (nvbuf) {
 451                 nvlist_free(nvbuf);
 452                 nvbuf = NULL;
 453                 lma = NULL;
 454         }
 455
 456         if (!list_empty(&scrub->os_inconsistent_items)) {
 457                 spin_lock(&scrub->os_lock);
 458                 if (likely(!list_empty(&scrub->os_inconsistent_items))) {
 459                         struct osd_inconsistent_item *oii;
 460
 461                         oii = list_entry(scrub->os_inconsistent_items.next,
 462                                 struct osd_inconsistent_item, oii_list);
 463                         *fid = oii->oii_cache.oic_fid;
 464                         *oid = oii->oii_cache.oic_dnode;
 465                         scrub->os_in_prior = 1;
 466                         spin_unlock(&scrub->os_lock);
 467
 468                         GOTO(out, rc = 0);
 469                 }
 470                 spin_unlock(&scrub->os_lock);
 471         }
 472
 473         if (!scrub->os_full_speed && !osd_scrub_has_window(it)) {
 474                 memset(&lwi, 0, sizeof(lwi));
 475                 l_wait_event(thread->t_ctl_waitq,
 476                              osd_scrub_wakeup(scrub, it),
 477                              &lwi);
 478         }
 479
 480         if (unlikely(!thread_is_running(thread)))
 481                 GOTO(out, rc = SCRUB_NEXT_EXIT);
 482
 483         rc = -dmu_object_next(dev->od_os, &scrub->os_pos_current, B_FALSE, 0);
 484         if (rc)
 485                 GOTO(out, rc = (rc == -ESRCH ? SCRUB_NEXT_BREAK : rc));
 486
 487         rc = __osd_xattr_load_by_oid(dev, scrub->os_pos_current, &nvbuf);
 488         if (rc == -ENOENT || rc == -EEXIST || rc == -ENODATA)
 489                 goto again;
 490
 491         if (rc)
 492                 GOTO(out, rc);
 493
 494         LASSERT(nvbuf != NULL);
 495         rc = -nvlist_lookup_byte_array(nvbuf, XATTR_NAME_LMA,
 496                                        (uchar_t **)&lma, &size);
 497         if (!rc) {
 498                 lustre_lma_swab(lma);
 499                 if (likely(!(lma->lma_compat & LMAC_NOT_IN_OI) &&
 500                            !(lma->lma_incompat & LMAI_AGENT))) {
 501                         *fid = lma->lma_self_fid;
 502                         *oid = scrub->os_pos_current;
 503
 504                         GOTO(out, rc = 0);
 505                 }
 506         }
 507
 508         if (!scrub->os_full_speed) {
 509                 spin_lock(&scrub->os_lock);
 510                 it->ooi_prefetched++;
 511                 if (it->ooi_waiting) {
 512                         it->ooi_waiting = 0;
 513                         wake_up_all(&thread->t_ctl_waitq);
 514                 }
 515                 spin_unlock(&scrub->os_lock);
 516         }
 517
 518         goto again;
 519
 520 out:
 521         if (nvbuf)
 522                 nvlist_free(nvbuf);
 523
 524         return rc;
 525 }
 526
 527 static int osd_scrub_exec(const struct lu_env *env, struct osd_device *dev,
 528                           const struct lu_fid *fid, uint64_t oid, int rc)
 529 {
 530         struct lustre_scrub *scrub = &dev->od_scrub;
 531         struct ptlrpc_thread *thread = &scrub->os_thread;
 532         struct osd_otable_it *it = dev->od_otable_it;
 533
 534         rc = osd_scrub_check_update(env, dev, fid, oid, rc);
 535         if (!scrub->os_in_prior) {
 536                 if (!scrub->os_full_speed) {
 537                         spin_lock(&scrub->os_lock);
 538                         it->ooi_prefetched++;
 539                         if (it->ooi_waiting) {
 540                                 it->ooi_waiting = 0;
 541                                 wake_up_all(&thread->t_ctl_waitq);
 542                         }
 543                         spin_unlock(&scrub->os_lock);
 544                 }
 545         } else {
 546                 scrub->os_in_prior = 0;
 547         }
 548
 549         if (rc)
 550                 return rc;
 551
 552         rc = scrub_checkpoint(env, scrub);
 553         if (rc) {
 554                 CDEBUG(D_LFSCK, "%s: fail to checkpoint, pos = %llu: "
 555                        "rc = %d\n", scrub->os_name, scrub->os_pos_current, rc);
 556                 /* Continue, as long as the scrub itself can go ahead. */
 557         }
 558
 559         return 0;
 560 }
 561
 562 static int osd_scrub_main(void *args)
 563 {
 564         struct lu_env env;
 565         struct osd_device *dev = (struct osd_device *)args;
 566         struct lustre_scrub *scrub = &dev->od_scrub;
 567         struct ptlrpc_thread *thread = &scrub->os_thread;
 568         struct lu_fid *fid;
 569         uint64_t oid;
 570         int rc = 0;
 571         ENTRY;
 572
 573         rc = lu_env_init(&env, LCT_LOCAL | LCT_DT_THREAD);
 574         if (rc) {
 575                 CDEBUG(D_LFSCK, "%s: OI scrub fail to init env: rc = %d\n",
 576                        scrub->os_name, rc);
 577                 GOTO(noenv, rc);
 578         }
 579
 580         rc = osd_scrub_prep(&env, dev);
 581         if (rc) {
 582                 CDEBUG(D_LFSCK, "%s: OI scrub fail to scrub prep: rc = %d\n",
 583                        scrub->os_name, rc);
 584                 GOTO(out, rc);
 585         }
 586
 587         if (!scrub->os_full_speed) {
 588                 struct l_wait_info lwi = { 0 };
 589                 struct osd_otable_it *it = dev->od_otable_it;
 590
 591                 l_wait_event(thread->t_ctl_waitq,
 592                              it->ooi_user_ready || !thread_is_running(thread),
 593                              &lwi);
 594                 if (unlikely(!thread_is_running(thread)))
 595                         GOTO(post, rc = 0);
 596
 597                 scrub->os_pos_current = it->ooi_pos;
 598         }
 599
 600         CDEBUG(D_LFSCK, "%s: OI scrub start, flags = 0x%x, pos = %llu\n",
 601                scrub->os_name, scrub->os_start_flags,
 602                scrub->os_pos_current);
 603
 604         fid = &osd_oti_get(&env)->oti_fid;
 605         while (!rc && thread_is_running(thread)) {
 606                 rc = osd_scrub_next(&env, dev, fid, &oid);
 607                 switch (rc) {
 608                 case SCRUB_NEXT_EXIT:
 609                         GOTO(post, rc = 0);
 610                 case SCRUB_NEXT_CRASH:
 611                         spin_lock(&scrub->os_lock);
 612                         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 613                         spin_unlock(&scrub->os_lock);
 614                         GOTO(out, rc = -EINVAL);
 615                 case SCRUB_NEXT_FATAL:
 616                         GOTO(post, rc = -EINVAL);
 617                 case SCRUB_NEXT_BREAK:
 618                         GOTO(post, rc = 1);
 619                 }
 620
 621                 rc = osd_scrub_exec(&env, dev, fid, oid, rc);
 622         }
 623
 624         GOTO(post, rc);
 625
 626 post:
 627         rc = osd_scrub_post(&env, dev, rc);
 628         CDEBUG(D_LFSCK, "%s: OI scrub: stop, pos = %llu: rc = %d\n",
 629                scrub->os_name, scrub->os_pos_current, rc);
 630
 631 out:
 632         while (!list_empty(&scrub->os_inconsistent_items)) {
