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_18_23 *ff = &info->oti_ff_old;
 167         struct dentry *dentry = &info->oti_obj_dentry;
 168         struct lu_fid *tfid = &info->oti_fid;
 169         bool fid_18_23 = false;
 170         handle_t *jh;
 171         int size = 0;
 172         int rc;
 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                 dquot_initialize(inode);
 211                 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
 212                 if (rc)
 213                         GOTO(stop, rc);
 214
 215                 fid_18_23 = true;
 216         } else if (rc != -ENODATA && rc < (int)sizeof(struct filter_fid_24_29)) {
 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 (fid_18_23) {
 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, dev->od_uuid, 0);
 476
 477         if (flags & SS_AUTO_FULL) {
 478                 scrub->os_full_speed = 1;
 479                 scrub->os_partial_scan = 0;
 480                 sf->sf_flags |= SF_AUTO;
 481         } else if (flags & SS_AUTO_PARTIAL) {
 482                 scrub->os_full_speed = 0;
 483                 scrub->os_partial_scan = 1;
 484                 sf->sf_flags |= SF_AUTO;
 485         } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
 486                                    SF_UPGRADE)) {
 487                 scrub->os_full_speed = 1;
 488                 scrub->os_partial_scan = 0;
 489         } else {
 490                 scrub->os_full_speed = 0;
 491                 scrub->os_partial_scan = 0;
 492         }
 493
 494         spin_lock(&scrub->os_lock);
 495         scrub->os_in_prior = 0;
 496         scrub->os_waiting = 0;
 497         scrub->os_paused = 0;
 498         scrub->os_in_join = 0;
 499         scrub->os_full_scrub = 0;
 500         spin_unlock(&scrub->os_lock);
 501         scrub->os_new_checked = 0;
 502         if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
 503                 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
 504         else if (sf->sf_pos_last_checkpoint != 0)
 505                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 506         else
 507                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 508
 509         scrub->os_pos_current = sf->sf_pos_latest_start;
 510         sf->sf_status = SS_SCANNING;
 511         sf->sf_time_latest_start = ktime_get_real_seconds();
 512         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 513         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
 514         rc = scrub_file_store(env, scrub);
 515         if (rc == 0) {
 516                 spin_lock(&scrub->os_lock);
 517                 thread_set_flags(thread, SVC_RUNNING);
 518                 spin_unlock(&scrub->os_lock);
 519                 wake_up_all(&thread->t_ctl_waitq);
 520         }
 521         up_write(&scrub->os_rwsem);
 522
 523         RETURN(rc);
 524 }
 525
 526 static int osd_scrub_post(const struct lu_env *env, struct osd_device *dev,
 527                           int result)
 528 {
 529         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
 530         struct scrub_file *sf = &scrub->os_file;
 531         int rc;
 532         ENTRY;
 533
 534         CDEBUG(D_LFSCK, "%s: OI scrub post with result = %d\n",
 535                osd_scrub2name(scrub), result);
 536
 537         down_write(&scrub->os_rwsem);
 538         spin_lock(&scrub->os_lock);
 539         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 540         spin_unlock(&scrub->os_lock);
 541         if (scrub->os_new_checked > 0) {
 542                 sf->sf_items_checked += scrub->os_new_checked;
 543                 scrub->os_new_checked = 0;
 544                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 545         }
 546         sf->sf_time_last_checkpoint = ktime_get_real_seconds();
 547         if (result > 0) {
 548                 dev->od_igif_inoi = 1;
 549                 dev->od_check_ff = 0;
 550                 sf->sf_status = SS_COMPLETED;
 551                 if (!(sf->sf_param & SP_DRYRUN)) {
 552                         memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 553                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 554                                           SF_UPGRADE | SF_AUTO);
 555                 }
 556                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 557                 sf->sf_success_count++;
 558         } else if (result == 0) {
 559                 if (scrub->os_paused)
 560                         sf->sf_status = SS_PAUSED;
 561                 else
 562                         sf->sf_status = SS_STOPPED;
 563         } else {
 564                 sf->sf_status = SS_FAILED;
 565         }
 566         sf->sf_run_time += ktime_get_seconds() -
 567                            scrub->os_time_last_checkpoint;
 568
 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
 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         bool has_lma = false;
 639         int rc;
 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 2.3 or older 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 or newer 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                      is_remote_parent_ino(dev, pos)))
 733                 RETURN(SCRUB_NEXT_CONTINUE);
 734
 735          /* Skip project quota inode since it is greater than s_first_ino. */
 736 #ifdef HAVE_PROJECT_QUOTA
 737         if (ldiskfs_has_feature_project(sb) &&
 738             pos == le32_to_cpu(LDISKFS_SB(sb)->s_es->s_prj_quota_inum))
 739                 RETURN(SCRUB_NEXT_CONTINUE);
 740 #endif
 741
 742         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 743         inode = osd_iget(info, dev, lid);
 744         if (IS_ERR(inode)) {
 745                 rc = PTR_ERR(inode);
 746                 /* The inode may be removed after bitmap searching, or the
 747                  * file is new created without inode initialized yet. */
 748                 if (rc == -ENOENT || rc == -ESTALE)
 749                         RETURN(SCRUB_NEXT_CONTINUE);
 750
 751                 CDEBUG(D_LFSCK, "%s: fail to read inode, ino# = %u: "
 752                        "rc = %d\n", osd_dev2name(dev), pos, rc);
 753                 RETURN(rc);
 754         }
 755
 756         /* It is an EA inode, no OI mapping for it, skip it. */
 757         if (osd_is_ea_inode(inode))
 758                 GOTO(put, rc = SCRUB_NEXT_CONTINUE);
 759
 760         if (scrub &&
 761             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 762                 /* Only skip it for the first OI scrub accessing. */
 763                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 764                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
 765         }
 766
 767         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
 768
 769         GOTO(put, rc);
 770
 771 put:
 772         iput(inode);
 773         return rc;
 774 }
 775
 776 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 777                           struct osd_iit_param *param,
 778                           struct osd_idmap_cache **oic, const bool noslot)
 779 {
 780         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
 781         struct ptlrpc_thread *thread = &scrub->os_thread;
 782         struct lu_fid *fid;
 783         struct osd_inode_id *lid;
 784         int rc;
 785
 786         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0)
 787                 wait_event_idle_timeout(
 788                         thread->t_ctl_waitq,
 789                         !list_empty(&scrub->os_inconsistent_items) ||
 790                         !thread_is_running(thread),
 791                         cfs_time_seconds(cfs_fail_val));
 792
 793         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 794                 spin_lock(&scrub->os_lock);
 795                 thread_set_flags(thread, SVC_STOPPING);
 796                 spin_unlock(&scrub->os_lock);
 797                 return SCRUB_NEXT_CRASH;
 798         }
 799
 800         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
 801                 return SCRUB_NEXT_FATAL;
 802
 803         if (unlikely(!thread_is_running(thread)))
 804                 return SCRUB_NEXT_EXIT;
 805
 806         if (!list_empty(&scrub->os_inconsistent_items)) {
 807                 spin_lock(&scrub->os_lock);
 808                 if (likely(!list_empty(&scrub->os_inconsistent_items))) {
 809                         struct osd_inconsistent_item *oii;
 810
 811                         oii = list_entry(scrub->os_inconsistent_items.next,
 812                                 struct osd_inconsistent_item, oii_list);
 813                         spin_unlock(&scrub->os_lock);
 814
 815                         *oic = &oii->oii_cache;
 816                         scrub->os_in_prior = 1;
 817
 818                         return 0;
 819                 }
 820                 spin_unlock(&scrub->os_lock);
 821         }
 822
 823         if (noslot)
 824                 return SCRUB_NEXT_WAIT;
 825
 826         rc = osd_iit_next(param, &scrub->os_pos_current);
 827         if (rc != 0)
 828                 return rc;
 829
 830         *oic = &dev->od_scrub.os_oic;
 831         fid = &(*oic)->oic_fid;
 832         lid = &(*oic)->oic_lid;
 833         rc = osd_iit_iget(info, dev, fid, lid,
 834                           scrub->os_pos_current, param->sb, true);
 835         return rc;
 836 }
 837
 838 static int osd_preload_next(struct osd_thread_info *info,
 839                             struct osd_device *dev, struct osd_iit_param *param,
 840                             struct osd_idmap_cache **oic, const bool noslot)
 841 {
 842         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
 843         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
 844         struct ptlrpc_thread *thread = &scrub->os_thread;
 845         int rc;
 846
 847         if (thread_is_running(thread) &&
 848             ooc->ooc_pos_preload >= scrub->os_pos_current)
 849                 return SCRUB_NEXT_EXIT;
 850
 851         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
 852         if (rc)
 853                 return rc;
 854
 855         rc = osd_iit_iget(info, dev,
 856                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
 857                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
 858                           ooc->ooc_pos_preload, param->sb, false);
 859         return rc;
 860 }
 861
 862 static inline int
 863 osd_scrub_wakeup(struct lustre_scrub *scrub, struct osd_otable_it *it)
 864 {
 865         spin_lock(&scrub->os_lock);
 866         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
 867             !list_empty(&scrub->os_inconsistent_items) ||
 868             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
 869                 scrub->os_waiting = 0;
 870         else
 871                 scrub->os_waiting = 1;
 872         spin_unlock(&scrub->os_lock);
 873
 874         return !scrub->os_waiting;
 875 }
 876
 877 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
 878                           struct osd_iit_param *param,
 879                           struct osd_idmap_cache *oic, bool *noslot, int rc)
 880 {
 881         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
 882         struct scrub_file *sf = &scrub->os_file;
