lustre/osd-ldiskfs/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, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2012, 2017, Intel Corporation.
  24  */
  25 /*
  26  * lustre/osd-ldiskfs/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 ldiskfs inode table to feed up layer LFSCK.
  34  *
  35  * Author: Fan Yong <yong.fan@whamcloud.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 <lustre_nodemap.h>
  47
  48 #include "osd_internal.h"
  49 #include "osd_oi.h"
  50 #include "osd_scrub.h"
  51
  52 #define OSD_OTABLE_MAX_HASH             0x00000000ffffffffULL
  53
  54 static inline int osd_scrub_has_window(struct lustre_scrub *scrub,
  55                                        struct osd_otable_cache *ooc)
  56 {
  57         return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
  58 }
  59
  60 /**
  61  * update/insert/delete the specified OI mapping (@fid @id) according to the ops
  62  *
  63  * \retval   1, changed nothing
  64  * \retval   0, changed successfully
  65  * \retval -ve, on error
  66  */
  67 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
  68                                      struct osd_device *dev,
  69                                      const struct lu_fid *fid,
  70                                      const struct osd_inode_id *id,
  71                                      int ops, bool force,
  72                                      enum oi_check_flags flags, bool *exist)
  73 {
  74         handle_t *th;
  75         int       rc;
  76         ENTRY;
  77
  78         if (dev->od_scrub.os_scrub.os_file.sf_param & SP_DRYRUN && !force)
  79                 RETURN(0);
  80
  81         /* DTO_INDEX_INSERT is enough for other two ops:
  82          * delete/update, but save stack. */
  83         th = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
  84                                 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
  85         if (IS_ERR(th)) {
  86                 rc = PTR_ERR(th);
  87                 CDEBUG(D_LFSCK, "%s: fail to start trans for scrub op %d "
  88                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
  89                        PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1,
  90                        rc);
  91                 RETURN(rc);
  92         }
  93
  94         switch (ops) {
  95         case DTO_INDEX_UPDATE:
  96                 rc = osd_oi_update(info, dev, fid, id, th, flags);
  97                 if (unlikely(rc == -ENOENT)) {
  98                         /* Some unlink thread may removed the OI mapping. */
  99                         rc = 1;
 100                 }
 101                 break;
 102         case DTO_INDEX_INSERT:
 103                 rc = osd_oi_insert(info, dev, fid, id, th, flags, exist);
 104                 if (unlikely(rc == -EEXIST)) {
 105                         rc = 1;
 106                         /* XXX: There are trouble things when adding OI
 107                          *      mapping for IGIF object, which may cause
 108                          *      multiple objects to be mapped to the same
 109                          *      IGIF formatted FID. Consider the following
 110                          *      situations:
 111                          *
 112                          *      1) The MDT is upgrading from 1.8 device.
 113                          *      The OI scrub generates IGIF FID1 for the
 114                          *      OBJ1 and adds the OI mapping.
 115                          *
 116                          *      2) For some reason, the OI scrub does not
 117                          *      process all the IGIF objects completely.
 118                          *
 119                          *      3) The MDT is backuped and restored against
 120                          *      this device.
 121                          *
 122                          *      4) When the MDT mounts up, the OI scrub will
 123                          *      try to rebuild the OI files. For some IGIF
 124                          *      object, OBJ2, which was not processed by the
 125                          *      OI scrub before the backup/restore, and the
 126                          *      new generated IGIF formatted FID may be just
 127                          *      the FID1, the same as OBJ1.
 128                          *
 129                          *      Under such case, the OI scrub cannot know how
 130                          *      to generate new FID for the OBJ2.
 131                          *
 132                          *      Currently, we do nothing for that. One possible
 133                          *      solution is to generate new normal FID for the
 134                          *      conflict object.
 135                          *
 136                          *      Anyway, it is rare, only exists in theory. */
 137                 }
 138                 break;
 139         case DTO_INDEX_DELETE:
 140                 rc = osd_oi_delete(info, dev, fid, th, flags);
 141                 if (rc == -ENOENT) {
 142                         /* It is normal that the unlink thread has removed the
 143                          * OI mapping already. */
 144                         rc = 1;
 145                 }
 146                 break;
 147         default:
 148                 LASSERTF(0, "Unexpected ops %d\n", ops);
 149                 break;
 150         }
 151
 152         ldiskfs_journal_stop(th);
 153         if (rc < 0)
 154                 CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d "
 155                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 156                        PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1,
 157                        rc);
 158
 159         RETURN(rc);
 160 }
 161
 162 static int
 163 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
 164                      struct inode *inode, const struct lu_fid *fid)
 165 {
 166         struct filter_fid_old   *ff      = &info->oti_ff;
 167         struct dentry           *dentry  = &info->oti_obj_dentry;
 168         struct lu_fid           *tfid    = &info->oti_fid;
 169         handle_t                *jh;
 170         int                      size    = 0;
 171         int                      rc;
 172         bool                     reset   = false;
 173         ENTRY;
 174
 175         if (dev->od_scrub.os_scrub.os_file.sf_param & SP_DRYRUN)
 176                 RETURN(0);
 177
 178         if (fid_is_idif(fid) && dev->od_index_in_idif == 0) {
 179                 struct ost_id *oi = &info->oti_ostid;
 180
 181                 fid_to_ostid(fid, oi);
 182                 ostid_to_fid(tfid, oi, 0);
 183         } else {
 184                 *tfid = *fid;
 185         }
 186
 187         /* We want the LMA to fit into the 256-byte OST inode, so operate
 188          * as following:
 189          * 1) read old XATTR_NAME_FID and save the parent FID;
 190          * 2) delete the old XATTR_NAME_FID;
 191          * 3) make new LMA and add it;
 192          * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
 193          *
 194          * Making the LMA to fit into the 256-byte OST inode can save time for
 195          * normal osd_check_lma() and for other OI scrub scanning in future.
 196          * So it is worth to make some slow conversion here. */
 197         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 198                                 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
 199         if (IS_ERR(jh)) {
 200                 rc = PTR_ERR(jh);
 201                 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
 202                        DFID": rc = %d\n", osd_name(dev), PFID(tfid), rc);
 203                 RETURN(rc);
 204         }
 205
 206         /* 1) read old XATTR_NAME_FID and save the parent FID */
 207         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 208         if (rc == sizeof(*ff)) {
 209                 /* 2) delete the old XATTR_NAME_FID */
 210                 ll_vfs_dq_init(inode);
 211                 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
 212                 if (rc)
 213                         GOTO(stop, rc);
 214
 215                 reset = true;
 216         } else if (rc != -ENODATA && rc != sizeof(struct filter_fid)) {
 217                 GOTO(stop, rc = -EINVAL);
 218         }
 219
 220         /* 3) make new LMA and add it */
 221         rc = osd_ea_fid_set(info, inode, tfid, LMAC_FID_ON_OST, 0);
 222         if (reset) {
 223                 if (rc)
 224                         /* If failed, we should try to add the old back. */
 225                         size = sizeof(*ff);
 226                 else
 227                         /* The new PFID EA will only contains ::ff_parent */
 228                         size = sizeof(ff->ff_parent);
 229         }
 230
 231         /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
 232         if (size > 0) {
 233                 int rc1;
 234
 235                 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
 236                                       XATTR_CREATE);
 237                 if (rc1 != 0 && rc == 0)
 238                         rc = rc1;
 239         }
 240
 241         GOTO(stop, rc);
 242
 243 stop:
 244         ldiskfs_journal_stop(jh);
 245         if (rc < 0)
 246                 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
 247                        osd_name(dev), PFID(tfid), rc);
 248         return rc;
 249 }
 250
 251 static int
 252 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
 253                        struct osd_idmap_cache *oic, int val)
 254 {
 255         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
 256         struct scrub_file            *sf     = &scrub->os_file;
 257         struct lu_fid                *fid    = &oic->oic_fid;
 258         struct osd_inode_id          *lid    = &oic->oic_lid;
 259         struct osd_inode_id          *lid2   = &info->oti_id;
 260         struct osd_inconsistent_item *oii    = NULL;
 261         struct inode                 *inode  = NULL;
 262         int                           ops    = DTO_INDEX_UPDATE;
 263         int                           rc;
 264         bool                          converted = false;
 265         bool                          exist     = false;
 266         ENTRY;
 267
 268         down_write(&scrub->os_rwsem);
 269         scrub->os_new_checked++;
 270         if (val < 0)
 271                 GOTO(out, rc = val);
 272
 273         if (scrub->os_in_prior)
 274                 oii = list_entry(oic, struct osd_inconsistent_item,
 275                                  oii_cache);
 276
 277         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
 278                 GOTO(out, rc = 0);
 279
 280         if (fid_is_igif(fid))
 281                 sf->sf_items_igif++;
 282
 283         if (val == SCRUB_NEXT_OSTOBJ_OLD) {
 284                 inode = osd_iget(info, dev, lid);
 285                 if (IS_ERR(inode)) {
 286                         rc = PTR_ERR(inode);
 287                         /* Someone removed the inode. */
 288                         if (rc == -ENOENT || rc == -ESTALE)
 289                                 rc = 0;
 290                         GOTO(out, rc);
 291                 }
 292
 293                 /* The inode has been reused as EA inode, ignore it. */
 294                 if (unlikely(osd_is_ea_inode(inode)))
 295                         GOTO(out, rc = 0);
 296
 297                 sf->sf_flags |= SF_UPGRADE;
 298                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
 299                 dev->od_check_ff = 1;
 300                 rc = osd_scrub_convert_ff(info, dev, inode, fid);
 301                 if (rc != 0)
 302                         GOTO(out, rc);
 303
 304                 converted = true;
 305         }
 306
 307         if ((val == SCRUB_NEXT_NOLMA) &&
 308             (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
 309                 GOTO(out, rc = 0);
 310
 311         if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA)) {
 312                 ops = DTO_INDEX_INSERT;
 313
 314                 goto iget;
 315         }
 316
 317         rc = osd_oi_lookup(info, dev, fid, lid2,
 318                 (val == SCRUB_NEXT_OSTOBJ ||
 319                  val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 320         if (rc != 0) {
 321                 if (rc == -ENOENT)
 322                         ops = DTO_INDEX_INSERT;
 323                 else if (rc != -ESTALE)
 324                         GOTO(out, rc);
 325
 326 iget:
 327                 if (inode == NULL) {
 328                         inode = osd_iget(info, dev, lid);
 329                         if (IS_ERR(inode)) {
 330                                 rc = PTR_ERR(inode);
 331                                 /* Someone removed the inode. */
 332                                 if (rc == -ENOENT || rc == -ESTALE)
 333                                         rc = 0;
 334                                 GOTO(out, rc);
 335                         }
 336
 337                         /* The inode has been reused as EA inode, ignore it. */
 338                         if (unlikely(osd_is_ea_inode(inode)))
 339                                 GOTO(out, rc = 0);
 340                 }
 341
 342                 if (!scrub->os_partial_scan)
 343                         scrub->os_full_speed = 1;
 344
 345                 switch (val) {
 346                 case SCRUB_NEXT_NOLMA:
 347                         sf->sf_flags |= SF_UPGRADE;
 348                         if (!(sf->sf_param & SP_DRYRUN)) {
 349                                 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
 350                                 if (rc != 0)
 351                                         GOTO(out, rc);
 352                         }
 353
 354                         if (!(sf->sf_flags & SF_INCONSISTENT))
 355                                 dev->od_igif_inoi = 0;
 356                         break;
 357                 case SCRUB_NEXT_OSTOBJ:
 358                         sf->sf_flags |= SF_INCONSISTENT;
 359                 case SCRUB_NEXT_OSTOBJ_OLD:
 360                         break;
 361                 default:
 362                         break;
 363                 }
 364         } else if (osd_id_eq(lid, lid2)) {
 365                 if (converted)
 366                         sf->sf_items_updated++;
 367
 368                 GOTO(out, rc = 0);
 369         } else {
 370                 if (!scrub->os_partial_scan)
 371                         scrub->os_full_speed = 1;
 372
 373                 sf->sf_flags |= SF_INCONSISTENT;
 374
 375                 /* XXX: If the device is restored from file-level backup, then
 376                  *      some IGIFs may have been already in OI files, and some
 377                  *      may be not yet. Means upgrading from 1.8 may be partly
 378                  *      processed, but some clients may hold some immobilized
 379                  *      IGIFs, and use them to access related objects. Under
 380                  *      such case, OSD does not know whether an given IGIF has
 381                  *      been processed or to be processed, and it also cannot
 382                  *      generate local ino#/gen# directly from the immobilized
 383                  *      IGIF because of the backup/restore. Then force OSD to
 384                  *      lookup the given IGIF in OI files, and if no entry,
 385                  *      then ask the client to retry after upgrading completed.
 386                  *      No better choice. */
 387                 dev->od_igif_inoi = 1;
 388         }
 389
 390         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
 391                         (val == SCRUB_NEXT_OSTOBJ ||
 392                          val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0,
 393                         &exist);
 394         if (rc == 0) {
 395                 if (scrub->os_in_prior)
 396                         sf->sf_items_updated_prior++;
 397                 else
 398                         sf->sf_items_updated++;
 399
 400                 if (ops == DTO_INDEX_INSERT && val == 0 && !exist) {
 401                         int idx = osd_oi_fid2idx(dev, fid);
 402
 403                         sf->sf_flags |= SF_RECREATED;
 404                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 405                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 406                 }
 407         }
 408
 409         GOTO(out, rc);
 410
 411 out:
 412         if (rc < 0) {
 413                 sf->sf_items_failed++;
 414                 if (sf->sf_pos_first_inconsistent == 0 ||
 415                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 416                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 417         } else {
 418                 rc = 0;
 419         }
 420
 421         /* There may be conflict unlink during the OI scrub,
 422          * if happend, then remove the new added OI mapping. */
 423         if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
 424             unlikely(ldiskfs_test_inode_state(inode,
 425                                               LDISKFS_STATE_LUSTRE_DESTROY)))
 426                 osd_scrub_refresh_mapping(info, dev, fid, lid,
 427                                 DTO_INDEX_DELETE, false,
 428                                 (val == SCRUB_NEXT_OSTOBJ ||
 429                                  val == SCRUB_NEXT_OSTOBJ_OLD) ?
