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_scrub_get_fid(struct osd_thread_info *info,
 707                              struct osd_device *dev, struct inode *inode,
 708                              struct lu_fid *fid, bool scrub)
 709 {
 710         struct lustre_mdt_attrs *lma     = &info->oti_mdt_attrs;
 711         int                      rc;
 712         bool                     has_lma = false;
 713
 714         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 715         if (rc == 0) {
 716                 has_lma = true;
 717                 if (lma->lma_compat & LMAC_NOT_IN_OI) {
 718                         ldiskfs_set_inode_state(inode,
 719                                                 LDISKFS_STATE_LUSTRE_NO_OI);
 720                         return SCRUB_NEXT_CONTINUE;
 721                 }
 722
 723                 *fid = lma->lma_self_fid;
 724                 if (fid_is_internal(&lma->lma_self_fid)) {
 725                         if (!scrub)
 726                                 rc = SCRUB_NEXT_CONTINUE;
 727                         return rc;
 728                 }
 729
 730                 if (!scrub)
 731                         return 0;
 732
 733                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
 734                         return 0;
 735
 736                 if (lma->lma_compat & LMAC_FID_ON_OST)
 737                         return SCRUB_NEXT_OSTOBJ;
 738
 739                 if (fid_is_idif(fid) || fid_is_last_id(fid))
 740                         return SCRUB_NEXT_OSTOBJ_OLD;
 741
 742                 if (lma->lma_incompat & LMAI_AGENT)
 743                         return SCRUB_NEXT_CONTINUE;
 744
 745                 /* Here, it may be MDT-object, or may be 2.4 OST-object.
 746                  * Fall through. */
 747         }
 748
 749         if (rc == -ENODATA || rc == 0) {
 750                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
 751                 if (rc == 0) {
 752                         if (scrub)
 753                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
 754                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 755                         return rc;
 756                 }
 757
 758                 if (rc > 0) {
 759                         if (!has_lma)
 760                                 /* It is FID-on-OST, but we do not know how
 761                                  * to generate its FID, ignore it directly. */
 762                                 rc = SCRUB_NEXT_CONTINUE;
 763                         else
 764                                 /* It is 2.4 OST-object. */
 765                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 766                         return rc;
 767                 }
 768
 769                 if (rc != -ENODATA)
 770                         return rc;
 771
 772                 if (!has_lma) {
 773                         if (dev->od_handle_nolma) {
 774                                 lu_igif_build(fid, inode->i_ino,
 775                                               inode->i_generation);
 776                                 if (scrub)
 777                                         rc = SCRUB_NEXT_NOLMA;
 778                                 else
 779                                         rc = 0;
 780                         } else {
 781                                 /* It may be FID-on-OST, or may be FID for
 782                                  * non-MDT0, anyway, we do not know how to
 783                                  * generate its FID, ignore it directly. */
 784                                 rc = SCRUB_NEXT_CONTINUE;
 785                         }
 786                         return rc;
 787                 }
 788
 789                 /* For OI scrub case only: the object has LMA but has no ff
 790                  * (or ff crashed). It may be MDT-object, may be OST-object
 791                  * with crashed ff. The last check is local FLDB. */
 792                 rc = osd_scrub_check_local_fldb(info, dev, fid);
 793         }
 794
 795         return rc;
 796 }
 797
 798 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 799                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 800                         struct super_block *sb, bool scrub)
 801 {
 802         struct inode *inode;
 803         int           rc;
 804         ENTRY;
 805
 806         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 807         inode = osd_iget(info, dev, lid);
 808         if (IS_ERR(inode)) {
 809                 rc = PTR_ERR(inode);
 810                 /* The inode may be removed after bitmap searching, or the
 811                  * file is new created without inode initialized yet. */
 812                 if (rc == -ENOENT || rc == -ESTALE)
 813                         RETURN(SCRUB_NEXT_CONTINUE);
 814
 815                 CERROR("%.16s: fail to read inode, ino# = %u, rc = %d\n",
 816                        LDISKFS_SB(sb)->s_es->s_volume_name, pos, rc);
 817                 RETURN(rc);
 818         }
 819
 820         /* If the inode has no OI mapping, then it is special locally used,
 821          * should be invisible to OI scrub or up layer LFSCK. */
 822         if (ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI))
 823                 GOTO(put, rc = SCRUB_NEXT_CONTINUE);
 824
 825         if (scrub &&
 826             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 827                 /* Only skip it for the first OI scrub accessing. */
 828                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 829                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
 830         }
 831
 832         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
 833
 834         GOTO(put, rc);
 835
 836 put:
 837         iput(inode);
 838         return rc;
 839 }
 840
 841 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 842                           struct osd_iit_param *param,
 843                           struct osd_idmap_cache **oic, int noslot)
 844 {
 845         struct osd_scrub     *scrub  = &dev->od_scrub;
 846         struct ptlrpc_thread *thread = &scrub->os_thread;
 847         struct lu_fid        *fid;
 848         struct osd_inode_id  *lid;
 849         int                   rc;
 850
 851         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
 852                 struct l_wait_info lwi;
 853
 854                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
 855                 l_wait_event(thread->t_ctl_waitq,
 856                              !cfs_list_empty(&scrub->os_inconsistent_items) ||
 857                              !thread_is_running(thread),
 858                              &lwi);
 859         }
 860
 861         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 862                 spin_lock(&scrub->os_lock);
 863                 thread_set_flags(thread, SVC_STOPPING);
 864                 spin_unlock(&scrub->os_lock);
 865                 return SCRUB_NEXT_CRASH;
 866         }
 867
 868         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
 869                 return SCRUB_NEXT_FATAL;
 870
 871         if (unlikely(!thread_is_running(thread)))
 872                 return SCRUB_NEXT_EXIT;
 873
 874         if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
 875                 struct osd_inconsistent_item *oii;
 876
 877                 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
 878                                      struct osd_inconsistent_item, oii_list);
 879                 *oic = &oii->oii_cache;
 880                 scrub->os_in_prior = 1;
 881                 return 0;
 882         }
 883
 884         if (noslot != 0)
 885                 return SCRUB_NEXT_WAIT;
 886
 887         rc = osd_iit_next(param, &scrub->os_pos_current);
 888         if (rc != 0)
 889                 return rc;
 890
 891         *oic = &scrub->os_oic;
 892         fid = &(*oic)->oic_fid;
 893         lid = &(*oic)->oic_lid;
 894         rc = osd_iit_iget(info, dev, fid, lid,
 895                           scrub->os_pos_current, param->sb, true);
 896         return rc;
 897 }
 898
 899 static int osd_preload_next(struct osd_thread_info *info,
 900                             struct osd_device *dev, struct osd_iit_param *param,
 901                             struct osd_idmap_cache **oic, int noslot)
 902 {
 903         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
 904         struct osd_scrub        *scrub;
 905         struct ptlrpc_thread    *thread;
 906         int                      rc;
 907
 908         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
 909         if (rc != 0)
 910                 return rc;
 911
 912         scrub = &dev->od_scrub;
 913         thread = &scrub->os_thread;
 914         if (thread_is_running(thread) &&
 915             ooc->ooc_pos_preload >= scrub->os_pos_current)
 916                 return SCRUB_NEXT_EXIT;
 917
 918         rc = osd_iit_iget(info, dev,
 919                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
 920                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
 921                           ooc->ooc_pos_preload, param->sb, false);
