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 static inline bool osd_scrub_has_window(struct osd_otable_it *it)
  59 {
  60         return it->ooi_prefetched < OTABLE_PREFETCH;
  61 }
  62
  63 /**
  64  * update/insert/delete the specified OI mapping (@fid @id) according to the ops
  65  *
  66  * \retval   1, changed nothing
  67  * \retval   0, changed successfully
  68  * \retval -ve, on error
  69  */
  70 int osd_scrub_refresh_mapping(const struct lu_env *env,
  71                               struct osd_device *dev,
  72                               const struct lu_fid *fid,
  73                               uint64_t oid, enum dt_txn_op ops,
  74                               bool force, const char *name)
  75 {
  76         struct osd_thread_info *info = osd_oti_get(env);
  77         struct zpl_direntry *zde = &info->oti_zde.lzd_reg;
  78         char *buf = info->oti_str;
  79         dmu_tx_t *tx = NULL;
  80         dnode_t *dn = NULL;
  81         uint64_t zapid;
  82         int rc;
  83         ENTRY;
  84
  85         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
  86                 GOTO(log, rc = 0);
  87
  88         tx = dmu_tx_create(dev->od_os);
  89         if (!tx)
  90                 GOTO(log, rc = -ENOMEM);
  91
  92         zapid = osd_get_name_n_idx(env, dev, fid, buf,
  93                                    sizeof(info->oti_str), &dn);
  94         osd_tx_hold_zap(tx, zapid, dn,
  95                         ops == DTO_INDEX_INSERT ? TRUE : FALSE, NULL);
  96         rc = -dmu_tx_assign(tx, TXG_WAIT);
  97         if (rc) {
  98                 dmu_tx_abort(tx);
  99                 GOTO(log, rc);
 100         }
 101
 102         switch (ops) {
 103         case DTO_INDEX_UPDATE:
 104                 zde->zde_pad = 0;
 105                 zde->zde_dnode = oid;
 106                 zde->zde_type = 0; /* The type in OI mapping is useless. */
 107                 rc = -zap_update(dev->od_os, zapid, buf, 8, sizeof(*zde) / 8,
 108                                  zde, tx);
 109                 if (unlikely(rc == -ENOENT)) {
 110                         /* Some unlink thread may removed the OI mapping. */
 111                         rc = 1;
 112                 }
 113                 break;
 114         case DTO_INDEX_INSERT:
 115                 zde->zde_pad = 0;
 116                 zde->zde_dnode = oid;
 117                 zde->zde_type = 0; /* The type in OI mapping is useless. */
 118                 rc = osd_zap_add(dev, zapid, dn, buf, 8, sizeof(*zde) / 8,
 119                                  zde, tx);
 120                 if (unlikely(rc == -EEXIST))
 121                         rc = 1;
 122                 break;
 123         case DTO_INDEX_DELETE:
 124                 rc = osd_zap_remove(dev, zapid, dn, buf, tx);
 125                 if (rc == -ENOENT) {
 126                         /* It is normal that the unlink thread has removed the
 127                          * OI mapping already. */
 128                         rc = 1;
 129                 }
 130                 break;
 131         default:
 132                 LASSERTF(0, "Unexpected ops %d\n", ops);
 133                 rc = -EINVAL;
 134                 break;
 135         }
 136
 137         dmu_tx_commit(tx);
 138         GOTO(log, rc);
 139
 140 log:
 141         CDEBUG(D_LFSCK, "%s: refresh OI map for scrub, op %d, force %s, "
 142                DFID" => %llu (%s): rc = %d\n", osd_name(dev), ops,
 143                force ? "yes" : "no", PFID(fid), oid, name ? name : "null", rc);
 144
 145         return rc;
 146 }
 147
 148 static int
 149 osd_scrub_check_update(const struct lu_env *env, struct osd_device *dev,
 150                        const struct lu_fid *fid, uint64_t oid, int val)
 151 {
 152         struct lustre_scrub *scrub = &dev->od_scrub;
 153         struct scrub_file *sf = &scrub->os_file;
 154         struct osd_inconsistent_item *oii = NULL;
 155         nvlist_t *nvbuf = NULL;
 156         dnode_t *dn = NULL;
 157         uint64_t oid2;
 158         int ops = DTO_INDEX_UPDATE;
 159         int rc;
 160         ENTRY;
 161
 162         down_write(&scrub->os_rwsem);
 163         scrub->os_new_checked++;
 164         if (val < 0)
 165                 GOTO(out, rc = val);
 166
 167         if (scrub->os_in_prior)
 168                 oii = list_entry(scrub->os_inconsistent_items.next,
 169                                  struct osd_inconsistent_item, oii_list);
 170
 171         if (oid < sf->sf_pos_latest_start && !oii)
 172                 GOTO(out, rc = 0);
 173
 174         if (oii && oii->oii_insert) {
 175                 ops = DTO_INDEX_INSERT;
 176                 goto zget;
 177         }
 178
 179         rc = osd_fid_lookup(env, dev, fid, &oid2);
 180         if (rc) {
 181                 if (rc != -ENOENT)
 182                         GOTO(out, rc);
 183
 184                 ops = DTO_INDEX_INSERT;
 185
 186 zget:
 187                 rc = __osd_obj2dnode(dev->od_os, oid, &dn);
 188                 if (rc) {
 189                         /* Someone removed the object by race. */
 190                         if (rc == -ENOENT || rc == -EEXIST)
 191                                 rc = 0;
 192                         GOTO(out, rc);
 193                 }
 194
 195                 spin_lock(&scrub->os_lock);
 196                 scrub->os_full_speed = 1;
 197                 spin_unlock(&scrub->os_lock);
 198
 199                 sf->sf_flags |= SF_INCONSISTENT;
 200         } else if (oid == oid2) {
 201                 GOTO(out, rc = 0);
 202         } else {
 203                 struct lustre_mdt_attrs *lma = NULL;
 204                 int size;
 205
 206                 rc = __osd_xattr_load_by_oid(dev, oid2, &nvbuf);
 207                 if (rc == -ENOENT || rc == -EEXIST || rc == -ENODATA)
 208                         goto update;
 209                 if (rc)
 210                         GOTO(out, rc);
 211
 212                 rc = -nvlist_lookup_byte_array(nvbuf, XATTR_NAME_LMA,
 213                                                (uchar_t **)&lma, &size);
 214                 if (rc == -ENOENT || rc == -EEXIST || rc == -ENODATA)
 215                         goto update;
 216                 if (rc)
 217                         GOTO(out, rc);
 218
 219                 lustre_lma_swab(lma);
 220                 if (unlikely(lu_fid_eq(&lma->lma_self_fid, fid))) {
 221                         CDEBUG(D_LFSCK, "%s: the FID "DFID" is used by "
 222                                "two objects: %llu and %llu (in OI)\n",
 223                                osd_name(dev), PFID(fid), oid, oid2);
 224
 225                         GOTO(out, rc = -EEXIST);
 226                 }
 227
 228 update:
 229                 spin_lock(&scrub->os_lock);
 230                 scrub->os_full_speed = 1;
 231                 spin_unlock(&scrub->os_lock);
 232                 sf->sf_flags |= SF_INCONSISTENT;
 233         }
 234
 235         rc = osd_scrub_refresh_mapping(env, dev, fid, oid, ops, false, NULL);
 236         if (!rc) {
 237                 if (scrub->os_in_prior)
 238                         sf->sf_items_updated_prior++;
 239                 else
 240                         sf->sf_items_updated++;
 241         }
 242
 243         GOTO(out, rc);
 244
 245 out:
 246         if (dev->od_is_ost) {
 247                 sa_handle_t *hdl;
 248                 uint64_t nlink, mode;
 249
 250                 rc = -sa_handle_get(dev->od_os, oid, NULL, SA_HDL_PRIVATE,
 251                                     &hdl);
 252                 if (rc)
 253                         GOTO(cleanup, rc);
 254
 255                 rc = -sa_lookup(hdl, SA_ZPL_MODE(dev), &mode, sizeof(mode));
 256                 if (rc || !S_ISREG(mode)) {
 257                         sa_handle_destroy(hdl);
 258                         GOTO(cleanup, rc);
 259                 }
 260
 261                 rc = -sa_lookup(hdl, SA_ZPL_LINKS(dev), &nlink, sizeof(nlink));
 262                 if (rc == 0 && nlink > 1)
 263                         scrub->os_has_ml_file = 1;
 264
 265                 sa_handle_destroy(hdl);
 266         }
 267
 268 cleanup:
 269         if (nvbuf)
 270                 nvlist_free(nvbuf);
 271
 272         if (rc < 0) {
 273                 sf->sf_items_failed++;
 274                 if (sf->sf_pos_first_inconsistent == 0 ||
 275                     sf->sf_pos_first_inconsistent > oid)
 276                         sf->sf_pos_first_inconsistent = oid;
 277         } else {
 278                 rc = 0;
 279         }
 280
 281         /* There may be conflict unlink during the OI scrub,
 282          * if happend, then remove the new added OI mapping. */
 283         if (ops == DTO_INDEX_INSERT && dn && dn->dn_free_txg)
 284                 osd_scrub_refresh_mapping(env, dev, fid, oid,
 285                                           DTO_INDEX_DELETE, false, NULL);
 286         up_write(&scrub->os_rwsem);
 287
 288         if (dn)
 289                 osd_dnode_rele(dn);
 290
 291         if (oii) {
 292                 spin_lock(&scrub->os_lock);
 293                 if (likely(!list_empty(&oii->oii_list)))
 294                         list_del(&oii->oii_list);
 295                 spin_unlock(&scrub->os_lock);
 296                 OBD_FREE_PTR(oii);
 297         }
 298
 299         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 300 }
 301
 302 /* iteration engine */
 303
 304 static inline int
 305 osd_scrub_wakeup(struct lustre_scrub *scrub, struct osd_otable_it *it)
 306 {
 307         spin_lock(&scrub->os_lock);
 308         if (osd_scrub_has_window(it) ||
 309             !list_empty(&scrub->os_inconsistent_items) ||
 310             it->ooi_waiting || kthread_should_stop())
 311                 scrub->os_waiting = 0;
 312         else
 313                 scrub->os_waiting = 1;
