lustre/osd-ldiskfs/osd_scrub.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2012, 2016, Intel Corporation.
  24  */
  25 /*
  26  * lustre/osd-ldiskfs/osd_scrub.c
  27  *
  28  * Top-level entry points into osd module
  29  *
  30  * The OI scrub is used for rebuilding Object Index files when restores MDT from
  31  * file-level backup.
  32  *
  33  * The otable based iterator scans ldiskfs inode table to feed up layer LFSCK.
  34  *
  35  * Author: Fan Yong <yong.fan@whamcloud.com>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_LFSCK
  39
  40 #include <linux/kthread.h>
  41 #include <lustre/lustre_idl.h>
  42 #include <lustre_disk.h>
  43 #include <dt_object.h>
  44 #include <linux/xattr.h>
  45
  46 #include "osd_internal.h"
  47 #include "osd_oi.h"
  48 #include "osd_scrub.h"
  49
  50 #define HALF_SEC        msecs_to_jiffies(MSEC_PER_SEC >> 1)
  51
  52 #define OSD_OTABLE_MAX_HASH             0x00000000ffffffffULL
  53
  54 #define SCRUB_NEXT_BREAK        1 /* exit current loop and process next group */
  55 #define SCRUB_NEXT_CONTINUE     2 /* skip current object and process next bit */
  56 #define SCRUB_NEXT_EXIT         3 /* exit all the loops */
  57 #define SCRUB_NEXT_WAIT         4 /* wait for free cache slot */
  58 #define SCRUB_NEXT_CRASH        5 /* simulate system crash during OI scrub */
  59 #define SCRUB_NEXT_FATAL        6 /* simulate failure during OI scrub */
  60 #define SCRUB_NEXT_NOSCRUB      7 /* new created object, no scrub on it */
  61 #define SCRUB_NEXT_NOLMA        8 /* the inode has no FID-in-LMA */
  62 #define SCRUB_NEXT_OSTOBJ       9 /* for OST-object */
  63 #define SCRUB_NEXT_OSTOBJ_OLD   10 /* old OST-object, no LMA or no FID-on-OST
  64                                     * flags in LMA */
  65
  66 /* misc functions */
  67
  68 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
  69 {
  70         return container_of0(scrub, struct osd_device, od_scrub);
  71 }
  72
  73 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
  74 {
  75         return osd_sb(osd_scrub2dev(scrub));
  76 }
  77
  78 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
  79                                        struct osd_otable_cache *ooc)
  80 {
  81         return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
  82 }
  83
  84 static inline const char *osd_scrub2name(struct osd_scrub *scrub)
  85 {
  86         return osd_dev2name(osd_scrub2dev(scrub));
  87 }
  88
  89 /**
  90  * update/insert/delete the specified OI mapping (@fid @id) according to the ops
  91  *
  92  * \retval   1, changed nothing
  93  * \retval   0, changed successfully
  94  * \retval -ve, on error
  95  */
  96 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
  97                                      struct osd_device *dev,
  98                                      const struct lu_fid *fid,
  99                                      const struct osd_inode_id *id,
 100                                      int ops, bool force,
 101                                      enum oi_check_flags flags, bool *exist)
 102 {
 103         handle_t *th;
 104         int       rc;
 105         ENTRY;
 106
 107         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
 108                 RETURN(0);
 109
 110         /* DTO_INDEX_INSERT is enough for other two ops:
 111          * delete/update, but save stack. */
 112         th = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 113                                 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
 114         if (IS_ERR(th)) {
 115                 rc = PTR_ERR(th);
 116                 CDEBUG(D_LFSCK, "%s: fail to start trans for scrub op %d "
 117                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 118                        PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1,
 119                        rc);
 120                 RETURN(rc);
 121         }
 122
 123         switch (ops) {
 124         case DTO_INDEX_UPDATE:
 125                 rc = osd_oi_update(info, dev, fid, id, th, flags);
 126                 if (unlikely(rc == -ENOENT)) {
 127                         /* Some unlink thread may removed the OI mapping. */
 128                         rc = 1;
 129                 }
 130                 break;
 131         case DTO_INDEX_INSERT:
 132                 rc = osd_oi_insert(info, dev, fid, id, th, flags, exist);
 133                 if (unlikely(rc == -EEXIST)) {
 134                         rc = 1;
 135                         /* XXX: There are trouble things when adding OI
 136                          *      mapping for IGIF object, which may cause
 137                          *      multiple objects to be mapped to the same
 138                          *      IGIF formatted FID. Consider the following
 139                          *      situations:
 140                          *
 141                          *      1) The MDT is upgrading from 1.8 device.
 142                          *      The OI scrub generates IGIF FID1 for the
 143                          *      OBJ1 and adds the OI mapping.
 144                          *
 145                          *      2) For some reason, the OI scrub does not
 146                          *      process all the IGIF objects completely.
 147                          *
 148                          *      3) The MDT is backuped and restored against
 149                          *      this device.
 150                          *
 151                          *      4) When the MDT mounts up, the OI scrub will
 152                          *      try to rebuild the OI files. For some IGIF
 153                          *      object, OBJ2, which was not processed by the
 154                          *      OI scrub before the backup/restore, and the
 155                          *      new generated IGIF formatted FID may be just
 156                          *      the FID1, the same as OBJ1.
 157                          *
 158                          *      Under such case, the OI scrub cannot know how
 159                          *      to generate new FID for the OBJ2.
 160                          *
 161                          *      Currently, we do nothing for that. One possible
 162                          *      solution is to generate new normal FID for the
 163                          *      conflict object.
 164                          *
 165                          *      Anyway, it is rare, only exists in theory. */
 166                 }
 167                 break;
 168         case DTO_INDEX_DELETE:
 169                 rc = osd_oi_delete(info, dev, fid, th, flags);
 170                 if (rc == -ENOENT) {
 171                         /* It is normal that the unlink thread has removed the
 172                          * OI mapping already. */
 173                         rc = 1;
 174                 }
 175                 break;
 176         default:
 177                 LASSERTF(0, "Unexpected ops %d\n", ops);
 178                 break;
 179         }
 180
 181         ldiskfs_journal_stop(th);
 182         if (rc < 0)
 183                 CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d "
 184                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 185                        PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1,
 186                        rc);
 187
 188         RETURN(rc);
 189 }
 190
 191 /* OI_scrub file ops */
 192
 193 static void osd_scrub_file_to_cpu(struct scrub_file *des,
 194                                   struct scrub_file *src)
 195 {
 196         memcpy(des->sf_uuid, src->sf_uuid, 16);
 197         des->sf_flags   = le64_to_cpu(src->sf_flags);
 198         des->sf_magic   = le32_to_cpu(src->sf_magic);
 199         des->sf_status  = le16_to_cpu(src->sf_status);
 200         des->sf_param   = le16_to_cpu(src->sf_param);
 201         des->sf_time_last_complete      =
 202                                 le64_to_cpu(src->sf_time_last_complete);
 203         des->sf_time_latest_start       =
 204                                 le64_to_cpu(src->sf_time_latest_start);
 205         des->sf_time_last_checkpoint    =
 206                                 le64_to_cpu(src->sf_time_last_checkpoint);
 207         des->sf_pos_latest_start        =
 208                                 le64_to_cpu(src->sf_pos_latest_start);
 209         des->sf_pos_last_checkpoint     =
 210                                 le64_to_cpu(src->sf_pos_last_checkpoint);
 211         des->sf_pos_first_inconsistent  =
 212                                 le64_to_cpu(src->sf_pos_first_inconsistent);