 922         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
 923          * ignore the failure, so it still need to skip the inode next time. */
 924         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
 925         return rc;
 926 }
 927
 928 static inline int
 929 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
 930 {
 931         spin_lock(&scrub->os_lock);
 932         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
 933             !cfs_list_empty(&scrub->os_inconsistent_items) ||
 934             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
 935                 scrub->os_waiting = 0;
 936         else
 937                 scrub->os_waiting = 1;
 938         spin_unlock(&scrub->os_lock);
 939
 940         return !scrub->os_waiting;
 941 }
 942
 943 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
 944                           struct osd_iit_param *param,
 945                           struct osd_idmap_cache *oic, int *noslot, int rc)
 946 {
 947         struct l_wait_info       lwi    = { 0 };
 948         struct osd_scrub        *scrub  = &dev->od_scrub;
 949         struct scrub_file       *sf     = &scrub->os_file;
 950         struct ptlrpc_thread    *thread = &scrub->os_thread;
 951         struct osd_otable_it    *it     = dev->od_otable_it;
 952         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
 953
 954         switch (rc) {
 955         case SCRUB_NEXT_CONTINUE:
 956                 goto next;
 957         case SCRUB_NEXT_WAIT:
 958                 goto wait;
 959         case SCRUB_NEXT_NOSCRUB:
 960                 down_write(&scrub->os_rwsem);
 961                 scrub->os_new_checked++;
 962                 sf->sf_items_noscrub++;
 963                 up_write(&scrub->os_rwsem);
 964                 goto next;
 965         }
 966
 967         rc = osd_scrub_check_update(info, dev, oic, rc);
 968         if (rc != 0)
 969                 return rc;
 970
 971         rc = osd_scrub_checkpoint(scrub);
 972         if (rc != 0) {
 973                 CERROR("%.16s: fail to checkpoint, pos = %u, rc = %d\n",
 974                        LDISKFS_SB(param->sb)->s_es->s_volume_name,
 975                        scrub->os_pos_current, rc);
 976                 /* Continue, as long as the scrub itself can go ahead. */
 977         }
 978
 979         if (scrub->os_in_prior) {
 980                 scrub->os_in_prior = 0;
 981                 return 0;
 982         }
 983
 984 next:
 985         scrub->os_pos_current = param->gbase + ++(param->offset);
 986
 987 wait:
 988         if (it != NULL && it->ooi_waiting &&
 989             ooc->ooc_pos_preload < scrub->os_pos_current) {
 990                 spin_lock(&scrub->os_lock);
 991                 it->ooi_waiting = 0;
 992                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
 993                 spin_unlock(&scrub->os_lock);
 994         }
 995
 996         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
 997                 return 0;
 998
 999         if (osd_scrub_has_window(scrub, ooc)) {
1000                 *noslot = 0;
1001                 return 0;
1002         }
1003
1004         l_wait_event(thread->t_ctl_waitq,
1005                      osd_scrub_wakeup(scrub, it),
1006                      &lwi);
1007
1008         if (osd_scrub_has_window(scrub, ooc))
1009                 *noslot = 0;
1010         else
1011                 *noslot = 1;
1012         return 0;
1013 }
1014
1015 static int osd_preload_exec(struct osd_thread_info *info,
1016                             struct osd_device *dev, struct osd_iit_param *param,
1017                             struct osd_idmap_cache *oic, int *noslot, int rc)
1018 {
1019         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1020
1021         if (rc == 0) {
1022                 ooc->ooc_cached_items++;
1023                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1024                                         ~OSD_OTABLE_IT_CACHE_MASK;
1025         }
1026         return rc > 0 ? 0 : rc;
1027 }
1028
1029 #define SCRUB_IT_ALL    1
1030 #define SCRUB_IT_CRASH  2
1031
1032 static int osd_inode_iteration(struct osd_thread_info *info,
1033                                struct osd_device *dev, __u32 max, bool preload)
1034 {
1035         osd_iit_next_policy   next;
1036         osd_iit_exec_policy   exec;
1037         __u32                *pos;
1038         __u32                *count;
1039         struct osd_iit_param  param;
1040         __u32                 limit;
1041         int                   noslot = 0;
1042         int                   rc;
1043         ENTRY;
1044
1045         if (!preload) {
1046                 struct osd_scrub *scrub = &dev->od_scrub;
1047
1048                 next = osd_scrub_next;
1049                 exec = osd_scrub_exec;
1050                 pos = &scrub->os_pos_current;
1051                 count = &scrub->os_new_checked;
1052         } else {
1053                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1054
1055                 next = osd_preload_next;
1056                 exec = osd_preload_exec;
1057                 pos = &ooc->ooc_pos_preload;
1058                 count = &ooc->ooc_cached_items;
1059         }
1060         param.sb = osd_sb(dev);
1061         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1062
1063         while (*pos <= limit && *count < max) {
1064                 struct osd_idmap_cache *oic = NULL;
1065
1066                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1067                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1068                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1069                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1070                 if (param.bitmap == NULL) {
1071                         CERROR("%.16s: fail to read bitmap for %u, "
1072                                "scrub will stop, urgent mode\n",
1073                                LDISKFS_SB(param.sb)->s_es->s_volume_name,
1074                                (__u32)param.bg);
1075                         RETURN(-EIO);
1076                 }
1077
1078                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1079                        *count < max) {
1080                         rc = next(info, dev, &param, &oic, noslot);
1081                         switch (rc) {
1082                         case SCRUB_NEXT_BREAK:
1083                                 goto next_group;
1084                         case SCRUB_NEXT_EXIT:
1085                                 brelse(param.bitmap);
1086                                 RETURN(0);
1087                         case SCRUB_NEXT_CRASH:
1088                                 brelse(param.bitmap);
1089                                 RETURN(SCRUB_IT_CRASH);
1090                         case SCRUB_NEXT_FATAL:
1091                                 brelse(param.bitmap);
1092                                 RETURN(-EINVAL);
1093                         }
1094
1095                         rc = exec(info, dev, &param, oic, &noslot, rc);
1096                         if (rc != 0) {
1097                                 brelse(param.bitmap);
1098                                 RETURN(rc);
1099                         }
1100                 }
1101
1102 next_group:
1103                 brelse(param.bitmap);
1104         }
1105
1106         if (*pos > limit)
1107                 RETURN(SCRUB_IT_ALL);
1108         RETURN(0);
1109 }
1110
1111 static int osd_otable_it_preload(const struct lu_env *env,
1112                                  struct osd_otable_it *it)
1113 {
1114         struct osd_device       *dev   = it->ooi_dev;
1115         struct osd_scrub        *scrub = &dev->od_scrub;
1116         struct osd_otable_cache *ooc   = &it->ooi_cache;
1117         int                      rc;
1118         ENTRY;
1119
1120         rc = osd_inode_iteration(osd_oti_get(env), dev,
1121                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1122         if (rc == SCRUB_IT_ALL)
1123                 it->ooi_all_cached = 1;
1124
1125         CDEBUG(D_LFSCK, "OSD pre-loaded: max = %u, preload = %u, rc = %d\n",
1126                le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count),
1127                ooc->ooc_pos_preload, rc);
1128
1129         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1130                 scrub->os_waiting = 0;
1131                 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
1132         }
1133
1134         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1135 }
1136
1137 static int osd_scrub_main(void *args)
1138 {
1139         struct lu_env         env;
1140         struct osd_device    *dev    = (struct osd_device *)args;
1141         struct osd_scrub     *scrub  = &dev->od_scrub;
1142         struct ptlrpc_thread *thread = &scrub->os_thread;
1143         struct super_block   *sb     = osd_sb(dev);
1144         int                   rc;
1145         ENTRY;
1146
1147         rc = lu_env_init(&env, LCT_LOCAL);
1148         if (rc != 0) {
1149                 CERROR("%.16s: OI scrub, fail to init env, rc = %d\n",
1150                        LDISKFS_SB(sb)->s_es->s_volume_name, rc);
1151                 GOTO(noenv, rc);
1152         }
1153
1154         rc = osd_scrub_prep(dev);