 314         spin_unlock(&scrub->os_lock);
 315
 316         return !scrub->os_waiting;
 317 }
 318
 319 static int osd_scrub_next(const struct lu_env *env, struct osd_device *dev,
 320                           struct lu_fid *fid, uint64_t *oid)
 321 {
 322         struct lustre_scrub *scrub = &dev->od_scrub;
 323         struct osd_otable_it *it = dev->od_otable_it;
 324         struct lustre_mdt_attrs *lma = NULL;
 325         nvlist_t *nvbuf = NULL;
 326         int size = 0;
 327         int rc = 0;
 328         ENTRY;
 329
 330         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
 331                 wait_var_event_timeout(
 332                         scrub,
 333                         !list_empty(&scrub->os_inconsistent_items) ||
 334                         kthread_should_stop(),
 335                         cfs_time_seconds(cfs_fail_val));
 336
 337                 if (kthread_should_stop())
 338                         RETURN(SCRUB_NEXT_EXIT);
 339         }
 340
 341         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 342                 spin_lock(&scrub->os_lock);
 343                 scrub->os_running = 0;
 344                 spin_unlock(&scrub->os_lock);
 345                 RETURN(SCRUB_NEXT_CRASH);
 346         }
 347
 348         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
 349                 RETURN(SCRUB_NEXT_FATAL);
 350
 351 again:
 352         if (nvbuf) {
 353                 nvlist_free(nvbuf);
 354                 nvbuf = NULL;
 355                 lma = NULL;
 356         }
 357
 358         if (!list_empty(&scrub->os_inconsistent_items)) {
 359                 spin_lock(&scrub->os_lock);
 360                 if (likely(!list_empty(&scrub->os_inconsistent_items))) {
 361                         struct osd_inconsistent_item *oii;
 362
 363                         oii = list_entry(scrub->os_inconsistent_items.next,
 364                                 struct osd_inconsistent_item, oii_list);
 365                         *fid = oii->oii_cache.oic_fid;
 366                         *oid = oii->oii_cache.oic_dnode;
 367                         scrub->os_in_prior = 1;
 368                         spin_unlock(&scrub->os_lock);
 369
 370                         GOTO(out, rc = 0);
 371                 }
 372                 spin_unlock(&scrub->os_lock);
 373         }
 374
 375         if (!scrub->os_full_speed && !osd_scrub_has_window(it))
 376                 wait_var_event(scrub, osd_scrub_wakeup(scrub, it));
 377
 378         if (kthread_should_stop())
 379                 GOTO(out, rc = SCRUB_NEXT_EXIT);
 380
 381         rc = -dmu_object_next(dev->od_os, &scrub->os_pos_current, B_FALSE, 0);
 382         if (rc)
 383                 GOTO(out, rc = (rc == -ESRCH ? SCRUB_NEXT_BREAK : rc));
 384
 385         rc = __osd_xattr_load_by_oid(dev, scrub->os_pos_current, &nvbuf);
 386         if (rc == -ENOENT || rc == -EEXIST || rc == -ENODATA)
 387                 goto again;
 388
 389         if (rc)
 390                 GOTO(out, rc);
 391
 392         LASSERT(nvbuf != NULL);
 393         rc = -nvlist_lookup_byte_array(nvbuf, XATTR_NAME_LMA,
 394                                        (uchar_t **)&lma, &size);
 395         if (!rc) {
 396                 lustre_lma_swab(lma);
 397                 if (likely(!(lma->lma_compat & LMAC_NOT_IN_OI) &&
 398                            !(lma->lma_incompat & LMAI_AGENT))) {
 399                         *fid = lma->lma_self_fid;
 400                         *oid = scrub->os_pos_current;
 401
 402                         GOTO(out, rc = 0);
 403                 }
 404         }
 405
 406         if (!scrub->os_full_speed) {
 407                 spin_lock(&scrub->os_lock);
 408                 it->ooi_prefetched++;
 409                 if (it->ooi_waiting) {
 410                         it->ooi_waiting = 0;
 411                         wake_up_var(scrub);
 412                 }
 413                 spin_unlock(&scrub->os_lock);
 414         }
 415
 416         goto again;
 417
 418 out:
 419         if (nvbuf)
 420                 nvlist_free(nvbuf);
 421
 422         return rc;
 423 }
 424
 425 static int osd_scrub_exec(const struct lu_env *env, struct osd_device *dev,
 426                           const struct lu_fid *fid, uint64_t oid, int rc)
 427 {
 428         struct lustre_scrub *scrub = &dev->od_scrub;
 429         struct osd_otable_it *it = dev->od_otable_it;
 430
 431         rc = osd_scrub_check_update(env, dev, fid, oid, rc);
 432         if (!scrub->os_in_prior) {
 433                 if (!scrub->os_full_speed) {
 434                         spin_lock(&scrub->os_lock);
 435                         it->ooi_prefetched++;
 436                         if (it->ooi_waiting) {
 437                                 it->ooi_waiting = 0;
 438                                 wake_up_var(scrub);
 439                         }
 440                         spin_unlock(&scrub->os_lock);
 441                 }
 442         } else {
 443                 spin_lock(&scrub->os_lock);
 444                 scrub->os_in_prior = 0;
 445                 spin_unlock(&scrub->os_lock);
 446         }
 447
 448         if (rc)
 449                 return rc;
 450
 451         rc = scrub_checkpoint(env, scrub);
 452         if (rc) {
 453                 CDEBUG(D_LFSCK, "%s: fail to checkpoint, pos = %llu: "
 454                        "rc = %d\n", scrub->os_name, scrub->os_pos_current, rc);
 455                 /* Continue, as long as the scrub itself can go ahead. */
 456         }
 457
 458         return 0;
 459 }
 460
 461 static int osd_scan_ml_file_main(const struct lu_env *env,
 462                                  struct osd_device *dev);
 463
 464 static int osd_scrub_main(void *args)
 465 {
 466         struct lu_env env;
 467         struct osd_device *dev = (struct osd_device *)args;
 468         struct lustre_scrub *scrub = &dev->od_scrub;
 469         struct lu_fid *fid;
 470         uint64_t oid;
 471         int rc = 0, ret;
 472         ENTRY;
 473
 474         rc = lu_env_init(&env, LCT_LOCAL | LCT_DT_THREAD);
 475         if (rc) {
 476                 CDEBUG(D_LFSCK, "%s: OI scrub fail to init env: rc = %d\n",
 477                        scrub->os_name, rc);
 478                 GOTO(noenv, rc);
 479         }
 480
 481         rc = scrub_thread_prep(&env, scrub, dev->od_uuid, 1);
 482         if (rc) {
 483                 CDEBUG(D_LFSCK, "%s: OI scrub fail to scrub prep: rc = %d\n",
 484                        scrub->os_name, rc);
 485                 GOTO(out, rc);
 486         }
 487
 488         if (!scrub->os_full_speed) {
 489                 struct osd_otable_it *it = dev->od_otable_it;
 490
 491                 wait_var_event(scrub,
 492                                it->ooi_user_ready ||
 493                                kthread_should_stop());
 494
 495                 if (kthread_should_stop())
 496                         GOTO(post, rc = 0);
 497
 498                 scrub->os_pos_current = it->ooi_pos;
 499         }
 500
 501         CDEBUG(D_LFSCK, "%s: OI scrub start, flags = 0x%x, pos = %llu\n",
 502                scrub->os_name, scrub->os_start_flags,
 503                scrub->os_pos_current);
 504
 505         fid = &osd_oti_get(&env)->oti_fid;
 506         while (!rc && !kthread_should_stop()) {
 507                 rc = osd_scrub_next(&env, dev, fid, &oid);
 508                 switch (rc) {
 509                 case SCRUB_NEXT_EXIT:
 510                         GOTO(post, rc = 0);
 511                 case SCRUB_NEXT_CRASH:
 512                         spin_lock(&scrub->os_lock);
 513                         scrub->os_running = 0;
 514                         spin_unlock(&scrub->os_lock);
 515                         GOTO(out, rc = -EINVAL);
 516                 case SCRUB_NEXT_FATAL:
 517                         GOTO(post, rc = -EINVAL);
 518                 case SCRUB_NEXT_BREAK:
 519                         GOTO(post, rc = 1);
 520                 }
 521
 522                 rc = osd_scrub_exec(&env, dev, fid, oid, rc);
 523         }
 524
 525         GOTO(post, rc);
 526
 527 post:
 528         if (scrub->os_has_ml_file) {
 529                 ret = osd_scan_ml_file_main(&env, dev);
 530                 if (ret != 0)
 531                         rc = ret;
 532         }
 533
 534         rc = scrub_thread_post(&env, &dev->od_scrub, rc);
 535         CDEBUG(D_LFSCK, "%s: OI scrub: stop, pos = %llu: rc = %d\n",
 536                scrub->os_name, scrub->os_pos_current, rc);
 537
 538 out:
 539         while (!list_empty(&scrub->os_inconsistent_items)) {
 540                 struct osd_inconsistent_item *oii;
 541
 542                 oii = list_entry(scrub->os_inconsistent_items.next,
 543                                  struct osd_inconsistent_item, oii_list);
 544                 list_del_init(&oii->oii_list);
 545                 OBD_FREE_PTR(oii);
 546         }
 547
 548         lu_env_fini(&env);
 549
 550 noenv:
 551         spin_lock(&scrub->os_lock);
 552         scrub->os_running = 0;
 553         spin_unlock(&scrub->os_lock);
 554         if (xchg(&scrub->os_task, NULL) == NULL)
 555                 /* scrub_stop is waiting, we need to synchronize */
 556                 wait_var_event(scrub, kthread_should_stop());
 557         wake_up_var(scrub);
 558         return rc;
 559 }
 560
 561 /* initial OI scrub */
 562
 563 struct osd_lf_map;
 564
 565 typedef int (*handle_dirent_t)(const struct lu_env *, struct osd_device *,
 566                                const char *, uint64_t, uint64_t,
 567                                enum osd_lf_flags, bool);
 568 static int osd_ios_varfid_hd(const struct lu_env *, struct osd_device *,
 569                              const char *, uint64_t, uint64_t,
 570                              enum osd_lf_flags, bool);