 633                 struct osd_inconsistent_item *oii;
 634
 635                 oii = list_entry(scrub->os_inconsistent_items.next,
 636                                  struct osd_inconsistent_item, oii_list);
 637                 list_del_init(&oii->oii_list);
 638                 OBD_FREE_PTR(oii);
 639         }
 640
 641         lu_env_fini(&env);
 642
 643 noenv:
 644         spin_lock(&scrub->os_lock);
 645         thread_set_flags(thread, SVC_STOPPED);
 646         wake_up_all(&thread->t_ctl_waitq);
 647         spin_unlock(&scrub->os_lock);
 648         return rc;
 649 }
 650
 651 /* initial OI scrub */
 652
 653 struct osd_lf_map;
 654
 655 typedef int (*handle_dirent_t)(const struct lu_env *, struct osd_device *,
 656                                const char *, uint64_t, uint64_t,
 657                                enum osd_lf_flags, bool);
 658 static int osd_ios_varfid_hd(const struct lu_env *, struct osd_device *,
 659                              const char *, uint64_t, uint64_t,
 660                              enum osd_lf_flags, bool);
 661 static int osd_ios_uld_hd(const struct lu_env *, struct osd_device *,
 662                           const char *, uint64_t, uint64_t,
 663                           enum osd_lf_flags, bool);
 664
 665 typedef int (*scan_dir_t)(const struct lu_env *, struct osd_device *,
 666                           uint64_t, handle_dirent_t, enum osd_lf_flags);
 667 static int osd_ios_general_sd(const struct lu_env *, struct osd_device *,
 668                               uint64_t, handle_dirent_t, enum osd_lf_flags);
 669 static int osd_ios_ROOT_sd(const struct lu_env *, struct osd_device *,
 670                            uint64_t, handle_dirent_t, enum osd_lf_flags);
 671
 672 struct osd_lf_map {
 673         char                    *olm_name;
 674         struct lu_fid            olm_fid;
 675         enum osd_lf_flags        olm_flags;
 676         scan_dir_t               olm_scan_dir;
 677         handle_dirent_t          olm_handle_dirent;
 678 };
 679
 680 /* Add the new introduced local files in the list in the future. */
 681 static const struct osd_lf_map osd_lf_maps[] = {
 682         /* CONFIGS */
 683         {
 684                 .olm_name               = MOUNT_CONFIGS_DIR,
 685                 .olm_fid                = {
 686                         .f_seq  = FID_SEQ_LOCAL_FILE,
 687                         .f_oid  = MGS_CONFIGS_OID,
 688                 },
 689                 .olm_flags              = OLF_SCAN_SUBITEMS,
 690                 .olm_scan_dir           = osd_ios_general_sd,
 691                 .olm_handle_dirent      = osd_ios_varfid_hd,
 692         },
 693
 694         /* NIDTBL_VERSIONS */
 695         {
 696                 .olm_name               = MGS_NIDTBL_DIR,
 697                 .olm_flags              = OLF_SCAN_SUBITEMS,
 698                 .olm_scan_dir           = osd_ios_general_sd,
 699                 .olm_handle_dirent      = osd_ios_varfid_hd,
 700         },
 701
 702         /* PENDING */
 703         {
 704                 .olm_name               = "PENDING",
 705         },
 706
 707         /* ROOT */
 708         {
 709                 .olm_name               = "ROOT",
 710                 .olm_fid                = {
 711                         .f_seq  = FID_SEQ_ROOT,
 712                         .f_oid  = FID_OID_ROOT,
 713                 },
 714                 .olm_flags              = OLF_SCAN_SUBITEMS,
 715                 .olm_scan_dir           = osd_ios_ROOT_sd,
 716         },
 717
 718         /* fld */
 719         {
 720                 .olm_name               = "fld",
 721                 .olm_fid                = {
 722                         .f_seq  = FID_SEQ_LOCAL_FILE,
 723                         .f_oid  = FLD_INDEX_OID,
 724                 },
 725         },
 726
 727         /* changelog_catalog */
 728         {
 729                 .olm_name               = CHANGELOG_CATALOG,
 730         },
 731
 732         /* changelog_users */
 733         {
 734                 .olm_name               = CHANGELOG_USERS,
 735         },
 736
 737         /* quota_master */
 738         {
 739                 .olm_name               = QMT_DIR,
 740                 .olm_flags              = OLF_SCAN_SUBITEMS,
 741                 .olm_scan_dir           = osd_ios_general_sd,
 742                 .olm_handle_dirent      = osd_ios_varfid_hd,
 743         },
 744
 745         /* quota_slave */
 746         {
 747                 .olm_name               = QSD_DIR,
 748                 .olm_flags              = OLF_SCAN_SUBITEMS,
 749                 .olm_scan_dir           = osd_ios_general_sd,
 750                 .olm_handle_dirent      = osd_ios_varfid_hd,
 751         },
 752
 753         /* LFSCK */
 754         {
 755                 .olm_name               = LFSCK_DIR,
 756                 .olm_flags              = OLF_SCAN_SUBITEMS | OLF_NOT_BACKUP,
 757                 .olm_scan_dir           = osd_ios_general_sd,
 758                 .olm_handle_dirent      = osd_ios_varfid_hd,
 759         },
 760
 761         /* lfsck_bookmark */
 762         {
 763                 .olm_name               = LFSCK_BOOKMARK,
 764         },
 765
 766         /* lfsck_layout */
 767         {
 768                 .olm_name               = LFSCK_LAYOUT,
 769         },
 770
 771         /* lfsck_namespace */
 772         {
 773                 .olm_name               = LFSCK_NAMESPACE,
 774         },
 775
 776         /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
 777          * and f_oid = index for their log files.  See lu_update_log{_dir}_fid()
 778          * for more details. */
 779
 780         /* update_log */
 781         {
 782                 .olm_name               = "update_log",
 783                 .olm_fid                = {
 784                         .f_seq  = FID_SEQ_UPDATE_LOG,
 785                 },
 786                 .olm_flags              = OLF_IDX_IN_FID,
 787         },
 788
 789         /* update_log_dir */
 790         {
 791                 .olm_name               = "update_log_dir",
 792                 .olm_fid        = {
 793                         .f_seq  = FID_SEQ_UPDATE_LOG_DIR,
 794                 },
 795                 .olm_flags              = OLF_SCAN_SUBITEMS | OLF_IDX_IN_FID,
 796                 .olm_scan_dir           = osd_ios_general_sd,
 797                 .olm_handle_dirent      = osd_ios_uld_hd,
 798         },
 799
 800         /* hsm_actions */
 801         {
 802                 .olm_name               = HSM_ACTIONS,
 803         },
 804
 805         /* nodemap */
 806         {
 807                 .olm_name               = LUSTRE_NODEMAP_NAME,
 808         },
 809
 810         /* index_backup */
 811         {
 812                 .olm_name               = INDEX_BACKUP_DIR,
 813                 .olm_fid                = {
 814                         .f_seq  = FID_SEQ_LOCAL_FILE,
 815                         .f_oid  = INDEX_BACKUP_OID,
 816                 },
 817                 .olm_flags              = OLF_SCAN_SUBITEMS | OLF_NOT_BACKUP,
