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, 2014, Intel Corporation.
  24  */
  25 /*
  26  * lustre/osd-ldiskfs/osd_scrub.c
  27  *
  28  * Top-level entry points into osd module
  29  *
  30  * The OI scrub is used for rebuilding Object Index files when restores MDT from
  31  * file-level backup.
  32  *
  33  * The otable based iterator scans ldiskfs inode table to feed up layer LFSCK.
  34  *
  35  * Author: Fan Yong <yong.fan@whamcloud.com>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_LFSCK
  39
  40 #include <linux/kthread.h>
  41 #include <lustre/lustre_idl.h>
  42 #include <lustre_disk.h>
  43 #include <dt_object.h>
  44 #include <linux/xattr.h>
  45
  46 #include "osd_internal.h"
  47 #include "osd_oi.h"
  48 #include "osd_scrub.h"
  49
  50 #define HALF_SEC        msecs_to_jiffies(MSEC_PER_SEC >> 1)
  51
  52 #define OSD_OTABLE_MAX_HASH             0x00000000ffffffffULL
  53
  54 #define SCRUB_NEXT_BREAK        1 /* exit current loop and process next group */
  55 #define SCRUB_NEXT_CONTINUE     2 /* skip current object and process next bit */
  56 #define SCRUB_NEXT_EXIT         3 /* exit all the loops */
  57 #define SCRUB_NEXT_WAIT         4 /* wait for free cache slot */
  58 #define SCRUB_NEXT_CRASH        5 /* simulate system crash during OI scrub */
  59 #define SCRUB_NEXT_FATAL        6 /* simulate failure during OI scrub */
  60 #define SCRUB_NEXT_NOSCRUB      7 /* new created object, no scrub on it */
  61 #define SCRUB_NEXT_NOLMA        8 /* the inode has no FID-in-LMA */
  62 #define SCRUB_NEXT_OSTOBJ       9 /* for OST-object */
  63 #define SCRUB_NEXT_OSTOBJ_OLD   10 /* old OST-object, no LMA or no FID-on-OST
  64                                     * flags in LMA */
  65
  66 /* misc functions */
  67
  68 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
  69 {
  70         return container_of0(scrub, struct osd_device, od_scrub);
  71 }
  72
  73 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
  74 {
  75         return osd_sb(osd_scrub2dev(scrub));
  76 }
  77
  78 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
  79                                        struct osd_otable_cache *ooc)
  80 {
  81         return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
  82 }
  83
  84 static inline const char *osd_scrub2name(struct osd_scrub *scrub)
  85 {
  86         return LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
  87 }
  88
  89 /**
  90  * update/insert/delete the specified OI mapping (@fid @id) according to the ops
  91  *
  92  * \retval   1, changed nothing
  93  * \retval   0, changed successfully
  94  * \retval -ve, on error
  95  */
  96 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
  97                                      struct osd_device *dev,
  98                                      const struct lu_fid *fid,
  99                                      const struct osd_inode_id *id,
 100                                      int ops, bool force,
 101                                      enum oi_check_flags flags)
 102 {
 103         handle_t *th;
 104         int       rc;
 105         ENTRY;
 106
 107         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
 108                 RETURN(0);
 109
 110         /* DTO_INDEX_INSERT is enough for other two ops:
 111          * delete/update, but save stack. */
 112         th = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 113                                 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
 114         if (IS_ERR(th)) {
 115                 rc = PTR_ERR(th);
 116                 CDEBUG(D_LFSCK, "%s: fail to start trans for scrub op %d "
 117                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 118                        PFID(fid), id->oii_ino, id->oii_gen, rc);
 119                 RETURN(rc);
 120         }
 121
 122         switch (ops) {
 123         case DTO_INDEX_UPDATE:
 124                 rc = osd_oi_update(info, dev, fid, id, th, flags);
 125                 if (unlikely(rc == -ENOENT)) {
 126                         /* Some unlink thread may removed the OI mapping. */
 127                         rc = 1;
 128                 }
 129                 break;
 130         case DTO_INDEX_INSERT:
 131                 rc = osd_oi_insert(info, dev, fid, id, th, flags);
 132                 if (unlikely(rc == -EEXIST)) {
 133                         rc = 1;
 134                         /* XXX: There are trouble things when adding OI
 135                          *      mapping for IGIF object, which may cause
 136                          *      multiple objects to be mapped to the same
 137                          *      IGIF formatted FID. Consider the following
 138                          *      situations:
 139                          *
 140                          *      1) The MDT is upgrading from 1.8 device.
 141                          *      The OI scrub generates IGIF FID1 for the
 142                          *      OBJ1 and adds the OI mapping.
 143                          *
 144                          *      2) For some reason, the OI scrub does not
 145                          *      process all the IGIF objects completely.
 146                          *
 147                          *      3) The MDT is backuped and restored against
 148                          *      this device.
 149                          *
 150                          *      4) When the MDT mounts up, the OI scrub will
 151                          *      try to rebuild the OI files. For some IGIF
 152                          *      object, OBJ2, which was not processed by the
 153                          *      OI scrub before the backup/restore, and the
 154                          *      new generated IGIF formatted FID may be just
 155                          *      the FID1, the same as OBJ1.
 156                          *
 157                          *      Under such case, the OI scrub cannot know how
 158                          *      to generate new FID for the OBJ2.
 159                          *
 160                          *      Currently, we do nothing for that. One possible
 161                          *      solution is to generate new normal FID for the
 162                          *      conflict object.
 163                          *
 164                          *      Anyway, it is rare, only exists in theory. */
 165                 }
 166                 break;
 167         case DTO_INDEX_DELETE:
 168                 rc = osd_oi_delete(info, dev, fid, th, flags);
 169                 if (rc == -ENOENT) {
 170                         /* It is normal that the unlink thread has removed the
 171                          * OI mapping already. */
 172                         rc = 1;
 173                 }
 174                 break;
 175         default:
 176                 LASSERTF(0, "Unexpected ops %d\n", ops);
 177                 break;
 178         }
 179
 180         ldiskfs_journal_stop(th);
 181         if (rc < 0)
 182                 CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d "
 183                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 184                        PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1,
 185                        rc);
 186
 187         RETURN(rc);
 188 }
 189
 190 /* OI_scrub file ops */
 191
 192 static void osd_scrub_file_to_cpu(struct scrub_file *des,
 193                                   struct scrub_file *src)
 194 {
 195         memcpy(des->sf_uuid, src->sf_uuid, 16);
 196         des->sf_flags   = le64_to_cpu(src->sf_flags);
 197         des->sf_magic   = le32_to_cpu(src->sf_magic);
 198         des->sf_status  = le16_to_cpu(src->sf_status);
 199         des->sf_param   = le16_to_cpu(src->sf_param);
 200         des->sf_time_last_complete      =
 201                                 le64_to_cpu(src->sf_time_last_complete);
 202         des->sf_time_latest_start       =
 203                                 le64_to_cpu(src->sf_time_latest_start);
 204         des->sf_time_last_checkpoint    =
 205                                 le64_to_cpu(src->sf_time_last_checkpoint);
 206         des->sf_pos_latest_start        =
 207                                 le64_to_cpu(src->sf_pos_latest_start);
 208         des->sf_pos_last_checkpoint     =
 209                                 le64_to_cpu(src->sf_pos_last_checkpoint);
 210         des->sf_pos_first_inconsistent  =
 211                                 le64_to_cpu(src->sf_pos_first_inconsistent);
 212         des->sf_items_checked           =
 213                                 le64_to_cpu(src->sf_items_checked);
 214         des->sf_items_updated           =
 215                                 le64_to_cpu(src->sf_items_updated);
 216         des->sf_items_failed            =
 217                                 le64_to_cpu(src->sf_items_failed);
 218         des->sf_items_updated_prior     =
 219                                 le64_to_cpu(src->sf_items_updated_prior);
 220         des->sf_run_time        = le32_to_cpu(src->sf_run_time);
 221         des->sf_success_count   = le32_to_cpu(src->sf_success_count);
 222         des->sf_oi_count        = le16_to_cpu(src->sf_oi_count);
 223         des->sf_internal_flags  = le16_to_cpu(src->sf_internal_flags);
 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         des->sf_internal_flags  = cpu_to_le16(src->sf_internal_flags);
 259         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 260 }
 261
 262 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
 263 {
 264         struct scrub_file *sf = &scrub->os_file;
 265
 266         memset(sf, 0, sizeof(*sf));
 267         memcpy(sf->sf_uuid, uuid, 16);
 268         sf->sf_magic = SCRUB_MAGIC_V1;
 269         sf->sf_status = SS_INIT;
 270 }
 271
 272 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
 273 {
 274         struct scrub_file *sf = &scrub->os_file;
 275
 276         CDEBUG(D_LFSCK, "%.16s: reset OI scrub file, old flags = "
 277                LPX64", add flags = "LPX64"\n",
 278                osd_scrub2name(scrub), sf->sf_flags, flags);
 279
 280         memcpy(sf->sf_uuid, uuid, 16);
 281         sf->sf_status = SS_INIT;
 282         sf->sf_flags |= flags;
 283         sf->sf_flags &= ~SF_AUTO;
 284         sf->sf_run_time = 0;
 285         sf->sf_time_latest_start = 0;
 286         sf->sf_time_last_checkpoint = 0;
 287         sf->sf_pos_latest_start = 0;
 288         sf->sf_pos_last_checkpoint = 0;
 289         sf->sf_pos_first_inconsistent = 0;
 290         sf->sf_items_checked = 0;
 291         sf->sf_items_updated = 0;
 292         sf->sf_items_failed = 0;
 293         if (!scrub->os_in_join)
 294                 sf->sf_items_updated_prior = 0;
 295
 296         sf->sf_items_noscrub = 0;
 297         sf->sf_items_igif = 0;
 298 }
 299
 300 static int osd_scrub_file_load(struct osd_scrub *scrub)
 301 {
 302         loff_t  pos  = 0;
 303         int     len  = sizeof(scrub->os_file_disk);
 304         int     rc;
 305
 306         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
 307         if (rc == len) {
 308                 struct scrub_file *sf = &scrub->os_file;
 309
 310                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
 311                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
 312                         CDEBUG(D_LFSCK, "%.16s: invalid scrub magic "
 313                                "0x%x != 0x%x\n", osd_scrub2name(scrub),
 314                                sf->sf_magic, SCRUB_MAGIC_V1);
 315                         /* Process it as new scrub file. */
 316                         rc = -ENOENT;
 317                 } else {
 318                         rc = 0;
 319                 }
 320         } else if (rc != 0) {
 321                 CDEBUG(D_LFSCK, "%.16s: fail to load scrub file, "
 322                        "expected = %d: rc = %d\n",
 323                        osd_scrub2name(scrub), len, rc);
 324                 if (rc > 0)
 325                         rc = -EFAULT;
 326         } else {
 327                 /* return -ENOENT for empty scrub file case. */
 328                 rc = -ENOENT;
 329         }
 330
 331         return rc;
 332 }
 333
 334 int osd_scrub_file_store(struct osd_scrub *scrub)
 335 {
 336         struct osd_device *dev;
 337         handle_t          *jh;
 338         loff_t             pos     = 0;
 339         int                len     = sizeof(scrub->os_file_disk);
 340         int                credits;
 341         int                rc;
 342
 343         dev = container_of0(scrub, struct osd_device, od_scrub);
 344         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
 345                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
 346         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits);
 347         if (IS_ERR(jh)) {
 348                 rc = PTR_ERR(jh);
 349                 CDEBUG(D_LFSCK, "%.16s: fail to start trans for scrub store: "
 350                        "rc = %d\n", osd_scrub2name(scrub), rc);
 351                 return rc;
 352         }
 353
 354         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
 355         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
 356                                       len, 0, &pos, jh);
 357         ldiskfs_journal_stop(jh);
 358         if (rc != 0)
 359                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file, "
 360                        "expected = %d: rc = %d\n",
 361                        osd_scrub2name(scrub), len, rc);
 362
 363         scrub->os_time_last_checkpoint = cfs_time_current();
 364         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
 365                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
 366         return rc;
 367 }
 368
 369 static int
 370 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
 371                      struct inode *inode, const struct lu_fid *fid)
 372 {
 373         struct filter_fid_old   *ff      = &info->oti_ff;
 374         struct dentry           *dentry  = &info->oti_obj_dentry;
 375         struct lu_fid           *tfid    = &info->oti_fid;
 376         handle_t                *jh;
 377         int                      size    = 0;
 378         int                      rc;
 379         bool                     removed = false;
 380         bool                     reset   = true;
 381         ENTRY;
 382
 383         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
 384                 RETURN(0);
 385
 386         if (fid_is_idif(fid) && dev->od_index_in_idif == 0) {
 387                 struct ost_id *oi = &info->oti_ostid;
 388
 389                 fid_to_ostid(fid, oi);
 390                 ostid_to_fid(tfid, oi, 0);
 391         } else {
 392                 *tfid = *fid;
 393         }
 394
 395         /* We want the LMA to fit into the 256-byte OST inode, so operate
 396          * as following:
 397          * 1) read old XATTR_NAME_FID and save the parent FID;
 398          * 2) delete the old XATTR_NAME_FID;
 399          * 3) make new LMA and add it;
 400          * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
 401          *
 402          * Making the LMA to fit into the 256-byte OST inode can save time for
 403          * normal osd_check_lma() and for other OI scrub scanning in future.
