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, 2013, 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_MDS
  39
  40 #include <lustre/lustre_idl.h>
  41 #include <lustre_disk.h>
  42 #include <dt_object.h>
  43
  44 #include "osd_internal.h"
  45 #include "osd_oi.h"
  46 #include "osd_scrub.h"
  47
  48 #define HALF_SEC        (CFS_HZ >> 1)
  49
  50 #define OSD_OTABLE_MAX_HASH             0x00000000ffffffffULL
  51
  52 #define SCRUB_NEXT_BREAK        1 /* exit current loop and process next group */
  53 #define SCRUB_NEXT_CONTINUE     2 /* skip current object and process next bit */
  54 #define SCRUB_NEXT_EXIT         3 /* exit all the loops */
  55 #define SCRUB_NEXT_WAIT         4 /* wait for free cache slot */
  56 #define SCRUB_NEXT_CRASH        5 /* simulate system crash during OI scrub */
  57 #define SCRUB_NEXT_FATAL        6 /* simulate failure during OI scrub */
  58 #define SCRUB_NEXT_NOSCRUB      7 /* new created object, no scrub on it */
  59 #define SCRUB_NEXT_NOLMA        8 /* the inode has no FID-in-LMA */
  60 #define SCRUB_NEXT_OSTOBJ       9 /* for OST-object */
  61 #define SCRUB_NEXT_OSTOBJ_OLD   10 /* old OST-object, no LMA or no FID-on-OST
  62                                     * flags 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 /**
  83  * update/insert/delete the specified OI mapping (@fid @id) according to the ops
  84  *
  85  * \retval   1, changed nothing
  86  * \retval   0, changed successfully
  87  * \retval -ve, on error
  88  */
  89 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
  90                                      struct osd_device *dev,
  91                                      const struct lu_fid *fid,
  92                                      const struct osd_inode_id *id,
  93                                      int ops, enum oi_check_flags flags)
  94 {
  95         const struct lu_env *env = info->oti_env;
  96         struct thandle      *th;
  97         struct osd_thandle  *oh;
  98         int                  rc;
  99         ENTRY;
 100
 101         th = dt_trans_create(env, &dev->od_dt_dev);
 102         if (IS_ERR(th))
 103                 RETURN(PTR_ERR(th));
 104
 105         oh = container_of0(th, struct osd_thandle, ot_super);
 106         LASSERT(oh->ot_handle == NULL);
 107
 108         switch (ops) {
 109         case DTO_INDEX_UPDATE:
 110                 osd_trans_declare_op(env, oh, OSD_OT_UPDATE,
 111                                      osd_dto_credits_noquota[DTO_INDEX_UPDATE]);
 112                 rc = dt_trans_start_local(env, &dev->od_dt_dev, th);
 113                 if (rc != 0)
 114                         GOTO(stop, rc);
 115
 116                 rc = osd_oi_update(info, dev, fid, id, th, flags);
 117                 if (unlikely(rc == -ENOENT)) {
 118                         /* Some unlink thread may removed the OI mapping. */
 119                         rc = 1;
 120                 }
 121                 break;
 122         case DTO_INDEX_INSERT:
 123                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
 124                                      osd_dto_credits_noquota[DTO_INDEX_INSERT]);
 125                 rc = dt_trans_start_local(env, &dev->od_dt_dev, th);
 126                 if (rc != 0)
 127                         GOTO(stop, rc);
 128
 129                 rc = osd_oi_insert(info, dev, fid, id, th, flags);
 130                 if (unlikely(rc == -EEXIST)) {
 131                         rc = 1;
 132                         /* XXX: There are trouble things when adding OI
 133                          *      mapping for IGIF object, which may cause
 134                          *      multiple objects to be mapped to the same
 135                          *      IGIF formatted FID. Consider the following
 136                          *      situations:
 137                          *
 138                          *      1) The MDT is upgrading from 1.8 device.
 139                          *      The OI scrub generates IGIF FID1 for the
 140                          *      OBJ1 and adds the OI mapping.
 141                          *
 142                          *      2) For some reason, the OI scrub does not
 143                          *      process all the IGIF objects completely.
 144                          *
 145                          *      3) The MDT is backuped and restored against
 146                          *      this device.
 147                          *
 148                          *      4) When the MDT mounts up, the OI scrub will
 149                          *      try to rebuild the OI files. For some IGIF
 150                          *      object, OBJ2, which was not processed by the
 151                          *      OI scrub before the backup/restore, and the
 152                          *      new generated IGIF formatted FID may be just
 153                          *      the FID1, the same as OBJ1.
 154                          *
 155                          *      Under such case, the OI scrub cannot know how
 156                          *      to generate new FID for the OBJ2.
 157                          *
 158                          *      Currently, we do nothing for that. One possible
 159                          *      solution is to generate new normal FID for the
 160                          *      conflict object.
 161                          *
 162                          *      Anyway, it is rare, only exists in theory. */
 163                 }
 164                 break;
 165         case DTO_INDEX_DELETE:
 166                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
 167                                      osd_dto_credits_noquota[DTO_INDEX_DELETE]);
 168                 rc = dt_trans_start_local(env, &dev->od_dt_dev, th);
 169                 if (rc != 0)
 170                         GOTO(stop, rc);
 171
 172                 rc = osd_oi_delete(info, dev, fid, th, flags);
 173                 if (rc == -ENOENT) {
 174                         /* It is normal that the unlink thread has removed the
 175                          * OI mapping already. */
 176                         rc = 1;
 177                 }
 178                 break;
 179         default:
 180                 LASSERTF(0, "Unexpected ops %d\n", ops);
 181                 break;
 182         }
 183
 184         GOTO(stop, rc);
 185
 186 stop:
 187         dt_trans_stop(env, &dev->od_dt_dev, th);
 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         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 225 }
 226
 227 static void osd_scrub_file_to_le(struct scrub_file *des,
 228                                  struct scrub_file *src)
 229 {
 230         memcpy(des->sf_uuid, src->sf_uuid, 16);
 231         des->sf_flags   = cpu_to_le64(src->sf_flags);
 232         des->sf_magic   = cpu_to_le32(src->sf_magic);
 233         des->sf_status  = cpu_to_le16(src->sf_status);
 234         des->sf_param   = cpu_to_le16(src->sf_param);
 235         des->sf_time_last_complete      =
 236                                 cpu_to_le64(src->sf_time_last_complete);
 237         des->sf_time_latest_start       =
 238                                 cpu_to_le64(src->sf_time_latest_start);
 239         des->sf_time_last_checkpoint    =
 240                                 cpu_to_le64(src->sf_time_last_checkpoint);
 241         des->sf_pos_latest_start        =
 242                                 cpu_to_le64(src->sf_pos_latest_start);
 243         des->sf_pos_last_checkpoint     =
 244                                 cpu_to_le64(src->sf_pos_last_checkpoint);
 245         des->sf_pos_first_inconsistent  =
 246                                 cpu_to_le64(src->sf_pos_first_inconsistent);
 247         des->sf_items_checked           =
 248                                 cpu_to_le64(src->sf_items_checked);
 249         des->sf_items_updated           =
 250                                 cpu_to_le64(src->sf_items_updated);
 251         des->sf_items_failed            =
 252                                 cpu_to_le64(src->sf_items_failed);
 253         des->sf_items_updated_prior     =
 254                                 cpu_to_le64(src->sf_items_updated_prior);
 255         des->sf_run_time        = cpu_to_le32(src->sf_run_time);
 256         des->sf_success_count   = cpu_to_le32(src->sf_success_count);
 257         des->sf_oi_count        = cpu_to_le16(src->sf_oi_count);
 258         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 259 }
 260
 261 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
 262 {
 263         struct scrub_file *sf = &scrub->os_file;
 264
 265         memset(sf, 0, sizeof(*sf));
 266         memcpy(sf->sf_uuid, uuid, 16);
 267         sf->sf_magic = SCRUB_MAGIC_V1;
 268         sf->sf_status = SS_INIT;
 269 }
 270
 271 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
 272 {
 273         struct scrub_file *sf = &scrub->os_file;
 274
 275         CDEBUG(D_LFSCK, "Reset OI scrub file, flags = "LPX64"\n", flags);
 276         memcpy(sf->sf_uuid, uuid, 16);
 277         sf->sf_status = SS_INIT;
 278         sf->sf_flags |= flags;
 279         sf->sf_param = 0;
 280         sf->sf_run_time = 0;
 281         sf->sf_time_latest_start = 0;
 282         sf->sf_time_last_checkpoint = 0;
 283         sf->sf_pos_latest_start = 0;
 284         sf->sf_pos_last_checkpoint = 0;
 285         sf->sf_pos_first_inconsistent = 0;
 286         sf->sf_items_checked = 0;
 287         sf->sf_items_updated = 0;
 288         sf->sf_items_failed = 0;
 289         sf->sf_items_updated_prior = 0;
 290         sf->sf_items_noscrub = 0;
 291         sf->sf_items_igif = 0;
 292 }
 293
 294 static int osd_scrub_file_load(struct osd_scrub *scrub)
 295 {
 296         loff_t  pos  = 0;
 297         char   *name = LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
 298         int     len  = sizeof(scrub->os_file_disk);
 299         int     rc;
 300
 301         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
 302         if (rc == len) {
 303                 struct scrub_file *sf = &scrub->os_file;
 304
 305                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
 306                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
 307                         CWARN("%.16s: invalid scrub magic 0x%x != 0x%x\n,",
 308                               name, sf->sf_magic, SCRUB_MAGIC_V1);
 309                         /* Process it as new scrub file. */
 310                         rc = -ENOENT;
 311                 } else {
 312                         rc = 0;
 313                 }
 314         } else if (rc != 0) {
 315                 CERROR("%.16s: fail to load scrub file, expected = %d, "
 316                        "rc = %d\n", name, len, rc);
 317                 if (rc > 0)
 318                         rc = -EFAULT;
 319         } else {