1155         if (rc != 0) {
1156                 CERROR("%.16s: OI scrub, fail to scrub prep, rc = %d\n",
1157                        LDISKFS_SB(sb)->s_es->s_volume_name, rc);
1158                 GOTO(out, rc);
1159         }
1160
1161         if (!scrub->os_full_speed) {
1162                 struct l_wait_info lwi = { 0 };
1163                 struct osd_otable_it *it = dev->od_otable_it;
1164                 struct osd_otable_cache *ooc = &it->ooi_cache;
1165
1166                 l_wait_event(thread->t_ctl_waitq,
1167                              it->ooi_user_ready || !thread_is_running(thread),
1168                              &lwi);
1169                 if (unlikely(!thread_is_running(thread)))
1170                         GOTO(post, rc = 0);
1171
1172                 scrub->os_pos_current = ooc->ooc_pos_preload;
1173         }
1174
1175         CDEBUG(D_LFSCK, "OI scrub: flags = 0x%x, pos = %u\n",
1176                scrub->os_start_flags, scrub->os_pos_current);
1177
1178         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1179         if (unlikely(rc == SCRUB_IT_CRASH))
1180                 GOTO(out, rc = -EINVAL);
1181         GOTO(post, rc);
1182
1183 post:
1184         osd_scrub_post(scrub, rc);
1185         CDEBUG(D_LFSCK, "OI scrub: stop, rc = %d, pos = %u\n",
1186                rc, scrub->os_pos_current);
1187
1188 out:
1189         while (!cfs_list_empty(&scrub->os_inconsistent_items)) {
1190                 struct osd_inconsistent_item *oii;
1191
1192                 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
1193                                      struct osd_inconsistent_item, oii_list);
1194                 cfs_list_del_init(&oii->oii_list);
1195                 OBD_FREE_PTR(oii);
1196         }
1197         lu_env_fini(&env);
1198
1199 noenv:
1200         spin_lock(&scrub->os_lock);
1201         thread_set_flags(thread, SVC_STOPPED);
1202         cfs_waitq_broadcast(&thread->t_ctl_waitq);
1203         spin_unlock(&scrub->os_lock);
1204         return rc;
1205 }
1206
1207 /* initial OI scrub */
1208
1209 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1210                          struct dentry *, filldir_t filldir);
1211
1212 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1213                                loff_t offset, __u64 ino, unsigned d_type);
1214 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1215                            loff_t offset, __u64 ino, unsigned d_type);
1216
1217 static int
1218 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1219                      struct dentry *dentry, filldir_t filldir);
1220 static int
1221 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1222                   struct dentry *dentry, filldir_t filldir);
1223
1224 static int
1225 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1226                      struct dentry *dentry, filldir_t filldir);
1227
1228 enum osd_lf_flags {
1229         OLF_SCAN_SUBITEMS       = 0x0001,
1230         OLF_HIDE_FID            = 0x0002,
1231         OLF_SHOW_NAME           = 0x0004,
1232         OLF_NO_OI               = 0x0008,
1233 };
1234
1235 struct osd_lf_map {
1236         char            *olm_name;
1237         struct lu_fid    olm_fid;
1238         __u16            olm_flags;
1239         scandir_t        olm_scandir;
1240         filldir_t        olm_filldir;
1241 };
1242
1243 /* Add the new introduced local files in the list in the future. */
1244 static const struct osd_lf_map osd_lf_maps[] = {
1245         /* CATALOGS */
1246         { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1247                 NULL, NULL },
1248
1249         /* CONFIGS */
1250         { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1251                 OLF_SCAN_SUBITEMS, osd_ios_general_scan,
1252                 osd_ios_varfid_fill },
1253
1254         /* NIDTBL_VERSIONS */
1255         { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1256                 osd_ios_general_scan, osd_ios_varfid_fill },
1257
1258         /* PENDING */
1259         { "PENDING", { 0, 0, 0 }, 0, NULL, NULL },
1260
1261         /* ROOT */
1262         { "ROOT", { FID_SEQ_ROOT, 1, 0 },
1263                 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, osd_ios_ROOT_scan, NULL },
1264
1265         /* changelog_catalog */
1266         { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, NULL, NULL },
1267
1268         /* changelog_users */
1269         { CHANGELOG_USERS, { 0, 0, 0 }, 0, NULL, NULL },
1270
1271         /* fld */
1272         { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1273                 NULL, NULL },
1274
1275         /* last_rcvd */
1276         { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1277                 NULL, NULL },
1278
1279         /* lfsck_bookmark */
1280         { "lfsck_bookmark", { 0, 0, 0 }, 0, NULL, NULL },
1281
1282         /* lov_objid */
1283         { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1284                 NULL, NULL },
1285
1286         /* lov_objseq */
1287         { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1288                 OLF_SHOW_NAME, NULL, NULL },
1289
1290         /* quota_master */
1291         { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1292                 osd_ios_general_scan, osd_ios_varfid_fill },
1293
1294         /* quota_slave */
1295         { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1296                 osd_ios_general_scan, osd_ios_varfid_fill },
1297
1298         /* seq_ctl */
1299         { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1300                 OLF_SHOW_NAME, NULL, NULL },
1301
1302         /* seq_srv */
1303         { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1304                 OLF_SHOW_NAME, NULL, NULL },
1305
1306         /* health_check */
1307         { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1308                 OLF_SHOW_NAME, NULL, NULL },
1309
1310         /* lfsck_namespace */
1311         { "lfsck_namespace", { 0, 0, 0 }, 0, NULL, NULL },
1312
1313         /* OBJECTS, upgrade from old device */
1314         { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, osd_ios_OBJECTS_scan, NULL },
1315
1316         /* lquota_v2.user, upgrade from old device */
1317         { "lquota_v2.user", { 0, 0, 0 }, 0, NULL, NULL },
1318
1319         /* lquota_v2.group, upgrade from old device */
1320         { "lquota_v2.group", { 0, 0, 0 }, 0, NULL, NULL },
1321
1322         /* LAST_GROUP, upgrade from old device */
1323         { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1324                 OLF_SHOW_NAME, NULL, NULL },
1325
1326         /* lost+found */
1327         { "lost+found", { 0, 0, 0 }, OLF_SCAN_SUBITEMS | OLF_NO_OI,
1328                 osd_ios_general_scan, osd_ios_lf_fill },
1329
1330         { NULL, { 0, 0, 0 }, 0, NULL, NULL }
1331 };
1332
1333 struct osd_ios_item {
1334         cfs_list_t       oii_list;
1335         struct dentry   *oii_dentry;
1336         scandir_t        oii_scandir;
1337         filldir_t        oii_filldir;
1338 };
1339
1340 struct osd_ios_filldir_buf {
1341         struct osd_thread_info  *oifb_info;
1342         struct osd_device       *oifb_dev;
1343         struct dentry           *oifb_dentry;
1344 };
1345
1346 static inline struct dentry *
1347 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1348 {
1349         struct dentry *dentry;
1350
1351         dentry = ll_lookup_one_len(name, parent, namelen);
1352         if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1353                 dput(dentry);
1354                 return ERR_PTR(-ENOENT);
1355         }
1356
1357         return dentry;
1358 }
1359
1360 static inline void
1361 osd_ios_llogname2fid(struct lu_fid *fid, const char *name, int namelen)
1362 {
1363         obd_id id = 0;
1364         int    i  = 0;
1365
1366         fid->f_seq = FID_SEQ_LLOG;
1367         while (i < namelen)
1368                 id = id * 10 + name[i++] - '0';
1369
1370         fid->f_oid = id & 0x00000000ffffffffULL;
1371         fid->f_ver = id >> 32;
1372 }
1373
1374 static inline void
1375 osd_ios_Oname2fid(struct lu_fid *fid, const char *name, int namelen)
1376 {
1377         __u64 seq = 0;
1378         int   i   = 0;
1379
1380         while (i < namelen)
1381                 seq = seq * 10 + name[i++] - '0';
1382
1383         lu_last_id_fid(fid, seq);
1384 }
1385
1386 static int
1387 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1388                  scandir_t scandir, filldir_t filldir)
1389 {
1390         struct osd_ios_item *item;
1391
1392         OBD_ALLOC_PTR(item);
1393         if (item == NULL)
1394                 return -ENOMEM;
1395
1396         CFS_INIT_LIST_HEAD(&item->oii_list);
1397         item->oii_dentry = dget(dentry);
1398         item->oii_scandir = scandir;
1399         item->oii_filldir = filldir;
1400         cfs_list_add_tail(&item->oii_list, &dev->od_ios_list);
1401         return 0;
1402 }
1403
1404 /**
1405  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1406  *
1407  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1408  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1409  * reference the inode, or fixed if it is missing or references another inode.