 818                 .olm_scan_dir           = osd_ios_general_sd,
 819                 .olm_handle_dirent      = osd_ios_varfid_hd,
 820         },
 821
 822         {
 823                 .olm_name               = NULL
 824         }
 825 };
 826
 827 /* Add the new introduced files under .lustre/ in the list in the future. */
 828 static const struct osd_lf_map osd_dl_maps[] = {
 829         /* .lustre/fid */
 830         {
 831                 .olm_name               = "fid",
 832                 .olm_fid                = {
 833                         .f_seq  = FID_SEQ_DOT_LUSTRE,
 834                         .f_oid  = FID_OID_DOT_LUSTRE_OBF,
 835                 },
 836         },
 837
 838         /* .lustre/lost+found */
 839         {
 840                 .olm_name               = "lost+found",
 841                 .olm_fid                = {
 842                         .f_seq  = FID_SEQ_DOT_LUSTRE,
 843                         .f_oid  = FID_OID_DOT_LUSTRE_LPF,
 844                 },
 845         },
 846
 847         {
 848                 .olm_name               = NULL
 849         }
 850 };
 851
 852 struct osd_ios_item {
 853         struct list_head        oii_list;
 854         uint64_t                oii_parent;
 855         enum osd_lf_flags       oii_flags;
 856         scan_dir_t              oii_scan_dir;
 857         handle_dirent_t         oii_handle_dirent;
 858 };
 859
 860 static int osd_ios_new_item(struct osd_device *dev, uint64_t parent,
 861                             enum osd_lf_flags flags, scan_dir_t scan_dir,
 862                             handle_dirent_t handle_dirent)
 863 {
 864         struct osd_ios_item *item;
 865
 866         OBD_ALLOC_PTR(item);
 867         if (!item) {
 868                 CWARN("%s: initial OI scrub failed to add item for %llu\n",
 869                       osd_name(dev), parent);
 870                 return -ENOMEM;
 871         }
 872
 873         INIT_LIST_HEAD(&item->oii_list);
 874         item->oii_parent = parent;
 875         item->oii_flags = flags;
 876         item->oii_scan_dir = scan_dir;
 877         item->oii_handle_dirent = handle_dirent;
 878         list_add_tail(&item->oii_list, &dev->od_ios_list);
 879
 880         return 0;
 881 }
 882
 883 static bool osd_index_need_recreate(const struct lu_env *env,
 884                                     struct osd_device *dev, uint64_t oid)
 885 {
 886         struct osd_thread_info *info = osd_oti_get(env);
 887         zap_attribute_t *za = &info->oti_za2;
 888         zap_cursor_t *zc = &info->oti_zc2;
 889         int rc;
 890         ENTRY;
 891
 892         zap_cursor_init_serialized(zc, dev->od_os, oid, 0);
 893         rc = -zap_cursor_retrieve(zc, za);
 894         zap_cursor_fini(zc);
 895         if (rc && rc != -ENOENT)
 896                 RETURN(true);
 897
 898         RETURN(false);
 899 }
 900
 901 static void osd_ios_index_register(const struct lu_env *env,
 902                                    struct osd_device *osd,
 903                                    const struct lu_fid *fid, uint64_t oid)
 904 {
 905         struct osd_thread_info *info = osd_oti_get(env);
 906         zap_attribute_t *za = &info->oti_za2;
 907         zap_cursor_t *zc = &info->oti_zc2;
 908         struct zap_leaf_entry *le;
 909         dnode_t *dn = NULL;
 910         sa_handle_t *hdl;
 911         __u64 mode = 0;
 912         __u32 keysize = 0;
 913         __u32 recsize = 0;
 914         int rc;
 915         ENTRY;
 916
 917         rc = __osd_obj2dnode(osd->od_os, oid, &dn);
 918         if (rc == -EEXIST || rc == -ENOENT)
 919                 RETURN_EXIT;
 920
 921         if (rc < 0)
 922                 GOTO(log, rc);
 923
 924         if (!osd_object_is_zap(dn))
 925                 GOTO(log, rc = 1);
 926
 927         rc = -sa_handle_get(osd->od_os, oid, NULL, SA_HDL_PRIVATE, &hdl);
 928         if (rc)
 929                 GOTO(log, rc);
 930
 931         rc = -sa_lookup(hdl, SA_ZPL_MODE(osd), &mode, sizeof(mode));
 932         sa_handle_destroy(hdl);
 933         if (rc)
 934                 GOTO(log, rc);
 935
 936         if (!S_ISREG(mode))
 937                 GOTO(log, rc = 1);
 938
 939         zap_cursor_init_serialized(zc, osd->od_os, oid, 0);
 940         rc = -zap_cursor_retrieve(zc, za);
 941         if (rc)
 942                 /* Skip empty index object */
 943                 GOTO(fini, rc = (rc == -ENOENT ? 1 : rc));
 944
 945         if (zc->zc_zap->zap_ismicro ||
 946             !(zap_f_phys(zc->zc_zap)->zap_flags & ZAP_FLAG_UINT64_KEY))
 947                 GOTO(fini, rc = 1);
 948
 949         le = ZAP_LEAF_ENTRY(zc->zc_leaf, 0);
 950         keysize = le->le_name_numints * 8;
 951         recsize = za->za_integer_length * za->za_num_integers;
 952         if (likely(keysize && recsize))
 953                 rc = osd_index_register(osd, fid, keysize, recsize);
 954
 955         GOTO(fini, rc);
 956
 957 fini:
 958         zap_cursor_fini(zc);
 959
 960 log:
 961         if (dn)
 962                 osd_dnode_rele(dn);
 963         if (rc < 0)
 964                 CWARN("%s: failed to register index "DFID" (%u/%u): rc = %d\n",
 965                       osd_name(osd), PFID(fid), keysize, recsize, rc);
 966         else if (!rc)
 967                 CDEBUG(D_LFSCK, "%s: registered index "DFID" (%u/%u)\n",
 968                        osd_name(osd), PFID(fid), keysize, recsize);
 969 }
 970
 971 static void osd_index_restore(const struct lu_env *env, struct osd_device *dev,
 972                               struct lustre_index_restore_unit *liru, void *buf,
 973                               int bufsize)
 974 {
 975         struct luz_direntry *zde = &osd_oti_get(env)->oti_zde;
 976         struct lu_fid *tgt_fid = &liru->liru_cfid;
 977         struct lu_fid bak_fid;
 978         int rc;
 979         ENTRY;
 980
 981         lustre_fid2lbx(buf, tgt_fid, bufsize);
 982         rc = -zap_lookup(dev->od_os, dev->od_index_backup_id, buf, 8,
 983                          sizeof(*zde) / 8, (void *)zde);
 984         if (rc)
 985                 GOTO(log, rc);
 986
 987         rc = osd_get_fid_by_oid(env, dev, zde->lzd_reg.zde_dnode, &bak_fid);
 988         if (rc)
 989                 GOTO(log, rc);
 990
 991         /* The OI mapping for index may be invalid, since it will be
 992          * re-created, not update the OI mapping, just cache it in RAM. */
 993         rc = osd_idc_find_and_init_with_oid(env, dev, tgt_fid,
 994                                             liru->liru_clid);
 995         if (!rc)
 996                 rc = lustre_index_restore(env, &dev->od_dt_dev,
 997                                 &liru->liru_pfid, tgt_fid, &bak_fid,
 998                                 liru->liru_name, &dev->od_index_backup_list,
 999                                 &dev->od_lock, buf, bufsize);
1000         GOTO(log, rc);
1001
1002 log:
1003         CDEBUG(D_WARNING, "%s: restore index '%s' with "DFID": rc = %d\n",
1004                osd_name(dev), liru->liru_name, PFID(tgt_fid), rc);
1005 }
1006
1007 /**
1008  * verify FID-in-LMA and OI entry for one object
1009  *
1010  * ios: Initial OI Scrub.