 883         struct ptlrpc_thread *thread = &scrub->os_thread;
 884         struct osd_otable_it *it = dev->od_otable_it;
 885         struct osd_otable_cache *ooc = it ? &it->ooi_cache : NULL;
 886
 887         switch (rc) {
 888         case SCRUB_NEXT_NOSCRUB:
 889                 down_write(&scrub->os_rwsem);
 890                 scrub->os_new_checked++;
 891                 sf->sf_items_noscrub++;
 892                 up_write(&scrub->os_rwsem);
 893         case SCRUB_NEXT_CONTINUE:
 894         case SCRUB_NEXT_WAIT:
 895                 goto wait;
 896         }
 897
 898         rc = osd_scrub_check_update(info, dev, oic, rc);
 899         if (rc != 0) {
 900                 scrub->os_in_prior = 0;
 901                 return rc;
 902         }
 903
 904         rc = scrub_checkpoint(info->oti_env, scrub);
 905         if (rc) {
 906                 CDEBUG(D_LFSCK, "%s: fail to checkpoint, pos = %llu: "
 907                        "rc = %d\n", osd_scrub2name(scrub),
 908                        scrub->os_pos_current, rc);
 909                 /* Continue, as long as the scrub itself can go ahead. */
 910         }
 911
 912         if (scrub->os_in_prior) {
 913                 scrub->os_in_prior = 0;
 914                 return 0;
 915         }
 916
 917 wait:
 918         if (it != NULL && it->ooi_waiting && ooc != NULL &&
 919             ooc->ooc_pos_preload < scrub->os_pos_current) {
 920                 spin_lock(&scrub->os_lock);
 921                 it->ooi_waiting = 0;
 922                 wake_up_all(&thread->t_ctl_waitq);
 923                 spin_unlock(&scrub->os_lock);
 924         }
 925
 926         if (rc == SCRUB_NEXT_CONTINUE)
 927                 return 0;
 928
 929         if (scrub->os_full_speed || !ooc || osd_scrub_has_window(scrub, ooc)) {
 930                 *noslot = false;
 931                 return 0;
 932         }
 933
 934         if (it != NULL)
 935                 wait_event_idle(thread->t_ctl_waitq,
 936                                 osd_scrub_wakeup(scrub, it));
 937
 938         if (!ooc || osd_scrub_has_window(scrub, ooc))
 939                 *noslot = false;
 940         else
 941                 *noslot = true;
 942         return 0;
 943 }
 944
 945 static int osd_preload_exec(struct osd_thread_info *info,
 946                             struct osd_device *dev, struct osd_iit_param *param,
 947                             struct osd_idmap_cache *oic, bool *noslot, int rc)
 948 {
 949         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
 950
 951         if (rc == 0) {
 952                 ooc->ooc_cached_items++;
 953                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
 954                                         ~OSD_OTABLE_IT_CACHE_MASK;
 955         }
 956         return rc > 0 ? 0 : rc;
 957 }
 958
 959 #define SCRUB_IT_ALL    1
 960 #define SCRUB_IT_CRASH  2
 961
 962 static void osd_scrub_join(const struct lu_env *env, struct osd_device *dev,
 963                            __u32 flags, bool inconsistent)
 964 {
 965         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
 966         struct ptlrpc_thread *thread = &scrub->os_thread;
 967         struct scrub_file    *sf     = &scrub->os_file;
 968         int                   rc;
 969         ENTRY;
 970
 971         LASSERT(!(flags & SS_AUTO_PARTIAL));
 972
 973         down_write(&scrub->os_rwsem);
 974         scrub->os_in_join = 1;
 975         if (flags & SS_SET_FAILOUT)
 976                 sf->sf_param |= SP_FAILOUT;
 977         else if (flags & SS_CLEAR_FAILOUT)
 978                 sf->sf_param &= ~SP_FAILOUT;
 979
 980         if (flags & SS_SET_DRYRUN)
 981                 sf->sf_param |= SP_DRYRUN;
 982         else if (flags & SS_CLEAR_DRYRUN)
 983                 sf->sf_param &= ~SP_DRYRUN;
 984
 985         if (flags & SS_RESET) {
 986                 scrub_file_reset(scrub, dev->od_uuid,
 987                                  inconsistent ? SF_INCONSISTENT : 0);
 988                 sf->sf_status = SS_SCANNING;
 989         }
 990
 991         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
 992                 scrub->os_full_speed = 1;
 993         else
 994                 scrub->os_full_speed = 0;
 995
 996         if (flags & SS_AUTO_FULL) {
 997                 sf->sf_flags |= SF_AUTO;
 998                 scrub->os_full_speed = 1;
 999         }
1000
1001         scrub->os_new_checked = 0;
1002         if (sf->sf_pos_last_checkpoint != 0)
1003                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1004         else
1005                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1006
1007         scrub->os_pos_current = sf->sf_pos_latest_start;
1008         sf->sf_time_latest_start = ktime_get_real_seconds();
1009         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1010         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1011         rc = scrub_file_store(env, scrub);
1012
1013         spin_lock(&scrub->os_lock);
1014         scrub->os_waiting = 0;
1015         scrub->os_paused = 0;
1016         scrub->os_partial_scan = 0;
1017         scrub->os_in_join = 0;
1018         scrub->os_full_scrub = 0;
1019         spin_unlock(&scrub->os_lock);
1020         wake_up_all(&thread->t_ctl_waitq);
1021         up_write(&scrub->os_rwsem);
1022
1023         CDEBUG(D_LFSCK, "%s: joined in the OI scrub with flag %u: rc = %d\n",
1024                osd_scrub2name(scrub), flags, rc);
1025
1026         EXIT;
1027 }
1028
1029 static int osd_inode_iteration(struct osd_thread_info *info,
1030                                struct osd_device *dev, __u32 max, bool preload)
1031 {
1032         struct lustre_scrub *scrub  = &dev->od_scrub.os_scrub;
1033         struct ptlrpc_thread *thread = &scrub->os_thread;
1034         struct scrub_file *sf = &scrub->os_file;
1035         osd_iit_next_policy next;
1036         osd_iit_exec_policy exec;
1037         __u64 *pos;
1038         __u64 *count;
1039         struct osd_iit_param *param;
1040         __u32 limit;
1041         int rc;
1042         bool noslot = true;
1043         ENTRY;
1044
1045         if (preload)
1046                 goto full;
1047
1048         param = &dev->od_scrub.os_iit_param;
1049         memset(param, 0, sizeof(*param));
1050         param->sb = osd_sb(dev);
1051
1052         while (scrub->os_partial_scan && !scrub->os_in_join) {
1053                 struct osd_idmap_cache *oic = NULL;
1054
1055                 rc = osd_scrub_next(info, dev, param, &oic, noslot);
1056                 switch (rc) {
1057                 case SCRUB_NEXT_EXIT:
1058                         RETURN(0);
1059                 case SCRUB_NEXT_CRASH:
1060                         RETURN(SCRUB_IT_CRASH);
1061                 case SCRUB_NEXT_FATAL:
1062                         RETURN(-EINVAL);
1063                 case SCRUB_NEXT_WAIT: {
1064                         struct kstatfs *ksfs = &info->oti_ksfs;
1065                         __u64 saved_flags;
1066
1067                         if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1068                             unlikely(sf->sf_items_updated_prior == 0))
1069                                 goto wait;
1070
1071                         if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1072                             scrub->os_full_scrub) {
1073                                 osd_scrub_join(info->oti_env, dev,
1074                                                SS_AUTO_FULL | SS_RESET, true);
1075                                 goto full;
1076                         }
1077
1078                         rc = param->sb->s_op->statfs(param->sb->s_root, ksfs);
1079                         if (rc == 0) {
1080                                 __u64 used = ksfs->f_files - ksfs->f_ffree;
1081
1082                                 used = div64_u64(used, sf->sf_items_updated_prior);
1083                                 /* If we hit too much inconsistent OI
1084                                  * mappings during the partial scan,
1085                                  * then scan the device completely. */
1086                                 if (used < dev->od_full_scrub_ratio) {
1087                                         osd_scrub_join(info->oti_env, dev,
1088                                                 SS_AUTO_FULL | SS_RESET, true);
1089                                         goto full;
1090                                 }
1091                         }
1092
1093 wait:
1094                         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1095                             cfs_fail_val > 0)
1096                                 continue;
1097
1098                         saved_flags = sf->sf_flags;
1099                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1100                                           SF_UPGRADE | SF_AUTO);
1101                         sf->sf_status = SS_COMPLETED;
1102                         wait_event_idle(
1103                                 thread->t_ctl_waitq,
1104                                 !thread_is_running(thread) ||
1105                                 !scrub->os_partial_scan ||
1106                                 scrub->os_in_join ||
1107                                 !list_empty(&scrub->os_inconsistent_items));
1108                         sf->sf_flags = saved_flags;
1109                         sf->sf_status = SS_SCANNING;
1110
1111                         if (unlikely(!thread_is_running(thread)))
1112                                 RETURN(0);
1113
1114                         if (!scrub->os_partial_scan || scrub->os_in_join)
1115                                 goto full;
1116
1117                         continue;
1118                 }
1119                 default:
1120                         LASSERTF(rc == 0, "rc = %d\n", rc);
1121
1122                         osd_scrub_exec(info, dev, param, oic, &noslot, rc);
1123                         break;
1124                 }
1125         }
1126
1127 full:
1128         if (!preload) {
1129                 wait_event_idle(thread->t_ctl_waitq,
1130                                 !thread_is_running(thread) ||
1131                                 !scrub->os_in_join);
1132
1133                 if (unlikely(!thread_is_running(thread)))
1134                         RETURN(0);
1135         }
1136
1137         noslot = false;
1138         if (!preload) {
1139                 next = osd_scrub_next;
1140                 exec = osd_scrub_exec;
1141                 pos = &scrub->os_pos_current;
1142                 count = &scrub->os_new_checked;
1143                 param->start = *pos;
1144                 param->bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param->sb);
1145                 param->offset =
1146                         (*pos - 1) % LDISKFS_INODES_PER_GROUP(param->sb);
1147                 param->gbase =
1148                         1 + param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
1149         } else {
1150                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1151
1152                 next = osd_preload_next;
1153                 exec = osd_preload_exec;
1154                 pos = &ooc->ooc_pos_preload;
1155                 count = &ooc->ooc_cached_items;
1156                 param = &dev->od_otable_it->ooi_iit_param;
1157         }
1158
1159         rc = 0;
