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, 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 #ifndef EXPORT_SYMTAB
  39 # define EXPORT_SYMTAB
  40 #endif
  41 #define DEBUG_SUBSYSTEM S_MDS
  42
  43 #include <lustre/lustre_idl.h>
  44 #include <lustre_disk.h>
  45 #include <dt_object.h>
  46
  47 #include "osd_internal.h"
  48 #include "osd_oi.h"
  49 #include "osd_scrub.h"
  50
  51 #define HALF_SEC        (CFS_HZ >> 1)
  52
  53 #define OSD_OTABLE_MAX_HASH             0x00000000ffffffffULL
  54
  55 #define SCRUB_NEXT_BREAK        1 /* exit current loop and process next group */
  56 #define SCRUB_NEXT_CONTINUE     2 /* skip current object and process next bit */
  57 #define SCRUB_NEXT_EXIT         3 /* exit all the loops */
  58 #define SCRUB_NEXT_WAIT         4 /* wait for free cache slot */
  59 #define SCRUB_NEXT_CRASH        5 /* simulate system crash during OI scrub */
  60 #define SCRUB_NEXT_FATAL        6 /* simulate failure during OI scrub */
  61 #define SCRUB_NEXT_NOSCRUB      7 /* new created object, no scrub on it */
  62 #define SCRUB_NEXT_NOLMA        8 /* the inode has no FID-in-LMA */
  63
  64 /* misc functions */
  65
  66 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
  67 {
  68         return container_of0(scrub, struct osd_device, od_scrub);
  69 }
  70
  71 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
  72 {
  73         return osd_sb(osd_scrub2dev(scrub));
  74 }
  75
  76 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
  77                                        struct osd_otable_cache *ooc)
  78 {
  79         return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
  80 }
  81
  82 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
  83                                      struct osd_device *dev,
  84                                      const struct lu_fid *fid,
  85                                      const struct osd_inode_id *id, int ops)
  86 {
  87         struct lu_fid         *oi_fid = &info->oti_fid2;
  88         struct osd_inode_id   *oi_id  = &info->oti_id2;
  89         struct iam_container  *bag;
  90         struct iam_path_descr *ipd;
  91         handle_t              *jh;
  92         int                    rc;
  93         ENTRY;
  94
  95         fid_cpu_to_be(oi_fid, fid);
  96         osd_id_pack(oi_id, id);
  97         jh = ldiskfs_journal_start_sb(osd_sb(dev),
  98                                       osd_dto_credits_noquota[ops]);
  99         if (IS_ERR(jh)) {
 100                 rc = PTR_ERR(jh);
 101                 CERROR("%.16s: fail to start trans for scrub store: rc = %d\n",
 102                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
 103                 RETURN(rc);
 104         }
 105
 106         bag = &osd_fid2oi(dev, fid)->oi_dir.od_container;
 107         ipd = osd_idx_ipd_get(info->oti_env, bag);
 108         if (unlikely(ipd == NULL)) {
 109                 ldiskfs_journal_stop(jh);
 110                 CERROR("%.16s: fail to get ipd for scrub store\n",
 111                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name);
 112                 RETURN(-ENOMEM);
 113         }
 114
 115         if (ops == DTO_INDEX_UPDATE) {
 116                 rc = iam_update(jh, bag, (const struct iam_key *)oi_fid,
 117                                 (struct iam_rec *)oi_id, ipd);
 118         } else {
 119                 rc = iam_insert(jh, bag, (const struct iam_key *)oi_fid,
 120                                 (struct iam_rec *)oi_id, ipd);
 121                 if (rc == -EEXIST) {
 122                         rc = 1;
 123                         /* XXX: There are trouble things when adding OI
 124                          *      mapping for IGIF object, which may cause
 125                          *      multiple objects to be mapped to the same
 126                          *      IGIF formatted FID. Consider the following
 127                          *      situations:
 128                          *
 129                          *      1) The MDT is upgrading from 1.8 device.
 130                          *      The OI scrub generates IGIF FID1 for the
 131                          *      OBJ1 and adds the OI mapping.
 132                          *
 133                          *      2) For some reason, the OI scrub does not
 134                          *      process all the IGIF objects completely.
 135                          *
 136                          *      3) The MDT is backuped and restored against
 137                          *      this device.
 138                          *
 139                          *      4) When the MDT mounts up, the OI scrub will
 140                          *      try to rebuild the OI files. For some IGIF
 141                          *      object, OBJ2, which was not processed by the
 142                          *      OI scrub before the backup/restore, and the
 143                          *      new generated IGIF formatted FID may be just
 144                          *      the FID1, the same as OBJ1.
 145                          *
 146                          *      Under such case, the OI scrub cannot know how
 147                          *      to generate new FID for the OBJ2.
 148                          *
 149                          *      Currently, we do nothing for that. One possible
 150                          *      solution is to generate new normal FID for the
 151                          *      conflict object.
 152                          *
 153                          *      Anyway, it is rare, only exists in theory. */
 154                 }
 155         }
 156         osd_ipd_put(info->oti_env, bag, ipd);
 157         ldiskfs_journal_stop(jh);
 158         RETURN(rc);
 159 }
 160
 161 /* OI_scrub file ops */
 162
 163 static void osd_scrub_file_to_cpu(struct scrub_file *des,
 164                                   struct scrub_file *src)
 165 {
 166         memcpy(des->sf_uuid, src->sf_uuid, 16);
 167         des->sf_flags   = le64_to_cpu(src->sf_flags);
 168         des->sf_magic   = le32_to_cpu(src->sf_magic);
 169         des->sf_status  = le16_to_cpu(src->sf_status);
 170         des->sf_param   = le16_to_cpu(src->sf_param);
 171         des->sf_time_last_complete      =
 172                                 le64_to_cpu(src->sf_time_last_complete);
 173         des->sf_time_latest_start       =
 174                                 le64_to_cpu(src->sf_time_latest_start);
 175         des->sf_time_last_checkpoint    =
 176                                 le64_to_cpu(src->sf_time_last_checkpoint);
 177         des->sf_pos_latest_start        =
 178                                 le64_to_cpu(src->sf_pos_latest_start);
 179         des->sf_pos_last_checkpoint     =
 180                                 le64_to_cpu(src->sf_pos_last_checkpoint);
 181         des->sf_pos_first_inconsistent  =
 182                                 le64_to_cpu(src->sf_pos_first_inconsistent);
 183         des->sf_items_checked           =
 184                                 le64_to_cpu(src->sf_items_checked);
 185         des->sf_items_updated           =
 186                                 le64_to_cpu(src->sf_items_updated);
 187         des->sf_items_failed            =
 188                                 le64_to_cpu(src->sf_items_failed);
 189         des->sf_items_updated_prior     =
 190                                 le64_to_cpu(src->sf_items_updated_prior);
 191         des->sf_run_time        = le32_to_cpu(src->sf_run_time);
 192         des->sf_success_count   = le32_to_cpu(src->sf_success_count);
 193         des->sf_oi_count        = le16_to_cpu(src->sf_oi_count);
 194         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 195 }
 196
 197 static void osd_scrub_file_to_le(struct scrub_file *des,
 198                                  struct scrub_file *src)
 199 {
 200         memcpy(des->sf_uuid, src->sf_uuid, 16);
 201         des->sf_flags   = cpu_to_le64(src->sf_flags);
 202         des->sf_magic   = cpu_to_le32(src->sf_magic);
 203         des->sf_status  = cpu_to_le16(src->sf_status);
 204         des->sf_param   = cpu_to_le16(src->sf_param);
 205         des->sf_time_last_complete      =
 206                                 cpu_to_le64(src->sf_time_last_complete);
 207         des->sf_time_latest_start       =
 208                                 cpu_to_le64(src->sf_time_latest_start);
 209         des->sf_time_last_checkpoint    =
 210                                 cpu_to_le64(src->sf_time_last_checkpoint);
 211         des->sf_pos_latest_start        =
 212                                 cpu_to_le64(src->sf_pos_latest_start);
 213         des->sf_pos_last_checkpoint     =
 214                                 cpu_to_le64(src->sf_pos_last_checkpoint);
 215         des->sf_pos_first_inconsistent  =
 216                                 cpu_to_le64(src->sf_pos_first_inconsistent);
 217         des->sf_items_checked           =
 218                                 cpu_to_le64(src->sf_items_checked);
 219         des->sf_items_updated           =
 220                                 cpu_to_le64(src->sf_items_updated);
 221         des->sf_items_failed            =
 222                                 cpu_to_le64(src->sf_items_failed);
 223         des->sf_items_updated_prior     =
 224                                 cpu_to_le64(src->sf_items_updated_prior);
 225         des->sf_run_time        = cpu_to_le32(src->sf_run_time);
 226         des->sf_success_count   = cpu_to_le32(src->sf_success_count);
 227         des->sf_oi_count        = cpu_to_le16(src->sf_oi_count);
 228         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 229 }
 230
 231 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
 232 {
 233         struct scrub_file *sf = &scrub->os_file;
 234
 235         memset(sf, 0, sizeof(*sf));
 236         memcpy(sf->sf_uuid, uuid, 16);
 237         sf->sf_magic = SCRUB_MAGIC_V1;
 238         sf->sf_status = SS_INIT;
 239 }
 240
 241 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
 242 {
 243         struct scrub_file *sf = &scrub->os_file;
 244
 245         CDEBUG(D_LFSCK, "Reset OI scrub file, flags = "LPX64"\n", flags);
 246         memcpy(sf->sf_uuid, uuid, 16);
 247         sf->sf_status = SS_INIT;
 248         sf->sf_flags |= flags;
 249         sf->sf_param = 0;
 250         sf->sf_run_time = 0;
 251         sf->sf_time_latest_start = 0;
 252         sf->sf_time_last_checkpoint = 0;
 253         sf->sf_pos_latest_start = 0;
 254         sf->sf_pos_last_checkpoint = 0;
 255         sf->sf_pos_first_inconsistent = 0;
 256         sf->sf_items_checked = 0;
 257         sf->sf_items_updated = 0;
 258         sf->sf_items_failed = 0;