 404          * So it is worth to make some slow conversion here. */
 405         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 406                                 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
 407         if (IS_ERR(jh)) {
 408                 rc = PTR_ERR(jh);
 409                 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
 410                        DFID": rc = %d\n", osd_name(dev), PFID(tfid), rc);
 411                 RETURN(rc);
 412         }
 413
 414         /* 1) read old XATTR_NAME_FID and save the parent FID */
 415         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 416         if (rc == sizeof(*ff)) {
 417                 /* 2) delete the old XATTR_NAME_FID */
 418                 ll_vfs_dq_init(inode);
 419                 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
 420                 if (rc != 0)
 421                         GOTO(stop, rc);
 422
 423                 removed = true;
 424         } else if (unlikely(rc == -ENODATA)) {
 425                 reset = false;
 426         } else if (rc != sizeof(struct filter_fid)) {
 427                 GOTO(stop, rc = -EINVAL);
 428         }
 429
 430         /* 3) make new LMA and add it */
 431         rc = osd_ea_fid_set(info, inode, tfid, LMAC_FID_ON_OST, 0);
 432         if (rc == 0 && reset)
 433                 size = sizeof(struct filter_fid);
 434         else if (rc != 0 && removed)
 435                 /* If failed, we should try to add the old back. */
 436                 size = sizeof(struct filter_fid_old);
 437
 438         /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
 439         if (size > 0) {
 440                 int rc1;
 441
 442                 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
 443                                       XATTR_CREATE);
 444                 if (rc1 != 0 && rc == 0)
 445                         rc = rc1;
 446         }
 447
 448         GOTO(stop, rc);
 449
 450 stop:
 451         ldiskfs_journal_stop(jh);
 452         if (rc < 0)
 453                 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
 454                        osd_name(dev), PFID(tfid), rc);
 455         return rc;
 456 }
 457
 458 static int
 459 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
 460                        struct osd_idmap_cache *oic, int val)
 461 {
 462         struct osd_scrub             *scrub  = &dev->od_scrub;
 463         struct scrub_file            *sf     = &scrub->os_file;
 464         struct lu_fid                *fid    = &oic->oic_fid;
 465         struct osd_inode_id          *lid    = &oic->oic_lid;
 466         struct osd_inode_id          *lid2   = &info->oti_id;
 467         struct osd_inconsistent_item *oii    = NULL;
 468         struct inode                 *inode  = NULL;
 469         int                           ops    = DTO_INDEX_UPDATE;
 470         int                           idx;
 471         int                           rc;
 472         bool                          converted = false;
 473         ENTRY;
 474
 475         down_write(&scrub->os_rwsem);
 476         scrub->os_new_checked++;
 477         if (val < 0)
 478                 GOTO(out, rc = val);
 479
 480         if (scrub->os_in_prior)
 481                 oii = list_entry(oic, struct osd_inconsistent_item,
 482                                  oii_cache);
 483
 484         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
 485                 GOTO(out, rc = 0);
 486
 487         if (fid_is_igif(fid))
 488                 sf->sf_items_igif++;
 489
 490         if (val == SCRUB_NEXT_OSTOBJ_OLD) {
 491                 inode = osd_iget(info, dev, lid);
 492                 if (IS_ERR(inode)) {
 493                         rc = PTR_ERR(inode);
 494                         /* Someone removed the inode. */
 495                         if (rc == -ENOENT || rc == -ESTALE)
 496                                 rc = 0;
 497                         GOTO(out, rc);
 498                 }
 499
 500                 sf->sf_flags |= SF_UPGRADE;
 501                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
 502                 dev->od_check_ff = 1;
 503                 rc = osd_scrub_convert_ff(info, dev, inode, fid);
 504                 if (rc != 0)
 505                         GOTO(out, rc);
 506
 507                 converted = true;
 508         }
 509
 510         if ((val == SCRUB_NEXT_NOLMA) &&
 511             (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
 512                 GOTO(out, rc = 0);
 513
 514         if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA)) {
 515                 ops = DTO_INDEX_INSERT;
 516
 517                 goto iget;
 518         }
 519
 520         rc = osd_oi_lookup(info, dev, fid, lid2,
 521                 (val == SCRUB_NEXT_OSTOBJ ||
 522                  val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 523         if (rc != 0) {
 524                 if (rc == -ENOENT)
 525                         ops = DTO_INDEX_INSERT;
 526                 else if (rc != -ESTALE)
 527                         GOTO(out, rc);
 528
 529 iget:
 530                 if (inode == NULL) {
 531                         inode = osd_iget(info, dev, lid);
 532                         if (IS_ERR(inode)) {
 533                                 rc = PTR_ERR(inode);
 534                                 /* Someone removed the inode. */
 535                                 if (rc == -ENOENT || rc == -ESTALE)
 536                                         rc = 0;
 537                                 GOTO(out, rc);
 538                         }
 539                 }
 540
 541                 if (!scrub->os_partial_scan)
 542                         scrub->os_full_speed = 1;
 543
 544                 idx = osd_oi_fid2idx(dev, fid);
 545                 switch (val) {
 546                 case SCRUB_NEXT_NOLMA:
 547                         sf->sf_flags |= SF_UPGRADE;
 548                         if (!(sf->sf_param & SP_DRYRUN)) {
 549                                 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
 550                                 if (rc != 0)
 551                                         GOTO(out, rc);
 552                         }
 553
 554                         if (!(sf->sf_flags & SF_INCONSISTENT))
 555                                 dev->od_igif_inoi = 0;
 556                         break;
 557                 case SCRUB_NEXT_OSTOBJ:
 558                         sf->sf_flags |= SF_INCONSISTENT;
 559                 case SCRUB_NEXT_OSTOBJ_OLD:
 560                         break;
 561                 default:
 562                         sf->sf_flags |= SF_RECREATED;
 563                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 564                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 565                         break;
 566                 }
 567         } else if (osd_id_eq(lid, lid2)) {
 568                 if (converted)
 569                         sf->sf_items_updated++;
 570
 571                 GOTO(out, rc = 0);
 572         } else {
 573                 if (!scrub->os_partial_scan)
 574                         scrub->os_full_speed = 1;
 575
 576                 sf->sf_flags |= SF_INCONSISTENT;
 577
 578                 /* XXX: If the device is restored from file-level backup, then
 579                  *      some IGIFs may have been already in OI files, and some
 580                  *      may be not yet. Means upgrading from 1.8 may be partly
 581                  *      processed, but some clients may hold some immobilized
 582                  *      IGIFs, and use them to access related objects. Under
 583                  *      such case, OSD does not know whether an given IGIF has
 584                  *      been processed or to be processed, and it also cannot
 585                  *      generate local ino#/gen# directly from the immobilized
 586                  *      IGIF because of the backup/restore. Then force OSD to
 587                  *      lookup the given IGIF in OI files, and if no entry,
 588                  *      then ask the client to retry after upgrading completed.
 589                  *      No better choice. */
 590                 dev->od_igif_inoi = 1;
 591         }
 592
 593         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
 594                         (val == SCRUB_NEXT_OSTOBJ ||
 595                          val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 596         if (rc == 0) {
 597                 if (scrub->os_in_prior)
 598                         sf->sf_items_updated_prior++;
 599                 else
 600                         sf->sf_items_updated++;
 601         }
 602
 603         GOTO(out, rc);
 604
 605 out:
 606         if (rc < 0) {
 607                 sf->sf_items_failed++;
 608                 if (sf->sf_pos_first_inconsistent == 0 ||
 609                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 610                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 611         } else {
 612                 rc = 0;
 613         }
 614
 615         /* There may be conflict unlink during the OI scrub,
 616          * if happend, then remove the new added OI mapping. */
 617         if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
 618             unlikely(inode->i_nlink == 0))
 619                 osd_scrub_refresh_mapping(info, dev, fid, lid,
 620                                 DTO_INDEX_DELETE, false,
 621                                 (val == SCRUB_NEXT_OSTOBJ ||
 622                                  val == SCRUB_NEXT_OSTOBJ_OLD) ?
 623                                 OI_KNOWN_ON_OST : 0);
 624         up_write(&scrub->os_rwsem);
 625
 626         if (inode != NULL && !IS_ERR(inode))
 627                 iput(inode);
 628
 629         if (oii != NULL) {
 630                 LASSERT(!list_empty(&oii->oii_list));
 631
 632                 spin_lock(&scrub->os_lock);
 633                 list_del_init(&oii->oii_list);
 634                 spin_unlock(&scrub->os_lock);
 635                 OBD_FREE_PTR(oii);
 636         }
 637         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 638 }
 639
 640 /* OI scrub APIs */
 641
 642 static int osd_scrub_prep(struct osd_device *dev)
 643 {
 644         struct osd_scrub     *scrub  = &dev->od_scrub;
 645         struct ptlrpc_thread *thread = &scrub->os_thread;
 646         struct scrub_file    *sf     = &scrub->os_file;
 647         __u32                 flags  = scrub->os_start_flags;
 648         int                   rc;
 649         bool                  drop_dryrun = false;
 650         ENTRY;
 651
 652         CDEBUG(D_LFSCK, "%.16s: OI scrub prep, flags = 0x%x\n",
 653                osd_scrub2name(scrub), flags);
 654
 655         down_write(&scrub->os_rwsem);
 656         if (flags & SS_SET_FAILOUT)
 657                 sf->sf_param |= SP_FAILOUT;
 658         else if (flags & SS_CLEAR_FAILOUT)
 659                 sf->sf_param &= ~SP_FAILOUT;
 660
 661         if (flags & SS_SET_DRYRUN) {
 662                 sf->sf_param |= SP_DRYRUN;
 663         } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
 664                 sf->sf_param &= ~SP_DRYRUN;
 665                 drop_dryrun = true;
 666         }
 667
 668         if (flags & SS_RESET)
 669                 osd_scrub_file_reset(scrub,
 670                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
 671
 672         if (flags & SS_AUTO_FULL) {
 673                 scrub->os_full_speed = 1;
 674                 scrub->os_partial_scan = 0;
 675                 sf->sf_flags |= SF_AUTO;
 676         } else if (flags & SS_AUTO_PARTIAL) {
 677                 scrub->os_full_speed = 0;
 678                 scrub->os_partial_scan = 1;
 679                 sf->sf_flags |= SF_AUTO;
 680         } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
 681                                    SF_UPGRADE)) {
 682                 scrub->os_full_speed = 1;
 683                 scrub->os_partial_scan = 0;
 684         } else {
 685                 scrub->os_full_speed = 0;
 686                 scrub->os_partial_scan = 0;
 687         }
 688
 689         spin_lock(&scrub->os_lock);
 690         scrub->os_in_prior = 0;
 691         scrub->os_waiting = 0;
 692         scrub->os_paused = 0;
 693         scrub->os_in_join = 0;
 694         scrub->os_full_scrub = 0;
 695         spin_unlock(&scrub->os_lock);
 696         scrub->os_new_checked = 0;
 697         if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
 698                 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
 699         else if (sf->sf_pos_last_checkpoint != 0)
 700                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 701         else
 702                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 703
 704         scrub->os_pos_current = sf->sf_pos_latest_start;
 705         sf->sf_status = SS_SCANNING;
 706         sf->sf_time_latest_start = cfs_time_current_sec();
 707         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 708         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
 709         rc = osd_scrub_file_store(scrub);
 710         if (rc == 0) {
 711                 spin_lock(&scrub->os_lock);
 712                 thread_set_flags(thread, SVC_RUNNING);
 713                 spin_unlock(&scrub->os_lock);
 714                 wake_up_all(&thread->t_ctl_waitq);
 715         }
 716         up_write(&scrub->os_rwsem);
 717
 718         RETURN(rc);
 719 }
 720
 721 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
 722 {
 723         struct scrub_file *sf = &scrub->os_file;
 724         int                rc;
 725
 726         if (likely(cfs_time_before(cfs_time_current(),
 727                                    scrub->os_time_next_checkpoint) ||
 728                    scrub->os_new_checked == 0))
 729                 return 0;
 730
 731         down_write(&scrub->os_rwsem);
 732         sf->sf_items_checked += scrub->os_new_checked;
 733         scrub->os_new_checked = 0;
 734         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 735         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 736         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 737                                             scrub->os_time_last_checkpoint);
 738         rc = osd_scrub_file_store(scrub);
 739         up_write(&scrub->os_rwsem);
 740
 741         return rc;
 742 }
 743
 744 static void osd_scrub_post(struct osd_scrub *scrub, int result)
 745 {
 746         struct scrub_file *sf = &scrub->os_file;
 747         ENTRY;
 748
 749         CDEBUG(D_LFSCK, "%.16s: OI scrub post, result = %d\n",
 750                osd_scrub2name(scrub), result);
 751
 752         down_write(&scrub->os_rwsem);
 753         spin_lock(&scrub->os_lock);
 754         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 755         spin_unlock(&scrub->os_lock);
 756         if (scrub->os_new_checked > 0) {
 757                 sf->sf_items_checked += scrub->os_new_checked;
 758                 scrub->os_new_checked = 0;
 759                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 760         }
 761         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 762         if (result > 0) {
 763                 struct osd_device *dev =
 764                         container_of0(scrub, struct osd_device, od_scrub);
 765
 766                 dev->od_igif_inoi = 1;
 767                 dev->od_check_ff = 0;
 768                 sf->sf_status = SS_COMPLETED;
 769                 if (!(sf->sf_param & SP_DRYRUN)) {
 770                         memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 771                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 772                                           SF_UPGRADE | SF_AUTO);
 773                 }