 430                                 OI_KNOWN_ON_OST : 0, NULL);
 431         up_write(&scrub->os_rwsem);
 432
 433         if (inode != NULL && !IS_ERR(inode))
 434                 iput(inode);
 435
 436         if (oii != NULL) {
 437                 spin_lock(&scrub->os_lock);
 438                 if (likely(!list_empty(&oii->oii_list)))
 439                         list_del(&oii->oii_list);
 440                 spin_unlock(&scrub->os_lock);
 441
 442                 OBD_FREE_PTR(oii);
 443         }
 444
 445         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 446 }
 447
 448 static int osd_scrub_prep(const struct lu_env *env, struct osd_device *dev)
 449 {
 450         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
 451         struct ptlrpc_thread *thread = &scrub->os_thread;
 452         struct scrub_file    *sf     = &scrub->os_file;
 453         __u32                 flags  = scrub->os_start_flags;
 454         int                   rc;
 455         bool                  drop_dryrun = false;
 456         ENTRY;
 457
 458         CDEBUG(D_LFSCK, "%s: OI scrub prep, flags = 0x%x\n",
 459                osd_scrub2name(scrub), flags);
 460
 461         down_write(&scrub->os_rwsem);
 462         if (flags & SS_SET_FAILOUT)
 463                 sf->sf_param |= SP_FAILOUT;
 464         else if (flags & SS_CLEAR_FAILOUT)
 465                 sf->sf_param &= ~SP_FAILOUT;
 466
 467         if (flags & SS_SET_DRYRUN) {
 468                 sf->sf_param |= SP_DRYRUN;
 469         } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
 470                 sf->sf_param &= ~SP_DRYRUN;
 471                 drop_dryrun = true;
 472         }
 473
 474         if (flags & SS_RESET)
 475                 scrub_file_reset(scrub,
 476                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
 477
 478         if (flags & SS_AUTO_FULL) {
 479                 scrub->os_full_speed = 1;
 480                 scrub->os_partial_scan = 0;
 481                 sf->sf_flags |= SF_AUTO;
 482         } else if (flags & SS_AUTO_PARTIAL) {
 483                 scrub->os_full_speed = 0;
 484                 scrub->os_partial_scan = 1;
 485                 sf->sf_flags |= SF_AUTO;
 486         } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
 487                                    SF_UPGRADE)) {
 488                 scrub->os_full_speed = 1;
 489                 scrub->os_partial_scan = 0;
 490         } else {
 491                 scrub->os_full_speed = 0;
 492                 scrub->os_partial_scan = 0;
 493         }
 494
 495         spin_lock(&scrub->os_lock);
 496         scrub->os_in_prior = 0;
 497         scrub->os_waiting = 0;
 498         scrub->os_paused = 0;
 499         scrub->os_in_join = 0;
 500         scrub->os_full_scrub = 0;
 501         spin_unlock(&scrub->os_lock);
 502         scrub->os_new_checked = 0;
 503         if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
 504                 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
 505         else if (sf->sf_pos_last_checkpoint != 0)
 506                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 507         else
 508                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 509
 510         scrub->os_pos_current = sf->sf_pos_latest_start;
 511         sf->sf_status = SS_SCANNING;
 512         sf->sf_time_latest_start = ktime_get_real_seconds();
 513         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 514         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
 515         rc = scrub_file_store(env, scrub);
 516         if (rc == 0) {
 517                 spin_lock(&scrub->os_lock);
 518                 thread_set_flags(thread, SVC_RUNNING);
 519                 spin_unlock(&scrub->os_lock);
 520                 wake_up_all(&thread->t_ctl_waitq);
 521         }
 522         up_write(&scrub->os_rwsem);
 523
 524         RETURN(rc);
 525 }
 526
 527 static int osd_scrub_post(const struct lu_env *env, struct osd_device *dev,
 528                           int result)
 529 {
 530         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
 531         struct scrub_file *sf = &scrub->os_file;
 532         int rc;
 533         ENTRY;
 534
 535         CDEBUG(D_LFSCK, "%s: OI scrub post with result = %d\n",
 536                osd_scrub2name(scrub), result);
 537
 538         down_write(&scrub->os_rwsem);
 539         spin_lock(&scrub->os_lock);
 540         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 541         spin_unlock(&scrub->os_lock);
 542         if (scrub->os_new_checked > 0) {
 543                 sf->sf_items_checked += scrub->os_new_checked;
 544                 scrub->os_new_checked = 0;
 545                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 546         }
 547         sf->sf_time_last_checkpoint = ktime_get_real_seconds();
 548         if (result > 0) {
 549                 dev->od_igif_inoi = 1;
 550                 dev->od_check_ff = 0;
 551                 sf->sf_status = SS_COMPLETED;
 552                 if (!(sf->sf_param & SP_DRYRUN)) {
 553                         memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 554                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 555                                           SF_UPGRADE | SF_AUTO);
 556                 }
 557                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 558                 sf->sf_success_count++;
 559         } else if (result == 0) {
 560                 if (scrub->os_paused)
 561                         sf->sf_status = SS_PAUSED;
 562                 else
 563                         sf->sf_status = SS_STOPPED;
 564         } else {
 565                 sf->sf_status = SS_FAILED;
 566         }
 567         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 568                                             scrub->os_time_last_checkpoint);
 569         rc = scrub_file_store(env, scrub);
 570         up_write(&scrub->os_rwsem);
 571
 572         RETURN(rc < 0 ? rc : result);
 573 }
 574
 575 /* iteration engine */
 576
 577 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 578                                    struct osd_device *dev,
 579                                    struct osd_iit_param *param,
 580                                    struct osd_idmap_cache **oic,
 581                                    const bool noslot);
 582
 583 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 584                                    struct osd_device *dev,
 585                                    struct osd_iit_param *param,
 586                                    struct osd_idmap_cache *oic,
 587                                    bool *noslot, int rc);
 588
 589 static int osd_iit_next(struct osd_iit_param *param, __u64 *pos)
 590 {
 591         __u32 offset;
 592
 593 again:
 594         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 595                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 596         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 597                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 598                 return SCRUB_NEXT_BREAK;
 599         }
 600
 601         offset = param->offset++;
 602         if (unlikely(*pos == param->gbase + offset && *pos != param->start)) {
 603                 /* We should NOT find the same object more than once. */
 604                 CERROR("%s: scan the same object multiple times at the pos: "
 605                        "group = %u, base = %u, offset = %u, start = %u\n",
 606                        param->sb->s_id, (__u32)param->bg, param->gbase,
 607                        offset, param->start);
 608                 goto again;
 609         }
 610
 611         *pos = param->gbase + offset;
 612         return 0;
 613 }
 614
 615 /**
 616  * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
 617  * \retval 0: FID-on-MDT
 618  */
 619 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
 620                                       struct osd_device *dev,
 621                                       struct lu_fid *fid)
 622 {
 623         /* XXX: The initial OI scrub will scan the top level /O to generate
 624          *      a small local FLDB according to the <seq>. If the given FID
 625          *      is in the local FLDB, then it is FID-on-OST; otherwise it's
 626          *      quite possible for FID-on-MDT. */
 627         if (dev->od_is_ost)
 628                 return SCRUB_NEXT_OSTOBJ_OLD;
 629         else
 630                 return 0;
 631 }
 632
 633 static int osd_scrub_get_fid(struct osd_thread_info *info,
 634                              struct osd_device *dev, struct inode *inode,
 635                              struct lu_fid *fid, bool scrub)
 636 {
 637         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
 638         int rc;
 639         bool has_lma = false;
 640
 641         rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
 642                          &info->oti_ost_attrs);
 643         if (rc == 0) {
 644                 has_lma = true;
 645                 if (lma->lma_compat & LMAC_NOT_IN_OI ||
 646                     lma->lma_incompat & LMAI_AGENT)
 647                         return SCRUB_NEXT_CONTINUE;
 648
 649                 *fid = lma->lma_self_fid;
 650                 if (!scrub)
 651                         return 0;
 652
 653                 if (lma->lma_compat & LMAC_FID_ON_OST)
 654                         return SCRUB_NEXT_OSTOBJ;
 655
 656                 if (fid_is_idif(fid))
 657                         return SCRUB_NEXT_OSTOBJ_OLD;
 658
 659                 /* For local object. */
 660                 if (fid_is_internal(fid))
 661                         return 0;
 662
 663                 /* For external visible MDT-object with non-normal FID. */
 664                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
 665                         return 0;
 666
 667                 /* For the object with normal FID, it may be MDT-object,
 668                  * or may be 2.4 OST-object, need further distinguish.
 669                  * Fall through to next section. */
 670         }
 671
 672         if (rc == -ENODATA || rc == 0) {
 673                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
 674                 if (rc == 0) {
 675                         if (scrub)
 676                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
 677                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 678                         return rc;
 679                 }
 680
 681                 if (rc > 0) {
 682                         if (!has_lma)
 683                                 /* It is FID-on-OST, but we do not know how
 684                                  * to generate its FID, ignore it directly. */
 685                                 rc = SCRUB_NEXT_CONTINUE;
 686                         else
 687                                 /* It is 2.4 OST-object. */
 688                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 689                         return rc;
 690                 }
 691
 692                 if (rc != -ENODATA)
 693                         return rc;
 694
 695                 if (!has_lma) {
 696                         if (dev->od_scrub.os_scrub.os_convert_igif) {
 697                                 lu_igif_build(fid, inode->i_ino,
 698                                               inode->i_generation);
 699                                 if (scrub)
 700                                         rc = SCRUB_NEXT_NOLMA;
 701                                 else
 702                                         rc = 0;
 703                         } else {
 704                                 /* It may be FID-on-OST, or may be FID for
 705                                  * non-MDT0, anyway, we do not know how to
 706                                  * generate its FID, ignore it directly. */
 707                                 rc = SCRUB_NEXT_CONTINUE;
 708                         }
 709                         return rc;
 710                 }
 711
 712                 /* For OI scrub case only: the object has LMA but has no ff
 713                  * (or ff crashed). It may be MDT-object, may be OST-object
 714                  * with crashed ff. The last check is local FLDB. */
 715                 rc = osd_scrub_check_local_fldb(info, dev, fid);
 716         }
 717
 718         return rc;
 719 }
 720
 721 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 722                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 723                         struct super_block *sb, bool scrub)
 724 {
 725         struct inode *inode;
 726         int           rc;
 727         ENTRY;
 728
 729         /* Not handle the backend root object and agent parent object.