1410  */
1411 static int
1412 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1413                  struct inode *inode, const struct lu_fid *fid, int flags)
1414 {
1415         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
1416         struct osd_inode_id     *id     = &info->oti_id;
1417         struct osd_inode_id     *id2    = &info->oti_id2;
1418         struct osd_scrub        *scrub  = &dev->od_scrub;
1419         struct scrub_file       *sf     = &scrub->os_file;
1420         struct lu_fid            tfid;
1421         int                      rc;
1422         ENTRY;
1423
1424         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1425         if (rc != 0 && rc != -ENODATA)
1426                 RETURN(rc);
1427
1428         osd_id_gen(id, inode->i_ino, inode->i_generation);
1429         if (rc == -ENODATA) {
1430                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID)
1431                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1432                 else
1433                         tfid = *fid;
1434                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1435                 if (rc != 0)
1436                         RETURN(rc);
1437         } else {
1438                 if (lma->lma_compat & LMAC_NOT_IN_OI)
1439                         RETURN(0);
1440
1441                 tfid = lma->lma_self_fid;
1442         }
1443
1444         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1445         if (rc != 0) {
1446                 if (rc != -ENOENT)
1447                         RETURN(rc);
1448
1449                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1450                                                DTO_INDEX_INSERT, 0);
1451                 if (rc > 0)
1452                         rc = 0;
1453
1454                 RETURN(rc);
1455         }
1456
1457         if (osd_id_eq_strict(id, id2))
1458                 RETURN(0);
1459
1460         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1461                 osd_scrub_file_reset(scrub,
1462                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1463                                      SF_INCONSISTENT);
1464                 rc = osd_scrub_file_store(scrub);
1465                 if (rc != 0)
1466                         RETURN(rc);
1467         }
1468
1469         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1470                                        DTO_INDEX_UPDATE, 0);
1471         if (rc > 0)
1472                 rc = 0;
1473
1474         RETURN(rc);
1475 }
1476
1477 /**
1478  * It scans the /lost+found, and for the OST-object (with filter_fid
1479  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1480  */
1481 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1482                            loff_t offset, __u64 ino, unsigned d_type)
1483 {
1484         struct osd_ios_filldir_buf *fill_buf = buf;
1485         struct osd_thread_info     *info     = fill_buf->oifb_info;
1486         struct osd_device          *dev      = fill_buf->oifb_dev;
1487         struct lu_fid              *fid      = &info->oti_fid;
1488         struct osd_scrub           *scrub    = &dev->od_scrub;
1489         struct dentry              *parent   = fill_buf->oifb_dentry;
1490         struct dentry              *child;
1491         struct inode               *dir      = parent->d_inode;
1492         struct inode               *inode;
1493         int                         rc;
1494         ENTRY;
1495
1496         /* skip any '.' started names */
1497         if (name[0] == '.')
1498                 RETURN(0);
1499
1500         scrub->os_lf_scanned++;
1501         child = osd_ios_lookup_one_len(name, parent, namelen);
1502         if (IS_ERR(child)) {
1503                 CWARN("%s: cannot lookup child '%.*s': rc = %d\n",
1504                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
1505                 RETURN(0);
1506         }
1507
1508         inode = child->d_inode;
1509         if (S_ISDIR(inode->i_mode)) {
1510                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1511                                       osd_ios_lf_fill);
1512                 if (rc != 0)
1513                         CWARN("%s: cannot add child '%.*s': rc = %d\n",
1514                               osd_name(dev), namelen, name, rc);
1515                 GOTO(put, rc);
1516         }
1517
1518         if (!S_ISREG(inode->i_mode))
1519                 GOTO(put, rc = 0);
1520
1521         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1522         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1523                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1524                 if (rc == 0) {
1525                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1526                                "/lost+found.\n", namelen, name, PFID(fid));
1527                         scrub->os_lf_repaired++;
1528                 } else {
1529                         CWARN("%s: cannot rename for '%.*s' "DFID": rc = %d\n",
1530                               osd_name(dev), namelen, name, PFID(fid), rc);
1531                 }
1532         }
1533
1534         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1535          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1536          *      can process them in furtuer. */
1537
1538         GOTO(put, rc);
1539
1540 put:
1541         if (rc < 0)
1542                 scrub->os_lf_failed++;
1543         dput(child);
1544         /* skip the failure to make the scanning to continue. */
1545         return 0;
1546 }
1547
1548 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1549                                loff_t offset, __u64 ino, unsigned d_type)
1550 {
1551         struct osd_ios_filldir_buf *fill_buf = buf;
1552         struct osd_device          *dev      = fill_buf->oifb_dev;
1553         struct dentry              *child;
1554         int                         rc;
1555         ENTRY;
1556
1557         /* skip any '.' started names */
1558         if (name[0] == '.')
1559                 RETURN(0);
1560
1561         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1562         if (IS_ERR(child))
1563                 RETURN(PTR_ERR(child));
1564
1565         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1566                               NULL, 0);
1567         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1568                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1569                                       osd_ios_varfid_fill);
1570         dput(child);
1571
1572         RETURN(rc);
1573 }
1574
1575 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
1576                              loff_t offset, __u64 ino, unsigned d_type)
1577 {
1578         struct osd_ios_filldir_buf *fill_buf = buf;
1579         struct osd_device          *dev      = fill_buf->oifb_dev;
1580         const struct osd_lf_map    *map;
1581         struct dentry              *child;
1582         int                         rc       = 0;
1583         ENTRY;
1584
1585         /* skip any '.' started names */
1586         if (name[0] == '.')
1587                 RETURN(0);
1588
1589         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
1590                 if (strlen(map->olm_name) != namelen)
1591                         continue;
1592
1593                 if (strncmp(map->olm_name, name, namelen) == 0)
1594                         break;
1595         }
1596
1597         if (map->olm_name == NULL)
1598                 RETURN(0);
1599
1600         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1601         if (IS_ERR(child))
1602                 RETURN(PTR_ERR(child));
1603
1604         if (!(map->olm_flags & OLF_NO_OI))
1605                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1606                                       &map->olm_fid, map->olm_flags);
1607         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
1608                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
1609                                       map->olm_filldir);
1610         dput(child);
1611
1612         RETURN(rc);
1613 }
1614
1615 static int
1616 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1617                      struct dentry *dentry, filldir_t filldir)
1618 {
1619         struct osd_ios_filldir_buf    buf   = { info, dev, dentry };
1620         struct file                  *filp  = &info->oti_it_ea.oie_file;
1621         struct inode                 *inode = dentry->d_inode;
1622         const struct file_operations *fops  = inode->i_fop;
1623         int                           rc;
1624         ENTRY;
1625
1626         LASSERT(filldir != NULL);
1627
1628         filp->f_pos = 0;
1629         filp->f_dentry = dentry;
1630         filp->f_mode = FMODE_64BITHASH;
1631         filp->f_mapping = inode->i_mapping;
1632         filp->f_op = fops;
1633         filp->private_data = NULL;
1634
1635         rc = fops->readdir(filp, &buf, filldir);
1636         fops->release(inode, filp);
1637
1638         RETURN(rc);
1639 }
1640
1641 static int
1642 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1643                   struct dentry *dentry, filldir_t filldir)
1644 {
1645         struct osd_scrub  *scrub  = &dev->od_scrub;
1646         struct scrub_file *sf     = &scrub->os_file;
1647         struct dentry     *child;
1648         int                rc;
1649         ENTRY;
1650
1651         /* It is existing MDT0 device. We only allow the case of object without
1652          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
1653          * can generate IGIF mode FID for the object and related OI mapping. If
1654          * it is on other MDTs, then becuase file-level backup/restore, related
1655          * OI mapping may be invalid already, we do not know which is the right
1656          * FID for the object. We only allow IGIF objects to reside on the MDT0.