1011  */
1012 static int osd_ios_scan_one(const struct lu_env *env, struct osd_device *dev,
1013                             const struct lu_fid *fid, uint64_t parent,
1014                             uint64_t oid, const char *name,
1015                             enum osd_lf_flags flags)
1016 {
1017         struct lustre_scrub *scrub = &dev->od_scrub;
1018         struct scrub_file *sf = &scrub->os_file;
1019         struct lustre_mdt_attrs *lma = NULL;
1020         nvlist_t *nvbuf = NULL;
1021         struct lu_fid tfid;
1022         uint64_t oid2 = 0;
1023         __u64 flag = 0;
1024         int size = 0;
1025         int op = 0;
1026         int rc;
1027         ENTRY;
1028
1029         rc = __osd_xattr_load_by_oid(dev, oid, &nvbuf);
1030         if (unlikely(rc == -ENOENT || rc == -EEXIST))
1031                 RETURN(0);
1032
1033         if (rc && rc != -ENODATA) {
1034                 CWARN("%s: initial OI scrub failed to get lma for %llu: "
1035                       "rc = %d\n", osd_name(dev), oid, rc);
1036
1037                 RETURN(rc);
1038         }
1039
1040         if (!rc) {
1041                 LASSERT(nvbuf != NULL);
1042                 rc = -nvlist_lookup_byte_array(nvbuf, XATTR_NAME_LMA,
1043                                                (uchar_t **)&lma, &size);
1044                 if (rc || size == 0) {
1045                         LASSERT(lma == NULL);
1046                         rc = -ENODATA;
1047                 } else {
1048                         LASSERTF(lma != NULL, "corrupted LMA, size %d\n", size);
1049                         lustre_lma_swab(lma);
1050                         if (lma->lma_compat & LMAC_NOT_IN_OI) {
1051                                 nvlist_free(nvbuf);
1052                                 RETURN(0);
1053                         }
1054
1055                         if (lma->lma_compat & LMAC_IDX_BACKUP &&
1056                             osd_index_need_recreate(env, dev, oid)) {
1057                                 if (parent == dev->od_root) {
1058                                         lu_local_obj_fid(&tfid,
1059                                                          OSD_FS_ROOT_OID);
1060                                 } else {
1061                                         rc = osd_get_fid_by_oid(env, dev,
1062                                                                 parent, &tfid);
1063                                         if (rc) {
1064                                                 nvlist_free(nvbuf);
1065                                                 RETURN(rc);
1066                                         }
1067                                 }
1068
1069                                 rc = lustre_liru_new(
1070                                                 &dev->od_index_restore_list,
1071                                                 &tfid, &lma->lma_self_fid, oid,
1072                                                 name, strlen(name));
1073                                 nvlist_free(nvbuf);
1074                                 RETURN(rc);
1075                         }
1076
1077                         tfid = lma->lma_self_fid;
1078                         if (!(flags & OLF_NOT_BACKUP))
1079                                 osd_ios_index_register(env, dev, &tfid, oid);
1080                 }
1081                 nvlist_free(nvbuf);
1082         }
1083
1084         if (rc == -ENODATA) {
1085                 if (!fid) {
1086                         /* Skip the object without FID-in-LMA */
1087                         CDEBUG(D_LFSCK, "%s: %llu has no FID-in-LMA, skip it\n",
1088                                osd_name(dev), oid);
1089
1090                         RETURN(0);
1091                 }
1092
1093                 LASSERT(!fid_is_zero(fid));
1094
1095                 tfid = *fid;
1096                 if (flags & OLF_IDX_IN_FID) {
1097                         LASSERT(dev->od_index >= 0);
1098
1099                         tfid.f_oid = dev->od_index;
1100                 }
1101         }
1102
1103         rc = osd_fid_lookup(env, dev, &tfid, &oid2);
1104         if (rc) {
1105                 if (rc != -ENOENT) {
1106                         CWARN("%s: initial OI scrub failed to lookup fid for "
1107                               DFID"=>%llu: rc = %d\n",
1108                               osd_name(dev), PFID(&tfid), oid, rc);
1109
1110                         RETURN(rc);
1111                 }
1112
1113                 flag = SF_RECREATED;
1114                 op = DTO_INDEX_INSERT;
1115         } else {
1116                 if (oid == oid2)
1117                         RETURN(0);
1118
1119                 flag = SF_INCONSISTENT;
1120                 op = DTO_INDEX_UPDATE;
1121         }
1122
1123         if (!(sf->sf_flags & flag)) {
1124                 scrub_file_reset(scrub, dev->od_uuid, flag);
1125                 rc = scrub_file_store(env, scrub);
1126                 if (rc)
1127                         RETURN(rc);
1128         }
1129
1130         rc = osd_scrub_refresh_mapping(env, dev, &tfid, oid, op, true, name);
1131
1132         RETURN(rc > 0 ? 0 : rc);
1133 }
1134
1135 static int osd_ios_varfid_hd(const struct lu_env *env, struct osd_device *dev,
1136                              const char *name, uint64_t parent, uint64_t oid,
1137                              enum osd_lf_flags flags, bool is_dir)
1138 {
1139         int rc;
1140         ENTRY;
1141
1142         rc = osd_ios_scan_one(env, dev, NULL, parent, oid, name, 0);
1143         if (!rc && is_dir)
1144                 rc = osd_ios_new_item(dev, oid, flags, osd_ios_general_sd,
1145                                       osd_ios_varfid_hd);
1146
1147         RETURN(rc);
1148 }
1149
1150 static int osd_ios_uld_hd(const struct lu_env *env, struct osd_device *dev,
1151                           const char *name, uint64_t parent, uint64_t oid,
1152                           enum osd_lf_flags flags, bool is_dir)
1153 {
1154         struct lu_fid tfid;
1155         int rc;
1156         ENTRY;
1157
1158         /* skip any non-DFID format name */
1159         if (name[0] != '[')
1160                 RETURN(0);
1161
1162         /* skip the start '[' */
1163         sscanf(&name[1], SFID, RFID(&tfid));
1164         if (fid_is_sane(&tfid))
1165                 rc = osd_ios_scan_one(env, dev, &tfid, parent, oid, name, 0);
1166         else
1167                 rc = -EIO;
1168
1169         RETURN(rc);
1170 }
1171
1172 /*
1173  * General scanner for the directories execpt /ROOT during initial OI scrub.
1174  * It scans the name entries under the given directory one by one. For each
1175  * entry, verifies its OI mapping via the given @handle_dirent.