 259         sf->sf_items_updated_prior = 0;
 260         sf->sf_items_noscrub = 0;
 261         sf->sf_items_igif = 0;
 262 }
 263
 264 static int osd_scrub_file_load(struct osd_scrub *scrub)
 265 {
 266         loff_t  pos  = 0;
 267         char   *name = LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
 268         int     len  = sizeof(scrub->os_file_disk);
 269         int     rc;
 270
 271         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
 272         if (rc == len) {
 273                 struct scrub_file *sf = &scrub->os_file;
 274
 275                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
 276                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
 277                         CWARN("%.16s: invalid scrub magic 0x%x != 0x%x\n,",
 278                               name, sf->sf_magic, SCRUB_MAGIC_V1);
 279                         /* Process it as new scrub file. */
 280                         rc = -ENOENT;
 281                 } else {
 282                         rc = 0;
 283                 }
 284         } else if (rc != 0) {
 285                 CERROR("%.16s: fail to load scrub file, expected = %d, "
 286                        "rc = %d\n", name, len, rc);
 287                 if (rc > 0)
 288                         rc = -EFAULT;
 289         } else {
 290                 /* return -ENOENT for empty scrub file case. */
 291                 rc = -ENOENT;
 292         }
 293
 294         return rc;
 295 }
 296
 297 int osd_scrub_file_store(struct osd_scrub *scrub)
 298 {
 299         struct osd_device *dev;
 300         handle_t          *jh;
 301         loff_t             pos     = 0;
 302         int                len     = sizeof(scrub->os_file_disk);
 303         int                credits;
 304         int                rc;
 305
 306         dev = container_of0(scrub, struct osd_device, od_scrub);
 307         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
 308                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
 309         jh = ldiskfs_journal_start_sb(osd_sb(dev), credits);
 310         if (IS_ERR(jh)) {
 311                 rc = PTR_ERR(jh);
 312                 CERROR("%.16s: fail to start trans for scrub store, rc = %d\n",
 313                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,rc);
 314                 return rc;
 315         }
 316
 317         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
 318         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
 319                                       len, 0, &pos, jh);
 320         ldiskfs_journal_stop(jh);
 321         if (rc != 0)
 322                 CERROR("%.16s: fail to store scrub file, expected = %d, "
 323                        "rc = %d\n",
 324                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
 325                        len, rc);
 326         scrub->os_time_last_checkpoint = cfs_time_current();
 327         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
 328                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
 329         return rc;
 330 }
 331
 332 /* OI scrub APIs */
 333
 334 static int osd_scrub_prep(struct osd_device *dev)
 335 {
 336         struct osd_scrub     *scrub  = &dev->od_scrub;
 337         struct ptlrpc_thread *thread = &scrub->os_thread;
 338         struct scrub_file    *sf     = &scrub->os_file;
 339         __u32                 flags  = scrub->os_start_flags;
 340         int                   rc;
 341         ENTRY;
 342
 343         down_write(&scrub->os_rwsem);
 344         if (flags & SS_SET_FAILOUT)
 345                 sf->sf_param |= SP_FAILOUT;
 346
 347         if (flags & SS_CLEAR_FAILOUT)
 348                 sf->sf_param &= ~SP_FAILOUT;
 349
 350         if (flags & SS_RESET)
 351                 osd_scrub_file_reset(scrub,
 352                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
 353
 354         if (flags & SS_AUTO) {
 355                 scrub->os_full_speed = 1;
 356                 sf->sf_flags |= SF_AUTO;
 357         } else {
 358                 scrub->os_full_speed = 0;
 359         }
 360
 361         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
 362                 scrub->os_full_speed = 1;
 363
 364         scrub->os_in_prior = 0;
 365         scrub->os_waiting = 0;
 366         scrub->os_paused = 0;
 367         scrub->os_new_checked = 0;
 368         if (sf->sf_pos_last_checkpoint != 0)
 369                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 370         else
 371                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 372
 373         scrub->os_pos_current = sf->sf_pos_latest_start;
 374         sf->sf_status = SS_SCANNING;
 375         sf->sf_time_latest_start = cfs_time_current_sec();
 376         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 377         rc = osd_scrub_file_store(scrub);
 378         if (rc == 0) {
 379                 spin_lock(&scrub->os_lock);
 380                 thread_set_flags(thread, SVC_RUNNING);
 381                 spin_unlock(&scrub->os_lock);
 382                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
 383         }
 384         up_write(&scrub->os_rwsem);
 385
 386         RETURN(rc);
 387 }
 388
 389 static int
 390 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
 391                        struct osd_idmap_cache *oic, int val)
 392 {
 393         struct osd_scrub             *scrub  = &dev->od_scrub;
 394         struct scrub_file            *sf     = &scrub->os_file;
 395         struct lu_fid                *fid    = &oic->oic_fid;
 396         struct osd_inode_id          *lid    = &oic->oic_lid;
 397         struct osd_inode_id          *lid2   = &info->oti_id;
 398         struct osd_inconsistent_item *oii    = NULL;
 399         struct inode                 *inode  = NULL;
 400         int                           ops    = DTO_INDEX_UPDATE;
 401         int                           idx;
 402         int                           rc;
 403         ENTRY;
 404
 405         down_write(&scrub->os_rwsem);
 406         scrub->os_new_checked++;
 407         if (val < 0)
 408                 GOTO(out, rc = val);
 409
 410         if (scrub->os_in_prior)
 411                 oii = cfs_list_entry(oic, struct osd_inconsistent_item,
 412                                      oii_cache);
 413
 414         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
 415                 GOTO(out, rc = 0);
 416
 417         if (fid_is_igif(fid))
 418                 sf->sf_items_igif++;
 419
 420         if ((val == SCRUB_NEXT_NOLMA) &&
 421             (!dev->od_handle_nolma || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
 422                 GOTO(out, rc = 0);
 423
 424         if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA))
 425                 goto iget;
 426
 427         /* XXX: Currently, no FID-in-LMA for OST object, so osd_oi_lookup()
 428          *      without checking FLD is enough.
 429          *
 430          *      It should be updated if FID-in-LMA for OSD object introduced
 431          *      in the future. */
 432         rc = osd_oi_lookup(info, dev, fid, lid2, false);
 433         if (rc != 0) {
 434                 if (rc != -ENOENT)
 435                         GOTO(out, rc);
 436
 437 iget:
 438                 inode = osd_iget(info, dev, lid);
 439                 if (IS_ERR(inode)) {
 440                         rc = PTR_ERR(inode);
 441                         /* Someone removed the inode. */
 442                         if (rc == -ENOENT || rc == -ESTALE)
 443                                 rc = 0;
 444                         GOTO(out, rc);
 445                 }
 446
 447                 /* Prevent the inode to be unlinked during OI scrub. */
 448                 mutex_lock(&inode->i_mutex);
 449                 if (unlikely(inode->i_nlink == 0)) {
 450                         mutex_unlock(&inode->i_mutex);
 451                         iput(inode);
 452                         GOTO(out, rc = 0);
 453                 }
 454
 455                 ops = DTO_INDEX_INSERT;
 456                 idx = osd_oi_fid2idx(dev, fid);
 457                 if (val == SCRUB_NEXT_NOLMA) {
 458                         sf->sf_flags |= SF_UPGRADE;
 459                         scrub->os_full_speed = 1;
 460                         rc = osd_ea_fid_set(info, inode, fid, 0);
 461                         if (rc != 0)
 462                                 GOTO(out, rc);
 463                 } else {
 464                         sf->sf_flags |= SF_RECREATED | SF_INCONSISTENT;
 465                         scrub->os_full_speed = 1;
 466                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 467                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 468                 }
 469         } else if (osd_id_eq(lid, lid2)) {
 470                 GOTO(out, rc = 0);
 471         } else {
 472                 sf->sf_flags |= SF_INCONSISTENT;
 473                 scrub->os_full_speed = 1;
 474         }
 475
 476         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops);
 477         if (rc == 0) {
 478                 if (scrub->os_in_prior)
 479                         sf->sf_items_updated_prior++;
 480                 else
 481                         sf->sf_items_updated++;
 482         }
 483
 484         GOTO(out, rc);
 485
 486 out:
 487         if (rc < 0) {
 488                 sf->sf_items_failed++;
 489                 if (sf->sf_pos_first_inconsistent == 0 ||
 490                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 491                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 492         } else {
 493                 rc = 0;
 494         }
 495
 496         if (ops == DTO_INDEX_INSERT) {
 497                 mutex_unlock(&inode->i_mutex);
 498                 iput(inode);
 499         }
 500         up_write(&scrub->os_rwsem);
 501
 502         if (oii != NULL) {
 503                 LASSERT(!cfs_list_empty(&oii->oii_list));
 504
 505                 spin_lock(&scrub->os_lock);
 506                 cfs_list_del_init(&oii->oii_list);
 507                 spin_unlock(&scrub->os_lock);
 508                 OBD_FREE_PTR(oii);
 509         }
 510         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 511 }
 512
 513 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
 514 {
 515         struct scrub_file *sf = &scrub->os_file;
 516         int                rc;
 517
 518         if (likely(cfs_time_before(cfs_time_current(),
 519                                    scrub->os_time_next_checkpoint) ||
 520                    scrub->os_new_checked == 0))
 521                 return 0;
 522
 523         down_write(&scrub->os_rwsem);