 730          * They are neither visible to namespace nor have OI mappings. */
 731         if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
 732                      pos == osd_remote_parent_ino(dev)))
 733                 RETURN(SCRUB_NEXT_CONTINUE);
 734
 735         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 736         inode = osd_iget(info, dev, lid);
 737         if (IS_ERR(inode)) {
 738                 rc = PTR_ERR(inode);
 739                 /* The inode may be removed after bitmap searching, or the
 740                  * file is new created without inode initialized yet. */
 741                 if (rc == -ENOENT || rc == -ESTALE)
 742                         RETURN(SCRUB_NEXT_CONTINUE);
 743
 744                 CDEBUG(D_LFSCK, "%s: fail to read inode, ino# = %u: "
 745                        "rc = %d\n", osd_dev2name(dev), pos, rc);
 746                 RETURN(rc);
 747         }
 748
 749         /* It is an EA inode, no OI mapping for it, skip it. */
 750         if (osd_is_ea_inode(inode))
 751                 GOTO(put, rc = SCRUB_NEXT_CONTINUE);
 752
 753         if (scrub &&
 754             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 755                 /* Only skip it for the first OI scrub accessing. */
 756                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 757                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
 758         }
 759
 760         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
 761
 762         GOTO(put, rc);
 763
 764 put:
 765         iput(inode);
 766         return rc;
 767 }
 768
 769 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 770                           struct osd_iit_param *param,
 771                           struct osd_idmap_cache **oic, const bool noslot)
 772 {
 773         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
 774         struct ptlrpc_thread *thread = &scrub->os_thread;
 775         struct lu_fid *fid;
 776         struct osd_inode_id *lid;
 777         int rc;
 778
 779         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
 780                 struct l_wait_info lwi;
 781
 782                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
 783                 if (likely(lwi.lwi_timeout > 0))
 784                         l_wait_event(thread->t_ctl_waitq,
 785                                 !list_empty(&scrub->os_inconsistent_items) ||
 786                                 !thread_is_running(thread),
 787                                 &lwi);
 788         }
 789
 790         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 791                 spin_lock(&scrub->os_lock);
 792                 thread_set_flags(thread, SVC_STOPPING);
 793                 spin_unlock(&scrub->os_lock);
 794                 return SCRUB_NEXT_CRASH;
 795         }
 796
 797         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
 798                 return SCRUB_NEXT_FATAL;
 799
 800         if (unlikely(!thread_is_running(thread)))
 801                 return SCRUB_NEXT_EXIT;
 802
 803         if (!list_empty(&scrub->os_inconsistent_items)) {
 804                 spin_lock(&scrub->os_lock);
 805                 if (likely(!list_empty(&scrub->os_inconsistent_items))) {
 806                         struct osd_inconsistent_item *oii;
 807
 808                         oii = list_entry(scrub->os_inconsistent_items.next,
 809                                 struct osd_inconsistent_item, oii_list);
 810                         spin_unlock(&scrub->os_lock);
 811
 812                         *oic = &oii->oii_cache;
 813                         scrub->os_in_prior = 1;
 814
 815                         return 0;
 816                 }
 817                 spin_unlock(&scrub->os_lock);
 818         }
 819
 820         if (noslot)
 821                 return SCRUB_NEXT_WAIT;
 822
 823         rc = osd_iit_next(param, &scrub->os_pos_current);
 824         if (rc != 0)
 825                 return rc;
 826
 827         *oic = &dev->od_scrub.os_oic;
 828         fid = &(*oic)->oic_fid;
 829         lid = &(*oic)->oic_lid;
 830         rc = osd_iit_iget(info, dev, fid, lid,
 831                           scrub->os_pos_current, param->sb, true);
 832         return rc;
 833 }
 834
 835 static int osd_preload_next(struct osd_thread_info *info,
 836                             struct osd_device *dev, struct osd_iit_param *param,
 837                             struct osd_idmap_cache **oic, const bool noslot)
 838 {
 839         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
 840         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
 841         struct ptlrpc_thread *thread = &scrub->os_thread;
 842         int rc;
 843
 844         if (thread_is_running(thread) &&
 845             ooc->ooc_pos_preload >= scrub->os_pos_current)
 846                 return SCRUB_NEXT_EXIT;
 847
 848         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
 849         if (rc)
 850                 return rc;
 851
 852         rc = osd_iit_iget(info, dev,
 853                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
 854                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
 855                           ooc->ooc_pos_preload, param->sb, false);
 856         return rc;
 857 }
 858
 859 static inline int
 860 osd_scrub_wakeup(struct lustre_scrub *scrub, struct osd_otable_it *it)
 861 {
 862         spin_lock(&scrub->os_lock);
 863         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
 864             !list_empty(&scrub->os_inconsistent_items) ||
 865             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
 866                 scrub->os_waiting = 0;
 867         else
 868                 scrub->os_waiting = 1;
 869         spin_unlock(&scrub->os_lock);
 870
 871         return !scrub->os_waiting;
 872 }
 873
 874 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
 875                           struct osd_iit_param *param,
 876                           struct osd_idmap_cache *oic, bool *noslot, int rc)
 877 {
 878         struct l_wait_info lwi = { 0 };
 879         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
 880         struct scrub_file *sf = &scrub->os_file;
 881         struct ptlrpc_thread *thread = &scrub->os_thread;
 882         struct osd_otable_it *it = dev->od_otable_it;
 883         struct osd_otable_cache *ooc = it ? &it->ooi_cache : NULL;
 884
 885         switch (rc) {
 886         case SCRUB_NEXT_NOSCRUB:
 887                 down_write(&scrub->os_rwsem);
 888                 scrub->os_new_checked++;
 889                 sf->sf_items_noscrub++;
 890                 up_write(&scrub->os_rwsem);
 891         case SCRUB_NEXT_CONTINUE:
 892         case SCRUB_NEXT_WAIT:
 893                 goto wait;
 894         }
 895
 896         rc = osd_scrub_check_update(info, dev, oic, rc);
 897         if (rc != 0) {
 898                 scrub->os_in_prior = 0;
 899                 return rc;
 900         }
 901
 902         rc = scrub_checkpoint(info->oti_env, scrub);
 903         if (rc) {
 904                 CDEBUG(D_LFSCK, "%s: fail to checkpoint, pos = %llu: "
 905                        "rc = %d\n", osd_scrub2name(scrub),
 906                        scrub->os_pos_current, rc);
 907                 /* Continue, as long as the scrub itself can go ahead. */
 908         }
 909
 910         if (scrub->os_in_prior) {
 911                 scrub->os_in_prior = 0;
 912                 return 0;
 913         }
 914
 915 wait:
 916         if (it != NULL && it->ooi_waiting && ooc != NULL &&
 917             ooc->ooc_pos_preload < scrub->os_pos_current) {
 918                 spin_lock(&scrub->os_lock);
 919                 it->ooi_waiting = 0;
 920                 wake_up_all(&thread->t_ctl_waitq);
 921                 spin_unlock(&scrub->os_lock);
 922         }
 923
 924         if (rc == SCRUB_NEXT_CONTINUE)
 925                 return 0;
 926
 927         if (scrub->os_full_speed || !ooc || osd_scrub_has_window(scrub, ooc)) {
 928                 *noslot = false;
 929                 return 0;
 930         }
 931
 932         if (it != NULL)
 933                 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
 934                              &lwi);
 935
 936         if (!ooc || osd_scrub_has_window(scrub, ooc))
 937                 *noslot = false;
 938         else
 939                 *noslot = true;
 940         return 0;
 941 }
 942
 943 static int osd_preload_exec(struct osd_thread_info *info,
 944                             struct osd_device *dev, struct osd_iit_param *param,
 945                             struct osd_idmap_cache *oic, bool *noslot, int rc)
 946 {
 947         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
 948
 949         if (rc == 0) {
 950                 ooc->ooc_cached_items++;
 951                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
 952                                         ~OSD_OTABLE_IT_CACHE_MASK;
 953         }
 954         return rc > 0 ? 0 : rc;
 955 }
 956
 957 #define SCRUB_IT_ALL    1
 958 #define SCRUB_IT_CRASH  2
 959
 960 static void osd_scrub_join(const struct lu_env *env, struct osd_device *dev,
 961                            __u32 flags, bool inconsistent)
 962 {
 963         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
 964         struct ptlrpc_thread *thread = &scrub->os_thread;
 965         struct scrub_file    *sf     = &scrub->os_file;
 966         int                   rc;
 967         ENTRY;
 968
 969         LASSERT(!(flags & SS_AUTO_PARTIAL));
 970
 971         down_write(&scrub->os_rwsem);
 972         scrub->os_in_join = 1;
 973         if (flags & SS_SET_FAILOUT)
 974                 sf->sf_param |= SP_FAILOUT;
 975         else if (flags & SS_CLEAR_FAILOUT)
 976                 sf->sf_param &= ~SP_FAILOUT;
 977
 978         if (flags & SS_SET_DRYRUN)
 979                 sf->sf_param |= SP_DRYRUN;
 980         else if (flags & SS_CLEAR_DRYRUN)
 981                 sf->sf_param &= ~SP_DRYRUN;
 982
 983         if (flags & SS_RESET) {
 984                 scrub_file_reset(scrub, LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
 985                         inconsistent ? SF_INCONSISTENT : 0);
 986                 sf->sf_status = SS_SCANNING;
 987         }
 988
 989         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
 990                 scrub->os_full_speed = 1;
 991         else
 992                 scrub->os_full_speed = 0;
 993
 994         if (flags & SS_AUTO_FULL) {
 995                 sf->sf_flags |= SF_AUTO;
 996                 scrub->os_full_speed = 1;
 997         }
 998
 999         scrub->os_new_checked = 0;
1000         if (sf->sf_pos_last_checkpoint != 0)
1001                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1002         else
1003                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1004
1005         scrub->os_pos_current = sf->sf_pos_latest_start;
1006         sf->sf_time_latest_start = ktime_get_real_seconds();
1007         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1008         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1009         rc = scrub_file_store(env, scrub);
1010
1011         spin_lock(&scrub->os_lock);
1012         scrub->os_waiting = 0;
1013         scrub->os_paused = 0;
1014         scrub->os_partial_scan = 0;
1015         scrub->os_in_join = 0;
1016         scrub->os_full_scrub = 0;
1017         spin_unlock(&scrub->os_lock);
1018         wake_up_all(&thread->t_ctl_waitq);
1019         up_write(&scrub->os_rwsem);
1020
1021         CDEBUG(D_LFSCK, "%s: joined in the OI scrub with flag %u: rc = %d\n",
1022                osd_scrub2name(scrub), flags, rc);
1023
1024         EXIT;
1025 }
1026
1027 static int osd_inode_iteration(struct osd_thread_info *info,
1028                                struct osd_device *dev, __u32 max, bool preload)
1029 {
1030         struct lustre_scrub *scrub  = &dev->od_scrub.os_scrub;
1031         struct ptlrpc_thread *thread = &scrub->os_thread;
1032         struct scrub_file *sf = &scrub->os_file;
1033         osd_iit_next_policy next;
1034         osd_iit_exec_policy exec;
1035         __u64 *pos;
1036         __u64 *count;
1037         struct osd_iit_param *param;
1038         struct l_wait_info lwi = { 0 };
1039         __u32 limit;
1040         int rc;
1041         bool noslot = true;
1042         ENTRY;
1043
1044         if (preload)
1045                 goto full;
1046
1047         param = &dev->od_scrub.os_iit_param;
1048         memset(param, 0, sizeof(*param));
1049         param->sb = osd_sb(dev);
1050
1051         while (scrub->os_partial_scan && !scrub->os_in_join) {
1052                 struct osd_idmap_cache *oic = NULL;
1053
1054                 rc = osd_scrub_next(info, dev, param, &oic, noslot);
1055                 switch (rc) {
1056                 case SCRUB_NEXT_EXIT:
1057                         RETURN(0);
1058                 case SCRUB_NEXT_CRASH:
1059                         RETURN(SCRUB_IT_CRASH);
1060                 case SCRUB_NEXT_FATAL:
1061                         RETURN(-EINVAL);
1062                 case SCRUB_NEXT_WAIT: {
1063                         struct kstatfs *ksfs = &info->oti_ksfs;
1064                         __u64 saved_flags;
1065
1066                         if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1067                             unlikely(sf->sf_items_updated_prior == 0))
1068                                 goto wait;
1069
1070                         if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1071                             scrub->os_full_scrub) {
1072                                 osd_scrub_join(info->oti_env, dev,
1073                                                SS_AUTO_FULL | SS_RESET, true);
1074                                 goto full;
1075                         }
1076
1077                         rc = param->sb->s_op->statfs(param->sb->s_root, ksfs);
1078                         if (rc == 0) {
1079                                 __u64 used = ksfs->f_files - ksfs->f_ffree;
1080
1081                                 do_div(used, sf->sf_items_updated_prior);
1082                                 /* If we hit too much inconsistent OI
1083                                  * mappings during the partial scan,
1084                                  * then scan the device completely. */
1085                                 if (used < dev->od_full_scrub_ratio) {
1086                                         osd_scrub_join(info->oti_env, dev,
1087                                                 SS_AUTO_FULL | SS_RESET, true);
1088                                         goto full;
1089                                 }
1090                         }
1091
1092 wait:
1093                         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1094                             cfs_fail_val > 0)
1095                                 continue;
1096
1097                         saved_flags = sf->sf_flags;
1098                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1099                                           SF_UPGRADE | SF_AUTO);
1100                         sf->sf_status = SS_COMPLETED;
1101                         l_wait_event(thread->t_ctl_waitq,
1102                                      !thread_is_running(thread) ||
1103                                      !scrub->os_partial_scan ||
1104                                      scrub->os_in_join ||
1105                                      !list_empty(&scrub->os_inconsistent_items),