 774                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 775                 sf->sf_success_count++;
 776         } else if (result == 0) {
 777                 if (scrub->os_paused)
 778                         sf->sf_status = SS_PAUSED;
 779                 else
 780                         sf->sf_status = SS_STOPPED;
 781         } else {
 782                 sf->sf_status = SS_FAILED;
 783         }
 784         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 785                                             scrub->os_time_last_checkpoint);
 786         result = osd_scrub_file_store(scrub);
 787         up_write(&scrub->os_rwsem);
 788
 789         EXIT;
 790 }
 791
 792 /* iteration engine */
 793
 794 struct osd_iit_param {
 795         struct super_block *sb;
 796         struct buffer_head *bitmap;
 797         ldiskfs_group_t bg;
 798         __u32 gbase;
 799         __u32 offset;
 800 };
 801
 802 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 803                                    struct osd_device *dev,
 804                                    struct osd_iit_param *param,
 805                                    struct osd_idmap_cache **oic,
 806                                    const bool noslot);
 807
 808 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 809                                    struct osd_device *dev,
 810                                    struct osd_iit_param *param,
 811                                    struct osd_idmap_cache *oic,
 812                                    bool *noslot, int rc);
 813
 814 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 815 {
 816         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 817                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 818         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 819                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 820                 return SCRUB_NEXT_BREAK;
 821         } else {
 822                 *pos = param->gbase + param->offset;
 823                 return 0;
 824         }
 825 }
 826
 827 /**
 828  * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
 829  * \retval 0: FID-on-MDT
 830  */
 831 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
 832                                       struct osd_device *dev,
 833                                       struct lu_fid *fid)
 834 {
 835         /* XXX: The initial OI scrub will scan the top level /O to generate
 836          *      a small local FLDB according to the <seq>. If the given FID
 837          *      is in the local FLDB, then it is FID-on-OST; otherwise it's
 838          *      quite possible for FID-on-MDT. */
 839         if (dev->od_is_ost)
 840                 return SCRUB_NEXT_OSTOBJ_OLD;
 841         else
 842                 return 0;
 843 }
 844
 845 static int osd_scrub_get_fid(struct osd_thread_info *info,
 846                              struct osd_device *dev, struct inode *inode,
 847                              struct lu_fid *fid, bool scrub)
 848 {
 849         struct lustre_mdt_attrs *lma     = &info->oti_mdt_attrs;
 850         int                      rc;
 851         bool                     has_lma = false;
 852
 853         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 854         if (rc == 0) {
 855                 has_lma = true;
 856                 if (lma->lma_compat & LMAC_NOT_IN_OI ||
 857                     lma->lma_incompat & LMAI_AGENT)
 858                         return SCRUB_NEXT_CONTINUE;
 859
 860                 *fid = lma->lma_self_fid;
 861                 if (!scrub)
 862                         return 0;
 863
 864                 if (lma->lma_compat & LMAC_FID_ON_OST)
 865                         return SCRUB_NEXT_OSTOBJ;
 866
 867                 if (fid_is_idif(fid))
 868                         return SCRUB_NEXT_OSTOBJ_OLD;
 869
 870                 /* For local object. */
 871                 if (fid_is_internal(fid))
 872                         return 0;
 873
 874                 /* For external visible MDT-object with non-normal FID. */
 875                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
 876                         return 0;
 877
 878                 /* For the object with normal FID, it may be MDT-object,
 879                  * or may be 2.4 OST-object, need further distinguish.
 880                  * Fall through to next section. */
 881         }
 882
 883         if (rc == -ENODATA || rc == 0) {
 884                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
 885                 if (rc == 0) {
 886                         if (scrub)
 887                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
 888                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 889                         return rc;
 890                 }
 891
 892                 if (rc > 0) {
 893                         if (!has_lma)
 894                                 /* It is FID-on-OST, but we do not know how
 895                                  * to generate its FID, ignore it directly. */
 896                                 rc = SCRUB_NEXT_CONTINUE;
 897                         else
 898                                 /* It is 2.4 OST-object. */
 899                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 900                         return rc;
 901                 }
 902
 903                 if (rc != -ENODATA)
 904                         return rc;
 905
 906                 if (!has_lma) {
 907                         if (dev->od_scrub.os_convert_igif) {
 908                                 lu_igif_build(fid, inode->i_ino,
 909                                               inode->i_generation);
 910                                 if (scrub)
 911                                         rc = SCRUB_NEXT_NOLMA;
 912                                 else
 913                                         rc = 0;
 914                         } else {
 915                                 /* It may be FID-on-OST, or may be FID for
 916                                  * non-MDT0, anyway, we do not know how to
 917                                  * generate its FID, ignore it directly. */
 918                                 rc = SCRUB_NEXT_CONTINUE;
 919                         }
 920                         return rc;
 921                 }
 922
 923                 /* For OI scrub case only: the object has LMA but has no ff
 924                  * (or ff crashed). It may be MDT-object, may be OST-object
 925                  * with crashed ff. The last check is local FLDB. */
 926                 rc = osd_scrub_check_local_fldb(info, dev, fid);
 927         }
 928
 929         return rc;
 930 }
 931
 932 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 933                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 934                         struct super_block *sb, bool scrub)
 935 {
 936         struct inode *inode;
 937         int           rc;
 938         ENTRY;
 939
 940         /* Not handle the backend root object and agent parent object.
 941          * They are neither visible to namespace nor have OI mappings. */
 942         if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
 943                      pos == osd_remote_parent_ino(dev)))
 944                 RETURN(SCRUB_NEXT_CONTINUE);
 945
 946         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 947         inode = osd_iget(info, dev, lid);
 948         if (IS_ERR(inode)) {
 949                 rc = PTR_ERR(inode);
 950                 /* The inode may be removed after bitmap searching, or the
 951                  * file is new created without inode initialized yet. */
 952                 if (rc == -ENOENT || rc == -ESTALE)
 953                         RETURN(SCRUB_NEXT_CONTINUE);
 954
 955                 CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: "
 956                        "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name,
 957                        pos, rc);
 958                 RETURN(rc);
 959         }
 960
 961         if (scrub &&
 962             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 963                 /* Only skip it for the first OI scrub accessing. */
 964                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 965                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
 966         }
 967
 968         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
 969
 970         GOTO(put, rc);
 971
 972 put:
 973         iput(inode);
 974         return rc;
 975 }
 976
 977 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 978                           struct osd_iit_param *param,
 979                           struct osd_idmap_cache **oic, const bool noslot)
 980 {
 981         struct osd_scrub     *scrub  = &dev->od_scrub;
 982         struct ptlrpc_thread *thread = &scrub->os_thread;
 983         struct lu_fid        *fid;
 984         struct osd_inode_id  *lid;
 985         int                   rc;
 986
 987         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
 988                 struct l_wait_info lwi;
 989
 990                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
 991                 if (likely(lwi.lwi_timeout > 0))
 992                         l_wait_event(thread->t_ctl_waitq,
 993                                 !list_empty(&scrub->os_inconsistent_items) ||
 994                                 !thread_is_running(thread),
 995                                 &lwi);
 996         }
 997
 998         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 999                 spin_lock(&scrub->os_lock);
1000                 thread_set_flags(thread, SVC_STOPPING);
1001                 spin_unlock(&scrub->os_lock);
1002                 return SCRUB_NEXT_CRASH;
1003         }
1004
1005         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
1006                 return SCRUB_NEXT_FATAL;
1007
1008         if (unlikely(!thread_is_running(thread)))
1009                 return SCRUB_NEXT_EXIT;
1010
1011         if (!list_empty(&scrub->os_inconsistent_items)) {
1012                 struct osd_inconsistent_item *oii;
1013
1014                 oii = list_entry(scrub->os_inconsistent_items.next,
1015                                  struct osd_inconsistent_item, oii_list);
1016                 *oic = &oii->oii_cache;
1017                 scrub->os_in_prior = 1;
1018                 return 0;
1019         }
1020
1021         if (noslot)
1022                 return SCRUB_NEXT_WAIT;
1023
1024         rc = osd_iit_next(param, &scrub->os_pos_current);
1025         if (rc != 0)
1026                 return rc;
1027
1028         *oic = &scrub->os_oic;
1029         fid = &(*oic)->oic_fid;
1030         lid = &(*oic)->oic_lid;
1031         rc = osd_iit_iget(info, dev, fid, lid,
1032                           scrub->os_pos_current, param->sb, true);
1033         return rc;
1034 }
1035
1036 static int osd_preload_next(struct osd_thread_info *info,
1037                             struct osd_device *dev, struct osd_iit_param *param,
1038                             struct osd_idmap_cache **oic, const bool noslot)
1039 {
1040         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
1041         struct osd_scrub        *scrub;
1042         struct ptlrpc_thread    *thread;
1043         int                      rc;
1044
1045         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1046         if (rc != 0)
1047                 return rc;
1048
1049         scrub = &dev->od_scrub;
1050         thread = &scrub->os_thread;
1051         if (thread_is_running(thread) &&
1052             ooc->ooc_pos_preload >= scrub->os_pos_current)
1053                 return SCRUB_NEXT_EXIT;
1054
1055         rc = osd_iit_iget(info, dev,
1056                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1057                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1058                           ooc->ooc_pos_preload, param->sb, false);
1059         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1060          * ignore the failure, so it still need to skip the inode next time. */
1061         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1062         return rc;
1063 }
1064
1065 static inline int
1066 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1067 {
1068         spin_lock(&scrub->os_lock);
1069         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1070             !list_empty(&scrub->os_inconsistent_items) ||
1071             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1072                 scrub->os_waiting = 0;
1073         else
1074                 scrub->os_waiting = 1;
1075         spin_unlock(&scrub->os_lock);
1076
1077         return !scrub->os_waiting;
1078 }
1079
1080 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1081                           struct osd_iit_param *param,
1082                           struct osd_idmap_cache *oic, bool *noslot, int rc)
1083 {
1084         struct l_wait_info       lwi    = { 0 };
1085         struct osd_scrub        *scrub  = &dev->od_scrub;
1086         struct scrub_file       *sf     = &scrub->os_file;
1087         struct ptlrpc_thread    *thread = &scrub->os_thread;
1088         struct osd_otable_it    *it     = dev->od_otable_it;
1089         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
1090
1091         switch (rc) {
1092         case SCRUB_NEXT_CONTINUE:
1093                 goto next;
1094         case SCRUB_NEXT_WAIT:
1095                 goto wait;
1096         case SCRUB_NEXT_NOSCRUB:
1097                 down_write(&scrub->os_rwsem);
1098                 scrub->os_new_checked++;
1099                 sf->sf_items_noscrub++;
1100                 up_write(&scrub->os_rwsem);
1101                 goto next;
1102         }
1103
1104         rc = osd_scrub_check_update(info, dev, oic, rc);
1105         if (rc != 0)
1106                 return rc;
1107
1108         rc = osd_scrub_checkpoint(scrub);
1109         if (rc != 0) {
1110                 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1111                        "rc = %d\n", osd_scrub2name(scrub),
1112                        scrub->os_pos_current, rc);
1113                 /* Continue, as long as the scrub itself can go ahead. */
1114         }
1115
1116         if (scrub->os_in_prior) {
1117                 scrub->os_in_prior = 0;
1118                 return 0;
1119         }
1120
1121 next:
1122         scrub->os_pos_current = param->gbase + ++(param->offset);
1123
1124 wait:
1125         if (it != NULL && it->ooi_waiting && ooc != NULL &&
1126             ooc->ooc_pos_preload < scrub->os_pos_current) {
1127                 spin_lock(&scrub->os_lock);
1128                 it->ooi_waiting = 0;
1129                 wake_up_all(&thread->t_ctl_waitq);
1130                 spin_unlock(&scrub->os_lock);
1131         }
1132
1133         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1134                 return 0;
1135
1136         if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1137                 *noslot = false;
1138                 return 0;
1139         }
1140
1141         if (it != NULL)
1142                 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1143                              &lwi);
1144
1145         if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1146                 *noslot = false;
1147         else
1148                 *noslot = true;
1149         return 0;
1150 }
1151
1152 static int osd_preload_exec(struct osd_thread_info *info,
1153                             struct osd_device *dev, struct osd_iit_param *param,
1154                             struct osd_idmap_cache *oic, bool *noslot, int rc)
1155 {
1156         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1157
