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,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2012, 2016, 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 <lustre/lustre_idl.h>
  42 #include <lustre_disk.h>
  43 #include <dt_object.h>
  44 #include <linux/xattr.h>
  45
  46 #include "osd_internal.h"
  47 #include "osd_oi.h"
  48 #include "osd_scrub.h"
  49
  50 #define HALF_SEC        msecs_to_jiffies(MSEC_PER_SEC >> 1)
  51
  52 #define OSD_OTABLE_MAX_HASH             0x00000000ffffffffULL
  53
  54 #define SCRUB_NEXT_BREAK        1 /* exit current loop and process next group */
  55 #define SCRUB_NEXT_CONTINUE     2 /* skip current object and process next bit */
  56 #define SCRUB_NEXT_EXIT         3 /* exit all the loops */
  57 #define SCRUB_NEXT_WAIT         4 /* wait for free cache slot */
  58 #define SCRUB_NEXT_CRASH        5 /* simulate system crash during OI scrub */
  59 #define SCRUB_NEXT_FATAL        6 /* simulate failure during OI scrub */
  60 #define SCRUB_NEXT_NOSCRUB      7 /* new created object, no scrub on it */
  61 #define SCRUB_NEXT_NOLMA        8 /* the inode has no FID-in-LMA */
  62 #define SCRUB_NEXT_OSTOBJ       9 /* for OST-object */
  63 #define SCRUB_NEXT_OSTOBJ_OLD   10 /* old OST-object, no LMA or no FID-on-OST
  64                                     * flags in LMA */
  65
  66 /* misc functions */
  67
  68 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
  69 {
  70         return container_of0(scrub, struct osd_device, od_scrub);
  71 }
  72
  73 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
  74 {
  75         return osd_sb(osd_scrub2dev(scrub));
  76 }
  77
  78 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
  79                                        struct osd_otable_cache *ooc)
  80 {
  81         return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
  82 }
  83
  84 static inline const char *osd_scrub2name(struct osd_scrub *scrub)
  85 {
  86         return osd_dev2name(osd_scrub2dev(scrub));
  87 }
  88
  89 /**
  90  * update/insert/delete the specified OI mapping (@fid @id) according to the ops
  91  *
  92  * \retval   1, changed nothing
  93  * \retval   0, changed successfully
  94  * \retval -ve, on error
  95  */
  96 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
  97                                      struct osd_device *dev,
  98                                      const struct lu_fid *fid,
  99                                      const struct osd_inode_id *id,
 100                                      int ops, bool force,
 101                                      enum oi_check_flags flags, bool *exist)
 102 {
 103         handle_t *th;
 104         int       rc;
 105         ENTRY;
 106
 107         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
 108                 RETURN(0);
 109
 110         /* DTO_INDEX_INSERT is enough for other two ops:
 111          * delete/update, but save stack. */
 112         th = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 113                                 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
 114         if (IS_ERR(th)) {
 115                 rc = PTR_ERR(th);
 116                 CDEBUG(D_LFSCK, "%s: fail to start trans for scrub op %d "
 117                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 118                        PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1,
 119                        rc);
 120                 RETURN(rc);
 121         }
 122
 123         switch (ops) {
 124         case DTO_INDEX_UPDATE:
 125                 rc = osd_oi_update(info, dev, fid, id, th, flags);
 126                 if (unlikely(rc == -ENOENT)) {
 127                         /* Some unlink thread may removed the OI mapping. */
 128                         rc = 1;
 129                 }
 130                 break;
 131         case DTO_INDEX_INSERT:
 132                 rc = osd_oi_insert(info, dev, fid, id, th, flags, exist);
 133                 if (unlikely(rc == -EEXIST)) {
 134                         rc = 1;
 135                         /* XXX: There are trouble things when adding OI
 136                          *      mapping for IGIF object, which may cause
 137                          *      multiple objects to be mapped to the same
 138                          *      IGIF formatted FID. Consider the following
 139                          *      situations:
 140                          *
 141                          *      1) The MDT is upgrading from 1.8 device.
 142                          *      The OI scrub generates IGIF FID1 for the
 143                          *      OBJ1 and adds the OI mapping.
 144                          *
 145                          *      2) For some reason, the OI scrub does not
 146                          *      process all the IGIF objects completely.
 147                          *
 148                          *      3) The MDT is backuped and restored against
 149                          *      this device.
 150                          *
 151                          *      4) When the MDT mounts up, the OI scrub will
 152                          *      try to rebuild the OI files. For some IGIF
 153                          *      object, OBJ2, which was not processed by the
 154                          *      OI scrub before the backup/restore, and the
 155                          *      new generated IGIF formatted FID may be just
 156                          *      the FID1, the same as OBJ1.
 157                          *
 158                          *      Under such case, the OI scrub cannot know how
 159                          *      to generate new FID for the OBJ2.
 160                          *
 161                          *      Currently, we do nothing for that. One possible
 162                          *      solution is to generate new normal FID for the
 163                          *      conflict object.
 164                          *
 165                          *      Anyway, it is rare, only exists in theory. */
 166                 }
 167                 break;
 168         case DTO_INDEX_DELETE:
 169                 rc = osd_oi_delete(info, dev, fid, th, flags);
 170                 if (rc == -ENOENT) {
 171                         /* It is normal that the unlink thread has removed the
 172                          * OI mapping already. */
 173                         rc = 1;
 174                 }
 175                 break;
 176         default:
 177                 LASSERTF(0, "Unexpected ops %d\n", ops);
 178                 break;
 179         }
 180
 181         ldiskfs_journal_stop(th);
 182         if (rc < 0)
 183                 CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d "
 184                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 185                        PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1,
 186                        rc);
 187
 188         RETURN(rc);
 189 }
 190
 191 /* OI_scrub file ops */
 192
 193 static void osd_scrub_file_to_cpu(struct scrub_file *des,
 194                                   struct scrub_file *src)
 195 {
 196         memcpy(des->sf_uuid, src->sf_uuid, 16);
 197         des->sf_flags   = le64_to_cpu(src->sf_flags);
 198         des->sf_magic   = le32_to_cpu(src->sf_magic);
 199         des->sf_status  = le16_to_cpu(src->sf_status);
 200         des->sf_param   = le16_to_cpu(src->sf_param);
 201         des->sf_time_last_complete      =
 202                                 le64_to_cpu(src->sf_time_last_complete);
 203         des->sf_time_latest_start       =
 204                                 le64_to_cpu(src->sf_time_latest_start);
 205         des->sf_time_last_checkpoint    =
 206                                 le64_to_cpu(src->sf_time_last_checkpoint);
 207         des->sf_pos_latest_start        =
 208                                 le64_to_cpu(src->sf_pos_latest_start);
 209         des->sf_pos_last_checkpoint     =
 210                                 le64_to_cpu(src->sf_pos_last_checkpoint);
 211         des->sf_pos_first_inconsistent  =
 212                                 le64_to_cpu(src->sf_pos_first_inconsistent);
 213         des->sf_items_checked           =
 214                                 le64_to_cpu(src->sf_items_checked);
 215         des->sf_items_updated           =
 216                                 le64_to_cpu(src->sf_items_updated);
 217         des->sf_items_failed            =
 218                                 le64_to_cpu(src->sf_items_failed);
 219         des->sf_items_updated_prior     =
 220                                 le64_to_cpu(src->sf_items_updated_prior);
 221         des->sf_run_time        = le32_to_cpu(src->sf_run_time);
 222         des->sf_success_count   = le32_to_cpu(src->sf_success_count);
 223         des->sf_oi_count        = le16_to_cpu(src->sf_oi_count);
 224         des->sf_internal_flags  = le16_to_cpu(src->sf_internal_flags);
 225         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 226 }
 227
 228 static void osd_scrub_file_to_le(struct scrub_file *des,
 229                                  struct scrub_file *src)
 230 {
 231         memcpy(des->sf_uuid, src->sf_uuid, 16);
 232         des->sf_flags   = cpu_to_le64(src->sf_flags);
 233         des->sf_magic   = cpu_to_le32(src->sf_magic);
 234         des->sf_status  = cpu_to_le16(src->sf_status);
 235         des->sf_param   = cpu_to_le16(src->sf_param);
 236         des->sf_time_last_complete      =
 237                                 cpu_to_le64(src->sf_time_last_complete);
 238         des->sf_time_latest_start       =
 239                                 cpu_to_le64(src->sf_time_latest_start);
 240         des->sf_time_last_checkpoint    =
 241                                 cpu_to_le64(src->sf_time_last_checkpoint);
 242         des->sf_pos_latest_start        =
 243                                 cpu_to_le64(src->sf_pos_latest_start);
 244         des->sf_pos_last_checkpoint     =
 245                                 cpu_to_le64(src->sf_pos_last_checkpoint);
 246         des->sf_pos_first_inconsistent  =
 247                                 cpu_to_le64(src->sf_pos_first_inconsistent);
 248         des->sf_items_checked           =
 249                                 cpu_to_le64(src->sf_items_checked);
 250         des->sf_items_updated           =
 251                                 cpu_to_le64(src->sf_items_updated);
 252         des->sf_items_failed            =
 253                                 cpu_to_le64(src->sf_items_failed);
 254         des->sf_items_updated_prior     =
 255                                 cpu_to_le64(src->sf_items_updated_prior);
 256         des->sf_run_time        = cpu_to_le32(src->sf_run_time);
 257         des->sf_success_count   = cpu_to_le32(src->sf_success_count);
 258         des->sf_oi_count        = cpu_to_le16(src->sf_oi_count);
 259         des->sf_internal_flags  = cpu_to_le16(src->sf_internal_flags);
 260         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 261 }
 262
 263 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
 264 {
 265         struct scrub_file *sf = &scrub->os_file;
 266
 267         memset(sf, 0, sizeof(*sf));
 268         memcpy(sf->sf_uuid, uuid, 16);
 269         sf->sf_magic = SCRUB_MAGIC_V1;
 270         sf->sf_status = SS_INIT;
 271 }
 272
 273 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
 274 {
 275         struct scrub_file *sf = &scrub->os_file;
 276
 277         CDEBUG(D_LFSCK, "%s: reset OI scrub file, old flags = "
 278                "%#llx, add flags = %#llx\n",
 279                osd_scrub2name(scrub), sf->sf_flags, flags);
 280
 281         memcpy(sf->sf_uuid, uuid, 16);
 282         sf->sf_status = SS_INIT;
 283         sf->sf_flags |= flags;
 284         sf->sf_flags &= ~SF_AUTO;
 285         sf->sf_run_time = 0;
 286         sf->sf_time_latest_start = 0;
 287         sf->sf_time_last_checkpoint = 0;
 288         sf->sf_pos_latest_start = 0;
 289         sf->sf_pos_last_checkpoint = 0;
 290         sf->sf_pos_first_inconsistent = 0;
 291         sf->sf_items_checked = 0;
 292         sf->sf_items_updated = 0;
 293         sf->sf_items_failed = 0;
 294         if (!scrub->os_in_join)
 295                 sf->sf_items_updated_prior = 0;
 296
 297         sf->sf_items_noscrub = 0;
 298         sf->sf_items_igif = 0;
 299 }
 300
 301 static int osd_scrub_file_load(struct osd_scrub *scrub)
 302 {
 303         loff_t  pos  = 0;
 304         int     len  = sizeof(scrub->os_file_disk);
 305         int     rc;
 306
 307         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
 308         if (rc == len) {
 309                 struct scrub_file *sf = &scrub->os_file;
 310
 311                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
 312                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
 313                         CDEBUG(D_LFSCK, "%s: invalid scrub magic "
 314                                "0x%x != 0x%x\n", osd_scrub2name(scrub),
 315                                sf->sf_magic, SCRUB_MAGIC_V1);
 316                         /* Process it as new scrub file. */
 317                         rc = -ENOENT;
 318                 } else {
 319                         rc = 0;
 320                 }
 321         } else if (rc != 0) {
 322                 CDEBUG(D_LFSCK, "%s: fail to load scrub file, "
 323                        "expected = %d: rc = %d\n",
 324                        osd_scrub2name(scrub), len, rc);
 325                 if (rc > 0)
 326                         rc = -EFAULT;
 327         } else {
 328                 /* return -ENOENT for empty scrub file case. */
 329                 rc = -ENOENT;
 330         }
 331
 332         return rc;
 333 }
 334
 335 int osd_scrub_file_store(struct osd_scrub *scrub)
 336 {
 337         struct osd_device *dev;
 338         handle_t          *jh;
 339         loff_t             pos     = 0;
 340         int                len     = sizeof(scrub->os_file_disk);
 341         int                credits;
 342         int                rc;
 343
 344         dev = container_of0(scrub, struct osd_device, od_scrub);
 345         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
 346                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
 347         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits);
 348         if (IS_ERR(jh)) {
 349                 rc = PTR_ERR(jh);
 350                 CDEBUG(D_LFSCK, "%s: fail to start trans for scrub store: "
 351                        "rc = %d\n", osd_scrub2name(scrub), rc);
 352                 return rc;
 353         }
 354
 355         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
 356         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
 357                                       len, 0, &pos, jh);
 358         ldiskfs_journal_stop(jh);
 359         if (rc != 0)
 360                 CDEBUG(D_LFSCK, "%s: fail to store scrub file, "
 361                        "expected = %d: rc = %d\n",
 362                        osd_scrub2name(scrub), len, rc);
 363
 364         scrub->os_time_last_checkpoint = cfs_time_current();
 365         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
 366                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
 367         return rc;
 368 }
 369
 370 static int
 371 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
 372                      struct inode *inode, const struct lu_fid *fid)
 373 {
 374         struct filter_fid_old   *ff      = &info->oti_ff;
 375         struct dentry           *dentry  = &info->oti_obj_dentry;
 376         struct lu_fid           *tfid    = &info->oti_fid;
 377         handle_t                *jh;
 378         int                      size    = 0;
 379         int                      rc;
 380         bool                     reset   = false;
 381         ENTRY;
 382
 383         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
 384                 RETURN(0);
 385
 386         if (fid_is_idif(fid) && dev->od_index_in_idif == 0) {
 387                 struct ost_id *oi = &info->oti_ostid;
 388
 389                 fid_to_ostid(fid, oi);
 390                 ostid_to_fid(tfid, oi, 0);
 391         } else {
 392                 *tfid = *fid;
 393         }
 394
 395         /* We want the LMA to fit into the 256-byte OST inode, so operate
 396          * as following:
 397          * 1) read old XATTR_NAME_FID and save the parent FID;
 398          * 2) delete the old XATTR_NAME_FID;
 399          * 3) make new LMA and add it;
 400          * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
 401          *
 402          * Making the LMA to fit into the 256-byte OST inode can save time for
 403          * normal osd_check_lma() and for other OI scrub scanning in future.