 213         des->sf_items_checked           =
 214                                 le64_to_cpu(src->sf_items_checked);
 215         des->sf_items_updated           =
 216                                 le64_to_cpu(src->sf_items_updated);
 217         des->sf_items_failed            =
 218                                 le64_to_cpu(src->sf_items_failed);
 219         des->sf_items_updated_prior     =
 220                                 le64_to_cpu(src->sf_items_updated_prior);
 221         des->sf_run_time        = le32_to_cpu(src->sf_run_time);
 222         des->sf_success_count   = le32_to_cpu(src->sf_success_count);
 223         des->sf_oi_count        = le16_to_cpu(src->sf_oi_count);
 224         des->sf_internal_flags  = le16_to_cpu(src->sf_internal_flags);
 225         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 226 }
 227
 228 static void osd_scrub_file_to_le(struct scrub_file *des,
 229                                  struct scrub_file *src)
 230 {
 231         memcpy(des->sf_uuid, src->sf_uuid, 16);
 232         des->sf_flags   = cpu_to_le64(src->sf_flags);
 233         des->sf_magic   = cpu_to_le32(src->sf_magic);
 234         des->sf_status  = cpu_to_le16(src->sf_status);
 235         des->sf_param   = cpu_to_le16(src->sf_param);
 236         des->sf_time_last_complete      =
 237                                 cpu_to_le64(src->sf_time_last_complete);
 238         des->sf_time_latest_start       =
 239                                 cpu_to_le64(src->sf_time_latest_start);
 240         des->sf_time_last_checkpoint    =
 241                                 cpu_to_le64(src->sf_time_last_checkpoint);
 242         des->sf_pos_latest_start        =
 243                                 cpu_to_le64(src->sf_pos_latest_start);
 244         des->sf_pos_last_checkpoint     =
 245                                 cpu_to_le64(src->sf_pos_last_checkpoint);
 246         des->sf_pos_first_inconsistent  =
 247                                 cpu_to_le64(src->sf_pos_first_inconsistent);
 248         des->sf_items_checked           =
 249                                 cpu_to_le64(src->sf_items_checked);
 250         des->sf_items_updated           =
 251                                 cpu_to_le64(src->sf_items_updated);
 252         des->sf_items_failed            =
 253                                 cpu_to_le64(src->sf_items_failed);
 254         des->sf_items_updated_prior     =
 255                                 cpu_to_le64(src->sf_items_updated_prior);
 256         des->sf_run_time        = cpu_to_le32(src->sf_run_time);
 257         des->sf_success_count   = cpu_to_le32(src->sf_success_count);
 258         des->sf_oi_count        = cpu_to_le16(src->sf_oi_count);
 259         des->sf_internal_flags  = cpu_to_le16(src->sf_internal_flags);
 260         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 261 }
 262
 263 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
 264 {
 265         struct scrub_file *sf = &scrub->os_file;
 266
 267         memset(sf, 0, sizeof(*sf));
 268         memcpy(sf->sf_uuid, uuid, 16);
 269         sf->sf_magic = SCRUB_MAGIC_V1;
 270         sf->sf_status = SS_INIT;
 271 }
 272
 273 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
 274 {
 275         struct scrub_file *sf = &scrub->os_file;
 276
 277         CDEBUG(D_LFSCK, "%s: reset OI scrub file, old flags = "
 278                "%#llx, add flags = %#llx\n",
 279                osd_scrub2name(scrub), sf->sf_flags, flags);
 280
 281         memcpy(sf->sf_uuid, uuid, 16);
 282         sf->sf_status = SS_INIT;
 283         sf->sf_flags |= flags;
 284         sf->sf_flags &= ~SF_AUTO;
 285         sf->sf_run_time = 0;
 286         sf->sf_time_latest_start = 0;
 287         sf->sf_time_last_checkpoint = 0;
 288         sf->sf_pos_latest_start = 0;
 289         sf->sf_pos_last_checkpoint = 0;
 290         sf->sf_pos_first_inconsistent = 0;
 291         sf->sf_items_checked = 0;
 292         sf->sf_items_updated = 0;
 293         sf->sf_items_failed = 0;
 294         if (!scrub->os_in_join)
 295                 sf->sf_items_updated_prior = 0;
 296
 297         sf->sf_items_noscrub = 0;
 298         sf->sf_items_igif = 0;
 299 }
 300
 301 static int osd_scrub_file_load(struct osd_scrub *scrub)
 302 {
 303         loff_t  pos  = 0;
 304         int     len  = sizeof(scrub->os_file_disk);
 305         int     rc;
 306
 307         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
 308         if (rc == len) {
 309                 struct scrub_file *sf = &scrub->os_file;
 310
 311                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
 312                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
 313                         CDEBUG(D_LFSCK, "%s: invalid scrub magic "
 314                                "0x%x != 0x%x\n", osd_scrub2name(scrub),
 315                                sf->sf_magic, SCRUB_MAGIC_V1);
 316                         /* Process it as new scrub file. */
 317                         rc = -ENOENT;
 318                 } else {
 319                         rc = 0;
 320                 }
 321         } else if (rc != 0) {
 322                 CDEBUG(D_LFSCK, "%s: fail to load scrub file, "
 323                        "expected = %d: rc = %d\n",
 324                        osd_scrub2name(scrub), len, rc);
 325                 if (rc > 0)
 326                         rc = -EFAULT;
 327         } else {
 328                 /* return -ENOENT for empty scrub file case. */
 329                 rc = -ENOENT;
 330         }
 331
 332         return rc;
 333 }
 334
 335 int osd_scrub_file_store(struct osd_scrub *scrub)
 336 {
 337         struct osd_device *dev;
 338         handle_t          *jh;
 339         loff_t             pos     = 0;
 340         int                len     = sizeof(scrub->os_file_disk);
 341         int                credits;
 342         int                rc;
 343
 344         dev = container_of0(scrub, struct osd_device, od_scrub);
 345         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
 346                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
 347         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits);
 348         if (IS_ERR(jh)) {
 349                 rc = PTR_ERR(jh);
 350                 CDEBUG(D_LFSCK, "%s: fail to start trans for scrub store: "
 351                        "rc = %d\n", osd_scrub2name(scrub), rc);
 352                 return rc;
 353         }
 354
 355         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
 356         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
 357                                       len, 0, &pos, jh);
 358         ldiskfs_journal_stop(jh);
 359         if (rc != 0)
 360                 CDEBUG(D_LFSCK, "%s: fail to store scrub file, "
 361                        "expected = %d: rc = %d\n",
 362                        osd_scrub2name(scrub), len, rc);
 363
 364         scrub->os_time_last_checkpoint = cfs_time_current();
 365         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
 366                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
 367         return rc;
 368 }
 369
 370 static int
 371 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
 372                      struct inode *inode, const struct lu_fid *fid)
 373 {
 374         struct filter_fid_old   *ff      = &info->oti_ff;
 375         struct dentry           *dentry  = &info->oti_obj_dentry;
 376         struct lu_fid           *tfid    = &info->oti_fid;
 377         handle_t                *jh;
 378         int                      size    = 0;
 379         int                      rc;
 380         bool                     reset   = false;
 381         ENTRY;
 382
 383         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
 384                 RETURN(0);
 385
 386         if (fid_is_idif(fid) && dev->od_index_in_idif == 0) {
 387                 struct ost_id *oi = &info->oti_ostid;
 388
 389                 fid_to_ostid(fid, oi);
 390                 ostid_to_fid(tfid, oi, 0);
 391         } else {
 392                 *tfid = *fid;
 393         }
 394
 395         /* We want the LMA to fit into the 256-byte OST inode, so operate
 396          * as following:
 397          * 1) read old XATTR_NAME_FID and save the parent FID;
 398          * 2) delete the old XATTR_NAME_FID;
 399          * 3) make new LMA and add it;
 400          * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
 401          *
 402          * Making the LMA to fit into the 256-byte OST inode can save time for
 403          * normal osd_check_lma() and for other OI scrub scanning in future.