 320                 /* return -ENOENT for empty scrub file case. */
 321                 rc = -ENOENT;
 322         }
 323
 324         return rc;
 325 }
 326
 327 int osd_scrub_file_store(struct osd_scrub *scrub)
 328 {
 329         struct osd_device *dev;
 330         handle_t          *jh;
 331         loff_t             pos     = 0;
 332         int                len     = sizeof(scrub->os_file_disk);
 333         int                credits;
 334         int                rc;
 335
 336         dev = container_of0(scrub, struct osd_device, od_scrub);
 337         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
 338                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
 339         jh = ldiskfs_journal_start_sb(osd_sb(dev), credits);
 340         if (IS_ERR(jh)) {
 341                 rc = PTR_ERR(jh);
 342                 CERROR("%.16s: fail to start trans for scrub store, rc = %d\n",
 343                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,rc);
 344                 return rc;
 345         }
 346
 347         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
 348         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
 349                                       len, 0, &pos, jh);
 350         ldiskfs_journal_stop(jh);
 351         if (rc != 0)
 352                 CERROR("%.16s: fail to store scrub file, expected = %d, "
 353                        "rc = %d\n",
 354                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
 355                        len, rc);
 356         scrub->os_time_last_checkpoint = cfs_time_current();
 357         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
 358                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
 359         return rc;
 360 }
 361
 362 /* OI scrub APIs */
 363
 364 static int osd_scrub_prep(struct osd_device *dev)
 365 {
 366         struct osd_scrub     *scrub  = &dev->od_scrub;
 367         struct ptlrpc_thread *thread = &scrub->os_thread;
 368         struct scrub_file    *sf     = &scrub->os_file;
 369         __u32                 flags  = scrub->os_start_flags;
 370         int                   rc;
 371         ENTRY;
 372
 373         down_write(&scrub->os_rwsem);
 374         if (flags & SS_SET_FAILOUT)
 375                 sf->sf_param |= SP_FAILOUT;
 376
 377         if (flags & SS_CLEAR_FAILOUT)
 378                 sf->sf_param &= ~SP_FAILOUT;
 379
 380         if (flags & SS_RESET)
 381                 osd_scrub_file_reset(scrub,
 382                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
 383
 384         if (flags & SS_AUTO) {
 385                 scrub->os_full_speed = 1;
 386                 sf->sf_flags |= SF_AUTO;
 387         } else {
 388                 scrub->os_full_speed = 0;
 389         }
 390
 391         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
 392                 scrub->os_full_speed = 1;
 393
 394         scrub->os_in_prior = 0;
 395         scrub->os_waiting = 0;
 396         scrub->os_paused = 0;
 397         scrub->os_new_checked = 0;
 398         if (sf->sf_pos_last_checkpoint != 0)
 399                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 400         else
 401                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 402
 403         scrub->os_pos_current = sf->sf_pos_latest_start;
 404         sf->sf_status = SS_SCANNING;
 405         sf->sf_time_latest_start = cfs_time_current_sec();
 406         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 407         rc = osd_scrub_file_store(scrub);
 408         if (rc == 0) {
 409                 spin_lock(&scrub->os_lock);
 410                 thread_set_flags(thread, SVC_RUNNING);
 411                 spin_unlock(&scrub->os_lock);
 412                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
 413         }
 414         up_write(&scrub->os_rwsem);
 415
 416         RETURN(rc);
 417 }
 418
 419 static int
 420 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
 421                        struct osd_idmap_cache *oic, int val)
 422 {
 423         struct osd_scrub             *scrub  = &dev->od_scrub;
 424         struct scrub_file            *sf     = &scrub->os_file;
 425         struct lu_fid                *fid    = &oic->oic_fid;
 426         struct osd_inode_id          *lid    = &oic->oic_lid;
 427         struct osd_inode_id          *lid2   = &info->oti_id;
 428         struct osd_inconsistent_item *oii    = NULL;
 429         struct inode                 *inode  = NULL;
 430         int                           ops    = DTO_INDEX_UPDATE;
 431         int                           idx;
 432         int                           rc;
 433         ENTRY;
 434
 435         down_write(&scrub->os_rwsem);
 436         scrub->os_new_checked++;
 437         if (val < 0)
 438                 GOTO(out, rc = val);
 439
 440         if (scrub->os_in_prior)
 441                 oii = cfs_list_entry(oic, struct osd_inconsistent_item,
 442                                      oii_cache);
 443
 444         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
 445                 GOTO(out, rc = 0);
 446
 447         if (fid_is_igif(fid))
 448                 sf->sf_items_igif++;
 449
 450         if (val == SCRUB_NEXT_OSTOBJ_OLD) {
 451                 inode = osd_iget(info, dev, lid);
 452                 if (IS_ERR(inode)) {
 453                         rc = PTR_ERR(inode);
 454                         /* Someone removed the inode. */
 455                         if (rc == -ENOENT || rc == -ESTALE)
 456                                 rc = 0;
 457                         GOTO(out, rc);
 458                 }
 459
 460                 sf->sf_flags |= SF_UPGRADE;
 461                 rc = osd_ea_fid_set(info, inode, fid,
 462                                     LMAC_FID_ON_OST, 0);
 463                 if (rc != 0)
 464                         GOTO(out, rc);
 465         }
 466
 467         if ((val == SCRUB_NEXT_NOLMA) &&
 468             (!dev->od_handle_nolma || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
 469                 GOTO(out, rc = 0);
 470
 471         if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA))
 472                 goto iget;
 473
 474         rc = osd_oi_lookup(info, dev, fid, lid2,
 475                 (val == SCRUB_NEXT_OSTOBJ ||
 476                  val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 477         if (rc != 0) {
 478                 if (rc != -ENOENT)
 479                         GOTO(out, rc);
 480
 481 iget:
 482                 if (inode == NULL) {
 483                         inode = osd_iget(info, dev, lid);
 484                         if (IS_ERR(inode)) {
 485                                 rc = PTR_ERR(inode);
 486                                 /* Someone removed the inode. */
 487                                 if (rc == -ENOENT || rc == -ESTALE)
 488                                         rc = 0;
 489                                 GOTO(out, rc);
 490                         }
 491                 }
 492
 493                 scrub->os_full_speed = 1;
 494                 ops = DTO_INDEX_INSERT;
 495                 idx = osd_oi_fid2idx(dev, fid);
 496                 switch (val) {
 497                 case SCRUB_NEXT_NOLMA:
 498                         sf->sf_flags |= SF_UPGRADE;
 499                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
 500                         if (rc != 0)
 501                                 GOTO(out, rc);
 502
 503                         if (!(sf->sf_flags & SF_INCONSISTENT))
 504                                 dev->od_igif_inoi = 0;
 505                         break;
 506                 case SCRUB_NEXT_OSTOBJ:
 507                         sf->sf_flags |= SF_INCONSISTENT;
 508                 case SCRUB_NEXT_OSTOBJ_OLD:
 509                         break;
 510                 default:
 511                         sf->sf_flags |= SF_RECREATED;
 512                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 513                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 514                         break;
 515                 }
 516         } else if (osd_id_eq(lid, lid2)) {
 517                 GOTO(out, rc = 0);
 518         } else {
 519                 scrub->os_full_speed = 1;
 520                 sf->sf_flags |= SF_INCONSISTENT;
 521
 522                 /* XXX: If the device is restored from file-level backup, then
 523                  *      some IGIFs may have been already in OI files, and some
 524                  *      may be not yet. Means upgrading from 1.8 may be partly
 525                  *      processed, but some clients may hold some immobilized
 526                  *      IGIFs, and use them to access related objects. Under
 527                  *      such case, OSD does not know whether an given IGIF has
 528                  *      been processed or to be processed, and it also cannot
 529                  *      generate local ino#/gen# directly from the immobilized
 530                  *      IGIF because of the backup/restore. Then force OSD to
 531                  *      lookup the given IGIF in OI files, and if no entry,
 532                  *      then ask the client to retry after upgrading completed.
 533                  *      No better choice. */
 534                 dev->od_igif_inoi = 1;
 535         }
 536
 537         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops,
 538                         (val == SCRUB_NEXT_OSTOBJ ||
 539                          val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 540         if (rc == 0) {
 541                 if (scrub->os_in_prior)
 542                         sf->sf_items_updated_prior++;
 543                 else
 544                         sf->sf_items_updated++;
 545
 546                 /* The target has been changed, need to be re-loaded. */
 547                 lu_object_purge(info->oti_env, osd2lu_dev(dev), fid);
 548         }
 549
 550         GOTO(out, rc);
 551
 552 out:
 553         if (rc < 0) {
 554                 sf->sf_items_failed++;
 555                 if (sf->sf_pos_first_inconsistent == 0 ||
 556                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 557                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 558         } else {
 559                 rc = 0;
 560         }
 561
 562         /* There may be conflict unlink during the OI scrub,
 563          * if happend, then remove the new added OI mapping. */
 564         if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
 565             unlikely(inode->i_nlink == 0))
 566                 osd_scrub_refresh_mapping(info, dev, fid, lid,
 567                                 DTO_INDEX_DELETE,
 568                                 (val == SCRUB_NEXT_OSTOBJ ||
 569                                  val == SCRUB_NEXT_OSTOBJ_OLD) ?