 404          * So it is worth to make some slow conversion here. */
 405         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 406                                 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
 407         if (IS_ERR(jh)) {
 408                 rc = PTR_ERR(jh);
 409                 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
 410                        DFID": rc = %d\n", osd_name(dev), PFID(tfid), rc);
 411                 RETURN(rc);
 412         }
 413
 414         /* 1) read old XATTR_NAME_FID and save the parent FID */
 415         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 416         if (rc == sizeof(*ff)) {
 417                 /* 2) delete the old XATTR_NAME_FID */
 418                 ll_vfs_dq_init(inode);
 419                 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
 420                 if (rc)
 421                         GOTO(stop, rc);
 422
 423                 reset = true;
 424         } else if (rc != -ENODATA && rc != sizeof(struct filter_fid)) {
 425                 GOTO(stop, rc = -EINVAL);
 426         }
 427
 428         /* 3) make new LMA and add it */
 429         rc = osd_ea_fid_set(info, inode, tfid, LMAC_FID_ON_OST, 0);
 430         if (reset) {
 431                 if (rc)
 432                         /* If failed, we should try to add the old back. */
 433                         size = sizeof(*ff);
 434                 else
 435                         /* The new PFID EA will only contains ::ff_parent */
 436                         size = sizeof(ff->ff_parent);
 437         }
 438
 439         /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
 440         if (size > 0) {
 441                 int rc1;
 442
 443                 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
 444                                       XATTR_CREATE);
 445                 if (rc1 != 0 && rc == 0)
 446                         rc = rc1;
 447         }
 448
 449         GOTO(stop, rc);
 450
 451 stop:
 452         ldiskfs_journal_stop(jh);
 453         if (rc < 0)
 454                 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
 455                        osd_name(dev), PFID(tfid), rc);
 456         return rc;
 457 }
 458
 459 static int
 460 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
 461                        struct osd_idmap_cache *oic, int val)
 462 {
 463         struct osd_scrub             *scrub  = &dev->od_scrub;
 464         struct scrub_file            *sf     = &scrub->os_file;
 465         struct lu_fid                *fid    = &oic->oic_fid;
 466         struct osd_inode_id          *lid    = &oic->oic_lid;
 467         struct osd_inode_id          *lid2   = &info->oti_id;
 468         struct osd_inconsistent_item *oii    = NULL;
 469         struct inode                 *inode  = NULL;
 470         int                           ops    = DTO_INDEX_UPDATE;
 471         int                           rc;
 472         bool                          converted = false;
 473         bool                          exist     = false;
 474         ENTRY;
 475
 476         down_write(&scrub->os_rwsem);
 477         scrub->os_new_checked++;
 478         if (val < 0)
 479                 GOTO(out, rc = val);
 480
 481         if (scrub->os_in_prior)
 482                 oii = list_entry(oic, struct osd_inconsistent_item,
 483                                  oii_cache);
 484
 485         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
 486                 GOTO(out, rc = 0);
 487
 488         if (fid_is_igif(fid))
 489                 sf->sf_items_igif++;
 490
 491         if (val == SCRUB_NEXT_OSTOBJ_OLD) {
 492                 inode = osd_iget(info, dev, lid);
 493                 if (IS_ERR(inode)) {
 494                         rc = PTR_ERR(inode);
 495                         /* Someone removed the inode. */
 496                         if (rc == -ENOENT || rc == -ESTALE)
 497                                 rc = 0;
 498                         GOTO(out, rc);
 499                 }
 500
 501                 /* The inode has been reused as EA inode, ignore it. */
 502                 if (unlikely(osd_is_ea_inode(inode)))
 503                         GOTO(out, rc = 0);
 504
 505                 sf->sf_flags |= SF_UPGRADE;
 506                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
 507                 dev->od_check_ff = 1;
 508                 rc = osd_scrub_convert_ff(info, dev, inode, fid);
 509                 if (rc != 0)
 510                         GOTO(out, rc);
 511
 512                 converted = true;
 513         }
 514
 515         if ((val == SCRUB_NEXT_NOLMA) &&
 516             (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
 517                 GOTO(out, rc = 0);
 518
 519         if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA)) {
 520                 ops = DTO_INDEX_INSERT;
 521
 522                 goto iget;
 523         }
 524
 525         rc = osd_oi_lookup(info, dev, fid, lid2,
 526                 (val == SCRUB_NEXT_OSTOBJ ||
 527                  val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 528         if (rc != 0) {
 529                 if (rc == -ENOENT)
 530                         ops = DTO_INDEX_INSERT;
 531                 else if (rc != -ESTALE)
 532                         GOTO(out, rc);
 533
 534 iget:
 535                 if (inode == NULL) {
 536                         inode = osd_iget(info, dev, lid);
 537                         if (IS_ERR(inode)) {
 538                                 rc = PTR_ERR(inode);
 539                                 /* Someone removed the inode. */
 540                                 if (rc == -ENOENT || rc == -ESTALE)
 541                                         rc = 0;
 542                                 GOTO(out, rc);
 543                         }
 544
 545                         /* The inode has been reused as EA inode, ignore it. */
 546                         if (unlikely(osd_is_ea_inode(inode)))
 547                                 GOTO(out, rc = 0);
 548                 }
 549
 550                 if (!scrub->os_partial_scan)
 551                         scrub->os_full_speed = 1;
 552
 553                 switch (val) {
 554                 case SCRUB_NEXT_NOLMA:
 555                         sf->sf_flags |= SF_UPGRADE;
 556                         if (!(sf->sf_param & SP_DRYRUN)) {
 557                                 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
 558                                 if (rc != 0)
 559                                         GOTO(out, rc);
 560                         }
 561
 562                         if (!(sf->sf_flags & SF_INCONSISTENT))
 563                                 dev->od_igif_inoi = 0;
 564                         break;
 565                 case SCRUB_NEXT_OSTOBJ:
 566                         sf->sf_flags |= SF_INCONSISTENT;
 567                 case SCRUB_NEXT_OSTOBJ_OLD:
 568                         break;
 569                 default:
 570                         break;
 571                 }
 572         } else if (osd_id_eq(lid, lid2)) {
 573                 if (converted)
 574                         sf->sf_items_updated++;
 575
 576                 GOTO(out, rc = 0);
 577         } else {
 578                 if (!scrub->os_partial_scan)
 579                         scrub->os_full_speed = 1;
 580
 581                 sf->sf_flags |= SF_INCONSISTENT;
 582
 583                 /* XXX: If the device is restored from file-level backup, then
 584                  *      some IGIFs may have been already in OI files, and some
 585                  *      may be not yet. Means upgrading from 1.8 may be partly
 586                  *      processed, but some clients may hold some immobilized
 587                  *      IGIFs, and use them to access related objects. Under
 588                  *      such case, OSD does not know whether an given IGIF has
 589                  *      been processed or to be processed, and it also cannot
 590                  *      generate local ino#/gen# directly from the immobilized
 591                  *      IGIF because of the backup/restore. Then force OSD to
 592                  *      lookup the given IGIF in OI files, and if no entry,
 593                  *      then ask the client to retry after upgrading completed.
 594                  *      No better choice. */
 595                 dev->od_igif_inoi = 1;
 596         }
 597
 598         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
 599                         (val == SCRUB_NEXT_OSTOBJ ||
 600                          val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0,
 601                         &exist);
 602         if (rc == 0) {
 603                 if (scrub->os_in_prior)
 604                         sf->sf_items_updated_prior++;
 605                 else
 606                         sf->sf_items_updated++;
 607
 608                 if (ops == DTO_INDEX_INSERT && val == 0 && !exist) {
 609                         int idx = osd_oi_fid2idx(dev, fid);
 610
 611                         sf->sf_flags |= SF_RECREATED;
 612                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 613                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 614                 }
 615         }
 616
 617         GOTO(out, rc);
 618
 619 out:
 620         if (rc < 0) {
 621                 sf->sf_items_failed++;
 622                 if (sf->sf_pos_first_inconsistent == 0 ||
 623                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 624                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 625         } else {
 626                 rc = 0;
 627         }
 628
 629         /* There may be conflict unlink during the OI scrub,
 630          * if happend, then remove the new added OI mapping. */
 631         if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
 632             unlikely(ldiskfs_test_inode_state(inode,
 633                                               LDISKFS_STATE_LUSTRE_DESTROY)))
 634                 osd_scrub_refresh_mapping(info, dev, fid, lid,
 635                                 DTO_INDEX_DELETE, false,
 636                                 (val == SCRUB_NEXT_OSTOBJ ||
 637                                  val == SCRUB_NEXT_OSTOBJ_OLD) ?
 638                                 OI_KNOWN_ON_OST : 0, NULL);
 639         up_write(&scrub->os_rwsem);
 640
 641         if (inode != NULL && !IS_ERR(inode))
 642                 iput(inode);
 643
 644         if (oii != NULL) {
 645                 spin_lock(&scrub->os_lock);
 646                 if (likely(!list_empty(&oii->oii_list)))
 647                         list_del(&oii->oii_list);
 648                 spin_unlock(&scrub->os_lock);
 649
 650                 OBD_FREE_PTR(oii);
 651         }
 652
 653         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 654 }
 655
 656 /* OI scrub APIs */
 657
 658 static int osd_scrub_prep(struct osd_device *dev)
 659 {
 660         struct osd_scrub     *scrub  = &dev->od_scrub;
 661         struct ptlrpc_thread *thread = &scrub->os_thread;
 662         struct scrub_file    *sf     = &scrub->os_file;
 663         __u32                 flags  = scrub->os_start_flags;
 664         int                   rc;
 665         bool                  drop_dryrun = false;
 666         ENTRY;
 667
 668         CDEBUG(D_LFSCK, "%s: OI scrub prep, flags = 0x%x\n",
 669                osd_scrub2name(scrub), flags);
 670
 671         down_write(&scrub->os_rwsem);
 672         if (flags & SS_SET_FAILOUT)
 673                 sf->sf_param |= SP_FAILOUT;
 674         else if (flags & SS_CLEAR_FAILOUT)
 675                 sf->sf_param &= ~SP_FAILOUT;
 676
 677         if (flags & SS_SET_DRYRUN) {
 678                 sf->sf_param |= SP_DRYRUN;
 679         } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
 680                 sf->sf_param &= ~SP_DRYRUN;
 681                 drop_dryrun = true;
 682         }
 683
 684         if (flags & SS_RESET)
 685                 osd_scrub_file_reset(scrub,
 686                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
 687
 688         if (flags & SS_AUTO_FULL) {
 689                 scrub->os_full_speed = 1;
 690                 scrub->os_partial_scan = 0;
 691                 sf->sf_flags |= SF_AUTO;
 692         } else if (flags & SS_AUTO_PARTIAL) {
 693                 scrub->os_full_speed = 0;
 694                 scrub->os_partial_scan = 1;
 695                 sf->sf_flags |= SF_AUTO;
 696         } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
 697                                    SF_UPGRADE)) {
 698                 scrub->os_full_speed = 1;
 699                 scrub->os_partial_scan = 0;
 700         } else {
 701                 scrub->os_full_speed = 0;
 702                 scrub->os_partial_scan = 0;
 703         }
 704
 705         spin_lock(&scrub->os_lock);
 706         scrub->os_in_prior = 0;
 707         scrub->os_waiting = 0;
 708         scrub->os_paused = 0;
 709         scrub->os_in_join = 0;
 710         scrub->os_full_scrub = 0;
 711         spin_unlock(&scrub->os_lock);
 712         scrub->os_new_checked = 0;
 713         if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
 714                 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
 715         else if (sf->sf_pos_last_checkpoint != 0)
 716                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 717         else
 718                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 719
 720         scrub->os_pos_current = sf->sf_pos_latest_start;
 721         sf->sf_status = SS_SCANNING;
 722         sf->sf_time_latest_start = cfs_time_current_sec();
 723         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 724         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
 725         rc = osd_scrub_file_store(scrub);
 726         if (rc == 0) {
 727                 spin_lock(&scrub->os_lock);
 728                 thread_set_flags(thread, SVC_RUNNING);
 729                 spin_unlock(&scrub->os_lock);
 730                 wake_up_all(&thread->t_ctl_waitq);
 731         }
 732         up_write(&scrub->os_rwsem);
 733
 734         RETURN(rc);
 735 }
 736
 737 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
 738 {
 739         struct scrub_file *sf = &scrub->os_file;
 740         int                rc;
 741
 742         if (likely(cfs_time_before(cfs_time_current(),
 743                                    scrub->os_time_next_checkpoint) ||
 744                    scrub->os_new_checked == 0))
 745                 return 0;
 746
 747         down_write(&scrub->os_rwsem);
 748         sf->sf_items_checked += scrub->os_new_checked;
 749         scrub->os_new_checked = 0;
 750         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 751         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 752         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 753                                             scrub->os_time_last_checkpoint);
 754         rc = osd_scrub_file_store(scrub);
 755         up_write(&scrub->os_rwsem);
 756
 757         return rc;
 758 }
 759
 760 static int osd_scrub_post(struct osd_scrub *scrub, int result)
 761 {
 762         struct scrub_file *sf = &scrub->os_file;
 763         int rc;
 764         ENTRY;
 765
 766         CDEBUG(D_LFSCK, "%s: OI scrub post, result = %d\n",
 767                osd_scrub2name(scrub), result);
 768
 769         down_write(&scrub->os_rwsem);
 770         spin_lock(&scrub->os_lock);
 771         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 772         spin_unlock(&scrub->os_lock);
 773         if (scrub->os_new_checked > 0) {
 774                 sf->sf_items_checked += scrub->os_new_checked;
 775                 scrub->os_new_checked = 0;
 776                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 777         }
 778         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 779         if (result > 0) {
 780                 struct osd_device *dev =
 781                         container_of0(scrub, struct osd_device, od_scrub);
 782
 783                 dev->od_igif_inoi = 1;
 784                 dev->od_check_ff = 0;
 785                 sf->sf_status = SS_COMPLETED;
 786                 if (!(sf->sf_param & SP_DRYRUN)) {
 787                         memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 788                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 789                                           SF_UPGRADE | SF_AUTO);
 790                 }
 791                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 792                 sf->sf_success_count++;
 793         } else if (result == 0) {
 794                 if (scrub->os_paused)
 795                         sf->sf_status = SS_PAUSED;
 796                 else
 797                         sf->sf_status = SS_STOPPED;
 798         } else {
 799                 sf->sf_status = SS_FAILED;
 800         }
 801         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 802                                             scrub->os_time_last_checkpoint);
 803         rc = osd_scrub_file_store(scrub);
 804         up_write(&scrub->os_rwsem);
 805
 806         RETURN(rc < 0 ? rc : result);
 807 }
 808
 809 /* iteration engine */
 810
 811 struct osd_iit_param {
 812         struct super_block *sb;
 813         struct buffer_head *bitmap;
 814         ldiskfs_group_t bg;
 815         __u32 gbase;
 816         __u32 offset;
 817         __u32 start;
 818 };
 819
 820 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 821                                    struct osd_device *dev,
 822                                    struct osd_iit_param *param,
 823                                    struct osd_idmap_cache **oic,
 824                                    const bool noslot);
 825
 826 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 827                                    struct osd_device *dev,
 828                                    struct osd_iit_param *param,
 829                                    struct osd_idmap_cache *oic,
 830                                    bool *noslot, int rc);
 831
 832 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 833 {
 834         __u32 offset;
 835
 836 again:
 837         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 838                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 839         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 840                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 841                 return SCRUB_NEXT_BREAK;
 842         }
 843
 844         offset = param->offset++;
 845         if (unlikely(*pos == param->gbase + offset && *pos != param->start)) {
 846                 /* We should NOT find the same object more than once. */
 847                 CERROR("%s: scan the same object multiple times at the pos: "
 848                        "group = %u, base = %u, offset = %u, start = %u\n",
 849                        param->sb->s_id, (__u32)param->bg, param->gbase,
 850                        offset, param->start);
 851                 goto again;
 852         }
 853
 854         *pos = param->gbase + offset;
 855         return 0;
 856 }
 857
 858 /**
 859  * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
 860  * \retval 0: FID-on-MDT
 861  */
 862 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
 863                                       struct osd_device *dev,
 864                                       struct lu_fid *fid)
 865 {
 866         /* XXX: The initial OI scrub will scan the top level /O to generate
 867          *      a small local FLDB according to the <seq>. If the given FID
 868          *      is in the local FLDB, then it is FID-on-OST; otherwise it's
 869          *      quite possible for FID-on-MDT. */
 870         if (dev->od_is_ost)
 871                 return SCRUB_NEXT_OSTOBJ_OLD;
 872         else
 873                 return 0;
 874 }
 875
 876 static int osd_scrub_get_fid(struct osd_thread_info *info,
 877                              struct osd_device *dev, struct inode *inode,
 878                              struct lu_fid *fid, bool scrub)
 879 {
 880         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
 881         int rc;
 882         bool has_lma = false;
 883
 884         rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
 885                          &info->oti_ost_attrs);
 886         if (rc == 0) {
 887                 has_lma = true;
 888                 if (lma->lma_compat & LMAC_NOT_IN_OI ||
 889                     lma->lma_incompat & LMAI_AGENT)
 890                         return SCRUB_NEXT_CONTINUE;
 891
 892                 *fid = lma->lma_self_fid;
 893                 if (!scrub)
 894                         return 0;
 895
 896                 if (lma->lma_compat & LMAC_FID_ON_OST)
 897                         return SCRUB_NEXT_OSTOBJ;
 898
 899                 if (fid_is_idif(fid))
 900                         return SCRUB_NEXT_OSTOBJ_OLD;
 901
 902                 /* For local object. */
 903                 if (fid_is_internal(fid))
 904                         return 0;
 905
 906                 /* For external visible MDT-object with non-normal FID. */
 907                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
 908                         return 0;
 909
 910                 /* For the object with normal FID, it may be MDT-object,
 911                  * or may be 2.4 OST-object, need further distinguish.