 404          * So it is worth to make some slow conversion here. */
 405         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 406                                 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
 407         if (IS_ERR(jh)) {
 408                 rc = PTR_ERR(jh);
 409                 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
 410                        DFID": rc = %d\n", osd_name(dev), PFID(tfid), rc);
 411                 RETURN(rc);
 412         }
 413
 414         /* 1) read old XATTR_NAME_FID and save the parent FID */
 415         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 416         if (rc == sizeof(*ff)) {
 417                 /* 2) delete the old XATTR_NAME_FID */
 418                 ll_vfs_dq_init(inode);
 419                 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
 420                 if (rc)
 421                         GOTO(stop, rc);
 422
 423                 reset = true;
 424         } else if (rc != -ENODATA && rc != sizeof(struct filter_fid)) {
 425                 GOTO(stop, rc = -EINVAL);
 426         }
 427
 428         /* 3) make new LMA and add it */
 429         rc = osd_ea_fid_set(info, inode, tfid, LMAC_FID_ON_OST, 0);
 430         if (reset) {
 431                 if (rc)
 432                         /* If failed, we should try to add the old back. */
 433                         size = sizeof(*ff);
 434                 else
 435                         /* The new PFID EA will only contains ::ff_parent */
 436                         size = sizeof(ff->ff_parent);
 437         }
 438
 439         /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
 440         if (size > 0) {
 441                 int rc1;
 442
 443                 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
 444                                       XATTR_CREATE);
 445                 if (rc1 != 0 && rc == 0)
 446                         rc = rc1;
 447         }
 448
 449         GOTO(stop, rc);
 450
 451 stop:
 452         ldiskfs_journal_stop(jh);
 453         if (rc < 0)
 454                 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
 455                        osd_name(dev), PFID(tfid), rc);
 456         return rc;
 457 }
 458
 459 static int
 460 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
 461                        struct osd_idmap_cache *oic, int val)
 462 {
 463         struct osd_scrub             *scrub  = &dev->od_scrub;
 464         struct scrub_file            *sf     = &scrub->os_file;
 465         struct lu_fid                *fid    = &oic->oic_fid;
 466         struct osd_inode_id          *lid    = &oic->oic_lid;
 467         struct osd_inode_id          *lid2   = &info->oti_id;
 468         struct osd_inconsistent_item *oii    = NULL;
 469         struct inode                 *inode  = NULL;
 470         int                           ops    = DTO_INDEX_UPDATE;
 471         int                           rc;
 472         bool                          converted = false;
 473         bool                          exist     = false;
 474         ENTRY;
 475
 476         down_write(&scrub->os_rwsem);
 477         scrub->os_new_checked++;
 478         if (val < 0)
 479                 GOTO(out, rc = val);
 480
 481         if (scrub->os_in_prior)
 482                 oii = list_entry(oic, struct osd_inconsistent_item,
 483                                  oii_cache);
 484
 485         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
 486                 GOTO(out, rc = 0);
 487
 488         if (fid_is_igif(fid))
 489                 sf->sf_items_igif++;
 490
 491         if (val == SCRUB_NEXT_OSTOBJ_OLD) {
 492                 inode = osd_iget(info, dev, lid);
 493                 if (IS_ERR(inode)) {
 494                         rc = PTR_ERR(inode);
 495                         /* Someone removed the inode. */
 496                         if (rc == -ENOENT || rc == -ESTALE)
 497                                 rc = 0;
 498                         GOTO(out, rc);
 499                 }
 500
 501                 /* The inode has been reused as EA inode, ignore it. */
 502                 if (unlikely(osd_is_ea_inode(inode)))
 503                         GOTO(out, rc = 0);
 504
 505                 sf->sf_flags |= SF_UPGRADE;
 506                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
 507                 dev->od_check_ff = 1;
 508                 rc = osd_scrub_convert_ff(info, dev, inode, fid);
 509                 if (rc != 0)
 510                         GOTO(out, rc);
 511
 512                 converted = true;
 513         }
 514
 515         if ((val == SCRUB_NEXT_NOLMA) &&
 516             (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
 517                 GOTO(out, rc = 0);
 518
 519         if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA)) {
 520                 ops = DTO_INDEX_INSERT;
 521
 522                 goto iget;
 523         }
 524
 525         rc = osd_oi_lookup(info, dev, fid, lid2,
 526                 (val == SCRUB_NEXT_OSTOBJ ||
 527                  val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 528         if (rc != 0) {
 529                 if (rc == -ENOENT)
 530                         ops = DTO_INDEX_INSERT;
 531                 else if (rc != -ESTALE)
 532                         GOTO(out, rc);
 533
 534 iget:
 535                 if (inode == NULL) {
 536                         inode = osd_iget(info, dev, lid);
 537                         if (IS_ERR(inode)) {
 538                                 rc = PTR_ERR(inode);
 539                                 /* Someone removed the inode. */
 540                                 if (rc == -ENOENT || rc == -ESTALE)
 541                                         rc = 0;
 542                                 GOTO(out, rc);
 543                         }
 544
 545                         /* The inode has been reused as EA inode, ignore it. */
 546                         if (unlikely(osd_is_ea_inode(inode)))
 547                                 GOTO(out, rc = 0);
 548                 }
 549
 550                 if (!scrub->os_partial_scan)
 551                         scrub->os_full_speed = 1;
 552
 553                 switch (val) {
 554                 case SCRUB_NEXT_NOLMA:
 555                         sf->sf_flags |= SF_UPGRADE;
 556                         if (!(sf->sf_param & SP_DRYRUN)) {
 557                                 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
 558                                 if (rc != 0)
 559                                         GOTO(out, rc);
 560                         }
 561
 562                         if (!(sf->sf_flags & SF_INCONSISTENT))
 563                                 dev->od_igif_inoi = 0;
 564                         break;
 565                 case SCRUB_NEXT_OSTOBJ:
 566                         sf->sf_flags |= SF_INCONSISTENT;
 567                 case SCRUB_NEXT_OSTOBJ_OLD:
 568                         break;
 569                 default:
 570                         break;
 571                 }
 572         } else if (osd_id_eq(lid, lid2)) {
 573                 if (converted)
 574                         sf->sf_items_updated++;
 575
 576                 GOTO(out, rc = 0);
 577         } else {
 578                 if (!scrub->os_partial_scan)
 579                         scrub->os_full_speed = 1;
 580
 581                 sf->sf_flags |= SF_INCONSISTENT;
 582
 583                 /* XXX: If the device is restored from file-level backup, then
 584                  *      some IGIFs may have been already in OI files, and some
 585                  *      may be not yet. Means upgrading from 1.8 may be partly
 586                  *      processed, but some clients may hold some immobilized
 587                  *      IGIFs, and use them to access related objects. Under
 588                  *      such case, OSD does not know whether an given IGIF has
 589                  *      been processed or to be processed, and it also cannot
 590                  *      generate local ino#/gen# directly from the immobilized
 591                  *      IGIF because of the backup/restore. Then force OSD to
 592                  *      lookup the given IGIF in OI files, and if no entry,
 593                  *      then ask the client to retry after upgrading completed.
 594                  *      No better choice. */
 595                 dev->od_igif_inoi = 1;
 596         }
 597
 598         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
 599                         (val == SCRUB_NEXT_OSTOBJ ||
 600                          val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0,
 601                         &exist);
 602         if (rc == 0) {
 603                 if (scrub->os_in_prior)
 604                         sf->sf_items_updated_prior++;
 605                 else
 606                         sf->sf_items_updated++;
 607
 608                 if (ops == DTO_INDEX_INSERT && val == 0 && !exist) {
 609                         int idx = osd_oi_fid2idx(dev, fid);
 610
 611                         sf->sf_flags |= SF_RECREATED;
 612                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 613                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 614                 }
 615         }
 616
 617         GOTO(out, rc);
 618
 619 out:
 620         if (rc < 0) {
 621                 sf->sf_items_failed++;
 622                 if (sf->sf_pos_first_inconsistent == 0 ||
 623                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 624                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 625         } else {
 626                 rc = 0;
 627         }
 628
 629         /* There may be conflict unlink during the OI scrub,
 630          * if happend, then remove the new added OI mapping. */
 631         if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
 632             unlikely(ldiskfs_test_inode_state(inode,
 633                                               LDISKFS_STATE_LUSTRE_DESTROY)))
 634                 osd_scrub_refresh_mapping(info, dev, fid, lid,
 635                                 DTO_INDEX_DELETE, false,
 636                                 (val == SCRUB_NEXT_OSTOBJ ||
 637                                  val == SCRUB_NEXT_OSTOBJ_OLD) ?