1657          *
1658          * XXX: For the case of object on non-MDT0 device with neither LMA nor
1659          *      "fid" xattr, then something crashed. We cannot re-generate the
1660          *      FID directly, instead, the OI scrub will scan the OI structure
1661          *      and try to re-generate the LMA from the OI mapping. But if the
1662          *      OI mapping crashed or lost also, then we have to give up under
1663          *      double failure cases. */
1664         dev->od_handle_nolma = 1;
1665         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
1666                                        strlen(dot_lustre_name));
1667         if (IS_ERR(child)) {
1668                 rc = PTR_ERR(child);
1669                 if (rc == -ENOENT) {
1670                         /* It is 1.8 MDT device. */
1671                         if (!(sf->sf_flags & SF_UPGRADE)) {
1672                                 osd_scrub_file_reset(scrub,
1673                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1674                                         SF_UPGRADE);
1675                                 rc = osd_scrub_file_store(scrub);
1676                         } else {
1677                                 rc = 0;
1678                         }
1679                 }
1680         } else {
1681                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
1682                  * so the client will get IGIF for the ".lustre" object when
1683                  * the MDT restart.
1684                  *
1685                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
1686                  * it does not know whether there are some old clients cached
1687                  * the ".lustre" IGIF during the upgrading. Two choices:
1688                  *
1689                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
1690                  *    It will allow the old connected clients to access the
1691                  *    ".lustre" with cached IGIF. But it will cause others
1692                  *    on the MDT failed to check "fid_is_dot_lustre()".
1693                  *
1694                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
1695                  *    for ".lustre" in spite of whether there are some clients
1696                  *    cached the ".lustre" IGIF or not. It enables the check
1697                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
1698                  *    that the old connected clients cannot access the ".lustre"
1699                  *    with the cached IGIF.
1700                  *
1701                  * Usually, it is rare case for the old connected clients
1702                  * to access the ".lustre" with cached IGIF. So we prefer
1703                  * to the solution 2). */
1704                 rc = osd_ios_scan_one(info, dev, child->d_inode,
1705                                       &LU_DOT_LUSTRE_FID, 0);
1706                 dput(child);
1707         }
1708
1709         RETURN(rc);
1710 }
1711
1712 static int
1713 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1714                      struct dentry *dentry, filldir_t filldir)
1715 {
1716         struct dentry *child;
1717         int            rc;
1718         ENTRY;
1719
1720         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
1721         if (!IS_ERR(child)) {
1722                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1723                 dput(child);
1724         } else {
1725                 rc = PTR_ERR(child);
1726         }
1727
1728         if (rc != 0 && rc != -ENOENT)
1729                 RETURN(rc);
1730
1731         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
1732         if (!IS_ERR(child)) {
1733                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1734                 dput(child);
1735         } else {
1736                 rc = PTR_ERR(child);
1737         }
1738
1739         if (rc == -ENOENT)
1740                 rc = 0;
1741
1742         RETURN(rc);
1743 }
1744
1745 static int osd_initial_OI_scrub(struct osd_thread_info *info,
1746                                 struct osd_device *dev)
1747 {
1748         struct osd_ios_item     *item    = NULL;
1749         scandir_t                scandir = osd_ios_general_scan;
1750         filldir_t                filldir = osd_ios_root_fill;
1751         struct dentry           *dentry  = osd_sb(dev)->s_root;
1752         const struct osd_lf_map *map     = osd_lf_maps;
1753         int                      rc;
1754         ENTRY;
1755
1756         while (1) {
1757                 rc = scandir(info, dev, dentry, filldir);
1758                 if (item != NULL) {
1759                         dput(item->oii_dentry);
1760                         OBD_FREE_PTR(item);
1761                 }
1762
1763                 if (rc != 0)
1764                         break;
1765
1766                 if (cfs_list_empty(&dev->od_ios_list))
1767                         break;
1768
1769                 item = cfs_list_entry(dev->od_ios_list.next,
1770                                       struct osd_ios_item, oii_list);
1771                 cfs_list_del_init(&item->oii_list);
1772
1773                 LASSERT(item->oii_scandir != NULL);
1774                 scandir = item->oii_scandir;
1775                 filldir = item->oii_filldir;
1776                 dentry = item->oii_dentry;
1777         }
1778
1779         while (!cfs_list_empty(&dev->od_ios_list)) {
1780                 item = cfs_list_entry(dev->od_ios_list.next,
1781                                       struct osd_ios_item, oii_list);
1782                 cfs_list_del_init(&item->oii_list);
1783                 dput(item->oii_dentry);
1784                 OBD_FREE_PTR(item);
1785         }
1786
1787         if (rc != 0)
1788                 RETURN(rc);
1789
1790         /* There maybe the case that the object has been removed, but its OI
1791          * mapping is still in the OI file, such as the "CATALOGS" after MDT
1792          * file-level backup/restore. So here cleanup the stale OI mappings. */
1793         while (map->olm_name != NULL) {
1794                 struct dentry *child;
1795
1796                 if (fid_is_zero(&map->olm_fid)) {
1797                         map++;
1798                         continue;
1799                 }
1800
1801                 child = osd_ios_lookup_one_len(map->olm_name,
1802                                                osd_sb(dev)->s_root,
1803                                                strlen(map->olm_name));
1804                 if (!IS_ERR(child))
1805                         dput(child);
1806                 else if (PTR_ERR(child) == -ENOENT)
1807                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
1808                                                   NULL, DTO_INDEX_DELETE, 0);
1809                 map++;
1810         }
1811
1812         RETURN(0);
1813 }
1814
1815 char *osd_lf_fid2name(const struct lu_fid *fid)
1816 {
1817         const struct osd_lf_map *map = osd_lf_maps;
1818
1819         while (map->olm_name != NULL) {
1820                 if (!lu_fid_eq(fid, &map->olm_fid)) {
1821                         map++;
1822                         continue;
1823                 }
1824
1825                 if (map->olm_flags & OLF_SHOW_NAME)
1826                         return map->olm_name;
1827                 else
1828                         return "";
1829         }
1830
1831         return NULL;
1832 }
1833
1834 /* OI scrub start/stop */
1835
1836 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
1837 {
1838         struct osd_scrub     *scrub  = &dev->od_scrub;
1839         struct ptlrpc_thread *thread = &scrub->os_thread;
1840         struct l_wait_info    lwi    = { 0 };
1841         int                   rc;
1842         ENTRY;
1843
1844 again:
1845         /* os_lock: sync status between stop and scrub thread */
1846         spin_lock(&scrub->os_lock);
1847         if (thread_is_running(thread)) {
1848                 spin_unlock(&scrub->os_lock);
1849                 RETURN(-EALREADY);
1850         } else if (unlikely(thread_is_stopping(thread))) {
1851                 spin_unlock(&scrub->os_lock);
1852                 l_wait_event(thread->t_ctl_waitq,
1853                              thread_is_stopped(thread),
1854                              &lwi);
1855                 goto again;
1856         }
1857         spin_unlock(&scrub->os_lock);
1858
1859         if (scrub->os_file.sf_status == SS_COMPLETED)
1860                 flags |= SS_RESET;
1861
1862         scrub->os_start_flags = flags;
1863         thread_set_flags(thread, 0);
1864         rc = PTR_ERR(kthread_run(osd_scrub_main, dev, "OI_scrub"));
1865         if (IS_ERR_VALUE(rc)) {
1866                 CERROR("%.16s: cannot start iteration thread, rc = %d\n",
1867                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
1868                 RETURN(rc);
1869         }
1870
1871         l_wait_event(thread->t_ctl_waitq,