 524         sf->sf_items_checked += scrub->os_new_checked;
 525         scrub->os_new_checked = 0;
 526         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 527         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 528         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 529                                             scrub->os_time_last_checkpoint);
 530         rc = osd_scrub_file_store(scrub);
 531         up_write(&scrub->os_rwsem);
 532
 533         return rc;
 534 }
 535
 536 static void osd_scrub_post(struct osd_scrub *scrub, int result)
 537 {
 538         struct scrub_file *sf = &scrub->os_file;
 539         ENTRY;
 540
 541         down_write(&scrub->os_rwsem);
 542         spin_lock(&scrub->os_lock);
 543         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 544         spin_unlock(&scrub->os_lock);
 545         if (scrub->os_new_checked > 0) {
 546                 sf->sf_items_checked += scrub->os_new_checked;
 547                 scrub->os_new_checked = 0;
 548                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 549         }
 550         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 551         if (result > 0) {
 552                 sf->sf_status = SS_COMPLETED;
 553                 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 554                 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 555                                   SF_UPGRADE | SF_AUTO);
 556                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 557                 sf->sf_success_count++;
 558         } else if (result == 0) {
 559                 if (scrub->os_paused)
 560                         sf->sf_status = SS_PAUSED;
 561                 else
 562                         sf->sf_status = SS_STOPPED;
 563         } else {
 564                 sf->sf_status = SS_FAILED;
 565         }
 566         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 567                                             scrub->os_time_last_checkpoint);
 568         result = osd_scrub_file_store(scrub);
 569         if (result < 0)
 570                 CERROR("%.16s: fail to osd_scrub_post, rc = %d\n",
 571                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
 572                        result);
 573         up_write(&scrub->os_rwsem);
 574
 575         EXIT;
 576 }
 577
 578 /* iteration engine */
 579
 580 struct osd_iit_param {
 581         struct super_block *sb;
 582         struct buffer_head *bitmap;
 583         ldiskfs_group_t bg;
 584         __u32 gbase;
 585         __u32 offset;
 586 };
 587
 588 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 589                                    struct osd_device *dev,
 590                                    struct osd_iit_param *param,
 591                                    struct osd_idmap_cache **oic,
 592                                    int noslot);
 593
 594 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 595                                    struct osd_device *dev,
 596                                    struct osd_iit_param *param,
 597                                    struct osd_idmap_cache *oic,
 598                                    int *noslot, int rc);
 599
 600 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 601 {
 602         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 603                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 604         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 605                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 606                 return SCRUB_NEXT_BREAK;
 607         } else {
 608                 *pos = param->gbase + param->offset;
 609                 return 0;
 610         }
 611 }
 612
 613 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 614                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 615                         struct super_block *sb, bool scrub)
 616 {
 617         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 618         struct inode            *inode;
 619         int                      rc;
 620
 621         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 622         inode = osd_iget(info, dev, lid);
 623         if (IS_ERR(inode)) {
 624                 rc = PTR_ERR(inode);
 625                 /* The inode may be removed after bitmap searching, or the
 626                  * file is new created without inode initialized yet. */
 627                 if (rc == -ENOENT || rc == -ESTALE)
 628                         return SCRUB_NEXT_CONTINUE;
 629
 630                 CERROR("%.16s: fail to read inode, ino# = %u, rc = %d\n",
 631                        LDISKFS_SB(sb)->s_es->s_volume_name, pos, rc);
 632                 return rc;
 633         }
 634
 635         /* If the inode has no OI mapping, then it is special locally used,
 636          * should be invisible to OI scrub or up layer LFSCK. */
 637         if (ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI)) {
 638                 iput(inode);
 639                 return SCRUB_NEXT_CONTINUE;
 640         }
 641
 642         if (scrub &&
 643             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 644                 /* Only skip it for the first OI scrub accessing. */
 645                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 646                 iput(inode);
 647                 return SCRUB_NEXT_NOSCRUB;
 648         }
 649
 650         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 651         if (rc == 0) {
 652                 if (fid_is_llog(&lma->lma_self_fid) ||
 653                     (!scrub && fid_is_internal(&lma->lma_self_fid)))
 654                         rc = SCRUB_NEXT_CONTINUE;
 655                 else
 656                         *fid = lma->lma_self_fid;
 657         } else if (rc == -ENODATA) {
 658                 lu_igif_build(fid, inode->i_ino, inode->i_generation);
 659                 if (scrub)
 660                         rc = SCRUB_NEXT_NOLMA;
 661                 else
 662                         rc = 0;
 663         }
 664         iput(inode);
 665         return rc;
 666 }
 667
 668 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 669                           struct osd_iit_param *param,
 670                           struct osd_idmap_cache **oic, int noslot)
 671 {
 672         struct osd_scrub     *scrub  = &dev->od_scrub;
 673         struct ptlrpc_thread *thread = &scrub->os_thread;
 674         struct lu_fid        *fid;
 675         struct osd_inode_id  *lid;
 676         int                   rc;
 677
 678         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
 679                 struct l_wait_info lwi;
 680
 681                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
 682                 l_wait_event(thread->t_ctl_waitq,
 683                              !cfs_list_empty(&scrub->os_inconsistent_items) ||
 684                              !thread_is_running(thread),
 685                              &lwi);
 686         }
 687
 688         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 689                 spin_lock(&scrub->os_lock);
 690                 thread_set_flags(thread, SVC_STOPPING);
 691                 spin_unlock(&scrub->os_lock);
 692                 return SCRUB_NEXT_CRASH;
 693         }
 694
 695         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
 696                 return SCRUB_NEXT_FATAL;
 697
 698         if (unlikely(!thread_is_running(thread)))
 699                 return SCRUB_NEXT_EXIT;
 700
 701         if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
 702                 struct osd_inconsistent_item *oii;
 703
 704                 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
 705                                      struct osd_inconsistent_item, oii_list);
 706                 *oic = &oii->oii_cache;
 707                 scrub->os_in_prior = 1;
 708                 return 0;
 709         }
 710
 711         if (noslot != 0)
 712                 return SCRUB_NEXT_WAIT;
 713
 714         rc = osd_iit_next(param, &scrub->os_pos_current);
 715         if (rc != 0)
 716                 return rc;
 717
 718         *oic = &scrub->os_oic;
 719         fid = &(*oic)->oic_fid;
 720         lid = &(*oic)->oic_lid;
 721         rc = osd_iit_iget(info, dev, fid, lid,
 722                           scrub->os_pos_current, param->sb, true);
 723         return rc;
 724 }
 725
 726 static int osd_preload_next(struct osd_thread_info *info,
 727                             struct osd_device *dev, struct osd_iit_param *param,
 728                             struct osd_idmap_cache **oic, int noslot)
 729 {
 730         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
 731         struct osd_scrub        *scrub;
 732         struct ptlrpc_thread    *thread;
 733         int                      rc;
 734
 735         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
 736         if (rc != 0)
 737                 return rc;
 738
 739         scrub = &dev->od_scrub;
 740         thread = &scrub->os_thread;
 741         if (thread_is_running(thread) &&
 742             ooc->ooc_pos_preload >= scrub->os_pos_current)
 743                 return SCRUB_NEXT_EXIT;
 744
 745         rc = osd_iit_iget(info, dev,
 746                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
 747                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
 748                           ooc->ooc_pos_preload, param->sb, false);
 749         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
 750          * ignore the failure, so it still need to skip the inode next time. */
 751         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
 752         return rc;
 753 }
 754
 755 static inline int
 756 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
 757 {
 758         spin_lock(&scrub->os_lock);
 759         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
 760             !cfs_list_empty(&scrub->os_inconsistent_items) ||
 761             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
 762                 scrub->os_waiting = 0;
 763         else
 764                 scrub->os_waiting = 1;
 765         spin_unlock(&scrub->os_lock);
 766
 767         return !scrub->os_waiting;
 768 }
 769
 770 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
 771                           struct osd_iit_param *param,
 772                           struct osd_idmap_cache *oic, int *noslot, int rc)
 773 {
 774         struct l_wait_info       lwi    = { 0 };
 775         struct osd_scrub        *scrub  = &dev->od_scrub;
 776         struct scrub_file       *sf     = &scrub->os_file;
 777         struct ptlrpc_thread    *thread = &scrub->os_thread;