 638                                 OI_KNOWN_ON_OST : 0, NULL);
 639         up_write(&scrub->os_rwsem);
 640
 641         if (inode != NULL && !IS_ERR(inode))
 642                 iput(inode);
 643
 644         if (oii != NULL) {
 645                 spin_lock(&scrub->os_lock);
 646                 if (likely(!list_empty(&oii->oii_list)))
 647                         list_del(&oii->oii_list);
 648                 spin_unlock(&scrub->os_lock);
 649
 650                 OBD_FREE_PTR(oii);
 651         }
 652
 653         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 654 }
 655
 656 /* OI scrub APIs */
 657
 658 static int osd_scrub_prep(struct osd_device *dev)
 659 {
 660         struct osd_scrub     *scrub  = &dev->od_scrub;
 661         struct ptlrpc_thread *thread = &scrub->os_thread;
 662         struct scrub_file    *sf     = &scrub->os_file;
 663         __u32                 flags  = scrub->os_start_flags;
 664         int                   rc;
 665         bool                  drop_dryrun = false;
 666         ENTRY;
 667
 668         CDEBUG(D_LFSCK, "%s: OI scrub prep, flags = 0x%x\n",
 669                osd_scrub2name(scrub), flags);
 670
 671         down_write(&scrub->os_rwsem);
 672         if (flags & SS_SET_FAILOUT)
 673                 sf->sf_param |= SP_FAILOUT;
 674         else if (flags & SS_CLEAR_FAILOUT)
 675                 sf->sf_param &= ~SP_FAILOUT;
 676
 677         if (flags & SS_SET_DRYRUN) {
 678                 sf->sf_param |= SP_DRYRUN;
 679         } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
 680                 sf->sf_param &= ~SP_DRYRUN;
 681                 drop_dryrun = true;
 682         }
 683
 684         if (flags & SS_RESET)
 685                 osd_scrub_file_reset(scrub,
 686                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
 687
 688         if (flags & SS_AUTO_FULL) {
 689                 scrub->os_full_speed = 1;
 690                 scrub->os_partial_scan = 0;
 691                 sf->sf_flags |= SF_AUTO;
 692         } else if (flags & SS_AUTO_PARTIAL) {
 693                 scrub->os_full_speed = 0;
 694                 scrub->os_partial_scan = 1;
 695                 sf->sf_flags |= SF_AUTO;
 696         } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
 697                                    SF_UPGRADE)) {
 698                 scrub->os_full_speed = 1;
 699                 scrub->os_partial_scan = 0;
 700         } else {
 701                 scrub->os_full_speed = 0;
 702                 scrub->os_partial_scan = 0;
 703         }
 704
 705         spin_lock(&scrub->os_lock);
 706         scrub->os_in_prior = 0;
 707         scrub->os_waiting = 0;
 708         scrub->os_paused = 0;
 709         scrub->os_in_join = 0;
 710         scrub->os_full_scrub = 0;
 711         spin_unlock(&scrub->os_lock);
 712         scrub->os_new_checked = 0;
 713         if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
 714                 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
 715         else if (sf->sf_pos_last_checkpoint != 0)
 716                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 717         else
 718                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 719
 720         scrub->os_pos_current = sf->sf_pos_latest_start;
 721         sf->sf_status = SS_SCANNING;
 722         sf->sf_time_latest_start = cfs_time_current_sec();
 723         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 724         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
 725         rc = osd_scrub_file_store(scrub);
 726         if (rc == 0) {
 727                 spin_lock(&scrub->os_lock);
 728                 thread_set_flags(thread, SVC_RUNNING);
 729                 spin_unlock(&scrub->os_lock);
 730                 wake_up_all(&thread->t_ctl_waitq);
 731         }
 732         up_write(&scrub->os_rwsem);
 733
 734         RETURN(rc);
 735 }
 736
 737 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
 738 {
 739         struct scrub_file *sf = &scrub->os_file;
 740         int                rc;
 741
 742         if (likely(cfs_time_before(cfs_time_current(),
 743                                    scrub->os_time_next_checkpoint) ||
 744                    scrub->os_new_checked == 0))
 745                 return 0;
 746
 747         down_write(&scrub->os_rwsem);
 748         sf->sf_items_checked += scrub->os_new_checked;
 749         scrub->os_new_checked = 0;
 750         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 751         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 752         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 753                                             scrub->os_time_last_checkpoint);
 754         rc = osd_scrub_file_store(scrub);
 755         up_write(&scrub->os_rwsem);
 756
 757         return rc;
 758 }
 759
 760 static int osd_scrub_post(struct osd_scrub *scrub, int result)
 761 {
 762         struct scrub_file *sf = &scrub->os_file;
 763         int rc;
 764         ENTRY;
 765
 766         CDEBUG(D_LFSCK, "%s: OI scrub post, result = %d\n",
 767                osd_scrub2name(scrub), result);
 768
 769         down_write(&scrub->os_rwsem);
 770         spin_lock(&scrub->os_lock);
 771         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 772         spin_unlock(&scrub->os_lock);
 773         if (scrub->os_new_checked > 0) {
 774                 sf->sf_items_checked += scrub->os_new_checked;
 775                 scrub->os_new_checked = 0;
 776                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 777         }
 778         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 779         if (result > 0) {
 780                 struct osd_device *dev =
 781                         container_of0(scrub, struct osd_device, od_scrub);
 782
 783                 dev->od_igif_inoi = 1;
 784                 dev->od_check_ff = 0;
 785                 sf->sf_status = SS_COMPLETED;
 786                 if (!(sf->sf_param & SP_DRYRUN)) {
 787                         memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 788                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 789                                           SF_UPGRADE | SF_AUTO);
 790                 }
 791                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 792                 sf->sf_success_count++;
 793         } else if (result == 0) {
 794                 if (scrub->os_paused)
 795                         sf->sf_status = SS_PAUSED;
 796                 else
 797                         sf->sf_status = SS_STOPPED;
 798         } else {
 799                 sf->sf_status = SS_FAILED;
 800         }
 801         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 802                                             scrub->os_time_last_checkpoint);
 803         rc = osd_scrub_file_store(scrub);
 804         up_write(&scrub->os_rwsem);
 805
 806         RETURN(rc < 0 ? rc : result);
 807 }
 808
 809 /* iteration engine */
 810
 811 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 812                                    struct osd_device *dev,
 813                                    struct osd_iit_param *param,
 814                                    struct osd_idmap_cache **oic,
 815                                    const bool noslot);
 816
 817 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 818                                    struct osd_device *dev,
 819                                    struct osd_iit_param *param,
 820                                    struct osd_idmap_cache *oic,
 821                                    bool *noslot, int rc);
 822
 823 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 824 {
 825         __u32 offset;
 826
 827 again:
 828         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 829                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 830         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 831                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 832                 return SCRUB_NEXT_BREAK;
 833         }
 834
 835         offset = param->offset++;
 836         if (unlikely(*pos == param->gbase + offset && *pos != param->start)) {
 837                 /* We should NOT find the same object more than once. */
 838                 CERROR("%s: scan the same object multiple times at the pos: "
 839                        "group = %u, base = %u, offset = %u, start = %u\n",
 840                        param->sb->s_id, (__u32)param->bg, param->gbase,
 841                        offset, param->start);
 842                 goto again;
 843         }
 844
 845         *pos = param->gbase + offset;
 846         return 0;
 847 }
 848
 849 /**
 850  * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
 851  * \retval 0: FID-on-MDT
 852  */
 853 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
 854                                       struct osd_device *dev,
 855                                       struct lu_fid *fid)
 856 {
 857         /* XXX: The initial OI scrub will scan the top level /O to generate
 858          *      a small local FLDB according to the <seq>. If the given FID
 859          *      is in the local FLDB, then it is FID-on-OST; otherwise it's
 860          *      quite possible for FID-on-MDT. */
 861         if (dev->od_is_ost)
 862                 return SCRUB_NEXT_OSTOBJ_OLD;
 863         else
 864                 return 0;
 865 }
 866
 867 static int osd_scrub_get_fid(struct osd_thread_info *info,
 868                              struct osd_device *dev, struct inode *inode,
 869                              struct lu_fid *fid, bool scrub)
 870 {
 871         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
 872         int rc;
 873         bool has_lma = false;
 874
 875         rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
 876                          &info->oti_ost_attrs);
 877         if (rc == 0) {
 878                 has_lma = true;
 879                 if (lma->lma_compat & LMAC_NOT_IN_OI ||
 880                     lma->lma_incompat & LMAI_AGENT)
 881                         return SCRUB_NEXT_CONTINUE;
 882
 883                 *fid = lma->lma_self_fid;
 884                 if (!scrub)
 885                         return 0;
 886
 887                 if (lma->lma_compat & LMAC_FID_ON_OST)
 888                         return SCRUB_NEXT_OSTOBJ;
 889
 890                 if (fid_is_idif(fid))
 891                         return SCRUB_NEXT_OSTOBJ_OLD;
 892
 893                 /* For local object. */
 894                 if (fid_is_internal(fid))
 895                         return 0;
 896
 897                 /* For external visible MDT-object with non-normal FID. */
 898                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
 899                         return 0;
 900
 901                 /* For the object with normal FID, it may be MDT-object,
 902                  * or may be 2.4 OST-object, need further distinguish.
 903                  * Fall through to next section. */
 904         }
 905
 906         if (rc == -ENODATA || rc == 0) {
 907                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
 908                 if (rc == 0) {
 909                         if (scrub)
 910                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
 911                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 912                         return rc;
 913                 }
 914
 915                 if (rc > 0) {
 916                         if (!has_lma)
 917                                 /* It is FID-on-OST, but we do not know how
 918                                  * to generate its FID, ignore it directly. */
 919                                 rc = SCRUB_NEXT_CONTINUE;
 920                         else
 921                                 /* It is 2.4 OST-object. */
 922                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 923                         return rc;
 924                 }
 925
 926                 if (rc != -ENODATA)
 927                         return rc;
 928
 929                 if (!has_lma) {
 930                         if (dev->od_scrub.os_convert_igif) {
 931                                 lu_igif_build(fid, inode->i_ino,
 932                                               inode->i_generation);
 933                                 if (scrub)
 934                                         rc = SCRUB_NEXT_NOLMA;
 935                                 else
 936                                         rc = 0;
 937                         } else {
 938                                 /* It may be FID-on-OST, or may be FID for
 939                                  * non-MDT0, anyway, we do not know how to
 940                                  * generate its FID, ignore it directly. */
 941                                 rc = SCRUB_NEXT_CONTINUE;
 942                         }
 943                         return rc;
 944                 }
 945
 946                 /* For OI scrub case only: the object has LMA but has no ff
 947                  * (or ff crashed). It may be MDT-object, may be OST-object
 948                  * with crashed ff. The last check is local FLDB. */
 949                 rc = osd_scrub_check_local_fldb(info, dev, fid);
 950         }
 951
 952         return rc;
 953 }
 954
 955 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 956                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 957                         struct super_block *sb, bool scrub)
 958 {
 959         struct inode *inode;
 960         int           rc;
 961         ENTRY;
 962
 963         /* Not handle the backend root object and agent parent object.