1872                      thread_is_running(thread) || thread_is_stopped(thread),
1873                      &lwi);
1874
1875         RETURN(0);
1876 }
1877
1878 int osd_scrub_start(struct osd_device *dev)
1879 {
1880         int rc;
1881         ENTRY;
1882
1883         /* od_otable_mutex: prevent curcurrent start/stop */
1884         mutex_lock(&dev->od_otable_mutex);
1885         rc = do_osd_scrub_start(dev, SS_AUTO);
1886         mutex_unlock(&dev->od_otable_mutex);
1887
1888         RETURN(rc == -EALREADY ? 0 : rc);
1889 }
1890
1891 static void do_osd_scrub_stop(struct osd_scrub *scrub)
1892 {
1893         struct ptlrpc_thread *thread = &scrub->os_thread;
1894         struct l_wait_info    lwi    = { 0 };
1895
1896         /* os_lock: sync status between stop and scrub thread */
1897         spin_lock(&scrub->os_lock);
1898         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
1899                 thread_set_flags(thread, SVC_STOPPING);
1900                 spin_unlock(&scrub->os_lock);
1901                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
1902                 l_wait_event(thread->t_ctl_waitq,
1903                              thread_is_stopped(thread),
1904                              &lwi);
1905                 /* Do not skip the last lock/unlock, which can guarantee that
1906                  * the caller cannot return until the OI scrub thread exit. */
1907                 spin_lock(&scrub->os_lock);
1908         }
1909         spin_unlock(&scrub->os_lock);
1910 }
1911
1912 static void osd_scrub_stop(struct osd_device *dev)
1913 {
1914         /* od_otable_mutex: prevent curcurrent start/stop */
1915         mutex_lock(&dev->od_otable_mutex);
1916         dev->od_scrub.os_paused = 1;
1917         do_osd_scrub_stop(&dev->od_scrub);
1918         mutex_unlock(&dev->od_otable_mutex);
1919 }
1920
1921 /* OI scrub setup/cleanup */
1922
1923 static const char osd_scrub_name[] = "OI_scrub";
1924
1925 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
1926 {
1927         struct osd_thread_info     *info   = osd_oti_get(env);
1928         struct osd_scrub           *scrub  = &dev->od_scrub;
1929         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
1930         struct scrub_file          *sf     = &scrub->os_file;
1931         struct super_block         *sb     = osd_sb(dev);
1932         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
1933         struct lvfs_run_ctxt        saved;
1934         struct file                *filp;
1935         struct inode               *inode;
1936         struct lu_fid              *fid    = &info->oti_fid;
1937         int                         dirty  = 0;
1938         int                         rc     = 0;
1939         ENTRY;
1940
1941         memset(scrub, 0, sizeof(*scrub));
1942         OBD_SET_CTXT_MAGIC(ctxt);
1943         ctxt->pwdmnt = dev->od_mnt;
1944         ctxt->pwd = dev->od_mnt->mnt_root;
1945         ctxt->fs = get_ds();
1946
1947         cfs_waitq_init(&scrub->os_thread.t_ctl_waitq);
1948         init_rwsem(&scrub->os_rwsem);
1949         spin_lock_init(&scrub->os_lock);
1950         CFS_INIT_LIST_HEAD(&scrub->os_inconsistent_items);
1951
1952         push_ctxt(&saved, ctxt, NULL);
1953         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
1954         if (IS_ERR(filp)) {
1955                 pop_ctxt(&saved, ctxt, NULL);
1956                 RETURN(PTR_ERR(filp));
1957         }
1958
1959         inode = filp->f_dentry->d_inode;
1960         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI);
1961         /* 'What the @fid is' is not imporatant, because the object
1962          * has no OI mapping, and only is visible inside the OSD.*/
1963         lu_igif_build(fid, inode->i_ino, inode->i_generation);
1964         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
1965         if (rc != 0) {
1966                 filp_close(filp, 0);
1967                 pop_ctxt(&saved, ctxt, NULL);
1968                 RETURN(rc);
1969         }
1970
1971         scrub->os_inode = igrab(inode);
1972         filp_close(filp, 0);
1973         pop_ctxt(&saved, ctxt, NULL);
1974
1975         rc = osd_scrub_file_load(scrub);
1976         if (rc == -ENOENT) {
1977                 osd_scrub_file_init(scrub, es->s_uuid);
1978                 dirty = 1;
1979         } else if (rc != 0) {
1980                 RETURN(rc);
1981         } else {
1982                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
1983                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
1984                         dirty = 1;
1985                 } else if (sf->sf_status == SS_SCANNING) {
1986                         sf->sf_status = SS_CRASHED;
1987                         dirty = 1;
1988                 }
1989         }
1990
1991         if (sf->sf_pos_last_checkpoint != 0)
1992                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
1993         else
1994                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
1995
1996         if (dirty != 0) {
1997                 rc = osd_scrub_file_store(scrub);
1998                 if (rc != 0)
1999                         RETURN(rc);
2000         }
2001
2002         /* Initialize OI files. */
2003         rc = osd_oi_init(info, dev);
2004         if (rc < 0)
2005                 RETURN(rc);
2006
2007         rc = osd_initial_OI_scrub(info, dev);
2008         if (rc == 0) {
2009                 if ((sf->sf_flags & SF_UPGRADE) &&
2010                    !(sf->sf_flags & SF_INCONSISTENT))
2011                         /* The 'od_igif_inoi' will be set after the
2012                          * upgrading completed, needs NOT remount. */
2013                         dev->od_igif_inoi = 0;
2014                 else
2015                         /* The 'od_igif_inoi' will be set under the
2016                          * following cases:
2017                          * 1) new created system, or
2018                          * 2) restored from file-level backup, or
2019                          * 3) the upgrading completed.
2020                          *
2021                          * The 'od_igif_inoi' may be cleared by OI scrub
2022                          * later if found that the system is upgrading. */
2023                         dev->od_igif_inoi = 1;
2024
2025                 if (!dev->od_noscrub &&
2026                     ((sf->sf_status == SS_PAUSED) ||
2027                      (sf->sf_status == SS_CRASHED &&
2028                       sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2029                                       SF_UPGRADE | SF_AUTO)) ||
2030                      (sf->sf_status == SS_INIT &&
2031                       sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2032                                       SF_UPGRADE))))
2033                         rc = osd_scrub_start(dev);
2034         }
2035
2036         /* it is possible that dcache entries may keep objects after they are
2037          * deleted by OSD. While it looks safe this can cause object data to
2038          * stay until umount causing failures in tests calculating free space,
2039          * e.g. replay-ost-single. Since those dcache entries are not used
2040          * anymore let's just free them after use here */
2041         shrink_dcache_sb(sb);
2042
2043         RETURN(rc);
2044 }
2045
2046 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2047 {
2048         struct osd_scrub *scrub = &dev->od_scrub;
2049
2050         LASSERT(dev->od_otable_it == NULL);
2051
2052         if (scrub->os_inode != NULL) {
2053                 osd_scrub_stop(dev);
2054                 iput(scrub->os_inode);
2055                 scrub->os_inode = NULL;
2056         }
2057         if (dev->od_oi_table != NULL)
2058                 osd_oi_fini(osd_oti_get(env), dev);
2059 }
2060
2061 /* object table based iteration APIs */
2062
2063 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2064                                        struct dt_object *dt, __u32 attr,
2065                                        struct lustre_capa *capa)
2066 {
2067         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2068         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2069         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2070         struct osd_scrub       *scrub = &dev->od_scrub;
2071         struct osd_otable_it   *it;
2072         __u32                   start = 0;
2073         int                     rc;
2074         ENTRY;
2075
2076         /* od_otable_mutex: prevent curcurrent init/fini */
2077         mutex_lock(&dev->od_otable_mutex);
2078         if (dev->od_otable_it != NULL)
2079                 GOTO(out, it = ERR_PTR(-EALREADY));
2080
2081         OBD_ALLOC_PTR(it);
2082         if (it == NULL)
2083                 GOTO(out, it = ERR_PTR(-ENOMEM));
2084
2085         dev->od_otable_it = it;
2086         it->ooi_dev = dev;
2087         it->ooi_cache.ooc_consumer_idx = -1;
2088         if (flags & DOIF_OUTUSED)