1176  */
1177 static int osd_ios_general_sd(const struct lu_env *env, struct osd_device *dev,
1178                               uint64_t parent, handle_dirent_t handle_dirent,
1179                               enum osd_lf_flags flags)
1180 {
1181         struct osd_thread_info *info = osd_oti_get(env);
1182         struct luz_direntry *zde = &info->oti_zde;
1183         zap_attribute_t *za = &info->oti_za;
1184         zap_cursor_t *zc = &info->oti_zc;
1185         int rc;
1186         ENTRY;
1187
1188         zap_cursor_init_serialized(zc, dev->od_os, parent, 0);
1189         rc = -zap_cursor_retrieve(zc, za);
1190         if (rc == -ENOENT)
1191                 zap_cursor_advance(zc);
1192         else if (rc)
1193                 GOTO(log, rc);
1194
1195         while (1) {
1196                 rc = -zap_cursor_retrieve(zc, za);
1197                 if (rc)
1198                         GOTO(log, rc = (rc == -ENOENT ? 0 : rc));
1199
1200                 /* skip the entry started with '.' */
1201                 if (likely(za->za_name[0] != '.')) {
1202                         rc = osd_zap_lookup(dev, parent, NULL, za->za_name,
1203                                         za->za_integer_length,
1204                                         sizeof(*zde) / za->za_integer_length,
1205                                         (void *)zde);
1206                         if (rc) {
1207                                 CWARN("%s: initial OI scrub failed to lookup "
1208                                       "%s under %llu: rc = %d\n",
1209                                       osd_name(dev), za->za_name, parent, rc);
1210                                 continue;
1211                         }
1212
1213                         rc = handle_dirent(env, dev, za->za_name, parent,
1214                                         zde->lzd_reg.zde_dnode, flags,
1215                                         S_ISDIR(DTTOIF(zde->lzd_reg.zde_type)) ?
1216                                         true : false);
1217                         CDEBUG(D_LFSCK, "%s: initial OI scrub handled %s under "
1218                                "%llu: rc = %d\n",
1219                                osd_name(dev), za->za_name, parent, rc);
1220                 }
1221
1222                 zap_cursor_advance(zc);
1223         }
1224
1225 log:
1226         if (rc)
1227                 CWARN("%s: initial OI scrub failed to scan the directory %llu: "
1228                       "rc = %d\n", osd_name(dev), parent, rc);
1229         zap_cursor_fini(zc);
1230
1231         return rc;
1232 }
1233
1234 /*
1235  * The scanner for /ROOT directory. It is not all the items under /ROOT will
1236  * be scanned during the initial OI scrub, instead, only the .lustre and the
1237  * sub-items under .lustre will be handled.
1238  */
1239 static int osd_ios_ROOT_sd(const struct lu_env *env, struct osd_device *dev,
1240                            uint64_t parent, handle_dirent_t handle_dirent,
1241                            enum osd_lf_flags flags)
1242 {
1243         struct luz_direntry *zde = &osd_oti_get(env)->oti_zde;
1244         const struct osd_lf_map *map;
1245         uint64_t oid;
1246         int rc;
1247         int rc1 = 0;
1248         ENTRY;
1249
1250         rc = osd_zap_lookup(dev, parent, NULL, dot_lustre_name, 8,
1251                             sizeof(*zde) / 8, (void *)zde);
1252         if (rc == -ENOENT) {
1253                 /* The .lustre directory is lost. That is not fatal. It can
1254                  * be re-created in the subsequent MDT start processing. */
1255                 RETURN(0);
1256         }
1257
1258         if (rc) {
1259                 CWARN("%s: initial OI scrub failed to find .lustre: "
1260                       "rc = %d\n", osd_name(dev), rc);
1261
1262                 RETURN(rc);
1263         }
1264
1265         oid = zde->lzd_reg.zde_dnode;
1266         rc = osd_ios_scan_one(env, dev, &LU_DOT_LUSTRE_FID, parent, oid,
1267                               dot_lustre_name, 0);
1268         if (rc)
1269                 RETURN(rc);
1270
1271         for (map = osd_dl_maps; map->olm_name; map++) {
1272                 rc = osd_zap_lookup(dev, oid, NULL, map->olm_name, 8,
1273                                     sizeof(*zde) / 8, (void *)zde);
1274                 if (rc) {
1275                         if (rc != -ENOENT)
1276                                 CWARN("%s: initial OI scrub failed to find"
1277                                       "the entry %s under .lustre: rc = %d\n",
1278                                       osd_name(dev), map->olm_name, rc);
1279                         else if (!fid_is_zero(&map->olm_fid))
1280                                 /* Try to remove the stale OI mapping. */
1281                                 osd_scrub_refresh_mapping(env, dev,
1282                                                 &map->olm_fid, 0,
1283                                                 DTO_INDEX_DELETE, true,
1284                                                 map->olm_name);
1285                         continue;
1286                 }
1287
1288                 rc = osd_ios_scan_one(env, dev, &map->olm_fid, oid,
1289                                       zde->lzd_reg.zde_dnode, map->olm_name,
1290                                       map->olm_flags);
1291                 if (rc)
1292                         rc1 = rc;
1293         }
1294
1295         RETURN(rc1);
1296 }
1297
1298 static void osd_initial_OI_scrub(const struct lu_env *env,
1299                                  struct osd_device *dev)
1300 {
1301         struct luz_direntry *zde = &osd_oti_get(env)->oti_zde;
1302         const struct osd_lf_map *map;
1303         int rc;
1304         ENTRY;
1305
1306         for (map = osd_lf_maps; map->olm_name; map++) {
1307                 rc = osd_zap_lookup(dev, dev->od_root, NULL, map->olm_name, 8,
1308                                     sizeof(*zde) / 8, (void *)zde);
1309                 if (rc) {
1310                         if (rc != -ENOENT)
1311                                 CWARN("%s: initial OI scrub failed "
1312                                       "to find the entry %s: rc = %d\n",
1313                                       osd_name(dev), map->olm_name, rc);
1314                         else if (!fid_is_zero(&map->olm_fid))
1315                                 /* Try to remove the stale OI mapping. */
1316                                 osd_scrub_refresh_mapping(env, dev,
1317                                                 &map->olm_fid, 0,
1318                                                 DTO_INDEX_DELETE, true,
1319                                                 map->olm_name);
1320                         continue;
1321                 }
1322
1323                 rc = osd_ios_scan_one(env, dev, &map->olm_fid, dev->od_root,
1324                                       zde->lzd_reg.zde_dnode, map->olm_name,
1325                                       map->olm_flags);
1326                 if (!rc && map->olm_flags & OLF_SCAN_SUBITEMS)
1327                         osd_ios_new_item(dev, zde->lzd_reg.zde_dnode,
1328                                          map->olm_flags, map->olm_scan_dir,
1329                                          map->olm_handle_dirent);
1330         }
1331
1332         while (!list_empty(&dev->od_ios_list)) {