 570                                 OI_KNOWN_ON_OST : 0);
 571         up_write(&scrub->os_rwsem);
 572
 573         if (inode != NULL && !IS_ERR(inode))
 574                 iput(inode);
 575
 576         if (oii != NULL) {
 577                 LASSERT(!cfs_list_empty(&oii->oii_list));
 578
 579                 spin_lock(&scrub->os_lock);
 580                 cfs_list_del_init(&oii->oii_list);
 581                 spin_unlock(&scrub->os_lock);
 582                 OBD_FREE_PTR(oii);
 583         }
 584         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 585 }
 586
 587 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
 588 {
 589         struct scrub_file *sf = &scrub->os_file;
 590         int                rc;
 591
 592         if (likely(cfs_time_before(cfs_time_current(),
 593                                    scrub->os_time_next_checkpoint) ||
 594                    scrub->os_new_checked == 0))
 595                 return 0;
 596
 597         down_write(&scrub->os_rwsem);
 598         sf->sf_items_checked += scrub->os_new_checked;
 599         scrub->os_new_checked = 0;
 600         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 601         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 602         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 603                                             scrub->os_time_last_checkpoint);
 604         rc = osd_scrub_file_store(scrub);
 605         up_write(&scrub->os_rwsem);
 606
 607         return rc;
 608 }
 609
 610 static void osd_scrub_post(struct osd_scrub *scrub, int result)
 611 {
 612         struct scrub_file *sf = &scrub->os_file;
 613         ENTRY;
 614
 615         down_write(&scrub->os_rwsem);
 616         spin_lock(&scrub->os_lock);
 617         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 618         spin_unlock(&scrub->os_lock);
 619         if (scrub->os_new_checked > 0) {
 620                 sf->sf_items_checked += scrub->os_new_checked;
 621                 scrub->os_new_checked = 0;
 622                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 623         }
 624         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 625         if (result > 0) {
 626                 struct osd_device *dev =
 627                         container_of0(scrub, struct osd_device, od_scrub);
 628
 629                 dev->od_igif_inoi = 1;
 630                 sf->sf_status = SS_COMPLETED;
 631                 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 632                 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 633                                   SF_UPGRADE | SF_AUTO);
 634                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 635                 sf->sf_success_count++;
 636         } else if (result == 0) {
 637                 if (scrub->os_paused)
 638                         sf->sf_status = SS_PAUSED;
 639                 else
 640                         sf->sf_status = SS_STOPPED;
 641         } else {
 642                 sf->sf_status = SS_FAILED;
 643         }
 644         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 645                                             scrub->os_time_last_checkpoint);
 646         result = osd_scrub_file_store(scrub);
 647         if (result < 0)
 648                 CERROR("%.16s: fail to osd_scrub_post, rc = %d\n",
 649                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
 650                        result);
 651         up_write(&scrub->os_rwsem);
 652
 653         EXIT;
 654 }
 655
 656 /* iteration engine */
 657
 658 struct osd_iit_param {
 659         struct super_block *sb;
 660         struct buffer_head *bitmap;
 661         ldiskfs_group_t bg;
 662         __u32 gbase;
 663         __u32 offset;
 664 };
 665
 666 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 667                                    struct osd_device *dev,
 668                                    struct osd_iit_param *param,
 669                                    struct osd_idmap_cache **oic,
 670                                    int noslot);
 671
 672 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 673                                    struct osd_device *dev,
 674                                    struct osd_iit_param *param,
 675                                    struct osd_idmap_cache *oic,
 676                                    int *noslot, int rc);
 677
 678 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 679 {
 680         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 681                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 682         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 683                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 684                 return SCRUB_NEXT_BREAK;
 685         } else {
 686                 *pos = param->gbase + param->offset;
 687                 return 0;
 688         }
 689 }
 690
 691 /**
 692  * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
 693  * \retval 0: FID-on-MDT
 694  */
 695 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
 696                                       struct osd_device *dev,
 697                                       struct lu_fid *fid)
 698 {
 699         /* XXX: The initial OI scrub will scan the top level /O to generate
 700          *      a small local FLDB according to the <seq>. If the given FID
 701          *      is in the local FLDB, then it is FID-on-OST; otherwise it's
 702          *      quite possible for FID-on-MDT. */
 703         return 0;
 704 }
 705
 706 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 707                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 708                         struct super_block *sb, bool scrub)
 709 {
 710         struct lustre_mdt_attrs *lma            = &info->oti_mdt_attrs;
 711         struct inode            *inode;
 712         int                      rc             = 0;
 713         bool                     has_lma        = false;
 714         ENTRY;
 715
 716         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 717         inode = osd_iget(info, dev, lid);
 718         if (IS_ERR(inode)) {
 719                 rc = PTR_ERR(inode);
 720                 /* The inode may be removed after bitmap searching, or the
 721                  * file is new created without inode initialized yet. */
 722                 if (rc == -ENOENT || rc == -ESTALE)
 723                         RETURN(SCRUB_NEXT_CONTINUE);
 724
 725                 CERROR("%.16s: fail to read inode, ino# = %u, rc = %d\n",
 726                        LDISKFS_SB(sb)->s_es->s_volume_name, pos, rc);
 727                 RETURN(rc);
 728         }
 729
 730         /* If the inode has no OI mapping, then it is special locally used,
 731          * should be invisible to OI scrub or up layer LFSCK. */
 732         if (ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI))
 733                 GOTO(put, rc = SCRUB_NEXT_CONTINUE);
 734
 735         if (scrub &&
 736             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 737                 /* Only skip it for the first OI scrub accessing. */
 738                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 739                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
 740         }
 741
 742         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 743         if (rc == 0) {
 744                 has_lma = true;
 745                 if (lma->lma_compat & LMAC_NOT_IN_OI) {
 746                         ldiskfs_set_inode_state(inode,
 747                                                 LDISKFS_STATE_LUSTRE_NO_OI);
 748                         GOTO(put, rc = SCRUB_NEXT_CONTINUE);
 749                 }
 750
 751                 if (fid_is_llog(&lma->lma_self_fid))
 752                         GOTO(put, rc = SCRUB_NEXT_CONTINUE);
 753
 754                 *fid = lma->lma_self_fid;
 755                 if (fid_is_internal(&lma->lma_self_fid)) {
 756                         if (!scrub)
 757                                 rc = SCRUB_NEXT_CONTINUE;
 758                         GOTO(put, rc);
 759                 }
 760
 761                 if (!scrub)
 762                         GOTO(put, rc);
 763
 764                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
 765                         GOTO(put, rc);
 766
 767                 if (lma->lma_compat & LMAC_FID_ON_OST || fid_is_last_id(fid))
 768                         GOTO(put, rc = SCRUB_NEXT_OSTOBJ);
 769
 770                 if (fid_is_idif(fid))
 771                         GOTO(put, rc = SCRUB_NEXT_OSTOBJ_OLD);
 772
 773                 if (lma->lma_incompat & LMAI_AGENT)
 774                         GOTO(put, rc = SCRUB_NEXT_CONTINUE);
 775
 776                 /* Here, it may be MDT-object, or may be 2.4 OST-object.
 777                  * Fall through. */
 778         }
 779
 780         if (rc == -ENODATA || rc == 0) {
 781                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
 782                 if (rc == 0) {
 783                         if (scrub)
 784                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
 785                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 786                         GOTO(put, rc);
 787                 }
 788
 789                 if (rc > 0) {
 790                         if (!has_lma)
 791                                 /* It is FID-on-OST, but we do not know how
 792                                  * to generate its FID, ignore it directly. */
 793                                 rc = SCRUB_NEXT_CONTINUE;
 794                         else
 795                                 /* It is 2.4 OST-object. */
 796                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 797                         GOTO(put, rc);
 798                 }
 799
 800                 if (rc != -ENODATA)
 801                         GOTO(put, rc);
 802
 803                 if (!has_lma) {
 804                         if (dev->od_handle_nolma) {
 805                                 lu_igif_build(fid, inode->i_ino,
 806                                               inode->i_generation);
 807                                 if (scrub)
 808                                         rc = SCRUB_NEXT_NOLMA;
 809                                 else
 810                                         rc = 0;
 811                         } else {
 812                                 /* It may be FID-on-OST, or may be FID for
 813                                  * non-MDT0, anyway, we do not know how to
 814                                  * generate its FID, ignore it directly. */
 815                                 rc = SCRUB_NEXT_CONTINUE;
 816                         }
 817                         GOTO(put, rc);
 818                 }
 819
 820                 /* For OI scrub case only: the object has LMA but has no ff
 821                  * (or ff crashed). It may be MDT-object, may be OST-object
 822                  * with crashed ff. The last check is local FLDB. */
 823                 rc = osd_scrub_check_local_fldb(info, dev, fid);
 824         }
 825
 826         GOTO(put, rc);
 827
 828 put:
 829         iput(inode);
 830         return rc;
 831 }
 832
 833 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 834                           struct osd_iit_param *param,
 835                           struct osd_idmap_cache **oic, int noslot)
 836 {
 837         struct osd_scrub     *scrub  = &dev->od_scrub;
 838         struct ptlrpc_thread *thread = &scrub->os_thread;
 839         struct lu_fid        *fid;
 840         struct osd_inode_id  *lid;
 841         int                   rc;
 842
 843         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
 844                 struct l_wait_info lwi;
 845
 846                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
 847                 l_wait_event(thread->t_ctl_waitq,
 848                              !cfs_list_empty(&scrub->os_inconsistent_items) ||
 849                              !thread_is_running(thread),
 850                              &lwi);
 851         }
 852
 853         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 854                 spin_lock(&scrub->os_lock);
 855                 thread_set_flags(thread, SVC_STOPPING);
 856                 spin_unlock(&scrub->os_lock);
 857                 return SCRUB_NEXT_CRASH;
 858         }
 859
 860         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
 861                 return SCRUB_NEXT_FATAL;
 862
 863         if (unlikely(!thread_is_running(thread)))
 864                 return SCRUB_NEXT_EXIT;
 865
 866         if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
 867                 struct osd_inconsistent_item *oii;
 868
 869                 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
 870                                      struct osd_inconsistent_item, oii_list);
 871                 *oic = &oii->oii_cache;
 872                 scrub->os_in_prior = 1;
 873                 return 0;
 874         }
 875
 876         if (noslot != 0)
 877                 return SCRUB_NEXT_WAIT;
 878