 571 static int osd_ios_uld_hd(const struct lu_env *, struct osd_device *,
 572                           const char *, uint64_t, uint64_t,
 573                           enum osd_lf_flags, bool);
 574
 575 typedef int (*scan_dir_t)(const struct lu_env *, struct osd_device *,
 576                           uint64_t, handle_dirent_t, enum osd_lf_flags);
 577 static int osd_ios_general_sd(const struct lu_env *, struct osd_device *,
 578                               uint64_t, handle_dirent_t, enum osd_lf_flags);
 579 static int osd_ios_ROOT_sd(const struct lu_env *, struct osd_device *,
 580                            uint64_t, handle_dirent_t, enum osd_lf_flags);
 581
 582 struct osd_lf_map {
 583         char                    *olm_name;
 584         struct lu_fid            olm_fid;
 585         enum osd_lf_flags        olm_flags;
 586         scan_dir_t               olm_scan_dir;
 587         handle_dirent_t          olm_handle_dirent;
 588 };
 589
 590 /* Add the new introduced local files in the list in the future. */
 591 static const struct osd_lf_map osd_lf_maps[] = {
 592         /* CONFIGS */
 593         {
 594                 .olm_name               = MOUNT_CONFIGS_DIR,
 595                 .olm_fid                = {
 596                         .f_seq  = FID_SEQ_LOCAL_FILE,
 597                         .f_oid  = MGS_CONFIGS_OID,
 598                 },
 599                 .olm_flags              = OLF_SCAN_SUBITEMS,
 600                 .olm_scan_dir           = osd_ios_general_sd,
 601                 .olm_handle_dirent      = osd_ios_varfid_hd,
 602         },
 603
 604         /* NIDTBL_VERSIONS */
 605         {
 606                 .olm_name               = MGS_NIDTBL_DIR,
 607                 .olm_flags              = OLF_SCAN_SUBITEMS,
 608                 .olm_scan_dir           = osd_ios_general_sd,
 609                 .olm_handle_dirent      = osd_ios_varfid_hd,
 610         },
 611
 612         /* PENDING */
 613         {
 614                 .olm_name               = MDT_ORPHAN_DIR,
 615         },
 616
 617         /* ROOT */
 618         {
 619                 .olm_name               = "ROOT",
 620                 .olm_fid                = {
 621                         .f_seq  = FID_SEQ_ROOT,
 622                         .f_oid  = FID_OID_ROOT,
 623                 },
 624                 .olm_flags              = OLF_SCAN_SUBITEMS,
 625                 .olm_scan_dir           = osd_ios_ROOT_sd,
 626         },
 627
 628         /* fld */
 629         {
 630                 .olm_name               = "fld",
 631                 .olm_fid                = {
 632                         .f_seq  = FID_SEQ_LOCAL_FILE,
 633                         .f_oid  = FLD_INDEX_OID,
 634                 },
 635         },
 636
 637         /* changelog_catalog */
 638         {
 639                 .olm_name               = CHANGELOG_CATALOG,
 640         },
 641
 642         /* changelog_users */
 643         {
 644                 .olm_name               = CHANGELOG_USERS,
 645         },
 646
 647         /* quota_master */
 648         {
 649                 .olm_name               = QMT_DIR,
 650                 .olm_flags              = OLF_SCAN_SUBITEMS,
 651                 .olm_scan_dir           = osd_ios_general_sd,
 652                 .olm_handle_dirent      = osd_ios_varfid_hd,
 653         },
 654
 655         /* quota_slave */
 656         {
 657                 .olm_name               = QSD_DIR,
 658                 .olm_flags              = OLF_SCAN_SUBITEMS,
 659                 .olm_scan_dir           = osd_ios_general_sd,
 660                 .olm_handle_dirent      = osd_ios_varfid_hd,
 661         },
 662
 663         /* LFSCK */
 664         {
 665                 .olm_name               = LFSCK_DIR,
 666                 .olm_flags              = OLF_SCAN_SUBITEMS | OLF_NOT_BACKUP,
 667                 .olm_scan_dir           = osd_ios_general_sd,
 668                 .olm_handle_dirent      = osd_ios_varfid_hd,
 669         },
 670
 671         /* lfsck_bookmark */
 672         {
 673                 .olm_name               = LFSCK_BOOKMARK,
 674         },
 675
 676         /* lfsck_layout */
 677         {
 678                 .olm_name               = LFSCK_LAYOUT,
 679         },
 680
 681         /* lfsck_namespace */
 682         {
 683                 .olm_name               = LFSCK_NAMESPACE,
 684         },
 685
 686         /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
 687          * and f_oid = index for their log files.  See lu_update_log{_dir}_fid()
 688          * for more details. */
 689
 690         /* update_log */
 691         {
 692                 .olm_name               = "update_log",
 693                 .olm_fid                = {
 694                         .f_seq  = FID_SEQ_UPDATE_LOG,
 695                 },
 696                 .olm_flags              = OLF_IDX_IN_FID,
 697         },
 698
 699         /* update_log_dir */
 700         {
 701                 .olm_name               = "update_log_dir",
 702                 .olm_fid        = {
 703                         .f_seq  = FID_SEQ_UPDATE_LOG_DIR,
 704                 },
 705                 .olm_flags              = OLF_SCAN_SUBITEMS | OLF_IDX_IN_FID,
 706                 .olm_scan_dir           = osd_ios_general_sd,
 707                 .olm_handle_dirent      = osd_ios_uld_hd,
 708         },
 709
 710         /* hsm_actions */
 711         {
 712                 .olm_name               = HSM_ACTIONS,
 713         },
 714
 715         /* nodemap */
 716         {
 717                 .olm_name               = LUSTRE_NODEMAP_NAME,
 718         },
 719
 720         /* index_backup */
 721         {
 722                 .olm_name               = INDEX_BACKUP_DIR,
 723                 .olm_fid                = {
 724                         .f_seq  = FID_SEQ_LOCAL_FILE,
 725                         .f_oid  = INDEX_BACKUP_OID,
 726                 },
 727                 .olm_flags              = OLF_SCAN_SUBITEMS | OLF_NOT_BACKUP,
 728                 .olm_scan_dir           = osd_ios_general_sd,
 729                 .olm_handle_dirent      = osd_ios_varfid_hd,
 730         },
 731
 732         {
 733                 .olm_name               = NULL
 734         }
 735 };
 736
 737 /* Add the new introduced files under .lustre/ in the list in the future. */
 738 static const struct osd_lf_map osd_dl_maps[] = {
 739         /* .lustre/fid */
 740         {
 741                 .olm_name               = "fid",
 742                 .olm_fid                = {
 743                         .f_seq  = FID_SEQ_DOT_LUSTRE,
 744                         .f_oid  = FID_OID_DOT_LUSTRE_OBF,
 745                 },
 746         },
 747
 748         /* .lustre/lost+found */
 749         {
 750                 .olm_name               = "lost+found",
 751                 .olm_fid                = {
 752                         .f_seq  = FID_SEQ_DOT_LUSTRE,
 753                         .f_oid  = FID_OID_DOT_LUSTRE_LPF,
 754                 },
 755         },
 756
 757         {
 758                 .olm_name               = NULL
 759         }
 760 };
 761
 762 struct osd_ios_item {
 763         struct list_head        oii_list;
 764         uint64_t                oii_parent;
 765         enum osd_lf_flags       oii_flags;
 766         scan_dir_t              oii_scan_dir;
 767         handle_dirent_t         oii_handle_dirent;
 768 };
 769
 770 static int osd_ios_new_item(struct osd_device *dev, uint64_t parent,
 771                             enum osd_lf_flags flags, scan_dir_t scan_dir,
 772                             handle_dirent_t handle_dirent)
 773 {
 774         struct osd_ios_item *item;
 775
 776         OBD_ALLOC_PTR(item);
 777         if (!item) {
 778                 CWARN("%s: initial OI scrub failed to add item for %llu\n",
 779                       osd_name(dev), parent);
 780                 return -ENOMEM;
 781         }
 782
 783         INIT_LIST_HEAD(&item->oii_list);
 784         item->oii_parent = parent;
 785         item->oii_flags = flags;
 786         item->oii_scan_dir = scan_dir;
 787         item->oii_handle_dirent = handle_dirent;
 788         list_add_tail(&item->oii_list, &dev->od_ios_list);
 789
 790         return 0;
 791 }
 792
 793 static bool osd_index_need_recreate(const struct lu_env *env,
 794                                     struct osd_device *dev, uint64_t oid)
 795 {
 796         struct osd_thread_info *info = osd_oti_get(env);
 797         zap_attribute_t *za = &info->oti_za2;
 798         zap_cursor_t *zc = &info->oti_zc2;
 799         int rc;
 800         ENTRY;
 801
 802         zap_cursor_init_serialized(zc, dev->od_os, oid, 0);
 803         rc = -zap_cursor_retrieve(zc, za);
 804         zap_cursor_fini(zc);
 805         if (rc && rc != -ENOENT)
 806                 RETURN(true);
 807
 808         RETURN(false);
 809 }
 810
 811 static void osd_ios_index_register(const struct lu_env *env,
 812                                    struct osd_device *osd,
 813                                    const struct lu_fid *fid, uint64_t oid)
 814 {
 815         struct osd_thread_info *info = osd_oti_get(env);
 816         zap_attribute_t *za = &info->oti_za2;
 817         zap_cursor_t *zc = &info->oti_zc2;
 818         struct zap_leaf_entry *le;
 819         dnode_t *dn = NULL;
 820         sa_handle_t *hdl;
 821         __u64 mode = 0;
 822         __u32 keysize = 0;
 823         __u32 recsize = 0;
 824         int rc;
 825         ENTRY;
 826
 827         rc = __osd_obj2dnode(osd->od_os, oid, &dn);
 828         if (rc == -EEXIST || rc == -ENOENT)
 829                 RETURN_EXIT;
 830
 831         if (rc < 0)
 832                 GOTO(log, rc);
 833
 834         if (!osd_object_is_zap(dn))
 835                 GOTO(log, rc = 1);
 836
 837         rc = -sa_handle_get(osd->od_os, oid, NULL, SA_HDL_PRIVATE, &hdl);