 964          * They are neither visible to namespace nor have OI mappings. */
 965         if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
 966                      is_remote_parent_ino(dev, pos)))
 967                 RETURN(SCRUB_NEXT_CONTINUE);
 968
 969         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 970         inode = osd_iget(info, dev, lid);
 971         if (IS_ERR(inode)) {
 972                 rc = PTR_ERR(inode);
 973                 /* The inode may be removed after bitmap searching, or the
 974                  * file is new created without inode initialized yet. */
 975                 if (rc == -ENOENT || rc == -ESTALE)
 976                         RETURN(SCRUB_NEXT_CONTINUE);
 977
 978                 CDEBUG(D_LFSCK, "%s: fail to read inode, ino# = %u: "
 979                        "rc = %d\n", osd_dev2name(dev), pos, rc);
 980                 RETURN(rc);
 981         }
 982
 983         /* It is an EA inode, no OI mapping for it, skip it. */
 984         if (osd_is_ea_inode(inode))
 985                 GOTO(put, rc = SCRUB_NEXT_CONTINUE);
 986
 987         if (scrub &&
 988             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 989                 /* Only skip it for the first OI scrub accessing. */
 990                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 991                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
 992         }
 993
 994         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
 995
 996         GOTO(put, rc);
 997
 998 put:
 999         iput(inode);
1000         return rc;
1001 }
1002
1003 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
1004                           struct osd_iit_param *param,
1005                           struct osd_idmap_cache **oic, const bool noslot)
1006 {
1007         struct osd_scrub     *scrub  = &dev->od_scrub;
1008         struct ptlrpc_thread *thread = &scrub->os_thread;
1009         struct lu_fid        *fid;
1010         struct osd_inode_id  *lid;
1011         int                   rc;
1012
1013         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
1014                 struct l_wait_info lwi;
1015
1016                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
1017                 if (likely(lwi.lwi_timeout > 0))
1018                         l_wait_event(thread->t_ctl_waitq,
1019                                 !list_empty(&scrub->os_inconsistent_items) ||
1020                                 !thread_is_running(thread),
1021                                 &lwi);
1022         }
1023
1024         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
1025                 spin_lock(&scrub->os_lock);
1026                 thread_set_flags(thread, SVC_STOPPING);
1027                 spin_unlock(&scrub->os_lock);
1028                 return SCRUB_NEXT_CRASH;
1029         }
1030
1031         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
1032                 return SCRUB_NEXT_FATAL;
1033
1034         if (unlikely(!thread_is_running(thread)))
1035                 return SCRUB_NEXT_EXIT;
1036
1037         if (!list_empty(&scrub->os_inconsistent_items)) {
1038                 spin_lock(&scrub->os_lock);
1039                 if (likely(!list_empty(&scrub->os_inconsistent_items))) {
1040                         struct osd_inconsistent_item *oii;
1041
1042                         oii = list_entry(scrub->os_inconsistent_items.next,
1043                                 struct osd_inconsistent_item, oii_list);
1044                         spin_unlock(&scrub->os_lock);
1045
1046                         *oic = &oii->oii_cache;
1047                         scrub->os_in_prior = 1;
1048
1049                         return 0;
1050                 }
1051                 spin_unlock(&scrub->os_lock);
1052         }
1053
1054         if (noslot)
1055                 return SCRUB_NEXT_WAIT;
1056
1057         rc = osd_iit_next(param, &scrub->os_pos_current);
1058         if (rc != 0)
1059                 return rc;
1060
1061         *oic = &scrub->os_oic;
1062         fid = &(*oic)->oic_fid;
1063         lid = &(*oic)->oic_lid;
1064         rc = osd_iit_iget(info, dev, fid, lid,
1065                           scrub->os_pos_current, param->sb, true);
1066         return rc;
1067 }
1068
1069 static int osd_preload_next(struct osd_thread_info *info,
1070                             struct osd_device *dev, struct osd_iit_param *param,
1071                             struct osd_idmap_cache **oic, const bool noslot)
1072 {
1073         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1074         struct osd_scrub *scrub = &dev->od_scrub;
1075         struct ptlrpc_thread *thread = &scrub->os_thread;
1076         int rc;
1077
1078         if (thread_is_running(thread) &&
1079             ooc->ooc_pos_preload >= scrub->os_pos_current)
1080                 return SCRUB_NEXT_EXIT;
1081
1082         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1083         if (rc)
1084                 return rc;
1085
1086         rc = osd_iit_iget(info, dev,
1087                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1088                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1089                           ooc->ooc_pos_preload, param->sb, false);
1090         return rc;
1091 }
1092
1093 static inline int
1094 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1095 {
1096         spin_lock(&scrub->os_lock);
1097         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1098             !list_empty(&scrub->os_inconsistent_items) ||
1099             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1100                 scrub->os_waiting = 0;
1101         else
1102                 scrub->os_waiting = 1;
1103         spin_unlock(&scrub->os_lock);
1104
1105         return !scrub->os_waiting;
1106 }
1107
1108 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1109                           struct osd_iit_param *param,
1110                           struct osd_idmap_cache *oic, bool *noslot, int rc)
1111 {
1112         struct l_wait_info       lwi    = { 0 };
1113         struct osd_scrub        *scrub  = &dev->od_scrub;
1114         struct scrub_file       *sf     = &scrub->os_file;
1115         struct ptlrpc_thread    *thread = &scrub->os_thread;
1116         struct osd_otable_it    *it     = dev->od_otable_it;
1117         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
1118
1119         switch (rc) {
1120         case SCRUB_NEXT_NOSCRUB:
1121                 down_write(&scrub->os_rwsem);
1122                 scrub->os_new_checked++;
1123                 sf->sf_items_noscrub++;
1124                 up_write(&scrub->os_rwsem);
1125         case SCRUB_NEXT_CONTINUE:
1126         case SCRUB_NEXT_WAIT:
1127                 goto wait;
1128         }
1129
1130         rc = osd_scrub_check_update(info, dev, oic, rc);
1131         if (rc != 0) {
1132                 scrub->os_in_prior = 0;
1133                 return rc;
1134         }
1135
1136         rc = osd_scrub_checkpoint(scrub);
1137         if (rc != 0) {
1138                 CDEBUG(D_LFSCK, "%s: fail to checkpoint, pos = %u: "
1139                        "rc = %d\n", osd_scrub2name(scrub),
1140                        scrub->os_pos_current, rc);
1141                 /* Continue, as long as the scrub itself can go ahead. */
1142         }
1143
1144         if (scrub->os_in_prior) {
1145                 scrub->os_in_prior = 0;
1146                 return 0;
1147         }
1148
1149 wait:
1150         if (it != NULL && it->ooi_waiting && ooc != NULL &&
1151             ooc->ooc_pos_preload < scrub->os_pos_current) {
1152                 spin_lock(&scrub->os_lock);
1153                 it->ooi_waiting = 0;
1154                 wake_up_all(&thread->t_ctl_waitq);
1155                 spin_unlock(&scrub->os_lock);
1156         }
1157
1158         if (rc == SCRUB_NEXT_CONTINUE)
1159                 return 0;
1160
1161         if (scrub->os_full_speed || !ooc || osd_scrub_has_window(scrub, ooc)) {
1162                 *noslot = false;
1163                 return 0;
1164         }
1165
1166         if (it != NULL)
1167                 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1168                              &lwi);
1169
1170         if (!ooc || osd_scrub_has_window(scrub, ooc))
1171                 *noslot = false;
1172         else
1173                 *noslot = true;
1174         return 0;
1175 }
1176
1177 static int osd_preload_exec(struct osd_thread_info *info,
1178                             struct osd_device *dev, struct osd_iit_param *param,
1179                             struct osd_idmap_cache *oic, bool *noslot, int rc)
1180 {
1181         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1182
1183         if (rc == 0) {
1184                 ooc->ooc_cached_items++;
1185                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1186                                         ~OSD_OTABLE_IT_CACHE_MASK;
1187         }
1188         return rc > 0 ? 0 : rc;
1189 }
1190
1191 #define SCRUB_IT_ALL    1
1192 #define SCRUB_IT_CRASH  2
1193
1194 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1195                            bool inconsistent)
1196 {
1197         struct osd_scrub     *scrub  = &dev->od_scrub;
1198         struct ptlrpc_thread *thread = &scrub->os_thread;
1199         struct scrub_file    *sf     = &scrub->os_file;
1200         int                   rc;
1201         ENTRY;
1202
1203         LASSERT(!(flags & SS_AUTO_PARTIAL));
1204
1205         down_write(&scrub->os_rwsem);
1206         scrub->os_in_join = 1;