1333                 struct osd_ios_item *item;
1334
1335                 item = list_entry(dev->od_ios_list.next,
1336                                   struct osd_ios_item, oii_list);
1337                 list_del_init(&item->oii_list);
1338                 item->oii_scan_dir(env, dev, item->oii_parent,
1339                                    item->oii_handle_dirent, item->oii_flags);
1340                 OBD_FREE_PTR(item);
1341         }
1342
1343         if (!list_empty(&dev->od_index_restore_list)) {
1344                 char *buf;
1345
1346                 OBD_ALLOC_LARGE(buf, INDEX_BACKUP_BUFSIZE);
1347                 if (!buf)
1348                         CERROR("%s: not enough RAM for rebuild index\n",
1349                                osd_name(dev));
1350
1351                 while (!list_empty(&dev->od_index_restore_list)) {
1352                         struct lustre_index_restore_unit *liru;
1353
1354                         liru = list_entry(dev->od_index_restore_list.next,
1355                                           struct lustre_index_restore_unit,
1356                                           liru_link);
1357                         list_del(&liru->liru_link);
1358                         if (buf)
1359                                 osd_index_restore(env, dev, liru, buf,
1360                                                   INDEX_BACKUP_BUFSIZE);
1361                         OBD_FREE(liru, liru->liru_len);
1362                 }
1363
1364                 if (buf)
1365                         OBD_FREE_LARGE(buf, INDEX_BACKUP_BUFSIZE);
1366         }
1367
1368         EXIT;
1369 }
1370
1371 /* OI scrub start/stop */
1372
1373 int osd_scrub_start(const struct lu_env *env, struct osd_device *dev,
1374                     __u32 flags)
1375 {
1376         int rc;
1377         ENTRY;
1378
1379         if (dev->od_dt_dev.dd_rdonly)
1380                 RETURN(-EROFS);
1381
1382         /* od_otable_sem: prevent concurrent start/stop */
1383         down(&dev->od_otable_sem);
1384         rc = scrub_start(osd_scrub_main, &dev->od_scrub, dev, flags);
1385         up(&dev->od_otable_sem);
1386
1387         RETURN(rc == -EALREADY ? 0 : rc);
1388 }
1389
1390 static void osd_scrub_stop(struct osd_device *dev)
1391 {
1392         struct lustre_scrub *scrub = &dev->od_scrub;
1393         ENTRY;
1394
1395         /* od_otable_sem: prevent concurrent start/stop */
1396         down(&dev->od_otable_sem);
1397         scrub->os_paused = 1;
1398         scrub_stop(scrub);
1399         up(&dev->od_otable_sem);
1400
1401         EXIT;
1402 }
1403
1404 /* OI scrub setup/cleanup */
1405
1406 static const char osd_scrub_name[] = "OI_scrub";
1407
1408 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
1409 {
1410         struct osd_thread_info *info = osd_oti_get(env);
1411         struct lustre_scrub *scrub = &dev->od_scrub;
1412         struct scrub_file *sf = &scrub->os_file;
1413         struct lu_fid *fid = &info->oti_fid;
1414         struct dt_object *obj;
1415         uint64_t oid;
1416         int rc = 0;
1417         bool dirty = false;
1418         ENTRY;
1419
1420         memcpy(dev->od_uuid,
1421                &dsl_dataset_phys(dev->od_os->os_dsl_dataset)->ds_guid,
1422                sizeof(dsl_dataset_phys(dev->od_os->os_dsl_dataset)->ds_guid));
1423         memset(&dev->od_scrub, 0, sizeof(struct lustre_scrub));
1424         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
1425         init_rwsem(&scrub->os_rwsem);
1426         spin_lock_init(&scrub->os_lock);
1427         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
1428         scrub->os_name = osd_name(dev);
1429
1430         /* 'What the @fid is' is not imporatant, because the object
1431          * has no OI mapping, and only is visible inside the OSD.*/
1432         fid->f_seq = FID_SEQ_IGIF_MAX;
1433         if (dev->od_is_ost)
1434                 fid->f_oid = ((1 << 31) | dev->od_index) + 1;
1435         else
1436                 fid->f_oid = dev->od_index + 1;
1437         fid->f_ver = 0;
1438         rc = osd_obj_find_or_create(env, dev, dev->od_root,
1439                                     osd_scrub_name, &oid, fid, false);
1440         if (rc)
1441                 RETURN(rc);
1442
1443         rc = osd_idc_find_and_init_with_oid(env, dev, fid, oid);
1444         if (rc)
1445                 RETURN(rc);
1446
1447         obj = lu2dt(lu_object_find_slice(env, osd2lu_dev(dev), fid, NULL));
1448         if (IS_ERR_OR_NULL(obj))
1449                 RETURN(obj ? PTR_ERR(obj) : -ENOENT);
1450
1451         scrub->os_obj = obj;
1452         rc = scrub_file_load(env, scrub);
1453         if (rc == -ENOENT || rc == -EFAULT) {
1454                 scrub_file_init(scrub, dev->od_uuid);
1455                 dirty = true;
1456         } else if (rc < 0) {
1457                 GOTO(cleanup_obj, rc);
1458         } else {
1459                 if (memcmp(sf->sf_uuid, dev->od_uuid, 16) != 0) {
1460                         struct obd_uuid *old_uuid;
1461                         struct obd_uuid *new_uuid;
1462
1463                         OBD_ALLOC_PTR(old_uuid);
1464                         OBD_ALLOC_PTR(new_uuid);
1465                         if (!old_uuid || !new_uuid) {
1466                                 CERROR("%s: UUID has been changed, but"
1467                                        "failed to allocate RAM for report\n",
1468                                        osd_name(dev));
1469                         } else {
1470                                 class_uuid_unparse(sf->sf_uuid, old_uuid);
1471                                 class_uuid_unparse(dev->od_uuid, new_uuid);
1472                                 CDEBUG(D_LFSCK, "%s: UUID has been changed "
1473                                        "from %s to %s\n", osd_name(dev),
1474                                        old_uuid->uuid, new_uuid->uuid);
1475                         }
1476                         scrub_file_reset(scrub, dev->od_uuid, SF_INCONSISTENT);
1477                         dirty = true;
1478                         if (old_uuid)
1479                                 OBD_FREE_PTR(old_uuid);
1480                         if (new_uuid)
1481                                 OBD_FREE_PTR(new_uuid);
1482                 } else if (sf->sf_status == SS_SCANNING) {
1483                         sf->sf_status = SS_CRASHED;
1484                         dirty = true;
1485                 }
1486
1487                 if ((sf->sf_oi_count & (sf->sf_oi_count - 1)) != 0) {
1488                         LCONSOLE_WARN("%s: invalid oi count %d, set it to %d\n",
1489                                       osd_name(dev), sf->sf_oi_count,
1490                                       osd_oi_count);
1491                         sf->sf_oi_count = osd_oi_count;
1492                         dirty = true;
1493                 }
1494         }
1495
1496         if (sf->sf_pos_last_checkpoint != 0)
1497                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
1498         else
1499                 scrub->os_pos_current = 1;
1500
1501         if (dirty) {
1502                 rc = scrub_file_store(env, scrub);
1503                 if (rc)
1504                         GOTO(cleanup_obj, rc);
1505         }
1506
1507         /* Initialize OI files. */
1508         rc = osd_oi_init(env, dev);
1509         if (rc < 0)