 778         struct osd_otable_it    *it     = dev->od_otable_it;
 779         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
 780
 781         switch (rc) {
 782         case SCRUB_NEXT_CONTINUE:
 783                 goto next;
 784         case SCRUB_NEXT_WAIT:
 785                 goto wait;
 786         case SCRUB_NEXT_NOSCRUB:
 787                 down_write(&scrub->os_rwsem);
 788                 scrub->os_new_checked++;
 789                 sf->sf_items_noscrub++;
 790                 up_write(&scrub->os_rwsem);
 791                 goto next;
 792         }
 793
 794         rc = osd_scrub_check_update(info, dev, oic, rc);
 795         if (rc != 0)
 796                 return rc;
 797
 798         rc = osd_scrub_checkpoint(scrub);
 799         if (rc != 0) {
 800                 CERROR("%.16s: fail to checkpoint, pos = %u, rc = %d\n",
 801                        LDISKFS_SB(param->sb)->s_es->s_volume_name,
 802                        scrub->os_pos_current, rc);
 803                 /* Continue, as long as the scrub itself can go ahead. */
 804         }
 805
 806         if (scrub->os_in_prior) {
 807                 scrub->os_in_prior = 0;
 808                 return 0;
 809         }
 810
 811 next:
 812         scrub->os_pos_current = param->gbase + ++(param->offset);
 813
 814 wait:
 815         if (it != NULL && it->ooi_waiting &&
 816             ooc->ooc_pos_preload < scrub->os_pos_current) {
 817                 spin_lock(&scrub->os_lock);
 818                 it->ooi_waiting = 0;
 819                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
 820                 spin_unlock(&scrub->os_lock);
 821         }
 822
 823         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
 824                 return 0;
 825
 826         if (osd_scrub_has_window(scrub, ooc)) {
 827                 *noslot = 0;
 828                 return 0;
 829         }
 830
 831         l_wait_event(thread->t_ctl_waitq,
 832                      osd_scrub_wakeup(scrub, it),
 833                      &lwi);
 834
 835         if (osd_scrub_has_window(scrub, ooc))
 836                 *noslot = 0;
 837         else
 838                 *noslot = 1;
 839         return 0;
 840 }
 841
 842 static int osd_preload_exec(struct osd_thread_info *info,
 843                             struct osd_device *dev, struct osd_iit_param *param,
 844                             struct osd_idmap_cache *oic, int *noslot, int rc)
 845 {
 846         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
 847
 848         if (rc == 0) {
 849                 ooc->ooc_cached_items++;
 850                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
 851                                         ~OSD_OTABLE_IT_CACHE_MASK;
 852         }
 853         return rc > 0 ? 0 : rc;
 854 }
 855
 856 #define SCRUB_IT_ALL    1
 857 #define SCRUB_IT_CRASH  2
 858
 859 static int osd_inode_iteration(struct osd_thread_info *info,
 860                                struct osd_device *dev, __u32 max, bool preload)
 861 {
 862         osd_iit_next_policy   next;
 863         osd_iit_exec_policy   exec;
 864         __u32                *pos;
 865         __u32                *count;
 866         struct osd_iit_param  param;
 867         __u32                 limit;
 868         int                   noslot = 0;
 869         int                   rc;
 870         ENTRY;
 871
 872         if (!preload) {
 873                 struct osd_scrub *scrub = &dev->od_scrub;
 874
 875                 next = osd_scrub_next;
 876                 exec = osd_scrub_exec;
 877                 pos = &scrub->os_pos_current;
 878                 count = &scrub->os_new_checked;
 879         } else {
 880                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
 881
 882                 next = osd_preload_next;
 883                 exec = osd_preload_exec;
 884                 pos = &ooc->ooc_pos_preload;
 885                 count = &ooc->ooc_cached_items;
 886         }
 887         param.sb = osd_sb(dev);
 888         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
 889
 890         while (*pos <= limit && *count < max) {
 891                 struct osd_idmap_cache *oic = NULL;
 892
 893                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
 894                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
 895                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
 896                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
 897                 if (param.bitmap == NULL) {
 898                         CERROR("%.16s: fail to read bitmap for %u, "
 899                                "scrub will stop, urgent mode\n",
 900                                LDISKFS_SB(param.sb)->s_es->s_volume_name,
 901                                (__u32)param.bg);
 902                         RETURN(-EIO);
 903                 }
 904
 905                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
 906                        *count < max) {
 907                         rc = next(info, dev, &param, &oic, noslot);
 908                         switch (rc) {
 909                         case SCRUB_NEXT_BREAK:
 910                                 goto next_group;
 911                         case SCRUB_NEXT_EXIT:
 912                                 brelse(param.bitmap);
 913                                 RETURN(0);
 914                         case SCRUB_NEXT_CRASH:
 915                                 brelse(param.bitmap);
 916                                 RETURN(SCRUB_IT_CRASH);
 917                         case SCRUB_NEXT_FATAL:
 918                                 brelse(param.bitmap);
 919                                 RETURN(-EINVAL);
 920                         }
 921
 922                         rc = exec(info, dev, &param, oic, &noslot, rc);
 923                         if (rc != 0) {
 924                                 brelse(param.bitmap);
 925                                 RETURN(rc);
 926                         }
 927
 928                         if (preload && dev->od_otable_it->ooi_stopping) {
 929                                 brelse(param.bitmap);
 930                                 RETURN(0);
 931                         }
 932                 }
 933
 934 next_group:
 935                 brelse(param.bitmap);
 936
 937                 if (preload && dev->od_otable_it->ooi_stopping)
 938                         RETURN(0);
 939         }
 940
 941         if (*pos > limit)
 942                 RETURN(SCRUB_IT_ALL);
 943         RETURN(0);
 944 }
 945
 946 static int osd_otable_it_preload(const struct lu_env *env,
 947                                  struct osd_otable_it *it)
 948 {
 949         struct osd_device       *dev   = it->ooi_dev;
 950         struct osd_scrub        *scrub = &dev->od_scrub;
 951         struct osd_otable_cache *ooc   = &it->ooi_cache;
 952         int                      rc;
 953         ENTRY;
 954
 955         rc = osd_inode_iteration(osd_oti_get(env), dev,
 956                                  OSD_OTABLE_IT_CACHE_SIZE, true);
 957         if (rc == SCRUB_IT_ALL)
 958                 it->ooi_all_cached = 1;
 959
 960         CDEBUG(D_LFSCK, "OSD pre-loaded: max = %u, preload = %u, rc = %d\n",
 961                le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count),
 962                ooc->ooc_pos_preload, rc);
 963
 964         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
 965                 scrub->os_waiting = 0;
 966                 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
 967         }
 968
 969         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
 970 }
 971
 972 static int osd_scrub_main(void *args)
 973 {
 974         struct lu_env         env;
 975         struct osd_device    *dev    = (struct osd_device *)args;
 976         struct osd_scrub     *scrub  = &dev->od_scrub;
 977         struct ptlrpc_thread *thread = &scrub->os_thread;
 978         struct super_block   *sb     = osd_sb(dev);
 979         int                   rc;
 980         ENTRY;
 981
 982         cfs_daemonize("OI_scrub");
 983         rc = lu_env_init(&env, LCT_DT_THREAD);
 984         if (rc != 0) {
 985                 CERROR("%.16s: OI scrub, fail to init env, rc = %d\n",
 986                        LDISKFS_SB(sb)->s_es->s_volume_name, rc);
 987                 GOTO(noenv, rc);
 988         }
 989
 990         rc = osd_scrub_prep(dev);
 991         if (rc != 0) {
 992                 CERROR("%.16s: OI scrub, fail to scrub prep, rc = %d\n",
 993                        LDISKFS_SB(sb)->s_es->s_volume_name, rc);
 994                 GOTO(out, rc);
 995         }
 996
 997         if (!scrub->os_full_speed) {
 998                 struct l_wait_info lwi = { 0 };
 999                 struct osd_otable_it *it = dev->od_otable_it;
1000                 struct osd_otable_cache *ooc = &it->ooi_cache;
1001
1002                 l_wait_event(thread->t_ctl_waitq,
1003                              it->ooi_user_ready || !thread_is_running(thread),
1004                              &lwi);
1005                 if (unlikely(!thread_is_running(thread)))
1006                         GOTO(post, rc = 0);
1007
1008                 LASSERT(scrub->os_pos_current >= ooc->ooc_pos_preload);
1009                 scrub->os_pos_current = ooc->ooc_pos_preload;
1010         }
1011
1012         CDEBUG(D_LFSCK, "OI scrub: flags = 0x%x, pos = %u\n",
1013                scrub->os_start_flags, scrub->os_pos_current);
1014
1015         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1016         if (unlikely(rc == SCRUB_IT_CRASH))
1017                 GOTO(out, rc = -EINVAL);
1018         GOTO(post, rc);
1019
1020 post:
1021         osd_scrub_post(scrub, rc);
1022         CDEBUG(D_LFSCK, "OI scrub: stop, rc = %d, pos = %u\n",
1023                rc, scrub->os_pos_current);
1024
1025 out:
1026         while (!cfs_list_empty(&scrub->os_inconsistent_items)) {
1027                 struct osd_inconsistent_item *oii;
1028
1029                 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
1030                                      struct osd_inconsistent_item, oii_list);
1031                 cfs_list_del_init(&oii->oii_list);
1032                 OBD_FREE_PTR(oii);
1033         }
1034         lu_env_fini(&env);
1035
1036 noenv:
1037         spin_lock(&scrub->os_lock);
1038         thread_set_flags(thread, SVC_STOPPED);
1039         cfs_waitq_broadcast(&thread->t_ctl_waitq);
1040         spin_unlock(&scrub->os_lock);
1041         return rc;
1042 }
1043
1044 /* initial OI scrub */
1045
1046 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1047                          struct dentry *, filldir_t filldir);
1048