 912                  * Fall through to next section. */
 913         }
 914
 915         if (rc == -ENODATA || rc == 0) {
 916                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
 917                 if (rc == 0) {
 918                         if (scrub)
 919                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
 920                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 921                         return rc;
 922                 }
 923
 924                 if (rc > 0) {
 925                         if (!has_lma)
 926                                 /* It is FID-on-OST, but we do not know how
 927                                  * to generate its FID, ignore it directly. */
 928                                 rc = SCRUB_NEXT_CONTINUE;
 929                         else
 930                                 /* It is 2.4 OST-object. */
 931                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 932                         return rc;
 933                 }
 934
 935                 if (rc != -ENODATA)
 936                         return rc;
 937
 938                 if (!has_lma) {
 939                         if (dev->od_scrub.os_convert_igif) {
 940                                 lu_igif_build(fid, inode->i_ino,
 941                                               inode->i_generation);
 942                                 if (scrub)
 943                                         rc = SCRUB_NEXT_NOLMA;
 944                                 else
 945                                         rc = 0;
 946                         } else {
 947                                 /* It may be FID-on-OST, or may be FID for
 948                                  * non-MDT0, anyway, we do not know how to
 949                                  * generate its FID, ignore it directly. */
 950                                 rc = SCRUB_NEXT_CONTINUE;
 951                         }
 952                         return rc;
 953                 }
 954
 955                 /* For OI scrub case only: the object has LMA but has no ff
 956                  * (or ff crashed). It may be MDT-object, may be OST-object
 957                  * with crashed ff. The last check is local FLDB. */
 958                 rc = osd_scrub_check_local_fldb(info, dev, fid);
 959         }
 960
 961         return rc;
 962 }
 963
 964 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 965                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 966                         struct super_block *sb, bool scrub)
 967 {
 968         struct inode *inode;
 969         int           rc;
 970         ENTRY;
 971
 972         /* Not handle the backend root object and agent parent object.
 973          * They are neither visible to namespace nor have OI mappings. */
 974         if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
 975                      pos == osd_remote_parent_ino(dev)))
 976                 RETURN(SCRUB_NEXT_CONTINUE);
 977
 978         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 979         inode = osd_iget(info, dev, lid);
 980         if (IS_ERR(inode)) {
 981                 rc = PTR_ERR(inode);
 982                 /* The inode may be removed after bitmap searching, or the
 983                  * file is new created without inode initialized yet. */
 984                 if (rc == -ENOENT || rc == -ESTALE)
 985                         RETURN(SCRUB_NEXT_CONTINUE);
 986
 987                 CDEBUG(D_LFSCK, "%s: fail to read inode, ino# = %u: "
 988                        "rc = %d\n", osd_dev2name(dev), pos, rc);
 989                 RETURN(rc);
 990         }
 991
 992         /* It is an EA inode, no OI mapping for it, skip it. */
 993         if (osd_is_ea_inode(inode))
 994                 GOTO(put, rc = SCRUB_NEXT_CONTINUE);
 995
 996         if (scrub &&
 997             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 998                 /* Only skip it for the first OI scrub accessing. */
 999                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
1000                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
1001         }
1002
1003         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
1004
1005         GOTO(put, rc);
1006
1007 put:
1008         iput(inode);
1009         return rc;
1010 }
1011
1012 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
1013                           struct osd_iit_param *param,
1014                           struct osd_idmap_cache **oic, const bool noslot)
1015 {
1016         struct osd_scrub     *scrub  = &dev->od_scrub;
1017         struct ptlrpc_thread *thread = &scrub->os_thread;
1018         struct lu_fid        *fid;
1019         struct osd_inode_id  *lid;
1020         int                   rc;
1021
1022         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
1023                 struct l_wait_info lwi;
1024
1025                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
1026                 if (likely(lwi.lwi_timeout > 0))
1027                         l_wait_event(thread->t_ctl_waitq,
1028                                 !list_empty(&scrub->os_inconsistent_items) ||
1029                                 !thread_is_running(thread),
1030                                 &lwi);
1031         }
1032
1033         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
1034                 spin_lock(&scrub->os_lock);
1035                 thread_set_flags(thread, SVC_STOPPING);
1036                 spin_unlock(&scrub->os_lock);
1037                 return SCRUB_NEXT_CRASH;
1038         }
1039
1040         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
1041                 return SCRUB_NEXT_FATAL;
1042
1043         if (unlikely(!thread_is_running(thread)))
1044                 return SCRUB_NEXT_EXIT;
1045
1046         if (!list_empty(&scrub->os_inconsistent_items)) {
1047                 spin_lock(&scrub->os_lock);
1048                 if (likely(!list_empty(&scrub->os_inconsistent_items))) {
1049                         struct osd_inconsistent_item *oii;
1050
1051                         oii = list_entry(scrub->os_inconsistent_items.next,
1052                                 struct osd_inconsistent_item, oii_list);
1053                         spin_unlock(&scrub->os_lock);
1054
1055                         *oic = &oii->oii_cache;
1056                         scrub->os_in_prior = 1;
1057
1058                         return 0;
1059                 }
1060                 spin_unlock(&scrub->os_lock);
1061         }
1062
1063         if (noslot)
1064                 return SCRUB_NEXT_WAIT;
1065
1066         rc = osd_iit_next(param, &scrub->os_pos_current);
1067         if (rc != 0)
1068                 return rc;
1069
1070         *oic = &scrub->os_oic;
1071         fid = &(*oic)->oic_fid;
1072         lid = &(*oic)->oic_lid;
1073         rc = osd_iit_iget(info, dev, fid, lid,
1074                           scrub->os_pos_current, param->sb, true);
1075         return rc;
1076 }
1077
1078 static int osd_preload_next(struct osd_thread_info *info,
1079                             struct osd_device *dev, struct osd_iit_param *param,
1080                             struct osd_idmap_cache **oic, const bool noslot)
1081 {
1082         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
1083         struct osd_scrub        *scrub;
1084         struct ptlrpc_thread    *thread;
1085         int                      rc;
1086
1087         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1088         if (rc != 0)
1089                 return rc;
1090
1091         scrub = &dev->od_scrub;
1092         thread = &scrub->os_thread;
1093         if (thread_is_running(thread) &&
1094             ooc->ooc_pos_preload >= scrub->os_pos_current)
1095                 return SCRUB_NEXT_EXIT;
1096
1097         rc = osd_iit_iget(info, dev,
1098                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1099                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1100                           ooc->ooc_pos_preload, param->sb, false);
1101         return rc;
1102 }
1103
1104 static inline int
1105 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1106 {
1107         spin_lock(&scrub->os_lock);
1108         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1109             !list_empty(&scrub->os_inconsistent_items) ||
1110             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1111                 scrub->os_waiting = 0;
1112         else
1113                 scrub->os_waiting = 1;
1114         spin_unlock(&scrub->os_lock);
1115
1116         return !scrub->os_waiting;
1117 }
1118
1119 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1120                           struct osd_iit_param *param,
1121                           struct osd_idmap_cache *oic, bool *noslot, int rc)
1122 {
1123         struct l_wait_info       lwi    = { 0 };
1124         struct osd_scrub        *scrub  = &dev->od_scrub;
1125         struct scrub_file       *sf     = &scrub->os_file;
1126         struct ptlrpc_thread    *thread = &scrub->os_thread;
1127         struct osd_otable_it    *it     = dev->od_otable_it;
1128         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
1129
1130         switch (rc) {
1131         case SCRUB_NEXT_NOSCRUB:
1132                 down_write(&scrub->os_rwsem);
1133                 scrub->os_new_checked++;
1134                 sf->sf_items_noscrub++;
1135                 up_write(&scrub->os_rwsem);
1136         case SCRUB_NEXT_CONTINUE:
1137         case SCRUB_NEXT_WAIT:
1138                 goto wait;
1139         }
1140
1141         rc = osd_scrub_check_update(info, dev, oic, rc);
1142         if (rc != 0) {
1143                 scrub->os_in_prior = 0;
1144                 return rc;
1145         }
1146
1147         rc = osd_scrub_checkpoint(scrub);
1148         if (rc != 0) {
1149                 CDEBUG(D_LFSCK, "%s: fail to checkpoint, pos = %u: "
1150                        "rc = %d\n", osd_scrub2name(scrub),
1151                        scrub->os_pos_current, rc);
1152                 /* Continue, as long as the scrub itself can go ahead. */
1153         }
1154
1155         if (scrub->os_in_prior) {
1156                 scrub->os_in_prior = 0;
1157                 return 0;
1158         }
1159
1160 wait:
1161         if (it != NULL && it->ooi_waiting && ooc != NULL &&
1162             ooc->ooc_pos_preload < scrub->os_pos_current) {
1163                 spin_lock(&scrub->os_lock);
1164                 it->ooi_waiting = 0;
1165                 wake_up_all(&thread->t_ctl_waitq);
1166                 spin_unlock(&scrub->os_lock);
1167         }
1168
1169         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1170                 return 0;
1171
1172         if (!ooc || osd_scrub_has_window(scrub, ooc)) {
1173                 *noslot = false;
1174                 return 0;
1175         }
1176
1177         if (it != NULL)
1178                 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1179                              &lwi);
1180
1181         if (!ooc || osd_scrub_has_window(scrub, ooc))
1182                 *noslot = false;
1183         else
1184                 *noslot = true;
1185         return 0;
1186 }
1187
1188 static int osd_preload_exec(struct osd_thread_info *info,
1189                             struct osd_device *dev, struct osd_iit_param *param,
1190                             struct osd_idmap_cache *oic, bool *noslot, int rc)
1191 {
1192         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1193
1194         if (rc == 0) {
1195                 ooc->ooc_cached_items++;
1196                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1197                                         ~OSD_OTABLE_IT_CACHE_MASK;
1198         }
1199         return rc > 0 ? 0 : rc;
1200 }
1201
1202 #define SCRUB_IT_ALL    1
1203 #define SCRUB_IT_CRASH  2
1204
1205 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1206                            bool inconsistent)
1207 {
1208         struct osd_scrub     *scrub  = &dev->od_scrub;
1209         struct ptlrpc_thread *thread = &scrub->os_thread;
1210         struct scrub_file    *sf     = &scrub->os_file;
1211         int                   rc;
1212         ENTRY;
1213
1214         LASSERT(!(flags & SS_AUTO_PARTIAL));
1215
1216         down_write(&scrub->os_rwsem);
1217         scrub->os_in_join = 1;
1218         if (flags & SS_SET_FAILOUT)
1219                 sf->sf_param |= SP_FAILOUT;
1220         else if (flags & SS_CLEAR_FAILOUT)
1221                 sf->sf_param &= ~SP_FAILOUT;
1222
1223         if (flags & SS_SET_DRYRUN)
1224                 sf->sf_param |= SP_DRYRUN;
1225         else if (flags & SS_CLEAR_DRYRUN)
1226                 sf->sf_param &= ~SP_DRYRUN;
1227
1228         if (flags & SS_RESET) {
1229                 osd_scrub_file_reset(scrub,
1230                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1231                         inconsistent ? SF_INCONSISTENT : 0);
1232                 sf->sf_status = SS_SCANNING;
1233         }
1234
1235         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1236                 scrub->os_full_speed = 1;
1237         else
1238                 scrub->os_full_speed = 0;
1239
1240         if (flags & SS_AUTO_FULL) {
1241                 sf->sf_flags |= SF_AUTO;
1242                 scrub->os_full_speed = 1;
1243         }
1244
1245         scrub->os_new_checked = 0;
1246         if (sf->sf_pos_last_checkpoint != 0)
1247                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1248         else
1249                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1250
1251         scrub->os_pos_current = sf->sf_pos_latest_start;
1252         sf->sf_time_latest_start = cfs_time_current_sec();
1253         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1254         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1255         rc = osd_scrub_file_store(scrub);
1256         if (rc != 0)
1257                 CDEBUG(D_LFSCK, "%s: fail to store scrub file when join "
1258                        "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1259
1260         spin_lock(&scrub->os_lock);
1261         scrub->os_waiting = 0;
1262         scrub->os_paused = 0;
1263         scrub->os_partial_scan = 0;
1264         scrub->os_in_join = 0;
1265         scrub->os_full_scrub = 0;
1266         spin_unlock(&scrub->os_lock);
1267         wake_up_all(&thread->t_ctl_waitq);
1268         up_write(&scrub->os_rwsem);
1269
1270         EXIT;
1271 }
1272
1273 static int osd_inode_iteration(struct osd_thread_info *info,