 879         rc = osd_iit_next(param, &scrub->os_pos_current);
 880         if (rc != 0)
 881                 return rc;
 882
 883         *oic = &scrub->os_oic;
 884         fid = &(*oic)->oic_fid;
 885         lid = &(*oic)->oic_lid;
 886         rc = osd_iit_iget(info, dev, fid, lid,
 887                           scrub->os_pos_current, param->sb, true);
 888         return rc;
 889 }
 890
 891 static int osd_preload_next(struct osd_thread_info *info,
 892                             struct osd_device *dev, struct osd_iit_param *param,
 893                             struct osd_idmap_cache **oic, int noslot)
 894 {
 895         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
 896         struct osd_scrub        *scrub;
 897         struct ptlrpc_thread    *thread;
 898         int                      rc;
 899
 900         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
 901         if (rc != 0)
 902                 return rc;
 903
 904         scrub = &dev->od_scrub;
 905         thread = &scrub->os_thread;
 906         if (thread_is_running(thread) &&
 907             ooc->ooc_pos_preload >= scrub->os_pos_current)
 908                 return SCRUB_NEXT_EXIT;
 909
 910         rc = osd_iit_iget(info, dev,
 911                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
 912                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
 913                           ooc->ooc_pos_preload, param->sb, false);
 914         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
 915          * ignore the failure, so it still need to skip the inode next time. */
 916         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
 917         return rc;
 918 }
 919
 920 static inline int
 921 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
 922 {
 923         spin_lock(&scrub->os_lock);
 924         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
 925             !cfs_list_empty(&scrub->os_inconsistent_items) ||
 926             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
 927                 scrub->os_waiting = 0;
 928         else
 929                 scrub->os_waiting = 1;
 930         spin_unlock(&scrub->os_lock);
 931
 932         return !scrub->os_waiting;
 933 }
 934
 935 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
 936                           struct osd_iit_param *param,
 937                           struct osd_idmap_cache *oic, int *noslot, int rc)
 938 {
 939         struct l_wait_info       lwi    = { 0 };
 940         struct osd_scrub        *scrub  = &dev->od_scrub;
 941         struct scrub_file       *sf     = &scrub->os_file;
 942         struct ptlrpc_thread    *thread = &scrub->os_thread;
 943         struct osd_otable_it    *it     = dev->od_otable_it;
 944         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
 945
 946         switch (rc) {
 947         case SCRUB_NEXT_CONTINUE:
 948                 goto next;
 949         case SCRUB_NEXT_WAIT:
 950                 goto wait;
 951         case SCRUB_NEXT_NOSCRUB:
 952                 down_write(&scrub->os_rwsem);
 953                 scrub->os_new_checked++;
 954                 sf->sf_items_noscrub++;
 955                 up_write(&scrub->os_rwsem);
 956                 goto next;
 957         }
 958
 959         rc = osd_scrub_check_update(info, dev, oic, rc);
 960         if (rc != 0)
 961                 return rc;
 962
 963         rc = osd_scrub_checkpoint(scrub);
 964         if (rc != 0) {
 965                 CERROR("%.16s: fail to checkpoint, pos = %u, rc = %d\n",
 966                        LDISKFS_SB(param->sb)->s_es->s_volume_name,
 967                        scrub->os_pos_current, rc);
 968                 /* Continue, as long as the scrub itself can go ahead. */
 969         }
 970
 971         if (scrub->os_in_prior) {
 972                 scrub->os_in_prior = 0;
 973                 return 0;
 974         }
 975
 976 next:
 977         scrub->os_pos_current = param->gbase + ++(param->offset);
 978
 979 wait:
 980         if (it != NULL && it->ooi_waiting &&
 981             ooc->ooc_pos_preload < scrub->os_pos_current) {
 982                 spin_lock(&scrub->os_lock);
 983                 it->ooi_waiting = 0;
 984                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
 985                 spin_unlock(&scrub->os_lock);
 986         }
 987
 988         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
 989                 return 0;
 990
 991         if (osd_scrub_has_window(scrub, ooc)) {
 992                 *noslot = 0;
 993                 return 0;
 994         }
 995
 996         l_wait_event(thread->t_ctl_waitq,
 997                      osd_scrub_wakeup(scrub, it),
 998                      &lwi);
 999
1000         if (osd_scrub_has_window(scrub, ooc))
1001                 *noslot = 0;
1002         else
1003                 *noslot = 1;
1004         return 0;
1005 }
1006
1007 static int osd_preload_exec(struct osd_thread_info *info,
1008                             struct osd_device *dev, struct osd_iit_param *param,
1009                             struct osd_idmap_cache *oic, int *noslot, int rc)
1010 {
1011         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1012
1013         if (rc == 0) {
1014                 ooc->ooc_cached_items++;
1015                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1016                                         ~OSD_OTABLE_IT_CACHE_MASK;
1017         }
1018         return rc > 0 ? 0 : rc;
1019 }
1020
1021 #define SCRUB_IT_ALL    1
1022 #define SCRUB_IT_CRASH  2
1023
1024 static int osd_inode_iteration(struct osd_thread_info *info,
1025                                struct osd_device *dev, __u32 max, bool preload)
1026 {
1027         osd_iit_next_policy   next;
1028         osd_iit_exec_policy   exec;
1029         __u32                *pos;
1030         __u32                *count;
1031         struct osd_iit_param  param;
1032         __u32                 limit;
1033         int                   noslot = 0;
1034         int                   rc;
1035         ENTRY;
1036
1037         if (!preload) {
1038                 struct osd_scrub *scrub = &dev->od_scrub;
1039
1040                 next = osd_scrub_next;
1041                 exec = osd_scrub_exec;
1042                 pos = &scrub->os_pos_current;
1043                 count = &scrub->os_new_checked;
1044         } else {
1045                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1046
1047                 next = osd_preload_next;
1048                 exec = osd_preload_exec;
1049                 pos = &ooc->ooc_pos_preload;
1050                 count = &ooc->ooc_cached_items;
1051         }
1052         param.sb = osd_sb(dev);
1053         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1054
1055         while (*pos <= limit && *count < max) {
1056                 struct osd_idmap_cache *oic = NULL;
1057
1058                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1059                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1060                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1061                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1062                 if (param.bitmap == NULL) {
1063                         CERROR("%.16s: fail to read bitmap for %u, "
1064                                "scrub will stop, urgent mode\n",
1065                                LDISKFS_SB(param.sb)->s_es->s_volume_name,
1066                                (__u32)param.bg);
1067                         RETURN(-EIO);
1068                 }
1069
1070                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1071                        *count < max) {
1072                         rc = next(info, dev, &param, &oic, noslot);
1073                         switch (rc) {
1074                         case SCRUB_NEXT_BREAK:
1075                                 goto next_group;
1076                         case SCRUB_NEXT_EXIT:
1077                                 brelse(param.bitmap);
1078                                 RETURN(0);
1079                         case SCRUB_NEXT_CRASH:
1080                                 brelse(param.bitmap);
1081                                 RETURN(SCRUB_IT_CRASH);
1082                         case SCRUB_NEXT_FATAL:
1083                                 brelse(param.bitmap);
1084                                 RETURN(-EINVAL);
1085                         }
1086
1087                         rc = exec(info, dev, &param, oic, &noslot, rc);
1088                         if (rc != 0) {
1089                                 brelse(param.bitmap);
1090                                 RETURN(rc);
1091                         }
1092                 }
1093
1094 next_group:
1095                 brelse(param.bitmap);
1096         }
1097
1098         if (*pos > limit)
1099                 RETURN(SCRUB_IT_ALL);
1100         RETURN(0);
1101 }
1102
1103 static int osd_otable_it_preload(const struct lu_env *env,
1104                                  struct osd_otable_it *it)
1105 {
1106         struct osd_device       *dev   = it->ooi_dev;
1107         struct osd_scrub        *scrub = &dev->od_scrub;
1108         struct osd_otable_cache *ooc   = &it->ooi_cache;
1109         int                      rc;
1110         ENTRY;
1111
1112         rc = osd_inode_iteration(osd_oti_get(env), dev,
1113                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1114         if (rc == SCRUB_IT_ALL)
1115                 it->ooi_all_cached = 1;
1116
1117         CDEBUG(D_LFSCK, "OSD pre-loaded: max = %u, preload = %u, rc = %d\n",
1118                le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count),
1119                ooc->ooc_pos_preload, rc);
1120
1121         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1122                 scrub->os_waiting = 0;
1123                 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
1124         }
1125
1126         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1127 }
1128
1129 static int osd_scrub_main(void *args)
1130 {
1131         struct lu_env         env;
1132         struct osd_device    *dev    = (struct osd_device *)args;
1133         struct osd_scrub     *scrub  = &dev->od_scrub;
1134         struct ptlrpc_thread *thread = &scrub->os_thread;
1135         struct super_block   *sb     = osd_sb(dev);
1136         int                   rc;
1137         ENTRY;
1138
1139         rc = lu_env_init(&env, LCT_LOCAL);
1140         if (rc != 0) {
1141                 CERROR("%.16s: OI scrub, fail to init env, rc = %d\n",
1142                        LDISKFS_SB(sb)->s_es->s_volume_name, rc);
1143                 GOTO(noenv, rc);
1144         }
1145
1146         rc = osd_scrub_prep(dev);
1147         if (rc != 0) {
1148                 CERROR("%.16s: OI scrub, fail to scrub prep, rc = %d\n",
1149                        LDISKFS_SB(sb)->s_es->s_volume_name, rc);
1150                 GOTO(out, rc);
1151         }
1152
1153         if (!scrub->os_full_speed) {
1154                 struct l_wait_info lwi = { 0 };
1155                 struct osd_otable_it *it = dev->od_otable_it;
1156                 struct osd_otable_cache *ooc = &it->ooi_cache;
1157
1158                 l_wait_event(thread->t_ctl_waitq,
1159                              it->ooi_user_ready || !thread_is_running(thread),
1160                              &lwi);
1161                 if (unlikely(!thread_is_running(thread)))
1162                         GOTO(post, rc = 0);
1163
1164                 scrub->os_pos_current = ooc->ooc_pos_preload;
1165         }
1166
1167         CDEBUG(D_LFSCK, "OI scrub: flags = 0x%x, pos = %u\n",
1168                scrub->os_start_flags, scrub->os_pos_current);
1169
1170         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1171         if (unlikely(rc == SCRUB_IT_CRASH))
1172                 GOTO(out, rc = -EINVAL);
1173         GOTO(post, rc);
1174
1175 post:
1176         osd_scrub_post(scrub, rc);
1177         CDEBUG(D_LFSCK, "OI scrub: stop, rc = %d, pos = %u\n",
1178                rc, scrub->os_pos_current);
1179
1180 out:
1181         while (!cfs_list_empty(&scrub->os_inconsistent_items)) {
1182                 struct osd_inconsistent_item *oii;
1183
1184                 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
1185                                      struct osd_inconsistent_item, oii_list);
1186                 cfs_list_del_init(&oii->oii_list);
1187                 OBD_FREE_PTR(oii);
1188         }
1189         lu_env_fini(&env);
1190
1191 noenv:
1192         spin_lock(&scrub->os_lock);
1193         thread_set_flags(thread, SVC_STOPPED);
1194         cfs_waitq_broadcast(&thread->t_ctl_waitq);
1195         spin_unlock(&scrub->os_lock);
1196         return rc;
1197 }
1198
1199 /* initial OI scrub */
1200
1201 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1202                          struct dentry *, filldir_t filldir);
1203
1204 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1205                                loff_t offset, __u64 ino, unsigned d_type);
1206
1207 static int
1208 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1209                      struct dentry *dentry, filldir_t filldir);
1210 static int