1160         limit = le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count);
1161         while (*pos <= limit && *count < max) {
1162                 struct ldiskfs_group_desc *desc;
1163                 bool next_group = false;
1164
1165                 desc = ldiskfs_get_group_desc(param->sb, param->bg, NULL);
1166                 if (!desc)
1167                         RETURN(-EIO);
1168
1169                 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1170                         next_group = true;
1171                         goto next_group;
1172                 }
1173
1174                 param->bitmap = ldiskfs_read_inode_bitmap(param->sb, param->bg);
1175                 if (!param->bitmap) {
1176                         CERROR("%s: fail to read bitmap for %u, "
1177                                "scrub will stop, urgent mode\n",
1178                                osd_scrub2name(scrub), (__u32)param->bg);
1179                         RETURN(-EIO);
1180                 }
1181
1182                 do {
1183                         struct osd_idmap_cache *oic = NULL;
1184
1185                         if (param->offset +
1186                                 ldiskfs_itable_unused_count(param->sb, desc) >=
1187                             LDISKFS_INODES_PER_GROUP(param->sb)) {
1188                                 next_group = true;
1189                                 goto next_group;
1190                         }
1191
1192                         rc = next(info, dev, param, &oic, noslot);
1193                         switch (rc) {
1194                         case SCRUB_NEXT_BREAK:
1195                                 next_group = true;
1196                                 goto next_group;
1197                         case SCRUB_NEXT_EXIT:
1198                                 brelse(param->bitmap);
1199                                 RETURN(0);
1200                         case SCRUB_NEXT_CRASH:
1201                                 brelse(param->bitmap);
1202                                 RETURN(SCRUB_IT_CRASH);
1203                         case SCRUB_NEXT_FATAL:
1204                                 brelse(param->bitmap);
1205                                 RETURN(-EINVAL);
1206                         }
1207
1208                         rc = exec(info, dev, param, oic, &noslot, rc);
1209                 } while (!rc && *pos <= limit && *count < max);
1210
1211 next_group:
1212                 if (param->bitmap) {
1213                         brelse(param->bitmap);
1214                         param->bitmap = NULL;
1215                 }
1216
1217                 if (rc < 0)
1218                         GOTO(out, rc);
1219
1220                 if (next_group) {
1221                         param->bg++;
1222                         param->offset = 0;
1223                         param->gbase = 1 +
1224                                 param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
1225                         *pos = param->gbase;
1226                         param->start = *pos;
1227                 }
1228         }
1229
1230         if (*pos > limit)
1231                 RETURN(SCRUB_IT_ALL);
1232
1233 out:
1234         RETURN(rc);
1235 }
1236
1237 static int osd_otable_it_preload(const struct lu_env *env,
1238                                  struct osd_otable_it *it)
1239 {
1240         struct osd_device *dev = it->ooi_dev;
1241         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
1242         struct osd_otable_cache *ooc   = &it->ooi_cache;
1243         int                      rc;
1244         ENTRY;
1245
1246         rc = osd_inode_iteration(osd_oti_get(env), dev,
1247                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1248         if (rc == SCRUB_IT_ALL)
1249                 it->ooi_all_cached = 1;
1250
1251         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1252                 spin_lock(&scrub->os_lock);
1253                 scrub->os_waiting = 0;
1254                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1255                 spin_unlock(&scrub->os_lock);
1256         }
1257
1258         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1259 }
1260
1261 static int osd_scrub_main(void *args)
1262 {
1263         struct lu_env env;
1264         struct osd_device *dev = (struct osd_device *)args;
1265         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
1266         struct ptlrpc_thread *thread = &scrub->os_thread;
1267         int rc;
1268         ENTRY;
1269
1270         rc = lu_env_init(&env, LCT_LOCAL | LCT_DT_THREAD);
1271         if (rc != 0) {
1272                 CDEBUG(D_LFSCK, "%s: OI scrub fail to init env: rc = %d\n",
1273                        osd_scrub2name(scrub), rc);
1274                 GOTO(noenv, rc);
1275         }
1276
1277         rc = osd_scrub_prep(&env, dev);
1278         if (rc != 0) {
1279                 CDEBUG(D_LFSCK, "%s: OI scrub fail to scrub prep: rc = %d\n",
1280                        osd_scrub2name(scrub), rc);
1281                 GOTO(out, rc);
1282         }
1283
1284         if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1285                 struct osd_otable_it *it = dev->od_otable_it;
1286                 struct osd_otable_cache *ooc = &it->ooi_cache;
1287
1288                 wait_event_idle(thread->t_ctl_waitq,
1289                                 it->ooi_user_ready ||
1290                                 !thread_is_running(thread));
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       = MDT_ORPHAN_DIR,
1422                 .olm_namelen    = sizeof(MDT_ORPHAN_DIR) - 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         /* please keep it as first member */
1728         struct dir_context       ctx;
1729         struct osd_thread_info  *oifb_info;
1730         struct osd_device       *oifb_dev;
1731         struct dentry           *oifb_dentry;
1732         int                      oifb_items;
1733 };
1734
1735 static int
1736 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1737                  scandir_t scandir, filldir_t filldir)
1738 {
1739         struct osd_ios_item *item;
1740         ENTRY;
1741
1742         OBD_ALLOC_PTR(item);
1743         if (item == NULL)
1744                 RETURN(-ENOMEM);
1745
1746         INIT_LIST_HEAD(&item->oii_list);
1747         item->oii_dentry = dget(dentry);
1748         item->oii_scandir = scandir;
1749         item->oii_filldir = filldir;
1750         list_add_tail(&item->oii_list, &dev->od_ios_list);
1751
1752         RETURN(0);
1753 }
1754
1755 static bool osd_index_need_recreate(const struct lu_env *env,
1756                                     struct osd_device *dev, struct inode *inode)
1757 {
1758         struct osd_directory *iam = &osd_oti_get(env)->oti_iam;
1759         struct iam_container *bag = &iam->od_container;
1760         int rc;
1761         ENTRY;
1762
1763         rc = iam_container_init(bag, &iam->od_descr, inode);
1764         if (rc)
1765                 RETURN(true);
1766
1767         rc = iam_container_setup(bag);
1768         iam_container_fini(bag);
1769         if (rc)
1770                 RETURN(true);
1771
1772         RETURN(false);
1773 }
1774
1775 static void osd_ios_index_register(const struct lu_env *env,
1776                                    struct osd_device *osd,
1777                                    const struct lu_fid *fid,
1778                                    struct inode *inode)
1779 {
1780         struct osd_directory *iam = &osd_oti_get(env)->oti_iam;
1781         struct iam_container *bag = &iam->od_container;
1782         struct super_block *sb = osd_sb(osd);
1783         struct iam_descr *descr;
1784         __u32 keysize = 0;
1785         __u32 recsize = 0;
1786         int rc;
1787         ENTRY;
1788
1789         /* Index must be a regular file. */
1790         if (!S_ISREG(inode->i_mode))
1791                 RETURN_EXIT;
1792
1793         /* Index's size must be block aligned. */
1794         if (inode->i_size < sb->s_blocksize ||
1795             (inode->i_size & (sb->s_blocksize - 1)) != 0)
1796                 RETURN_EXIT;
1797
1798         iam_container_init(bag, &iam->od_descr, inode);
1799         rc = iam_container_setup(bag);
1800         if (rc)
1801                 GOTO(fini, rc = 1);
1802
1803         descr = bag->ic_descr;
1804         /* May be regular file with IAM_LFIX_ROOT_MAGIC matched
1805          * coincidentally, or corrupted index object, skip it. */
1806         if (descr->id_ptr_size != 4)
1807                 GOTO(fini, rc = 1);
1808
1809         keysize = descr->id_key_size;
1810         recsize = descr->id_rec_size;
1811         rc = osd_index_register(osd, fid, keysize, recsize);
1812
1813         GOTO(fini, rc);
1814
1815 fini:
1816         iam_container_fini(bag);
1817         if (!rc)
1818                 CDEBUG(D_LFSCK, "%s: index object "DFID" (%u/%u) registered\n",
1819                        osd_name(osd), PFID(fid), keysize, recsize);
1820 }
1821
1822 static void osd_index_restore(const struct lu_env *env, struct osd_device *dev,
1823                               struct lustre_index_restore_unit *liru,
1824                               void *buf, int bufsize)
1825 {
1826         struct osd_thread_info *info = osd_oti_get(env);
1827         struct osd_inode_id *id = &info->oti_id;
1828         struct lu_fid *tgt_fid = &liru->liru_cfid;
1829         struct inode *bak_inode = NULL;
1830         struct ldiskfs_dir_entry_2 *de = NULL;
1831         struct buffer_head *bh = NULL;
1832         struct dentry *dentry;
1833         char *name = buf;
1834         struct lu_fid bak_fid;
1835         int rc;
1836         ENTRY;
1837
1838         lustre_fid2lbx(name, tgt_fid, bufsize);
1839         dentry = osd_child_dentry_by_inode(env, dev->od_index_backup_inode,
1840                                            name, strlen(name));
1841         bh = osd_ldiskfs_find_entry(dev->od_index_backup_inode,
1842                                     &dentry->d_name, &de, NULL, NULL);
1843         if (IS_ERR(bh))
1844                 GOTO(log, rc = PTR_ERR(bh));
1845
1846         osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
1847         brelse(bh);
1848         bak_inode = osd_iget_fid(info, dev, id, &bak_fid);
1849         if (IS_ERR(bak_inode))
1850                 GOTO(log, rc = PTR_ERR(bak_inode));
1851
1852         iput(bak_inode);
1853         /* The OI mapping for index may be invalid, since it will be
1854          * re-created, not update the OI mapping, just cache it in RAM. */
1855         osd_id_gen(id, liru->liru_clid, OSD_OII_NOGEN);
1856         osd_add_oi_cache(info, dev, id, tgt_fid);
1857         rc = lustre_index_restore(env, &dev->od_dt_dev, &liru->liru_pfid,
1858                                   tgt_fid, &bak_fid, liru->liru_name,
1859                                   &dev->od_index_backup_list, &dev->od_lock,
1860                                   buf, bufsize);
1861         GOTO(log, rc);
1862
1863 log:
1864         CDEBUG(D_WARNING, "%s: restore index '%s' with "DFID": rc = %d\n",
1865                osd_name(dev), liru->liru_name, PFID(tgt_fid), rc);
1866 }
1867
1868 /**
1869  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1870  *
1871  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1872  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1873  * reference the inode, or fixed if it is missing or references another inode.