2089                 it->ooi_used_outside = 1;
2090
2091         if (flags & DOIF_RESET)
2092                 start |= SS_RESET;
2093
2094         if (valid & DOIV_ERROR_HANDLE) {
2095                 if (flags & DOIF_FAILOUT)
2096                         start |= SS_SET_FAILOUT;
2097                 else
2098                         start |= SS_CLEAR_FAILOUT;
2099         }
2100
2101         rc = do_osd_scrub_start(dev, start);
2102         if (rc < 0 && rc != -EALREADY) {
2103                 dev->od_otable_it = NULL;
2104                 OBD_FREE_PTR(it);
2105                 GOTO(out, it = ERR_PTR(rc));
2106         }
2107
2108         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2109
2110         GOTO(out, it);
2111
2112 out:
2113         mutex_unlock(&dev->od_otable_mutex);
2114         return (struct dt_it *)it;
2115 }
2116
2117 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2118 {
2119         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2120         struct osd_device    *dev = it->ooi_dev;
2121
2122         /* od_otable_mutex: prevent curcurrent init/fini */
2123         mutex_lock(&dev->od_otable_mutex);
2124         do_osd_scrub_stop(&dev->od_scrub);
2125         LASSERT(dev->od_otable_it == it);
2126
2127         dev->od_otable_it = NULL;
2128         mutex_unlock(&dev->od_otable_mutex);
2129         OBD_FREE_PTR(it);
2130 }
2131
2132 static int osd_otable_it_get(const struct lu_env *env,
2133                              struct dt_it *di, const struct dt_key *key)
2134 {
2135         return 0;
2136 }
2137
2138 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2139 {
2140 }
2141
2142 static inline int
2143 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2144 {
2145         spin_lock(&scrub->os_lock);
2146         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2147             scrub->os_waiting ||
2148             !thread_is_running(&scrub->os_thread))
2149                 it->ooi_waiting = 0;
2150         else
2151                 it->ooi_waiting = 1;
2152         spin_unlock(&scrub->os_lock);
2153
2154         return !it->ooi_waiting;
2155 }
2156
2157 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2158 {
2159         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2160         struct osd_device       *dev    = it->ooi_dev;
2161         struct osd_scrub        *scrub  = &dev->od_scrub;
2162         struct osd_otable_cache *ooc    = &it->ooi_cache;
2163         struct ptlrpc_thread    *thread = &scrub->os_thread;
2164         struct l_wait_info       lwi    = { 0 };
2165         int                      rc;
2166         ENTRY;
2167
2168         LASSERT(it->ooi_user_ready);
2169
2170 again:
2171         if (!thread_is_running(thread) && !it->ooi_used_outside)
2172                 RETURN(1);
2173
2174         if (ooc->ooc_cached_items > 0) {
2175                 ooc->ooc_cached_items--;
2176                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2177                                         ~OSD_OTABLE_IT_CACHE_MASK;
2178                 RETURN(0);
2179         }
2180
2181         if (it->ooi_all_cached) {
2182                 l_wait_event(thread->t_ctl_waitq,
2183                              !thread_is_running(thread),
2184                              &lwi);
2185                 RETURN(1);
2186         }
2187
2188         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2189                 spin_lock(&scrub->os_lock);
2190                 scrub->os_waiting = 0;
2191                 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
2192                 spin_unlock(&scrub->os_lock);
2193         }
2194
2195         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2196                 l_wait_event(thread->t_ctl_waitq,
2197                              osd_otable_it_wakeup(scrub, it),
2198                              &lwi);
2199
2200         if (!thread_is_running(thread) && !it->ooi_used_outside)
2201                 RETURN(1);
2202
2203         rc = osd_otable_it_preload(env, it);
2204         if (rc >= 0)
2205                 goto again;
2206
2207         RETURN(rc);
2208 }
2209
2210 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2211                                         const struct dt_it *di)
2212 {
2213         return NULL;
2214 }
2215
2216 static int osd_otable_it_key_size(const struct lu_env *env,
2217                                   const struct dt_it *di)
2218 {
2219         return sizeof(__u64);
2220 }
2221
2222 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2223                              struct dt_rec *rec, __u32 attr)
2224 {
2225         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2226         struct osd_otable_cache *ooc = &it->ooi_cache;
2227
2228         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2229
2230         /* Filter out Invald FID already. */
2231         LASSERTF(fid_is_sane((struct lu_fid *)rec),
2232                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2233                  PFID((struct lu_fid *)rec),
2234                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2235
2236         return 0;
2237 }
2238
2239 static __u64 osd_otable_it_store(const struct lu_env *env,
2240                                  const struct dt_it *di)
2241 {
2242         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2243         struct osd_otable_cache *ooc = &it->ooi_cache;
2244         __u64                    hash;
2245
2246         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2247                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2248         else
2249                 hash = ooc->ooc_pos_preload;
2250         return hash;
2251 }
2252
2253 /**
2254  * Set the OSD layer iteration start position as the specified hash.
2255  */
2256 static int osd_otable_it_load(const struct lu_env *env,
2257                               const struct dt_it *di, __u64 hash)
2258 {
2259         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
2260         struct osd_device       *dev   = it->ooi_dev;
2261         struct osd_otable_cache *ooc   = &it->ooi_cache;
2262         struct osd_scrub        *scrub = &dev->od_scrub;
2263         int                      rc;
2264         ENTRY;
2265
2266         /* Forbid to set iteration position after iteration started. */
2267         if (it->ooi_user_ready)
2268                 RETURN(-EPERM);
2269
2270         if (hash > OSD_OTABLE_MAX_HASH)
2271                 hash = OSD_OTABLE_MAX_HASH;
2272
2273         ooc->ooc_pos_preload = hash;
2274         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2275                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2276
2277         it->ooi_user_ready = 1;
2278         if (!scrub->os_full_speed)
2279                 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
2280
2281         /* Unplug OSD layer iteration by the first next() call. */
2282         rc = osd_otable_it_next(env, (struct dt_it *)it);
2283
2284         RETURN(rc);
2285 }
2286
2287 static int osd_otable_it_key_rec(const struct lu_env *env,
2288                                  const struct dt_it *di, void *key_rec)
2289 {
2290         return 0;
2291 }
2292
2293 const struct dt_index_operations osd_otable_ops = {
2294         .dio_it = {
2295                 .init     = osd_otable_it_init,
2296                 .fini     = osd_otable_it_fini,
2297                 .get      = osd_otable_it_get,
2298                 .put      = osd_otable_it_put,
2299                 .next     = osd_otable_it_next,
2300                 .key      = osd_otable_it_key,
2301                 .key_size = osd_otable_it_key_size,
2302                 .rec      = osd_otable_it_rec,
2303                 .store    = osd_otable_it_store,
2304                 .load     = osd_otable_it_load,
2305                 .key_rec  = osd_otable_it_key_rec,
2306         }
2307 };
2308
2309 /* high priority inconsistent items list APIs */
2310
2311 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2312                    int insert)
2313 {
2314         struct osd_inconsistent_item *oii;
2315         struct osd_scrub             *scrub  = &dev->od_scrub;
2316         struct ptlrpc_thread         *thread = &scrub->os_thread;
2317         int                           wakeup = 0;
2318         ENTRY;
2319
2320         OBD_ALLOC_PTR(oii);
2321         if (unlikely(oii == NULL))
2322                 RETURN(-ENOMEM);
2323
2324         CFS_INIT_LIST_HEAD(&oii->oii_list);
2325         oii->oii_cache = *oic;
2326         oii->oii_insert = insert;
2327
2328         spin_lock(&scrub->os_lock);
2329         if (unlikely(!thread_is_running(thread))) {
2330                 spin_unlock(&scrub->os_lock);
2331                 OBD_FREE_PTR(oii);
2332                 RETURN(-EAGAIN);
2333         }
2334
2335         if (cfs_list_empty(&scrub->os_inconsistent_items))
2336                 wakeup = 1;
2337         cfs_list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2338         spin_unlock(&scrub->os_lock);
2339