1106                                      &lwi);
1107                         sf->sf_flags = saved_flags;
1108                         sf->sf_status = SS_SCANNING;
1109
1110                         if (unlikely(!thread_is_running(thread)))
1111                                 RETURN(0);
1112
1113                         if (!scrub->os_partial_scan || scrub->os_in_join)
1114                                 goto full;
1115
1116                         continue;
1117                 }
1118                 default:
1119                         LASSERTF(rc == 0, "rc = %d\n", rc);
1120
1121                         osd_scrub_exec(info, dev, param, oic, &noslot, rc);
1122                         break;
1123                 }
1124         }
1125
1126 full:
1127         if (!preload) {
1128                 l_wait_event(thread->t_ctl_waitq,
1129                              !thread_is_running(thread) || !scrub->os_in_join,
1130                              &lwi);
1131
1132                 if (unlikely(!thread_is_running(thread)))
1133                         RETURN(0);
1134         }
1135
1136         noslot = false;
1137         if (!preload) {
1138                 next = osd_scrub_next;
1139                 exec = osd_scrub_exec;
1140                 pos = &scrub->os_pos_current;
1141                 count = &scrub->os_new_checked;
1142                 param->start = *pos;
1143                 param->bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param->sb);
1144                 param->offset =
1145                         (*pos - 1) % LDISKFS_INODES_PER_GROUP(param->sb);
1146                 param->gbase =
1147                         1 + param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
1148         } else {
1149                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1150
1151                 next = osd_preload_next;
1152                 exec = osd_preload_exec;
1153                 pos = &ooc->ooc_pos_preload;
1154                 count = &ooc->ooc_cached_items;
1155                 param = &dev->od_otable_it->ooi_iit_param;
1156         }
1157
1158         rc = 0;
1159         limit = le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count);
1160         while (*pos <= limit && *count < max) {
1161                 struct ldiskfs_group_desc *desc;
1162                 bool next_group = false;
1163
1164                 desc = ldiskfs_get_group_desc(param->sb, param->bg, NULL);
1165                 if (!desc)
1166                         RETURN(-EIO);
1167
1168                 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1169                         next_group = true;
1170                         goto next_group;
1171                 }
1172
1173                 param->bitmap = ldiskfs_read_inode_bitmap(param->sb, param->bg);
1174                 if (!param->bitmap) {
1175                         CERROR("%s: fail to read bitmap for %u, "
1176                                "scrub will stop, urgent mode\n",
1177                                osd_scrub2name(scrub), (__u32)param->bg);
1178                         RETURN(-EIO);
1179                 }
1180
1181                 do {
1182                         struct osd_idmap_cache *oic = NULL;
1183
1184                         if (param->offset +
1185                                 ldiskfs_itable_unused_count(param->sb, desc) >=
1186                             LDISKFS_INODES_PER_GROUP(param->sb)) {
1187                                 next_group = true;
1188                                 goto next_group;
1189                         }
1190
1191                         rc = next(info, dev, param, &oic, noslot);
1192                         switch (rc) {
1193                         case SCRUB_NEXT_BREAK:
1194                                 next_group = true;
1195                                 goto next_group;
1196                         case SCRUB_NEXT_EXIT:
1197                                 brelse(param->bitmap);
1198                                 RETURN(0);
1199                         case SCRUB_NEXT_CRASH:
1200                                 brelse(param->bitmap);
1201                                 RETURN(SCRUB_IT_CRASH);
1202                         case SCRUB_NEXT_FATAL:
1203                                 brelse(param->bitmap);
1204                                 RETURN(-EINVAL);
1205                         }
1206
1207                         rc = exec(info, dev, param, oic, &noslot, rc);
1208                 } while (!rc && *pos <= limit && *count < max);
1209
1210 next_group:
1211                 if (param->bitmap) {
1212                         brelse(param->bitmap);
1213                         param->bitmap = NULL;
1214                 }
1215
1216                 if (rc < 0)
1217                         GOTO(out, rc);
1218
1219                 if (next_group) {
1220                         param->bg++;
1221                         param->offset = 0;
1222                         param->gbase = 1 +
1223                                 param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
1224                         *pos = param->gbase;
1225                         param->start = *pos;
1226                 }
1227         }
1228
1229         if (*pos > limit)
1230                 RETURN(SCRUB_IT_ALL);
1231
1232 out:
1233         RETURN(rc);
1234 }
1235
1236 static int osd_otable_it_preload(const struct lu_env *env,
1237                                  struct osd_otable_it *it)
1238 {
1239         struct osd_device *dev = it->ooi_dev;
1240         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
1241         struct osd_otable_cache *ooc   = &it->ooi_cache;
1242         int                      rc;
1243         ENTRY;
1244
1245         rc = osd_inode_iteration(osd_oti_get(env), dev,
1246                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1247         if (rc == SCRUB_IT_ALL)
1248                 it->ooi_all_cached = 1;
1249
1250         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1251                 spin_lock(&scrub->os_lock);
1252                 scrub->os_waiting = 0;
1253                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1254                 spin_unlock(&scrub->os_lock);
1255         }
1256
1257         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1258 }
1259
1260 static int osd_scrub_main(void *args)
1261 {
1262         struct lu_env env;
1263         struct osd_device *dev = (struct osd_device *)args;
1264         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
1265         struct ptlrpc_thread *thread = &scrub->os_thread;
1266         int rc;
1267         ENTRY;
1268
1269         rc = lu_env_init(&env, LCT_LOCAL | LCT_DT_THREAD);
1270         if (rc != 0) {
1271                 CDEBUG(D_LFSCK, "%s: OI scrub fail to init env: rc = %d\n",
1272                        osd_scrub2name(scrub), rc);
1273                 GOTO(noenv, rc);
1274         }
1275
1276         rc = osd_scrub_prep(&env, dev);
1277         if (rc != 0) {
1278                 CDEBUG(D_LFSCK, "%s: OI scrub fail to scrub prep: rc = %d\n",
1279                        osd_scrub2name(scrub), rc);
1280                 GOTO(out, rc);
1281         }
1282
1283         if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1284                 struct l_wait_info lwi = { 0 };
1285                 struct osd_otable_it *it = dev->od_otable_it;
1286                 struct osd_otable_cache *ooc = &it->ooi_cache;
1287
1288                 l_wait_event(thread->t_ctl_waitq,
1289                              it->ooi_user_ready || !thread_is_running(thread),
1290                              &lwi);
1291                 if (unlikely(!thread_is_running(thread)))
1292                         GOTO(post, rc = 0);
1293
1294                 scrub->os_pos_current = ooc->ooc_pos_preload;
1295         }
1296
1297         CDEBUG(D_LFSCK, "%s: OI scrub start, flags = 0x%x, pos = %llu\n",
1298                osd_scrub2name(scrub), scrub->os_start_flags,
1299                scrub->os_pos_current);
1300
1301         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1302         if (unlikely(rc == SCRUB_IT_CRASH)) {
1303                 spin_lock(&scrub->os_lock);
1304                 thread_set_flags(&scrub->os_thread, SVC_STOPPING);
1305                 spin_unlock(&scrub->os_lock);
1306                 GOTO(out, rc = -EINVAL);
1307         }
1308
1309         GOTO(post, rc);
1310
1311 post:
1312         rc = osd_scrub_post(&env, dev, rc);
1313         CDEBUG(D_LFSCK, "%s: OI scrub: stop, pos = %llu: rc = %d\n",
1314                osd_scrub2name(scrub), scrub->os_pos_current, rc);
1315
1316 out:
1317         while (!list_empty(&scrub->os_inconsistent_items)) {
1318                 struct osd_inconsistent_item *oii;
1319
1320                 oii = list_entry(scrub->os_inconsistent_items.next,
1321                                  struct osd_inconsistent_item, oii_list);
1322                 list_del_init(&oii->oii_list);
1323                 OBD_FREE_PTR(oii);
1324         }
1325         lu_env_fini(&env);
1326
1327 noenv:
1328         spin_lock(&scrub->os_lock);
1329         thread_set_flags(thread, SVC_STOPPED);
1330         wake_up_all(&thread->t_ctl_waitq);
1331         spin_unlock(&scrub->os_lock);
1332         return rc;
1333 }
1334
1335 /* initial OI scrub */
1336
1337 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1338                          struct dentry *, filldir_t filldir);
1339
1340 #ifdef HAVE_FILLDIR_USE_CTX
1341 static int osd_ios_varfid_fill(struct dir_context *buf, const char *name,
1342                                int namelen, loff_t offset, __u64 ino,
1343                                unsigned d_type);
1344 static int osd_ios_lf_fill(struct dir_context *buf, const char *name,
1345                            int namelen, loff_t offset, __u64 ino,
1346                            unsigned d_type);
1347 static int osd_ios_dl_fill(struct dir_context *buf, const char *name,
1348                            int namelen, loff_t offset, __u64 ino,
1349                            unsigned d_type);
1350 static int osd_ios_uld_fill(struct dir_context *buf, const char *name,
1351                             int namelen, loff_t offset, __u64 ino,
1352                             unsigned d_type);
1353 #else
1354 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1355                                loff_t offset, __u64 ino, unsigned d_type);
1356 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1357                            loff_t offset, __u64 ino, unsigned d_type);
1358 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1359                            loff_t offset, __u64 ino, unsigned d_type);
1360 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1361                             loff_t offset, __u64 ino, unsigned d_type);
1362 #endif
1363
1364 static int
1365 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1366                      struct dentry *dentry, filldir_t filldir);
1367 static int
1368 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1369                   struct dentry *dentry, filldir_t filldir);
1370
1371 static int
1372 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1373                      struct dentry *dentry, filldir_t filldir);
1374
1375 struct osd_lf_map {
1376         char            *olm_name;
1377         struct lu_fid    olm_fid;
1378         __u16            olm_flags;
1379         __u16            olm_namelen;
1380         scandir_t        olm_scandir;
1381         filldir_t        olm_filldir;
1382 };
1383
1384 /* Add the new introduced local files in the list in the future. */
1385 static const struct osd_lf_map osd_lf_maps[] = {
1386         /* CATALOGS */
1387         {
1388                 .olm_name       = CATLIST,
1389                 .olm_fid        = {
1390                         .f_seq  = FID_SEQ_LOCAL_FILE,
1391                         .f_oid  = LLOG_CATALOGS_OID,
1392                 },
1393                 .olm_flags      = OLF_SHOW_NAME,
1394                 .olm_namelen    = sizeof(CATLIST) - 1,
1395         },
1396
1397         /* CONFIGS */
1398         {
1399                 .olm_name       = MOUNT_CONFIGS_DIR,
1400                 .olm_fid        = {
1401                         .f_seq  = FID_SEQ_LOCAL_FILE,
1402                         .f_oid  = MGS_CONFIGS_OID,
1403                 },
1404                 .olm_flags      = OLF_SCAN_SUBITEMS,
1405                 .olm_namelen    = sizeof(MOUNT_CONFIGS_DIR) - 1,
1406                 .olm_scandir    = osd_ios_general_scan,
1407                 .olm_filldir    = osd_ios_varfid_fill,
1408         },
1409
1410         /* NIDTBL_VERSIONS */
1411         {
1412                 .olm_name       = MGS_NIDTBL_DIR,
1413                 .olm_flags      = OLF_SCAN_SUBITEMS,
1414                 .olm_namelen    = sizeof(MGS_NIDTBL_DIR) - 1,
1415                 .olm_scandir    = osd_ios_general_scan,
1416                 .olm_filldir    = osd_ios_varfid_fill,
1417         },
1418
1419         /* PENDING */
1420         {
1421                 .olm_name       = "PENDING",
1422                 .olm_namelen    = sizeof("PENDING") - 1,
1423         },
1424
1425         /* ROOT */
1426         {
1427                 .olm_name       = "ROOT",
1428                 .olm_fid        = {
1429                         .f_seq  = FID_SEQ_ROOT,
1430                         .f_oid  = FID_OID_ROOT,
1431                 },
1432                 .olm_flags      = OLF_SCAN_SUBITEMS | OLF_HIDE_FID,
1433                 .olm_namelen    = sizeof("ROOT") - 1,
1434                 .olm_scandir    = osd_ios_ROOT_scan,
1435         },
1436
1437         /* changelog_catalog */
1438         {
1439                 .olm_name       = CHANGELOG_CATALOG,
1440                 .olm_namelen    = sizeof(CHANGELOG_CATALOG) - 1,
1441         },
1442
1443         /* changelog_users */
1444         {
1445                 .olm_name       = CHANGELOG_USERS,
1446                 .olm_namelen    = sizeof(CHANGELOG_USERS) - 1,
1447         },
1448
1449         /* fld */
1450         {
1451                 .olm_name       = "fld",
1452                 .olm_fid        = {
1453                         .f_seq  = FID_SEQ_LOCAL_FILE,
1454                         .f_oid  = FLD_INDEX_OID,
1455                 },
1456                 .olm_flags      = OLF_SHOW_NAME,
1457                 .olm_namelen    = sizeof("fld") - 1,
1458         },
1459
1460         /* last_rcvd */
1461         {
1462                 .olm_name       = LAST_RCVD,
1463                 .olm_fid        = {
1464                         .f_seq  = FID_SEQ_LOCAL_FILE,
1465                         .f_oid  = LAST_RECV_OID,
1466                 },
1467                 .olm_flags      = OLF_SHOW_NAME,
1468                 .olm_namelen    = sizeof(LAST_RCVD) - 1,
1469         },
1470
1471         /* reply_data */
1472         {
1473                 .olm_name       = REPLY_DATA,
1474                 .olm_fid        = {
1475                         .f_seq  = FID_SEQ_LOCAL_FILE,
1476                         .f_oid  = REPLY_DATA_OID,
1477                 },
1478                 .olm_flags      = OLF_SHOW_NAME,
1479                 .olm_namelen    = sizeof(REPLY_DATA) - 1,
1480         },
1481
1482         /* lov_objid */
1483         {
1484                 .olm_name       = LOV_OBJID,
1485                 .olm_fid        = {
1486                         .f_seq  = FID_SEQ_LOCAL_FILE,
1487                         .f_oid  = MDD_LOV_OBJ_OID,