1274                                struct osd_device *dev, __u32 max, bool preload)
1275 {
1276         struct osd_scrub     *scrub  = &dev->od_scrub;
1277         struct ptlrpc_thread *thread = &scrub->os_thread;
1278         struct scrub_file    *sf     = &scrub->os_file;
1279         osd_iit_next_policy   next;
1280         osd_iit_exec_policy   exec;
1281         __u32                *pos;
1282         __u32                *count;
1283         struct osd_iit_param  param  = { NULL };
1284         struct l_wait_info    lwi    = { 0 };
1285         __u32                 limit;
1286         int                   rc;
1287         bool                  noslot = true;
1288         ENTRY;
1289
1290         param.sb = osd_sb(dev);
1291         if (preload)
1292                 goto full;
1293
1294         while (scrub->os_partial_scan && !scrub->os_in_join) {
1295                 struct osd_idmap_cache *oic = NULL;
1296
1297                 rc = osd_scrub_next(info, dev, &param, &oic, noslot);
1298                 switch (rc) {
1299                 case SCRUB_NEXT_EXIT:
1300                         RETURN(0);
1301                 case SCRUB_NEXT_CRASH:
1302                         RETURN(SCRUB_IT_CRASH);
1303                 case SCRUB_NEXT_FATAL:
1304                         RETURN(-EINVAL);
1305                 case SCRUB_NEXT_WAIT: {
1306                         struct kstatfs *ksfs = &info->oti_ksfs;
1307                         __u64 saved_flags;
1308
1309                         if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1310                             unlikely(sf->sf_items_updated_prior == 0))
1311                                 goto wait;
1312
1313                         if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1314                             scrub->os_full_scrub) {
1315                                 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1316                                                true);
1317                                 goto full;
1318                         }
1319
1320                         rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1321                         if (rc == 0) {
1322                                 __u64 used = ksfs->f_files - ksfs->f_ffree;
1323
1324                                 do_div(used, sf->sf_items_updated_prior);
1325                                 /* If we hit too much inconsistent OI
1326                                  * mappings during the partial scan,
1327                                  * then scan the device completely. */
1328                                 if (used < dev->od_full_scrub_ratio) {
1329                                         osd_scrub_join(dev,
1330                                                 SS_AUTO_FULL | SS_RESET, true);
1331                                         goto full;
1332                                 }
1333                         }
1334
1335 wait:
1336                         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1337                             cfs_fail_val > 0)
1338                                 continue;
1339
1340                         saved_flags = sf->sf_flags;
1341                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1342                                           SF_UPGRADE | SF_AUTO);
1343                         sf->sf_status = SS_COMPLETED;
1344                         l_wait_event(thread->t_ctl_waitq,
1345                                      !thread_is_running(thread) ||
1346                                      !scrub->os_partial_scan ||
1347                                      scrub->os_in_join ||
1348                                      !list_empty(&scrub->os_inconsistent_items),
1349                                      &lwi);
1350                         sf->sf_flags = saved_flags;
1351                         sf->sf_status = SS_SCANNING;
1352
1353                         if (unlikely(!thread_is_running(thread)))
1354                                 RETURN(0);
1355
1356                         if (!scrub->os_partial_scan || scrub->os_in_join)
1357                                 goto full;
1358
1359                         continue;
1360                 }
1361                 default:
1362                         LASSERTF(rc == 0, "rc = %d\n", rc);
1363
1364                         osd_scrub_exec(info, dev, &param, oic, &noslot, rc);
1365                         break;
1366                 }
1367         }
1368
1369 full:
1370         if (!preload) {
1371                 l_wait_event(thread->t_ctl_waitq,
1372                              !thread_is_running(thread) || !scrub->os_in_join,
1373                              &lwi);
1374
1375                 if (unlikely(!thread_is_running(thread)))
1376                         RETURN(0);
1377         }
1378
1379         noslot = false;
1380         if (!preload) {
1381                 next = osd_scrub_next;
1382                 exec = osd_scrub_exec;
1383                 pos = &scrub->os_pos_current;
1384                 count = &scrub->os_new_checked;
1385         } else {
1386                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1387
1388                 next = osd_preload_next;
1389                 exec = osd_preload_exec;
1390                 pos = &ooc->ooc_pos_preload;
1391                 count = &ooc->ooc_cached_items;
1392         }
1393
1394         rc = 0;
1395         param.start = *pos;
1396         param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1397         param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1398         param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1399         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1400
1401         while (*pos <= limit && *count < max) {
1402                 struct ldiskfs_group_desc *desc;
1403                 bool next_group = false;
1404
1405                 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1406                 if (!desc)
1407                         RETURN(-EIO);
1408
1409                 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1410                         next_group = true;
1411                         goto next_group;
1412                 }
1413
1414                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1415                 if (!param.bitmap) {
1416                         CERROR("%s: fail to read bitmap for %u, "
1417                                "scrub will stop, urgent mode\n",
1418                                osd_scrub2name(scrub), (__u32)param.bg);
1419                         RETURN(-EIO);
1420                 }
1421
1422                 do {
1423                         struct osd_idmap_cache *oic = NULL;
1424
1425                         if (param.offset +
1426                                 ldiskfs_itable_unused_count(param.sb, desc) >=
1427                             LDISKFS_INODES_PER_GROUP(param.sb)) {
1428                                 next_group = true;
1429                                 goto next_group;
1430                         }
1431
1432                         rc = next(info, dev, &param, &oic, noslot);
1433                         switch (rc) {
1434                         case SCRUB_NEXT_BREAK:
1435                                 next_group = true;
1436                                 goto next_group;
1437                         case SCRUB_NEXT_EXIT:
1438                                 brelse(param.bitmap);
1439                                 RETURN(0);
1440                         case SCRUB_NEXT_CRASH:
1441                                 brelse(param.bitmap);
1442                                 RETURN(SCRUB_IT_CRASH);
1443                         case SCRUB_NEXT_FATAL:
1444                                 brelse(param.bitmap);
1445                                 RETURN(-EINVAL);
1446                         }
1447
1448                         rc = exec(info, dev, &param, oic, &noslot, rc);
1449                 } while (!rc && *pos <= limit && *count < max);
1450
1451 next_group:
1452                 if (param.bitmap) {
1453                         brelse(param.bitmap);
1454                         param.bitmap = NULL;
1455                 }
1456
1457                 if (rc < 0)
1458                         GOTO(out, rc);
1459
1460                 if (next_group) {
1461                         param.bg++;
1462                         param.offset = 0;
1463                         param.gbase = 1 +
1464                                 param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1465                         *pos = param.gbase;
1466                         param.start = *pos;
1467                 }
1468         }
1469
1470         if (*pos > limit)
1471                 RETURN(SCRUB_IT_ALL);
1472
1473 out:
1474         /* For preload case, increase the iteration cursor,
1475          * then we will not scan the last object repeatedly. */
1476         if (preload)
1477                 dev->od_otable_it->ooi_cache.ooc_pos_preload++;
1478
1479         RETURN(rc);
1480 }
1481
1482 static int osd_otable_it_preload(const struct lu_env *env,
1483                                  struct osd_otable_it *it)
1484 {
1485         struct osd_device       *dev   = it->ooi_dev;
1486         struct osd_scrub        *scrub = &dev->od_scrub;
1487         struct osd_otable_cache *ooc   = &it->ooi_cache;
1488         int                      rc;
1489         ENTRY;
1490
1491         rc = osd_inode_iteration(osd_oti_get(env), dev,
1492                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1493         if (rc == SCRUB_IT_ALL)
1494                 it->ooi_all_cached = 1;
1495
1496         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1497                 spin_lock(&scrub->os_lock);
1498                 scrub->os_waiting = 0;
1499                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1500                 spin_unlock(&scrub->os_lock);
1501         }
1502
1503         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1504 }
1505
1506 static int osd_scrub_main(void *args)
1507 {
1508         struct lu_env         env;
1509         struct osd_device    *dev    = (struct osd_device *)args;
1510         struct osd_scrub     *scrub  = &dev->od_scrub;
1511         struct ptlrpc_thread *thread = &scrub->os_thread;
1512         int                   rc;
1513         ENTRY;
1514
1515         rc = lu_env_init(&env, LCT_LOCAL);
1516         if (rc != 0) {
1517                 CDEBUG(D_LFSCK, "%s: OI scrub fail to init env: rc = %d\n",
1518                        osd_scrub2name(scrub), rc);
1519                 GOTO(noenv, rc);
1520         }
1521
1522         rc = osd_scrub_prep(dev);
1523         if (rc != 0) {
1524                 CDEBUG(D_LFSCK, "%s: OI scrub fail to scrub prep: rc = %d\n",
1525                        osd_scrub2name(scrub), rc);
1526                 GOTO(out, rc);
1527         }
1528
1529         if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1530                 struct l_wait_info lwi = { 0 };
1531                 struct osd_otable_it *it = dev->od_otable_it;
1532                 struct osd_otable_cache *ooc = &it->ooi_cache;
1533
1534                 l_wait_event(thread->t_ctl_waitq,
1535                              it->ooi_user_ready || !thread_is_running(thread),
1536                              &lwi);
1537                 if (unlikely(!thread_is_running(thread)))
1538                         GOTO(post, rc = 0);
1539
1540                 scrub->os_pos_current = ooc->ooc_pos_preload;
1541         }
1542
1543         CDEBUG(D_LFSCK, "%s: OI scrub start, flags = 0x%x, pos = %u\n",
1544                osd_scrub2name(scrub), scrub->os_start_flags,
1545                scrub->os_pos_current);
1546
1547         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1548         if (unlikely(rc == SCRUB_IT_CRASH))
1549                 GOTO(out, rc = -EINVAL);
1550         GOTO(post, rc);
1551
1552 post:
1553         rc = osd_scrub_post(scrub, rc);
1554         CDEBUG(D_LFSCK, "%s: OI scrub: stop, pos = %u: rc = %d\n",
1555                osd_scrub2name(scrub), scrub->os_pos_current, rc);
1556
1557 out:
1558         while (!list_empty(&scrub->os_inconsistent_items)) {
1559                 struct osd_inconsistent_item *oii;
1560
1561                 oii = list_entry(scrub->os_inconsistent_items.next,
1562                                      struct osd_inconsistent_item, oii_list);
1563                 list_del_init(&oii->oii_list);
1564                 OBD_FREE_PTR(oii);
1565         }
1566         lu_env_fini(&env);
1567
1568 noenv:
1569         spin_lock(&scrub->os_lock);
1570         thread_set_flags(thread, SVC_STOPPED);
1571         wake_up_all(&thread->t_ctl_waitq);
1572         spin_unlock(&scrub->os_lock);
1573         return rc;
1574 }
1575
1576 /* initial OI scrub */
1577
1578 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1579                          struct dentry *, filldir_t filldir);
1580
1581 #ifdef HAVE_FILLDIR_USE_CTX
1582 static int osd_ios_varfid_fill(struct dir_context *buf, const char *name,
1583                                int namelen, loff_t offset, __u64 ino,
1584                                unsigned d_type);
1585 static int osd_ios_lf_fill(struct dir_context *buf, const char *name,
1586                            int namelen, loff_t offset, __u64 ino,
1587                            unsigned d_type);
1588 static int osd_ios_dl_fill(struct dir_context *buf, const char *name,
1589                            int namelen, loff_t offset, __u64 ino,
1590                            unsigned d_type);
1591 static int osd_ios_uld_fill(struct dir_context *buf, const char *name,
1592                             int namelen, loff_t offset, __u64 ino,
1593                             unsigned d_type);
1594 #else
1595 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1596                                loff_t offset, __u64 ino, unsigned d_type);
1597 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1598                            loff_t offset, __u64 ino, unsigned d_type);
1599 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1600                            loff_t offset, __u64 ino, unsigned d_type);
1601 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1602                             loff_t offset, __u64 ino, unsigned d_type);
1603 #endif
1604
1605 static int
1606 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1607                      struct dentry *dentry, filldir_t filldir);
1608 static int
1609 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1610                   struct dentry *dentry, filldir_t filldir);
1611
1612 static int