1158         if (rc == 0) {
1159                 ooc->ooc_cached_items++;
1160                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1161                                         ~OSD_OTABLE_IT_CACHE_MASK;
1162         }
1163         return rc > 0 ? 0 : rc;
1164 }
1165
1166 #define SCRUB_IT_ALL    1
1167 #define SCRUB_IT_CRASH  2
1168
1169 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1170                            bool inconsistent)
1171 {
1172         struct osd_scrub     *scrub  = &dev->od_scrub;
1173         struct ptlrpc_thread *thread = &scrub->os_thread;
1174         struct scrub_file    *sf     = &scrub->os_file;
1175         int                   rc;
1176         ENTRY;
1177
1178         LASSERT(!(flags & SS_AUTO_PARTIAL));
1179
1180         down_write(&scrub->os_rwsem);
1181         scrub->os_in_join = 1;
1182         if (flags & SS_SET_FAILOUT)
1183                 sf->sf_param |= SP_FAILOUT;
1184         else if (flags & SS_CLEAR_FAILOUT)
1185                 sf->sf_param &= ~SP_FAILOUT;
1186
1187         if (flags & SS_SET_DRYRUN)
1188                 sf->sf_param |= SP_DRYRUN;
1189         else if (flags & SS_CLEAR_DRYRUN)
1190                 sf->sf_param &= ~SP_DRYRUN;
1191
1192         if (flags & SS_RESET) {
1193                 osd_scrub_file_reset(scrub,
1194                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1195                         inconsistent ? SF_INCONSISTENT : 0);
1196                 sf->sf_status = SS_SCANNING;
1197         }
1198
1199         if (flags & SS_AUTO_FULL) {
1200                 sf->sf_flags |= SF_AUTO;
1201                 scrub->os_full_speed = 1;
1202         }
1203
1204         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1205                 scrub->os_full_speed = 1;
1206         else
1207                 scrub->os_full_speed = 0;
1208
1209         scrub->os_new_checked = 0;
1210         if (sf->sf_pos_last_checkpoint != 0)
1211                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1212         else
1213                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1214
1215         scrub->os_pos_current = sf->sf_pos_latest_start;
1216         sf->sf_time_latest_start = cfs_time_current_sec();
1217         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1218         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1219         rc = osd_scrub_file_store(scrub);
1220         if (rc != 0)
1221                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
1222                        "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1223
1224         spin_lock(&scrub->os_lock);
1225         scrub->os_waiting = 0;
1226         scrub->os_paused = 0;
1227         scrub->os_partial_scan = 0;
1228         scrub->os_in_join = 0;
1229         scrub->os_full_scrub = 0;
1230         spin_unlock(&scrub->os_lock);
1231         wake_up_all(&thread->t_ctl_waitq);
1232         up_write(&scrub->os_rwsem);
1233
1234         EXIT;
1235 }
1236
1237 static int osd_inode_iteration(struct osd_thread_info *info,
1238                                struct osd_device *dev, __u32 max, bool preload)
1239 {
1240         struct osd_scrub     *scrub  = &dev->od_scrub;
1241         struct ptlrpc_thread *thread = &scrub->os_thread;
1242         struct scrub_file    *sf     = &scrub->os_file;
1243         osd_iit_next_policy   next;
1244         osd_iit_exec_policy   exec;
1245         __u32                *pos;
1246         __u32                *count;
1247         struct osd_iit_param  param  = { NULL };
1248         struct l_wait_info    lwi    = { 0 };
1249         __u32                 limit;
1250         int                   rc;
1251         bool                  noslot = true;
1252         ENTRY;
1253
1254         param.sb = osd_sb(dev);
1255         if (preload)
1256                 goto full;
1257
1258         while (scrub->os_partial_scan && !scrub->os_in_join) {
1259                 struct osd_idmap_cache *oic = NULL;
1260
1261                 rc = osd_scrub_next(info, dev, &param, &oic, noslot);
1262                 switch (rc) {
1263                 case SCRUB_NEXT_EXIT:
1264                         RETURN(0);
1265                 case SCRUB_NEXT_CRASH:
1266                         RETURN(SCRUB_IT_CRASH);
1267                 case SCRUB_NEXT_FATAL:
1268                         RETURN(-EINVAL);
1269                 case SCRUB_NEXT_WAIT: {
1270                         struct kstatfs *ksfs = &info->oti_ksfs;
1271                         __u64 saved_flags;
1272
1273                         if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1274                             unlikely(sf->sf_items_updated_prior == 0))
1275                                 goto wait;
1276
1277                         if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1278                             scrub->os_full_scrub) {
1279                                 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1280                                                true);
1281                                 goto full;
1282                         }
1283
1284                         rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1285                         if (rc == 0) {
1286                                 __u64 used = ksfs->f_files - ksfs->f_ffree;
1287
1288                                 do_div(used, sf->sf_items_updated_prior);
1289                                 /* If we hit too much inconsistent OI
1290                                  * mappings during the partial scan,
1291                                  * then scan the device completely. */
1292                                 if (used < dev->od_full_scrub_ratio) {
1293                                         osd_scrub_join(dev,
1294                                                 SS_AUTO_FULL | SS_RESET, true);
1295                                         goto full;
1296                                 }
1297                         }
1298
1299 wait:
1300                         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1301                             cfs_fail_val > 0)
1302                                 continue;
1303
1304                         saved_flags = sf->sf_flags;
1305                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1306                                           SF_UPGRADE | SF_AUTO);
1307                         sf->sf_status = SS_COMPLETED;
1308                         l_wait_event(thread->t_ctl_waitq,
1309                                      !thread_is_running(thread) ||
1310                                      !scrub->os_partial_scan ||
1311                                      scrub->os_in_join ||
1312                                      !list_empty(&scrub->os_inconsistent_items),
1313                                      &lwi);
1314                         sf->sf_flags = saved_flags;
1315                         sf->sf_status = SS_SCANNING;
1316
1317                         if (unlikely(!thread_is_running(thread)))
1318                                 RETURN(0);
1319
1320                         if (!scrub->os_partial_scan || scrub->os_in_join)
1321                                 goto full;
1322
1323                         continue;
1324                 }
1325                 default:
1326                         LASSERTF(rc == 0, "rc = %d\n", rc);
1327
1328                         osd_scrub_exec(info, dev, &param, oic, &noslot, rc);
1329                         break;
1330                 }
1331         }
1332
1333 full:
1334         if (!preload) {
1335                 l_wait_event(thread->t_ctl_waitq,
1336                              !thread_is_running(thread) || !scrub->os_in_join,
1337                              &lwi);
1338
1339                 if (unlikely(!thread_is_running(thread)))
1340                         RETURN(0);
1341         }
1342
1343         noslot = false;
1344         if (!preload) {
1345                 next = osd_scrub_next;
1346                 exec = osd_scrub_exec;
1347                 pos = &scrub->os_pos_current;
1348                 count = &scrub->os_new_checked;
1349         } else {
1350                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1351
1352                 next = osd_preload_next;
1353                 exec = osd_preload_exec;
1354                 pos = &ooc->ooc_pos_preload;
1355                 count = &ooc->ooc_cached_items;
1356         }
1357         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1358
1359         while (*pos <= limit && *count < max) {
1360                 struct osd_idmap_cache *oic = NULL;
1361                 struct ldiskfs_group_desc *desc;
1362
1363                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1364                 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1365                 if (desc == NULL)
1366                         RETURN(-EIO);
1367
1368                 ldiskfs_lock_group(param.sb, param.bg);
1369                 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1370                         ldiskfs_unlock_group(param.sb, param.bg);
1371                         *pos = 1 + (param.bg + 1) *
1372                                 LDISKFS_INODES_PER_GROUP(param.sb);
1373                         continue;
1374                 }
1375                 ldiskfs_unlock_group(param.sb, param.bg);
1376
1377                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1378                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1379                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1380                 if (param.bitmap == NULL) {
1381                         CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1382                                "scrub will stop, urgent mode\n",
1383                                osd_scrub2name(scrub), (__u32)param.bg);
1384                         RETURN(-EIO);
1385                 }
1386
1387                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1388                        *count < max) {
1389                         if (param.offset +
1390                                 ldiskfs_itable_unused_count(param.sb, desc) >
1391                             LDISKFS_INODES_PER_GROUP(param.sb))
1392                                 goto next_group;
1393
1394                         rc = next(info, dev, &param, &oic, noslot);
1395                         switch (rc) {
1396                         case SCRUB_NEXT_BREAK:
1397                                 goto next_group;
1398                         case SCRUB_NEXT_EXIT:
1399                                 brelse(param.bitmap);
1400                                 RETURN(0);
1401                         case SCRUB_NEXT_CRASH:
1402                                 brelse(param.bitmap);
1403                                 RETURN(SCRUB_IT_CRASH);
1404                         case SCRUB_NEXT_FATAL:
1405                                 brelse(param.bitmap);
1406                                 RETURN(-EINVAL);
1407                         }
1408
1409                         rc = exec(info, dev, &param, oic, &noslot, rc);
1410                         if (rc != 0) {
1411                                 brelse(param.bitmap);
1412                                 RETURN(rc);
1413                         }
1414                 }
1415
1416 next_group:
1417                 brelse(param.bitmap);
1418         }
1419
1420         if (*pos > limit)
1421                 RETURN(SCRUB_IT_ALL);
1422         RETURN(0);
1423 }
1424
1425 static int osd_otable_it_preload(const struct lu_env *env,
1426                                  struct osd_otable_it *it)
1427 {
1428         struct osd_device       *dev   = it->ooi_dev;
1429         struct osd_scrub        *scrub = &dev->od_scrub;
1430         struct osd_otable_cache *ooc   = &it->ooi_cache;
1431         int                      rc;
1432         ENTRY;
1433
1434         rc = osd_inode_iteration(osd_oti_get(env), dev,
1435                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1436         if (rc == SCRUB_IT_ALL)
1437                 it->ooi_all_cached = 1;
1438
1439         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1440                 spin_lock(&scrub->os_lock);
1441                 scrub->os_waiting = 0;
1442                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1443                 spin_unlock(&scrub->os_lock);
1444         }
1445
1446         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1447 }
1448
1449 static int osd_scrub_main(void *args)
1450 {
1451         struct lu_env         env;
1452         struct osd_device    *dev    = (struct osd_device *)args;
1453         struct osd_scrub     *scrub  = &dev->od_scrub;
1454         struct ptlrpc_thread *thread = &scrub->os_thread;
1455         int                   rc;
1456         ENTRY;
1457
1458         rc = lu_env_init(&env, LCT_LOCAL);
1459         if (rc != 0) {
1460                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1461                        osd_scrub2name(scrub), rc);
1462                 GOTO(noenv, rc);
1463         }
1464
1465         rc = osd_scrub_prep(dev);
1466         if (rc != 0) {
1467                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1468                        osd_scrub2name(scrub), rc);
1469                 GOTO(out, rc);
1470         }
1471
1472         if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1473                 struct l_wait_info lwi = { 0 };
1474                 struct osd_otable_it *it = dev->od_otable_it;
1475                 struct osd_otable_cache *ooc = &it->ooi_cache;
1476
1477                 l_wait_event(thread->t_ctl_waitq,
1478                              it->ooi_user_ready || !thread_is_running(thread),
1479                              &lwi);
1480                 if (unlikely(!thread_is_running(thread)))
1481                         GOTO(post, rc = 0);
1482
1483                 scrub->os_pos_current = ooc->ooc_pos_preload;
1484         }
1485
1486         CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1487                osd_scrub2name(scrub), scrub->os_start_flags,
1488                scrub->os_pos_current);
1489
1490         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1491         if (unlikely(rc == SCRUB_IT_CRASH))
1492                 GOTO(out, rc = -EINVAL);
1493         GOTO(post, rc);
1494
1495 post:
1496         osd_scrub_post(scrub, rc);
1497         CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1498                osd_scrub2name(scrub), scrub->os_pos_current, rc);
1499
1500 out:
1501         while (!list_empty(&scrub->os_inconsistent_items)) {
1502                 struct osd_inconsistent_item *oii;
1503
1504                 oii = list_entry(scrub->os_inconsistent_items.next,
1505                                      struct osd_inconsistent_item, oii_list);
1506                 list_del_init(&oii->oii_list);
1507                 OBD_FREE_PTR(oii);
1508         }
1509         lu_env_fini(&env);
1510
1511 noenv:
1512         spin_lock(&scrub->os_lock);
1513         thread_set_flags(thread, SVC_STOPPED);
1514         wake_up_all(&thread->t_ctl_waitq);
1515         spin_unlock(&scrub->os_lock);
1516         return rc;
1517 }
1518
1519 /* initial OI scrub */
1520
1521 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1522                          struct dentry *, filldir_t filldir);
1523
1524 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1525                                loff_t offset, __u64 ino, unsigned d_type);
1526 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1527                            loff_t offset, __u64 ino, unsigned d_type);
1528 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1529                            loff_t offset, __u64 ino, unsigned d_type);