1488                 },
1489                 .olm_flags      = OLF_SHOW_NAME,
1490                 .olm_namelen    = sizeof(LOV_OBJID) - 1,
1491         },
1492
1493         /* lov_objseq */
1494         {
1495                 .olm_name       = LOV_OBJSEQ,
1496                 .olm_fid        = {
1497                         .f_seq  = FID_SEQ_LOCAL_FILE,
1498                         .f_oid  = MDD_LOV_OBJ_OSEQ,
1499                 },
1500                 .olm_flags      = OLF_SHOW_NAME,
1501                 .olm_namelen    = sizeof(LOV_OBJSEQ) - 1,
1502         },
1503
1504         /* quota_master */
1505         {
1506                 .olm_name       = QMT_DIR,
1507                 .olm_flags      = OLF_SCAN_SUBITEMS,
1508                 .olm_namelen    = sizeof(QMT_DIR) - 1,
1509                 .olm_scandir    = osd_ios_general_scan,
1510                 .olm_filldir    = osd_ios_varfid_fill,
1511         },
1512
1513         /* quota_slave */
1514         {
1515                 .olm_name       = QSD_DIR,
1516                 .olm_flags      = OLF_SCAN_SUBITEMS,
1517                 .olm_namelen    = sizeof(QSD_DIR) - 1,
1518                 .olm_scandir    = osd_ios_general_scan,
1519                 .olm_filldir    = osd_ios_varfid_fill,
1520         },
1521
1522         /* seq_ctl */
1523         {
1524                 .olm_name       = "seq_ctl",
1525                 .olm_fid        = {
1526                         .f_seq  = FID_SEQ_LOCAL_FILE,
1527                         .f_oid  = FID_SEQ_CTL_OID,
1528                 },
1529                 .olm_flags      = OLF_SHOW_NAME,
1530                 .olm_namelen    = sizeof("seq_ctl") - 1,
1531         },
1532
1533         /* seq_srv */
1534         {
1535                 .olm_name       = "seq_srv",
1536                 .olm_fid        = {
1537                         .f_seq  = FID_SEQ_LOCAL_FILE,
1538                         .f_oid  = FID_SEQ_SRV_OID,
1539                 },
1540                 .olm_flags      = OLF_SHOW_NAME,
1541                 .olm_namelen    = sizeof("seq_srv") - 1,
1542         },
1543
1544         /* health_check */
1545         {
1546                 .olm_name       = HEALTH_CHECK,
1547                 .olm_fid        = {
1548                         .f_seq  = FID_SEQ_LOCAL_FILE,
1549                         .f_oid  = OFD_HEALTH_CHECK_OID,
1550                 },
1551                 .olm_flags      = OLF_SHOW_NAME,
1552                 .olm_namelen    = sizeof(HEALTH_CHECK) - 1,
1553         },
1554
1555         /* LFSCK */
1556         {
1557                 .olm_name       = LFSCK_DIR,
1558                 .olm_flags      = OLF_SCAN_SUBITEMS,
1559                 .olm_namelen    = sizeof(LFSCK_DIR) - 1,
1560                 .olm_scandir    = osd_ios_general_scan,
1561                 .olm_filldir    = osd_ios_varfid_fill,
1562         },
1563
1564         /* lfsck_bookmark */
1565         {
1566                 .olm_name       = LFSCK_BOOKMARK,
1567                 .olm_namelen    = sizeof(LFSCK_BOOKMARK) - 1,
1568         },
1569
1570         /* lfsck_layout */
1571         {
1572                 .olm_name       = LFSCK_LAYOUT,
1573                 .olm_namelen    = sizeof(LFSCK_LAYOUT) - 1,
1574         },
1575
1576         /* lfsck_namespace */
1577         {
1578                 .olm_name       = LFSCK_NAMESPACE,
1579                 .olm_namelen    = sizeof(LFSCK_NAMESPACE) - 1,
1580         },
1581
1582         /* OBJECTS, upgrade from old device */
1583         {
1584                 .olm_name       = OBJECTS,
1585                 .olm_flags      = OLF_SCAN_SUBITEMS,
1586                 .olm_namelen    = sizeof(OBJECTS) - 1,
1587                 .olm_scandir    = osd_ios_OBJECTS_scan,
1588         },
1589
1590         /* lquota_v2.user, upgrade from old device */
1591         {
1592                 .olm_name       = "lquota_v2.user",
1593                 .olm_namelen    = sizeof("lquota_v2.user") - 1,
1594         },
1595
1596         /* lquota_v2.group, upgrade from old device */
1597         {
1598                 .olm_name       = "lquota_v2.group",
1599                 .olm_namelen    = sizeof("lquota_v2.group") - 1,
1600         },
1601
1602         /* LAST_GROUP, upgrade from old device */
1603         {
1604                 .olm_name       = "LAST_GROUP",
1605                 .olm_fid        = {
1606                         .f_seq  = FID_SEQ_LOCAL_FILE,
1607                         .f_oid  = OFD_LAST_GROUP_OID,
1608                 },
1609                 .olm_flags      = OLF_SHOW_NAME,
1610                 .olm_namelen    = sizeof("LAST_GROUP") - 1,
1611         },
1612
1613         /* committed batchid for cross-MDT operation */
1614         {
1615                 .olm_name       = "BATCHID",
1616                 .olm_fid        = {
1617                         .f_seq  = FID_SEQ_LOCAL_FILE,
1618                         .f_oid  = BATCHID_COMMITTED_OID,
1619                 },
1620                 .olm_flags      = OLF_SHOW_NAME,
1621                 .olm_namelen    = sizeof("BATCHID") - 1,
1622         },
1623
1624         /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1625          * and f_oid = index for their log files.  See lu_update_log{_dir}_fid()
1626          * for more details. */
1627
1628         /* update_log */
1629         {
1630                 .olm_name       = "update_log",
1631                 .olm_fid        = {
1632                         .f_seq  = FID_SEQ_UPDATE_LOG,
1633                 },
1634                 .olm_flags      = OLF_SHOW_NAME | OLF_IDX_IN_FID,
1635                 .olm_namelen    = sizeof("update_log") - 1,
1636         },
1637
1638         /* update_log_dir */
1639         {
1640                 .olm_name       = "update_log_dir",
1641                 .olm_fid        = {
1642                         .f_seq  = FID_SEQ_UPDATE_LOG_DIR,
1643                 },
1644                 .olm_flags      = OLF_SHOW_NAME | OLF_SCAN_SUBITEMS |
1645                                   OLF_IDX_IN_FID,
1646                 .olm_namelen    = sizeof("update_log_dir") - 1,
1647                 .olm_scandir    = osd_ios_general_scan,
1648                 .olm_filldir    = osd_ios_uld_fill,
1649         },
1650
1651         /* lost+found */
1652         {
1653                 .olm_name       = "lost+found",
1654                 .olm_fid        = {
1655                         .f_seq  = FID_SEQ_LOCAL_FILE,
1656                         .f_oid  = OSD_LPF_OID,
1657                 },
1658                 .olm_flags      = OLF_SCAN_SUBITEMS,
1659                 .olm_namelen    = sizeof("lost+found") - 1,
1660                 .olm_scandir    = osd_ios_general_scan,
1661                 .olm_filldir    = osd_ios_lf_fill,
1662         },
1663
1664         /* hsm_actions */
1665         {
1666                 .olm_name       = HSM_ACTIONS,
1667         },
1668
1669         /* nodemap */
1670         {
1671                 .olm_name       = LUSTRE_NODEMAP_NAME,
1672         },
1673
1674         /* index_backup */
1675         {
1676                 .olm_name       = INDEX_BACKUP_DIR,
1677                 .olm_fid        = {
1678                         .f_seq  = FID_SEQ_LOCAL_FILE,
1679                         .f_oid  = INDEX_BACKUP_OID,
1680                 },
1681                 .olm_flags      = OLF_SCAN_SUBITEMS | OLF_NOT_BACKUP,
1682                 .olm_namelen    = sizeof(INDEX_BACKUP_DIR) - 1,
1683                 .olm_scandir    = osd_ios_general_scan,
1684                 .olm_filldir    = osd_ios_varfid_fill,
1685         },
1686
1687         {
1688                 .olm_name       = NULL
1689         }
1690 };
1691
1692 /* Add the new introduced files under .lustre/ in the list in the future. */
1693 static const struct osd_lf_map osd_dl_maps[] = {
1694         /* .lustre/fid */
1695         {
1696                 .olm_name       = "fid",
1697                 .olm_fid        = {
1698                         .f_seq  = FID_SEQ_DOT_LUSTRE,
1699                         .f_oid  = FID_OID_DOT_LUSTRE_OBF,
1700                 },
1701                 .olm_namelen    = sizeof("fid") - 1,
1702         },
1703
1704         /* .lustre/lost+found */
1705         {
1706                 .olm_name       = "lost+found",
1707                 .olm_fid        = {
1708                         .f_seq  = FID_SEQ_DOT_LUSTRE,
1709                         .f_oid  = FID_OID_DOT_LUSTRE_LPF,
1710                 },
1711                 .olm_namelen    = sizeof("lost+found") - 1,
1712         },
1713
1714         {
1715                 .olm_name       = NULL
1716         }
1717 };
1718
1719 struct osd_ios_item {
1720         struct list_head oii_list;
1721         struct dentry   *oii_dentry;
1722         scandir_t        oii_scandir;
1723         filldir_t        oii_filldir;
1724 };
1725
1726 struct osd_ios_filldir_buf {
1727 #ifdef HAVE_DIR_CONTEXT
1728         /* please keep it as first member */
1729         struct dir_context       ctx;
1730 #endif
1731         struct osd_thread_info  *oifb_info;
1732         struct osd_device       *oifb_dev;
1733         struct dentry           *oifb_dentry;
1734         int                      oifb_items;
1735 };
1736
1737 static inline struct dentry *
1738 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1739 {
1740         struct dentry *dentry;
1741
1742         dentry = ll_lookup_one_len(name, parent, namelen);
1743         if (IS_ERR(dentry)) {
1744                 int rc = PTR_ERR(dentry);
1745
1746                 if (rc != -ENOENT)
1747                         CERROR("Fail to find %.*s in %.*s (%lu/%u): rc = %d\n",
1748                                namelen, name, parent->d_name.len,
1749                                parent->d_name.name, parent->d_inode->i_ino,
1750                                parent->d_inode->i_generation, rc);
1751
1752                 return dentry;
1753         }
1754
1755         if (dentry->d_inode == NULL) {
1756                 dput(dentry);
1757                 return ERR_PTR(-ENOENT);
1758         }
1759
1760         return dentry;
1761 }
1762
1763 static int
1764 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1765                  scandir_t scandir, filldir_t filldir)
1766 {
1767         struct osd_ios_item *item;
1768         ENTRY;
1769
1770         OBD_ALLOC_PTR(item);
1771         if (item == NULL)
1772                 RETURN(-ENOMEM);
1773
1774         INIT_LIST_HEAD(&item->oii_list);
1775         item->oii_dentry = dget(dentry);
1776         item->oii_scandir = scandir;
1777         item->oii_filldir = filldir;
1778         list_add_tail(&item->oii_list, &dev->od_ios_list);
1779
1780         RETURN(0);
1781 }
1782
1783 static bool osd_index_need_recreate(const struct lu_env *env,
1784                                     struct osd_device *dev, struct inode *inode)
1785 {
1786         struct osd_directory *iam = &osd_oti_get(env)->oti_iam;
1787         struct iam_container *bag = &iam->od_container;
1788         int rc;
1789         ENTRY;
1790
1791         rc = iam_container_init(bag, &iam->od_descr, inode);
1792         if (rc)
1793                 RETURN(true);
1794
1795         rc = iam_container_setup(bag);
1796         iam_container_fini(bag);
1797         if (rc)
1798                 RETURN(true);
1799
1800         RETURN(false);
1801 }
1802
1803 static void osd_ios_index_register(const struct lu_env *env,
1804                                    struct osd_device *osd,
1805                                    const struct lu_fid *fid,
1806                                    struct inode *inode)
1807 {
1808         struct osd_directory *iam = &osd_oti_get(env)->oti_iam;
1809         struct iam_container *bag = &iam->od_container;
1810         struct super_block *sb = osd_sb(osd);
1811         struct iam_descr *descr;
1812         __u32 keysize = 0;
1813         __u32 recsize = 0;
1814         int rc;
1815         ENTRY;
1816
1817         /* Index must be a regular file. */
1818         if (!S_ISREG(inode->i_mode))
1819                 RETURN_EXIT;
1820
1821         /* Index's size must be block aligned. */
1822         if (inode->i_size < sb->s_blocksize ||
1823             (inode->i_size & (sb->s_blocksize - 1)) != 0)
1824                 RETURN_EXIT;
1825
1826         iam_container_init(bag, &iam->od_descr, inode);
1827         rc = iam_container_setup(bag);
1828         if (rc)
1829                 GOTO(fini, rc = 1);
1830
1831         descr = bag->ic_descr;
1832         /* May be regular file with IAM_LFIX_ROOT_MAGIC matched
1833          * coincidentally, or corrupted index object, skip it. */
1834         if (descr->id_ptr_size != 4)
1835                 GOTO(fini, rc = 1);
1836
1837         keysize = descr->id_key_size;
1838         recsize = descr->id_rec_size;
1839         rc = osd_index_register(osd, fid, keysize, recsize);
1840
1841         GOTO(fini, rc);
1842
1843 fini:
1844         iam_container_fini(bag);
1845         if (!rc)
1846                 CDEBUG(D_LFSCK, "%s: index object "DFID" (%u/%u) registered\n",
1847                        osd_name(osd), PFID(fid), keysize, recsize);
1848 }
1849
1850 static void osd_index_restore(const struct lu_env *env, struct osd_device *dev,
1851                               struct lustre_index_restore_unit *liru,
1852                               void *buf, int bufsize)
1853 {
1854         struct osd_thread_info *info = osd_oti_get(env);
1855         struct osd_inode_id *id = &info->oti_id;
1856         struct lu_fid *tgt_fid = &liru->liru_cfid;
1857         struct inode *bak_inode = NULL;
1858         struct ldiskfs_dir_entry_2 *de = NULL;
1859         struct buffer_head *bh = NULL;
1860         struct dentry *dentry;
1861         char *name = buf;
1862         struct lu_fid bak_fid;
1863         int rc;
1864         ENTRY;
1865
1866         lustre_fid2lbx(name, tgt_fid, bufsize);
1867         dentry = osd_child_dentry_by_inode(env, dev->od_index_backup_inode,
1868                                            name, strlen(name));
1869         bh = osd_ldiskfs_find_entry(dev->od_index_backup_inode,
1870                                     &dentry->d_name, &de, NULL, NULL);
1871         if (IS_ERR(bh))
1872                 GOTO(log, rc = PTR_ERR(bh));
1873
1874         osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
1875         brelse(bh);
1876         bak_inode = osd_iget_fid(info, dev, id, &bak_fid);
1877         if (IS_ERR(bak_inode))
1878                 GOTO(log, rc = PTR_ERR(bak_inode));
1879
1880         iput(bak_inode);
1881         /* The OI mapping for index may be invalid, since it will be
1882          * re-created, not update the OI mapping, just cache it in RAM. */
1883         osd_id_gen(id, liru->liru_clid, OSD_OII_NOGEN);
1884         osd_add_oi_cache(info, dev, id, tgt_fid);
1885         rc = lustre_index_restore(env, &dev->od_dt_dev, &liru->liru_pfid,
1886                                   tgt_fid, &bak_fid, liru->liru_name,
1887                                   &dev->od_index_backup_list, &dev->od_lock,
1888                                   buf, bufsize);
1889         GOTO(log, rc);
1890
1891 log:
1892         CDEBUG(D_WARNING, "%s: restore index '%s' with "DFID": rc = %d\n",
1893                osd_name(dev), liru->liru_name, PFID(tgt_fid), rc);
1894 }
1895
1896 /**
1897  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1898  *
1899  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1900  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1901  * reference the inode, or fixed if it is missing or references another inode.