1613 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1614                      struct dentry *dentry, filldir_t filldir);
1615
1616 enum osd_lf_flags {
1617         OLF_SCAN_SUBITEMS       = 0x0001,
1618         OLF_HIDE_FID            = 0x0002,
1619         OLF_SHOW_NAME           = 0x0004,
1620         OLF_NO_OI               = 0x0008,
1621         OLF_IDX_IN_FID          = 0x0010,
1622 };
1623
1624 struct osd_lf_map {
1625         char            *olm_name;
1626         struct lu_fid    olm_fid;
1627         __u16            olm_flags;
1628         __u16            olm_namelen;
1629         scandir_t        olm_scandir;
1630         filldir_t        olm_filldir;
1631 };
1632
1633 /* Add the new introduced local files in the list in the future. */
1634 static const struct osd_lf_map osd_lf_maps[] = {
1635         /* CATALOGS */
1636         {
1637                 .olm_name       = CATLIST,
1638                 .olm_fid        = {
1639                         .f_seq  = FID_SEQ_LOCAL_FILE,
1640                         .f_oid  = LLOG_CATALOGS_OID,
1641                 },
1642                 .olm_flags      = OLF_SHOW_NAME,
1643                 .olm_namelen    = sizeof(CATLIST) - 1,
1644         },
1645
1646         /* CONFIGS */
1647         {
1648                 .olm_name       = MOUNT_CONFIGS_DIR,
1649                 .olm_fid        = {
1650                         .f_seq  = FID_SEQ_LOCAL_FILE,
1651                         .f_oid  = MGS_CONFIGS_OID,
1652                 },
1653                 .olm_flags      = OLF_SCAN_SUBITEMS,
1654                 .olm_namelen    = sizeof(MOUNT_CONFIGS_DIR) - 1,
1655                 .olm_scandir    = osd_ios_general_scan,
1656                 .olm_filldir    = osd_ios_varfid_fill,
1657         },
1658
1659         /* NIDTBL_VERSIONS */
1660         {
1661                 .olm_name       = MGS_NIDTBL_DIR,
1662                 .olm_flags      = OLF_SCAN_SUBITEMS,
1663                 .olm_namelen    = sizeof(MGS_NIDTBL_DIR) - 1,
1664                 .olm_scandir    = osd_ios_general_scan,
1665                 .olm_filldir    = osd_ios_varfid_fill,
1666         },
1667
1668         /* PENDING */
1669         {
1670                 .olm_name       = "PENDING",
1671                 .olm_namelen    = sizeof("PENDING") - 1,
1672         },
1673
1674         /* ROOT */
1675         {
1676                 .olm_name       = "ROOT",
1677                 .olm_fid        = {
1678                         .f_seq  = FID_SEQ_ROOT,
1679                         .f_oid  = FID_OID_ROOT,
1680                 },
1681                 .olm_flags      = OLF_SCAN_SUBITEMS | OLF_HIDE_FID,
1682                 .olm_namelen    = sizeof("ROOT") - 1,
1683                 .olm_scandir    = osd_ios_ROOT_scan,
1684         },
1685
1686         /* changelog_catalog */
1687         {
1688                 .olm_name       = CHANGELOG_CATALOG,
1689                 .olm_namelen    = sizeof(CHANGELOG_CATALOG) - 1,
1690         },
1691
1692         /* changelog_users */
1693         {
1694                 .olm_name       = CHANGELOG_USERS,
1695                 .olm_namelen    = sizeof(CHANGELOG_USERS) - 1,
1696         },
1697
1698         /* fld */
1699         {
1700                 .olm_name       = "fld",
1701                 .olm_fid        = {
1702                         .f_seq  = FID_SEQ_LOCAL_FILE,
1703                         .f_oid  = FLD_INDEX_OID,
1704                 },
1705                 .olm_flags      = OLF_SHOW_NAME,
1706                 .olm_namelen    = sizeof("fld") - 1,
1707         },
1708
1709         /* last_rcvd */
1710         {
1711                 .olm_name       = LAST_RCVD,
1712                 .olm_fid        = {
1713                         .f_seq  = FID_SEQ_LOCAL_FILE,
1714                         .f_oid  = LAST_RECV_OID,
1715                 },
1716                 .olm_flags      = OLF_SHOW_NAME,
1717                 .olm_namelen    = sizeof(LAST_RCVD) - 1,
1718         },
1719
1720         /* reply_data */
1721         {
1722                 .olm_name       = REPLY_DATA,
1723                 .olm_fid        = {
1724                         .f_seq  = FID_SEQ_LOCAL_FILE,
1725                         .f_oid  = REPLY_DATA_OID,
1726                 },
1727                 .olm_flags      = OLF_SHOW_NAME,
1728                 .olm_namelen    = sizeof(REPLY_DATA) - 1,
1729         },
1730
1731         /* lov_objid */
1732         {
1733                 .olm_name       = LOV_OBJID,
1734                 .olm_fid        = {
1735                         .f_seq  = FID_SEQ_LOCAL_FILE,
1736                         .f_oid  = MDD_LOV_OBJ_OID,
1737                 },
1738                 .olm_flags      = OLF_SHOW_NAME,
1739                 .olm_namelen    = sizeof(LOV_OBJID) - 1,
1740         },
1741
1742         /* lov_objseq */
1743         {
1744                 .olm_name       = LOV_OBJSEQ,
1745                 .olm_fid        = {
1746                         .f_seq  = FID_SEQ_LOCAL_FILE,
1747                         .f_oid  = MDD_LOV_OBJ_OSEQ,
1748                 },
1749                 .olm_flags      = OLF_SHOW_NAME,
1750                 .olm_namelen    = sizeof(LOV_OBJSEQ) - 1,
1751         },
1752
1753         /* quota_master */
1754         {
1755                 .olm_name       = QMT_DIR,
1756                 .olm_flags      = OLF_SCAN_SUBITEMS,
1757                 .olm_namelen    = sizeof(QMT_DIR) - 1,
1758                 .olm_scandir    = osd_ios_general_scan,
1759                 .olm_filldir    = osd_ios_varfid_fill,
1760         },
1761
1762         /* quota_slave */
1763         {
1764                 .olm_name       = QSD_DIR,
1765                 .olm_flags      = OLF_SCAN_SUBITEMS,
1766                 .olm_namelen    = sizeof(QSD_DIR) - 1,
1767                 .olm_scandir    = osd_ios_general_scan,
1768                 .olm_filldir    = osd_ios_varfid_fill,
1769         },
1770
1771         /* seq_ctl */
1772         {
1773                 .olm_name       = "seq_ctl",
1774                 .olm_fid        = {
1775                         .f_seq  = FID_SEQ_LOCAL_FILE,
1776                         .f_oid  = FID_SEQ_CTL_OID,
1777                 },
1778                 .olm_flags      = OLF_SHOW_NAME,
1779                 .olm_namelen    = sizeof("seq_ctl") - 1,
1780         },
1781
1782         /* seq_srv */
1783         {
1784                 .olm_name       = "seq_srv",
1785                 .olm_fid        = {
1786                         .f_seq  = FID_SEQ_LOCAL_FILE,
1787                         .f_oid  = FID_SEQ_SRV_OID,
1788                 },
1789                 .olm_flags      = OLF_SHOW_NAME,
1790                 .olm_namelen    = sizeof("seq_srv") - 1,
1791         },
1792
1793         /* health_check */
1794         {
1795                 .olm_name       = HEALTH_CHECK,
1796                 .olm_fid        = {
1797                         .f_seq  = FID_SEQ_LOCAL_FILE,
1798                         .f_oid  = OFD_HEALTH_CHECK_OID,
1799                 },
1800                 .olm_flags      = OLF_SHOW_NAME,
1801                 .olm_namelen    = sizeof(HEALTH_CHECK) - 1,
1802         },
1803
1804         /* LFSCK */
1805         {
1806                 .olm_name       = LFSCK_DIR,
1807                 .olm_namelen    = sizeof(LFSCK_DIR) - 1,
1808                 .olm_scandir    = osd_ios_general_scan,
1809                 .olm_filldir    = osd_ios_varfid_fill,
1810         },
1811
1812         /* lfsck_bookmark */
1813         {
1814                 .olm_name       = LFSCK_BOOKMARK,
1815                 .olm_namelen    = sizeof(LFSCK_BOOKMARK) - 1,
1816         },
1817
1818         /* lfsck_layout */
1819         {
1820                 .olm_name       = LFSCK_LAYOUT,
1821                 .olm_namelen    = sizeof(LFSCK_LAYOUT) - 1,
1822         },
1823
1824         /* lfsck_namespace */
1825         {
1826                 .olm_name       = LFSCK_NAMESPACE,
1827                 .olm_namelen    = sizeof(LFSCK_NAMESPACE) - 1,
1828         },
1829
1830         /* OBJECTS, upgrade from old device */
1831         {
1832                 .olm_name       = OBJECTS,
1833                 .olm_flags      = OLF_SCAN_SUBITEMS,
1834                 .olm_namelen    = sizeof(OBJECTS) - 1,
1835                 .olm_scandir    = osd_ios_OBJECTS_scan,
1836         },
1837
1838         /* lquota_v2.user, upgrade from old device */
1839         {
1840                 .olm_name       = "lquota_v2.user",
1841                 .olm_namelen    = sizeof("lquota_v2.user") - 1,
1842         },
1843
1844         /* lquota_v2.group, upgrade from old device */
1845         {
1846                 .olm_name       = "lquota_v2.group",
1847                 .olm_namelen    = sizeof("lquota_v2.group") - 1,
1848         },
1849
1850         /* LAST_GROUP, upgrade from old device */
1851         {
1852                 .olm_name       = "LAST_GROUP",
1853                 .olm_fid        = {
1854                         .f_seq  = FID_SEQ_LOCAL_FILE,
1855                         .f_oid  = OFD_LAST_GROUP_OID,
1856                 },
1857                 .olm_flags      = OLF_SHOW_NAME,
1858                 .olm_namelen    = sizeof("LAST_GROUP") - 1,
1859         },
1860
1861         /* committed batchid for cross-MDT operation */
1862         {
1863                 .olm_name       = "BATCHID",
1864                 .olm_fid        = {
1865                         .f_seq  = FID_SEQ_LOCAL_FILE,
1866                         .f_oid  = BATCHID_COMMITTED_OID,
1867                 },
1868                 .olm_flags      = OLF_SHOW_NAME,
1869                 .olm_namelen    = sizeof("BATCHID") - 1,
1870         },
1871
1872         /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1873          * and f_oid = index for their log files.  See lu_update_log{_dir}_fid()
1874          * for more details. */
1875
1876         /* update_log */
1877         {
1878                 .olm_name       = "update_log",
1879                 .olm_fid        = {
1880                         .f_seq  = FID_SEQ_UPDATE_LOG,
1881                 },
1882                 .olm_flags      = OLF_SHOW_NAME | OLF_IDX_IN_FID,
1883                 .olm_namelen    = sizeof("update_log") - 1,
1884         },
1885
1886         /* update_log_dir */
1887         {
1888                 .olm_name       = "update_log_dir",
1889                 .olm_fid        = {
1890                         .f_seq  = FID_SEQ_UPDATE_LOG_DIR,
1891                 },
1892                 .olm_flags      = OLF_SHOW_NAME | OLF_SCAN_SUBITEMS |
1893                                   OLF_IDX_IN_FID,
1894                 .olm_namelen    = sizeof("update_log_dir") - 1,
1895                 .olm_scandir    = osd_ios_general_scan,
1896                 .olm_filldir    = osd_ios_uld_fill,
1897         },
1898
1899         /* lost+found */
1900         {
1901                 .olm_name       = "lost+found",
1902                 .olm_fid        = {
1903                         .f_seq  = FID_SEQ_LOCAL_FILE,
1904                         .f_oid  = OSD_LPF_OID,
1905                 },
1906                 .olm_flags      = OLF_SCAN_SUBITEMS,
1907                 .olm_namelen    = sizeof("lost+found") - 1,
1908                 .olm_scandir    = osd_ios_general_scan,
1909                 .olm_filldir    = osd_ios_lf_fill,
1910         },
1911
1912         {
1913                 .olm_name       = NULL
1914         }
1915 };
1916
1917 /* Add the new introduced files under .lustre/ in the list in the future. */
1918 static const struct osd_lf_map osd_dl_maps[] = {
1919         /* .lustre/fid */
1920         {
1921                 .olm_name       = "fid",
1922                 .olm_fid        = {
1923                         .f_seq  = FID_SEQ_DOT_LUSTRE,
1924                         .f_oid  = FID_OID_DOT_LUSTRE_OBF,
1925                 },
1926                 .olm_namelen    = sizeof("fid") - 1,
1927         },
1928         /* .lustre/lost+found */
1929         {
1930                 .olm_name       = "lost+found",
1931                 .olm_fid        = {
1932                         .f_seq  = FID_SEQ_DOT_LUSTRE,
1933                         .f_oid  = FID_OID_DOT_LUSTRE_LPF,
1934                 },
1935                 .olm_namelen    = sizeof("lost+found") - 1,
1936         },
1937         {
1938                 .olm_name       = NULL
1939         }
1940 };
1941
1942 struct osd_ios_item {
1943         struct list_head oii_list;
1944         struct dentry   *oii_dentry;
1945         scandir_t        oii_scandir;
1946         filldir_t        oii_filldir;
1947 };
1948
1949 struct osd_ios_filldir_buf {
1950 #ifdef HAVE_DIR_CONTEXT
1951         /* please keep it as first member */
1952         struct dir_context       ctx;
1953 #endif
1954         struct osd_thread_info  *oifb_info;
1955         struct osd_device       *oifb_dev;
1956         struct dentry           *oifb_dentry;
1957 };
1958
1959 static inline struct dentry *
1960 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1961 {
1962         struct dentry *dentry;
1963
1964         dentry = ll_lookup_one_len(name, parent, namelen);
1965         if (IS_ERR(dentry)) {
1966                 int rc = PTR_ERR(dentry);
1967
1968                 if (rc != -ENOENT)
1969                         CERROR("Fail to find %.*s in %.*s (%lu/%u): rc = %d\n",
1970                                namelen, name, parent->d_name.len,
1971                                parent->d_name.name, parent->d_inode->i_ino,
1972                                parent->d_inode->i_generation, rc);
1973
1974                 return dentry;
1975         }
1976
1977         if (dentry->d_inode == NULL) {
1978                 dput(dentry);
1979                 return ERR_PTR(-ENOENT);
1980         }
1981
1982         return dentry;
1983 }
1984
1985 static int
1986 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1987                  scandir_t scandir, filldir_t filldir)
1988 {
1989         struct osd_ios_item *item;
1990         ENTRY;
1991
1992         OBD_ALLOC_PTR(item);
1993         if (item == NULL)
1994                 RETURN(-ENOMEM);
1995
1996         INIT_LIST_HEAD(&item->oii_list);
1997         item->oii_dentry = dget(dentry);
1998         item->oii_scandir = scandir;
1999         item->oii_filldir = filldir;
2000         list_add_tail(&item->oii_list, &dev->od_ios_list);
2001
2002         RETURN(0);
2003 }
2004
2005 /**
2006  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
2007  *
2008  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
2009  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
2010  * reference the inode, or fixed if it is missing or references another inode.