1530 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1531                             loff_t offset, __u64 ino, unsigned d_type);
1532
1533 static int
1534 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1535                      struct dentry *dentry, filldir_t filldir);
1536 static int
1537 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1538                   struct dentry *dentry, filldir_t filldir);
1539
1540 static int
1541 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1542                      struct dentry *dentry, filldir_t filldir);
1543
1544 enum osd_lf_flags {
1545         OLF_SCAN_SUBITEMS       = 0x0001,
1546         OLF_HIDE_FID            = 0x0002,
1547         OLF_SHOW_NAME           = 0x0004,
1548         OLF_NO_OI               = 0x0008,
1549         OLF_IDX_IN_FID          = 0x0010,
1550 };
1551
1552 struct osd_lf_map {
1553         char            *olm_name;
1554         struct lu_fid    olm_fid;
1555         __u16            olm_flags;
1556         __u16            olm_namelen;
1557         scandir_t        olm_scandir;
1558         filldir_t        olm_filldir;
1559 };
1560
1561 /* Add the new introduced local files in the list in the future. */
1562 static const struct osd_lf_map osd_lf_maps[] = {
1563         /* CATALOGS */
1564         { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1565                 sizeof(CATLIST) - 1, NULL, NULL },
1566
1567         /* CONFIGS */
1568         { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1569                 OLF_SCAN_SUBITEMS, sizeof(MOUNT_CONFIGS_DIR) - 1,
1570                 osd_ios_general_scan, osd_ios_varfid_fill },
1571
1572         /* NIDTBL_VERSIONS */
1573         { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1574                 sizeof(MGS_NIDTBL_DIR) - 1, osd_ios_general_scan,
1575                 osd_ios_varfid_fill },
1576
1577         /* PENDING */
1578         { "PENDING", { 0, 0, 0 }, 0, sizeof("PENDING") - 1, NULL, NULL },
1579
1580         /* ROOT */
1581         { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 },
1582                 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, sizeof("ROOT") - 1,
1583                 osd_ios_ROOT_scan, NULL },
1584
1585         /* changelog_catalog */
1586         { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, sizeof(CHANGELOG_CATALOG) - 1,
1587                 NULL, NULL },
1588
1589         /* changelog_users */
1590         { CHANGELOG_USERS, { 0, 0, 0 }, 0, sizeof(CHANGELOG_USERS) - 1,
1591                 NULL, NULL },
1592
1593         /* fld */
1594         { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1595                 sizeof("fld") - 1, NULL, NULL },
1596
1597         /* last_rcvd */
1598         { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1599                 sizeof(LAST_RCVD) - 1, NULL, NULL },
1600
1601         /* reply_data */
1602         { REPLY_DATA, { FID_SEQ_LOCAL_FILE, REPLY_DATA_OID, 0 }, OLF_SHOW_NAME,
1603                 sizeof(REPLY_DATA) - 1, NULL, NULL },
1604
1605         /* lov_objid */
1606         { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1607                 sizeof(LOV_OBJID) - 1, NULL, NULL },
1608
1609         /* lov_objseq */
1610         { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1611                 OLF_SHOW_NAME, sizeof(LOV_OBJSEQ) - 1, NULL, NULL },
1612
1613         /* quota_master */
1614         { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QMT_DIR) - 1,
1615                 osd_ios_general_scan, osd_ios_varfid_fill },
1616
1617         /* quota_slave */
1618         { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QSD_DIR) - 1,
1619                 osd_ios_general_scan, osd_ios_varfid_fill },
1620
1621         /* seq_ctl */
1622         { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1623                 OLF_SHOW_NAME, sizeof("seq_ctl") - 1, NULL, NULL },
1624
1625         /* seq_srv */
1626         { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1627                 OLF_SHOW_NAME, sizeof("seq_srv") - 1, NULL, NULL },
1628
1629         /* health_check */
1630         { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1631                 OLF_SHOW_NAME, sizeof(HEALTH_CHECK) - 1, NULL, NULL },
1632
1633         /* LFSCK */
1634         { LFSCK_DIR, { 0, 0, 0 }, 0, sizeof(LFSCK_DIR) - 1,
1635                 osd_ios_general_scan, osd_ios_varfid_fill },
1636
1637         /* lfsck_bookmark */
1638         { LFSCK_BOOKMARK, { 0, 0, 0 }, 0, sizeof(LFSCK_BOOKMARK) - 1,
1639                 NULL, NULL },
1640
1641         /* lfsck_layout */
1642         { LFSCK_LAYOUT, { 0, 0, 0 }, 0, sizeof(LFSCK_LAYOUT) - 1,
1643                 NULL, NULL },
1644
1645         /* lfsck_namespace */
1646         { LFSCK_NAMESPACE, { 0, 0, 0 }, 0, sizeof(LFSCK_NAMESPACE) - 1,
1647                 NULL, NULL },
1648
1649         /* OBJECTS, upgrade from old device */
1650         { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(OBJECTS) - 1,
1651                 osd_ios_OBJECTS_scan, NULL },
1652
1653         /* lquota_v2.user, upgrade from old device */
1654         { "lquota_v2.user", { 0, 0, 0 }, 0, sizeof("lquota_v2.user") - 1,
1655                 NULL, NULL },
1656
1657         /* lquota_v2.group, upgrade from old device */
1658         { "lquota_v2.group", { 0, 0, 0 }, 0, sizeof("lquota_v2.group") - 1,
1659                 NULL, NULL },
1660
1661         /* LAST_GROUP, upgrade from old device */
1662         { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1663                 OLF_SHOW_NAME, sizeof("LAST_GROUP") - 1, NULL, NULL },
1664
1665         /* committed batchid for cross-MDT operation */
1666         { "BATCHID", { FID_SEQ_LOCAL_FILE, BATCHID_COMMITTED_OID, 0 },
1667                 OLF_SHOW_NAME, sizeof("BATCHID") - 1, NULL, NULL },
1668
1669         /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1670          * and f_oid = index for their log files.  See lu_update_log{_dir}_fid()
1671          * for more details. */
1672
1673         /* update_log */
1674         { "update_log", { FID_SEQ_UPDATE_LOG, 0, 0 },
1675                 OLF_SHOW_NAME | OLF_IDX_IN_FID, sizeof("update_log") - 1,
1676                 NULL, NULL },
1677
1678         /* update_log_dir */
1679         { "update_log_dir", { FID_SEQ_UPDATE_LOG_DIR, 0, 0 },
1680                 OLF_SHOW_NAME | OLF_SCAN_SUBITEMS | OLF_IDX_IN_FID,
1681                 sizeof("update_log_dir") - 1,
1682                 osd_ios_general_scan, osd_ios_uld_fill },
1683
1684         /* lost+found */
1685         { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 },
1686                 OLF_SCAN_SUBITEMS, sizeof("lost+found") - 1,
1687                 osd_ios_general_scan, osd_ios_lf_fill },
1688
1689         { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1690 };
1691
1692 /* Add the new introduced files under .lustre/ in the list in the future. */
1693 static const struct osd_lf_map osd_dl_maps[] = {
1694         /* .lustre/fid */
1695         { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1696                 sizeof("fid") - 1, NULL, NULL },
1697
1698         /* .lustre/lost+found */
1699         { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1700                 sizeof("lost+found") - 1, NULL, NULL },
1701
1702         { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1703 };
1704
1705 struct osd_ios_item {
1706         struct list_head oii_list;
1707         struct dentry   *oii_dentry;
1708         scandir_t        oii_scandir;
1709         filldir_t        oii_filldir;
1710 };
1711
1712 struct osd_ios_filldir_buf {
1713 #ifdef HAVE_DIR_CONTEXT
1714         /* please keep it as first member */
1715         struct dir_context       ctx;
1716 #endif
1717         struct osd_thread_info  *oifb_info;
1718         struct osd_device       *oifb_dev;
1719         struct dentry           *oifb_dentry;
1720 };
1721
1722 static inline struct dentry *
1723 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1724 {
1725         struct dentry *dentry;
1726
1727         dentry = ll_lookup_one_len(name, parent, namelen);
1728         if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1729                 dput(dentry);
1730                 return ERR_PTR(-ENOENT);
1731         }
1732
1733         return dentry;
1734 }
1735
1736 static int
1737 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1738                  scandir_t scandir, filldir_t filldir)
1739 {
1740         struct osd_ios_item *item;
1741         ENTRY;
1742
1743         OBD_ALLOC_PTR(item);
1744         if (item == NULL)
1745                 RETURN(-ENOMEM);
1746
1747         INIT_LIST_HEAD(&item->oii_list);
1748         item->oii_dentry = dget(dentry);
1749         item->oii_scandir = scandir;
1750         item->oii_filldir = filldir;
1751         list_add_tail(&item->oii_list, &dev->od_ios_list);
1752
1753         RETURN(0);
1754 }
1755
1756 /**
1757  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1758  *
1759  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1760  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1761  * reference the inode, or fixed if it is missing or references another inode.
1762  */
1763 static int
1764 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1765                  struct inode *inode, const struct lu_fid *fid, int flags)
1766 {
1767         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
1768         struct osd_inode_id     *id     = &info->oti_id;
1769         struct osd_inode_id     *id2    = &info->oti_id2;
1770         struct osd_scrub        *scrub  = &dev->od_scrub;
1771         struct scrub_file       *sf     = &scrub->os_file;
1772         struct lu_fid            tfid;
1773         int                      rc;
1774         ENTRY;
1775
1776         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1777         if (rc != 0 && rc != -ENODATA) {
1778                 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1779                        "rc = %d\n", osd_name(dev), rc);
1780
1781                 RETURN(rc);
1782         }
1783
1784         osd_id_gen(id, inode->i_ino, inode->i_generation);
1785         if (rc == -ENODATA) {
1786                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
1787                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1788                 } else {
1789                         tfid = *fid;
1790                         if (flags & OLF_IDX_IN_FID) {
1791                                 LASSERT(dev->od_index >= 0);
1792
1793                                 tfid.f_oid = dev->od_index;
1794                         }
1795                 }
1796                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1797                 if (rc != 0) {
1798                         CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1799                               "scrub: rc = %d\n", osd_name(dev), rc);
1800
1801                         RETURN(rc);
1802                 }
1803         } else {
1804                 if (lma->lma_compat & LMAC_NOT_IN_OI)
1805                         RETURN(0);
1806
1807                 tfid = lma->lma_self_fid;
1808         }
1809
1810         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1811         if (rc != 0) {
1812                 if (rc != -ENOENT)
1813                         RETURN(rc);
1814
1815                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1816                                                DTO_INDEX_INSERT, true, 0);
1817                 if (rc > 0)
1818                         rc = 0;
1819
1820                 RETURN(rc);
1821         }
1822
1823         if (osd_id_eq_strict(id, id2))
1824                 RETURN(0);
1825
1826         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1827                 osd_scrub_file_reset(scrub,
1828                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1829                                      SF_INCONSISTENT);
1830                 rc = osd_scrub_file_store(scrub);
1831                 if (rc != 0)
1832                         RETURN(rc);
1833         }
1834
1835         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1836                                        DTO_INDEX_UPDATE, true, 0);
1837         if (rc > 0)
1838                 rc = 0;
1839
1840         RETURN(rc);
1841 }
1842
1843 /**
1844  * It scans the /lost+found, and for the OST-object (with filter_fid
1845  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1846  */
1847 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1848                            loff_t offset, __u64 ino, unsigned d_type)
1849 {
1850         struct osd_ios_filldir_buf *fill_buf = buf;
1851         struct osd_thread_info     *info     = fill_buf->oifb_info;
1852         struct osd_device          *dev      = fill_buf->oifb_dev;
1853         struct lu_fid              *fid      = &info->oti_fid;
1854         struct osd_scrub           *scrub    = &dev->od_scrub;
1855         struct dentry              *parent   = fill_buf->oifb_dentry;
1856         struct dentry              *child;
1857         struct inode               *dir      = parent->d_inode;
1858         struct inode               *inode;
1859         int                         rc;
1860         ENTRY;
1861
1862         /* skip any '.' started names */
1863         if (name[0] == '.')
1864                 RETURN(0);
1865
1866         scrub->os_lf_scanned++;
1867         child = osd_ios_lookup_one_len(name, parent, namelen);
1868         if (IS_ERR(child)) {
1869                 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1870                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
1871                 RETURN(0);
1872         }
1873
1874         inode = child->d_inode;
1875         if (S_ISDIR(inode->i_mode)) {
1876                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1877                                       osd_ios_lf_fill);
1878                 if (rc != 0)
1879                         CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1880                               "rc = %d\n", osd_name(dev), namelen, name, rc);
1881                 GOTO(put, rc);
1882         }
1883
1884         if (!S_ISREG(inode->i_mode))
1885                 GOTO(put, rc = 0);
1886
1887         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1888         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1889                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1890                 if (rc == 0) {
1891                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1892                                "/lost+found.\n", namelen, name, PFID(fid));
1893                         scrub->os_lf_repaired++;
1894                 } else {
1895                         CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1896                                DFID": rc = %d\n",
1897                                osd_name(dev), namelen, name, PFID(fid), rc);
1898                 }
1899         }
1900
1901         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1902          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1903          *      can process them in furtuer. */
1904
1905         GOTO(put, rc);
1906
1907 put:
1908         if (rc < 0)
1909                 scrub->os_lf_failed++;
1910         dput(child);
1911         /* skip the failure to make the scanning to continue. */
1912         return 0;
1913 }
1914
1915 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1916                                loff_t offset, __u64 ino, unsigned d_type)
1917 {
1918         struct osd_ios_filldir_buf *fill_buf = buf;
1919         struct osd_device          *dev      = fill_buf->oifb_dev;
1920         struct dentry              *child;
1921         int                         rc;
1922         ENTRY;
1923
1924         /* skip any '.' started names */
1925         if (name[0] == '.')