1211 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1212                   struct dentry *dentry, filldir_t filldir);
1213
1214 static int
1215 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1216                      struct dentry *dentry, filldir_t filldir);
1217
1218 enum osd_lf_flags {
1219         OLF_SCAN_SUBITEMS       = 0x0001,
1220         OLF_HIDE_FID            = 0x0002,
1221         OLF_SHOW_NAME           = 0x0004,
1222 };
1223
1224 struct osd_lf_map {
1225         char            *olm_name;
1226         struct lu_fid    olm_fid;
1227         __u16            olm_flags;
1228         scandir_t        olm_scandir;
1229         filldir_t        olm_filldir;
1230 };
1231
1232 /* Add the new introduced local files in the list in the future. */
1233 static const struct osd_lf_map osd_lf_maps[] = {
1234         /* CATALOGS */
1235         { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1236                 NULL, NULL },
1237
1238         /* CONFIGS */
1239         { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1240                 OLF_SCAN_SUBITEMS, osd_ios_general_scan,
1241                 osd_ios_varfid_fill },
1242
1243         /* NIDTBL_VERSIONS */
1244         { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1245                 osd_ios_general_scan, osd_ios_varfid_fill },
1246
1247         /* PENDING */
1248         { "PENDING", { 0, 0, 0 }, 0, NULL, NULL },
1249
1250         /* ROOT */
1251         { "ROOT", { FID_SEQ_ROOT, 1, 0 },
1252                 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, osd_ios_ROOT_scan, NULL },
1253
1254         /* changelog_catalog */
1255         { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, NULL, NULL },
1256
1257         /* changelog_users */
1258         { CHANGELOG_USERS, { 0, 0, 0 }, 0, NULL, NULL },
1259
1260         /* fld */
1261         { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1262                 NULL, NULL },
1263
1264         /* last_rcvd */
1265         { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1266                 NULL, NULL },
1267
1268         /* lfsck_bookmark */
1269         { "lfsck_bookmark", { 0, 0, 0 }, 0, NULL, NULL },
1270
1271         /* lov_objid */
1272         { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1273                 NULL, NULL },
1274
1275         /* lov_objseq */
1276         { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1277                 OLF_SHOW_NAME, NULL, NULL },
1278
1279         /* quota_master */
1280         { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1281                 osd_ios_general_scan, osd_ios_varfid_fill },
1282
1283         /* quota_slave */
1284         { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1285                 osd_ios_general_scan, osd_ios_varfid_fill },
1286
1287         /* seq_ctl */
1288         { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1289                 OLF_SHOW_NAME, NULL, NULL },
1290
1291         /* seq_srv */
1292         { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1293                 OLF_SHOW_NAME, NULL, NULL },
1294
1295         /* health_check */
1296         { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1297                 OLF_SHOW_NAME, NULL, NULL },
1298
1299         /* lfsck_namespace */
1300         { "lfsck_namespace", { 0, 0, 0 }, 0, NULL, NULL },
1301
1302         /* OBJECTS, upgrade from old device */
1303         { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, osd_ios_OBJECTS_scan, NULL },
1304
1305         /* lquota_v2.user, upgrade from old device */
1306         { "lquota_v2.user", { 0, 0, 0 }, 0, NULL, NULL },
1307
1308         /* lquota_v2.group, upgrade from old device */
1309         { "lquota_v2.group", { 0, 0, 0 }, 0, NULL, NULL },
1310
1311         /* LAST_GROUP, upgrade from old device */
1312         { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1313                 OLF_SHOW_NAME, NULL, NULL },
1314
1315         { NULL, { 0, 0, 0 }, 0, NULL, NULL }
1316 };
1317
1318 struct osd_ios_item {
1319         cfs_list_t       oii_list;
1320         struct dentry   *oii_dentry;
1321         scandir_t        oii_scandir;
1322         filldir_t        oii_filldir;
1323 };
1324
1325 struct osd_ios_filldir_buf {
1326         struct osd_thread_info  *oifb_info;
1327         struct osd_device       *oifb_dev;
1328         struct dentry           *oifb_dentry;
1329 };
1330
1331 static inline struct dentry *
1332 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1333 {
1334         struct dentry *dentry;
1335
1336         dentry = ll_lookup_one_len(name, parent, namelen);
1337         if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1338                 dput(dentry);
1339                 return ERR_PTR(-ENOENT);
1340         }
1341
1342         return dentry;
1343 }
1344
1345 static inline void
1346 osd_ios_llogname2fid(struct lu_fid *fid, const char *name, int namelen)
1347 {
1348         obd_id id = 0;
1349         int    i  = 0;
1350
1351         fid->f_seq = FID_SEQ_LLOG;
1352         while (i < namelen)
1353                 id = id * 10 + name[i++] - '0';
1354
1355         fid->f_oid = id & 0x00000000ffffffffULL;
1356         fid->f_ver = id >> 32;
1357 }
1358
1359 static inline void
1360 osd_ios_Oname2fid(struct lu_fid *fid, const char *name, int namelen)
1361 {
1362         __u64 seq = 0;
1363         int   i   = 0;
1364
1365         while (i < namelen)
1366                 seq = seq * 10 + name[i++] - '0';
1367
1368         lu_last_id_fid(fid, seq);
1369 }
1370
1371 static int
1372 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1373                  scandir_t scandir, filldir_t filldir)
1374 {
1375         struct osd_ios_item *item;
1376
1377         OBD_ALLOC_PTR(item);
1378         if (item == NULL)
1379                 return -ENOMEM;
1380
1381         CFS_INIT_LIST_HEAD(&item->oii_list);
1382         item->oii_dentry = dget(dentry);
1383         item->oii_scandir = scandir;
1384         item->oii_filldir = filldir;
1385         cfs_list_add_tail(&item->oii_list, &dev->od_ios_list);
1386         return 0;
1387 }
1388
1389 /**
1390  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1391  *
1392  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1393  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1394  * reference the inode, or fixed if it is missing or references another inode.
1395  */
1396 static int
1397 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1398                  struct inode *inode, const struct lu_fid *fid, int flags)
1399 {
1400         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
1401         struct osd_inode_id     *id     = &info->oti_id;
1402         struct osd_inode_id     *id2    = &info->oti_id2;
1403         struct osd_scrub        *scrub  = &dev->od_scrub;
1404         struct scrub_file       *sf     = &scrub->os_file;
1405         struct lu_fid            tfid;
1406         int                      rc;
1407         ENTRY;
1408
1409         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1410         if (rc != 0 && rc != -ENODATA)
1411                 RETURN(rc);
1412
1413         osd_id_gen(id, inode->i_ino, inode->i_generation);
1414         if (rc == -ENODATA) {
1415                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID)
1416                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1417                 else
1418                         tfid = *fid;
1419                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1420                 if (rc != 0)
1421                         RETURN(rc);
1422         } else {
1423                 if (lma->lma_compat & LMAC_NOT_IN_OI)
1424                         RETURN(0);
1425
1426                 tfid = lma->lma_self_fid;
1427         }
1428
1429         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1430         if (rc != 0) {
1431                 if (rc != -ENOENT)
1432                         RETURN(rc);
1433
1434                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1435                                                DTO_INDEX_INSERT, 0);
1436                 if (rc > 0)
1437                         rc = 0;
1438
1439                 RETURN(rc);
1440         }
1441
1442         if (osd_id_eq_strict(id, id2))
1443                 RETURN(0);
1444
1445         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1446                 osd_scrub_file_reset(scrub,
1447                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1448                                      SF_INCONSISTENT);
1449                 rc = osd_scrub_file_store(scrub);
1450                 if (rc != 0)
1451                         RETURN(rc);
1452         }
1453
1454         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1455                                        DTO_INDEX_UPDATE, 0);
1456         if (rc > 0)
1457                 rc = 0;
1458
1459         RETURN(rc);
1460 }
1461
1462 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1463                                loff_t offset, __u64 ino, unsigned d_type)
1464 {
1465         struct osd_ios_filldir_buf *fill_buf = buf;
1466         struct osd_device          *dev      = fill_buf->oifb_dev;
1467         struct dentry              *child;
1468         int                         rc;
1469         ENTRY;
1470
1471         /* skip any '.' started names */
1472         if (name[0] == '.')
1473                 RETURN(0);
1474
1475         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1476         if (IS_ERR(child))
1477                 RETURN(PTR_ERR(child));
1478
1479         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1480                               NULL, 0);
1481         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1482                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1483                                       osd_ios_varfid_fill);
1484         dput(child);
1485
1486         RETURN(rc);
1487 }
1488
1489 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
1490                              loff_t offset, __u64 ino, unsigned d_type)
1491 {
1492         struct osd_ios_filldir_buf *fill_buf = buf;
1493         struct osd_device          *dev      = fill_buf->oifb_dev;
1494         const struct osd_lf_map    *map;
1495         struct dentry              *child;
1496         int                         rc       = 0;
1497         ENTRY;
1498
1499         /* skip any '.' started names */
1500         if (name[0] == '.')
1501                 RETURN(0);
1502
1503         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
1504                 if (strlen(map->olm_name) != namelen)
1505                         continue;
1506
1507                 if (strncmp(map->olm_name, name, namelen) == 0)
1508                         break;
1509         }
1510
1511         if (map->olm_name == NULL)
1512                 RETURN(0);
1513
1514         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1515         if (IS_ERR(child))
1516                 RETURN(PTR_ERR(child));
1517
1518         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1519                               &map->olm_fid, map->olm_flags);
1520         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
1521                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
1522                                       map->olm_filldir);
1523         dput(child);
1524
1525         RETURN(rc);
1526 }
1527
1528 static int
1529 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1530                      struct dentry *dentry, filldir_t filldir)
1531 {
1532         struct osd_ios_filldir_buf    buf   = { info, dev, dentry };
1533         struct file                  *filp  = &info->oti_it_ea.oie_file;
1534         struct inode                 *inode = dentry->d_inode;
1535         const struct file_operations *fops  = inode->i_fop;
1536         int                           rc;
1537         ENTRY;
1538
1539         LASSERT(filldir != NULL);
1540
1541         filp->f_pos = 0;
1542         filp->f_dentry = dentry;
1543         filp->f_mode = FMODE_64BITHASH;
1544         filp->f_mapping = inode->i_mapping;
1545         filp->f_op = fops;
1546         filp->private_data = NULL;
1547
1548         rc = fops->readdir(filp, &buf, filldir);
1549         fops->release(inode, filp);
1550
1551         RETURN(rc);
1552 }
1553
1554 static int
1555 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1556                   struct dentry *dentry, filldir_t filldir)
1557 {
1558         struct osd_scrub  *scrub  = &dev->od_scrub;
1559         struct scrub_file *sf     = &scrub->os_file;
1560         struct dentry     *child;
1561         int                rc;
1562         ENTRY;
1563
1564         /* It is existing MDT0 device. We only allow the case of object without
1565          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
1566          * can generate IGIF mode FID for the object and related OI mapping. If
1567          * it is on other MDTs, then becuase file-level backup/restore, related
1568          * OI mapping may be invalid already, we do not know which is the right
1569          * FID for the object. We only allow IGIF objects to reside on the MDT0.