1510                 GOTO(cleanup_obj, rc);
1511
1512         if (!dev->od_dt_dev.dd_rdonly)
1513                 osd_initial_OI_scrub(env, dev);
1514
1515         if (!dev->od_dt_dev.dd_rdonly &&
1516             dev->od_auto_scrub_interval != AS_NEVER &&
1517             ((sf->sf_status == SS_PAUSED) ||
1518              (sf->sf_status == SS_CRASHED &&
1519               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
1520                               SF_UPGRADE | SF_AUTO)) ||
1521              (sf->sf_status == SS_INIT &&
1522               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
1523                               SF_UPGRADE))))
1524                 rc = osd_scrub_start(env, dev, SS_AUTO_FULL);
1525
1526         if (rc)
1527                 GOTO(cleanup_oi, rc);
1528
1529         RETURN(0);
1530
1531 cleanup_oi:
1532         osd_oi_fini(env, dev);
1533 cleanup_obj:
1534         dt_object_put_nocache(env, scrub->os_obj);
1535         scrub->os_obj = NULL;
1536
1537         return rc;
1538 }
1539
1540 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
1541 {
1542         struct lustre_scrub *scrub = &dev->od_scrub;
1543
1544         LASSERT(!dev->od_otable_it);
1545
1546         if (scrub->os_obj) {
1547                 osd_scrub_stop(dev);
1548                 dt_object_put_nocache(env, scrub->os_obj);
1549                 scrub->os_obj = NULL;
1550         }
1551
1552         if (dev->od_oi_table)
1553                 osd_oi_fini(env, dev);
1554 }
1555
1556 /* object table based iteration APIs */
1557
1558 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
1559                                        struct dt_object *dt, __u32 attr)
1560 {
1561         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
1562         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
1563         struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1564         struct lustre_scrub *scrub = &dev->od_scrub;
1565         struct osd_otable_it *it;
1566         __u32 start = 0;
1567         int rc;
1568         ENTRY;
1569
1570         if (dev->od_dt_dev.dd_rdonly)
1571                 RETURN(ERR_PTR(-EROFS));
1572
1573         /* od_otable_sem: prevent concurrent init/fini */
1574         down(&dev->od_otable_sem);
1575         if (dev->od_otable_it)
1576                 GOTO(out, it = ERR_PTR(-EALREADY));
1577
1578         OBD_ALLOC_PTR(it);
1579         if (!it)
1580                 GOTO(out, it = ERR_PTR(-ENOMEM));
1581
1582         if (flags & DOIF_OUTUSED)
1583                 it->ooi_used_outside = 1;
1584
1585         if (flags & DOIF_RESET)
1586                 start |= SS_RESET;
1587
1588         if (valid & DOIV_ERROR_HANDLE) {
1589                 if (flags & DOIF_FAILOUT)
1590                         start |= SS_SET_FAILOUT;
1591                 else
1592                         start |= SS_CLEAR_FAILOUT;
1593         }
1594
1595         if (valid & DOIV_DRYRUN) {
1596                 if (flags & DOIF_DRYRUN)
1597                         start |= SS_SET_DRYRUN;
1598                 else
1599                         start |= SS_CLEAR_DRYRUN;
1600         }
1601
1602         /* XXX: dmu_object_next() does NOT find dnodes allocated
1603          *      in the current non-committed txg, so we force txg
1604          *      commit to find all existing dnodes ... */
1605         txg_wait_synced(dmu_objset_pool(dev->od_os), 0ULL);
1606
1607         dev->od_otable_it = it;
1608         it->ooi_dev = dev;
1609         rc = scrub_start(osd_scrub_main, scrub, dev, start & ~SS_AUTO_PARTIAL);
1610         if (rc == -EALREADY) {
1611                 it->ooi_pos = 1;
1612         } else if (rc < 0) {
1613                 dev->od_otable_it = NULL;
1614                 OBD_FREE_PTR(it);
1615                 it = ERR_PTR(rc);
1616         } else {
1617                 it->ooi_pos = scrub->os_pos_current;
1618         }
1619
1620         GOTO(out, it);
1621
1622 out:
1623         up(&dev->od_otable_sem);
1624         return (struct dt_it *)it;
1625 }
1626
1627 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
1628 {
1629         struct osd_otable_it *it = (struct osd_otable_it *)di;
1630         struct osd_device *dev = it->ooi_dev;
1631
1632         /* od_otable_sem: prevent concurrent init/fini */
1633         down(&dev->od_otable_sem);
1634         scrub_stop(&dev->od_scrub);
1635         LASSERT(dev->od_otable_it == it);
1636
1637         dev->od_otable_it = NULL;
1638         up(&dev->od_otable_sem);
1639         OBD_FREE_PTR(it);
1640 }
1641
1642 static int osd_otable_it_get(const struct lu_env *env,
1643                              struct dt_it *di, const struct dt_key *key)
1644 {
1645         return 0;
1646 }
1647
1648 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
1649 {
1650 }
1651
1652 static void osd_otable_it_preload(const struct lu_env *env,
1653                                   struct osd_otable_it *it)
1654 {
1655         struct osd_device *dev = it->ooi_dev;
1656         int rc;
1657
1658         /* can go negative on the very first access to the iterator
1659          * or if some non-Lustre objects were found */
1660         if (unlikely(it->ooi_prefetched < 0))
1661                 it->ooi_prefetched = 0;
1662
1663         if (it->ooi_prefetched >= (OTABLE_PREFETCH >> 1))
1664                 return;
1665
1666         if (it->ooi_prefetched_dnode == 0)
1667                 it->ooi_prefetched_dnode = it->ooi_pos;
1668
1669         while (it->ooi_prefetched < OTABLE_PREFETCH) {
1670                 rc = -dmu_object_next(dev->od_os, &it->ooi_prefetched_dnode,
1671                                       B_FALSE, 0);
1672                 if (rc)
1673                         break;
1674
1675                 osd_dmu_prefetch(dev->od_os, it->ooi_prefetched_dnode,
1676                                  0, 0, 0, ZIO_PRIORITY_ASYNC_READ);
1677                 it->ooi_prefetched++;
1678         }
1679 }
1680
1681 static inline int
1682 osd_otable_it_wakeup(struct lustre_scrub *scrub, struct osd_otable_it *it)
1683 {
1684         spin_lock(&scrub->os_lock);
1685         if (it->ooi_pos < scrub->os_pos_current || scrub->os_waiting ||
1686             !thread_is_running(&scrub->os_thread))
1687                 it->ooi_waiting = 0;
1688         else
1689                 it->ooi_waiting = 1;
1690         spin_unlock(&scrub->os_lock);
1691
1692         return !it->ooi_waiting;
1693 }
1694
1695 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
1696 {
1697         struct osd_otable_it *it = (struct osd_otable_it *)di;
1698         struct osd_device *dev = it->ooi_dev;
1699         struct lustre_scrub *scrub = &dev->od_scrub;
1700         struct ptlrpc_thread *thread = &scrub->os_thread;
1701         struct l_wait_info lwi = { 0 };
1702         struct lustre_mdt_attrs *lma = NULL;
1703         nvlist_t *nvbuf = NULL;
1704         int size = 0;
1705         int rc;
1706         ENTRY;
1707
1708         LASSERT(it->ooi_user_ready);
1709         fid_zero(&it->ooi_fid);
1710
1711         if (unlikely(it->ooi_all_cached))
1712                 RETURN(1);
1713
1714 again:
1715         if (nvbuf) {
1716                 nvlist_free(nvbuf);
1717                 nvbuf = NULL;
1718                 lma = NULL;
1719                 size = 0;
1720         }
1721
1722         if (it->ooi_pos >= scrub->os_pos_current)
1723                 l_wait_event(thread->t_ctl_waitq,