 838         if (rc)
 839                 GOTO(log, rc);
 840
 841         rc = -sa_lookup(hdl, SA_ZPL_MODE(osd), &mode, sizeof(mode));
 842         sa_handle_destroy(hdl);
 843         if (rc)
 844                 GOTO(log, rc);
 845
 846         if (!S_ISREG(mode))
 847                 GOTO(log, rc = 1);
 848
 849         zap_cursor_init_serialized(zc, osd->od_os, oid, 0);
 850         rc = -zap_cursor_retrieve(zc, za);
 851         if (rc)
 852                 /* Skip empty index object */
 853                 GOTO(fini, rc = (rc == -ENOENT ? 1 : rc));
 854
 855         if (zc->zc_zap->zap_ismicro ||
 856             !(zap_f_phys(zc->zc_zap)->zap_flags & ZAP_FLAG_UINT64_KEY))
 857                 GOTO(fini, rc = 1);
 858
 859         le = ZAP_LEAF_ENTRY(zc->zc_leaf, 0);
 860         keysize = le->le_name_numints * 8;
 861         recsize = za->za_integer_length * za->za_num_integers;
 862         if (likely(keysize && recsize))
 863                 rc = osd_index_register(osd, fid, keysize, recsize);
 864
 865         GOTO(fini, rc);
 866
 867 fini:
 868         zap_cursor_fini(zc);
 869
 870 log:
 871         if (dn)
 872                 osd_dnode_rele(dn);
 873         if (rc < 0)
 874                 CWARN("%s: failed to register index "DFID" (%u/%u): rc = %d\n",
 875                       osd_name(osd), PFID(fid), keysize, recsize, rc);
 876         else if (!rc)
 877                 CDEBUG(D_LFSCK, "%s: registered index "DFID" (%u/%u)\n",
 878                        osd_name(osd), PFID(fid), keysize, recsize);
 879 }
 880
 881 static void osd_index_restore(const struct lu_env *env, struct osd_device *dev,
 882                               struct lustre_index_restore_unit *liru, void *buf,
 883                               int bufsize)
 884 {
 885         struct luz_direntry *zde = &osd_oti_get(env)->oti_zde;
 886         struct lu_fid *tgt_fid = &liru->liru_cfid;
 887         struct lu_fid bak_fid;
 888         int rc;
 889         ENTRY;
 890
 891         lustre_fid2lbx(buf, tgt_fid, bufsize);
 892         rc = -zap_lookup(dev->od_os, dev->od_index_backup_id, buf, 8,
 893                          sizeof(*zde) / 8, (void *)zde);
 894         if (rc)
 895                 GOTO(log, rc);
 896
 897         rc = osd_get_fid_by_oid(env, dev, zde->lzd_reg.zde_dnode, &bak_fid);
 898         if (rc)
 899                 GOTO(log, rc);
 900
 901         /* The OI mapping for index may be invalid, since it will be
 902          * re-created, not update the OI mapping, just cache it in RAM. */
 903         rc = osd_idc_find_and_init_with_oid(env, dev, tgt_fid,
 904                                             liru->liru_clid);
 905         if (!rc)
 906                 rc = lustre_index_restore(env, &dev->od_dt_dev,
 907                                 &liru->liru_pfid, tgt_fid, &bak_fid,
 908                                 liru->liru_name, &dev->od_index_backup_list,
 909                                 &dev->od_lock, buf, bufsize);
 910         GOTO(log, rc);
 911
 912 log:
 913         CDEBUG(D_WARNING, "%s: restore index '%s' with "DFID": rc = %d\n",
 914                osd_name(dev), liru->liru_name, PFID(tgt_fid), rc);
 915 }
 916
 917 /**
 918  * verify FID-in-LMA and OI entry for one object
 919  *
 920  * ios: Initial OI Scrub.
 921  */
 922 static int osd_ios_scan_one(const struct lu_env *env, struct osd_device *dev,
 923                             const struct lu_fid *fid, uint64_t parent,
 924                             uint64_t oid, const char *name,
 925                             enum osd_lf_flags flags)
 926 {
 927         struct lustre_scrub *scrub = &dev->od_scrub;
 928         struct scrub_file *sf = &scrub->os_file;
 929         struct lustre_mdt_attrs *lma = NULL;
 930         nvlist_t *nvbuf = NULL;
 931         struct lu_fid tfid;
 932         uint64_t oid2 = 0;
 933         __u64 flag = 0;
 934         int size = 0;
 935         int op = 0;
 936         int rc;
 937         ENTRY;
 938
 939         rc = __osd_xattr_load_by_oid(dev, oid, &nvbuf);
 940         if (unlikely(rc == -ENOENT || rc == -EEXIST))
 941                 RETURN(0);
 942
 943         if (rc && rc != -ENODATA) {
 944                 CWARN("%s: initial OI scrub failed to get lma for %llu: "
 945                       "rc = %d\n", osd_name(dev), oid, rc);
 946
 947                 RETURN(rc);
 948         }
 949
 950         if (!rc) {
 951                 LASSERT(nvbuf != NULL);
 952                 rc = -nvlist_lookup_byte_array(nvbuf, XATTR_NAME_LMA,
 953                                                (uchar_t **)&lma, &size);
 954                 if (rc || size == 0) {
 955                         LASSERT(lma == NULL);
 956                         rc = -ENODATA;
 957                 } else {
 958                         LASSERTF(lma != NULL, "corrupted LMA, size %d\n", size);
 959                         lustre_lma_swab(lma);
 960                         if (lma->lma_compat & LMAC_NOT_IN_OI) {
 961                                 nvlist_free(nvbuf);
 962                                 RETURN(0);
 963                         }
 964
 965                         if (lma->lma_compat & LMAC_IDX_BACKUP &&
 966                             osd_index_need_recreate(env, dev, oid)) {
 967                                 if (parent == dev->od_root) {
 968                                         lu_local_obj_fid(&tfid,
 969                                                          OSD_FS_ROOT_OID);
 970                                 } else {
 971                                         rc = osd_get_fid_by_oid(env, dev,
 972                                                                 parent, &tfid);
 973                                         if (rc) {
 974                                                 nvlist_free(nvbuf);
 975                                                 RETURN(rc);
 976                                         }
 977                                 }
 978
 979                                 rc = lustre_liru_new(
 980                                                 &dev->od_index_restore_list,
 981                                                 &tfid, &lma->lma_self_fid, oid,
 982                                                 name, strlen(name));
 983                                 nvlist_free(nvbuf);
 984                                 RETURN(rc);
 985                         }
 986
 987                         tfid = lma->lma_self_fid;
 988                         if (!(flags & OLF_NOT_BACKUP))
 989                                 osd_ios_index_register(env, dev, &tfid, oid);
 990                 }
 991                 nvlist_free(nvbuf);
 992         }
 993
 994         if (rc == -ENODATA) {
 995                 if (!fid) {
 996                         /* Skip the object without FID-in-LMA */
 997                         CDEBUG(D_LFSCK, "%s: %llu has no FID-in-LMA, skip it\n",
 998                                osd_name(dev), oid);
 999
1000                         RETURN(0);
1001                 }
1002
1003                 LASSERT(!fid_is_zero(fid));
1004
1005                 tfid = *fid;
1006                 if (flags & OLF_IDX_IN_FID) {
1007                         LASSERT(dev->od_index >= 0);
1008
1009                         tfid.f_oid = dev->od_index;
1010                 }
1011         }
1012
1013         rc = osd_fid_lookup(env, dev, &tfid, &oid2);
1014         if (rc) {
1015                 if (rc != -ENOENT) {
1016                         CWARN("%s: initial OI scrub failed to lookup fid for "
1017                               DFID"=>%llu: rc = %d\n",
1018                               osd_name(dev), PFID(&tfid), oid, rc);
1019
1020                         RETURN(rc);
1021                 }
1022
1023                 flag = SF_RECREATED;
1024                 op = DTO_INDEX_INSERT;
1025         } else {
1026                 if (oid == oid2)
1027                         RETURN(0);
1028
1029                 flag = SF_INCONSISTENT;
1030                 op = DTO_INDEX_UPDATE;
1031         }
1032
1033         if (!(sf->sf_flags & flag)) {
1034                 scrub_file_reset(scrub, dev->od_uuid, flag);
1035                 rc = scrub_file_store(env, scrub);
1036                 if (rc)
1037                         RETURN(rc);
1038         }
1039
1040         rc = osd_scrub_refresh_mapping(env, dev, &tfid, oid, op, true, name);
1041
1042         RETURN(rc > 0 ? 0 : rc);
1043 }
1044
1045 static int osd_ios_varfid_hd(const struct lu_env *env, struct osd_device *dev,
1046                              const char *name, uint64_t parent, uint64_t oid,
1047                              enum osd_lf_flags flags, bool is_dir)
1048 {
1049         int rc;
1050         ENTRY;
1051
1052         rc = osd_ios_scan_one(env, dev, NULL, parent, oid, name, 0);
1053         if (!rc && is_dir)
1054                 rc = osd_ios_new_item(dev, oid, flags, osd_ios_general_sd,
1055                                       osd_ios_varfid_hd);
1056
1057         RETURN(rc);
1058 }
1059
1060 static int osd_ios_uld_hd(const struct lu_env *env, struct osd_device *dev,
1061                           const char *name, uint64_t parent, uint64_t oid,
1062                           enum osd_lf_flags flags, bool is_dir)
1063 {
1064         struct lu_fid tfid;
1065         int rc;
1066         ENTRY;
1067
1068         /* skip any non-DFID format name */
1069         if (name[0] != '[')
1070                 RETURN(0);
1071
1072         /* skip the start '[' */
1073         sscanf(&name[1], SFID, RFID(&tfid));
1074         if (fid_is_sane(&tfid))
1075                 rc = osd_ios_scan_one(env, dev, &tfid, parent, oid, name, 0);
1076         else
1077                 rc = -EIO;
1078
1079         RETURN(rc);
1080 }
1081
1082 /*
1083  * General scanner for the directories execpt /ROOT during initial OI scrub.