1049 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1050                                loff_t offset, __u64 ino, unsigned d_type);
1051
1052 static int
1053 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1054                      struct dentry *dentry, filldir_t filldir);
1055 static int
1056 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1057                   struct dentry *dentry, filldir_t filldir);
1058
1059 static int
1060 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1061                      struct dentry *dentry, filldir_t filldir);
1062
1063 enum osd_lf_flags {
1064         OLF_SCAN_SUBITEMS       = 0x0001,
1065         OLF_HIDE_FID            = 0x0002,
1066         OLF_SHOW_NAME           = 0x0004,
1067 };
1068
1069 struct osd_lf_map {
1070         char            *olm_name;
1071         struct lu_fid    olm_fid;
1072         __u16            olm_flags;
1073         scandir_t        olm_scandir;
1074         filldir_t        olm_filldir;
1075 };
1076
1077 /* Add the new introduced local files in the list in the future. */
1078 static const struct osd_lf_map osd_lf_maps[] = {
1079         /* CATALOGS */
1080         { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1081                 NULL, NULL },
1082
1083         /* CONFIGS */
1084         { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1085                 OLF_SCAN_SUBITEMS, osd_ios_general_scan,
1086                 osd_ios_varfid_fill },
1087
1088         /* NIDTBL_VERSIONS */
1089         { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1090                 osd_ios_general_scan, osd_ios_varfid_fill },
1091
1092         /* PENDING */
1093         { "PENDING", { FID_SEQ_LOCAL_FILE, MDD_ORPHAN_OID, 0 }, 0, NULL, NULL },
1094
1095         /* ROOT */
1096         { "ROOT", { FID_SEQ_ROOT, 1, 0 },
1097                 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, osd_ios_ROOT_scan, NULL },
1098
1099         /* changelog_catalog */
1100         { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, NULL, NULL },
1101
1102         /* changelog_users */
1103         { CHANGELOG_USERS, { 0, 0, 0 }, 0, NULL, NULL },
1104
1105         /* fld */
1106         { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1107                 NULL, NULL },
1108
1109         /* last_rcvd */
1110         { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1111                 NULL, NULL },
1112
1113         /* lfsck_bookmark */
1114         { "lfsck_bookmark", { FID_SEQ_LOCAL_FILE, LFSCK_BOOKMARK_OID, 0 }, 0,
1115                 NULL, NULL },
1116
1117         /* lov_objid */
1118         { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1119                 NULL, NULL },
1120
1121         /* lov_objseq */
1122         { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1123                 OLF_SHOW_NAME, NULL, NULL },
1124
1125         /* quota_master */
1126         { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1127                 osd_ios_general_scan, osd_ios_varfid_fill },
1128
1129         /* quota_slave */
1130         { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1131                 osd_ios_general_scan, osd_ios_varfid_fill },
1132
1133         /* seq-200000003-lastid */
1134         { "seq-200000003-lastid", { FID_SEQ_LOCAL_NAME, 1, 0 }, 0,
1135                 NULL, NULL },
1136
1137         /* seq_ctl */
1138         { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1139                 OLF_SHOW_NAME, NULL, NULL },
1140
1141         /* seq_srv */
1142         { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1143                 OLF_SHOW_NAME, NULL, NULL },
1144
1145         /* LAST_GROUP */
1146         { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1147                 OLF_SHOW_NAME, NULL, NULL },
1148
1149         /* health_check */
1150         { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1151                 OLF_SHOW_NAME, NULL, NULL },
1152
1153         /* lfsck_namespace */
1154         { "lfsck_namespace", { FID_SEQ_LOCAL_FILE, LFSCK_BOOKMARK_OID, 0 }, 0,
1155                 NULL, NULL },
1156
1157         /* OBJECTS, upgrade from old device */
1158         { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, osd_ios_OBJECTS_scan, NULL },
1159
1160         /* lquota_v2.user, upgrade from old device */
1161         { "lquota_v2.user", { 0, 0, 0 }, 0, NULL, NULL },
1162
1163         /* lquota_v2.group, upgrade from old device */
1164         { "lquota_v2.group", { 0, 0, 0 }, 0, NULL, NULL },
1165
1166         { NULL, { 0, 0, 0 }, 0, NULL, NULL }
1167 };
1168
1169 struct osd_ios_item {
1170         cfs_list_t       oii_list;
1171         struct dentry   *oii_dentry;
1172         scandir_t        oii_scandir;
1173         filldir_t        oii_filldir;
1174 };
1175
1176 struct osd_ios_filldir_buf {
1177         struct osd_thread_info  *oifb_info;
1178         struct osd_device       *oifb_dev;
1179         struct dentry           *oifb_dentry;
1180 };
1181
1182 static inline struct dentry *
1183 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1184 {
1185         struct dentry *dentry;
1186
1187         dentry = ll_lookup_one_len(name, parent, namelen);
1188         if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1189                 dput(dentry);
1190                 return ERR_PTR(-ENOENT);
1191         }
1192
1193         return dentry;
1194 }
1195
1196 static inline void
1197 osd_ios_llogname2fid(struct lu_fid *fid, const char *name, int namelen)
1198 {
1199         obd_id id = 0;
1200         int    i  = 0;
1201
1202         fid->f_seq = FID_SEQ_LLOG;
1203         while (i < namelen)
1204                 id = id * 10 + name[i++] - '0';
1205
1206         fid->f_oid = id & 0x00000000ffffffffULL;
1207         fid->f_ver = id >> 32;
1208 }
1209
1210 static inline void
1211 osd_ios_Oname2fid(struct lu_fid *fid, const char *name, int namelen)
1212 {
1213         __u64 seq = 0;
1214         int   i   = 0;
1215
1216         while (i < namelen)
1217                 seq = seq * 10 + name[i++] - '0';
1218
1219         lu_last_id_fid(fid, seq);
1220 }
1221
1222 static int
1223 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1224                  scandir_t scandir, filldir_t filldir)
1225 {
1226         struct osd_ios_item *item;
1227
1228         OBD_ALLOC_PTR(item);
1229         if (item == NULL)
1230                 return -ENOMEM;
1231
1232         CFS_INIT_LIST_HEAD(&item->oii_list);
1233         item->oii_dentry = dget(dentry);
1234         item->oii_scandir = scandir;
1235         item->oii_filldir = filldir;
1236         cfs_list_add_tail(&item->oii_list, &dev->od_ios_list);
1237         return 0;
1238 }
1239
1240 /**
1241  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1242  *
1243  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1244  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1245  * reference the inode, or fixed if it is missing or references another inode.
1246  */
1247 static int
1248 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1249                  struct inode *inode, const struct lu_fid *fid, int flags)
1250 {
1251         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
1252         struct osd_inode_id     *id     = &info->oti_id;
1253         struct osd_inode_id     *id2    = &info->oti_id2;
1254         struct osd_scrub        *scrub  = &dev->od_scrub;
1255         struct scrub_file       *sf     = &scrub->os_file;
1256         struct lu_fid            tfid;
1257         int                      rc;
1258         ENTRY;
1259
1260         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1261         if (rc != 0 && rc != -ENODATA)
1262                 RETURN(rc);
1263
1264         osd_id_gen(id, inode->i_ino, inode->i_generation);
1265         if (rc == -ENODATA) {
1266                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID)
1267                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1268                 else
1269                         tfid = *fid;
1270                 rc = osd_ea_fid_set(info, inode, &tfid, 0);
1271                 if (rc != 0)
1272                         RETURN(rc);
1273         } else {
1274                 tfid = lma->lma_self_fid;
1275         }
1276
1277         rc = __osd_oi_lookup(info, dev, &tfid, id2);
1278         if (rc != 0) {
1279                 if (rc != -ENOENT)
1280                         RETURN(rc);
1281
1282                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1283                                                DTO_INDEX_INSERT);
1284                 RETURN(rc);
1285         }
1286
1287         if (osd_id_eq_strict(id, id2))
1288                 RETURN(0);
1289
1290         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1291                 osd_scrub_file_reset(scrub,
1292                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1293                                      SF_INCONSISTENT);
1294                 rc = osd_scrub_file_store(scrub);
1295                 if (rc != 0)
1296                         RETURN(rc);
1297         }
1298
1299         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id, DTO_INDEX_UPDATE);
1300
1301         RETURN(rc);
1302 }
1303
1304 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1305                                loff_t offset, __u64 ino, unsigned d_type)
1306 {
1307         struct osd_ios_filldir_buf *fill_buf = buf;
1308         struct osd_device          *dev      = fill_buf->oifb_dev;
1309         struct dentry              *child;
1310         int                         rc;
1311         ENTRY;
1312
1313         /* skip any '.' started names */
1314         if (name[0] == '.')
1315                 RETURN(0);
1316
1317         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1318         if (IS_ERR(child))
1319                 RETURN(PTR_ERR(child));
1320
1321         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1322                               NULL, 0);
1323         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1324                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1325                                       osd_ios_varfid_fill);
1326         dput(child);
1327
1328         RETURN(rc);
1329 }
1330
1331 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
1332                              loff_t offset, __u64 ino, unsigned d_type)
1333 {
1334         struct osd_ios_filldir_buf *fill_buf = buf;
1335         struct osd_device          *dev      = fill_buf->oifb_dev;
1336         const struct osd_lf_map    *map;
1337         struct dentry              *child;
1338         int                         rc       = 0;
1339         ENTRY;
1340
1341         /* skip any '.' started names */
1342         if (name[0] == '.')