1874  */
1875 static int
1876 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1877                  struct inode *parent, struct inode *inode,
1878                  const struct lu_fid *fid, const char *name,
1879                  int namelen, int flags)
1880 {
1881         struct lustre_mdt_attrs *lma    = &info->oti_ost_attrs.loa_lma;
1882         struct osd_inode_id     *id     = &info->oti_id;
1883         struct osd_inode_id     *id2    = &info->oti_id2;
1884         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
1885         struct scrub_file       *sf     = &scrub->os_file;
1886         struct lu_fid            tfid;
1887         int                      rc;
1888         ENTRY;
1889
1890         if (!inode) {
1891                 CDEBUG(D_INODE, "%s: child '%.*s' lacks inode: rc = -2\n",
1892                        osd_name(dev), namelen, name);
1893                 RETURN(-ENOENT);
1894         }
1895
1896         rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
1897                          &info->oti_ost_attrs);
1898         if (rc != 0 && rc != -ENODATA) {
1899                 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1900                        "rc = %d\n", osd_name(dev), rc);
1901
1902                 RETURN(rc);
1903         }
1904
1905         osd_id_gen(id, inode->i_ino, inode->i_generation);
1906         if (rc == -ENODATA) {
1907                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
1908                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1909                 } else {
1910                         tfid = *fid;
1911                         if (flags & OLF_IDX_IN_FID) {
1912                                 LASSERT(dev->od_index >= 0);
1913
1914                                 tfid.f_oid = dev->od_index;
1915                         }
1916                 }
1917                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1918                 if (rc != 0) {
1919                         CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1920                               "scrub: rc = %d\n", osd_name(dev), rc);
1921
1922                         RETURN(rc);
1923                 }
1924         } else {
1925                 if (lma->lma_compat & LMAC_NOT_IN_OI)
1926                         RETURN(0);
1927
1928                 tfid = lma->lma_self_fid;
1929                 if (lma->lma_compat & LMAC_IDX_BACKUP &&
1930                     osd_index_need_recreate(info->oti_env, dev, inode)) {
1931                         struct lu_fid *pfid = &info->oti_fid3;
1932
1933                         if (parent == osd_sb(dev)->s_root->d_inode) {
1934                                 lu_local_obj_fid(pfid, OSD_FS_ROOT_OID);
1935                         } else {
1936                                 rc = osd_scrub_get_fid(info, dev, parent, pfid,
1937                                                        false);
1938                                 if (rc)
1939                                         RETURN(rc);
1940                         }
1941
1942                         rc = lustre_liru_new(&dev->od_index_restore_list, pfid,
1943                                         &tfid, inode->i_ino, name, namelen);
1944
1945                         RETURN(rc);
1946                 }
1947
1948                 if (!(flags & OLF_NOT_BACKUP))
1949                         osd_ios_index_register(info->oti_env, dev, &tfid,
1950                                                inode);
1951         }
1952
1953         /* Since this called from iterate_dir() the inode lock will be taken */
1954         rc = osd_oi_lookup(info, dev, &tfid, id2, OI_LOCKED);
1955         if (rc != 0) {
1956                 if (rc != -ENOENT)
1957                         RETURN(rc);
1958
1959                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1960                                                DTO_INDEX_INSERT, true, 0, NULL);
1961                 if (rc > 0)
1962                         rc = 0;
1963
1964                 RETURN(rc);
1965         }
1966
1967         if (osd_id_eq_strict(id, id2))
1968                 RETURN(0);
1969
1970         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1971                 scrub_file_reset(scrub, dev->od_uuid, SF_INCONSISTENT);
1972                 rc = scrub_file_store(info->oti_env, scrub);
1973                 if (rc != 0)
1974                         RETURN(rc);
1975         }
1976
1977         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1978                                        DTO_INDEX_UPDATE, true, 0, NULL);
1979         if (rc > 0)
1980                 rc = 0;
1981
1982         RETURN(rc);
1983 }
1984
1985 /**
1986  * It scans the /lost+found, and for the OST-object (with filter_fid
1987  * or filter_fid_18_23), move them back to its proper /O/<seq>/d<x>.
1988  */
1989 #ifdef HAVE_FILLDIR_USE_CTX
1990 static int osd_ios_lf_fill(struct dir_context *buf,
1991 #else
1992 static int osd_ios_lf_fill(void *buf,
1993 #endif
1994                            const char *name, int namelen,
1995                            loff_t offset, __u64 ino, unsigned d_type)
1996 {
1997         struct osd_ios_filldir_buf *fill_buf =
1998                 (struct osd_ios_filldir_buf *)buf;
1999         struct osd_thread_info     *info     = fill_buf->oifb_info;
2000         struct osd_device          *dev      = fill_buf->oifb_dev;
2001         struct lu_fid              *fid      = &info->oti_fid;
2002         struct osd_scrub           *scrub    = &dev->od_scrub;
2003         struct dentry              *parent   = fill_buf->oifb_dentry;
2004         struct dentry              *child;
2005         struct inode               *dir      = parent->d_inode;
2006         struct inode               *inode;
2007         int                         rc;
2008         ENTRY;
2009
2010         fill_buf->oifb_items++;
2011
2012         /* skip any '.' started names */
2013         if (name[0] == '.')
2014                 RETURN(0);
2015
2016         scrub->os_lf_scanned++;
2017         child = osd_lookup_one_len(dev, name, parent, namelen);
2018         if (IS_ERR(child)) {
2019                 rc = PTR_ERR(child);
2020                 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
2021                       osd_name(dev), namelen, name, rc);
2022                 RETURN(rc);
2023         } else if (!child->d_inode) {
2024                 dput(child);
2025                 CDEBUG(D_INODE, "%s: child '%.*s' lacks inode\n",
2026                        osd_name(dev), namelen, name);
2027                 RETURN(-ENOENT);
2028         }
2029
2030         inode = child->d_inode;
2031         if (S_ISDIR(inode->i_mode)) {
2032                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2033                                       osd_ios_lf_fill);
2034                 if (rc != 0)
2035                         CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
2036                               "rc = %d\n", osd_name(dev), namelen, name, rc);
2037                 GOTO(put, rc);
2038         }
2039
2040         if (!S_ISREG(inode->i_mode))
2041                 GOTO(put, rc = 0);
2042
2043         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
2044         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
2045                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
2046                 if (rc == 0) {
2047                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
2048                                "/lost+found.\n", namelen, name, PFID(fid));
2049                         scrub->os_lf_repaired++;
2050                 } else {
2051                         CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
2052                                DFID": rc = %d\n",
2053                                osd_name(dev), namelen, name, PFID(fid), rc);
2054                 }
2055         }
2056
2057         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
2058          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
2059          *      can process them in furtuer. */
2060
2061         GOTO(put, rc);
2062
2063 put:
2064         if (rc < 0)
2065                 scrub->os_lf_failed++;
2066         dput(child);
2067         /* skip the failure to make the scanning to continue. */
2068         return 0;
2069 }
2070
2071 #ifdef HAVE_FILLDIR_USE_CTX
2072 static int osd_ios_varfid_fill(struct dir_context *buf,
2073 #else
2074 static int osd_ios_varfid_fill(void *buf,
2075 #endif
2076                                const char *name, int namelen,
2077                                loff_t offset, __u64 ino, unsigned d_type)
2078 {
2079         struct osd_ios_filldir_buf *fill_buf =
2080                 (struct osd_ios_filldir_buf *)buf;
2081         struct osd_device          *dev      = fill_buf->oifb_dev;
2082         struct dentry              *child;
2083         int                         rc;
2084         ENTRY;
2085
2086         fill_buf->oifb_items++;
2087
2088         /* skip any '.' started names */
2089         if (name[0] == '.')