1570          *
1571          * XXX: For the case of object on non-MDT0 device with neither LMA nor
1572          *      "fid" xattr, then something crashed. We cannot re-generate the
1573          *      FID directly, instead, the OI scrub will scan the OI structure
1574          *      and try to re-generate the LMA from the OI mapping. But if the
1575          *      OI mapping crashed or lost also, then we have to give up under
1576          *      double failure cases. */
1577         dev->od_handle_nolma = 1;
1578         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
1579                                        strlen(dot_lustre_name));
1580         if (IS_ERR(child)) {
1581                 rc = PTR_ERR(child);
1582                 if (rc == -ENOENT) {
1583                         /* It is 1.8 MDT device. */
1584                         if (!(sf->sf_flags & SF_UPGRADE)) {
1585                                 osd_scrub_file_reset(scrub,
1586                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1587                                         SF_UPGRADE);
1588                                 rc = osd_scrub_file_store(scrub);
1589                         } else {
1590                                 rc = 0;
1591                         }
1592                 }
1593         } else {
1594                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
1595                  * so the client will get IGIF for the ".lustre" object when
1596                  * the MDT restart.
1597                  *
1598                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
1599                  * it does not know whether there are some old clients cached
1600                  * the ".lustre" IGIF during the upgrading. Two choices:
1601                  *
1602                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
1603                  *    It will allow the old connected clients to access the
1604                  *    ".lustre" with cached IGIF. But it will cause others
1605                  *    on the MDT failed to check "fid_is_dot_lustre()".
1606                  *
1607                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
1608                  *    for ".lustre" in spite of whether there are some clients
1609                  *    cached the ".lustre" IGIF or not. It enables the check
1610                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
1611                  *    that the old connected clients cannot access the ".lustre"
1612                  *    with the cached IGIF.
1613                  *
1614                  * Usually, it is rare case for the old connected clients
1615                  * to access the ".lustre" with cached IGIF. So we prefer
1616                  * to the solution 2). */
1617                 rc = osd_ios_scan_one(info, dev, child->d_inode,
1618                                       &LU_DOT_LUSTRE_FID, 0);
1619                 dput(child);
1620         }
1621
1622         RETURN(rc);
1623 }
1624
1625 static int
1626 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1627                      struct dentry *dentry, filldir_t filldir)
1628 {
1629         struct dentry *child;
1630         int            rc;
1631         ENTRY;
1632
1633         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
1634         if (!IS_ERR(child)) {
1635                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1636                 dput(child);
1637         } else {
1638                 rc = PTR_ERR(child);
1639         }
1640
1641         if (rc != 0 && rc != -ENOENT)
1642                 RETURN(rc);
1643
1644         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
1645         if (!IS_ERR(child)) {
1646                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1647                 dput(child);
1648         } else {
1649                 rc = PTR_ERR(child);
1650         }
1651
1652         if (rc == -ENOENT)
1653                 rc = 0;
1654
1655         RETURN(rc);
1656 }
1657
1658 static int osd_initial_OI_scrub(struct osd_thread_info *info,
1659                                 struct osd_device *dev)
1660 {
1661         struct osd_ios_item     *item    = NULL;
1662         scandir_t                scandir = osd_ios_general_scan;
1663         filldir_t                filldir = osd_ios_root_fill;
1664         struct dentry           *dentry  = osd_sb(dev)->s_root;
1665         const struct osd_lf_map *map     = osd_lf_maps;
1666         int                      rc;
1667         ENTRY;
1668
1669         while (1) {
1670                 rc = scandir(info, dev, dentry, filldir);
1671                 if (item != NULL) {
1672                         dput(item->oii_dentry);
1673                         OBD_FREE_PTR(item);
1674                 }
1675
1676                 if (rc != 0)
1677                         break;
1678
1679                 if (cfs_list_empty(&dev->od_ios_list))
1680                         break;
1681
1682                 item = cfs_list_entry(dev->od_ios_list.next,
1683                                       struct osd_ios_item, oii_list);
1684                 cfs_list_del_init(&item->oii_list);
1685
1686                 LASSERT(item->oii_scandir != NULL);
1687                 scandir = item->oii_scandir;
1688                 filldir = item->oii_filldir;
1689                 dentry = item->oii_dentry;
1690         }
1691
1692         while (!cfs_list_empty(&dev->od_ios_list)) {
1693                 item = cfs_list_entry(dev->od_ios_list.next,
1694                                       struct osd_ios_item, oii_list);
1695                 cfs_list_del_init(&item->oii_list);
1696                 dput(item->oii_dentry);
1697                 OBD_FREE_PTR(item);
1698         }
1699
1700         if (rc != 0)
1701                 RETURN(rc);
1702
1703         /* There maybe the case that the object has been removed, but its OI
1704          * mapping is still in the OI file, such as the "CATALOGS" after MDT
1705          * file-level backup/restore. So here cleanup the stale OI mappings. */
1706         while (map->olm_name != NULL) {
1707                 struct dentry *child;
1708
1709                 if (fid_is_zero(&map->olm_fid)) {
1710                         map++;
1711                         continue;
1712                 }
1713
1714                 child = osd_ios_lookup_one_len(map->olm_name,
1715                                                osd_sb(dev)->s_root,
1716                                                strlen(map->olm_name));
1717                 if (!IS_ERR(child))
1718                         dput(child);
1719                 else if (PTR_ERR(child) == -ENOENT)
1720                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
1721                                                   NULL, DTO_INDEX_DELETE, 0);
1722                 map++;
1723         }
1724
1725         RETURN(0);
1726 }
1727
1728 char *osd_lf_fid2name(const struct lu_fid *fid)
1729 {
1730         const struct osd_lf_map *map = osd_lf_maps;
1731
1732         while (map->olm_name != NULL) {
1733                 if (!lu_fid_eq(fid, &map->olm_fid)) {
1734                         map++;
1735                         continue;
1736                 }
1737
1738                 if (map->olm_flags & OLF_SHOW_NAME)
1739                         return map->olm_name;
1740                 else
1741                         return "";
1742         }
1743
1744         return NULL;
1745 }
1746
1747 /* OI scrub start/stop */
1748
1749 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
1750 {
1751         struct osd_scrub     *scrub  = &dev->od_scrub;
1752         struct ptlrpc_thread *thread = &scrub->os_thread;
1753         struct l_wait_info    lwi    = { 0 };
1754         int                   rc;
1755         ENTRY;
1756
1757 again:
1758         /* os_lock: sync status between stop and scrub thread */
1759         spin_lock(&scrub->os_lock);
1760         if (thread_is_running(thread)) {
1761                 spin_unlock(&scrub->os_lock);
1762                 RETURN(-EALREADY);
1763         } else if (unlikely(thread_is_stopping(thread))) {
1764                 spin_unlock(&scrub->os_lock);
1765                 l_wait_event(thread->t_ctl_waitq,
1766                              thread_is_stopped(thread),
1767                              &lwi);
1768                 goto again;
1769         }
1770         spin_unlock(&scrub->os_lock);
1771
1772         if (scrub->os_file.sf_status == SS_COMPLETED)
1773                 flags |= SS_RESET;
1774
1775         scrub->os_start_flags = flags;
1776         thread_set_flags(thread, 0);
1777         rc = PTR_ERR(kthread_run(osd_scrub_main, dev, "OI_scrub"));
1778         if (IS_ERR_VALUE(rc)) {
1779                 CERROR("%.16s: cannot start iteration thread, rc = %d\n",
1780                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
1781                 RETURN(rc);
1782         }
1783
1784         l_wait_event(thread->t_ctl_waitq,
1785                      thread_is_running(thread) || thread_is_stopped(thread),
1786                      &lwi);
1787
1788         RETURN(0);
1789 }
1790
1791 int osd_scrub_start(struct osd_device *dev)
1792 {
1793         int rc;
1794         ENTRY;
1795
1796         /* od_otable_mutex: prevent curcurrent start/stop */
1797         mutex_lock(&dev->od_otable_mutex);
1798         rc = do_osd_scrub_start(dev, SS_AUTO);
1799         mutex_unlock(&dev->od_otable_mutex);
1800
1801         RETURN(rc == -EALREADY ? 0 : rc);
1802 }
1803
1804 static void do_osd_scrub_stop(struct osd_scrub *scrub)
1805 {
1806         struct ptlrpc_thread *thread = &scrub->os_thread;
1807         struct l_wait_info    lwi    = { 0 };
1808
1809         /* os_lock: sync status between stop and scrub thread */
1810         spin_lock(&scrub->os_lock);
1811         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
1812                 thread_set_flags(thread, SVC_STOPPING);
1813                 spin_unlock(&scrub->os_lock);
1814                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
1815                 l_wait_event(thread->t_ctl_waitq,
1816                              thread_is_stopped(thread),
1817                              &lwi);
1818                 /* Do not skip the last lock/unlock, which can guarantee that
1819                  * the caller cannot return until the OI scrub thread exit. */
1820                 spin_lock(&scrub->os_lock);
1821         }
1822         spin_unlock(&scrub->os_lock);
1823 }
1824
1825 static void osd_scrub_stop(struct osd_device *dev)
1826 {
1827         /* od_otable_mutex: prevent curcurrent start/stop */
1828         mutex_lock(&dev->od_otable_mutex);
1829         dev->od_scrub.os_paused = 1;
1830         do_osd_scrub_stop(&dev->od_scrub);
1831         mutex_unlock(&dev->od_otable_mutex);
1832 }
1833
1834 /* OI scrub setup/cleanup */
1835
1836 static const char osd_scrub_name[] = "OI_scrub";
1837
1838 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
1839 {
1840         struct osd_thread_info     *info   = osd_oti_get(env);
1841         struct osd_scrub           *scrub  = &dev->od_scrub;
1842         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
1843         struct scrub_file          *sf     = &scrub->os_file;
1844         struct super_block         *sb     = osd_sb(dev);
1845         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
1846         struct lvfs_run_ctxt        saved;
1847         struct file                *filp;
1848         struct inode               *inode;
1849         struct lu_fid              *fid    = &info->oti_fid;
1850         int                         dirty  = 0;
1851         int                         rc     = 0;
1852         ENTRY;
1853
1854         memset(scrub, 0, sizeof(*scrub));
1855         OBD_SET_CTXT_MAGIC(ctxt);
1856         ctxt->pwdmnt = dev->od_mnt;
1857         ctxt->pwd = dev->od_mnt->mnt_root;
1858         ctxt->fs = get_ds();
1859
1860         cfs_waitq_init(&scrub->os_thread.t_ctl_waitq);
1861         init_rwsem(&scrub->os_rwsem);
1862         spin_lock_init(&scrub->os_lock);
1863         CFS_INIT_LIST_HEAD(&scrub->os_inconsistent_items);
1864
1865         push_ctxt(&saved, ctxt, NULL);
1866         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
1867         if (IS_ERR(filp)) {
1868                 pop_ctxt(&saved, ctxt, NULL);
1869                 RETURN(PTR_ERR(filp));
1870         }
1871
1872         inode = filp->f_dentry->d_inode;
1873         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
1874         /* 'What the @fid is' is not imporatant, because the object
1875          * has no OI mapping, and only is visible inside the OSD.*/
1876         lu_igif_build(fid, inode->i_ino, inode->i_generation);
1877         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
1878         if (rc != 0) {
1879                 filp_close(filp, 0);
1880                 pop_ctxt(&saved, ctxt, NULL);
1881                 RETURN(rc);
1882         }
1883
1884         scrub->os_inode = igrab(inode);
1885         filp_close(filp, 0);
1886         pop_ctxt(&saved, ctxt, NULL);
1887
1888         rc = osd_scrub_file_load(scrub);
1889         if (rc == -ENOENT) {
1890                 osd_scrub_file_init(scrub, es->s_uuid);
1891                 dirty = 1;
1892         } else if (rc != 0) {
1893                 RETURN(rc);
1894         } else {
1895                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
1896                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
1897                         dirty = 1;
1898                 } else if (sf->sf_status == SS_SCANNING) {
1899                         sf->sf_status = SS_CRASHED;
1900                         dirty = 1;
1901                 }
1902         }
1903
1904         if (sf->sf_pos_last_checkpoint != 0)
1905                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
1906         else
1907                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
1908
1909         if (dirty != 0) {
1910                 rc = osd_scrub_file_store(scrub);
1911                 if (rc != 0)
1912                         RETURN(rc);
1913         }
1914
1915         /* Initialize OI files. */
1916         rc = osd_oi_init(info, dev);
1917         if (rc < 0)
1918                 RETURN(rc);
1919
1920         rc = osd_initial_OI_scrub(info, dev);
1921         if (rc == 0) {
1922                 if ((sf->sf_flags & SF_UPGRADE) &&
1923                    !(sf->sf_flags & SF_INCONSISTENT))
1924                         /* The 'od_igif_inoi' will be set after the
1925                          * upgrading completed, needs NOT remount. */
1926                         dev->od_igif_inoi = 0;
1927                 else
1928                         /* The 'od_igif_inoi' will be set under the
1929                          * following cases:
1930                          * 1) new created system, or
1931                          * 2) restored from file-level backup, or
1932                          * 3) the upgrading completed.