1207         if (flags & SS_SET_FAILOUT)
1208                 sf->sf_param |= SP_FAILOUT;
1209         else if (flags & SS_CLEAR_FAILOUT)
1210                 sf->sf_param &= ~SP_FAILOUT;
1211
1212         if (flags & SS_SET_DRYRUN)
1213                 sf->sf_param |= SP_DRYRUN;
1214         else if (flags & SS_CLEAR_DRYRUN)
1215                 sf->sf_param &= ~SP_DRYRUN;
1216
1217         if (flags & SS_RESET) {
1218                 osd_scrub_file_reset(scrub,
1219                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1220                         inconsistent ? SF_INCONSISTENT : 0);
1221                 sf->sf_status = SS_SCANNING;
1222         }
1223
1224         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1225                 scrub->os_full_speed = 1;
1226         else
1227                 scrub->os_full_speed = 0;
1228
1229         if (flags & SS_AUTO_FULL) {
1230                 sf->sf_flags |= SF_AUTO;
1231                 scrub->os_full_speed = 1;
1232         }
1233
1234         scrub->os_new_checked = 0;
1235         if (sf->sf_pos_last_checkpoint != 0)
1236                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1237         else
1238                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1239
1240         scrub->os_pos_current = sf->sf_pos_latest_start;
1241         sf->sf_time_latest_start = cfs_time_current_sec();
1242         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1243         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1244         rc = osd_scrub_file_store(scrub);
1245         if (rc != 0)
1246                 CDEBUG(D_LFSCK, "%s: fail to store scrub file when join "
1247                        "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1248
1249         spin_lock(&scrub->os_lock);
1250         scrub->os_waiting = 0;
1251         scrub->os_paused = 0;
1252         scrub->os_partial_scan = 0;
1253         scrub->os_in_join = 0;
1254         scrub->os_full_scrub = 0;
1255         spin_unlock(&scrub->os_lock);
1256         wake_up_all(&thread->t_ctl_waitq);
1257         up_write(&scrub->os_rwsem);
1258
1259         EXIT;
1260 }
1261
1262 static int osd_inode_iteration(struct osd_thread_info *info,
1263                                struct osd_device *dev, __u32 max, bool preload)
1264 {
1265         struct osd_scrub     *scrub  = &dev->od_scrub;
1266         struct ptlrpc_thread *thread = &scrub->os_thread;
1267         struct scrub_file    *sf     = &scrub->os_file;
1268         osd_iit_next_policy   next;
1269         osd_iit_exec_policy   exec;
1270         __u32                *pos;
1271         __u32                *count;
1272         struct osd_iit_param *param;
1273         struct l_wait_info    lwi    = { 0 };
1274         __u32                 limit =
1275                 le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count);
1276         int                   rc;
1277         bool                  noslot = true;
1278         ENTRY;
1279
1280         if (preload)
1281                 goto full;
1282
1283         param = &scrub->os_iit_param;
1284         memset(param, 0, sizeof(*param));
1285         param->sb = osd_sb(dev);
1286
1287         while (scrub->os_partial_scan && !scrub->os_in_join) {
1288                 struct osd_idmap_cache *oic = NULL;
1289
1290                 rc = osd_scrub_next(info, dev, param, &oic, noslot);
1291                 switch (rc) {
1292                 case SCRUB_NEXT_EXIT:
1293                         RETURN(0);
1294                 case SCRUB_NEXT_CRASH:
1295                         RETURN(SCRUB_IT_CRASH);
1296                 case SCRUB_NEXT_FATAL:
1297                         RETURN(-EINVAL);
1298                 case SCRUB_NEXT_WAIT: {
1299                         struct kstatfs *ksfs = &info->oti_ksfs;
1300                         __u64 saved_flags;
1301
1302                         if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1303                             unlikely(sf->sf_items_updated_prior == 0))
1304                                 goto wait;
1305
1306                         if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1307                             scrub->os_full_scrub) {
1308                                 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1309                                                true);
1310                                 goto full;
1311                         }
1312
1313                         rc = param->sb->s_op->statfs(param->sb->s_root, ksfs);
1314                         if (rc == 0) {
1315                                 __u64 used = ksfs->f_files - ksfs->f_ffree;
1316
1317                                 do_div(used, sf->sf_items_updated_prior);
1318                                 /* If we hit too much inconsistent OI
1319                                  * mappings during the partial scan,
1320                                  * then scan the device completely. */
1321                                 if (used < dev->od_full_scrub_ratio) {
1322                                         osd_scrub_join(dev,
1323                                                 SS_AUTO_FULL | SS_RESET, true);
1324                                         goto full;
1325                                 }
1326                         }
1327
1328 wait:
1329                         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1330                             cfs_fail_val > 0)
1331                                 continue;
1332
1333                         saved_flags = sf->sf_flags;
1334                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1335                                           SF_UPGRADE | SF_AUTO);
1336                         sf->sf_status = SS_COMPLETED;
1337                         l_wait_event(thread->t_ctl_waitq,
1338                                      !thread_is_running(thread) ||
1339                                      !scrub->os_partial_scan ||
1340                                      scrub->os_in_join ||
1341                                      !list_empty(&scrub->os_inconsistent_items),
1342                                      &lwi);
1343                         sf->sf_flags = saved_flags;
1344                         sf->sf_status = SS_SCANNING;
1345
1346                         if (unlikely(!thread_is_running(thread)))
1347                                 RETURN(0);
1348
1349                         if (!scrub->os_partial_scan || scrub->os_in_join)
1350                                 goto full;
1351
1352                         continue;
1353                 }
1354                 default:
1355                         LASSERTF(rc == 0, "rc = %d\n", rc);
1356
1357                         osd_scrub_exec(info, dev, param, oic, &noslot, rc);
1358                         break;
1359                 }
1360         }
1361
1362 full:
1363         if (!preload) {
1364                 l_wait_event(thread->t_ctl_waitq,
1365                              !thread_is_running(thread) || !scrub->os_in_join,
1366                              &lwi);
1367
1368                 if (unlikely(!thread_is_running(thread)))
1369                         RETURN(0);
1370         }
1371
1372         noslot = false;
1373         if (!preload) {
1374                 next = osd_scrub_next;
1375                 exec = osd_scrub_exec;
1376                 pos = &scrub->os_pos_current;
1377                 count = &scrub->os_new_checked;
1378                 param->start = *pos;
1379                 param->bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param->sb);
1380                 param->offset =
1381                         (*pos - 1) % LDISKFS_INODES_PER_GROUP(param->sb);
1382                 param->gbase =
1383                         1 + param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
1384         } else {
1385                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1386
1387                 next = osd_preload_next;
1388                 exec = osd_preload_exec;
1389                 pos = &ooc->ooc_pos_preload;
1390                 count = &ooc->ooc_cached_items;
1391                 param = &dev->od_otable_it->ooi_iit_param;
1392         }
1393
1394         rc = 0;
1395         while (*pos <= limit && *count < max) {
1396                 struct ldiskfs_group_desc *desc;
1397                 bool next_group = false;
1398
1399                 desc = ldiskfs_get_group_desc(param->sb, param->bg, NULL);
1400                 if (!desc)
1401                         RETURN(-EIO);
1402
1403                 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1404                         next_group = true;
1405                         goto next_group;
1406                 }
1407
1408                 param->bitmap = ldiskfs_read_inode_bitmap(param->sb, param->bg);
1409                 if (!param->bitmap) {
1410                         CERROR("%s: fail to read bitmap for %u, "
1411                                "scrub will stop, urgent mode\n",
1412                                osd_scrub2name(scrub), (__u32)param->bg);
1413                         RETURN(-EIO);
1414                 }
1415
1416                 do {
1417                         struct osd_idmap_cache *oic = NULL;
1418
1419                         if (param->offset +
1420                                 ldiskfs_itable_unused_count(param->sb, desc) >=
1421                             LDISKFS_INODES_PER_GROUP(param->sb)) {
1422                                 next_group = true;
1423                                 goto next_group;