2090                 RETURN(0);
2091
2092         child = osd_lookup_one_len(dev, name, fill_buf->oifb_dentry, namelen);
2093         if (IS_ERR(child))
2094                 RETURN(PTR_ERR(child));
2095
2096         rc = osd_ios_scan_one(fill_buf->oifb_info, dev,
2097                               fill_buf->oifb_dentry->d_inode, child->d_inode,
2098                               NULL, name, namelen, 0);
2099         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
2100                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2101                                       osd_ios_varfid_fill);
2102         dput(child);
2103
2104         RETURN(rc);
2105 }
2106
2107 #ifdef HAVE_FILLDIR_USE_CTX
2108 static int osd_ios_dl_fill(struct dir_context *buf,
2109 #else
2110 static int osd_ios_dl_fill(void *buf,
2111 #endif
2112                            const char *name, int namelen,
2113                            loff_t offset, __u64 ino, unsigned d_type)
2114 {
2115         struct osd_ios_filldir_buf *fill_buf =
2116                 (struct osd_ios_filldir_buf *)buf;
2117         struct osd_device          *dev      = fill_buf->oifb_dev;
2118         const struct osd_lf_map    *map;
2119         struct dentry              *child;
2120         int                         rc       = 0;
2121         ENTRY;
2122
2123         fill_buf->oifb_items++;
2124
2125         /* skip any '.' started names */
2126         if (name[0] == '.')
2127                 RETURN(0);
2128
2129         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
2130                 if (map->olm_namelen != namelen)
2131                         continue;
2132
2133                 if (strncmp(map->olm_name, name, namelen) == 0)
2134                         break;
2135         }
2136
2137         if (map->olm_name == NULL)
2138                 RETURN(0);
2139
2140         child = osd_lookup_one_len(dev, name, fill_buf->oifb_dentry, namelen);
2141         if (IS_ERR(child))
2142                 RETURN(PTR_ERR(child));
2143
2144         rc = osd_ios_scan_one(fill_buf->oifb_info, dev,
2145                               fill_buf->oifb_dentry->d_inode, child->d_inode,
2146                               &map->olm_fid, name, namelen, map->olm_flags);
2147         dput(child);
2148
2149         RETURN(rc);
2150 }
2151
2152 #ifdef HAVE_FILLDIR_USE_CTX
2153 static int osd_ios_uld_fill(struct dir_context *buf,
2154 #else
2155 static int osd_ios_uld_fill(void *buf,
2156 #endif
2157                             const char *name, int namelen,
2158                             loff_t offset, __u64 ino, unsigned d_type)
2159 {
2160         struct osd_ios_filldir_buf *fill_buf =
2161                 (struct osd_ios_filldir_buf *)buf;
2162         struct osd_device *dev = fill_buf->oifb_dev;
2163         struct dentry              *child;
2164         struct lu_fid               tfid;
2165         int                         rc       = 0;
2166         ENTRY;
2167
2168         fill_buf->oifb_items++;
2169
2170         /* skip any non-DFID format name */
2171         if (name[0] != '[')
2172                 RETURN(0);
2173
2174         child = osd_lookup_one_len(dev, name, fill_buf->oifb_dentry, namelen);
2175         if (IS_ERR(child))
2176                 RETURN(PTR_ERR(child));
2177
2178         /* skip the start '[' */
2179         sscanf(&name[1], SFID, RFID(&tfid));
2180         if (fid_is_sane(&tfid))
2181                 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
2182                                       fill_buf->oifb_dentry->d_inode,
2183                                       child->d_inode, &tfid, name, namelen, 0);
2184         else
2185                 rc = -EIO;
2186         dput(child);
2187
2188         RETURN(rc);
2189 }
2190
2191 #ifdef HAVE_FILLDIR_USE_CTX
2192 static int osd_ios_root_fill(struct dir_context *buf,
2193 #else
2194 static int osd_ios_root_fill(void *buf,
2195 #endif
2196                              const char *name, int namelen,
2197                              loff_t offset, __u64 ino, unsigned d_type)
2198 {
2199         struct osd_ios_filldir_buf *fill_buf =
2200                 (struct osd_ios_filldir_buf *)buf;
2201         struct osd_device          *dev      = fill_buf->oifb_dev;
2202         const struct osd_lf_map    *map;
2203         struct dentry              *child;
2204         int                         rc       = 0;
2205         ENTRY;
2206
2207         fill_buf->oifb_items++;
2208
2209         /* skip any '.' started names */
2210         if (name[0] == '.')
2211                 RETURN(0);
2212
2213         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2214                 if (map->olm_namelen != namelen)
2215                         continue;
2216
2217                 if (strncmp(map->olm_name, name, namelen) == 0)
2218                         break;
2219         }
2220
2221         if (map->olm_name == NULL)
2222                 RETURN(0);
2223
2224         child = osd_lookup_one_len(dev, name, fill_buf->oifb_dentry, namelen);
2225         if (IS_ERR(child))
2226                 RETURN(PTR_ERR(child));
2227         else if (!child->d_inode)
2228                 GOTO(out_put, rc = -ENOENT);
2229
2230         if (!(map->olm_flags & OLF_NO_OI))
2231                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev,
2232                                 fill_buf->oifb_dentry->d_inode, child->d_inode,
2233                                 &map->olm_fid, name, namelen, map->olm_flags);
2234         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2235                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2236                                       map->olm_filldir);
2237 out_put:
2238         dput(child);
2239
2240         RETURN(rc);
2241 }
2242
2243 static int
2244 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2245                      struct dentry *dentry, filldir_t filldir)
2246 {
2247         struct osd_ios_filldir_buf    buf   = {
2248                                                 .ctx.actor = filldir,
2249                                                 .oifb_info = info,
2250                                                 .oifb_dev = dev,
2251                                                 .oifb_dentry = dentry };
2252         struct file                  *filp  = &info->oti_file;
2253         struct inode                 *inode = dentry->d_inode;
2254         const struct file_operations *fops  = inode->i_fop;
2255         int                           rc;
2256         ENTRY;
2257
2258         LASSERT(filldir != NULL);
2259
2260         filp->f_pos = 0;
2261         filp->f_path.dentry = dentry;
2262         filp->f_flags |= O_NOATIME;
2263         filp->f_mode = FMODE_64BITHASH | FMODE_NONOTIFY;
2264         filp->f_mapping = inode->i_mapping;
2265         filp->f_op = fops;
2266         filp->private_data = NULL;
2267         filp->f_cred = current_cred();
2268         filp->f_inode = inode;
2269         rc = osd_security_file_alloc(filp);
2270         if (rc)
2271                 RETURN(rc);
2272
2273         do {
2274                 buf.oifb_items = 0;
2275                 rc = iterate_dir(filp, &buf.ctx);
2276         } while (rc >= 0 && buf.oifb_items > 0 &&
2277                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
2278         fops->release(inode, filp);
2279
2280         RETURN(rc);
2281 }
2282
2283 static int
2284 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2285                   struct dentry *dentry, filldir_t filldir)
2286 {
2287         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2288         struct scrub_file *sf = &scrub->os_file;
2289         struct dentry *child;
2290         int rc;
2291         ENTRY;
2292
2293         /* It is existing MDT0 device. We only allow the case of object without
2294          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2295          * can generate IGIF mode FID for the object and related OI mapping. If
2296          * it is on other MDTs, then becuase file-level backup/restore, related
2297          * OI mapping may be invalid already, we do not know which is the right
2298          * FID for the object. We only allow IGIF objects to reside on the MDT0.
2299          *
2300          * XXX: For the case of object on non-MDT0 device with neither LMA nor
2301          *      "fid" xattr, then something crashed. We cannot re-generate the
2302          *      FID directly, instead, the OI scrub will scan the OI structure
2303          *      and try to re-generate the LMA from the OI mapping. But if the
2304          *      OI mapping crashed or lost also, then we have to give up under
2305          *      double failure cases. */
2306         scrub->os_convert_igif = 1;
2307         child = osd_lookup_one_len_unlocked(dev, dot_lustre_name, dentry,
2308                                             strlen(dot_lustre_name));
2309         if (IS_ERR(child)) {
2310                 if (PTR_ERR(child) != -ENOENT)
2311                         RETURN(PTR_ERR(child));
2312                 goto out_scrub;
2313         }
2314
2315         /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2316          * so the client will get IGIF for the ".lustre" object when
2317          * the MDT restart.
2318          *
2319          * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2320          * it does not know whether there are some old clients cached
2321          * the ".lustre" IGIF during the upgrading. Two choices:
2322          *
2323          * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2324          *    It will allow the old connected clients to access the
2325          *    ".lustre" with cached IGIF. But it will cause others
2326          *    on the MDT failed to check "fid_is_dot_lustre()".
2327          *
2328          * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2329          *    for ".lustre" in spite of whether there are some clients
2330          *    cached the ".lustre" IGIF or not. It enables the check
2331          *    "fid_is_dot_lustre()" on the MDT, although it will cause
2332          *    that the old connected clients cannot access the ".lustre"
2333          *    with the cached IGIF.
2334          *
2335          * Usually, it is rare case for the old connected clients
2336          * to access the ".lustre" with cached IGIF. So we prefer
2337          * to the solution 2).