2340         if (wakeup != 0)
2341                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
2342
2343         RETURN(0);
2344 }
2345
2346 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2347                    struct osd_inode_id *id)
2348 {
2349         struct osd_scrub             *scrub = &dev->od_scrub;
2350         struct osd_inconsistent_item *oii;
2351         ENTRY;
2352
2353         spin_lock(&scrub->os_lock);
2354         cfs_list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2355                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2356                         *id = oii->oii_cache.oic_lid;
2357                         spin_unlock(&scrub->os_lock);
2358                         RETURN(0);
2359                 }
2360         }
2361         spin_unlock(&scrub->os_lock);
2362
2363         RETURN(-ENOENT);
2364 }
2365
2366 /* OI scrub dump */
2367
2368 static const char *scrub_status_names[] = {
2369         "init",
2370         "scanning",
2371         "completed",
2372         "failed",
2373         "stopped",
2374         "paused",
2375         "crashed",
2376         NULL
2377 };
2378
2379 static const char *scrub_flags_names[] = {
2380         "recreated",
2381         "inconsistent",
2382         "auto",
2383         "upgrade",
2384         NULL
2385 };
2386
2387 static const char *scrub_param_names[] = {
2388         "failout",
2389         NULL
2390 };
2391
2392 static int scrub_bits_dump(char **buf, int *len, int bits, const char *names[],
2393                            const char *prefix)
2394 {
2395         int save = *len;
2396         int flag;
2397         int rc;
2398         int i;
2399
2400         rc = snprintf(*buf, *len, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2401         if (rc <= 0)
2402                 return -ENOSPC;
2403
2404         *buf += rc;
2405         *len -= rc;
2406         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2407                 if (flag & bits) {
2408                         bits &= ~flag;
2409                         rc = snprintf(*buf, *len, "%s%c", names[i],
2410                                       bits != 0 ? ',' : '\n');
2411                         if (rc <= 0)
2412                                 return -ENOSPC;
2413
2414                         *buf += rc;
2415                         *len -= rc;
2416                 }
2417         }
2418         return save - *len;
2419 }
2420
2421 static int scrub_time_dump(char **buf, int *len, __u64 time, const char *prefix)
2422 {
2423         int rc;
2424
2425         if (time != 0)
2426                 rc = snprintf(*buf, *len, "%s: "LPU64" seconds\n", prefix,
2427                               cfs_time_current_sec() - time);
2428         else
2429                 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
2430         if (rc <= 0)
2431                 return -ENOSPC;
2432
2433         *buf += rc;
2434         *len -= rc;
2435         return rc;
2436 }
2437
2438 static int scrub_pos_dump(char **buf, int *len, __u64 pos, const char *prefix)
2439 {
2440         int rc;
2441
2442         if (pos != 0)
2443                 rc = snprintf(*buf, *len, "%s: "LPU64"\n", prefix, pos);
2444         else
2445                 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
2446         if (rc <= 0)
2447                 return -ENOSPC;
2448
2449         *buf += rc;
2450         *len -= rc;
2451         return rc;
2452 }
2453
2454 int osd_scrub_dump(struct osd_device *dev, char *buf, int len)
2455 {
2456         struct osd_scrub  *scrub   = &dev->od_scrub;
2457         struct scrub_file *sf      = &scrub->os_file;
2458         __u64              checked;
2459         __u64              speed;
2460         int                save    = len;
2461         int                ret     = -ENOSPC;
2462         int                rc;
2463
2464         down_read(&scrub->os_rwsem);
2465         rc = snprintf(buf, len,
2466                       "name: OI_scrub\n"
2467                       "magic: 0x%x\n"
2468                       "oi_files: %d\n"
2469                       "status: %s\n",
2470                       sf->sf_magic, (int)sf->sf_oi_count,
2471                       scrub_status_names[sf->sf_status]);
2472         if (rc <= 0)
2473                 goto out;
2474
2475         buf += rc;
2476         len -= rc;
2477         rc = scrub_bits_dump(&buf, &len, sf->sf_flags, scrub_flags_names,
2478                              "flags");
2479         if (rc < 0)
2480                 goto out;
2481
2482         rc = scrub_bits_dump(&buf, &len, sf->sf_param, scrub_param_names,
2483                              "param");
2484         if (rc < 0)
2485                 goto out;
2486
2487         rc = scrub_time_dump(&buf, &len, sf->sf_time_last_complete,
2488                              "time_since_last_completed");
2489         if (rc < 0)
2490                 goto out;
2491
2492         rc = scrub_time_dump(&buf, &len, sf->sf_time_latest_start,
2493                              "time_since_latest_start");
2494         if (rc < 0)
2495                 goto out;
2496
2497         rc = scrub_time_dump(&buf, &len, sf->sf_time_last_checkpoint,
2498                              "time_since_last_checkpoint");
2499         if (rc < 0)
2500                 goto out;
2501
2502         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_latest_start,
2503                             "latest_start_position");
2504         if (rc < 0)
2505                 goto out;
2506
2507         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_last_checkpoint,
2508                             "last_checkpoint_position");
2509         if (rc < 0)
2510                 goto out;
2511
2512         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_first_inconsistent,
2513                             "first_failure_position");
2514         if (rc < 0)
2515                 goto out;
2516
2517         checked = sf->sf_items_checked + scrub->os_new_checked;
2518         rc = snprintf(buf, len,
2519                       "checked: "LPU64"\n"
2520                       "updated: "LPU64"\n"
2521                       "failed: "LPU64"\n"
2522                       "prior_updated: "LPU64"\n"
2523                       "noscrub: "LPU64"\n"
2524                       "igif: "LPU64"\n"
2525                       "success_count: %u\n",
2526                       checked, sf->sf_items_updated, sf->sf_items_failed,
2527                       sf->sf_items_updated_prior, sf->sf_items_noscrub,
2528                       sf->sf_items_igif, sf->sf_success_count);
2529         if (rc <= 0)
2530                 goto out;
2531
2532         buf += rc;
2533         len -= rc;
2534         speed = checked;
2535         if (thread_is_running(&scrub->os_thread)) {
2536                 cfs_duration_t duration = cfs_time_current() -
2537                                           scrub->os_time_last_checkpoint;
2538                 __u64 new_checked = scrub->os_new_checked * CFS_HZ;
2539                 __u32 rtime = sf->sf_run_time +
2540                               cfs_duration_sec(duration + HALF_SEC);
2541
2542                 if (duration != 0)
2543                         do_div(new_checked, duration);
2544                 if (rtime != 0)
2545                         do_div(speed, rtime);
2546                 rc = snprintf(buf, len,
2547                               "run_time: %u seconds\n"
2548                               "average_speed: "LPU64" objects/sec\n"
2549                               "real-time_speed: "LPU64" objects/sec\n"
2550                               "current_position: %u\n"
2551                               "lf_scanned: "LPU64"\n"
2552                               "lf_reparied: "LPU64"\n"
2553                               "lf_failed: "LPU64"\n",
2554                               rtime, speed, new_checked, scrub->os_pos_current,
2555                               scrub->os_lf_scanned, scrub->os_lf_repaired,
2556                               scrub->os_lf_failed);
2557         } else {
2558                 if (sf->sf_run_time != 0)
2559                         do_div(speed, sf->sf_run_time);
2560                 rc = snprintf(buf, len,
2561                               "run_time: %u seconds\n"
2562                               "average_speed: "LPU64" objects/sec\n"
2563                               "real-time_speed: N/A\n"
2564                               "current_position: N/A\n"
2565                               "lf_scanned: "LPU64"\n"
2566                               "lf_reparied: "LPU64"\n"
2567                               "lf_failed: "LPU64"\n",
2568                               sf->sf_run_time, speed, scrub->os_lf_scanned,
2569                               scrub->os_lf_repaired, scrub->os_lf_failed);
2570         }
2571         if (rc <= 0)
2572                 goto out;
2573
2574         buf += rc;
2575         len -= rc;
2576         ret = save - len;
2577
2578 out:
2579         up_read(&scrub->os_rwsem);
2580         return ret;
2581 }