1933                          *
1934                          * The 'od_igif_inoi' may be cleared by OI scrub
1935                          * later if found that the system is upgrading. */
1936                         dev->od_igif_inoi = 1;
1937
1938                 if (!dev->od_noscrub &&
1939                     ((sf->sf_status == SS_PAUSED) ||
1940                      (sf->sf_status == SS_CRASHED &&
1941                       sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
1942                                       SF_UPGRADE | SF_AUTO)) ||
1943                      (sf->sf_status == SS_INIT &&
1944                       sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
1945                                       SF_UPGRADE))))
1946                         rc = osd_scrub_start(dev);
1947         }
1948
1949         /* it is possible that dcache entries may keep objects after they are
1950          * deleted by OSD. While it looks safe this can cause object data to
1951          * stay until umount causing failures in tests calculating free space,
1952          * e.g. replay-ost-single. Since those dcache entries are not used
1953          * anymore let's just free them after use here */
1954         shrink_dcache_sb(sb);
1955
1956         RETURN(rc);
1957 }
1958
1959 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
1960 {
1961         struct osd_scrub *scrub = &dev->od_scrub;
1962
1963         LASSERT(dev->od_otable_it == NULL);
1964
1965         if (scrub->os_inode != NULL) {
1966                 osd_scrub_stop(dev);
1967                 iput(scrub->os_inode);
1968                 scrub->os_inode = NULL;
1969         }
1970         if (dev->od_oi_table != NULL)
1971                 osd_oi_fini(osd_oti_get(env), dev);
1972 }
1973
1974 /* object table based iteration APIs */
1975
1976 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
1977                                        struct dt_object *dt, __u32 attr,
1978                                        struct lustre_capa *capa)
1979 {
1980         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
1981         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
1982         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
1983         struct osd_scrub       *scrub = &dev->od_scrub;
1984         struct osd_otable_it   *it;
1985         __u32                   start = 0;
1986         int                     rc;
1987         ENTRY;
1988
1989         /* od_otable_mutex: prevent curcurrent init/fini */
1990         mutex_lock(&dev->od_otable_mutex);
1991         if (dev->od_otable_it != NULL)
1992                 GOTO(out, it = ERR_PTR(-EALREADY));
1993
1994         OBD_ALLOC_PTR(it);
1995         if (it == NULL)
1996                 GOTO(out, it = ERR_PTR(-ENOMEM));
1997
1998         dev->od_otable_it = it;
1999         it->ooi_dev = dev;
2000         it->ooi_cache.ooc_consumer_idx = -1;
2001         if (flags & DOIF_OUTUSED)
2002                 it->ooi_used_outside = 1;
2003
2004         if (flags & DOIF_RESET)
2005                 start |= SS_RESET;
2006
2007         if (valid & DOIV_ERROR_HANDLE) {
2008                 if (flags & DOIF_FAILOUT)
2009                         start |= SS_SET_FAILOUT;
2010                 else
2011                         start |= SS_CLEAR_FAILOUT;
2012         }
2013
2014         rc = do_osd_scrub_start(dev, start);
2015         if (rc < 0 && rc != -EALREADY) {
2016                 dev->od_otable_it = NULL;
2017                 OBD_FREE_PTR(it);
2018                 GOTO(out, it = ERR_PTR(rc));
2019         }
2020
2021         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2022
2023         GOTO(out, it);
2024
2025 out:
2026         mutex_unlock(&dev->od_otable_mutex);
2027         return (struct dt_it *)it;
2028 }
2029
2030 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2031 {
2032         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2033         struct osd_device    *dev = it->ooi_dev;
2034
2035         /* od_otable_mutex: prevent curcurrent init/fini */
2036         mutex_lock(&dev->od_otable_mutex);
2037         do_osd_scrub_stop(&dev->od_scrub);
2038         LASSERT(dev->od_otable_it == it);
2039
2040         dev->od_otable_it = NULL;
2041         mutex_unlock(&dev->od_otable_mutex);
2042         OBD_FREE_PTR(it);
2043 }
2044
2045 static int osd_otable_it_get(const struct lu_env *env,
2046                              struct dt_it *di, const struct dt_key *key)
2047 {
2048         return 0;
2049 }
2050
2051 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2052 {
2053 }
2054
2055 static inline int
2056 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2057 {
2058         spin_lock(&scrub->os_lock);
2059         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2060             scrub->os_waiting ||
2061             !thread_is_running(&scrub->os_thread))
2062                 it->ooi_waiting = 0;
2063         else
2064                 it->ooi_waiting = 1;
2065         spin_unlock(&scrub->os_lock);
2066
2067         return !it->ooi_waiting;
2068 }
2069
2070 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2071 {
2072         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2073         struct osd_device       *dev    = it->ooi_dev;
2074         struct osd_scrub        *scrub  = &dev->od_scrub;
2075         struct osd_otable_cache *ooc    = &it->ooi_cache;
2076         struct ptlrpc_thread    *thread = &scrub->os_thread;
2077         struct l_wait_info       lwi    = { 0 };
2078         int                      rc;
2079         ENTRY;
2080
2081         LASSERT(it->ooi_user_ready);
2082
2083 again:
2084         if (!thread_is_running(thread) && !it->ooi_used_outside)
2085                 RETURN(1);
2086
2087         if (ooc->ooc_cached_items > 0) {
2088                 ooc->ooc_cached_items--;
2089                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2090                                         ~OSD_OTABLE_IT_CACHE_MASK;
2091                 RETURN(0);
2092         }
2093
2094         if (it->ooi_all_cached) {
2095                 l_wait_event(thread->t_ctl_waitq,
2096                              !thread_is_running(thread),
2097                              &lwi);
2098                 RETURN(1);
2099         }
2100
2101         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2102                 spin_lock(&scrub->os_lock);
2103                 scrub->os_waiting = 0;
2104                 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
2105                 spin_unlock(&scrub->os_lock);
2106         }
2107
2108         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2109                 l_wait_event(thread->t_ctl_waitq,
2110                              osd_otable_it_wakeup(scrub, it),
2111                              &lwi);
2112
2113         if (!thread_is_running(thread) && !it->ooi_used_outside)
2114                 RETURN(1);
2115
2116         rc = osd_otable_it_preload(env, it);
2117         if (rc >= 0)
2118                 goto again;
2119
2120         RETURN(rc);
2121 }
2122
2123 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2124                                         const struct dt_it *di)
2125 {
2126         return NULL;
2127 }
2128
2129 static int osd_otable_it_key_size(const struct lu_env *env,
2130                                   const struct dt_it *di)
2131 {
2132         return sizeof(__u64);
2133 }
2134
2135 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2136                              struct dt_rec *rec, __u32 attr)
2137 {
2138         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2139         struct osd_otable_cache *ooc = &it->ooi_cache;
2140
2141         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2142
2143         /* Filter out Invald FID already. */
2144         LASSERTF(fid_is_sane((struct lu_fid *)rec),
2145                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2146                  PFID((struct lu_fid *)rec),
2147                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2148
2149         return 0;
2150 }
2151
2152 static __u64 osd_otable_it_store(const struct lu_env *env,
2153                                  const struct dt_it *di)
2154 {
2155         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2156         struct osd_otable_cache *ooc = &it->ooi_cache;
2157         __u64                    hash;
2158
2159         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2160                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2161         else
2162                 hash = ooc->ooc_pos_preload;
2163         return hash;
2164 }
2165
2166 /**
2167  * Set the OSD layer iteration start position as the specified hash.