1343                 RETURN(0);
1344
1345         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
1346                 if (strlen(map->olm_name) != namelen)
1347                         continue;
1348
1349                 if (strncmp(map->olm_name, name, namelen) == 0)
1350                         break;
1351         }
1352
1353         if (map->olm_name == NULL)
1354                 RETURN(0);
1355
1356         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1357         if (IS_ERR(child))
1358                 RETURN(PTR_ERR(child));
1359
1360         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1361                               &map->olm_fid, map->olm_flags);
1362         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
1363                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
1364                                       map->olm_filldir);
1365         dput(child);
1366
1367         RETURN(rc);
1368 }
1369
1370 static int
1371 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1372                      struct dentry *dentry, filldir_t filldir)
1373 {
1374         struct osd_ios_filldir_buf    buf   = { info, dev, dentry };
1375         struct file                  *filp  = &info->oti_it_ea.oie_file;
1376         struct inode                 *inode = dentry->d_inode;
1377         const struct file_operations *fops  = inode->i_fop;
1378         int                           rc;
1379         ENTRY;
1380
1381         LASSERT(filldir != NULL);
1382
1383         filp->f_pos = 0;
1384         filp->f_dentry = dentry;
1385         filp->f_mode = FMODE_64BITHASH;
1386         filp->f_mapping = inode->i_mapping;
1387         filp->f_op = fops;
1388         filp->private_data = NULL;
1389
1390         rc = fops->readdir(filp, &buf, filldir);
1391         fops->release(inode, filp);
1392
1393         RETURN(rc);
1394 }
1395
1396 static int
1397 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1398                   struct dentry *dentry, filldir_t filldir)
1399 {
1400         struct osd_scrub  *scrub  = &dev->od_scrub;
1401         struct scrub_file *sf     = &scrub->os_file;
1402         struct dentry     *child;
1403         int                rc;
1404         ENTRY;
1405
1406         /* It is existing MDT device. */
1407         dev->od_handle_nolma = 1;
1408         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
1409                                        strlen(dot_lustre_name));
1410         if (IS_ERR(child)) {
1411                 rc = PTR_ERR(child);
1412                 if (rc == -ENOENT) {
1413                         /* It is 1.8 MDT device. */
1414                         if (!(sf->sf_flags & SF_UPGRADE)) {
1415                                 osd_scrub_file_reset(scrub,
1416                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1417                                         SF_UPGRADE);
1418                                 rc = osd_scrub_file_store(scrub);
1419                         } else {
1420                                 rc = 0;
1421                         }
1422                 }
1423         } else {
1424                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1425                 dput(child);
1426         }
1427
1428         RETURN(rc);
1429 }
1430
1431 static int
1432 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1433                      struct dentry *dentry, filldir_t filldir)
1434 {
1435         struct dentry *child;
1436         int            rc;
1437         ENTRY;
1438
1439         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
1440         if (!IS_ERR(child)) {
1441                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1442                 dput(child);
1443         } else {
1444                 rc = PTR_ERR(child);
1445         }
1446
1447         if (rc != 0 && rc != -ENOENT)
1448                 RETURN(rc);
1449
1450         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
1451         if (!IS_ERR(child)) {
1452                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1453                 dput(child);
1454         } else {
1455                 rc = PTR_ERR(child);
1456         }
1457
1458         if (rc == -ENOENT)
1459                 rc = 0;
1460
1461         RETURN(rc);
1462 }
1463
1464 static int osd_initial_OI_scrub(struct osd_thread_info *info,
1465                                 struct osd_device *dev)
1466 {
1467         struct osd_ios_item *item    = NULL;
1468         scandir_t            scandir = osd_ios_general_scan;
1469         filldir_t            filldir = osd_ios_root_fill;
1470         struct dentry       *dentry  = osd_sb(dev)->s_root;
1471         int                  rc;
1472         ENTRY;
1473
1474         while (1) {
1475                 rc = scandir(info, dev, dentry, filldir);
1476                 if (item != NULL) {
1477                         dput(item->oii_dentry);
1478                         OBD_FREE_PTR(item);
1479                 }
1480
1481                 if (rc != 0)
1482                         break;
1483
1484                 if (cfs_list_empty(&dev->od_ios_list))
1485                         break;
1486
1487                 item = cfs_list_entry(dev->od_ios_list.next,
1488                                       struct osd_ios_item, oii_list);
1489                 cfs_list_del_init(&item->oii_list);
1490
1491                 LASSERT(item->oii_scandir != NULL);
1492                 scandir = item->oii_scandir;
1493                 filldir = item->oii_filldir;
1494                 dentry = item->oii_dentry;
1495         }
1496
1497         while (!cfs_list_empty(&dev->od_ios_list)) {
1498                 item = cfs_list_entry(dev->od_ios_list.next,
1499                                       struct osd_ios_item, oii_list);
1500                 cfs_list_del_init(&item->oii_list);
1501                 dput(item->oii_dentry);
1502                 OBD_FREE_PTR(item);
1503         }
1504
1505         RETURN(rc);
1506 }
1507
1508 char *osd_lf_fid2name(const struct lu_fid *fid)
1509 {
1510         const struct osd_lf_map *map = osd_lf_maps;
1511
1512         while (map->olm_name != NULL) {
1513                 if (!lu_fid_eq(fid, &map->olm_fid)) {
1514                         map++;
1515                         continue;
1516                 }
1517
1518                 if (map->olm_flags & OLF_SHOW_NAME)
1519                         return map->olm_name;
1520                 else
1521                         return "";
1522         }
1523
1524         return NULL;
1525 }
1526
1527 /* OI scrub start/stop */
1528
1529 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
1530 {
1531         struct osd_scrub     *scrub  = &dev->od_scrub;
1532         struct ptlrpc_thread *thread = &scrub->os_thread;
1533         struct l_wait_info    lwi    = { 0 };
1534         int                   rc;
1535         ENTRY;
1536
1537 again:
1538         /* os_lock: sync status between stop and scrub thread */
1539         spin_lock(&scrub->os_lock);
1540         if (thread_is_running(thread)) {
1541                 spin_unlock(&scrub->os_lock);
1542                 RETURN(-EALREADY);
1543         } else if (unlikely(thread_is_stopping(thread))) {
1544                 spin_unlock(&scrub->os_lock);
1545                 l_wait_event(thread->t_ctl_waitq,
1546                              thread_is_stopped(thread),
1547                              &lwi);
1548                 goto again;
1549         }
1550         spin_unlock(&scrub->os_lock);
1551
1552         if (scrub->os_file.sf_status == SS_COMPLETED)
1553                 flags |= SS_RESET;
1554
1555         scrub->os_start_flags = flags;
1556         thread_set_flags(thread, 0);
1557         rc = cfs_create_thread(osd_scrub_main, dev, 0);
1558         if (rc < 0) {
1559                 CERROR("%.16s: cannot start iteration thread, rc = %d\n",
1560                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
1561                 RETURN(rc);
1562         }
1563
1564         l_wait_event(thread->t_ctl_waitq,
1565                      thread_is_running(thread) || thread_is_stopped(thread),
1566                      &lwi);
1567
1568         RETURN(0);
1569 }
1570
1571 int osd_scrub_start(struct osd_device *dev)
1572 {
1573         int rc;
1574         ENTRY;
1575
1576         /* od_otable_mutex: prevent curcurrent start/stop */
1577         mutex_lock(&dev->od_otable_mutex);
1578         rc = do_osd_scrub_start(dev, SS_AUTO);
1579         mutex_unlock(&dev->od_otable_mutex);
1580
1581         RETURN(rc == -EALREADY ? 0 : rc);
1582 }
1583
1584 static void do_osd_scrub_stop(struct osd_scrub *scrub)
1585 {
1586         struct ptlrpc_thread *thread = &scrub->os_thread;
1587         struct l_wait_info    lwi    = { 0 };
1588
1589         /* os_lock: sync status between stop and scrub thread */
1590         spin_lock(&scrub->os_lock);
1591         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
1592                 thread_set_flags(thread, SVC_STOPPING);
1593                 spin_unlock(&scrub->os_lock);
1594                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
1595                 l_wait_event(thread->t_ctl_waitq,
1596                              thread_is_stopped(thread),
1597                              &lwi);
1598                 /* Do not skip the last lock/unlock, which can guarantee that
1599                  * the caller cannot return until the OI scrub thread exit. */
1600                 spin_lock(&scrub->os_lock);
1601         }
1602         spin_unlock(&scrub->os_lock);
1603 }
1604
1605 static void osd_scrub_stop(struct osd_device *dev)
1606 {
1607         /* od_otable_mutex: prevent curcurrent start/stop */
1608         mutex_lock(&dev->od_otable_mutex);
1609         dev->od_scrub.os_paused = 1;
1610         do_osd_scrub_stop(&dev->od_scrub);
1611         mutex_unlock(&dev->od_otable_mutex);
1612 }
1613
1614 /* OI scrub setup/cleanup */
1615
1616 static const char osd_scrub_name[] = "OI_scrub";
1617
1618 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
1619 {
1620         struct osd_thread_info     *info   = osd_oti_get(env);
1621         struct osd_scrub           *scrub  = &dev->od_scrub;
1622         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
1623         struct scrub_file          *sf     = &scrub->os_file;
1624         struct super_block         *sb     = osd_sb(dev);
1625         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
1626         struct lvfs_run_ctxt        saved;
1627         struct file                *filp;
1628         int                         dirty  = 0;
1629         int                         rc     = 0;
1630         ENTRY;
1631
1632         memset(scrub, 0, sizeof(*scrub));
1633         OBD_SET_CTXT_MAGIC(ctxt);
1634         ctxt->pwdmnt = dev->od_mnt;
1635         ctxt->pwd = dev->od_mnt->mnt_root;
1636         ctxt->fs = get_ds();
1637
1638         cfs_waitq_init(&scrub->os_thread.t_ctl_waitq);