1926                 RETURN(0);
1927
1928         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1929         if (IS_ERR(child))
1930                 RETURN(PTR_ERR(child));
1931
1932         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1933                               NULL, 0);
1934         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1935                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1936                                       osd_ios_varfid_fill);
1937         dput(child);
1938
1939         RETURN(rc);
1940 }
1941
1942 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1943                            loff_t offset, __u64 ino, unsigned d_type)
1944 {
1945         struct osd_ios_filldir_buf *fill_buf = buf;
1946         struct osd_device          *dev      = fill_buf->oifb_dev;
1947         const struct osd_lf_map    *map;
1948         struct dentry              *child;
1949         int                         rc       = 0;
1950         ENTRY;
1951
1952         /* skip any '.' started names */
1953         if (name[0] == '.')
1954                 RETURN(0);
1955
1956         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
1957                 if (map->olm_namelen != namelen)
1958                         continue;
1959
1960                 if (strncmp(map->olm_name, name, namelen) == 0)
1961                         break;
1962         }
1963
1964         if (map->olm_name == NULL)
1965                 RETURN(0);
1966
1967         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1968         if (IS_ERR(child))
1969                 RETURN(PTR_ERR(child));
1970
1971         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1972                               &map->olm_fid, map->olm_flags);
1973         dput(child);
1974
1975         RETURN(rc);
1976 }
1977
1978 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1979                             loff_t offset, __u64 ino, unsigned d_type)
1980 {
1981         struct osd_ios_filldir_buf *fill_buf = buf;
1982         struct dentry              *child;
1983         struct lu_fid               tfid;
1984         int                         rc       = 0;
1985         ENTRY;
1986
1987         /* skip any non-DFID format name */
1988         if (name[0] != '[')
1989                 RETURN(0);
1990
1991         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1992         if (IS_ERR(child))
1993                 RETURN(PTR_ERR(child));
1994
1995         /* skip the start '[' */
1996         sscanf(&name[1], SFID, RFID(&tfid));
1997         if (fid_is_sane(&tfid))
1998                 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
1999                                       child->d_inode, &tfid, 0);
2000         else
2001                 rc = -EIO;
2002         dput(child);
2003
2004         RETURN(rc);
2005 }
2006
2007 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
2008                              loff_t offset, __u64 ino, unsigned d_type)
2009 {
2010         struct osd_ios_filldir_buf *fill_buf = buf;
2011         struct osd_device          *dev      = fill_buf->oifb_dev;
2012         const struct osd_lf_map    *map;
2013         struct dentry              *child;
2014         int                         rc       = 0;
2015         ENTRY;
2016
2017         /* skip any '.' started names */
2018         if (name[0] == '.')
2019                 RETURN(0);
2020
2021         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2022                 if (map->olm_namelen != namelen)
2023                         continue;
2024
2025                 if (strncmp(map->olm_name, name, namelen) == 0)
2026                         break;
2027         }
2028
2029         if (map->olm_name == NULL)
2030                 RETURN(0);
2031
2032         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2033         if (IS_ERR(child))
2034                 RETURN(PTR_ERR(child));
2035
2036         if (!(map->olm_flags & OLF_NO_OI))
2037                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2038                                       &map->olm_fid, map->olm_flags);
2039         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2040                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2041                                       map->olm_filldir);
2042         dput(child);
2043
2044         RETURN(rc);
2045 }
2046
2047 static int
2048 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2049                      struct dentry *dentry, filldir_t filldir)
2050 {
2051         struct osd_ios_filldir_buf    buf   = {
2052 #ifdef HAVE_DIR_CONTEXT
2053                                                 .ctx.actor = filldir,
2054 #endif
2055                                                 .oifb_info = info,
2056                                                 .oifb_dev = dev,
2057                                                 .oifb_dentry = dentry };
2058         struct file                  *filp  = &info->oti_file;
2059         struct inode                 *inode = dentry->d_inode;
2060         const struct file_operations *fops  = inode->i_fop;
2061         int                           rc;
2062         ENTRY;
2063
2064         LASSERT(filldir != NULL);
2065
2066         filp->f_pos = 0;
2067         filp->f_path.dentry = dentry;
2068         filp->f_mode = FMODE_64BITHASH;
2069         filp->f_mapping = inode->i_mapping;
2070         filp->f_op = fops;
2071         filp->private_data = NULL;
2072         set_file_inode(filp, inode);
2073
2074 #ifdef HAVE_DIR_CONTEXT
2075         buf.ctx.pos = filp->f_pos;
2076         rc = fops->iterate(filp, &buf.ctx);
2077         filp->f_pos = buf.ctx.pos;
2078 #else
2079         rc = fops->readdir(filp, &buf, filldir);
2080 #endif
2081         fops->release(inode, filp);
2082
2083         RETURN(rc);
2084 }
2085
2086 static int
2087 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2088                   struct dentry *dentry, filldir_t filldir)
2089 {
2090         struct osd_scrub  *scrub  = &dev->od_scrub;
2091         struct scrub_file *sf     = &scrub->os_file;
2092         struct dentry     *child;
2093         int                rc;
2094         ENTRY;
2095
2096         /* It is existing MDT0 device. We only allow the case of object without
2097          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2098          * can generate IGIF mode FID for the object and related OI mapping. If
2099          * it is on other MDTs, then becuase file-level backup/restore, related
2100          * OI mapping may be invalid already, we do not know which is the right
2101          * FID for the object. We only allow IGIF objects to reside on the MDT0.
2102          *
2103          * XXX: For the case of object on non-MDT0 device with neither LMA nor
2104          *      "fid" xattr, then something crashed. We cannot re-generate the
2105          *      FID directly, instead, the OI scrub will scan the OI structure
2106          *      and try to re-generate the LMA from the OI mapping. But if the
2107          *      OI mapping crashed or lost also, then we have to give up under
2108          *      double failure cases. */
2109         scrub->os_convert_igif = 1;
2110         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2111                                        strlen(dot_lustre_name));
2112         if (IS_ERR(child)) {
2113                 rc = PTR_ERR(child);
2114                 if (rc == -ENOENT) {
2115                         /* It is 1.8 MDT device. */
2116                         if (!(sf->sf_flags & SF_UPGRADE)) {
2117                                 osd_scrub_file_reset(scrub,
2118                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2119                                         SF_UPGRADE);
2120                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2121                                 rc = osd_scrub_file_store(scrub);
2122                         } else {
2123                                 rc = 0;
2124                         }
2125                 }
2126         } else {
2127                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2128                  * so the client will get IGIF for the ".lustre" object when
2129                  * the MDT restart.
2130                  *
2131                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2132                  * it does not know whether there are some old clients cached
2133                  * the ".lustre" IGIF during the upgrading. Two choices:
2134                  *
2135                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2136                  *    It will allow the old connected clients to access the
2137                  *    ".lustre" with cached IGIF. But it will cause others
2138                  *    on the MDT failed to check "fid_is_dot_lustre()".
2139                  *
2140                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2141                  *    for ".lustre" in spite of whether there are some clients
2142                  *    cached the ".lustre" IGIF or not. It enables the check
2143                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
2144                  *    that the old connected clients cannot access the ".lustre"
2145                  *    with the cached IGIF.
2146                  *
2147                  * Usually, it is rare case for the old connected clients
2148                  * to access the ".lustre" with cached IGIF. So we prefer
2149                  * to the solution 2). */
2150                 rc = osd_ios_scan_one(info, dev, child->d_inode,
2151                                       &LU_DOT_LUSTRE_FID, 0);
2152                 if (rc == 0)
2153                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2154                                               osd_ios_dl_fill);
2155                 dput(child);
2156         }
2157
2158         RETURN(rc);
2159 }
2160
2161 static int
2162 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2163                      struct dentry *dentry, filldir_t filldir)
2164 {
2165         struct osd_scrub  *scrub  = &dev->od_scrub;
2166         struct scrub_file *sf     = &scrub->os_file;
2167         struct dentry     *child;
2168         int                rc;
2169         ENTRY;
2170
2171         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2172                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2173                 rc = osd_scrub_file_store(scrub);
2174                 if (rc != 0)
2175                         RETURN(rc);
2176         }
2177
2178         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2179         if (!IS_ERR(child)) {
2180                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2181                 dput(child);
2182         } else {
2183                 rc = PTR_ERR(child);
2184         }
2185
2186         if (rc != 0 && rc != -ENOENT)
2187                 RETURN(rc);
2188
2189         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2190         if (!IS_ERR(child)) {
2191                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2192                 dput(child);
2193         } else {
2194                 rc = PTR_ERR(child);
2195         }
2196
2197         if (rc == -ENOENT)
2198                 rc = 0;
2199
2200         RETURN(rc);
2201 }
2202
2203 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2204                                 struct osd_device *dev)
2205 {
2206         struct osd_ios_item     *item    = NULL;
2207         scandir_t                scandir = osd_ios_general_scan;
2208         filldir_t                filldir = osd_ios_root_fill;
2209         struct dentry           *dentry  = osd_sb(dev)->s_root;
2210         const struct osd_lf_map *map     = osd_lf_maps;
2211         int                      rc;
2212         ENTRY;
2213
2214         /* Lookup IGIF in OI by force for initial OI scrub. */
2215         dev->od_igif_inoi = 1;
2216
2217         while (1) {
2218                 rc = scandir(info, dev, dentry, filldir);
2219                 if (item != NULL) {
2220                         dput(item->oii_dentry);
2221                         OBD_FREE_PTR(item);
2222                 }
2223
2224                 if (rc != 0)
2225                         break;
2226
2227                 if (list_empty(&dev->od_ios_list))
2228                         break;
2229
2230                 item = list_entry(dev->od_ios_list.next,
2231                                   struct osd_ios_item, oii_list);
2232                 list_del_init(&item->oii_list);
2233
2234                 LASSERT(item->oii_scandir != NULL);
2235                 scandir = item->oii_scandir;
2236                 filldir = item->oii_filldir;
2237                 dentry = item->oii_dentry;
2238         }
2239
2240         while (!list_empty(&dev->od_ios_list)) {
2241                 item = list_entry(dev->od_ios_list.next,
2242                                   struct osd_ios_item, oii_list);
2243                 list_del_init(&item->oii_list);
2244                 dput(item->oii_dentry);
2245                 OBD_FREE_PTR(item);
2246         }
2247
2248         if (rc != 0)
2249                 RETURN(rc);
2250
2251         /* There maybe the case that the object has been removed, but its OI
2252          * mapping is still in the OI file, such as the "CATALOGS" after MDT
2253          * file-level backup/restore. So here cleanup the stale OI mappings. */
2254         while (map->olm_name != NULL) {
2255                 struct dentry *child;
2256
2257                 if (fid_is_zero(&map->olm_fid)) {
2258                         map++;
2259                         continue;
2260                 }
2261
2262                 child = osd_ios_lookup_one_len(map->olm_name,
2263                                                osd_sb(dev)->s_root,
2264                                                map->olm_namelen);
2265                 if (!IS_ERR(child))
2266                         dput(child);
2267                 else if (PTR_ERR(child) == -ENOENT)
2268                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2269                                                   NULL, DTO_INDEX_DELETE,
2270                                                   true, 0);
2271                 map++;
2272         }
2273
2274         RETURN(0);
2275 }
2276
2277 char *osd_lf_fid2name(const struct lu_fid *fid)
2278 {
2279         const struct osd_lf_map *map = osd_lf_maps;
2280
2281         while (map->olm_name != NULL) {
2282                 if (!lu_fid_eq(fid, &map->olm_fid)) {
2283                         map++;
2284                         continue;
2285                 }
2286
2287                 if (map->olm_flags & OLF_SHOW_NAME)
2288                         return map->olm_name;
2289                 else
2290                         return "";
2291         }
2292
2293         return NULL;
2294 }
2295
2296 /* OI scrub start/stop */
2297
2298 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2299 {
2300         struct osd_scrub     *scrub  = &dev->od_scrub;
2301         struct ptlrpc_thread *thread = &scrub->os_thread;
2302         struct l_wait_info    lwi    = { 0 };
2303         struct task_struct   *task;
2304         int                   rc;
2305         ENTRY;
2306
2307         /* os_lock: sync status between stop and scrub thread */
2308         spin_lock(&scrub->os_lock);
2309
2310 again:
2311         if (thread_is_running(thread)) {
2312                 spin_unlock(&scrub->os_lock);
2313                 if (!(scrub->os_file.sf_flags & SF_AUTO) ||
2314                      (flags & (SS_AUTO_FULL | SS_AUTO_PARTIAL)))
2315                         RETURN(-EALREADY);
2316
2317                 osd_scrub_join(dev, flags, false);
2318                 spin_lock(&scrub->os_lock);
2319                 if (!thread_is_running(thread))
2320                         goto again;
2321
2322                 spin_unlock(&scrub->os_lock);
2323                 RETURN(0);
2324         }
2325
2326         if (unlikely(thread_is_stopping(thread))) {
2327                 spin_unlock(&scrub->os_lock);
2328                 l_wait_event(thread->t_ctl_waitq,
2329                              thread_is_stopped(thread),
2330                              &lwi);
2331                 spin_lock(&scrub->os_lock);
2332                 goto again;
2333         }
2334         spin_unlock(&scrub->os_lock);
2335
2336         if (scrub->os_file.sf_status == SS_COMPLETED) {
2337                 if (!(flags & SS_SET_FAILOUT))
2338                         flags |= SS_CLEAR_FAILOUT;
2339
2340                 if (!(flags & SS_SET_DRYRUN))