2168  */
2169 static int osd_otable_it_load(const struct lu_env *env,
2170                               const struct dt_it *di, __u64 hash)
2171 {
2172         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
2173         struct osd_device       *dev   = it->ooi_dev;
2174         struct osd_otable_cache *ooc   = &it->ooi_cache;
2175         struct osd_scrub        *scrub = &dev->od_scrub;
2176         int                      rc;
2177         ENTRY;
2178
2179         /* Forbid to set iteration position after iteration started. */
2180         if (it->ooi_user_ready)
2181                 RETURN(-EPERM);
2182
2183         if (hash > OSD_OTABLE_MAX_HASH)
2184                 hash = OSD_OTABLE_MAX_HASH;
2185
2186         ooc->ooc_pos_preload = hash;
2187         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2188                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2189
2190         it->ooi_user_ready = 1;
2191         if (!scrub->os_full_speed)
2192                 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
2193
2194         /* Unplug OSD layer iteration by the first next() call. */
2195         rc = osd_otable_it_next(env, (struct dt_it *)it);
2196
2197         RETURN(rc);
2198 }
2199
2200 static int osd_otable_it_key_rec(const struct lu_env *env,
2201                                  const struct dt_it *di, void *key_rec)
2202 {
2203         return 0;
2204 }
2205
2206 const struct dt_index_operations osd_otable_ops = {
2207         .dio_it = {
2208                 .init     = osd_otable_it_init,
2209                 .fini     = osd_otable_it_fini,
2210                 .get      = osd_otable_it_get,
2211                 .put      = osd_otable_it_put,
2212                 .next     = osd_otable_it_next,
2213                 .key      = osd_otable_it_key,
2214                 .key_size = osd_otable_it_key_size,
2215                 .rec      = osd_otable_it_rec,
2216                 .store    = osd_otable_it_store,
2217                 .load     = osd_otable_it_load,
2218                 .key_rec  = osd_otable_it_key_rec,
2219         }
2220 };
2221
2222 /* high priority inconsistent items list APIs */
2223
2224 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2225                    int insert)
2226 {
2227         struct osd_inconsistent_item *oii;
2228         struct osd_scrub             *scrub  = &dev->od_scrub;
2229         struct ptlrpc_thread         *thread = &scrub->os_thread;
2230         int                           wakeup = 0;
2231         ENTRY;
2232
2233         OBD_ALLOC_PTR(oii);
2234         if (unlikely(oii == NULL))
2235                 RETURN(-ENOMEM);
2236
2237         CFS_INIT_LIST_HEAD(&oii->oii_list);
2238         oii->oii_cache = *oic;
2239         oii->oii_insert = insert;
2240
2241         spin_lock(&scrub->os_lock);
2242         if (unlikely(!thread_is_running(thread))) {
2243                 spin_unlock(&scrub->os_lock);
2244                 OBD_FREE_PTR(oii);
2245                 RETURN(-EAGAIN);
2246         }
2247
2248         if (cfs_list_empty(&scrub->os_inconsistent_items))
2249                 wakeup = 1;
2250         cfs_list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2251         spin_unlock(&scrub->os_lock);
2252
2253         if (wakeup != 0)
2254                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
2255
2256         RETURN(0);
2257 }
2258
2259 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2260                    struct osd_inode_id *id)
2261 {
2262         struct osd_scrub             *scrub = &dev->od_scrub;
2263         struct osd_inconsistent_item *oii;
2264         ENTRY;
2265
2266         spin_lock(&scrub->os_lock);
2267         cfs_list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2268                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2269                         *id = oii->oii_cache.oic_lid;
2270                         spin_unlock(&scrub->os_lock);
2271                         RETURN(0);
2272                 }
2273         }
2274         spin_unlock(&scrub->os_lock);
2275
2276         RETURN(-ENOENT);
2277 }
2278
2279 /* OI scrub dump */
2280
2281 static const char *scrub_status_names[] = {
2282         "init",
2283         "scanning",
2284         "completed",
2285         "failed",
2286         "stopped",
2287         "paused",
2288         "crashed",
2289         NULL
2290 };
2291
2292 static const char *scrub_flags_names[] = {
2293         "recreated",
2294         "inconsistent",
2295         "auto",
2296         "upgrade",
2297         NULL
2298 };
2299
2300 static const char *scrub_param_names[] = {
2301         "failout",
2302         NULL
2303 };
2304
2305 static int scrub_bits_dump(char **buf, int *len, int bits, const char *names[],
2306                            const char *prefix)
2307 {
2308         int save = *len;
2309         int flag;
2310         int rc;
2311         int i;
2312
2313         rc = snprintf(*buf, *len, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2314         if (rc <= 0)
2315                 return -ENOSPC;
2316
2317         *buf += rc;
2318         *len -= rc;
2319         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2320                 if (flag & bits) {
2321                         bits &= ~flag;
2322                         rc = snprintf(*buf, *len, "%s%c", names[i],
2323                                       bits != 0 ? ',' : '\n');
2324                         if (rc <= 0)
2325                                 return -ENOSPC;
2326
2327                         *buf += rc;
2328                         *len -= rc;
2329                 }
2330         }
2331         return save - *len;
2332 }
2333
2334 static int scrub_time_dump(char **buf, int *len, __u64 time, const char *prefix)
2335 {
2336         int rc;
2337
2338         if (time != 0)
2339                 rc = snprintf(*buf, *len, "%s: "LPU64" seconds\n", prefix,
2340                               cfs_time_current_sec() - time);
2341         else
2342                 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
2343         if (rc <= 0)
2344                 return -ENOSPC;
2345
2346         *buf += rc;
2347         *len -= rc;
2348         return rc;
2349 }
2350
2351 static int scrub_pos_dump(char **buf, int *len, __u64 pos, const char *prefix)
2352 {
2353         int rc;
2354
2355         if (pos != 0)
2356                 rc = snprintf(*buf, *len, "%s: "LPU64"\n", prefix, pos);
2357         else
2358                 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
2359         if (rc <= 0)
2360                 return -ENOSPC;
2361
2362         *buf += rc;
2363         *len -= rc;
2364         return rc;
2365 }
2366
2367 int osd_scrub_dump(struct osd_device *dev, char *buf, int len)
2368 {
2369         struct osd_scrub  *scrub   = &dev->od_scrub;
2370         struct scrub_file *sf      = &scrub->os_file;
2371         __u64              checked;
2372         __u64              speed;
2373         int                save    = len;
2374         int                ret     = -ENOSPC;
2375         int                rc;
2376
2377         down_read(&scrub->os_rwsem);
2378         rc = snprintf(buf, len,
2379                       "name: OI_scrub\n"
2380                       "magic: 0x%x\n"
2381                       "oi_files: %d\n"
2382                       "status: %s\n",
2383                       sf->sf_magic, (int)sf->sf_oi_count,
2384                       scrub_status_names[sf->sf_status]);
2385         if (rc <= 0)
2386                 goto out;
2387
2388         buf += rc;
2389         len -= rc;
2390         rc = scrub_bits_dump(&buf, &len, sf->sf_flags, scrub_flags_names,
2391                              "flags");
2392         if (rc < 0)
2393                 goto out;
2394
2395         rc = scrub_bits_dump(&buf, &len, sf->sf_param, scrub_param_names,
2396                              "param");
2397         if (rc < 0)
2398                 goto out;
2399
2400         rc = scrub_time_dump(&buf, &len, sf->sf_time_last_complete,
2401                              "time_since_last_completed");
2402         if (rc < 0)
2403                 goto out;
2404
2405         rc = scrub_time_dump(&buf, &len, sf->sf_time_latest_start,
2406                              "time_since_latest_start");
2407         if (rc < 0)
2408                 goto out;
2409
2410         rc = scrub_time_dump(&buf, &len, sf->sf_time_last_checkpoint,
2411                              "time_since_last_checkpoint");
2412         if (rc < 0)
2413                 goto out;
2414
2415         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_latest_start,
2416                             "latest_start_position");
2417         if (rc < 0)
2418                 goto out;
2419
2420         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_last_checkpoint,
2421                             "last_checkpoint_position");
2422         if (rc < 0)
2423                 goto out;
2424
2425         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_first_inconsistent,
2426                             "first_failure_position");
2427         if (rc < 0)
2428                 goto out;
2429
2430         checked = sf->sf_items_checked + scrub->os_new_checked;
2431         rc = snprintf(buf, len,
2432                       "checked: "LPU64"\n"
2433                       "updated: "LPU64"\n"
2434                       "failed: "LPU64"\n"
2435                       "prior_updated: "LPU64"\n"
2436                       "noscrub: "LPU64"\n"
2437                       "igif: "LPU64"\n"
2438                       "success_count: %u\n",
2439                       checked, sf->sf_items_updated, sf->sf_items_failed,
2440                       sf->sf_items_updated_prior, sf->sf_items_noscrub,
2441                       sf->sf_items_igif, sf->sf_success_count);
2442         if (rc <= 0)
2443                 goto out;
2444
2445         buf += rc;
2446         len -= rc;
2447         speed = checked;
2448         if (thread_is_running(&scrub->os_thread)) {
2449                 cfs_duration_t duration = cfs_time_current() -
2450                                           scrub->os_time_last_checkpoint;
2451                 __u64 new_checked = scrub->os_new_checked * CFS_HZ;
2452                 __u32 rtime = sf->sf_run_time +
2453                               cfs_duration_sec(duration + HALF_SEC);
2454
2455                 if (duration != 0)
2456                         do_div(new_checked, duration);
2457                 if (rtime != 0)
2458                         do_div(speed, rtime);
2459                 rc = snprintf(buf, len,
2460                               "run_time: %u seconds\n"
2461                               "average_speed: "LPU64" objects/sec\n"
2462                               "real-time_speed: "LPU64" objects/sec\n"
2463                               "current_position: %u\n",
2464                               rtime, speed, new_checked, scrub->os_pos_current);
2465         } else {
2466                 if (sf->sf_run_time != 0)
2467                         do_div(speed, sf->sf_run_time);
2468                 rc = snprintf(buf, len,
2469                               "run_time: %u seconds\n"
2470                               "average_speed: "LPU64" objects/sec\n"
2471                               "real-time_speed: N/A\n"
2472                               "current_position: N/A\n",
2473                               sf->sf_run_time, speed);
2474         }
2475         if (rc <= 0)
2476                 goto out;
2477
2478         buf += rc;
2479         len -= rc;
2480         ret = save - len;
2481
2482 out:
2483         up_read(&scrub->os_rwsem);
2484         return ret;
2485 }