2011  */
2012 static int
2013 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
2014                  struct inode *inode, const struct lu_fid *fid, int flags)
2015 {
2016         struct lustre_mdt_attrs *lma    = &info->oti_ost_attrs.loa_lma;
2017         struct osd_inode_id     *id     = &info->oti_id;
2018         struct osd_inode_id     *id2    = &info->oti_id2;
2019         struct osd_scrub        *scrub  = &dev->od_scrub;
2020         struct scrub_file       *sf     = &scrub->os_file;
2021         struct lu_fid            tfid;
2022         int                      rc;
2023         ENTRY;
2024
2025         rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
2026                          &info->oti_ost_attrs);
2027         if (rc != 0 && rc != -ENODATA) {
2028                 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
2029                        "rc = %d\n", osd_name(dev), rc);
2030
2031                 RETURN(rc);
2032         }
2033
2034         osd_id_gen(id, inode->i_ino, inode->i_generation);
2035         if (rc == -ENODATA) {
2036                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
2037                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
2038                 } else {
2039                         tfid = *fid;
2040                         if (flags & OLF_IDX_IN_FID) {
2041                                 LASSERT(dev->od_index >= 0);
2042
2043                                 tfid.f_oid = dev->od_index;
2044                         }
2045                 }
2046                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
2047                 if (rc != 0) {
2048                         CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
2049                               "scrub: rc = %d\n", osd_name(dev), rc);
2050
2051                         RETURN(rc);
2052                 }
2053         } else {
2054                 if (lma->lma_compat & LMAC_NOT_IN_OI)
2055                         RETURN(0);
2056
2057                 tfid = lma->lma_self_fid;
2058         }
2059
2060         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
2061         if (rc != 0) {
2062                 if (rc != -ENOENT)
2063                         RETURN(rc);
2064
2065                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
2066                                                DTO_INDEX_INSERT, true, 0, NULL);
2067                 if (rc > 0)
2068                         rc = 0;
2069
2070                 RETURN(rc);
2071         }
2072
2073         if (osd_id_eq_strict(id, id2))
2074                 RETURN(0);
2075
2076         if (!(sf->sf_flags & SF_INCONSISTENT)) {
2077                 osd_scrub_file_reset(scrub,
2078                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2079                                      SF_INCONSISTENT);
2080                 rc = osd_scrub_file_store(scrub);
2081                 if (rc != 0)
2082                         RETURN(rc);
2083         }
2084
2085         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
2086                                        DTO_INDEX_UPDATE, true, 0, NULL);
2087         if (rc > 0)
2088                 rc = 0;
2089
2090         RETURN(rc);
2091 }
2092
2093 /**
2094  * It scans the /lost+found, and for the OST-object (with filter_fid
2095  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
2096  */
2097 #ifdef HAVE_FILLDIR_USE_CTX
2098 static int osd_ios_lf_fill(struct dir_context *buf,
2099 #else
2100 static int osd_ios_lf_fill(void *buf,
2101 #endif
2102                            const char *name, int namelen,
2103                            loff_t offset, __u64 ino, unsigned d_type)
2104 {
2105         struct osd_ios_filldir_buf *fill_buf =
2106                 (struct osd_ios_filldir_buf *)buf;
2107         struct osd_thread_info     *info     = fill_buf->oifb_info;
2108         struct osd_device          *dev      = fill_buf->oifb_dev;
2109         struct lu_fid              *fid      = &info->oti_fid;
2110         struct osd_scrub           *scrub    = &dev->od_scrub;
2111         struct dentry              *parent   = fill_buf->oifb_dentry;
2112         struct dentry              *child;
2113         struct inode               *dir      = parent->d_inode;
2114         struct inode               *inode;
2115         int                         rc;
2116         ENTRY;
2117
2118         /* skip any '.' started names */
2119         if (name[0] == '.')
2120                 RETURN(0);
2121
2122         scrub->os_lf_scanned++;
2123         child = osd_ios_lookup_one_len(name, parent, namelen);
2124         if (IS_ERR(child)) {
2125                 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
2126                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
2127                 RETURN(0);
2128         }
2129
2130         inode = child->d_inode;
2131         if (S_ISDIR(inode->i_mode)) {
2132                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2133                                       osd_ios_lf_fill);
2134                 if (rc != 0)
2135                         CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
2136                               "rc = %d\n", osd_name(dev), namelen, name, rc);
2137                 GOTO(put, rc);
2138         }
2139
2140         if (!S_ISREG(inode->i_mode))
2141                 GOTO(put, rc = 0);
2142
2143         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
2144         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
2145                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
2146                 if (rc == 0) {
2147                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
2148                                "/lost+found.\n", namelen, name, PFID(fid));
2149                         scrub->os_lf_repaired++;
2150                 } else {
2151                         CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
2152                                DFID": rc = %d\n",
2153                                osd_name(dev), namelen, name, PFID(fid), rc);
2154                 }
2155         }
2156
2157         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
2158          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
2159          *      can process them in furtuer. */
2160
2161         GOTO(put, rc);
2162
2163 put:
2164         if (rc < 0)
2165                 scrub->os_lf_failed++;
2166         dput(child);
2167         /* skip the failure to make the scanning to continue. */
2168         return 0;
2169 }
2170
2171 #ifdef HAVE_FILLDIR_USE_CTX
2172 static int osd_ios_varfid_fill(struct dir_context *buf,
2173 #else
2174 static int osd_ios_varfid_fill(void *buf,
2175 #endif
2176                                const char *name, int namelen,
2177                                loff_t offset, __u64 ino, unsigned d_type)
2178 {
2179         struct osd_ios_filldir_buf *fill_buf =
2180                 (struct osd_ios_filldir_buf *)buf;
2181         struct osd_device          *dev      = fill_buf->oifb_dev;
2182         struct dentry              *child;
2183         int                         rc;
2184         ENTRY;
2185
2186         /* skip any '.' started names */
2187         if (name[0] == '.')
2188                 RETURN(0);
2189
2190         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2191         if (IS_ERR(child))
2192                 RETURN(PTR_ERR(child));
2193
2194         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2195                               NULL, 0);
2196         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
2197                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2198                                       osd_ios_varfid_fill);
2199         dput(child);
2200
2201         RETURN(rc);
2202 }
2203
2204 #ifdef HAVE_FILLDIR_USE_CTX
2205 static int osd_ios_dl_fill(struct dir_context *buf,
2206 #else
2207 static int osd_ios_dl_fill(void *buf,
2208 #endif
2209                            const char *name, int namelen,
2210                            loff_t offset, __u64 ino, unsigned d_type)
2211 {
2212         struct osd_ios_filldir_buf *fill_buf =
2213                 (struct osd_ios_filldir_buf *)buf;
2214         struct osd_device          *dev      = fill_buf->oifb_dev;
2215         const struct osd_lf_map    *map;
2216         struct dentry              *child;
2217         int                         rc       = 0;
2218         ENTRY;
2219
2220         /* skip any '.' started names */
2221         if (name[0] == '.')
2222                 RETURN(0);
2223
2224         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
2225                 if (map->olm_namelen != namelen)
2226                         continue;
2227
2228                 if (strncmp(map->olm_name, name, namelen) == 0)
2229                         break;
2230         }
2231
2232         if (map->olm_name == NULL)
2233                 RETURN(0);
2234
2235         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2236         if (IS_ERR(child))
2237                 RETURN(PTR_ERR(child));
2238
2239         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2240                               &map->olm_fid, map->olm_flags);
2241         dput(child);
2242
2243         RETURN(rc);
2244 }
2245
2246 #ifdef HAVE_FILLDIR_USE_CTX
2247 static int osd_ios_uld_fill(struct dir_context *buf,
2248 #else
2249 static int osd_ios_uld_fill(void *buf,
2250 #endif
2251                             const char *name, int namelen,
2252                             loff_t offset, __u64 ino, unsigned d_type)
2253 {
2254         struct osd_ios_filldir_buf *fill_buf =
2255                 (struct osd_ios_filldir_buf *)buf;
2256         struct dentry              *child;
2257         struct lu_fid               tfid;
2258         int                         rc       = 0;
2259         ENTRY;
2260
2261         /* skip any non-DFID format name */
2262         if (name[0] != '[')
2263                 RETURN(0);
2264
2265         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2266         if (IS_ERR(child))
2267                 RETURN(PTR_ERR(child));
2268
2269         /* skip the start '[' */
2270         sscanf(&name[1], SFID, RFID(&tfid));
2271         if (fid_is_sane(&tfid))
2272                 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
2273                                       child->d_inode, &tfid, 0);
2274         else
2275                 rc = -EIO;
2276         dput(child);
2277
2278         RETURN(rc);
2279 }
2280
2281 #ifdef HAVE_FILLDIR_USE_CTX
2282 static int osd_ios_root_fill(struct dir_context *buf,
2283 #else
2284 static int osd_ios_root_fill(void *buf,
2285 #endif
2286                              const char *name, int namelen,
2287                              loff_t offset, __u64 ino, unsigned d_type)
2288 {
2289         struct osd_ios_filldir_buf *fill_buf =
2290                 (struct osd_ios_filldir_buf *)buf;
2291         struct osd_device          *dev      = fill_buf->oifb_dev;
2292         const struct osd_lf_map    *map;
2293         struct dentry              *child;
2294         int                         rc       = 0;
2295         ENTRY;
2296
2297         /* skip any '.' started names */
2298         if (name[0] == '.')
2299                 RETURN(0);
2300
2301         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2302                 if (map->olm_namelen != namelen)
2303                         continue;
2304
2305                 if (strncmp(map->olm_name, name, namelen) == 0)
2306                         break;
2307         }
2308
2309         if (map->olm_name == NULL)
2310                 RETURN(0);
2311
2312         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2313         if (IS_ERR(child))
2314                 RETURN(PTR_ERR(child));
2315
2316         if (!(map->olm_flags & OLF_NO_OI))
2317                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2318                                       &map->olm_fid, map->olm_flags);
2319         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2320                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2321                                       map->olm_filldir);
2322         dput(child);
2323
2324         RETURN(rc);
2325 }
2326
2327 static int
2328 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2329                      struct dentry *dentry, filldir_t filldir)
2330 {
2331         struct osd_ios_filldir_buf    buf   = {
2332 #ifdef HAVE_DIR_CONTEXT
2333                                                 .ctx.actor = filldir,
2334 #endif
2335                                                 .oifb_info = info,
2336                                                 .oifb_dev = dev,
2337                                                 .oifb_dentry = dentry };
2338         struct file                  *filp  = &info->oti_file;
2339         struct inode                 *inode = dentry->d_inode;
2340         const struct file_operations *fops  = inode->i_fop;
2341         int                           rc;
2342         ENTRY;
2343
2344         LASSERT(filldir != NULL);
2345
2346         filp->f_pos = 0;
2347         filp->f_path.dentry = dentry;
2348         filp->f_mode = FMODE_64BITHASH;
2349         filp->f_mapping = inode->i_mapping;
2350         filp->f_op = fops;
2351         filp->private_data = NULL;
2352         set_file_inode(filp, inode);
2353
2354 #ifdef HAVE_DIR_CONTEXT
2355         buf.ctx.pos = filp->f_pos;
2356         rc = fops->iterate(filp, &buf.ctx);
2357         filp->f_pos = buf.ctx.pos;
2358 #else
2359         rc = fops->readdir(filp, &buf, filldir);
2360 #endif
2361         fops->release(inode, filp);
2362
2363         RETURN(rc);
2364 }
2365
2366 static int
2367 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2368                   struct dentry *dentry, filldir_t filldir)
2369 {
2370         struct osd_scrub  *scrub  = &dev->od_scrub;
2371         struct scrub_file *sf     = &scrub->os_file;
2372         struct dentry     *child;
2373         int                rc;
2374         ENTRY;
2375
2376         /* It is existing MDT0 device. We only allow the case of object without
2377          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2378          * can generate IGIF mode FID for the object and related OI mapping. If
2379          * it is on other MDTs, then becuase file-level backup/restore, related
2380          * OI mapping may be invalid already, we do not know which is the right
2381          * FID for the object. We only allow IGIF objects to reside on the MDT0.
2382          *
2383          * XXX: For the case of object on non-MDT0 device with neither LMA nor
2384          *      "fid" xattr, then something crashed. We cannot re-generate the
2385          *      FID directly, instead, the OI scrub will scan the OI structure
2386          *      and try to re-generate the LMA from the OI mapping. But if the
2387          *      OI mapping crashed or lost also, then we have to give up under
2388          *      double failure cases. */
2389         scrub->os_convert_igif = 1;
2390         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2391                                        strlen(dot_lustre_name));
2392         if (IS_ERR(child)) {
2393                 rc = PTR_ERR(child);
2394                 if (rc == -ENOENT) {
2395                         /* It is 1.8 MDT device. */
2396                         if (!(sf->sf_flags & SF_UPGRADE)) {
2397                                 osd_scrub_file_reset(scrub,
2398                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2399                                         SF_UPGRADE);
2400                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2401                                 rc = osd_scrub_file_store(scrub);
2402                         } else {
2403                                 rc = 0;
2404                         }
2405                 }
2406         } else {
2407                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2408                  * so the client will get IGIF for the ".lustre" object when
2409                  * the MDT restart.
2410                  *
2411                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2412                  * it does not know whether there are some old clients cached
2413                  * the ".lustre" IGIF during the upgrading. Two choices:
2414                  *
2415                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2416                  *    It will allow the old connected clients to access the
2417                  *    ".lustre" with cached IGIF. But it will cause others
2418                  *    on the MDT failed to check "fid_is_dot_lustre()".
2419                  *
2420                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2421                  *    for ".lustre" in spite of whether there are some clients
2422                  *    cached the ".lustre" IGIF or not. It enables the check
2423                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
2424                  *    that the old connected clients cannot access the ".lustre"
2425                  *    with the cached IGIF.