2338          */
2339         inode_lock(dentry->d_inode);
2340         rc = osd_ios_scan_one(info, dev, dentry->d_inode,
2341                               child->d_inode, &LU_DOT_LUSTRE_FID,
2342                               dot_lustre_name,
2343                               strlen(dot_lustre_name), 0);
2344         inode_unlock(dentry->d_inode);
2345         if (rc == -ENOENT) {
2346 out_scrub:
2347                 /* It is 1.8 MDT device. */
2348                 if (!(sf->sf_flags & SF_UPGRADE)) {
2349                         scrub_file_reset(scrub, dev->od_uuid,
2350                                          SF_UPGRADE);
2351                         sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2352                         rc = scrub_file_store(info->oti_env, scrub);
2353                 } else {
2354                         rc = 0;
2355                 }
2356         } else if (rc == 0) {
2357                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2358                                       osd_ios_dl_fill);
2359         }
2360         dput(child);
2361
2362         RETURN(rc);
2363 }
2364
2365 static int
2366 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2367                      struct dentry *dentry, filldir_t filldir)
2368 {
2369         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2370         struct scrub_file *sf = &scrub->os_file;
2371         struct dentry *child;
2372         int rc;
2373         ENTRY;
2374
2375         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2376                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2377                 rc = scrub_file_store(info->oti_env, scrub);
2378                 if (rc != 0)
2379                         RETURN(rc);
2380         }
2381
2382         child = osd_lookup_one_len_unlocked(dev, ADMIN_USR, dentry,
2383                                             strlen(ADMIN_USR));
2384         if (IS_ERR(child)) {
2385                 rc = PTR_ERR(child);
2386         } else {
2387                 inode_lock(dentry->d_inode);
2388                 rc = osd_ios_scan_one(info, dev, dentry->d_inode,
2389                                       child->d_inode, NULL, ADMIN_USR,
2390                                       strlen(ADMIN_USR), 0);
2391                 inode_unlock(dentry->d_inode);
2392                 dput(child);
2393         }
2394
2395         if (rc != 0 && rc != -ENOENT)
2396                 GOTO(out, rc);
2397
2398         child = osd_lookup_one_len_unlocked(dev, ADMIN_GRP, dentry,
2399                                             strlen(ADMIN_GRP));
2400         if (IS_ERR(child))
2401                 GOTO(out, rc = PTR_ERR(child));
2402
2403         inode_lock(dentry->d_inode);
2404         rc = osd_ios_scan_one(info, dev, dentry->d_inode,
2405                               child->d_inode, NULL, ADMIN_GRP,
2406                               strlen(ADMIN_GRP), 0);
2407         inode_unlock(dentry->d_inode);
2408         dput(child);
2409 out:
2410         RETURN(rc == -ENOENT ? 0 : rc);
2411 }
2412
2413 static void osd_initial_OI_scrub(struct osd_thread_info *info,
2414                                  struct osd_device *dev)
2415 {
2416         struct osd_ios_item     *item    = NULL;
2417         scandir_t                scandir = osd_ios_general_scan;
2418         filldir_t                filldir = osd_ios_root_fill;
2419         struct dentry           *dentry  = osd_sb(dev)->s_root;
2420         const struct osd_lf_map *map     = osd_lf_maps;
2421         ENTRY;
2422
2423         /* Lookup IGIF in OI by force for initial OI scrub. */
2424         dev->od_igif_inoi = 1;
2425
2426         while (1) {
2427                 /* Don't take inode_lock here since scandir() callbacks
2428                  * can call VFS functions which may manully take the
2429                  * inode lock itself like iterate_dir(). Since this
2430                  * is the case it is best to leave the scandir()
2431                  * callbacks to managing the inode lock.
2432                  */
2433                 scandir(info, dev, dentry, filldir);
2434                 if (item != NULL) {
2435                         dput(item->oii_dentry);
2436                         OBD_FREE_PTR(item);
2437                 }
2438
2439                 if (list_empty(&dev->od_ios_list))
2440                         break;
2441
2442                 item = list_entry(dev->od_ios_list.next,
2443                                   struct osd_ios_item, oii_list);
2444                 list_del_init(&item->oii_list);
2445
2446                 LASSERT(item->oii_scandir != NULL);
2447                 scandir = item->oii_scandir;
2448                 filldir = item->oii_filldir;
2449                 dentry = item->oii_dentry;
2450         }
2451
2452         /* There maybe the case that the object has been removed, but its OI
2453          * mapping is still in the OI file, such as the "CATALOGS" after MDT
2454          * file-level backup/restore. So here cleanup the stale OI mappings. */
2455         while (map->olm_name != NULL) {
2456                 struct dentry *child;
2457
2458                 if (fid_is_zero(&map->olm_fid)) {
2459                         map++;
2460                         continue;
2461                 }
2462
2463                 child = osd_lookup_one_len_unlocked(dev, map->olm_name,
2464                                                     osd_sb(dev)->s_root,
2465                                                     map->olm_namelen);
2466                 if (PTR_ERR(child) == -ENOENT ||
2467                     (!IS_ERR(child) && !child->d_inode))
2468                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2469                                                   NULL, DTO_INDEX_DELETE,
2470                                                   true, 0, NULL);
2471                 if (!IS_ERR(child))
2472                         dput(child);
2473                 map++;
2474         }
2475
2476         if (!list_empty(&dev->od_index_restore_list)) {
2477                 char *buf;
2478
2479                 OBD_ALLOC_LARGE(buf, INDEX_BACKUP_BUFSIZE);
2480                 if (!buf)
2481                         CERROR("%s: not enough RAM for rebuild index\n",
2482                                osd_name(dev));
2483
2484                 while (!list_empty(&dev->od_index_restore_list)) {
2485                         struct lustre_index_restore_unit *liru;
2486
2487                         liru = list_entry(dev->od_index_restore_list.next,
2488                                           struct lustre_index_restore_unit,
2489                                           liru_link);
2490                         list_del(&liru->liru_link);
2491                         if (buf)
2492                                 osd_index_restore(info->oti_env, dev, liru,
2493                                                   buf, INDEX_BACKUP_BUFSIZE);
2494                         OBD_FREE(liru, liru->liru_len);
2495                 }
2496
2497                 if (buf)
2498                         OBD_FREE_LARGE(buf, INDEX_BACKUP_BUFSIZE);
2499         }
2500
2501         EXIT;
2502 }
2503
2504 char *osd_lf_fid2name(const struct lu_fid *fid)
2505 {
2506         const struct osd_lf_map *map = osd_lf_maps;
2507
2508         while (map->olm_name != NULL) {
2509                 if (!lu_fid_eq(fid, &map->olm_fid)) {
2510                         map++;
2511                         continue;
2512                 }
2513
2514                 if (map->olm_flags & OLF_SHOW_NAME)
2515                         return map->olm_name;
2516                 else
2517                         return "";
2518         }
2519
2520         return NULL;
2521 }
2522
2523 /* OI scrub start/stop */
2524
2525 int osd_scrub_start(const struct lu_env *env, struct osd_device *dev,
2526                     __u32 flags)
2527 {
2528         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2529         int rc;
2530         ENTRY;
2531
2532         if (dev->od_dt_dev.dd_rdonly)
2533                 RETURN(-EROFS);
2534
2535         /* od_otable_mutex: prevent curcurrent start/stop */
2536         mutex_lock(&dev->od_otable_mutex);
2537         rc = scrub_start(osd_scrub_main, scrub, dev, flags);
2538         if (rc == -EALREADY) {
2539                 rc = 0;
2540                 if ((scrub->os_file.sf_flags & SF_AUTO ||
2541                      scrub->os_partial_scan) &&
2542                     !(flags & SS_AUTO_PARTIAL))
2543                         osd_scrub_join(env, dev, flags, false);
2544         }
2545         mutex_unlock(&dev->od_otable_mutex);
2546
2547         RETURN(rc);
2548 }
2549
2550 void osd_scrub_stop(struct osd_device *dev)
2551 {
2552         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2553
2554         /* od_otable_mutex: prevent curcurrent start/stop */
2555         mutex_lock(&dev->od_otable_mutex);
2556         scrub->os_paused = 1;
2557         scrub_stop(scrub);
2558         mutex_unlock(&dev->od_otable_mutex);
2559 }
2560
2561 /* OI scrub setup/cleanup */
2562
2563 static const char osd_scrub_name[] = "OI_scrub";
2564
2565 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2566 {
2567         struct osd_thread_info *info = osd_oti_get(env);
2568         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2569         struct lvfs_run_ctxt *ctxt = &dev->od_scrub.os_ctxt;
2570         struct scrub_file *sf = &scrub->os_file;
2571         struct super_block *sb = osd_sb(dev);
2572         struct lvfs_run_ctxt saved;
2573         struct file *filp;
2574         struct inode *inode;
2575         struct lu_fid *fid = &info->oti_fid;
2576         struct osd_inode_id *id = &info->oti_id;
2577         struct dt_object *obj;
2578         bool dirty = false;
2579         bool restored = false;
2580         int rc = 0;
2581         ENTRY;
2582
2583         memset(&dev->od_scrub, 0, sizeof(struct osd_scrub));
2584         OBD_SET_CTXT_MAGIC(ctxt);
2585         ctxt->pwdmnt = dev->od_mnt;
2586         ctxt->pwd = dev->od_mnt->mnt_root;
2587         ctxt->fs = KERNEL_DS;
2588
2589         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2590         init_rwsem(&scrub->os_rwsem);
2591         spin_lock_init(&scrub->os_lock);
2592         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2593         scrub->os_name = osd_name(dev);
2594
2595         push_ctxt(&saved, ctxt);
2596         filp = filp_open(osd_scrub_name, O_RDWR |
2597                          (dev->od_dt_dev.dd_rdonly ? 0 : O_CREAT), 0644);
2598         if (IS_ERR(filp)) {
2599                 pop_ctxt(&saved, ctxt);
2600                 RETURN(PTR_ERR(filp));
2601         }
2602
2603         inode = file_inode(filp);
2604         if (!dev->od_dt_dev.dd_rdonly) {
2605                 /* 'What the @fid is' is not imporatant, because the object
2606                  * has no OI mapping, and only is visible inside the OSD.*/
2607                 lu_igif_build(fid, inode->i_ino, inode->i_generation);
2608                 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2609                 if (rc) {
2610                         filp_close(filp, NULL);
2611                         pop_ctxt(&saved, ctxt);
2612                         RETURN(rc);
2613                 }
2614         }
2615
2616         osd_id_gen(id, inode->i_ino, inode->i_generation);
2617         osd_add_oi_cache(info, dev, id, fid);
2618         filp_close(filp, NULL);
2619         pop_ctxt(&saved, ctxt);
2620
2621         obj = lu2dt(lu_object_find_slice(env, osd2lu_dev(dev), fid, NULL));
2622         if (IS_ERR_OR_NULL(obj))
2623                 RETURN(obj ? PTR_ERR(obj) : -ENOENT);
2624
2625 #ifndef HAVE_S_UUID_AS_UUID_T
2626         memcpy(dev->od_uuid.b, sb->s_uuid, UUID_SIZE);
2627 #else
2628         uuid_copy(&dev->od_uuid, &sb->s_uuid);
2629 #endif
2630         scrub->os_obj = obj;
2631         rc = scrub_file_load(env, scrub);
2632         if (rc == -ENOENT || rc == -EFAULT) {
2633                 scrub_file_init(scrub, dev->od_uuid);
2634                 /* If the "/O" dir does not exist when mount (indicated by
2635                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2636                  * then it is quite probably that the device is a new one,
2637                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2638                  *
2639                  * For the rare case that "/O" and "OI_scrub" both lost on
2640                  * an old device, it can be found and cleared later.