1724                              osd_otable_it_wakeup(scrub, it),
1725                              &lwi);
1726
1727         if (!thread_is_running(thread) && !it->ooi_used_outside)
1728                 GOTO(out, rc = 1);
1729
1730         rc = -dmu_object_next(dev->od_os, &it->ooi_pos, B_FALSE, 0);
1731         if (rc) {
1732                 if (unlikely(rc == -ESRCH)) {
1733                         it->ooi_all_cached = 1;
1734                         rc = 1;
1735                 }
1736
1737                 GOTO(out, rc);
1738         }
1739
1740         rc = __osd_xattr_load_by_oid(dev, it->ooi_pos, &nvbuf);
1741
1742         if (!scrub->os_full_speed)
1743                 spin_lock(&scrub->os_lock);
1744         it->ooi_prefetched--;
1745         if (!scrub->os_full_speed) {
1746                 if (scrub->os_waiting) {
1747                         scrub->os_waiting = 0;
1748                         wake_up_all(&thread->t_ctl_waitq);
1749                 }
1750                 spin_unlock(&scrub->os_lock);
1751         }
1752
1753         if (rc == -ENOENT || rc == -EEXIST || rc == -ENODATA)
1754                 goto again;
1755
1756         if (rc)
1757                 GOTO(out, rc);
1758
1759         LASSERT(nvbuf != NULL);
1760         rc = -nvlist_lookup_byte_array(nvbuf, XATTR_NAME_LMA,
1761                                        (uchar_t **)&lma, &size);
1762         if (rc || size == 0)
1763                 /* It is either non-Lustre object or OSD internal object,
1764                  * ignore it, go ahead */
1765                 goto again;
1766
1767         LASSERTF(lma != NULL, "corrupted LMA, size %d\n", size);
1768         lustre_lma_swab(lma);
1769         if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI ||
1770                      lma->lma_incompat & LMAI_AGENT))
1771                 goto again;
1772
1773         it->ooi_fid = lma->lma_self_fid;
1774
1775         GOTO(out, rc = 0);
1776
1777 out:
1778         if (nvbuf)
1779                 nvlist_free(nvbuf);
1780
1781         if (!rc && scrub->os_full_speed)
1782                 osd_otable_it_preload(env, it);
1783
1784         return rc;
1785 }
1786
1787 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
1788                                         const struct dt_it *di)
1789 {
1790         return NULL;
1791 }
1792
1793 static int osd_otable_it_key_size(const struct lu_env *env,
1794                                   const struct dt_it *di)
1795 {
1796         return sizeof(__u64);
1797 }
1798
1799 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
1800                              struct dt_rec *rec, __u32 attr)
1801 {
1802         struct osd_otable_it *it  = (struct osd_otable_it *)di;
1803         struct lu_fid *fid = (struct lu_fid *)rec;
1804
1805         *fid = it->ooi_fid;
1806         return 0;
1807 }
1808
1809 static __u64 osd_otable_it_store(const struct lu_env *env,
1810                                  const struct dt_it *di)
1811 {
1812         struct osd_otable_it *it = (struct osd_otable_it *)di;
1813
1814         return it->ooi_pos;
1815 }
1816
1817 /**
1818  * Set the OSD layer iteration start position as the specified hash.
1819  */
1820 static int osd_otable_it_load(const struct lu_env *env,
1821                               const struct dt_it *di, __u64 hash)
1822 {
1823         struct osd_otable_it *it = (struct osd_otable_it *)di;
1824         struct osd_device *dev = it->ooi_dev;
1825         struct lustre_scrub *scrub = &dev->od_scrub;
1826         int rc;
1827         ENTRY;
1828
1829         /* Forbid to set iteration position after iteration started. */
1830         if (it->ooi_user_ready)
1831                 RETURN(-EPERM);
1832
1833         if (hash > OSD_OTABLE_MAX_HASH)
1834                 hash = OSD_OTABLE_MAX_HASH;
1835
1836         /* The hash is the last checkpoint position,
1837          * we will start from the next one. */
1838         it->ooi_pos = hash + 1;
1839         it->ooi_prefetched = 0;
1840         it->ooi_prefetched_dnode = 0;
1841         it->ooi_user_ready = 1;
1842         if (!scrub->os_full_speed)
1843                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1844
1845         /* Unplug OSD layer iteration by the first next() call. */
1846         rc = osd_otable_it_next(env, (struct dt_it *)it);
1847
1848         RETURN(rc);
1849 }
1850
1851 static int osd_otable_it_key_rec(const struct lu_env *env,
1852                                  const struct dt_it *di, void *key_rec)
1853 {
1854         return 0;
1855 }
1856
1857 const struct dt_index_operations osd_otable_ops = {
1858         .dio_it = {
1859                 .init     = osd_otable_it_init,
1860                 .fini     = osd_otable_it_fini,
1861                 .get      = osd_otable_it_get,
1862                 .put      = osd_otable_it_put,
1863                 .next     = osd_otable_it_next,
1864                 .key      = osd_otable_it_key,
1865                 .key_size = osd_otable_it_key_size,
1866                 .rec      = osd_otable_it_rec,
1867                 .store    = osd_otable_it_store,
1868                 .load     = osd_otable_it_load,
1869                 .key_rec  = osd_otable_it_key_rec,
1870         }
1871 };
1872
1873 /* high priority inconsistent items list APIs */
1874
1875 int osd_oii_insert(const struct lu_env *env, struct osd_device *dev,
1876                    const struct lu_fid *fid, uint64_t oid, bool insert)
1877 {
1878         struct lustre_scrub *scrub = &dev->od_scrub;
1879         struct ptlrpc_thread *thread = &scrub->os_thread;
1880         struct osd_inconsistent_item *oii;
1881         bool wakeup = false;
1882         ENTRY;
1883
1884         osd_idc_find_and_init_with_oid(env, dev, fid, oid);
1885         OBD_ALLOC_PTR(oii);
1886         if (unlikely(!oii))
1887                 RETURN(-ENOMEM);
1888
1889         INIT_LIST_HEAD(&oii->oii_list);
1890         oii->oii_cache.oic_dev = dev;
1891         oii->oii_cache.oic_fid = *fid;
1892         oii->oii_cache.oic_dnode = oid;
1893         oii->oii_insert = insert;
1894
1895         spin_lock(&scrub->os_lock);
1896         if (unlikely(!thread_is_running(thread))) {
1897                 spin_unlock(&scrub->os_lock);
1898                 OBD_FREE_PTR(oii);
1899                 RETURN(-EAGAIN);
1900         }
1901
1902         if (list_empty(&scrub->os_inconsistent_items))
1903                 wakeup = true;
1904         list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
1905         spin_unlock(&scrub->os_lock);
1906
1907         if (wakeup)
1908                 wake_up_all(&thread->t_ctl_waitq);
1909
1910         RETURN(0);
1911 }
1912
1913 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
1914                    uint64_t *oid)
1915 {
1916         struct lustre_scrub *scrub = &dev->od_scrub;
1917         struct osd_inconsistent_item *oii;
1918         int ret = -ENOENT;
1919         ENTRY;
1920
1921         spin_lock(&scrub->os_lock);
1922         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
1923                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
1924                         *oid = oii->oii_cache.oic_dnode;
1925                         ret = 0;
1926                         break;
1927                 }
1928         }
1929         spin_unlock(&scrub->os_lock);
1930
1931         RETURN(ret);
1932 }