2426                  *
2427                  * Usually, it is rare case for the old connected clients
2428                  * to access the ".lustre" with cached IGIF. So we prefer
2429                  * to the solution 2). */
2430                 rc = osd_ios_scan_one(info, dev, child->d_inode,
2431                                       &LU_DOT_LUSTRE_FID, 0);
2432                 if (rc == 0)
2433                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2434                                               osd_ios_dl_fill);
2435                 dput(child);
2436         }
2437
2438         RETURN(rc);
2439 }
2440
2441 static int
2442 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2443                      struct dentry *dentry, filldir_t filldir)
2444 {
2445         struct osd_scrub  *scrub  = &dev->od_scrub;
2446         struct scrub_file *sf     = &scrub->os_file;
2447         struct dentry     *child;
2448         int                rc;
2449         ENTRY;
2450
2451         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2452                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2453                 rc = osd_scrub_file_store(scrub);
2454                 if (rc != 0)
2455                         RETURN(rc);
2456         }
2457
2458         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2459         if (!IS_ERR(child)) {
2460                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2461                 dput(child);
2462         } else {
2463                 rc = PTR_ERR(child);
2464         }
2465
2466         if (rc != 0 && rc != -ENOENT)
2467                 RETURN(rc);
2468
2469         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2470         if (!IS_ERR(child)) {
2471                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2472                 dput(child);
2473         } else {
2474                 rc = PTR_ERR(child);
2475         }
2476
2477         if (rc == -ENOENT)
2478                 rc = 0;
2479
2480         RETURN(rc);
2481 }
2482
2483 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2484                                 struct osd_device *dev)
2485 {
2486         struct osd_ios_item     *item    = NULL;
2487         scandir_t                scandir = osd_ios_general_scan;
2488         filldir_t                filldir = osd_ios_root_fill;
2489         struct dentry           *dentry  = osd_sb(dev)->s_root;
2490         const struct osd_lf_map *map     = osd_lf_maps;
2491         int                      rc;
2492         ENTRY;
2493
2494         /* Lookup IGIF in OI by force for initial OI scrub. */
2495         dev->od_igif_inoi = 1;
2496
2497         while (1) {
2498                 rc = scandir(info, dev, dentry, filldir);
2499                 if (item != NULL) {
2500                         dput(item->oii_dentry);
2501                         OBD_FREE_PTR(item);
2502                 }
2503
2504                 if (rc != 0)
2505                         break;
2506
2507                 if (list_empty(&dev->od_ios_list))
2508                         break;
2509
2510                 item = list_entry(dev->od_ios_list.next,
2511                                   struct osd_ios_item, oii_list);
2512                 list_del_init(&item->oii_list);
2513
2514                 LASSERT(item->oii_scandir != NULL);
2515                 scandir = item->oii_scandir;
2516                 filldir = item->oii_filldir;
2517                 dentry = item->oii_dentry;
2518         }
2519
2520         while (!list_empty(&dev->od_ios_list)) {
2521                 item = list_entry(dev->od_ios_list.next,
2522                                   struct osd_ios_item, oii_list);
2523                 list_del_init(&item->oii_list);
2524                 dput(item->oii_dentry);
2525                 OBD_FREE_PTR(item);
2526         }
2527
2528         if (rc != 0)
2529                 RETURN(rc);
2530
2531         /* There maybe the case that the object has been removed, but its OI
2532          * mapping is still in the OI file, such as the "CATALOGS" after MDT
2533          * file-level backup/restore. So here cleanup the stale OI mappings. */
2534         while (map->olm_name != NULL) {
2535                 struct dentry *child;
2536
2537                 if (fid_is_zero(&map->olm_fid)) {
2538                         map++;
2539                         continue;
2540                 }
2541
2542                 child = osd_ios_lookup_one_len(map->olm_name,
2543                                                osd_sb(dev)->s_root,
2544                                                map->olm_namelen);
2545                 if (!IS_ERR(child))
2546                         dput(child);
2547                 else if (PTR_ERR(child) == -ENOENT)
2548                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2549                                                   NULL, DTO_INDEX_DELETE,
2550                                                   true, 0, NULL);
2551                 map++;
2552         }
2553
2554         RETURN(0);
2555 }
2556
2557 char *osd_lf_fid2name(const struct lu_fid *fid)
2558 {
2559         const struct osd_lf_map *map = osd_lf_maps;
2560
2561         while (map->olm_name != NULL) {
2562                 if (!lu_fid_eq(fid, &map->olm_fid)) {
2563                         map++;
2564                         continue;
2565                 }
2566
2567                 if (map->olm_flags & OLF_SHOW_NAME)
2568                         return map->olm_name;
2569                 else
2570                         return "";
2571         }
2572
2573         return NULL;
2574 }
2575
2576 /* OI scrub start/stop */
2577
2578 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2579 {
2580         struct osd_scrub     *scrub  = &dev->od_scrub;
2581         struct ptlrpc_thread *thread = &scrub->os_thread;
2582         struct l_wait_info    lwi    = { 0 };
2583         struct task_struct   *task;
2584         int                   rc;
2585         ENTRY;
2586
2587         if (dev->od_dt_dev.dd_rdonly)
2588                 RETURN(-EROFS);
2589
2590         /* os_lock: sync status between stop and scrub thread */
2591         spin_lock(&scrub->os_lock);
2592
2593 again:
2594         if (thread_is_running(thread)) {
2595                 spin_unlock(&scrub->os_lock);
2596                 if (!(scrub->os_file.sf_flags & SF_AUTO ||
2597                       scrub->os_partial_scan) ||
2598                      (flags & SS_AUTO_PARTIAL))
2599                         RETURN(-EALREADY);
2600
2601                 osd_scrub_join(dev, flags, false);
2602                 spin_lock(&scrub->os_lock);
2603                 if (!thread_is_running(thread))
2604                         goto again;
2605
2606                 spin_unlock(&scrub->os_lock);
2607                 RETURN(0);
2608         }
2609
2610         if (unlikely(thread_is_stopping(thread))) {
2611                 spin_unlock(&scrub->os_lock);
2612                 l_wait_event(thread->t_ctl_waitq,
2613                              thread_is_stopped(thread),
2614                              &lwi);
2615                 spin_lock(&scrub->os_lock);
2616                 goto again;
2617         }
2618         spin_unlock(&scrub->os_lock);
2619
2620         if (scrub->os_file.sf_status == SS_COMPLETED) {
2621                 if (!(flags & SS_SET_FAILOUT))
2622                         flags |= SS_CLEAR_FAILOUT;
2623
2624                 if (!(flags & SS_SET_DRYRUN))
2625                         flags |= SS_CLEAR_DRYRUN;
2626
2627                 flags |= SS_RESET;
2628         }
2629
2630         scrub->os_start_flags = flags;
2631         thread_set_flags(thread, 0);
2632         task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2633         if (IS_ERR(task)) {
2634                 rc = PTR_ERR(task);
2635                 CERROR("%s: cannot start iteration thread: rc = %d\n",
2636                        osd_scrub2name(scrub), rc);
2637                 RETURN(rc);
2638         }
2639
2640         l_wait_event(thread->t_ctl_waitq,
2641                      thread_is_running(thread) || thread_is_stopped(thread),
2642                      &lwi);
2643
2644         RETURN(0);
2645 }
2646
2647 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2648 {
2649         int rc;
2650         ENTRY;
2651
2652         /* od_otable_mutex: prevent curcurrent start/stop */
2653         mutex_lock(&dev->od_otable_mutex);
2654         rc = do_osd_scrub_start(dev, flags);
2655         mutex_unlock(&dev->od_otable_mutex);
2656
2657         RETURN(rc == -EALREADY ? 0 : rc);
2658 }
2659
2660 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2661 {
2662         struct ptlrpc_thread *thread = &scrub->os_thread;
2663         struct l_wait_info    lwi    = { 0 };
2664
2665         /* os_lock: sync status between stop and scrub thread */
2666         spin_lock(&scrub->os_lock);
2667         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2668                 thread_set_flags(thread, SVC_STOPPING);
2669                 spin_unlock(&scrub->os_lock);
2670                 wake_up_all(&thread->t_ctl_waitq);
2671                 l_wait_event(thread->t_ctl_waitq,
2672                              thread_is_stopped(thread),
2673                              &lwi);
2674                 /* Do not skip the last lock/unlock, which can guarantee that
2675                  * the caller cannot return until the OI scrub thread exit. */
2676                 spin_lock(&scrub->os_lock);
2677         }
2678         spin_unlock(&scrub->os_lock);
2679 }
2680
2681 static void osd_scrub_stop(struct osd_device *dev)
2682 {
2683         /* od_otable_mutex: prevent curcurrent start/stop */
2684         mutex_lock(&dev->od_otable_mutex);
2685         dev->od_scrub.os_paused = 1;
2686         do_osd_scrub_stop(&dev->od_scrub);
2687         mutex_unlock(&dev->od_otable_mutex);
2688 }
2689
2690 /* OI scrub setup/cleanup */
2691
2692 static const char osd_scrub_name[] = "OI_scrub";
2693
2694 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2695 {
2696         struct osd_thread_info     *info   = osd_oti_get(env);
2697         struct osd_scrub           *scrub  = &dev->od_scrub;
2698         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
2699         struct scrub_file          *sf     = &scrub->os_file;
2700         struct super_block         *sb     = osd_sb(dev);
2701         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
2702         struct lvfs_run_ctxt        saved;
2703         struct file                *filp;
2704         struct inode               *inode;
2705         struct lu_fid              *fid    = &info->oti_fid;
2706         bool                        dirty  = false;
2707         bool                        restored = false;
2708         int                         rc     = 0;
2709         ENTRY;
2710
2711         memset(scrub, 0, sizeof(*scrub));
2712         OBD_SET_CTXT_MAGIC(ctxt);
2713         ctxt->pwdmnt = dev->od_mnt;
2714         ctxt->pwd = dev->od_mnt->mnt_root;
2715         ctxt->fs = get_ds();
2716
2717         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2718         init_rwsem(&scrub->os_rwsem);
2719         spin_lock_init(&scrub->os_lock);
2720         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2721
2722         push_ctxt(&saved, ctxt);
2723         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2724         if (IS_ERR(filp)) {
2725                 pop_ctxt(&saved, ctxt);
2726                 RETURN(PTR_ERR(filp));
2727         }
2728
2729         inode = file_inode(filp);
2730         /* 'What the @fid is' is not imporatant, because the object
2731          * has no OI mapping, and only is visible inside the OSD.*/
2732         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2733         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2734         if (rc != 0) {
2735                 filp_close(filp, NULL);
2736                 pop_ctxt(&saved, ctxt);
2737                 RETURN(rc);
2738         }
2739
2740         scrub->os_inode = igrab(inode);
2741         filp_close(filp, NULL);
2742         pop_ctxt(&saved, ctxt);
2743
2744         rc = osd_scrub_file_load(scrub);
2745         if (rc == -ENOENT) {
2746                 osd_scrub_file_init(scrub, es->s_uuid);
2747                 /* If the "/O" dir does not exist when mount (indicated by
2748                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2749                  * then it is quite probably that the device is a new one,
2750                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2751                  *
2752                  * For the rare case that "/O" and "OI_scrub" both lost on
2753                  * an old device, it can be found and cleared later.
2754                  *
2755                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2756                  * need to check "filter_fid_old" and to convert it to
2757                  * "filter_fid" for each object, and all the IGIF should
2758                  * have their FID mapping in OI files already. */
2759                 if (dev->od_maybe_new)
2760                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2761                 dirty = true;
2762         } else if (rc != 0) {
2763                 GOTO(cleanup_inode, rc);
2764         } else {
2765                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2766                         struct obd_uuid *old_uuid;
2767                         struct obd_uuid *new_uuid;
2768
2769                         OBD_ALLOC_PTR(old_uuid);
2770                         OBD_ALLOC_PTR(new_uuid);
2771                         if (old_uuid == NULL || new_uuid == NULL) {
2772                                 CERROR("%s: UUID has been changed, but"
2773                                        "failed to allocate RAM for report\n",
2774                                        osd_dev2name(dev));
2775                         } else {
2776                                 class_uuid_unparse(sf->sf_uuid, old_uuid);
2777                                 class_uuid_unparse(es->s_uuid, new_uuid);
2778                                 CERROR("%s: UUID has been changed from "
2779                                        "%s to %s\n", osd_dev2name(dev),
2780                                        old_uuid->uuid, new_uuid->uuid);
2781                         }
2782                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2783                         dirty = true;
2784                         restored = true;
2785                         if (old_uuid != NULL)
2786                                 OBD_FREE_PTR(old_uuid);
2787                         if (new_uuid != NULL)
2788                                 OBD_FREE_PTR(new_uuid);
2789                 } else if (sf->sf_status == SS_SCANNING) {
2790                         sf->sf_status = SS_CRASHED;
2791                         dirty = true;
2792                 }
2793         }
2794
2795         if (sf->sf_pos_last_checkpoint != 0)
2796                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2797         else
2798                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2799
2800         if (dirty) {
2801                 rc = osd_scrub_file_store(scrub);
2802                 if (rc != 0)
2803                         GOTO(cleanup_inode, rc);
2804         }
2805
2806         /* Initialize OI files. */
2807         rc = osd_oi_init(info, dev, restored);
2808         if (rc < 0)
2809                 GOTO(cleanup_inode, rc);
2810
2811         rc = osd_initial_OI_scrub(info, dev);
2812         if (rc != 0)
2813                 GOTO(cleanup_oi, rc);
2814
2815         if (sf->sf_flags & SF_UPGRADE ||
2816             !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2817               sf->sf_success_count > 0)) {
2818                 dev->od_igif_inoi = 0;
2819                 dev->od_check_ff = dev->od_is_ost;
2820         } else {
2821                 dev->od_igif_inoi = 1;
2822                 dev->od_check_ff = 0;
2823         }
2824
2825         if (sf->sf_flags & SF_INCONSISTENT)
2826                 /* The 'od_igif_inoi' will be set under the
2827                  * following cases:
2828                  * 1) new created system, or
2829                  * 2) restored from file-level backup, or
2830                  * 3) the upgrading completed.