2641                  *
2642                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2643                  * need to check "filter_fid_18_23" and to convert it to
2644                  * "filter_fid" for each object, and all the IGIF should
2645                  * have their FID mapping in OI files already. */
2646                 if (dev->od_maybe_new && rc == -ENOENT)
2647                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2648                 dirty = true;
2649         } else if (rc < 0) {
2650                 GOTO(cleanup_obj, rc);
2651         } else {
2652                 if (!uuid_equal(&sf->sf_uuid, &dev->od_uuid)) {
2653                         CDEBUG(D_LFSCK,
2654                                "%s: UUID has been changed from %pU to %pU\n",
2655                                osd_dev2name(dev), &sf->sf_uuid, &dev->od_uuid);
2656                         scrub_file_reset(scrub, dev->od_uuid, SF_INCONSISTENT);
2657                         dirty = true;
2658                         restored = true;
2659                 } else if (sf->sf_status == SS_SCANNING) {
2660                         sf->sf_status = SS_CRASHED;
2661                         dirty = true;
2662                 }
2663
2664                 if ((sf->sf_oi_count & (sf->sf_oi_count - 1)) != 0) {
2665                         LCONSOLE_WARN("%s: invalid oi count %d, set it to %d\n",
2666                                       osd_dev2name(dev), sf->sf_oi_count,
2667                                       osd_oi_count);
2668                         sf->sf_oi_count = osd_oi_count;
2669                         dirty = true;
2670                 }
2671         }
2672
2673         if (sf->sf_pos_last_checkpoint != 0)
2674                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2675         else
2676                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2677
2678         if (dirty) {
2679                 rc = scrub_file_store(env, scrub);
2680                 if (rc)
2681                         GOTO(cleanup_obj, rc);
2682         }
2683
2684         /* Initialize OI files. */
2685         rc = osd_oi_init(info, dev, restored);
2686         if (rc < 0)
2687                 GOTO(cleanup_obj, rc);
2688
2689         if (!dev->od_dt_dev.dd_rdonly)
2690                 osd_initial_OI_scrub(info, dev);
2691
2692         if (sf->sf_flags & SF_UPGRADE ||
2693             !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2694               sf->sf_success_count > 0)) {
2695                 dev->od_igif_inoi = 0;
2696                 dev->od_check_ff = dev->od_is_ost;
2697         } else {
2698                 dev->od_igif_inoi = 1;
2699                 dev->od_check_ff = 0;
2700         }
2701
2702         if (sf->sf_flags & SF_INCONSISTENT)
2703                 /* The 'od_igif_inoi' will be set under the
2704                  * following cases:
2705                  * 1) new created system, or
2706                  * 2) restored from file-level backup, or
2707                  * 3) the upgrading completed.
2708                  *
2709                  * The 'od_igif_inoi' may be cleared by OI scrub
2710                  * later if found that the system is upgrading. */
2711                 dev->od_igif_inoi = 1;
2712
2713         if (!dev->od_dt_dev.dd_rdonly &&
2714             dev->od_auto_scrub_interval != AS_NEVER &&
2715             ((sf->sf_status == SS_PAUSED) ||
2716              (sf->sf_status == SS_CRASHED &&
2717               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2718                               SF_UPGRADE | SF_AUTO)) ||
2719              (sf->sf_status == SS_INIT &&
2720               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2721                               SF_UPGRADE))))
2722                 rc = osd_scrub_start(env, dev, SS_AUTO_FULL);
2723
2724         if (rc != 0)
2725                 GOTO(cleanup_oi, rc);
2726
2727         /* it is possible that dcache entries may keep objects after they are
2728          * deleted by OSD. While it looks safe this can cause object data to
2729          * stay until umount causing failures in tests calculating free space,
2730          * e.g. replay-ost-single. Since those dcache entries are not used
2731          * anymore let's just free them after use here */
2732         shrink_dcache_sb(sb);
2733
2734         RETURN(0);
2735 cleanup_oi:
2736         osd_oi_fini(info, dev);
2737 cleanup_obj:
2738         dt_object_put_nocache(env, scrub->os_obj);
2739         scrub->os_obj = NULL;
2740
2741         return rc;
2742 }
2743
2744 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2745 {
2746         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2747
2748         LASSERT(dev->od_otable_it == NULL);
2749
2750         if (scrub->os_obj != NULL) {
2751                 osd_scrub_stop(dev);
2752                 dt_object_put_nocache(env, scrub->os_obj);
2753                 scrub->os_obj = NULL;
2754         }
2755         if (dev->od_oi_table != NULL)
2756                 osd_oi_fini(osd_oti_get(env), dev);
2757 }
2758
2759 /* object table based iteration APIs */
2760
2761 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2762                                        struct dt_object *dt, __u32 attr)
2763 {
2764         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2765         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2766         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2767         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2768         struct osd_otable_it   *it;
2769         __u32                   start = 0;
2770         int                     rc;
2771         ENTRY;
2772
2773         /* od_otable_mutex: prevent curcurrent init/fini */
2774         mutex_lock(&dev->od_otable_mutex);
2775         if (dev->od_otable_it != NULL)
2776                 GOTO(out, it = ERR_PTR(-EALREADY));
2777
2778         OBD_ALLOC_PTR(it);
2779         if (it == NULL)
2780                 GOTO(out, it = ERR_PTR(-ENOMEM));
2781
2782         dev->od_otable_it = it;
2783         it->ooi_dev = dev;
2784         it->ooi_cache.ooc_consumer_idx = -1;
2785         if (flags & DOIF_OUTUSED)
2786                 it->ooi_used_outside = 1;
2787
2788         if (flags & DOIF_RESET)
2789                 start |= SS_RESET;
2790
2791         if (valid & DOIV_ERROR_HANDLE) {
2792                 if (flags & DOIF_FAILOUT)
2793                         start |= SS_SET_FAILOUT;
2794                 else
2795                         start |= SS_CLEAR_FAILOUT;
2796         }
2797
2798         if (valid & DOIV_DRYRUN) {
2799                 if (flags & DOIF_DRYRUN)
2800                         start |= SS_SET_DRYRUN;
2801                 else
2802                         start |= SS_CLEAR_DRYRUN;
2803         }
2804
2805         rc = scrub_start(osd_scrub_main, scrub, dev, start & ~SS_AUTO_PARTIAL);
2806         if (rc == -EALREADY) {
2807                 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2808         } else  if (rc < 0) {
2809                 dev->od_otable_it = NULL;
2810                 OBD_FREE_PTR(it);
2811                 it = ERR_PTR(rc);
2812         } else {
2813                 /* We have to start from the begining. */
2814                 it->ooi_cache.ooc_pos_preload =
2815                         LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2816         }
2817
2818         GOTO(out, it);
2819
2820 out:
2821         mutex_unlock(&dev->od_otable_mutex);
2822         return (struct dt_it *)it;
2823 }
2824
2825 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2826 {
2827         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2828         struct osd_device    *dev = it->ooi_dev;
2829
2830         /* od_otable_mutex: prevent curcurrent init/fini */
2831         mutex_lock(&dev->od_otable_mutex);
2832         scrub_stop(&dev->od_scrub.os_scrub);
2833         LASSERT(dev->od_otable_it == it);
2834
2835         dev->od_otable_it = NULL;
2836         mutex_unlock(&dev->od_otable_mutex);
2837         OBD_FREE_PTR(it);
2838 }
2839
2840 static int osd_otable_it_get(const struct lu_env *env,
2841                              struct dt_it *di, const struct dt_key *key)
2842 {
2843         return 0;
2844 }
2845
2846 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2847 {
2848 }
2849
2850 static inline int
2851 osd_otable_it_wakeup(struct lustre_scrub *scrub, struct osd_otable_it *it)
2852 {
2853         spin_lock(&scrub->os_lock);
2854         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2855             scrub->os_waiting ||
2856             !thread_is_running(&scrub->os_thread))
2857                 it->ooi_waiting = 0;
2858         else
2859                 it->ooi_waiting = 1;
2860         spin_unlock(&scrub->os_lock);
2861
2862         return !it->ooi_waiting;
2863 }
2864
2865 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2866 {
2867         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2868         struct osd_device       *dev    = it->ooi_dev;
2869         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2870         struct osd_otable_cache *ooc    = &it->ooi_cache;
2871         struct ptlrpc_thread    *thread = &scrub->os_thread;
2872         int                      rc;
2873         ENTRY;
2874
2875         LASSERT(it->ooi_user_ready);
2876
2877 again:
2878         if (!thread_is_running(thread) && !it->ooi_used_outside)