1639         init_rwsem(&scrub->os_rwsem);
1640         spin_lock_init(&scrub->os_lock);
1641         CFS_INIT_LIST_HEAD(&scrub->os_inconsistent_items);
1642
1643         push_ctxt(&saved, ctxt, NULL);
1644         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
1645         if (IS_ERR(filp))
1646                 RETURN(PTR_ERR(filp));
1647
1648         scrub->os_inode = igrab(filp->f_dentry->d_inode);
1649         filp_close(filp, 0);
1650         pop_ctxt(&saved, ctxt, NULL);
1651         ldiskfs_set_inode_state(scrub->os_inode,
1652                                 LDISKFS_STATE_LUSTRE_NO_OI);
1653
1654         rc = osd_scrub_file_load(scrub);
1655         if (rc == -ENOENT) {
1656                 osd_scrub_file_init(scrub, es->s_uuid);
1657                 dirty = 1;
1658         } else if (rc != 0) {
1659                 RETURN(rc);
1660         } else {
1661                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
1662                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
1663                         dirty = 1;
1664                 } else if (sf->sf_status == SS_SCANNING) {
1665                         sf->sf_status = SS_CRASHED;
1666                         dirty = 1;
1667                 }
1668         }
1669
1670         if (sf->sf_pos_last_checkpoint != 0)
1671                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
1672         else
1673                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
1674
1675         if (dirty != 0) {
1676                 rc = osd_scrub_file_store(scrub);
1677                 if (rc != 0)
1678                         RETURN(rc);
1679         }
1680
1681         /* Initialize OI files. */
1682         rc = osd_oi_init(info, dev);
1683         if (rc < 0)
1684                 RETURN(rc);
1685
1686         rc = osd_initial_OI_scrub(info, dev);
1687         if (rc == 0 && !dev->od_noscrub &&
1688             ((sf->sf_status == SS_PAUSED) ||
1689              (sf->sf_status == SS_CRASHED &&
1690               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE |
1691                               SF_AUTO)) ||
1692              (sf->sf_status == SS_INIT &&
1693               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))))
1694                 rc = osd_scrub_start(dev);
1695
1696         RETURN(rc);
1697 }
1698
1699 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
1700 {
1701         struct osd_scrub *scrub = &dev->od_scrub;
1702
1703         LASSERT(dev->od_otable_it == NULL);
1704
1705         if (scrub->os_inode != NULL) {
1706                 osd_scrub_stop(dev);
1707                 iput(scrub->os_inode);
1708                 scrub->os_inode = NULL;
1709         }
1710         if (dev->od_oi_table != NULL)
1711                 osd_oi_fini(osd_oti_get(env), dev);
1712 }
1713
1714 /* object table based iteration APIs */
1715
1716 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
1717                                        struct dt_object *dt, __u32 attr,
1718                                        struct lustre_capa *capa)
1719 {
1720         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
1721         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
1722         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
1723         struct osd_scrub       *scrub = &dev->od_scrub;
1724         struct osd_otable_it   *it;
1725         __u32                   start = 0;
1726         int                     rc;
1727         ENTRY;
1728
1729         /* od_otable_mutex: prevent curcurrent init/fini */
1730         mutex_lock(&dev->od_otable_mutex);
1731         if (dev->od_otable_it != NULL)
1732                 GOTO(out, it = ERR_PTR(-EALREADY));
1733
1734         OBD_ALLOC_PTR(it);
1735         if (it == NULL)
1736                 GOTO(out, it = ERR_PTR(-ENOMEM));
1737
1738         dev->od_otable_it = it;
1739         it->ooi_dev = dev;
1740         it->ooi_pid = cfs_curproc_pid();
1741         it->ooi_cache.ooc_consumer_idx = -1;
1742         if (flags & DOIF_OUTUSED)
1743                 it->ooi_used_outside = 1;
1744
1745         if (flags & DOIF_RESET)
1746                 start |= SS_RESET;
1747
1748         if (valid & DOIV_ERROR_HANDLE) {
1749                 if (flags & DOIF_FAILOUT)
1750                         start |= SS_SET_FAILOUT;
1751                 else
1752                         start |= SS_CLEAR_FAILOUT;
1753         }
1754
1755         rc = do_osd_scrub_start(dev, start);
1756         if (rc < 0 && rc != -EALREADY) {
1757                 dev->od_otable_it = NULL;
1758                 OBD_FREE_PTR(it);
1759                 GOTO(out, it = ERR_PTR(rc));
1760         }
1761
1762         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
1763
1764         GOTO(out, it);
1765
1766 out:
1767         mutex_unlock(&dev->od_otable_mutex);
1768         return (struct dt_it *)it;
1769 }
1770
1771 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
1772 {
1773         struct osd_otable_it *it  = (struct osd_otable_it *)di;
1774         struct osd_device    *dev = it->ooi_dev;
1775
1776         /* od_otable_mutex: prevent curcurrent init/fini */
1777         mutex_lock(&dev->od_otable_mutex);
1778         do_osd_scrub_stop(&dev->od_scrub);
1779         LASSERT(dev->od_otable_it == it);
1780
1781         dev->od_otable_it = NULL;
1782         mutex_unlock(&dev->od_otable_mutex);
1783         OBD_FREE_PTR(it);
1784 }
1785
1786 static int osd_otable_it_get(const struct lu_env *env,
1787                              struct dt_it *di, const struct dt_key *key)
1788 {
1789         return 0;
1790 }
1791
1792 /**
1793  * It is hack here:
1794  *
1795  * Sometimes the otable-based iteration driver (LFSCK) may be blocked in OSD
1796  * layer when someone wants to stop/pause the iteration. Under such case, we
1797  * need some mechanism to notify the event and wakeup the blocker.
1798  */
1799 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
1800 {
1801         struct osd_otable_it *it  = (struct osd_otable_it *)di;
1802         struct osd_device    *dev = it->ooi_dev;
1803
1804         /* od_otable_mutex: prevent curcurrent init/fini */
1805         mutex_lock(&dev->od_otable_mutex);
1806         if (it->ooi_pid == cfs_curproc_pid()) {
1807                 dev->od_scrub.os_paused = 1;
1808         } else {
1809                 struct ptlrpc_thread *thread = &dev->od_scrub.os_thread;
1810
1811                 it->ooi_stopping = 1;
1812                 if (it->ooi_waiting)
1813                         cfs_waitq_broadcast(&thread->t_ctl_waitq);
1814         }
1815         mutex_unlock(&dev->od_otable_mutex);
1816 }
1817
1818 static inline int
1819 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1820 {
1821         spin_lock(&scrub->os_lock);
1822         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
1823             scrub->os_waiting || it->ooi_stopping ||
1824             !thread_is_running(&scrub->os_thread))
1825                 it->ooi_waiting = 0;
1826         else
1827                 it->ooi_waiting = 1;
1828         spin_unlock(&scrub->os_lock);
1829
1830         return !it->ooi_waiting;
1831 }
1832
1833 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
1834 {
1835         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
1836         struct osd_device       *dev    = it->ooi_dev;
1837         struct osd_scrub        *scrub  = &dev->od_scrub;
1838         struct osd_otable_cache *ooc    = &it->ooi_cache;
1839         struct ptlrpc_thread    *thread = &scrub->os_thread;
1840         struct l_wait_info       lwi    = { 0 };
1841         int                      rc;
1842         ENTRY;
1843
1844         LASSERT(it->ooi_user_ready);
1845
1846 again:
1847         if (!thread_is_running(thread) && !it->ooi_used_outside)
1848                 RETURN(1);
1849
1850         if (ooc->ooc_cached_items > 0) {
1851                 ooc->ooc_cached_items--;
1852                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
1853                                         ~OSD_OTABLE_IT_CACHE_MASK;
1854                 RETURN(0);
1855         }
1856
1857         if (it->ooi_all_cached) {
1858                 l_wait_event(thread->t_ctl_waitq,
1859                              !thread_is_running(thread),
1860                              &lwi);
1861                 RETURN(1);
1862         }
1863
1864         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1865                 spin_lock(&scrub->os_lock);
1866                 scrub->os_waiting = 0;
1867                 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
1868                 spin_unlock(&scrub->os_lock);
1869         }
1870
1871         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
1872                 l_wait_event(thread->t_ctl_waitq,
1873                              osd_otable_it_wakeup(scrub, it),
1874                              &lwi);
1875
1876         if (!thread_is_running(thread) && !it->ooi_used_outside)
1877                 RETURN(1);
1878
1879         if (it->ooi_stopping)
1880                 RETURN(0);
1881
1882         rc = osd_otable_it_preload(env, it);
1883         if (rc >= 0)
1884                 goto again;
1885
1886         RETURN(rc);
1887 }
1888
1889 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
1890                                         const struct dt_it *di)
1891 {
1892         return NULL;
1893 }
1894
1895 static int osd_otable_it_key_size(const struct lu_env *env,
1896                                   const struct dt_it *di)
1897 {
1898         return sizeof(__u64);
1899 }
1900
1901 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
1902                              struct dt_rec *rec, __u32 attr)
1903 {
1904         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
1905         struct osd_otable_cache *ooc = &it->ooi_cache;
1906
1907         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
1908
1909         /* Filter out Invald FID already. */
1910         LASSERTF(fid_is_sane((struct lu_fid *)rec),
1911                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
1912                  PFID((struct lu_fid *)rec),
1913                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
1914
1915         return 0;
1916 }
1917
1918 static __u64 osd_otable_it_store(const struct lu_env *env,
1919                                  const struct dt_it *di)
1920 {
1921         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
1922         struct osd_otable_cache *ooc = &it->ooi_cache;
1923         __u64                    hash;
1924
1925         if (it->ooi_user_ready)
1926                 hash = ooc->ooc_pos_preload;
1927         else if (likely(ooc->ooc_consumer_idx != -1))
1928                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
1929         else
1930                 hash = 0;
1931         return hash;
1932 }
1933
1934 /**
1935  * Set the OSD layer iteration start position as the specified hash.