2831                  *
2832                  * The 'od_igif_inoi' may be cleared by OI scrub
2833                  * later if found that the system is upgrading. */
2834                 dev->od_igif_inoi = 1;
2835
2836         if (!dev->od_dt_dev.dd_rdonly && !dev->od_noscrub &&
2837             ((sf->sf_status == SS_PAUSED) ||
2838              (sf->sf_status == SS_CRASHED &&
2839               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2840                               SF_UPGRADE | SF_AUTO)) ||
2841              (sf->sf_status == SS_INIT &&
2842               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2843                               SF_UPGRADE))))
2844                 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2845
2846         if (rc != 0)
2847                 GOTO(cleanup_oi, rc);
2848
2849         /* it is possible that dcache entries may keep objects after they are
2850          * deleted by OSD. While it looks safe this can cause object data to
2851          * stay until umount causing failures in tests calculating free space,
2852          * e.g. replay-ost-single. Since those dcache entries are not used
2853          * anymore let's just free them after use here */
2854         shrink_dcache_sb(sb);
2855
2856         RETURN(0);
2857 cleanup_oi:
2858         osd_oi_fini(info, dev);
2859 cleanup_inode:
2860         iput(scrub->os_inode);
2861         scrub->os_inode = NULL;
2862
2863         return rc;
2864 }
2865
2866 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2867 {
2868         struct osd_scrub *scrub = &dev->od_scrub;
2869
2870         LASSERT(dev->od_otable_it == NULL);
2871
2872         if (scrub->os_inode != NULL) {
2873                 osd_scrub_stop(dev);
2874                 iput(scrub->os_inode);
2875                 scrub->os_inode = NULL;
2876         }
2877         if (dev->od_oi_table != NULL)
2878                 osd_oi_fini(osd_oti_get(env), dev);
2879 }
2880
2881 /* object table based iteration APIs */
2882
2883 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2884                                        struct dt_object *dt, __u32 attr)
2885 {
2886         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2887         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2888         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2889         struct osd_scrub       *scrub = &dev->od_scrub;
2890         struct osd_otable_it   *it;
2891         __u32                   start = 0;
2892         int                     rc;
2893         ENTRY;
2894
2895         /* od_otable_mutex: prevent curcurrent init/fini */
2896         mutex_lock(&dev->od_otable_mutex);
2897         if (dev->od_otable_it != NULL)
2898                 GOTO(out, it = ERR_PTR(-EALREADY));
2899
2900         OBD_ALLOC_PTR(it);
2901         if (it == NULL)
2902                 GOTO(out, it = ERR_PTR(-ENOMEM));
2903
2904         dev->od_otable_it = it;
2905         it->ooi_dev = dev;
2906         it->ooi_cache.ooc_consumer_idx = -1;
2907         if (flags & DOIF_OUTUSED)
2908                 it->ooi_used_outside = 1;
2909
2910         if (flags & DOIF_RESET)
2911                 start |= SS_RESET;
2912
2913         if (valid & DOIV_ERROR_HANDLE) {
2914                 if (flags & DOIF_FAILOUT)
2915                         start |= SS_SET_FAILOUT;
2916                 else
2917                         start |= SS_CLEAR_FAILOUT;
2918         }
2919
2920         if (valid & DOIV_DRYRUN) {
2921                 if (flags & DOIF_DRYRUN)
2922                         start |= SS_SET_DRYRUN;
2923                 else
2924                         start |= SS_CLEAR_DRYRUN;
2925         }
2926
2927         rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2928         if (rc < 0 && rc != -EALREADY) {
2929                 dev->od_otable_it = NULL;
2930                 OBD_FREE_PTR(it);
2931                 GOTO(out, it = ERR_PTR(rc));
2932         }
2933
2934         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2935
2936         GOTO(out, it);
2937
2938 out:
2939         mutex_unlock(&dev->od_otable_mutex);
2940         return (struct dt_it *)it;
2941 }
2942
2943 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2944 {
2945         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2946         struct osd_device    *dev = it->ooi_dev;
2947
2948         /* od_otable_mutex: prevent curcurrent init/fini */
2949         mutex_lock(&dev->od_otable_mutex);
2950         do_osd_scrub_stop(&dev->od_scrub);
2951         LASSERT(dev->od_otable_it == it);
2952
2953         dev->od_otable_it = NULL;
2954         mutex_unlock(&dev->od_otable_mutex);
2955         OBD_FREE_PTR(it);
2956 }
2957
2958 static int osd_otable_it_get(const struct lu_env *env,
2959                              struct dt_it *di, const struct dt_key *key)
2960 {
2961         return 0;
2962 }
2963
2964 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2965 {
2966 }
2967
2968 static inline int
2969 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2970 {
2971         spin_lock(&scrub->os_lock);
2972         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2973             scrub->os_waiting ||
2974             !thread_is_running(&scrub->os_thread))
2975                 it->ooi_waiting = 0;
2976         else
2977                 it->ooi_waiting = 1;
2978         spin_unlock(&scrub->os_lock);
2979
2980         return !it->ooi_waiting;
2981 }
2982
2983 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2984 {
2985         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2986         struct osd_device       *dev    = it->ooi_dev;
2987         struct osd_scrub        *scrub  = &dev->od_scrub;
2988         struct osd_otable_cache *ooc    = &it->ooi_cache;
2989         struct ptlrpc_thread    *thread = &scrub->os_thread;
2990         struct l_wait_info       lwi    = { 0 };
2991         int                      rc;
2992         ENTRY;
2993
2994         LASSERT(it->ooi_user_ready);
2995
2996 again:
2997         if (!thread_is_running(thread) && !it->ooi_used_outside)
2998                 RETURN(1);
2999
3000         if (ooc->ooc_cached_items > 0) {
3001                 ooc->ooc_cached_items--;
3002                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
3003                                         ~OSD_OTABLE_IT_CACHE_MASK;
3004                 RETURN(0);
3005         }
3006
3007         if (it->ooi_all_cached) {
3008                 l_wait_event(thread->t_ctl_waitq,
3009                              !thread_is_running(thread),
3010                              &lwi);
3011                 RETURN(1);
3012         }
3013
3014         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
3015                 spin_lock(&scrub->os_lock);
3016                 scrub->os_waiting = 0;
3017                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
3018                 spin_unlock(&scrub->os_lock);
3019         }
3020
3021         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
3022                 l_wait_event(thread->t_ctl_waitq,
3023                              osd_otable_it_wakeup(scrub, it),
3024                              &lwi);
3025
3026         if (!thread_is_running(thread) && !it->ooi_used_outside)
3027                 RETURN(1);
3028
3029         rc = osd_otable_it_preload(env, it);
3030         if (rc >= 0)
3031                 goto again;
3032
3033         RETURN(rc);
3034 }
3035
3036 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
3037                                         const struct dt_it *di)
3038 {
3039         return NULL;
3040 }
3041
3042 static int osd_otable_it_key_size(const struct lu_env *env,
3043                                   const struct dt_it *di)
3044 {
3045         return sizeof(__u64);
3046 }
3047
3048 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
3049                              struct dt_rec *rec, __u32 attr)
3050 {
3051         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
3052         struct osd_otable_cache *ooc = &it->ooi_cache;
3053
3054         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
3055
3056         /* Filter out Invald FID already. */
3057         LASSERTF(fid_is_sane((struct lu_fid *)rec),
3058                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
3059                  PFID((struct lu_fid *)rec),
3060                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
3061
3062         return 0;
3063 }
3064
3065 static __u64 osd_otable_it_store(const struct lu_env *env,
3066                                  const struct dt_it *di)
3067 {
3068         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
3069         struct osd_otable_cache *ooc = &it->ooi_cache;
3070         __u64                    hash;
3071
3072         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
3073                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
3074         else
3075                 hash = ooc->ooc_pos_preload;
3076         return hash;
3077 }
3078
3079 /**
3080  * Set the OSD layer iteration start position as the specified hash.
3081  */
3082 static int osd_otable_it_load(const struct lu_env *env,
3083                               const struct dt_it *di, __u64 hash)
3084 {
3085         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
3086         struct osd_device       *dev   = it->ooi_dev;
3087         struct osd_otable_cache *ooc   = &it->ooi_cache;
3088         struct osd_scrub        *scrub = &dev->od_scrub;
3089         int                      rc;
3090         ENTRY;
3091
3092         /* Forbid to set iteration position after iteration started. */
3093         if (it->ooi_user_ready)
3094                 RETURN(-EPERM);
3095
3096         LASSERT(!scrub->os_partial_scan);
3097
3098         if (hash > OSD_OTABLE_MAX_HASH)
3099                 hash = OSD_OTABLE_MAX_HASH;
3100
3101         /* The hash is the last checkpoint position,
3102          * we will start from the next one. */
3103         ooc->ooc_pos_preload = hash + 1;
3104         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
3105                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
3106
3107         it->ooi_user_ready = 1;
3108         if (!scrub->os_full_speed)
3109                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
3110
3111         /* Unplug OSD layer iteration by the first next() call. */
3112         rc = osd_otable_it_next(env, (struct dt_it *)it);
3113
3114         RETURN(rc);
3115 }
3116
3117 static int osd_otable_it_key_rec(const struct lu_env *env,
3118                                  const struct dt_it *di, void *key_rec)
3119 {
3120         return 0;
3121 }
3122
3123 const struct dt_index_operations osd_otable_ops = {
3124         .dio_it = {
3125                 .init     = osd_otable_it_init,
3126                 .fini     = osd_otable_it_fini,
3127                 .get      = osd_otable_it_get,
3128                 .put      = osd_otable_it_put,
3129                 .next     = osd_otable_it_next,
3130                 .key      = osd_otable_it_key,
3131                 .key_size = osd_otable_it_key_size,
3132                 .rec      = osd_otable_it_rec,
3133                 .store    = osd_otable_it_store,
3134                 .load     = osd_otable_it_load,
3135                 .key_rec  = osd_otable_it_key_rec,
3136         }
3137 };
3138
3139 /* high priority inconsistent items list APIs */
3140
3141 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL  60
3142
3143 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
3144                    int insert)
3145 {
3146         struct osd_inconsistent_item *oii;
3147         struct osd_scrub             *scrub  = &dev->od_scrub;
3148         struct ptlrpc_thread         *thread = &scrub->os_thread;
3149         int                           wakeup = 0;
3150         ENTRY;
3151
3152         OBD_ALLOC_PTR(oii);
3153         if (unlikely(oii == NULL))
3154                 RETURN(-ENOMEM);
3155
3156         INIT_LIST_HEAD(&oii->oii_list);
3157         oii->oii_cache = *oic;
3158         oii->oii_insert = insert;
3159
3160         if (scrub->os_partial_scan) {
3161                 __u64 now = cfs_time_current_sec();
3162
3163                 /* If there haven't been errors in a long time,
3164                  * decay old count until either the errors are
3165                  * gone or we reach the current interval. */
3166                 while (unlikely(scrub->os_bad_oimap_count > 0 &&
3167                                 scrub->os_bad_oimap_time +
3168                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
3169                         scrub->os_bad_oimap_count >>= 1;
3170                         scrub->os_bad_oimap_time +=
3171                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
3172                 }
3173
3174                 scrub->os_bad_oimap_time = now;
3175                 if (++scrub->os_bad_oimap_count >
3176                     dev->od_full_scrub_threshold_rate)
3177                         scrub->os_full_scrub = 1;
3178         }
3179
3180         spin_lock(&scrub->os_lock);
3181         if (unlikely(!thread_is_running(thread))) {
3182                 spin_unlock(&scrub->os_lock);
3183                 OBD_FREE_PTR(oii);
3184                 RETURN(-EAGAIN);
3185         }
3186
3187         if (list_empty(&scrub->os_inconsistent_items))
3188                 wakeup = 1;
3189         list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
3190         spin_unlock(&scrub->os_lock);
3191
3192         if (wakeup != 0)
3193                 wake_up_all(&thread->t_ctl_waitq);
3194
3195         RETURN(0);
3196 }
3197
3198 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
3199                    struct osd_inode_id *id)
3200 {
3201         struct osd_scrub             *scrub = &dev->od_scrub;
3202         struct osd_inconsistent_item *oii;
3203         ENTRY;
3204
3205         spin_lock(&scrub->os_lock);
3206         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
3207                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
3208                         *id = oii->oii_cache.oic_lid;
3209                         spin_unlock(&scrub->os_lock);
3210                         RETURN(0);
3211                 }
3212         }
3213         spin_unlock(&scrub->os_lock);
3214
3215         RETURN(-ENOENT);
3216 }
3217
3218 /* OI scrub dump */
3219
3220 static const char *scrub_status_names[] = {
3221         "init",
3222         "scanning",
3223         "completed",
3224         "failed",
3225         "stopped",
3226         "paused",
3227         "crashed",
3228         NULL
3229 };
3230
3231 static const char *scrub_flags_names[] = {
3232         "recreated",
3233         "inconsistent",
3234         "auto",
3235         "upgrade",
3236         NULL
3237 };
3238
3239 static const char *scrub_param_names[] = {
3240         "failout",
3241         "dryrun",
3242         NULL
3243 };
3244
3245 static void scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
3246                             const char *prefix)
3247 {
3248         int flag;
3249         int i;
3250
3251         seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
3252
3253         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
3254                 if (flag & bits) {
3255                         bits &= ~flag;
3256                         seq_printf(m, "%s%c", names[i],
3257                                    bits != 0 ? ',' : '\n');
3258                 }
3259         }
3260 }
3261
3262 static void scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
3263 {
3264         if (time != 0)
3265                 seq_printf(m, "%s: %llu seconds\n", prefix,
3266                            cfs_time_current_sec() - time);
3267         else
3268                 seq_printf(m, "%s: N/A\n", prefix);
3269 }
3270
3271 static void scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
3272 {
3273         if (pos != 0)
3274                 seq_printf(m, "%s: %llu\n", prefix, pos);
3275         else
3276                 seq_printf(m, "%s: N/A\n", prefix);
3277 }
3278
3279 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
3280 {
3281         struct osd_scrub  *scrub   = &dev->od_scrub;
3282         struct scrub_file *sf      = &scrub->os_file;
3283         __u64              checked;
3284         __u64              speed;
3285
3286         down_read(&scrub->os_rwsem);
3287         seq_printf(m, "name: OI_scrub\n"
3288                    "magic: 0x%x\n"
3289                    "oi_files: %d\n"
3290                    "status: %s\n",
3291                    sf->sf_magic, (int)sf->sf_oi_count,
3292                    scrub_status_names[sf->sf_status]);
3293
3294         scrub_bits_dump(m, sf->sf_flags, scrub_flags_names, "flags");
3295
3296         scrub_bits_dump(m, sf->sf_param, scrub_param_names, "param");
3297
3298         scrub_time_dump(m, sf->sf_time_last_complete,
3299                         "time_since_last_completed");
3300
3301         scrub_time_dump(m, sf->sf_time_latest_start,
3302                         "time_since_latest_start");
3303
3304         scrub_time_dump(m, sf->sf_time_last_checkpoint,
3305                         "time_since_last_checkpoint");
3306
3307         scrub_pos_dump(m, sf->sf_pos_latest_start,
3308                         "latest_start_position");
3309
3310         scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
3311                         "last_checkpoint_position");
3312
3313         scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
3314                         "first_failure_position");
3315
3316         checked = sf->sf_items_checked + scrub->os_new_checked;
3317         seq_printf(m, "checked: %llu\n"
3318                    "updated: %llu\n"
3319                    "failed: %llu\n"
3320                    "prior_updated: %llu\n"
3321                    "noscrub: %llu\n"
3322                    "igif: %llu\n"
3323                    "success_count: %u\n",
3324                    checked, sf->sf_items_updated, sf->sf_items_failed,
3325                    sf->sf_items_updated_prior, sf->sf_items_noscrub,
3326                    sf->sf_items_igif, sf->sf_success_count);
3327
3328         speed = checked;
3329         if (thread_is_running(&scrub->os_thread)) {
3330                 cfs_duration_t duration = cfs_time_current() -
3331                                           scrub->os_time_last_checkpoint;
3332                 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
3333                                                      MSEC_PER_SEC);
3334                 __u32 rtime = sf->sf_run_time +
3335                               cfs_duration_sec(duration + HALF_SEC);
3336
3337                 if (duration != 0)
3338                         do_div(new_checked, duration);
3339                 if (rtime != 0)
3340                         do_div(speed, rtime);
3341                 seq_printf(m, "run_time: %u seconds\n"
3342                            "average_speed: %llu objects/sec\n"
3343                            "real-time_speed: %llu objects/sec\n"
3344                            "current_position: %u\n"
3345                            "lf_scanned: %llu\n"
3346                            "lf_repaired: %llu\n"
3347                            "lf_failed: %llu\n",
3348                            rtime, speed, new_checked, scrub->os_pos_current,
3349                            scrub->os_lf_scanned, scrub->os_lf_repaired,
3350                            scrub->os_lf_failed);
3351         } else {
3352                 if (sf->sf_run_time != 0)
3353                         do_div(speed, sf->sf_run_time);
3354                 seq_printf(m, "run_time: %u seconds\n"
3355                            "average_speed: %llu objects/sec\n"
3356                            "real-time_speed: N/A\n"
3357                            "current_position: N/A\n"
3358                            "lf_scanned: %llu\n"
3359                            "lf_repaired: %llu\n"
3360                            "lf_failed: %llu\n",
3361                            sf->sf_run_time, speed, scrub->os_lf_scanned,
3362                            scrub->os_lf_repaired, scrub->os_lf_failed);
3363         }
3364
3365         up_read(&scrub->os_rwsem);
3366         return 0;
3367 }