1902  */
1903 static int
1904 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1905                  struct inode *parent, struct inode *inode,
1906                  const struct lu_fid *fid, const char *name,
1907                  int namelen, int flags)
1908 {
1909         struct lustre_mdt_attrs *lma    = &info->oti_ost_attrs.loa_lma;
1910         struct osd_inode_id     *id     = &info->oti_id;
1911         struct osd_inode_id     *id2    = &info->oti_id2;
1912         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
1913         struct scrub_file       *sf     = &scrub->os_file;
1914         struct lu_fid            tfid;
1915         int                      rc;
1916         ENTRY;
1917
1918         rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
1919                          &info->oti_ost_attrs);
1920         if (rc != 0 && rc != -ENODATA) {
1921                 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1922                        "rc = %d\n", osd_name(dev), rc);
1923
1924                 RETURN(rc);
1925         }
1926
1927         osd_id_gen(id, inode->i_ino, inode->i_generation);
1928         if (rc == -ENODATA) {
1929                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
1930                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1931                 } else {
1932                         tfid = *fid;
1933                         if (flags & OLF_IDX_IN_FID) {
1934                                 LASSERT(dev->od_index >= 0);
1935
1936                                 tfid.f_oid = dev->od_index;
1937                         }
1938                 }
1939                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1940                 if (rc != 0) {
1941                         CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1942                               "scrub: rc = %d\n", osd_name(dev), rc);
1943
1944                         RETURN(rc);
1945                 }
1946         } else {
1947                 if (lma->lma_compat & LMAC_NOT_IN_OI)
1948                         RETURN(0);
1949
1950                 tfid = lma->lma_self_fid;
1951                 if (lma->lma_compat & LMAC_IDX_BACKUP &&
1952                     osd_index_need_recreate(info->oti_env, dev, inode)) {
1953                         struct lu_fid *pfid = &info->oti_fid3;
1954
1955                         if (parent == osd_sb(dev)->s_root->d_inode) {
1956                                 lu_local_obj_fid(pfid, OSD_FS_ROOT_OID);
1957                         } else {
1958                                 rc = osd_scrub_get_fid(info, dev, parent, pfid,
1959                                                        false);
1960                                 if (rc)
1961                                         RETURN(rc);
1962                         }
1963
1964                         rc = lustre_liru_new(&dev->od_index_restore_list, pfid,
1965                                         &tfid, inode->i_ino, name, namelen);
1966
1967                         RETURN(rc);
1968                 }
1969
1970                 if (!(flags & OLF_NOT_BACKUP))
1971                         osd_ios_index_register(info->oti_env, dev, &tfid,
1972                                                inode);
1973         }
1974
1975         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1976         if (rc != 0) {
1977                 if (rc != -ENOENT)
1978                         RETURN(rc);
1979
1980                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1981                                                DTO_INDEX_INSERT, true, 0, NULL);
1982                 if (rc > 0)
1983                         rc = 0;
1984
1985                 RETURN(rc);
1986         }
1987
1988         if (osd_id_eq_strict(id, id2))
1989                 RETURN(0);
1990
1991         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1992                 scrub_file_reset(scrub, LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1993                                  SF_INCONSISTENT);
1994                 rc = scrub_file_store(info->oti_env, scrub);
1995                 if (rc != 0)
1996                         RETURN(rc);
1997         }
1998
1999         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
2000                                        DTO_INDEX_UPDATE, true, 0, NULL);
2001         if (rc > 0)
2002                 rc = 0;
2003
2004         RETURN(rc);
2005 }
2006
2007 /**
2008  * It scans the /lost+found, and for the OST-object (with filter_fid
2009  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
2010  */
2011 #ifdef HAVE_FILLDIR_USE_CTX
2012 static int osd_ios_lf_fill(struct dir_context *buf,
2013 #else
2014 static int osd_ios_lf_fill(void *buf,
2015 #endif
2016                            const char *name, int namelen,
2017                            loff_t offset, __u64 ino, unsigned d_type)
2018 {
2019         struct osd_ios_filldir_buf *fill_buf =
2020                 (struct osd_ios_filldir_buf *)buf;
2021         struct osd_thread_info     *info     = fill_buf->oifb_info;
2022         struct osd_device          *dev      = fill_buf->oifb_dev;
2023         struct lu_fid              *fid      = &info->oti_fid;
2024         struct osd_scrub           *scrub    = &dev->od_scrub;
2025         struct dentry              *parent   = fill_buf->oifb_dentry;
2026         struct dentry              *child;
2027         struct inode               *dir      = parent->d_inode;
2028         struct inode               *inode;
2029         int                         rc;
2030         ENTRY;
2031
2032         fill_buf->oifb_items++;
2033
2034         /* skip any '.' started names */
2035         if (name[0] == '.')
2036                 RETURN(0);
2037
2038         scrub->os_lf_scanned++;
2039         child = osd_ios_lookup_one_len(name, parent, namelen);
2040         if (IS_ERR(child)) {
2041                 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
2042                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
2043                 RETURN(0);
2044         }
2045
2046         inode = child->d_inode;
2047         if (S_ISDIR(inode->i_mode)) {
2048                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2049                                       osd_ios_lf_fill);
2050                 if (rc != 0)
2051                         CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
2052                               "rc = %d\n", osd_name(dev), namelen, name, rc);
2053                 GOTO(put, rc);
2054         }
2055
2056         if (!S_ISREG(inode->i_mode))
2057                 GOTO(put, rc = 0);
2058
2059         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
2060         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
2061                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
2062                 if (rc == 0) {
2063                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
2064                                "/lost+found.\n", namelen, name, PFID(fid));
2065                         scrub->os_lf_repaired++;
2066                 } else {
2067                         CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
2068                                DFID": rc = %d\n",
2069                                osd_name(dev), namelen, name, PFID(fid), rc);
2070                 }
2071         }
2072
2073         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
2074          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
2075          *      can process them in furtuer. */
2076
2077         GOTO(put, rc);
2078
2079 put:
2080         if (rc < 0)
2081                 scrub->os_lf_failed++;
2082         dput(child);
2083         /* skip the failure to make the scanning to continue. */
2084         return 0;
2085 }
2086
2087 #ifdef HAVE_FILLDIR_USE_CTX
2088 static int osd_ios_varfid_fill(struct dir_context *buf,
2089 #else
2090 static int osd_ios_varfid_fill(void *buf,
2091 #endif
2092                                const char *name, int namelen,
2093                                loff_t offset, __u64 ino, unsigned d_type)
2094 {
2095         struct osd_ios_filldir_buf *fill_buf =
2096                 (struct osd_ios_filldir_buf *)buf;
2097         struct osd_device          *dev      = fill_buf->oifb_dev;
2098         struct dentry              *child;
2099         int                         rc;
2100         ENTRY;
2101
2102         fill_buf->oifb_items++;
2103
2104         /* skip any '.' started names */
2105         if (name[0] == '.')
2106                 RETURN(0);
2107
2108         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2109         if (IS_ERR(child))
2110                 RETURN(PTR_ERR(child));
2111
2112         rc = osd_ios_scan_one(fill_buf->oifb_info, dev,
2113                               fill_buf->oifb_dentry->d_inode, child->d_inode,
2114                               NULL, name, namelen, 0);
2115         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
2116                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2117                                       osd_ios_varfid_fill);
2118         dput(child);
2119
2120         RETURN(rc);
2121 }
2122
2123 #ifdef HAVE_FILLDIR_USE_CTX
2124 static int osd_ios_dl_fill(struct dir_context *buf,
2125 #else
2126 static int osd_ios_dl_fill(void *buf,
2127 #endif
2128                            const char *name, int namelen,
2129                            loff_t offset, __u64 ino, unsigned d_type)
2130 {
2131         struct osd_ios_filldir_buf *fill_buf =
2132                 (struct osd_ios_filldir_buf *)buf;
2133         struct osd_device          *dev      = fill_buf->oifb_dev;
2134         const struct osd_lf_map    *map;
2135         struct dentry              *child;
2136         int                         rc       = 0;
2137         ENTRY;
2138
2139         fill_buf->oifb_items++;
2140
2141         /* skip any '.' started names */
2142         if (name[0] == '.')
2143                 RETURN(0);
2144
2145         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
2146                 if (map->olm_namelen != namelen)
2147                         continue;
2148
2149                 if (strncmp(map->olm_name, name, namelen) == 0)
2150                         break;
2151         }
2152
2153         if (map->olm_name == NULL)
2154                 RETURN(0);
2155
2156         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2157         if (IS_ERR(child))
2158                 RETURN(PTR_ERR(child));
2159
2160         rc = osd_ios_scan_one(fill_buf->oifb_info, dev,
2161                               fill_buf->oifb_dentry->d_inode, child->d_inode,
2162                               &map->olm_fid, name, namelen, map->olm_flags);
2163         dput(child);
2164
2165         RETURN(rc);
2166 }
2167
2168 #ifdef HAVE_FILLDIR_USE_CTX
2169 static int osd_ios_uld_fill(struct dir_context *buf,
2170 #else
2171 static int osd_ios_uld_fill(void *buf,
2172 #endif
2173                             const char *name, int namelen,
2174                             loff_t offset, __u64 ino, unsigned d_type)
2175 {
2176         struct osd_ios_filldir_buf *fill_buf =
2177                 (struct osd_ios_filldir_buf *)buf;
2178         struct dentry              *child;
2179         struct lu_fid               tfid;
2180         int                         rc       = 0;
2181         ENTRY;
2182
2183         fill_buf->oifb_items++;
2184
2185         /* skip any non-DFID format name */
2186         if (name[0] != '[')
2187                 RETURN(0);
2188
2189         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2190         if (IS_ERR(child))
2191                 RETURN(PTR_ERR(child));
2192
2193         /* skip the start '[' */
2194         sscanf(&name[1], SFID, RFID(&tfid));
2195         if (fid_is_sane(&tfid))
2196                 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
2197                                       fill_buf->oifb_dentry->d_inode,
2198                                       child->d_inode, &tfid, name, namelen, 0);
2199         else
2200                 rc = -EIO;
2201         dput(child);
2202
2203         RETURN(rc);
2204 }
2205
2206 #ifdef HAVE_FILLDIR_USE_CTX
2207 static int osd_ios_root_fill(struct dir_context *buf,
2208 #else
2209 static int osd_ios_root_fill(void *buf,
2210 #endif
2211                              const char *name, int namelen,
2212                              loff_t offset, __u64 ino, unsigned d_type)
2213 {
2214         struct osd_ios_filldir_buf *fill_buf =
2215                 (struct osd_ios_filldir_buf *)buf;
2216         struct osd_device          *dev      = fill_buf->oifb_dev;
2217         const struct osd_lf_map    *map;
2218         struct dentry              *child;
2219         int                         rc       = 0;
2220         ENTRY;
2221
2222         fill_buf->oifb_items++;
2223
2224         /* skip any '.' started names */
2225         if (name[0] == '.')
2226                 RETURN(0);
2227
2228         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2229                 if (map->olm_namelen != namelen)
2230                         continue;
2231
2232                 if (strncmp(map->olm_name, name, namelen) == 0)
2233                         break;
2234         }
2235
2236         if (map->olm_name == NULL)
2237                 RETURN(0);
2238
2239         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2240         if (IS_ERR(child))
2241                 RETURN(PTR_ERR(child));
2242
2243         if (!(map->olm_flags & OLF_NO_OI))
2244                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev,
2245                                 fill_buf->oifb_dentry->d_inode, child->d_inode,
2246                                 &map->olm_fid, name, namelen, map->olm_flags);
2247         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2248                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2249                                       map->olm_filldir);
2250         dput(child);
2251
2252         RETURN(rc);
2253 }
2254
2255 static int
2256 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2257                      struct dentry *dentry, filldir_t filldir)
2258 {
2259         struct osd_ios_filldir_buf    buf   = {
2260 #ifdef HAVE_DIR_CONTEXT
2261                                                 .ctx.actor = filldir,
2262 #endif
2263                                                 .oifb_info = info,
2264                                                 .oifb_dev = dev,
2265                                                 .oifb_dentry = dentry };
2266         struct file                  *filp  = &info->oti_file;
2267         struct inode                 *inode = dentry->d_inode;
2268         const struct file_operations *fops  = inode->i_fop;
2269         int                           rc;
2270         ENTRY;
2271
2272         LASSERT(filldir != NULL);
2273
2274         filp->f_pos = 0;
2275         filp->f_path.dentry = dentry;
2276         filp->f_mode = FMODE_64BITHASH;
2277         filp->f_mapping = inode->i_mapping;
2278         filp->f_op = fops;
2279         filp->private_data = NULL;
2280         set_file_inode(filp, inode);
2281
2282         do {
2283                 buf.oifb_items = 0;
2284 #ifdef HAVE_DIR_CONTEXT
2285                 buf.ctx.pos = filp->f_pos;
2286                 rc = fops->iterate(filp, &buf.ctx);
2287                 filp->f_pos = buf.ctx.pos;
2288 #else
2289                 rc = fops->readdir(filp, &buf, filldir);
2290 #endif
2291         } while (rc >= 0 && buf.oifb_items > 0 &&
2292                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
2293         fops->release(inode, filp);
2294
2295         RETURN(rc);
2296 }
2297
2298 static int
2299 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2300                   struct dentry *dentry, filldir_t filldir)
2301 {
2302         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2303         struct scrub_file *sf = &scrub->os_file;
2304         struct dentry *child;
2305         int rc;
2306         ENTRY;
2307
2308         /* It is existing MDT0 device. We only allow the case of object without
2309          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2310          * can generate IGIF mode FID for the object and related OI mapping. If
2311          * it is on other MDTs, then becuase file-level backup/restore, related
2312          * OI mapping may be invalid already, we do not know which is the right
2313          * FID for the object. We only allow IGIF objects to reside on the MDT0.
2314          *
2315          * XXX: For the case of object on non-MDT0 device with neither LMA nor
2316          *      "fid" xattr, then something crashed. We cannot re-generate the
2317          *      FID directly, instead, the OI scrub will scan the OI structure
2318          *      and try to re-generate the LMA from the OI mapping. But if the
2319          *      OI mapping crashed or lost also, then we have to give up under
2320          *      double failure cases. */
2321         scrub->os_convert_igif = 1;
2322         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2323                                        strlen(dot_lustre_name));
2324         if (IS_ERR(child)) {
2325                 rc = PTR_ERR(child);
2326                 if (rc == -ENOENT) {
2327                         /* It is 1.8 MDT device. */
2328                         if (!(sf->sf_flags & SF_UPGRADE)) {
2329                                 scrub_file_reset(scrub,
2330                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2331                                         SF_UPGRADE);
2332                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2333                                 rc = scrub_file_store(info->oti_env, scrub);
2334                         } else {
2335                                 rc = 0;
2336                         }
2337                 }
2338         } else {
2339                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2340                  * so the client will get IGIF for the ".lustre" object when
2341                  * the MDT restart.
2342                  *
2343                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2344                  * it does not know whether there are some old clients cached
2345                  * the ".lustre" IGIF during the upgrading. Two choices:
2346                  *
2347                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2348                  *    It will allow the old connected clients to access the
2349                  *    ".lustre" with cached IGIF. But it will cause others
2350                  *    on the MDT failed to check "fid_is_dot_lustre()".