1936  */
1937 static int osd_otable_it_load(const struct lu_env *env,
1938                               const struct dt_it *di, __u64 hash)
1939 {
1940         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
1941         struct osd_device       *dev   = it->ooi_dev;
1942         struct osd_otable_cache *ooc   = &it->ooi_cache;
1943         struct osd_scrub        *scrub = &dev->od_scrub;
1944         int                      rc;
1945         ENTRY;
1946
1947         /* Forbid to set iteration position after iteration started. */
1948         if (it->ooi_user_ready)
1949                 RETURN(-EPERM);
1950
1951         if (hash > OSD_OTABLE_MAX_HASH)
1952                 hash = OSD_OTABLE_MAX_HASH;
1953
1954         /* Skip the one that has been processed last time. */
1955         if (ooc->ooc_pos_preload > hash)
1956                 ooc->ooc_pos_preload = hash;
1957
1958         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
1959                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1960
1961         it->ooi_user_ready = 1;
1962         if (!scrub->os_full_speed)
1963                 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
1964
1965         /* Unplug OSD layer iteration by the first next() call. */
1966         rc = osd_otable_it_next(env, (struct dt_it *)it);
1967
1968         RETURN(rc);
1969 }
1970
1971 static int osd_otable_it_key_rec(const struct lu_env *env,
1972                                  const struct dt_it *di, void *key_rec)
1973 {
1974         return 0;
1975 }
1976
1977 const struct dt_index_operations osd_otable_ops = {
1978         .dio_it = {
1979                 .init     = osd_otable_it_init,
1980                 .fini     = osd_otable_it_fini,
1981                 .get      = osd_otable_it_get,
1982                 .put      = osd_otable_it_put,
1983                 .next     = osd_otable_it_next,
1984                 .key      = osd_otable_it_key,
1985                 .key_size = osd_otable_it_key_size,
1986                 .rec      = osd_otable_it_rec,
1987                 .store    = osd_otable_it_store,
1988                 .load     = osd_otable_it_load,
1989                 .key_rec  = osd_otable_it_key_rec,
1990         }
1991 };
1992
1993 /* high priority inconsistent items list APIs */
1994
1995 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
1996                    int insert)
1997 {
1998         struct osd_inconsistent_item *oii;
1999         struct osd_scrub             *scrub  = &dev->od_scrub;
2000         struct ptlrpc_thread         *thread = &scrub->os_thread;
2001         int                           wakeup = 0;
2002         ENTRY;
2003
2004         OBD_ALLOC_PTR(oii);
2005         if (unlikely(oii == NULL))
2006                 RETURN(-ENOMEM);
2007
2008         CFS_INIT_LIST_HEAD(&oii->oii_list);
2009         oii->oii_cache = *oic;
2010         oii->oii_insert = insert;
2011
2012         spin_lock(&scrub->os_lock);
2013         if (unlikely(!thread_is_running(thread))) {
2014                 spin_unlock(&scrub->os_lock);
2015                 OBD_FREE_PTR(oii);
2016                 RETURN(-EAGAIN);
2017         }
2018
2019         if (cfs_list_empty(&scrub->os_inconsistent_items))
2020                 wakeup = 1;
2021         cfs_list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2022         spin_unlock(&scrub->os_lock);
2023
2024         if (wakeup != 0)
2025                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
2026
2027         RETURN(0);
2028 }
2029
2030 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2031                    struct osd_inode_id *id)
2032 {
2033         struct osd_scrub             *scrub = &dev->od_scrub;
2034         struct osd_inconsistent_item *oii;
2035         ENTRY;
2036
2037         spin_lock(&scrub->os_lock);
2038         cfs_list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2039                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2040                         *id = oii->oii_cache.oic_lid;
2041                         spin_unlock(&scrub->os_lock);
2042                         RETURN(0);
2043                 }
2044         }
2045         spin_unlock(&scrub->os_lock);
2046
2047         RETURN(-ENOENT);
2048 }
2049
2050 /* OI scrub dump */
2051
2052 static const char *scrub_status_names[] = {
2053         "init",
2054         "scanning",
2055         "completed",
2056         "failed",
2057         "stopped",
2058         "paused",
2059         "crashed",
2060         NULL
2061 };
2062
2063 static const char *scrub_flags_names[] = {
2064         "recreated",
2065         "inconsistent",
2066         "auto",
2067         "upgrade",
2068         NULL
2069 };
2070
2071 static const char *scrub_param_names[] = {
2072         "failout",
2073         NULL
2074 };
2075
2076 static int scrub_bits_dump(char **buf, int *len, int bits, const char *names[],
2077                            const char *prefix)
2078 {
2079         int save = *len;
2080         int flag;
2081         int rc;
2082         int i;
2083
2084         rc = snprintf(*buf, *len, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2085         if (rc <= 0)
2086                 return -ENOSPC;
2087
2088         *buf += rc;
2089         *len -= rc;
2090         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2091                 if (flag & bits) {
2092                         bits &= ~flag;
2093                         rc = snprintf(*buf, *len, "%s%c", names[i],
2094                                       bits != 0 ? ',' : '\n');
2095                         if (rc <= 0)
2096                                 return -ENOSPC;
2097
2098                         *buf += rc;
2099                         *len -= rc;
2100                 }
2101         }
2102         return save - *len;
2103 }
2104
2105 static int scrub_time_dump(char **buf, int *len, __u64 time, const char *prefix)
2106 {
2107         int rc;
2108
2109         if (time != 0)
2110                 rc = snprintf(*buf, *len, "%s: "LPU64" seconds\n", prefix,
2111                               cfs_time_current_sec() - time);
2112         else
2113                 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
2114         if (rc <= 0)
2115                 return -ENOSPC;
2116
2117         *buf += rc;
2118         *len -= rc;
2119         return rc;
2120 }
2121
2122 static int scrub_pos_dump(char **buf, int *len, __u64 pos, const char *prefix)
2123 {
2124         int rc;
2125
2126         if (pos != 0)
2127                 rc = snprintf(*buf, *len, "%s: "LPU64"\n", prefix, pos);
2128         else
2129                 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
2130         if (rc <= 0)
2131                 return -ENOSPC;
2132
2133         *buf += rc;
2134         *len -= rc;
2135         return rc;
2136 }
2137
2138 int osd_scrub_dump(struct osd_device *dev, char *buf, int len)
2139 {
2140         struct osd_scrub  *scrub   = &dev->od_scrub;
2141         struct scrub_file *sf      = &scrub->os_file;
2142         __u64              checked;
2143         __u64              speed;
2144         int                save    = len;
2145         int                ret     = -ENOSPC;
2146         int                rc;
2147
2148         down_read(&scrub->os_rwsem);
2149         rc = snprintf(buf, len,
2150                       "name: OI_scrub\n"
2151                       "magic: 0x%x\n"
2152                       "oi_files: %d\n"
2153                       "status: %s\n",
2154                       sf->sf_magic, (int)sf->sf_oi_count,
2155                       scrub_status_names[sf->sf_status]);
2156         if (rc <= 0)
2157                 goto out;
2158
2159         buf += rc;
2160         len -= rc;
2161         rc = scrub_bits_dump(&buf, &len, sf->sf_flags, scrub_flags_names,
2162                              "flags");
2163         if (rc < 0)
2164                 goto out;
2165
2166         rc = scrub_bits_dump(&buf, &len, sf->sf_param, scrub_param_names,
2167                              "param");
2168         if (rc < 0)
2169                 goto out;
2170
2171         rc = scrub_time_dump(&buf, &len, sf->sf_time_last_complete,
2172                              "time_since_last_completed");
2173         if (rc < 0)
2174                 goto out;
2175
2176         rc = scrub_time_dump(&buf, &len, sf->sf_time_latest_start,
2177                              "time_since_latest_start");
2178         if (rc < 0)
2179                 goto out;
2180
2181         rc = scrub_time_dump(&buf, &len, sf->sf_time_last_checkpoint,
2182                              "time_since_last_checkpoint");
2183         if (rc < 0)
2184                 goto out;
2185
2186         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_latest_start,
2187                             "latest_start_position");
2188         if (rc < 0)
2189                 goto out;
2190
2191         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_last_checkpoint,
2192                             "last_checkpoint_position");
2193         if (rc < 0)
2194                 goto out;
2195
2196         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_first_inconsistent,
2197                             "first_failure_position");
2198         if (rc < 0)
2199                 goto out;
2200
2201         checked = sf->sf_items_checked + scrub->os_new_checked;
2202         rc = snprintf(buf, len,
2203                       "checked: "LPU64"\n"
2204                       "updated: "LPU64"\n"
2205                       "failed: "LPU64"\n"
2206                       "prior_updated: "LPU64"\n"
2207                       "noscrub: "LPU64"\n"
2208                       "igif: "LPU64"\n"
2209                       "success_count: %u\n",
2210                       checked, sf->sf_items_updated, sf->sf_items_failed,
2211                       sf->sf_items_updated_prior, sf->sf_items_noscrub,
2212                       sf->sf_items_igif, sf->sf_success_count);
2213         if (rc <= 0)
2214                 goto out;
2215
2216         buf += rc;
2217         len -= rc;
2218         speed = checked;
2219         if (thread_is_running(&scrub->os_thread)) {
2220                 cfs_duration_t duration = cfs_time_current() -
2221                                           scrub->os_time_last_checkpoint;
2222                 __u64 new_checked = scrub->os_new_checked * CFS_HZ;
2223                 __u32 rtime = sf->sf_run_time +
2224                               cfs_duration_sec(duration + HALF_SEC);
2225
2226                 if (duration != 0)
2227                         do_div(new_checked, duration);
2228                 if (rtime != 0)
2229                         do_div(speed, rtime);
2230                 rc = snprintf(buf, len,
2231                               "run_time: %u seconds\n"
2232                               "average_speed: "LPU64" objects/sec\n"
2233                               "real-time_speed: "LPU64" objects/sec\n"
2234                               "current_position: %u\n",
2235                               rtime, speed, new_checked, scrub->os_pos_current);
2236         } else {
2237                 if (sf->sf_run_time != 0)
2238                         do_div(speed, sf->sf_run_time);
2239                 rc = snprintf(buf, len,
2240                               "run_time: %u seconds\n"
2241                               "average_speed: "LPU64" objects/sec\n"
2242                               "real-time_speed: N/A\n"
2243                               "current_position: N/A\n",
2244                               sf->sf_run_time, speed);
2245         }
2246         if (rc <= 0)
2247                 goto out;
2248
2249         buf += rc;
2250         len -= rc;
2251         ret = save - len;
2252
2253 out:
2254         up_read(&scrub->os_rwsem);
2255         return ret;
2256 }