1084  * It scans the name entries under the given directory one by one. For each
1085  * entry, verifies its OI mapping via the given @handle_dirent.
1086  */
1087 static int osd_ios_general_sd(const struct lu_env *env, struct osd_device *dev,
1088                               uint64_t parent, handle_dirent_t handle_dirent,
1089                               enum osd_lf_flags flags)
1090 {
1091         struct osd_thread_info *info = osd_oti_get(env);
1092         struct luz_direntry *zde = &info->oti_zde;
1093         zap_attribute_t *za = &info->oti_za;
1094         zap_cursor_t *zc = &info->oti_zc;
1095         int rc;
1096         ENTRY;
1097
1098         zap_cursor_init_serialized(zc, dev->od_os, parent, 0);
1099         rc = -zap_cursor_retrieve(zc, za);
1100         if (rc == -ENOENT)
1101                 zap_cursor_advance(zc);
1102         else if (rc)
1103                 GOTO(log, rc);
1104
1105         while (1) {
1106                 rc = -zap_cursor_retrieve(zc, za);
1107                 if (rc)
1108                         GOTO(log, rc = (rc == -ENOENT ? 0 : rc));
1109
1110                 /* skip the entry started with '.' */
1111                 if (likely(za->za_name[0] != '.')) {
1112                         rc = osd_zap_lookup(dev, parent, NULL, za->za_name,
1113                                         za->za_integer_length,
1114                                         sizeof(*zde) / za->za_integer_length,
1115                                         (void *)zde);
1116                         if (rc) {
1117                                 CWARN("%s: initial OI scrub failed to lookup "
1118                                       "%s under %llu: rc = %d\n",
1119                                       osd_name(dev), za->za_name, parent, rc);
1120                                 continue;
1121                         }
1122
1123                         rc = handle_dirent(env, dev, za->za_name, parent,
1124                                         zde->lzd_reg.zde_dnode, flags,
1125                                         S_ISDIR(DTTOIF(zde->lzd_reg.zde_type)) ?
1126                                         true : false);
1127                         CDEBUG(D_LFSCK, "%s: initial OI scrub handled %s under "
1128                                "%llu: rc = %d\n",
1129                                osd_name(dev), za->za_name, parent, rc);
1130                 }
1131
1132                 zap_cursor_advance(zc);
1133         }
1134
1135 log:
1136         if (rc)
1137                 CWARN("%s: initial OI scrub failed to scan the directory %llu: "
1138                       "rc = %d\n", osd_name(dev), parent, rc);
1139         zap_cursor_fini(zc);
1140
1141         return rc;
1142 }
1143
1144 /*
1145  * The scanner for /ROOT directory. It is not all the items under /ROOT will
1146  * be scanned during the initial OI scrub, instead, only the .lustre and the
1147  * sub-items under .lustre will be handled.
1148  */
1149 static int osd_ios_ROOT_sd(const struct lu_env *env, struct osd_device *dev,
1150                            uint64_t parent, handle_dirent_t handle_dirent,
1151                            enum osd_lf_flags flags)
1152 {
1153         struct luz_direntry *zde = &osd_oti_get(env)->oti_zde;
1154         const struct osd_lf_map *map;
1155         uint64_t oid;
1156         int rc;
1157         int rc1 = 0;
1158         ENTRY;
1159
1160         rc = osd_zap_lookup(dev, parent, NULL, dot_lustre_name, 8,
1161                             sizeof(*zde) / 8, (void *)zde);
1162         if (rc == -ENOENT) {
1163                 /* The .lustre directory is lost. That is not fatal. It can
1164                  * be re-created in the subsequent MDT start processing. */
1165                 RETURN(0);
1166         }
1167
1168         if (rc) {
1169                 CWARN("%s: initial OI scrub failed to find .lustre: "
1170                       "rc = %d\n", osd_name(dev), rc);
1171
1172                 RETURN(rc);
1173         }
1174
1175         oid = zde->lzd_reg.zde_dnode;
1176         rc = osd_ios_scan_one(env, dev, &LU_DOT_LUSTRE_FID, parent, oid,
1177                               dot_lustre_name, 0);
1178         if (rc)
1179                 RETURN(rc);
1180
1181         for (map = osd_dl_maps; map->olm_name; map++) {
1182                 rc = osd_zap_lookup(dev, oid, NULL, map->olm_name, 8,
1183                                     sizeof(*zde) / 8, (void *)zde);
1184                 if (rc) {
1185                         if (rc != -ENOENT)
1186                                 CWARN("%s: initial OI scrub failed to find the entry %s under .lustre: rc = %d\n",
1187                                       osd_name(dev), map->olm_name, rc);
1188                         else if (!fid_is_zero(&map->olm_fid))
1189                                 /* Try to remove the stale OI mapping. */
1190                                 osd_scrub_refresh_mapping(env, dev,
1191                                                 &map->olm_fid, 0,
1192                                                 DTO_INDEX_DELETE, true,
1193                                                 map->olm_name);
1194                         continue;
1195                 }
1196
1197                 rc = osd_ios_scan_one(env, dev, &map->olm_fid, oid,
1198                                       zde->lzd_reg.zde_dnode, map->olm_name,
1199                                       map->olm_flags);
1200                 if (rc)
1201                         rc1 = rc;
1202         }
1203
1204         RETURN(rc1);
1205 }
1206
1207 static void osd_initial_OI_scrub(const struct lu_env *env,
1208                                  struct osd_device *dev)
1209 {
1210         struct luz_direntry *zde = &osd_oti_get(env)->oti_zde;
1211         const struct osd_lf_map *map;
1212         int rc;
1213         ENTRY;
1214
1215         for (map = osd_lf_maps; map->olm_name; map++) {
1216                 rc = osd_zap_lookup(dev, dev->od_root, NULL, map->olm_name, 8,
1217                                     sizeof(*zde) / 8, (void *)zde);
1218                 if (rc) {
1219                         if (rc != -ENOENT)
1220                                 CWARN("%s: initial OI scrub failed "
1221                                       "to find the entry %s: rc = %d\n",
1222                                       osd_name(dev), map->olm_name, rc);
1223                         else if (!fid_is_zero(&map->olm_fid))
1224                                 /* Try to remove the stale OI mapping. */
1225                                 osd_scrub_refresh_mapping(env, dev,
1226                                                 &map->olm_fid, 0,
1227                                                 DTO_INDEX_DELETE, true,
1228                                                 map->olm_name);
1229                         continue;
1230                 }
1231
1232                 rc = osd_ios_scan_one(env, dev, &map->olm_fid, dev->od_root,
1233                                       zde->lzd_reg.zde_dnode, map->olm_name,
1234                                       map->olm_flags);
1235                 if (!rc && map->olm_flags & OLF_SCAN_SUBITEMS)
1236                         osd_ios_new_item(dev, zde->lzd_reg.zde_dnode,
1237                                          map->olm_flags, map->olm_scan_dir,
1238                                          map->olm_handle_dirent);
1239         }
1240
1241         while (!list_empty(&dev->od_ios_list)) {
1242                 struct osd_ios_item *item;
1243
1244                 item = list_entry(dev->od_ios_list.next,
1245                                   struct osd_ios_item, oii_list);
1246                 list_del_init(&item->oii_list);
1247                 item->oii_scan_dir(env, dev, item->oii_parent,
1248                                    item->oii_handle_dirent, item->oii_flags);
1249                 OBD_FREE_PTR(item);
1250         }
1251
1252         if (!list_empty(&dev->od_index_restore_list)) {
1253                 char *buf;
1254
1255                 OBD_ALLOC_LARGE(buf, INDEX_BACKUP_BUFSIZE);
1256                 if (!buf)
1257                         CERROR("%s: not enough RAM for rebuild index\n",
1258                                osd_name(dev));
1259
1260                 while (!list_empty(&dev->od_index_restore_list)) {
1261                         struct lustre_index_restore_unit *liru;
1262
1263                         liru = list_entry(dev->od_index_restore_list.next,
1264                                           struct lustre_index_restore_unit,
1265                                           liru_link);
1266                         list_del(&liru->liru_link);
1267                         if (buf)
1268                                 osd_index_restore(env, dev, liru, buf,
1269                                                   INDEX_BACKUP_BUFSIZE);
1270                         OBD_FREE(liru, liru->liru_len);
1271                 }
1272
1273                 if (buf)
1274                         OBD_FREE_LARGE(buf, INDEX_BACKUP_BUFSIZE);
1275         }
1276