2351                  *
2352                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2353                  *    for ".lustre" in spite of whether there are some clients
2354                  *    cached the ".lustre" IGIF or not. It enables the check
2355                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
2356                  *    that the old connected clients cannot access the ".lustre"
2357                  *    with the cached IGIF.
2358                  *
2359                  * Usually, it is rare case for the old connected clients
2360                  * to access the ".lustre" with cached IGIF. So we prefer
2361                  * to the solution 2). */
2362                 rc = osd_ios_scan_one(info, dev, dentry->d_inode,
2363                                       child->d_inode, &LU_DOT_LUSTRE_FID,
2364                                       dot_lustre_name,
2365                                       strlen(dot_lustre_name), 0);
2366                 if (rc == 0)
2367                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2368                                               osd_ios_dl_fill);
2369                 dput(child);
2370         }
2371
2372         RETURN(rc);
2373 }
2374
2375 static int
2376 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2377                      struct dentry *dentry, filldir_t filldir)
2378 {
2379         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2380         struct scrub_file *sf = &scrub->os_file;
2381         struct dentry *child;
2382         int rc;
2383         ENTRY;
2384
2385         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2386                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2387                 rc = scrub_file_store(info->oti_env, scrub);
2388                 if (rc != 0)
2389                         RETURN(rc);
2390         }
2391
2392         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2393         if (!IS_ERR(child)) {
2394                 rc = osd_ios_scan_one(info, dev, dentry->d_inode,
2395                                       child->d_inode, NULL, ADMIN_USR,
2396                                       strlen(ADMIN_USR), 0);
2397                 dput(child);
2398         } else {
2399                 rc = PTR_ERR(child);
2400         }
2401
2402         if (rc != 0 && rc != -ENOENT)
2403                 RETURN(rc);
2404
2405         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2406         if (!IS_ERR(child)) {
2407                 rc = osd_ios_scan_one(info, dev, dentry->d_inode,
2408                                       child->d_inode, NULL, ADMIN_GRP,
2409                                       strlen(ADMIN_GRP), 0);
2410                 dput(child);
2411         } else {
2412                 rc = PTR_ERR(child);
2413         }
2414
2415         if (rc == -ENOENT)
2416                 rc = 0;
2417
2418         RETURN(rc);
2419 }
2420
2421 static void osd_initial_OI_scrub(struct osd_thread_info *info,
2422                                  struct osd_device *dev)
2423 {
2424         struct osd_ios_item     *item    = NULL;
2425         scandir_t                scandir = osd_ios_general_scan;
2426         filldir_t                filldir = osd_ios_root_fill;
2427         struct dentry           *dentry  = osd_sb(dev)->s_root;
2428         const struct osd_lf_map *map     = osd_lf_maps;
2429         ENTRY;
2430
2431         /* Lookup IGIF in OI by force for initial OI scrub. */
2432         dev->od_igif_inoi = 1;
2433
2434         while (1) {
2435                 scandir(info, dev, dentry, filldir);
2436                 if (item != NULL) {
2437                         dput(item->oii_dentry);
2438                         OBD_FREE_PTR(item);
2439                 }
2440
2441                 if (list_empty(&dev->od_ios_list))
2442                         break;
2443
2444                 item = list_entry(dev->od_ios_list.next,
2445                                   struct osd_ios_item, oii_list);
2446                 list_del_init(&item->oii_list);
2447
2448                 LASSERT(item->oii_scandir != NULL);
2449                 scandir = item->oii_scandir;
2450                 filldir = item->oii_filldir;
2451                 dentry = item->oii_dentry;
2452         }
2453
2454         /* There maybe the case that the object has been removed, but its OI
2455          * mapping is still in the OI file, such as the "CATALOGS" after MDT
2456          * file-level backup/restore. So here cleanup the stale OI mappings. */
2457         while (map->olm_name != NULL) {
2458                 struct dentry *child;
2459
2460                 if (fid_is_zero(&map->olm_fid)) {
2461                         map++;
2462                         continue;
2463                 }
2464
2465                 child = osd_ios_lookup_one_len(map->olm_name,
2466                                                osd_sb(dev)->s_root,
2467                                                map->olm_namelen);
2468                 if (!IS_ERR(child))
2469                         dput(child);
2470                 else if (PTR_ERR(child) == -ENOENT)
2471                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2472                                                   NULL, DTO_INDEX_DELETE,
2473                                                   true, 0, NULL);
2474                 map++;
2475         }
2476
2477         if (!list_empty(&dev->od_index_restore_list)) {
2478                 char *buf;
2479
2480                 OBD_ALLOC_LARGE(buf, INDEX_BACKUP_BUFSIZE);
2481                 if (!buf)
2482                         CERROR("%s: not enough RAM for rebuild index\n",
2483                                osd_name(dev));
2484
2485                 while (!list_empty(&dev->od_index_restore_list)) {
2486                         struct lustre_index_restore_unit *liru;
2487
2488                         liru = list_entry(dev->od_index_restore_list.next,
2489                                           struct lustre_index_restore_unit,
2490                                           liru_link);
2491                         list_del(&liru->liru_link);
2492                         if (buf)
2493                                 osd_index_restore(info->oti_env, dev, liru,
2494                                                   buf, INDEX_BACKUP_BUFSIZE);
2495                         OBD_FREE(liru, liru->liru_len);
2496                 }
2497
2498                 if (buf)
2499                         OBD_FREE_LARGE(buf, INDEX_BACKUP_BUFSIZE);
2500         }
2501
2502         EXIT;
2503 }
2504
2505 char *osd_lf_fid2name(const struct lu_fid *fid)
2506 {
2507         const struct osd_lf_map *map = osd_lf_maps;
2508
2509         while (map->olm_name != NULL) {
2510                 if (!lu_fid_eq(fid, &map->olm_fid)) {
2511                         map++;
2512                         continue;
2513                 }
2514
2515                 if (map->olm_flags & OLF_SHOW_NAME)
2516                         return map->olm_name;
2517                 else
2518                         return "";
2519         }
2520
2521         return NULL;
2522 }
2523
2524 /* OI scrub start/stop */
2525
2526 int osd_scrub_start(const struct lu_env *env, struct osd_device *dev,
2527                     __u32 flags)
2528 {
2529         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2530         int rc;
2531         ENTRY;
2532
2533         if (dev->od_dt_dev.dd_rdonly)
2534                 RETURN(-EROFS);
2535
2536         /* od_otable_mutex: prevent curcurrent start/stop */
2537         mutex_lock(&dev->od_otable_mutex);
2538         rc = scrub_start(osd_scrub_main, scrub, dev, flags);
2539         if (rc == -EALREADY) {
2540                 rc = 0;
2541                 if ((scrub->os_file.sf_flags & SF_AUTO ||
2542                      scrub->os_partial_scan) &&
2543                     !(flags & SS_AUTO_PARTIAL))
2544                         osd_scrub_join(env, dev, flags, false);
2545         }
2546         mutex_unlock(&dev->od_otable_mutex);
2547
2548         RETURN(rc);
2549 }
2550
2551 static void osd_scrub_stop(struct osd_device *dev)
2552 {
2553         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2554
2555         /* od_otable_mutex: prevent curcurrent start/stop */
2556         mutex_lock(&dev->od_otable_mutex);
2557         scrub->os_paused = 1;
2558         scrub_stop(scrub);
2559         mutex_unlock(&dev->od_otable_mutex);
2560 }
2561
2562 /* OI scrub setup/cleanup */
2563
2564 static const char osd_scrub_name[] = "OI_scrub";
2565
2566 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2567 {
2568         struct osd_thread_info *info = osd_oti_get(env);
2569         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2570         struct lvfs_run_ctxt *ctxt = &dev->od_scrub.os_ctxt;
2571         struct scrub_file *sf = &scrub->os_file;
2572         struct super_block *sb = osd_sb(dev);
2573         struct ldiskfs_super_block *es = LDISKFS_SB(sb)->s_es;
2574         struct lvfs_run_ctxt saved;
2575         struct file *filp;
2576         struct inode *inode;
2577         struct lu_fid *fid = &info->oti_fid;
2578         struct osd_inode_id *id = &info->oti_id;
2579         struct dt_object *obj;
2580         bool dirty = false;
2581         bool restored = false;
2582         int rc = 0;
2583         ENTRY;
2584
2585         memset(&dev->od_scrub, 0, sizeof(struct osd_scrub));
2586         OBD_SET_CTXT_MAGIC(ctxt);
2587         ctxt->pwdmnt = dev->od_mnt;
2588         ctxt->pwd = dev->od_mnt->mnt_root;
2589         ctxt->fs = get_ds();
2590
2591         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2592         init_rwsem(&scrub->os_rwsem);
2593         spin_lock_init(&scrub->os_lock);
2594         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2595         scrub->os_name = osd_name(dev);
2596
2597         push_ctxt(&saved, ctxt);
2598         filp = filp_open(osd_scrub_name, O_RDWR |
2599                          (dev->od_dt_dev.dd_rdonly ? 0 : O_CREAT), 0644);
2600         if (IS_ERR(filp)) {
2601                 pop_ctxt(&saved, ctxt);
2602                 RETURN(PTR_ERR(filp));
2603         }
2604
2605         inode = file_inode(filp);
2606         if (!dev->od_dt_dev.dd_rdonly) {
2607                 /* 'What the @fid is' is not imporatant, because the object
2608                  * has no OI mapping, and only is visible inside the OSD.*/
2609                 lu_igif_build(fid, inode->i_ino, inode->i_generation);
2610                 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2611                 if (rc) {
2612                         filp_close(filp, NULL);
2613                         pop_ctxt(&saved, ctxt);
2614                         RETURN(rc);
2615                 }
2616         }
2617
2618         osd_id_gen(id, inode->i_ino, inode->i_generation);
2619         osd_add_oi_cache(info, dev, id, fid);
2620         filp_close(filp, NULL);
2621         pop_ctxt(&saved, ctxt);
2622
2623         obj = lu2dt(lu_object_find_slice(env, osd2lu_dev(dev), fid, NULL));
2624         if (IS_ERR_OR_NULL(obj))
2625                 RETURN(obj ? PTR_ERR(obj) : -ENOENT);
2626
2627         scrub->os_obj = obj;
2628         rc = scrub_file_load(env, scrub);
2629         if (rc == -ENOENT || rc == -EFAULT) {
2630                 scrub_file_init(scrub, es->s_uuid);
2631                 /* If the "/O" dir does not exist when mount (indicated by
2632                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2633                  * then it is quite probably that the device is a new one,
2634                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2635                  *
2636                  * For the rare case that "/O" and "OI_scrub" both lost on
2637                  * an old device, it can be found and cleared later.
2638                  *
2639                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2640                  * need to check "filter_fid_old" and to convert it to
2641                  * "filter_fid" for each object, and all the IGIF should
2642                  * have their FID mapping in OI files already. */
2643                 if (dev->od_maybe_new && rc == -ENOENT)
2644                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2645                 dirty = true;
2646         } else if (rc < 0) {
2647                 GOTO(cleanup_obj, rc);
2648         } else {
2649                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2650                         struct obd_uuid *old_uuid;
2651                         struct obd_uuid *new_uuid;
2652
2653                         OBD_ALLOC_PTR(old_uuid);
2654                         OBD_ALLOC_PTR(new_uuid);
2655                         if (old_uuid == NULL || new_uuid == NULL) {
2656                                 CERROR("%s: UUID has been changed, but"
2657                                        "failed to allocate RAM for report\n",
2658                                        osd_dev2name(dev));
2659                         } else {
2660                                 class_uuid_unparse(sf->sf_uuid, old_uuid);
2661                                 class_uuid_unparse(es->s_uuid, new_uuid);
2662                                 CDEBUG(D_LFSCK, "%s: UUID has been changed "
2663                                        "from %s to %s\n", osd_dev2name(dev),
2664                                        old_uuid->uuid, new_uuid->uuid);
2665                         }
2666                         scrub_file_reset(scrub, es->s_uuid, SF_INCONSISTENT);
2667                         dirty = true;
2668                         restored = true;
2669                         if (old_uuid != NULL)
2670                                 OBD_FREE_PTR(old_uuid);
2671                         if (new_uuid != NULL)
2672                                 OBD_FREE_PTR(new_uuid);
2673                 } else if (sf->sf_status == SS_SCANNING) {
2674                         sf->sf_status = SS_CRASHED;
2675                         dirty = true;
2676                 }
2677
2678                 if ((sf->sf_oi_count & (sf->sf_oi_count - 1)) != 0) {
2679                         LCONSOLE_WARN("%s: invalid oi count %d, set it to %d\n",
2680                                       osd_dev2name(dev), sf->sf_oi_count,
2681                                       osd_oi_count);
2682                         sf->sf_oi_count = osd_oi_count;
2683                         dirty = true;
2684                 }
2685         }
2686
2687         if (sf->sf_pos_last_checkpoint != 0)
2688                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2689         else
2690                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2691
2692         if (dirty) {
2693                 rc = scrub_file_store(env, scrub);
2694                 if (rc)
2695                         GOTO(cleanup_obj, rc);
2696         }
2697
2698         /* Initialize OI files. */
2699         rc = osd_oi_init(info, dev, restored);
2700         if (rc < 0)
2701                 GOTO(cleanup_obj, rc);
2702
2703         if (!dev->od_dt_dev.dd_rdonly)
2704                 osd_initial_OI_scrub(info, dev);
2705
2706         if (sf->sf_flags & SF_UPGRADE ||
2707             !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2708               sf->sf_success_count > 0)) {
2709                 dev->od_igif_inoi = 0;
2710                 dev->od_check_ff = dev->od_is_ost;
2711         } else {
2712                 dev->od_igif_inoi = 1;
2713                 dev->od_check_ff = 0;
2714         }
2715
2716         if (sf->sf_flags & SF_INCONSISTENT)
2717                 /* The 'od_igif_inoi' will be set under the
2718                  * following cases:
2719                  * 1) new created system, or
2720                  * 2) restored from file-level backup, or
2721                  * 3) the upgrading completed.