2341                         flags |= SS_CLEAR_DRYRUN;
2342
2343                 flags |= SS_RESET;
2344         }
2345
2346         scrub->os_start_flags = flags;
2347         thread_set_flags(thread, 0);
2348         task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2349         if (IS_ERR(task)) {
2350                 rc = PTR_ERR(task);
2351                 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2352                        osd_scrub2name(scrub), rc);
2353                 RETURN(rc);
2354         }
2355
2356         l_wait_event(thread->t_ctl_waitq,
2357                      thread_is_running(thread) || thread_is_stopped(thread),
2358                      &lwi);
2359
2360         RETURN(0);
2361 }
2362
2363 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2364 {
2365         int rc;
2366         ENTRY;
2367
2368         /* od_otable_mutex: prevent curcurrent start/stop */
2369         mutex_lock(&dev->od_otable_mutex);
2370         rc = do_osd_scrub_start(dev, flags);
2371         mutex_unlock(&dev->od_otable_mutex);
2372
2373         RETURN(rc == -EALREADY ? 0 : rc);
2374 }
2375
2376 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2377 {
2378         struct ptlrpc_thread *thread = &scrub->os_thread;
2379         struct l_wait_info    lwi    = { 0 };
2380
2381         /* os_lock: sync status between stop and scrub thread */
2382         spin_lock(&scrub->os_lock);
2383         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2384                 thread_set_flags(thread, SVC_STOPPING);
2385                 spin_unlock(&scrub->os_lock);
2386                 wake_up_all(&thread->t_ctl_waitq);
2387                 l_wait_event(thread->t_ctl_waitq,
2388                              thread_is_stopped(thread),
2389                              &lwi);
2390                 /* Do not skip the last lock/unlock, which can guarantee that
2391                  * the caller cannot return until the OI scrub thread exit. */
2392                 spin_lock(&scrub->os_lock);
2393         }
2394         spin_unlock(&scrub->os_lock);
2395 }
2396
2397 static void osd_scrub_stop(struct osd_device *dev)
2398 {
2399         /* od_otable_mutex: prevent curcurrent start/stop */
2400         mutex_lock(&dev->od_otable_mutex);
2401         dev->od_scrub.os_paused = 1;
2402         do_osd_scrub_stop(&dev->od_scrub);
2403         mutex_unlock(&dev->od_otable_mutex);
2404 }
2405
2406 /* OI scrub setup/cleanup */
2407
2408 static const char osd_scrub_name[] = "OI_scrub";
2409
2410 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2411 {
2412         struct osd_thread_info     *info   = osd_oti_get(env);
2413         struct osd_scrub           *scrub  = &dev->od_scrub;
2414         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
2415         struct scrub_file          *sf     = &scrub->os_file;
2416         struct super_block         *sb     = osd_sb(dev);
2417         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
2418         struct lvfs_run_ctxt        saved;
2419         struct file                *filp;
2420         struct inode               *inode;
2421         struct lu_fid              *fid    = &info->oti_fid;
2422         int                         dirty  = 0;
2423         int                         rc     = 0;
2424         ENTRY;
2425
2426         memset(scrub, 0, sizeof(*scrub));
2427         OBD_SET_CTXT_MAGIC(ctxt);
2428         ctxt->pwdmnt = dev->od_mnt;
2429         ctxt->pwd = dev->od_mnt->mnt_root;
2430         ctxt->fs = get_ds();
2431
2432         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2433         init_rwsem(&scrub->os_rwsem);
2434         spin_lock_init(&scrub->os_lock);
2435         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2436
2437         push_ctxt(&saved, ctxt);
2438         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2439         if (IS_ERR(filp)) {
2440                 pop_ctxt(&saved, ctxt);
2441                 RETURN(PTR_ERR(filp));
2442         }
2443
2444         inode = filp->f_path.dentry->d_inode;
2445         /* 'What the @fid is' is not imporatant, because the object
2446          * has no OI mapping, and only is visible inside the OSD.*/
2447         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2448         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2449         if (rc != 0) {
2450                 filp_close(filp, NULL);
2451                 pop_ctxt(&saved, ctxt);
2452                 RETURN(rc);
2453         }
2454
2455         scrub->os_inode = igrab(inode);
2456         filp_close(filp, NULL);
2457         pop_ctxt(&saved, ctxt);
2458
2459         rc = osd_scrub_file_load(scrub);
2460         if (rc == -ENOENT) {
2461                 osd_scrub_file_init(scrub, es->s_uuid);
2462                 /* If the "/O" dir does not exist when mount (indicated by
2463                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2464                  * then it is quite probably that the device is a new one,
2465                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2466                  *
2467                  * For the rare case that "/O" and "OI_scrub" both lost on
2468                  * an old device, it can be found and cleared later.
2469                  *
2470                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2471                  * need to check "filter_fid_old" and to convert it to
2472                  * "filter_fid" for each object, and all the IGIF should
2473                  * have their FID mapping in OI files already. */
2474                 if (dev->od_maybe_new)
2475                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2476                 dirty = 1;
2477         } else if (rc != 0) {
2478                 GOTO(cleanup_inode, rc);
2479         } else {
2480                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2481                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2482                         dirty = 1;
2483                 } else if (sf->sf_status == SS_SCANNING) {
2484                         sf->sf_status = SS_CRASHED;
2485                         dirty = 1;
2486                 }
2487         }
2488
2489         if (sf->sf_pos_last_checkpoint != 0)
2490                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2491         else
2492                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2493
2494         if (dirty != 0) {
2495                 rc = osd_scrub_file_store(scrub);
2496                 if (rc != 0)
2497                         GOTO(cleanup_inode, rc);
2498         }
2499
2500         /* Initialize OI files. */
2501         rc = osd_oi_init(info, dev);
2502         if (rc < 0)
2503                 GOTO(cleanup_inode, rc);
2504
2505         rc = osd_initial_OI_scrub(info, dev);
2506         if (rc != 0)
2507                 GOTO(cleanup_oi, rc);
2508
2509         if (sf->sf_flags & SF_UPGRADE ||
2510             !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2511               sf->sf_success_count > 0)) {
2512                 dev->od_igif_inoi = 0;
2513                 dev->od_check_ff = dev->od_is_ost;
2514         } else {
2515                 dev->od_igif_inoi = 1;
2516                 dev->od_check_ff = 0;
2517         }
2518
2519         if (sf->sf_flags & SF_INCONSISTENT)
2520                 /* The 'od_igif_inoi' will be set under the
2521                  * following cases:
2522                  * 1) new created system, or
2523                  * 2) restored from file-level backup, or
2524                  * 3) the upgrading completed.
2525                  *
2526                  * The 'od_igif_inoi' may be cleared by OI scrub
2527                  * later if found that the system is upgrading. */
2528                 dev->od_igif_inoi = 1;
2529
2530         if (!dev->od_noscrub &&
2531             ((sf->sf_status == SS_PAUSED) ||
2532              (sf->sf_status == SS_CRASHED &&
2533               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2534                               SF_UPGRADE | SF_AUTO)) ||
2535              (sf->sf_status == SS_INIT &&
2536               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2537                               SF_UPGRADE))))
2538                 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2539
2540         if (rc != 0)
2541                 GOTO(cleanup_oi, rc);
2542
2543         /* it is possible that dcache entries may keep objects after they are
2544          * deleted by OSD. While it looks safe this can cause object data to
2545          * stay until umount causing failures in tests calculating free space,
2546          * e.g. replay-ost-single. Since those dcache entries are not used
2547          * anymore let's just free them after use here */
2548         shrink_dcache_sb(sb);
2549
2550         RETURN(0);
2551 cleanup_oi:
2552         osd_oi_fini(info, dev);
2553 cleanup_inode:
2554         iput(scrub->os_inode);
2555         scrub->os_inode = NULL;
2556
2557         return rc;
2558 }
2559
2560 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2561 {
2562         struct osd_scrub *scrub = &dev->od_scrub;
2563
2564         LASSERT(dev->od_otable_it == NULL);
2565
2566         if (scrub->os_inode != NULL) {
2567                 osd_scrub_stop(dev);
2568                 iput(scrub->os_inode);
2569                 scrub->os_inode = NULL;
2570         }
2571         if (dev->od_oi_table != NULL)
2572                 osd_oi_fini(osd_oti_get(env), dev);
2573 }
2574
2575 /* object table based iteration APIs */
2576
2577 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2578                                        struct dt_object *dt, __u32 attr)
2579 {
2580         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2581         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2582         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2583         struct osd_scrub       *scrub = &dev->od_scrub;
2584         struct osd_otable_it   *it;
2585         __u32                   start = 0;
2586         int                     rc;
2587         ENTRY;
2588
2589         /* od_otable_mutex: prevent curcurrent init/fini */
2590         mutex_lock(&dev->od_otable_mutex);
2591         if (dev->od_otable_it != NULL)
2592                 GOTO(out, it = ERR_PTR(-EALREADY));
2593
2594         OBD_ALLOC_PTR(it);
2595         if (it == NULL)
2596                 GOTO(out, it = ERR_PTR(-ENOMEM));
2597
2598         dev->od_otable_it = it;
2599         it->ooi_dev = dev;
2600         it->ooi_cache.ooc_consumer_idx = -1;
2601         if (flags & DOIF_OUTUSED)
2602                 it->ooi_used_outside = 1;
2603
2604         if (flags & DOIF_RESET)
2605                 start |= SS_RESET;
2606
2607         if (valid & DOIV_ERROR_HANDLE) {
2608                 if (flags & DOIF_FAILOUT)
2609                         start |= SS_SET_FAILOUT;
2610                 else
2611                         start |= SS_CLEAR_FAILOUT;
2612         }
2613
2614         if (valid & DOIV_DRYRUN) {
2615                 if (flags & DOIF_DRYRUN)
2616                         start |= SS_SET_DRYRUN;
2617                 else
2618                         start |= SS_CLEAR_DRYRUN;
2619         }
2620
2621         rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2622         if (rc < 0 && rc != -EALREADY) {
2623                 dev->od_otable_it = NULL;
2624                 OBD_FREE_PTR(it);
2625                 GOTO(out, it = ERR_PTR(rc));
2626         }
2627
2628         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2629
2630         GOTO(out, it);
2631
2632 out:
2633         mutex_unlock(&dev->od_otable_mutex);
2634         return (struct dt_it *)it;
2635 }
2636
2637 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2638 {
2639         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2640         struct osd_device    *dev = it->ooi_dev;
2641
2642         /* od_otable_mutex: prevent curcurrent init/fini */
2643         mutex_lock(&dev->od_otable_mutex);
2644         do_osd_scrub_stop(&dev->od_scrub);
2645         LASSERT(dev->od_otable_it == it);
2646
2647         dev->od_otable_it = NULL;
2648         mutex_unlock(&dev->od_otable_mutex);
2649         OBD_FREE_PTR(it);
2650 }
2651
2652 static int osd_otable_it_get(const struct lu_env *env,
2653                              struct dt_it *di, const struct dt_key *key)
2654 {
2655         return 0;
2656 }
2657
2658 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2659 {
2660 }
2661
2662 static inline int
2663 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2664 {
2665         spin_lock(&scrub->os_lock);
2666         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2667             scrub->os_waiting ||
2668             !thread_is_running(&scrub->os_thread))
2669                 it->ooi_waiting = 0;
2670         else
2671                 it->ooi_waiting = 1;
2672         spin_unlock(&scrub->os_lock);
2673
2674         return !it->ooi_waiting;
2675 }
2676
2677 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2678 {
2679         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2680         struct osd_device       *dev    = it->ooi_dev;
2681         struct osd_scrub        *scrub  = &dev->od_scrub;
2682         struct osd_otable_cache *ooc    = &it->ooi_cache;
2683         struct ptlrpc_thread    *thread = &scrub->os_thread;
2684         struct l_wait_info       lwi    = { 0 };
2685         int                      rc;
2686         ENTRY;
2687
2688         LASSERT(it->ooi_user_ready);
2689
2690 again:
2691         if (!thread_is_running(thread) && !it->ooi_used_outside)
2692                 RETURN(1);
2693
2694         if (ooc->ooc_cached_items > 0) {
2695                 ooc->ooc_cached_items--;
2696                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2697                                         ~OSD_OTABLE_IT_CACHE_MASK;
2698                 RETURN(0);
2699         }
2700
2701         if (it->ooi_all_cached) {
2702                 l_wait_event(thread->t_ctl_waitq,
2703                              !thread_is_running(thread),
2704                              &lwi);
2705                 RETURN(1);
2706         }
2707
2708         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2709                 spin_lock(&scrub->os_lock);
2710                 scrub->os_waiting = 0;
2711                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2712                 spin_unlock(&scrub->os_lock);
2713         }
2714
2715         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2716                 l_wait_event(thread->t_ctl_waitq,
2717                              osd_otable_it_wakeup(scrub, it),
2718                              &lwi);
2719
2720         if (!thread_is_running(thread) && !it->ooi_used_outside)
2721                 RETURN(1);
2722
2723         rc = osd_otable_it_preload(env, it);
2724         if (rc >= 0)
2725                 goto again;
2726
2727         RETURN(rc);
2728 }
2729
2730 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2731                                         const struct dt_it *di)
2732 {
2733         return NULL;
2734 }
2735
2736 static int osd_otable_it_key_size(const struct lu_env *env,
2737                                   const struct dt_it *di)
2738 {
2739         return sizeof(__u64);
2740 }
2741
2742 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2743                              struct dt_rec *rec, __u32 attr)
2744 {
2745         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2746         struct osd_otable_cache *ooc = &it->ooi_cache;
2747
2748         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2749
2750         /* Filter out Invald FID already. */
2751         LASSERTF(fid_is_sane((struct lu_fid *)rec),
2752                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2753                  PFID((struct lu_fid *)rec),
2754                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2755
2756         return 0;
2757 }
2758
2759 static __u64 osd_otable_it_store(const struct lu_env *env,
2760                                  const struct dt_it *di)
2761 {
2762         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2763         struct osd_otable_cache *ooc = &it->ooi_cache;
2764         __u64                    hash;
2765
2766         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2767                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2768         else
2769                 hash = ooc->ooc_pos_preload;
2770         return hash;
2771 }
2772
2773 /**
2774  * Set the OSD layer iteration start position as the specified hash.