1277         EXIT;
1278 }
1279
1280 /* OI scrub start/stop */
1281
1282 int osd_scrub_start(const struct lu_env *env, struct osd_device *dev,
1283                     __u32 flags)
1284 {
1285         int rc;
1286         ENTRY;
1287
1288         if (dev->od_dt_dev.dd_rdonly)
1289                 RETURN(-EROFS);
1290
1291         /* od_otable_sem: prevent concurrent start/stop */
1292         down(&dev->od_otable_sem);
1293         rc = scrub_start(osd_scrub_main, &dev->od_scrub, dev, flags);
1294         up(&dev->od_otable_sem);
1295
1296         RETURN(rc == -EALREADY ? 0 : rc);
1297 }
1298
1299 void osd_scrub_stop(struct osd_device *dev)
1300 {
1301         struct lustre_scrub *scrub = &dev->od_scrub;
1302         ENTRY;
1303
1304         /* od_otable_sem: prevent concurrent start/stop */
1305         down(&dev->od_otable_sem);
1306         spin_lock(&scrub->os_lock);
1307         scrub->os_paused = 1;
1308         spin_unlock(&scrub->os_lock);
1309         scrub_stop(scrub);
1310         up(&dev->od_otable_sem);
1311
1312         EXIT;
1313 }
1314
1315 /* OI scrub setup/cleanup */
1316
1317 static const char osd_scrub_name[] = "OI_scrub";
1318
1319 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev,
1320                     bool resetoi)
1321 {
1322         struct osd_thread_info *info = osd_oti_get(env);
1323         struct lustre_scrub *scrub = &dev->od_scrub;
1324         struct scrub_file *sf = &scrub->os_file;
1325         struct lu_fid *fid = &info->oti_fid;
1326         struct dt_object *obj;
1327         uint64_t oid;
1328         int rc = 0;
1329         bool dirty = false;
1330         ENTRY;
1331
1332         memcpy(dev->od_uuid.b,
1333                &dsl_dataset_phys(dev->od_os->os_dsl_dataset)->ds_guid,
1334                sizeof(dsl_dataset_phys(dev->od_os->os_dsl_dataset)->ds_guid));
1335         memset(&dev->od_scrub, 0, sizeof(struct lustre_scrub));
1336         init_rwsem(&scrub->os_rwsem);
1337         spin_lock_init(&scrub->os_lock);
1338         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
1339         scrub->os_name = osd_name(dev);
1340
1341         /* 'What the @fid is' is not imporatant, because the object
1342          * has no OI mapping, and only is visible inside the OSD.*/
1343         fid->f_seq = FID_SEQ_IGIF_MAX;
1344         if (dev->od_is_ost)
1345                 fid->f_oid = ((1 << 31) | dev->od_index) + 1;
1346         else
1347                 fid->f_oid = dev->od_index + 1;
1348         fid->f_ver = 0;
1349         rc = osd_obj_find_or_create(env, dev, dev->od_root,
1350                                     osd_scrub_name, &oid, fid, false);
1351         if (rc)
1352                 RETURN(rc);
1353
1354         rc = osd_idc_find_and_init_with_oid(env, dev, fid, oid);
1355         if (rc)
1356                 RETURN(rc);
1357
1358         obj = lu2dt(lu_object_find_slice(env, osd2lu_dev(dev), fid, NULL));
1359         if (IS_ERR_OR_NULL(obj))
1360                 RETURN(obj ? PTR_ERR(obj) : -ENOENT);
1361
1362         obj->do_body_ops = &osd_body_scrub_ops;
1363         scrub->os_obj = obj;
1364         rc = scrub_file_load(env, scrub);
1365         if (rc == -ENOENT || rc == -EFAULT) {
1366                 scrub_file_init(scrub, dev->od_uuid);
1367                 dirty = true;
1368         } else if (rc < 0) {
1369                 GOTO(cleanup_obj, rc);
1370         } else {
1371                 if (!uuid_equal(&sf->sf_uuid, &dev->od_uuid)) {
1372                         CDEBUG(D_LFSCK,
1373                                "%s: UUID has been changed from %pU to %pU\n",
1374                                osd_name(dev), &sf->sf_uuid, &dev->od_uuid);
1375                         scrub_file_reset(scrub, dev->od_uuid, SF_INCONSISTENT);
1376                         dirty = true;
1377                 } else if (sf->sf_status == SS_SCANNING) {
1378                         sf->sf_status = SS_CRASHED;
1379                         dirty = true;
1380                 }
1381
1382                 if ((sf->sf_oi_count & (sf->sf_oi_count - 1)) != 0) {
1383                         LCONSOLE_WARN("%s: invalid oi count %d, set it to %d\n",
1384                                       osd_name(dev), sf->sf_oi_count,
1385                                       osd_oi_count);
1386                         sf->sf_oi_count = osd_oi_count;
1387                         dirty = true;
1388                 }
1389         }
1390
1391         if (sf->sf_pos_last_checkpoint != 0)
1392                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
1393         else
1394                 scrub->os_pos_current = 1;
1395
1396         if (dirty) {
1397                 rc = scrub_file_store(env, scrub);
1398                 if (rc)
1399                         GOTO(cleanup_obj, rc);
1400         }
1401
1402         /* Initialize OI files. */
1403         rc = osd_oi_init(env, dev, resetoi);
1404         if (rc < 0)
1405                 GOTO(cleanup_obj, rc);
1406
1407         if (!dev->od_dt_dev.dd_rdonly)
1408                 osd_initial_OI_scrub(env, dev);
1409
1410         if (!dev->od_dt_dev.dd_rdonly &&
1411             dev->od_auto_scrub_interval != AS_NEVER &&
1412             ((sf->sf_status == SS_PAUSED) ||
1413              (sf->sf_status == SS_CRASHED &&
1414               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
1415                               SF_UPGRADE | SF_AUTO)) ||
1416              (sf->sf_status == SS_INIT &&
1417               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
1418                               SF_UPGRADE))))
1419                 rc = osd_scrub_start(env, dev, SS_AUTO_FULL);
1420
1421         if (rc)
1422                 GOTO(cleanup_oi, rc);
1423
1424         RETURN(0);
1425
1426 cleanup_oi:
1427         osd_oi_fini(env, dev);
1428 cleanup_obj:
1429         dt_object_put_nocache(env, scrub->os_obj);
1430         scrub->os_obj = NULL;
1431
1432         return rc;
1433 }
1434
1435 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
1436 {
1437         struct lustre_scrub *scrub = &dev->od_scrub;
1438
1439         LASSERT(!dev->od_otable_it);
1440
1441         if (scrub->os_obj) {
1442                 osd_scrub_stop(dev);
1443                 dt_object_put_nocache(env, scrub->os_obj);
1444                 scrub->os_obj = NULL;
1445         }
1446
1447         if (dev->od_oi_table)
1448                 osd_oi_fini(env, dev);
1449 }
1450
1451 /* object table based iteration APIs */
1452
1453 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
1454                                        struct dt_object *dt, __u32 attr)
1455 {
1456         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
1457         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
1458         struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1459         struct lustre_scrub *scrub = &dev->od_scrub;
1460         struct osd_otable_it *it;
1461         __u32 start = 0;
1462         int rc;
1463         ENTRY;
1464
1465         if (dev->od_dt_dev.dd_rdonly)
1466                 RETURN(ERR_PTR(-EROFS));
1467
1468         /* od_otable_sem: prevent concurrent init/fini */
1469         down(&dev->od_otable_sem);
1470         if (dev->od_otable_it)
1471                 GOTO(out, it = ERR_PTR(-EALREADY));
1472
1473         OBD_ALLOC_PTR(it);
1474         if (!it)
1475                 GOTO(out, it = ERR_PTR(-ENOMEM));
1476
1477         if (flags & DOIF_OUTUSED)
1478                 it->ooi_used_outside = 1;
1479
1480         if (flags & DOIF_RESET)
1481                 start |= SS_RESET;
1482
1483         if (valid & DOIV_ERROR_HANDLE) {
1484                 if (flags & DOIF_FAILOUT)
1485                         start |= SS_SET_FAILOUT;
1486                 else
1487                         start |= SS_CLEAR_FAILOUT;
1488         }
1489
1490         if (valid & DOIV_DRYRUN) {
1491                 if (flags & DOIF_DRYRUN)
1492                         start |= SS_SET_DRYRUN;
1493                 else
1494                         start |= SS_CLEAR_DRYRUN;
1495         }
1496
1497         /* XXX: dmu_object_next() does NOT find dnodes allocated
1498          *      in the current non-committed txg, so we force txg
1499          *      commit to find all existing dnodes ... */
1500         txg_wait_synced(dmu_objset_pool(dev->od_os), 0ULL);
1501
1502         dev->od_otable_it = it;
1503         it->ooi_dev = dev;
1504         rc = scrub_start(osd_scrub_main, scrub, dev, start & ~SS_AUTO_PARTIAL);
1505         if (rc == -EALREADY) {
1506                 it->ooi_pos = 1;
1507         } else if (rc < 0) {
1508                 dev->od_otable_it = NULL;
1509                 OBD_FREE_PTR(it);
1510                 it = ERR_PTR(rc);
1511         } else {
1512                 it->ooi_pos = scrub->os_pos_current;
1513         }
1514
1515         GOTO(out, it);
1516
1517 out:
1518         up(&dev->od_otable_sem);
1519         return (struct dt_it *)it;
1520 }
1521
1522 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
1523 {
1524         struct osd_otable_it *it = (struct osd_otable_it *)di;