1424                         }
1425
1426                         rc = next(info, dev, param, &oic, noslot);
1427                         switch (rc) {
1428                         case SCRUB_NEXT_BREAK:
1429                                 next_group = true;
1430                                 goto next_group;
1431                         case SCRUB_NEXT_EXIT:
1432                                 brelse(param->bitmap);
1433                                 RETURN(0);
1434                         case SCRUB_NEXT_CRASH:
1435                                 brelse(param->bitmap);
1436                                 RETURN(SCRUB_IT_CRASH);
1437                         case SCRUB_NEXT_FATAL:
1438                                 brelse(param->bitmap);
1439                                 RETURN(-EINVAL);
1440                         }
1441
1442                         rc = exec(info, dev, param, oic, &noslot, rc);
1443                 } while (!rc && *pos <= limit && *count < max);
1444
1445 next_group:
1446                 if (param->bitmap) {
1447                         brelse(param->bitmap);
1448                         param->bitmap = NULL;
1449                 }
1450
1451                 if (rc < 0)
1452                         GOTO(out, rc);
1453
1454                 if (next_group) {
1455                         param->bg++;
1456                         param->offset = 0;
1457                         param->gbase = 1 +
1458                                 param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
1459                         *pos = param->gbase;
1460                         param->start = *pos;
1461                 }
1462         }
1463
1464         if (*pos > limit)
1465                 RETURN(SCRUB_IT_ALL);
1466
1467 out:
1468         RETURN(rc);
1469 }
1470
1471 static int osd_otable_it_preload(const struct lu_env *env,
1472                                  struct osd_otable_it *it)
1473 {
1474         struct osd_device       *dev   = it->ooi_dev;
1475         struct osd_scrub        *scrub = &dev->od_scrub;
1476         struct osd_otable_cache *ooc   = &it->ooi_cache;
1477         int                      rc;
1478         ENTRY;
1479
1480         rc = osd_inode_iteration(osd_oti_get(env), dev,
1481                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1482         if (rc == SCRUB_IT_ALL)
1483                 it->ooi_all_cached = 1;
1484
1485         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1486                 spin_lock(&scrub->os_lock);
1487                 scrub->os_waiting = 0;
1488                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1489                 spin_unlock(&scrub->os_lock);
1490         }
1491
1492         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1493 }
1494
1495 static int osd_scrub_main(void *args)
1496 {
1497         struct lu_env         env;
1498         struct osd_device    *dev    = (struct osd_device *)args;
1499         struct osd_scrub     *scrub  = &dev->od_scrub;
1500         struct ptlrpc_thread *thread = &scrub->os_thread;
1501         int                   rc;
1502         ENTRY;
1503
1504         rc = lu_env_init(&env, LCT_LOCAL);
1505         if (rc != 0) {
1506                 CDEBUG(D_LFSCK, "%s: OI scrub fail to init env: rc = %d\n",
1507                        osd_scrub2name(scrub), rc);
1508                 GOTO(noenv, rc);
1509         }
1510
1511         rc = osd_scrub_prep(dev);
1512         if (rc != 0) {
1513                 CDEBUG(D_LFSCK, "%s: OI scrub fail to scrub prep: rc = %d\n",
1514                        osd_scrub2name(scrub), rc);
1515                 GOTO(out, rc);
1516         }
1517
1518         if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1519                 struct l_wait_info lwi = { 0 };
1520                 struct osd_otable_it *it = dev->od_otable_it;
1521                 struct osd_otable_cache *ooc = &it->ooi_cache;
1522
1523                 l_wait_event(thread->t_ctl_waitq,
1524                              it->ooi_user_ready || !thread_is_running(thread),
1525                              &lwi);
1526                 if (unlikely(!thread_is_running(thread)))
1527                         GOTO(post, rc = 0);
1528
1529                 scrub->os_pos_current = ooc->ooc_pos_preload;
1530         }
1531
1532         CDEBUG(D_LFSCK, "%s: OI scrub start, flags = 0x%x, pos = %u\n",
1533                osd_scrub2name(scrub), scrub->os_start_flags,
1534                scrub->os_pos_current);
1535
1536         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1537         if (unlikely(rc == SCRUB_IT_CRASH))
1538                 GOTO(out, rc = -EINVAL);
1539         GOTO(post, rc);
1540
1541 post:
1542         rc = osd_scrub_post(scrub, rc);
1543         CDEBUG(D_LFSCK, "%s: OI scrub: stop, pos = %u: rc = %d\n",
1544                osd_scrub2name(scrub), scrub->os_pos_current, rc);
1545
1546 out:
1547         while (!list_empty(&scrub->os_inconsistent_items)) {
1548                 struct osd_inconsistent_item *oii;
1549
1550                 oii = list_entry(scrub->os_inconsistent_items.next,
1551                                      struct osd_inconsistent_item, oii_list);
1552                 list_del_init(&oii->oii_list);
1553                 OBD_FREE_PTR(oii);
1554         }
1555         lu_env_fini(&env);
1556
1557 noenv:
1558         spin_lock(&scrub->os_lock);
1559         thread_set_flags(thread, SVC_STOPPED);
1560         wake_up_all(&thread->t_ctl_waitq);
1561         spin_unlock(&scrub->os_lock);
1562         return rc;
1563 }
1564
1565 /* initial OI scrub */
1566
1567 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1568                          struct dentry *, filldir_t filldir);
1569
1570 #ifdef HAVE_FILLDIR_USE_CTX
1571 static int osd_ios_varfid_fill(struct dir_context *buf, const char *name,
1572                                int namelen, loff_t offset, __u64 ino,
1573                                unsigned d_type);
1574 static int osd_ios_lf_fill(struct dir_context *buf, const char *name,
1575                            int namelen, loff_t offset, __u64 ino,
1576                            unsigned d_type);
1577 static int osd_ios_dl_fill(struct dir_context *buf, const char *name,
1578                            int namelen, loff_t offset, __u64 ino,
1579                            unsigned d_type);
1580 static int osd_ios_uld_fill(struct dir_context *buf, const char *name,
1581                             int namelen, loff_t offset, __u64 ino,
1582                             unsigned d_type);
1583 #else
1584 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1585                                loff_t offset, __u64 ino, unsigned d_type);
1586 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1587                            loff_t offset, __u64 ino, unsigned d_type);
1588 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1589                            loff_t offset, __u64 ino, unsigned d_type);
1590 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1591                             loff_t offset, __u64 ino, unsigned d_type);
1592 #endif
1593
1594 static int
1595 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1596                      struct dentry *dentry, filldir_t filldir);
1597 static int
1598 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1599                   struct dentry *dentry, filldir_t filldir);
1600
1601 static int
1602 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1603                      struct dentry *dentry, filldir_t filldir);
1604
1605 enum osd_lf_flags {
1606         OLF_SCAN_SUBITEMS       = 0x0001,
1607         OLF_HIDE_FID            = 0x0002,
1608         OLF_SHOW_NAME           = 0x0004,
1609         OLF_NO_OI               = 0x0008,
1610         OLF_IDX_IN_FID          = 0x0010,
1611 };
1612
1613 struct osd_lf_map {
1614         char            *olm_name;
1615         struct lu_fid    olm_fid;
1616         __u16            olm_flags;
1617         __u16            olm_namelen;
1618         scandir_t        olm_scandir;
1619         filldir_t        olm_filldir;
1620 };
1621
1622 /* Add the new introduced local files in the list in the future. */
1623 static const struct osd_lf_map osd_lf_maps[] = {
1624         /* CATALOGS */
1625         {
1626                 .olm_name       = CATLIST,
1627                 .olm_fid        = {
1628                         .f_seq  = FID_SEQ_LOCAL_FILE,
1629                         .f_oid  = LLOG_CATALOGS_OID,
1630                 },
1631                 .olm_flags      = OLF_SHOW_NAME,
1632                 .olm_namelen    = sizeof(CATLIST) - 1,
1633         },
1634
1635         /* CONFIGS */
1636         {
1637                 .olm_name       = MOUNT_CONFIGS_DIR,
1638                 .olm_fid        = {
1639                         .f_seq  = FID_SEQ_LOCAL_FILE,
1640                         .f_oid  = MGS_CONFIGS_OID,
1641                 },
1642                 .olm_flags      = OLF_SCAN_SUBITEMS,
1643                 .olm_namelen    = sizeof(MOUNT_CONFIGS_DIR) - 1,
1644                 .olm_scandir    = osd_ios_general_scan,
1645                 .olm_filldir    = osd_ios_varfid_fill,
1646         },
1647
1648         /* NIDTBL_VERSIONS */
1649         {
1650                 .olm_name       = MGS_NIDTBL_DIR,
1651                 .olm_flags      = OLF_SCAN_SUBITEMS,
1652                 .olm_namelen    = sizeof(MGS_NIDTBL_DIR) - 1,
1653                 .olm_scandir    = osd_ios_general_scan,
1654                 .olm_filldir    = osd_ios_varfid_fill,
1655         },
1656
1657         /* PENDING */
1658         {