2722                  *
2723                  * The 'od_igif_inoi' may be cleared by OI scrub
2724                  * later if found that the system is upgrading. */
2725                 dev->od_igif_inoi = 1;
2726
2727         if (!dev->od_dt_dev.dd_rdonly &&
2728             dev->od_auto_scrub_interval != AS_NEVER &&
2729             ((sf->sf_status == SS_PAUSED) ||
2730              (sf->sf_status == SS_CRASHED &&
2731               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2732                               SF_UPGRADE | SF_AUTO)) ||
2733              (sf->sf_status == SS_INIT &&
2734               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2735                               SF_UPGRADE))))
2736                 rc = osd_scrub_start(env, dev, SS_AUTO_FULL);
2737
2738         if (rc != 0)
2739                 GOTO(cleanup_oi, rc);
2740
2741         /* it is possible that dcache entries may keep objects after they are
2742          * deleted by OSD. While it looks safe this can cause object data to
2743          * stay until umount causing failures in tests calculating free space,
2744          * e.g. replay-ost-single. Since those dcache entries are not used
2745          * anymore let's just free them after use here */
2746         shrink_dcache_sb(sb);
2747
2748         RETURN(0);
2749 cleanup_oi:
2750         osd_oi_fini(info, dev);
2751 cleanup_obj:
2752         dt_object_put_nocache(env, scrub->os_obj);
2753         scrub->os_obj = NULL;
2754
2755         return rc;
2756 }
2757
2758 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2759 {
2760         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2761
2762         LASSERT(dev->od_otable_it == NULL);
2763
2764         if (scrub->os_obj != NULL) {
2765                 osd_scrub_stop(dev);
2766                 dt_object_put_nocache(env, scrub->os_obj);
2767                 scrub->os_obj = NULL;
2768         }
2769         if (dev->od_oi_table != NULL)
2770                 osd_oi_fini(osd_oti_get(env), dev);
2771 }
2772
2773 /* object table based iteration APIs */
2774
2775 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2776                                        struct dt_object *dt, __u32 attr)
2777 {
2778         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2779         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2780         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2781         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2782         struct osd_otable_it   *it;
2783         __u32                   start = 0;
2784         int                     rc;
2785         ENTRY;
2786
2787         /* od_otable_mutex: prevent curcurrent init/fini */
2788         mutex_lock(&dev->od_otable_mutex);
2789         if (dev->od_otable_it != NULL)
2790                 GOTO(out, it = ERR_PTR(-EALREADY));
2791
2792         OBD_ALLOC_PTR(it);
2793         if (it == NULL)
2794                 GOTO(out, it = ERR_PTR(-ENOMEM));
2795
2796         dev->od_otable_it = it;
2797         it->ooi_dev = dev;
2798         it->ooi_cache.ooc_consumer_idx = -1;
2799         if (flags & DOIF_OUTUSED)
2800                 it->ooi_used_outside = 1;
2801
2802         if (flags & DOIF_RESET)
2803                 start |= SS_RESET;
2804
2805         if (valid & DOIV_ERROR_HANDLE) {
2806                 if (flags & DOIF_FAILOUT)
2807                         start |= SS_SET_FAILOUT;
2808                 else
2809                         start |= SS_CLEAR_FAILOUT;
2810         }
2811
2812         if (valid & DOIV_DRYRUN) {
2813                 if (flags & DOIF_DRYRUN)
2814                         start |= SS_SET_DRYRUN;
2815                 else
2816                         start |= SS_CLEAR_DRYRUN;
2817         }
2818
2819         rc = scrub_start(osd_scrub_main, scrub, dev, start & ~SS_AUTO_PARTIAL);
2820         if (rc == -EALREADY) {
2821                 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2822         } else  if (rc < 0) {
2823                 dev->od_otable_it = NULL;
2824                 OBD_FREE_PTR(it);
2825                 it = ERR_PTR(rc);
2826         } else {
2827                 /* We have to start from the begining. */
2828                 it->ooi_cache.ooc_pos_preload =
2829                         LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2830         }
2831
2832         GOTO(out, it);
2833
2834 out:
2835         mutex_unlock(&dev->od_otable_mutex);
2836         return (struct dt_it *)it;
2837 }
2838
2839 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2840 {
2841         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2842         struct osd_device    *dev = it->ooi_dev;
2843
2844         /* od_otable_mutex: prevent curcurrent init/fini */
2845         mutex_lock(&dev->od_otable_mutex);
2846         scrub_stop(&dev->od_scrub.os_scrub);
2847         LASSERT(dev->od_otable_it == it);
2848
2849         dev->od_otable_it = NULL;
2850         mutex_unlock(&dev->od_otable_mutex);
2851         OBD_FREE_PTR(it);
2852 }
2853
2854 static int osd_otable_it_get(const struct lu_env *env,
2855                              struct dt_it *di, const struct dt_key *key)
2856 {
2857         return 0;
2858 }
2859
2860 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2861 {
2862 }
2863
2864 static inline int
2865 osd_otable_it_wakeup(struct lustre_scrub *scrub, struct osd_otable_it *it)
2866 {
2867         spin_lock(&scrub->os_lock);
2868         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2869             scrub->os_waiting ||
2870             !thread_is_running(&scrub->os_thread))
2871                 it->ooi_waiting = 0;
2872         else
2873                 it->ooi_waiting = 1;
2874         spin_unlock(&scrub->os_lock);
2875
2876         return !it->ooi_waiting;
2877 }
2878
2879 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2880 {
2881         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2882         struct osd_device       *dev    = it->ooi_dev;
2883         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2884         struct osd_otable_cache *ooc    = &it->ooi_cache;
2885         struct ptlrpc_thread    *thread = &scrub->os_thread;
2886         struct l_wait_info       lwi    = { 0 };
2887         int                      rc;
2888         ENTRY;
2889
2890         LASSERT(it->ooi_user_ready);
2891
2892 again:
2893         if (!thread_is_running(thread) && !it->ooi_used_outside)
2894                 RETURN(1);
2895
2896         if (ooc->ooc_cached_items > 0) {
2897                 ooc->ooc_cached_items--;
2898                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2899                                         ~OSD_OTABLE_IT_CACHE_MASK;
2900                 RETURN(0);
2901         }
2902
2903         if (it->ooi_all_cached) {
2904                 l_wait_event(thread->t_ctl_waitq,
2905                              !thread_is_running(thread),
2906                              &lwi);
2907                 RETURN(1);
2908         }
2909
2910         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2911                 spin_lock(&scrub->os_lock);
2912                 scrub->os_waiting = 0;
2913                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2914                 spin_unlock(&scrub->os_lock);
2915         }
2916
2917         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2918                 l_wait_event(thread->t_ctl_waitq,
2919                              osd_otable_it_wakeup(scrub, it),
2920                              &lwi);
2921
2922         if (!thread_is_running(thread) && !it->ooi_used_outside)
2923                 RETURN(1);
2924
2925         rc = osd_otable_it_preload(env, it);
2926         if (rc >= 0)
2927                 goto again;
2928
2929         RETURN(rc);
2930 }
2931
2932 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2933                                         const struct dt_it *di)
2934 {
2935         return NULL;
2936 }
2937
2938 static int osd_otable_it_key_size(const struct lu_env *env,
2939                                   const struct dt_it *di)
2940 {
2941         return sizeof(__u64);
2942 }
2943
2944 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2945                              struct dt_rec *rec, __u32 attr)
2946 {
2947         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2948         struct osd_otable_cache *ooc = &it->ooi_cache;
2949
2950         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2951
2952         /* Filter out Invald FID already. */
2953         LASSERTF(fid_is_sane((struct lu_fid *)rec),
2954                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2955                  PFID((struct lu_fid *)rec),
2956                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2957
2958         return 0;
2959 }
2960
2961 static __u64 osd_otable_it_store(const struct lu_env *env,
2962                                  const struct dt_it *di)
2963 {
2964         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2965         struct osd_otable_cache *ooc = &it->ooi_cache;
2966         __u64                    hash;
2967
2968         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2969                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2970         else
2971                 hash = ooc->ooc_pos_preload;
2972         return hash;
2973 }
2974
2975 /**
2976  * Set the OSD layer iteration start position as the specified hash.
2977  */
2978 static int osd_otable_it_load(const struct lu_env *env,
2979                               const struct dt_it *di, __u64 hash)
2980 {
2981         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
2982         struct osd_device       *dev   = it->ooi_dev;
2983         struct osd_otable_cache *ooc   = &it->ooi_cache;
2984         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2985         struct osd_iit_param    *param = &it->ooi_iit_param;
2986         int                      rc;
2987         ENTRY;
2988
2989         /* Forbid to set iteration position after iteration started. */
2990         if (it->ooi_user_ready)
2991                 RETURN(-EPERM);
2992
2993         LASSERT(!scrub->os_partial_scan);
2994
2995         if (hash > OSD_OTABLE_MAX_HASH)
2996                 hash = OSD_OTABLE_MAX_HASH;
2997
2998         /* The hash is the last checkpoint position,
2999          * we will start from the next one. */
3000         ooc->ooc_pos_preload = hash + 1;
3001         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
3002                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
3003
3004         it->ooi_user_ready = 1;
3005         if (!scrub->os_full_speed)
3006                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
3007
3008         memset(param, 0, sizeof(*param));
3009         param->sb = osd_sb(dev);
3010         param->start = ooc->ooc_pos_preload;
3011         param->bg = (ooc->ooc_pos_preload - 1) /
3012                     LDISKFS_INODES_PER_GROUP(param->sb);
3013         param->offset = (ooc->ooc_pos_preload - 1) %
3014                         LDISKFS_INODES_PER_GROUP(param->sb);
3015         param->gbase = 1 + param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
3016
3017         /* Unplug OSD layer iteration by the first next() call. */
3018         rc = osd_otable_it_next(env, (struct dt_it *)it);
3019
3020         RETURN(rc);
3021 }
3022
3023 static int osd_otable_it_key_rec(const struct lu_env *env,
3024                                  const struct dt_it *di, void *key_rec)
3025 {
3026         return 0;
3027 }
3028
3029 const struct dt_index_operations osd_otable_ops = {
3030         .dio_it = {
3031                 .init     = osd_otable_it_init,
3032                 .fini     = osd_otable_it_fini,
3033                 .get      = osd_otable_it_get,
3034                 .put      = osd_otable_it_put,
3035                 .next     = osd_otable_it_next,
3036                 .key      = osd_otable_it_key,
3037                 .key_size = osd_otable_it_key_size,
3038                 .rec      = osd_otable_it_rec,
3039                 .store    = osd_otable_it_store,
3040                 .load     = osd_otable_it_load,
3041                 .key_rec  = osd_otable_it_key_rec,
3042         }
3043 };
3044
3045 /* high priority inconsistent items list APIs */
3046
3047 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL  60
3048
3049 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
3050                    int insert)
3051 {
3052         struct osd_inconsistent_item *oii;
3053         struct osd_scrub *oscrub = &dev->od_scrub;
3054         struct lustre_scrub *lscrub = &oscrub->os_scrub;
3055         struct ptlrpc_thread *thread = &lscrub->os_thread;
3056         int wakeup = 0;
3057         ENTRY;
3058
3059         OBD_ALLOC_PTR(oii);
3060         if (unlikely(oii == NULL))
3061                 RETURN(-ENOMEM);
3062
3063         INIT_LIST_HEAD(&oii->oii_list);
3064         oii->oii_cache = *oic;
3065         oii->oii_insert = insert;
3066
3067         if (lscrub->os_partial_scan) {
3068                 __u64 now = ktime_get_real_seconds();
3069
3070                 /* If there haven't been errors in a long time,
3071                  * decay old count until either the errors are
3072                  * gone or we reach the current interval. */
3073                 while (unlikely(oscrub->os_bad_oimap_count > 0 &&
3074                                 oscrub->os_bad_oimap_time +
3075                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
3076                         oscrub->os_bad_oimap_count >>= 1;
3077                         oscrub->os_bad_oimap_time +=
3078                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
3079                 }
3080
3081                 oscrub->os_bad_oimap_time = now;
3082                 if (++oscrub->os_bad_oimap_count >
3083                     dev->od_full_scrub_threshold_rate)
3084                         lscrub->os_full_scrub = 1;
3085         }
3086
3087         spin_lock(&lscrub->os_lock);
3088         if (unlikely(!thread_is_running(thread))) {
3089                 spin_unlock(&lscrub->os_lock);
3090                 OBD_FREE_PTR(oii);
3091                 RETURN(-EAGAIN);
3092         }
3093
3094         if (list_empty(&lscrub->os_inconsistent_items))
3095                 wakeup = 1;
3096         list_add_tail(&oii->oii_list, &lscrub->os_inconsistent_items);
3097         spin_unlock(&lscrub->os_lock);
3098
3099         if (wakeup != 0)
3100                 wake_up_all(&thread->t_ctl_waitq);
3101
3102         RETURN(0);
3103 }
3104
3105 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
3106                    struct osd_inode_id *id)
3107 {
3108         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
3109         struct osd_inconsistent_item *oii;
3110         ENTRY;
3111
3112         spin_lock(&scrub->os_lock);
3113         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
3114                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
3115                         *id = oii->oii_cache.oic_lid;
3116                         spin_unlock(&scrub->os_lock);
3117                         RETURN(0);
3118                 }
3119         }
3120         spin_unlock(&scrub->os_lock);
3121
3122         RETURN(-ENOENT);
3123 }
3124
3125 void osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
3126 {
3127         struct osd_scrub *scrub = &dev->od_scrub;
3128
3129         scrub_dump(m, &scrub->os_scrub);
3130         seq_printf(m, "lf_scanned: %llu\n"
3131                    "lf_%s: %llu\n"
3132                    "lf_failed: %llu\n",
3133                    scrub->os_lf_scanned,
3134                    scrub->os_scrub.os_file.sf_param & SP_DRYRUN ?
3135                         "inconsistent" : "repaired",
3136                    scrub->os_lf_repaired,
3137                    scrub->os_lf_failed);
3138 }