1525         struct osd_device *dev = it->ooi_dev;
1526
1527         /* od_otable_sem: prevent concurrent init/fini */
1528         down(&dev->od_otable_sem);
1529         scrub_stop(&dev->od_scrub);
1530         LASSERT(dev->od_otable_it == it);
1531
1532         dev->od_otable_it = NULL;
1533         up(&dev->od_otable_sem);
1534         OBD_FREE_PTR(it);
1535 }
1536
1537 static int osd_otable_it_get(const struct lu_env *env,
1538                              struct dt_it *di, const struct dt_key *key)
1539 {
1540         return 0;
1541 }
1542
1543 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
1544 {
1545 }
1546
1547 static void osd_otable_it_preload(const struct lu_env *env,
1548                                   struct osd_otable_it *it)
1549 {
1550         struct osd_device *dev = it->ooi_dev;
1551         int rc;
1552
1553         /* can go negative on the very first access to the iterator
1554          * or if some non-Lustre objects were found */
1555         if (unlikely(it->ooi_prefetched < 0))
1556                 it->ooi_prefetched = 0;
1557
1558         if (it->ooi_prefetched >= (OTABLE_PREFETCH >> 1))
1559                 return;
1560
1561         if (it->ooi_prefetched_dnode == 0)
1562                 it->ooi_prefetched_dnode = it->ooi_pos;
1563
1564         while (it->ooi_prefetched < OTABLE_PREFETCH) {
1565                 rc = -dmu_object_next(dev->od_os, &it->ooi_prefetched_dnode,
1566                                       B_FALSE, 0);
1567                 if (rc)
1568                         break;
1569
1570                 osd_dmu_prefetch(dev->od_os, it->ooi_prefetched_dnode,
1571                                  0, 0, 0, ZIO_PRIORITY_ASYNC_READ);
1572                 it->ooi_prefetched++;
1573         }
1574 }
1575
1576 static inline int
1577 osd_otable_it_wakeup(struct lustre_scrub *scrub, struct osd_otable_it *it)
1578 {
1579         spin_lock(&scrub->os_lock);
1580         if (it->ooi_pos < scrub->os_pos_current || scrub->os_waiting ||
1581             !scrub->os_running)
1582                 it->ooi_waiting = 0;
1583         else
1584                 it->ooi_waiting = 1;
1585         spin_unlock(&scrub->os_lock);
1586
1587         return !it->ooi_waiting;
1588 }
1589
1590 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
1591 {
1592         struct osd_otable_it *it = (struct osd_otable_it *)di;
1593         struct osd_device *dev = it->ooi_dev;
1594         struct lustre_scrub *scrub = &dev->od_scrub;
1595         struct lustre_mdt_attrs *lma = NULL;
1596         nvlist_t *nvbuf = NULL;
1597         int rc, size = 0;
1598         bool locked;
1599         ENTRY;
1600
1601         LASSERT(it->ooi_user_ready);
1602         fid_zero(&it->ooi_fid);
1603
1604         if (unlikely(it->ooi_all_cached))
1605                 RETURN(1);
1606
1607 again:
1608         if (nvbuf) {
1609                 nvlist_free(nvbuf);
1610                 nvbuf = NULL;
1611                 lma = NULL;
1612                 size = 0;
1613         }
1614
1615         if (it->ooi_pos >= scrub->os_pos_current)
1616                 wait_var_event(scrub,
1617                                osd_otable_it_wakeup(scrub, it));
1618
1619         if (!scrub->os_running && !it->ooi_used_outside)
1620                 GOTO(out, rc = 1);
1621
1622         rc = -dmu_object_next(dev->od_os, &it->ooi_pos, B_FALSE, 0);
1623         if (rc) {
1624                 if (unlikely(rc == -ESRCH)) {
1625                         it->ooi_all_cached = 1;
1626                         rc = 1;
1627                 }
1628
1629                 GOTO(out, rc);
1630         }
1631
1632         rc = __osd_xattr_load_by_oid(dev, it->ooi_pos, &nvbuf);
1633
1634         locked = false;
1635         if (!scrub->os_full_speed) {
1636                 spin_lock(&scrub->os_lock);
1637                 locked = true;
1638         }
1639         it->ooi_prefetched--;
1640         if (!scrub->os_full_speed) {
1641                 if (scrub->os_waiting) {
1642                         scrub->os_waiting = 0;
1643                         wake_up_var(scrub);
1644                 }
1645         }
1646         if (locked)
1647                 spin_unlock(&scrub->os_lock);
1648
1649         if (rc == -ENOENT || rc == -EEXIST || rc == -ENODATA)
1650                 goto again;
1651
1652         if (rc)
1653                 GOTO(out, rc);
1654
1655         LASSERT(nvbuf != NULL);
1656         rc = -nvlist_lookup_byte_array(nvbuf, XATTR_NAME_LMA,
1657                                        (uchar_t **)&lma, &size);
1658         if (rc || size == 0)
1659                 /* It is either non-Lustre object or OSD internal object,
1660                  * ignore it, go ahead */
1661                 goto again;
1662
1663         LASSERTF(lma != NULL, "corrupted LMA, size %d\n", size);
1664         lustre_lma_swab(lma);
1665         if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI ||
1666                      lma->lma_incompat & LMAI_AGENT))
1667                 goto again;
1668
1669         it->ooi_fid = lma->lma_self_fid;
1670
1671         GOTO(out, rc = 0);
1672
1673 out:
1674         if (nvbuf)
1675                 nvlist_free(nvbuf);
1676
1677         if (!rc && scrub->os_full_speed)
1678                 osd_otable_it_preload(env, it);
1679
1680         return rc;
1681 }
1682
1683 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
1684                                         const struct dt_it *di)
1685 {
1686         return NULL;
1687 }
1688
1689 static int osd_otable_it_key_size(const struct lu_env *env,
1690                                   const struct dt_it *di)
1691 {
1692         return sizeof(__u64);
1693 }
1694
1695 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
1696                              struct dt_rec *rec, __u32 attr)
1697 {
1698         struct osd_otable_it *it  = (struct osd_otable_it *)di;
1699         struct lu_fid *fid = (struct lu_fid *)rec;
1700
1701         *fid = it->ooi_fid;
1702         return 0;
1703 }
1704
1705 static __u64 osd_otable_it_store(const struct lu_env *env,
1706                                  const struct dt_it *di)
1707 {
1708         struct osd_otable_it *it = (struct osd_otable_it *)di;
1709
1710         return it->ooi_pos;
1711 }
1712
1713 /**
1714  * Set the OSD layer iteration start position as the specified hash.
1715  */
1716 static int osd_otable_it_load(const struct lu_env *env,
1717                               const struct dt_it *di, __u64 hash)
1718 {
1719         struct osd_otable_it *it = (struct osd_otable_it *)di;
1720         struct osd_device *dev = it->ooi_dev;
1721         struct lustre_scrub *scrub = &dev->od_scrub;
1722         int rc;
1723         ENTRY;
1724
1725         /* Forbid to set iteration position after iteration started. */
1726         if (it->ooi_user_ready)
1727                 RETURN(-EPERM);
1728
1729         if (hash > OSD_OTABLE_MAX_HASH)
1730                 hash = OSD_OTABLE_MAX_HASH;
1731
1732         /* The hash is the last checkpoint position,
1733          * we will start from the next one. */
1734         it->ooi_pos = hash + 1;
1735         it->ooi_prefetched = 0;
1736         it->ooi_prefetched_dnode = 0;
1737         it->ooi_user_ready = 1;
1738         if (!scrub->os_full_speed)
1739                 wake_up_var(scrub);
1740
1741         /* Unplug OSD layer iteration by the first next() call. */
1742         rc = osd_otable_it_next(env, (struct dt_it *)it);
1743
1744         RETURN(rc);
1745 }
1746
1747 static int osd_otable_it_key_rec(const struct lu_env *env,
1748                                  const struct dt_it *di, void *key_rec)
1749 {
1750         return 0;
1751 }
1752
1753 const struct dt_index_operations osd_otable_ops = {
1754         .dio_it = {
1755                 .init     = osd_otable_it_init,
1756                 .fini     = osd_otable_it_fini,
1757                 .get      = osd_otable_it_get,
1758                 .put      = osd_otable_it_put,
1759                 .next     = osd_otable_it_next,
1760                 .key      = osd_otable_it_key,
1761                 .key_size = osd_otable_it_key_size,
1762                 .rec      = osd_otable_it_rec,
1763                 .store    = osd_otable_it_store,
1764                 .load     = osd_otable_it_load,
1765                 .key_rec  = osd_otable_it_key_rec,
1766         }
1767 };
1768
1769 /* high priority inconsistent items list APIs */
1770
1771 int osd_oii_insert(const struct lu_env *env, struct osd_device *dev,
1772                    const struct lu_fid *fid, uint64_t oid, bool insert)
1773 {
1774         struct lustre_scrub *scrub = &dev->od_scrub;
1775         struct osd_inconsistent_item *oii;
1776         bool wakeup = false;
1777         ENTRY;
1778
1779         osd_idc_find_and_init_with_oid(env, dev, fid, oid);
1780         OBD_ALLOC_PTR(oii);
1781         if (unlikely(!oii))
1782                 RETURN(-ENOMEM);
1783
1784         INIT_LIST_HEAD(&oii->oii_list);
1785         oii->oii_cache.oic_dev = dev;
1786         oii->oii_cache.oic_fid = *fid;
1787         oii->oii_cache.oic_dnode = oid;