2775  */
2776 static int osd_otable_it_load(const struct lu_env *env,
2777                               const struct dt_it *di, __u64 hash)
2778 {
2779         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
2780         struct osd_device       *dev   = it->ooi_dev;
2781         struct osd_otable_cache *ooc   = &it->ooi_cache;
2782         struct osd_scrub        *scrub = &dev->od_scrub;
2783         int                      rc;
2784         ENTRY;
2785
2786         /* Forbid to set iteration position after iteration started. */
2787         if (it->ooi_user_ready)
2788                 RETURN(-EPERM);
2789
2790         if (hash > OSD_OTABLE_MAX_HASH)
2791                 hash = OSD_OTABLE_MAX_HASH;
2792
2793         ooc->ooc_pos_preload = hash;
2794         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2795                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2796
2797         it->ooi_user_ready = 1;
2798         if (!scrub->os_full_speed)
2799                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2800
2801         /* Unplug OSD layer iteration by the first next() call. */
2802         rc = osd_otable_it_next(env, (struct dt_it *)it);
2803
2804         RETURN(rc);
2805 }
2806
2807 static int osd_otable_it_key_rec(const struct lu_env *env,
2808                                  const struct dt_it *di, void *key_rec)
2809 {
2810         return 0;
2811 }
2812
2813 const struct dt_index_operations osd_otable_ops = {
2814         .dio_it = {
2815                 .init     = osd_otable_it_init,
2816                 .fini     = osd_otable_it_fini,
2817                 .get      = osd_otable_it_get,
2818                 .put      = osd_otable_it_put,
2819                 .next     = osd_otable_it_next,
2820                 .key      = osd_otable_it_key,
2821                 .key_size = osd_otable_it_key_size,
2822                 .rec      = osd_otable_it_rec,
2823                 .store    = osd_otable_it_store,
2824                 .load     = osd_otable_it_load,
2825                 .key_rec  = osd_otable_it_key_rec,
2826         }
2827 };
2828
2829 /* high priority inconsistent items list APIs */
2830
2831 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL  60
2832
2833 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2834                    int insert)
2835 {
2836         struct osd_inconsistent_item *oii;
2837         struct osd_scrub             *scrub  = &dev->od_scrub;
2838         struct ptlrpc_thread         *thread = &scrub->os_thread;
2839         int                           wakeup = 0;
2840         ENTRY;
2841
2842         OBD_ALLOC_PTR(oii);
2843         if (unlikely(oii == NULL))
2844                 RETURN(-ENOMEM);
2845
2846         INIT_LIST_HEAD(&oii->oii_list);
2847         oii->oii_cache = *oic;
2848         oii->oii_insert = insert;
2849
2850         if (scrub->os_partial_scan) {
2851                 __u64 now = cfs_time_current_sec();
2852
2853                 /* If there haven't been errors in a long time,
2854                  * decay old count until either the errors are
2855                  * gone or we reach the current interval. */
2856                 while (unlikely(scrub->os_bad_oimap_count > 0 &&
2857                                 scrub->os_bad_oimap_time +
2858                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
2859                         scrub->os_bad_oimap_count >>= 1;
2860                         scrub->os_bad_oimap_time +=
2861                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
2862                 }
2863
2864                 scrub->os_bad_oimap_time = now;
2865                 if (++scrub->os_bad_oimap_count >
2866                     dev->od_full_scrub_threshold_rate)
2867                         scrub->os_full_scrub = 1;
2868         }
2869
2870         spin_lock(&scrub->os_lock);
2871         if (unlikely(!thread_is_running(thread))) {
2872                 spin_unlock(&scrub->os_lock);
2873                 OBD_FREE_PTR(oii);
2874                 RETURN(-EAGAIN);
2875         }
2876
2877         if (list_empty(&scrub->os_inconsistent_items))
2878                 wakeup = 1;
2879         list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2880         spin_unlock(&scrub->os_lock);
2881
2882         if (wakeup != 0)
2883                 wake_up_all(&thread->t_ctl_waitq);
2884
2885         RETURN(0);
2886 }
2887
2888 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2889                    struct osd_inode_id *id)
2890 {
2891         struct osd_scrub             *scrub = &dev->od_scrub;
2892         struct osd_inconsistent_item *oii;
2893         ENTRY;
2894
2895         spin_lock(&scrub->os_lock);
2896         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2897                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2898                         *id = oii->oii_cache.oic_lid;
2899                         spin_unlock(&scrub->os_lock);
2900                         RETURN(0);
2901                 }
2902         }
2903         spin_unlock(&scrub->os_lock);
2904
2905         RETURN(-ENOENT);
2906 }
2907
2908 /* OI scrub dump */
2909
2910 static const char *scrub_status_names[] = {
2911         "init",
2912         "scanning",
2913         "completed",
2914         "failed",
2915         "stopped",
2916         "paused",
2917         "crashed",
2918         NULL
2919 };
2920
2921 static const char *scrub_flags_names[] = {
2922         "recreated",
2923         "inconsistent",
2924         "auto",
2925         "upgrade",
2926         NULL
2927 };
2928
2929 static const char *scrub_param_names[] = {
2930         "failout",
2931         "dryrun",
2932         NULL
2933 };
2934
2935 static int scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
2936                            const char *prefix)
2937 {
2938         int flag;
2939         int rc;
2940         int i;
2941
2942         rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2943         if (rc < 0)
2944                 return rc;
2945
2946         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2947                 if (flag & bits) {
2948                         bits &= ~flag;
2949                         rc = seq_printf(m, "%s%c", names[i],
2950                                         bits != 0 ? ',' : '\n');
2951                         if (rc < 0)
2952                                 return rc;
2953                 }
2954         }
2955         return 0;
2956 }
2957
2958 static int scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
2959 {
2960         int rc;
2961
2962         if (time != 0)
2963                 rc = seq_printf(m, "%s: "LPU64" seconds\n", prefix,
2964                               cfs_time_current_sec() - time);
2965         else
2966                 rc = seq_printf(m, "%s: N/A\n", prefix);
2967         return rc;
2968 }
2969
2970 static int scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
2971 {
2972         int rc;
2973
2974         if (pos != 0)
2975                 rc = seq_printf(m, "%s: "LPU64"\n", prefix, pos);
2976         else
2977                 rc = seq_printf(m, "%s: N/A\n", prefix);
2978         return rc;
2979 }
2980
2981 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
2982 {
2983         struct osd_scrub  *scrub   = &dev->od_scrub;
2984         struct scrub_file *sf      = &scrub->os_file;
2985         __u64              checked;
2986         __u64              speed;
2987         int                rc;
2988
2989         down_read(&scrub->os_rwsem);
2990         rc = seq_printf(m, "name: OI_scrub\n"
2991                         "magic: 0x%x\n"
2992                         "oi_files: %d\n"
2993                         "status: %s\n",
2994                         sf->sf_magic, (int)sf->sf_oi_count,
2995                         scrub_status_names[sf->sf_status]);
2996         if (rc < 0)
2997                 goto out;
2998
2999         rc = scrub_bits_dump(m, sf->sf_flags, scrub_flags_names,
3000                              "flags");
3001         if (rc < 0)
3002                 goto out;
3003
3004         rc = scrub_bits_dump(m, sf->sf_param, scrub_param_names,
3005                              "param");
3006         if (rc < 0)
3007                 goto out;
3008
3009         rc = scrub_time_dump(m, sf->sf_time_last_complete,
3010                              "time_since_last_completed");
3011         if (rc < 0)
3012                 goto out;
3013
3014         rc = scrub_time_dump(m, sf->sf_time_latest_start,
3015                              "time_since_latest_start");
3016         if (rc < 0)
3017                 goto out;
3018
3019         rc = scrub_time_dump(m, sf->sf_time_last_checkpoint,
3020                              "time_since_last_checkpoint");
3021         if (rc < 0)
3022                 goto out;
3023
3024         rc = scrub_pos_dump(m, sf->sf_pos_latest_start,
3025                             "latest_start_position");
3026         if (rc < 0)
3027                 goto out;
3028
3029         rc = scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
3030                             "last_checkpoint_position");
3031         if (rc < 0)
3032                 goto out;
3033
3034         rc = scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
3035                             "first_failure_position");
3036         if (rc < 0)
3037                 goto out;
3038
3039         checked = sf->sf_items_checked + scrub->os_new_checked;
3040         rc = seq_printf(m, "checked: "LPU64"\n"
3041                       "updated: "LPU64"\n"
3042                       "failed: "LPU64"\n"
3043                       "prior_updated: "LPU64"\n"
3044                       "noscrub: "LPU64"\n"
3045                       "igif: "LPU64"\n"
3046                       "success_count: %u\n",
3047                       checked, sf->sf_items_updated, sf->sf_items_failed,
3048                       sf->sf_items_updated_prior, sf->sf_items_noscrub,
3049                       sf->sf_items_igif, sf->sf_success_count);
3050         if (rc < 0)
3051                 goto out;
3052
3053         speed = checked;
3054         if (thread_is_running(&scrub->os_thread)) {
3055                 cfs_duration_t duration = cfs_time_current() -
3056                                           scrub->os_time_last_checkpoint;
3057                 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
3058                                                      MSEC_PER_SEC);
3059                 __u32 rtime = sf->sf_run_time +
3060                               cfs_duration_sec(duration + HALF_SEC);
3061
3062                 if (duration != 0)
3063                         do_div(new_checked, duration);
3064                 if (rtime != 0)
3065                         do_div(speed, rtime);
3066                 rc = seq_printf(m, "run_time: %u seconds\n"
3067                               "average_speed: "LPU64" objects/sec\n"
3068                               "real-time_speed: "LPU64" objects/sec\n"
3069                               "current_position: %u\n"
3070                               "lf_scanned: "LPU64"\n"
3071                               "lf_repaired: "LPU64"\n"
3072                               "lf_failed: "LPU64"\n",
3073                               rtime, speed, new_checked, scrub->os_pos_current,
3074                               scrub->os_lf_scanned, scrub->os_lf_repaired,
3075                               scrub->os_lf_failed);
3076         } else {
3077                 if (sf->sf_run_time != 0)
3078                         do_div(speed, sf->sf_run_time);
3079                 rc = seq_printf(m, "run_time: %u seconds\n"
3080                               "average_speed: "LPU64" objects/sec\n"
3081                               "real-time_speed: N/A\n"
3082                               "current_position: N/A\n"
3083                               "lf_scanned: "LPU64"\n"
3084                               "lf_repaired: "LPU64"\n"
3085                               "lf_failed: "LPU64"\n",
3086                               sf->sf_run_time, speed, scrub->os_lf_scanned,
3087                               scrub->os_lf_repaired, scrub->os_lf_failed);
3088         }
3089
3090 out:
3091         up_read(&scrub->os_rwsem);
3092         return (rc < 0 ? -ENOSPC : 0);
3093 }