2879                 RETURN(1);
2880
2881         if (ooc->ooc_cached_items > 0) {
2882                 ooc->ooc_cached_items--;
2883                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2884                                         ~OSD_OTABLE_IT_CACHE_MASK;
2885                 RETURN(0);
2886         }
2887
2888         if (it->ooi_all_cached) {
2889                 wait_event_idle(thread->t_ctl_waitq,
2890                                 !thread_is_running(thread));
2891                 RETURN(1);
2892         }
2893
2894         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2895                 spin_lock(&scrub->os_lock);
2896                 scrub->os_waiting = 0;
2897                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2898                 spin_unlock(&scrub->os_lock);
2899         }
2900
2901         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2902                 wait_event_idle(thread->t_ctl_waitq,
2903                                 osd_otable_it_wakeup(scrub, it));
2904
2905         if (!thread_is_running(thread) && !it->ooi_used_outside)
2906                 RETURN(1);
2907
2908         rc = osd_otable_it_preload(env, it);
2909         if (rc >= 0)
2910                 goto again;
2911
2912         RETURN(rc);
2913 }
2914
2915 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2916                                         const struct dt_it *di)
2917 {
2918         return NULL;
2919 }
2920
2921 static int osd_otable_it_key_size(const struct lu_env *env,
2922                                   const struct dt_it *di)
2923 {
2924         return sizeof(__u64);
2925 }
2926
2927 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2928                              struct dt_rec *rec, __u32 attr)
2929 {
2930         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2931         struct osd_otable_cache *ooc = &it->ooi_cache;
2932
2933         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2934
2935         /* Filter out Invald FID already. */
2936         LASSERTF(fid_is_sane((struct lu_fid *)rec),
2937                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2938                  PFID((struct lu_fid *)rec),
2939                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2940
2941         return 0;
2942 }
2943
2944 static __u64 osd_otable_it_store(const struct lu_env *env,
2945                                  const struct dt_it *di)
2946 {
2947         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2948         struct osd_otable_cache *ooc = &it->ooi_cache;
2949         __u64                    hash;
2950
2951         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2952                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2953         else
2954                 hash = ooc->ooc_pos_preload;
2955         return hash;
2956 }
2957
2958 /**
2959  * Set the OSD layer iteration start position as the specified hash.
2960  */
2961 static int osd_otable_it_load(const struct lu_env *env,
2962                               const struct dt_it *di, __u64 hash)
2963 {
2964         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
2965         struct osd_device       *dev   = it->ooi_dev;
2966         struct osd_otable_cache *ooc   = &it->ooi_cache;
2967         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2968         struct osd_iit_param    *param = &it->ooi_iit_param;
2969         int                      rc;
2970         ENTRY;
2971
2972         /* Forbid to set iteration position after iteration started. */
2973         if (it->ooi_user_ready)
2974                 RETURN(-EPERM);
2975
2976         LASSERT(!scrub->os_partial_scan);
2977
2978         if (hash > OSD_OTABLE_MAX_HASH)
2979                 hash = OSD_OTABLE_MAX_HASH;
2980
2981         /* The hash is the last checkpoint position,
2982          * we will start from the next one. */
2983         ooc->ooc_pos_preload = hash + 1;
2984         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2985                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2986
2987         it->ooi_user_ready = 1;
2988         if (!scrub->os_full_speed)
2989                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2990
2991         memset(param, 0, sizeof(*param));
2992         param->sb = osd_sb(dev);
2993         param->start = ooc->ooc_pos_preload;
2994         param->bg = (ooc->ooc_pos_preload - 1) /
2995                     LDISKFS_INODES_PER_GROUP(param->sb);
2996         param->offset = (ooc->ooc_pos_preload - 1) %
2997                         LDISKFS_INODES_PER_GROUP(param->sb);
2998         param->gbase = 1 + param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
2999
3000         /* Unplug OSD layer iteration by the first next() call. */
3001         rc = osd_otable_it_next(env, (struct dt_it *)it);
3002
3003         RETURN(rc);
3004 }
3005
3006 static int osd_otable_it_key_rec(const struct lu_env *env,
3007                                  const struct dt_it *di, void *key_rec)
3008 {
3009         return 0;
3010 }
3011
3012 const struct dt_index_operations osd_otable_ops = {
3013         .dio_it = {
3014                 .init     = osd_otable_it_init,
3015                 .fini     = osd_otable_it_fini,
3016                 .get      = osd_otable_it_get,
3017                 .put      = osd_otable_it_put,
3018                 .next     = osd_otable_it_next,
3019                 .key      = osd_otable_it_key,
3020                 .key_size = osd_otable_it_key_size,
3021                 .rec      = osd_otable_it_rec,
3022                 .store    = osd_otable_it_store,
3023                 .load     = osd_otable_it_load,
3024                 .key_rec  = osd_otable_it_key_rec,
3025         }
3026 };
3027
3028 /* high priority inconsistent items list APIs */
3029
3030 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL  60
3031
3032 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
3033                    int insert)
3034 {
3035         struct osd_inconsistent_item *oii;
3036         struct osd_scrub *oscrub = &dev->od_scrub;
3037         struct lustre_scrub *lscrub = &oscrub->os_scrub;
3038         struct ptlrpc_thread *thread = &lscrub->os_thread;
3039         int wakeup = 0;
3040         ENTRY;
3041
3042         OBD_ALLOC_PTR(oii);
3043         if (unlikely(oii == NULL))
3044                 RETURN(-ENOMEM);
3045
3046         INIT_LIST_HEAD(&oii->oii_list);
3047         oii->oii_cache = *oic;
3048         oii->oii_insert = insert;
3049
3050         if (lscrub->os_partial_scan) {
3051                 __u64 now = ktime_get_real_seconds();
3052
3053                 /* If there haven't been errors in a long time,
3054                  * decay old count until either the errors are
3055                  * gone or we reach the current interval. */
3056                 while (unlikely(oscrub->os_bad_oimap_count > 0 &&
3057                                 oscrub->os_bad_oimap_time +
3058                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
3059                         oscrub->os_bad_oimap_count >>= 1;
3060                         oscrub->os_bad_oimap_time +=
3061                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
3062                 }
3063
3064                 oscrub->os_bad_oimap_time = now;
3065                 if (++oscrub->os_bad_oimap_count >
3066                     dev->od_full_scrub_threshold_rate)
3067                         lscrub->os_full_scrub = 1;
3068         }
3069
3070         spin_lock(&lscrub->os_lock);
3071         if (unlikely(!thread_is_running(thread))) {
3072                 spin_unlock(&lscrub->os_lock);
3073                 OBD_FREE_PTR(oii);
3074                 RETURN(-EAGAIN);
3075         }
3076
3077         if (list_empty(&lscrub->os_inconsistent_items))
3078                 wakeup = 1;
3079         list_add_tail(&oii->oii_list, &lscrub->os_inconsistent_items);
3080         spin_unlock(&lscrub->os_lock);
3081
3082         if (wakeup != 0)
3083                 wake_up_all(&thread->t_ctl_waitq);
3084
3085         RETURN(0);
3086 }
3087
3088 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
3089                    struct osd_inode_id *id)
3090 {
3091         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
3092         struct osd_inconsistent_item *oii;
3093         ENTRY;
3094
3095         spin_lock(&scrub->os_lock);
3096         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
3097                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
3098                         *id = oii->oii_cache.oic_lid;
3099                         spin_unlock(&scrub->os_lock);
3100                         RETURN(0);
3101                 }
3102         }
3103         spin_unlock(&scrub->os_lock);
3104
3105         RETURN(-ENOENT);
3106 }
3107
3108 void osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
3109 {
3110         struct osd_scrub *scrub = &dev->od_scrub;
3111
3112         scrub_dump(m, &scrub->os_scrub);
3113         seq_printf(m, "lf_scanned: %llu\n"
3114                    "lf_%s: %llu\n"
3115                    "lf_failed: %llu\n",
3116                    scrub->os_lf_scanned,
3117                    scrub->os_scrub.os_file.sf_param & SP_DRYRUN ?
3118                         "inconsistent" : "repaired",
3119                    scrub->os_lf_repaired,
3120                    scrub->os_lf_failed);
3121 }