1788         oii->oii_insert = insert;
1789
1790         spin_lock(&scrub->os_lock);
1791         if (!scrub->os_running) {
1792                 spin_unlock(&scrub->os_lock);
1793                 OBD_FREE_PTR(oii);
1794                 RETURN(-EAGAIN);
1795         }
1796
1797         if (list_empty(&scrub->os_inconsistent_items))
1798                 wakeup = true;
1799         list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
1800         spin_unlock(&scrub->os_lock);
1801
1802         if (wakeup)
1803                 wake_up_var(scrub);
1804
1805         RETURN(0);
1806 }
1807
1808 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
1809                    uint64_t *oid)
1810 {
1811         struct lustre_scrub *scrub = &dev->od_scrub;
1812         struct osd_inconsistent_item *oii;
1813         int ret = -ENOENT;
1814         ENTRY;
1815
1816         spin_lock(&scrub->os_lock);
1817         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
1818                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
1819                         *oid = oii->oii_cache.oic_dnode;
1820                         ret = 0;
1821                         break;
1822                 }
1823         }
1824         spin_unlock(&scrub->os_lock);
1825
1826         RETURN(ret);
1827 }
1828
1829 typedef int (*scan_dir_helper_t)(const struct lu_env *env,
1830                                  struct osd_device *dev, uint64_t dir_oid,
1831                                  struct osd_zap_it *ozi);
1832
1833 static int osd_scan_dir(const struct lu_env *env, struct osd_device *dev,
1834                         uint64_t id, scan_dir_helper_t cb)
1835 {
1836         struct osd_zap_it *it;
1837         struct luz_direntry *zde;
1838         zap_attribute_t *za;
1839         int rc;
1840
1841         ENTRY;
1842
1843         OBD_SLAB_ALLOC_PTR_GFP(it, osd_zapit_cachep, GFP_NOFS);
1844         if (it == NULL)
1845                 RETURN(-ENOMEM);
1846
1847         rc = osd_zap_cursor_init(&it->ozi_zc, dev->od_os, id, 0);
1848         if (rc != 0)
1849                 GOTO(out, rc);
1850
1851         za = &it->ozi_za;
1852         zde = &it->ozi_zde;
1853         while (1) {
1854                 rc = -zap_cursor_retrieve(it->ozi_zc, za);
1855                 if (unlikely(rc)) {
1856                         if (rc == -ENOENT)
1857                                 rc = 0;
1858
1859                         break;
1860                 }
1861
1862                 if (name_is_dot_or_dotdot(za->za_name, strlen(za->za_name))) {
1863                         zap_cursor_advance(it->ozi_zc);
1864                         continue;
1865                 }
1866
1867                 strncpy(it->ozi_name, za->za_name, sizeof(it->ozi_name));
1868                 if (za->za_integer_length != 8) {
1869                         rc = -EIO;
1870                         break;
1871                 }
1872
1873                 rc = osd_zap_lookup(dev, it->ozi_zc->zc_zapobj, NULL,
1874                                     za->za_name, za->za_integer_length,
1875                                     sizeof(*zde) / za->za_integer_length, zde);
1876                 if (rc)
1877                         break;
1878
1879                 rc = cb(env, dev, id, it);
1880                 if (rc)
1881                         break;
1882
1883                 zap_cursor_advance(it->ozi_zc);
1884         }
1885         osd_zap_cursor_fini(it->ozi_zc);
1886
1887 out:
1888         OBD_SLAB_FREE_PTR(it, osd_zapit_cachep);
1889         RETURN(rc);
1890 }
1891
1892 static int osd_remove_ml_file(const struct lu_env *env, struct osd_device *dev,
1893                               uint64_t dir, uint64_t id, struct lu_fid *fid,
1894                               char *name)
1895 {
1896         struct osd_thread_info *info = osd_oti_get(env);
1897         struct dt_object *dt;
1898         struct osd_object *obj = NULL;
1899         dmu_tx_t *tx;
1900         sa_handle_t *hdl;
1901         uint64_t nlink;
1902         int rc;
1903
1904         rc = -sa_handle_get(dev->od_os, id, NULL, SA_HDL_PRIVATE, &hdl);
1905         if (rc)
1906                 RETURN(rc);
1907
1908         dt = lu2dt(lu_object_find_slice(env, osd2lu_dev(dev), fid, NULL));
1909         if (IS_ERR(dt))
1910                 RETURN(PTR_ERR(dt));
1911
1912         if (dt) {
1913                 obj = osd_dt_obj(dt);
1914                 down_read(&obj->oo_guard);
1915         }
1916
1917         rc = -sa_lookup(hdl, SA_ZPL_LINKS(dev), &nlink, sizeof(nlink));
1918         if (rc)
1919                 GOTO(out, rc);
1920
1921         if (nlink <= 1) {
1922                 CERROR("%s: multi-link file O/%s/%s/%s has nlink %llu\n",
1923                        osd_name(dev), info->oti_seq_name, info->oti_dir_name,
1924                        name, nlink);
1925                 GOTO(out, rc = 0);
1926         }
1927
1928         tx = dmu_tx_create(dev->od_os);
1929         if (!tx) {
1930                 CERROR("%s: fail to create tx to remove multi-link file!\n",
1931                        osd_name(dev));
1932                 GOTO(out, rc = -ENOMEM);
1933         }
1934
1935         dmu_tx_hold_zap(tx, dir, FALSE, NULL);
1936         rc = -dmu_tx_assign(tx, TXG_WAIT);
1937         if (rc)
1938                 GOTO(abort, rc);
1939
1940         nlink--;
1941         rc = -sa_update(hdl, SA_ZPL_LINKS(dev), &nlink, sizeof(nlink), tx);
1942         if (rc)
1943                 GOTO(abort, rc);
1944
1945         rc = -zap_remove(dev->od_os, dir, name, tx);
1946         if (rc)
1947                 GOTO(abort, rc);
1948
1949         dmu_tx_commit(tx);
1950         GOTO(out, rc);
1951
1952 abort:
1953         dmu_tx_abort(tx);
1954
1955 out:
1956         if (dt) {
1957                 up_read(&obj->oo_guard);
1958                 dt_object_put_nocache(env, dt);
1959         }
1960
1961         sa_handle_destroy(hdl);
1962         RETURN(rc);
1963 }
1964
1965 static int osd_scan_ml_file(const struct lu_env *env, struct osd_device *dev,
1966                             uint64_t dir_oid, struct osd_zap_it *ozi)
1967 {
1968         struct osd_thread_info *info = osd_oti_get(env);
1969         struct lu_fid *fid = &info->oti_fid;
1970         struct ost_id *ostid = &info->oti_ostid;
1971         char name[32];
1972         u64 seq;
1973         int rc = 0;
1974
1975         ENTRY;
1976
1977         rc = osd_get_fid_by_oid(env, dev, ozi->ozi_zde.lzd_reg.zde_dnode, fid);
1978         if (rc)
1979                 RETURN(rc);
1980
1981         seq = fid_seq(fid);
1982         fid_to_ostid(fid, ostid);
1983
1984         snprintf(name, sizeof(name), (fid_seq_is_rsvd(seq) ||
1985                                       fid_seq_is_mdt0(seq)) ? "%llu" : "%llx",
1986                                       fid_seq_is_idif(seq) ? 0 : seq);
1987         if (strcmp(info->oti_seq_name, name) != 0)
1988                 GOTO(fix, rc);
1989
1990         snprintf(name, sizeof(name), "d%d",
1991                 (int)ostid_id(ostid) % OSD_OST_MAP_SIZE);
1992         if (strcmp(info->oti_dir_name, name) != 0)
1993                 GOTO(fix, rc);
1994
1995         snprintf(name, sizeof(name), "%llu", ostid_id(ostid));
1996         if (strcmp(ozi->ozi_name, name) == 0)
1997                 RETURN(0);
1998
1999 fix:
2000         CDEBUG(D_LFSCK, "%s: the file O/%s/%s/%s is corrupted\n",
2001                osd_name(dev), info->oti_seq_name, info->oti_dir_name,
2002                ozi->ozi_name);
2003
2004         rc = osd_remove_ml_file(env, dev, dir_oid,
2005                                 ozi->ozi_zde.lzd_reg.zde_dnode, fid,
2006                                 ozi->ozi_name);
2007         RETURN(rc);
2008 }
2009
2010 static int osd_scan_ml_file_dir(const struct lu_env *env,
2011                                 struct osd_device *dev, uint64_t dir_oid,
2012                                 struct osd_zap_it *ozi)
2013 {
2014         struct osd_thread_info *info = osd_oti_get(env);
2015
2016         if (!S_ISDIR(cpu_to_le16(DTTOIF(ozi->ozi_zde.lzd_reg.zde_type))))
2017                 return 0;
2018
2019         info->oti_dir_name = ozi->ozi_name;
2020         return osd_scan_dir(env, dev, ozi->ozi_zde.lzd_reg.zde_dnode,
2021                             osd_scan_ml_file);
2022 }
2023
2024 static int osd_scan_ml_file_seq(const struct lu_env *env,
2025                                 struct osd_device *dev, uint64_t dir_oid,
2026                                 struct osd_zap_it *ozi)
2027 {
2028         struct osd_thread_info *info = osd_oti_get(env);
2029
2030         if (!S_ISDIR(cpu_to_le16(DTTOIF(ozi->ozi_zde.lzd_reg.zde_type))))
2031                 return 0;
2032
2033         info->oti_seq_name = ozi->ozi_name;
2034         return osd_scan_dir(env, dev, ozi->ozi_zde.lzd_reg.zde_dnode,
2035                             osd_scan_ml_file_dir);
2036 }
2037
2038 static int osd_scan_ml_file_main(const struct lu_env *env,
2039                                  struct osd_device *dev)
2040 {
2041         return osd_scan_dir(env, dev, dev->od_O_id, osd_scan_ml_file_seq);
2042 }