1659                 .olm_name       = "PENDING",
1660                 .olm_namelen    = sizeof("PENDING") - 1,
1661         },
1662
1663         /* ROOT */
1664         {
1665                 .olm_name       = "ROOT",
1666                 .olm_fid        = {
1667                         .f_seq  = FID_SEQ_ROOT,
1668                         .f_oid  = FID_OID_ROOT,
1669                 },
1670                 .olm_flags      = OLF_SCAN_SUBITEMS | OLF_HIDE_FID,
1671                 .olm_namelen    = sizeof("ROOT") - 1,
1672                 .olm_scandir    = osd_ios_ROOT_scan,
1673         },
1674
1675         /* changelog_catalog */
1676         {
1677                 .olm_name       = CHANGELOG_CATALOG,
1678                 .olm_namelen    = sizeof(CHANGELOG_CATALOG) - 1,
1679         },
1680
1681         /* changelog_users */
1682         {
1683                 .olm_name       = CHANGELOG_USERS,
1684                 .olm_namelen    = sizeof(CHANGELOG_USERS) - 1,
1685         },
1686
1687         /* fld */
1688         {
1689                 .olm_name       = "fld",
1690                 .olm_fid        = {
1691                         .f_seq  = FID_SEQ_LOCAL_FILE,
1692                         .f_oid  = FLD_INDEX_OID,
1693                 },
1694                 .olm_flags      = OLF_SHOW_NAME,
1695                 .olm_namelen    = sizeof("fld") - 1,
1696         },
1697
1698         /* last_rcvd */
1699         {
1700                 .olm_name       = LAST_RCVD,
1701                 .olm_fid        = {
1702                         .f_seq  = FID_SEQ_LOCAL_FILE,
1703                         .f_oid  = LAST_RECV_OID,
1704                 },
1705                 .olm_flags      = OLF_SHOW_NAME,
1706                 .olm_namelen    = sizeof(LAST_RCVD) - 1,
1707         },
1708
1709         /* reply_data */
1710         {
1711                 .olm_name       = REPLY_DATA,
1712                 .olm_fid        = {
1713                         .f_seq  = FID_SEQ_LOCAL_FILE,
1714                         .f_oid  = REPLY_DATA_OID,
1715                 },
1716                 .olm_flags      = OLF_SHOW_NAME,
1717                 .olm_namelen    = sizeof(REPLY_DATA) - 1,
1718         },
1719
1720         /* lov_objid */
1721         {
1722                 .olm_name       = LOV_OBJID,
1723                 .olm_fid        = {
1724                         .f_seq  = FID_SEQ_LOCAL_FILE,
1725                         .f_oid  = MDD_LOV_OBJ_OID,
1726                 },
1727                 .olm_flags      = OLF_SHOW_NAME,
1728                 .olm_namelen    = sizeof(LOV_OBJID) - 1,
1729         },
1730
1731         /* lov_objseq */
1732         {
1733                 .olm_name       = LOV_OBJSEQ,
1734                 .olm_fid        = {
1735                         .f_seq  = FID_SEQ_LOCAL_FILE,
1736                         .f_oid  = MDD_LOV_OBJ_OSEQ,
1737                 },
1738                 .olm_flags      = OLF_SHOW_NAME,
1739                 .olm_namelen    = sizeof(LOV_OBJSEQ) - 1,
1740         },
1741
1742         /* quota_master */
1743         {
1744                 .olm_name       = QMT_DIR,
1745                 .olm_flags      = OLF_SCAN_SUBITEMS,
1746                 .olm_namelen    = sizeof(QMT_DIR) - 1,
1747                 .olm_scandir    = osd_ios_general_scan,
1748                 .olm_filldir    = osd_ios_varfid_fill,
1749         },
1750
1751         /* quota_slave */
1752         {
1753                 .olm_name       = QSD_DIR,
1754                 .olm_flags      = OLF_SCAN_SUBITEMS,
1755                 .olm_namelen    = sizeof(QSD_DIR) - 1,
1756                 .olm_scandir    = osd_ios_general_scan,
1757                 .olm_filldir    = osd_ios_varfid_fill,
1758         },
1759
1760         /* seq_ctl */
1761         {
1762                 .olm_name       = "seq_ctl",
1763                 .olm_fid        = {
1764                         .f_seq  = FID_SEQ_LOCAL_FILE,
1765                         .f_oid  = FID_SEQ_CTL_OID,
1766                 },
1767                 .olm_flags      = OLF_SHOW_NAME,
1768                 .olm_namelen    = sizeof("seq_ctl") - 1,
1769         },
1770
1771         /* seq_srv */
1772         {
1773                 .olm_name       = "seq_srv",
1774                 .olm_fid        = {
1775                         .f_seq  = FID_SEQ_LOCAL_FILE,
1776                         .f_oid  = FID_SEQ_SRV_OID,
1777                 },
1778                 .olm_flags      = OLF_SHOW_NAME,
1779                 .olm_namelen    = sizeof("seq_srv") - 1,
1780         },
1781
1782         /* health_check */
1783         {
1784                 .olm_name       = HEALTH_CHECK,
1785                 .olm_fid        = {
1786                         .f_seq  = FID_SEQ_LOCAL_FILE,
1787                         .f_oid  = OFD_HEALTH_CHECK_OID,
1788                 },
1789                 .olm_flags      = OLF_SHOW_NAME,
1790                 .olm_namelen    = sizeof(HEALTH_CHECK) - 1,
1791         },
1792
1793         /* LFSCK */
1794         {
1795                 .olm_name       = LFSCK_DIR,
1796                 .olm_namelen    = sizeof(LFSCK_DIR) - 1,
1797                 .olm_scandir    = osd_ios_general_scan,
1798                 .olm_filldir    = osd_ios_varfid_fill,
1799         },
1800
1801         /* lfsck_bookmark */
1802         {
1803                 .olm_name       = LFSCK_BOOKMARK,
1804                 .olm_namelen    = sizeof(LFSCK_BOOKMARK) - 1,
1805         },
1806
1807         /* lfsck_layout */
1808         {
1809                 .olm_name       = LFSCK_LAYOUT,
1810                 .olm_namelen    = sizeof(LFSCK_LAYOUT) - 1,
1811         },
1812
1813         /* lfsck_namespace */
1814         {
1815                 .olm_name       = LFSCK_NAMESPACE,
1816                 .olm_namelen    = sizeof(LFSCK_NAMESPACE) - 1,
1817         },
1818
1819         /* OBJECTS, upgrade from old device */
1820         {
1821                 .olm_name       = OBJECTS,
1822                 .olm_flags      = OLF_SCAN_SUBITEMS,
1823                 .olm_namelen    = sizeof(OBJECTS) - 1,
1824                 .olm_scandir    = osd_ios_OBJECTS_scan,
1825         },
1826
1827         /* lquota_v2.user, upgrade from old device */
1828         {
1829                 .olm_name       = "lquota_v2.user",
1830                 .olm_namelen    = sizeof("lquota_v2.user") - 1,
1831         },
1832
1833         /* lquota_v2.group, upgrade from old device */
1834         {
1835                 .olm_name       = "lquota_v2.group",
1836                 .olm_namelen    = sizeof("lquota_v2.group") - 1,
1837         },
1838
1839         /* LAST_GROUP, upgrade from old device */
1840         {
1841                 .olm_name       = "LAST_GROUP",
1842                 .olm_fid        = {
1843                         .f_seq  = FID_SEQ_LOCAL_FILE,
1844                         .f_oid  = OFD_LAST_GROUP_OID,
1845                 },
1846                 .olm_flags      = OLF_SHOW_NAME,
1847                 .olm_namelen    = sizeof("LAST_GROUP") - 1,
1848         },
1849
1850         /* committed batchid for cross-MDT operation */
1851         {
1852                 .olm_name       = "BATCHID",
1853                 .olm_fid        = {
1854                         .f_seq  = FID_SEQ_LOCAL_FILE,
1855                         .f_oid  = BATCHID_COMMITTED_OID,
1856                 },
1857                 .olm_flags      = OLF_SHOW_NAME,
1858                 .olm_namelen    = sizeof("BATCHID") - 1,
1859         },
1860
1861         /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1862          * and f_oid = index for their log files.  See lu_update_log{_dir}_fid()
1863          * for more details. */
1864
1865         /* update_log */
1866         {
1867                 .olm_name       = "update_log",
1868                 .olm_fid        = {
1869                         .f_seq  = FID_SEQ_UPDATE_LOG,
1870                 },
1871                 .olm_flags      = OLF_SHOW_NAME | OLF_IDX_IN_FID,
1872                 .olm_namelen    = sizeof("update_log") - 1,
1873         },
1874
1875         /* update_log_dir */
1876         {
1877                 .olm_name       = "update_log_dir",
1878                 .olm_fid        = {
1879                         .f_seq  = FID_SEQ_UPDATE_LOG_DIR,
1880                 },
1881                 .olm_flags      = OLF_SHOW_NAME | OLF_SCAN_SUBITEMS |
1882                                   OLF_IDX_IN_FID,
1883                 .olm_namelen    = sizeof("update_log_dir") - 1,
1884                 .olm_scandir    = osd_ios_general_scan,
1885                 .olm_filldir    = osd_ios_uld_fill,
1886         },
1887
1888         /* lost+found */
1889         {
1890                 .olm_name       = "lost+found",
1891                 .olm_fid        = {
1892                         .f_seq  = FID_SEQ_LOCAL_FILE,
1893                         .f_oid  = OSD_LPF_OID,
1894                 },
1895                 .olm_flags      = OLF_SCAN_SUBITEMS,
1896                 .olm_namelen    = sizeof("lost+found") - 1,
1897                 .olm_scandir    = osd_ios_general_scan,
1898                 .olm_filldir    = osd_ios_lf_fill,
1899         },
1900
1901         {
1902                 .olm_name       = NULL
1903         }
1904 };
1905
1906 /* Add the new introduced files under .lustre/ in the list in the future. */
1907 static const struct osd_lf_map osd_dl_maps[] = {
1908         /* .lustre/fid */
1909         {
1910                 .olm_name       = "fid",
1911                 .olm_fid        = {
1912                         .f_seq  = FID_SEQ_DOT_LUSTRE,
1913                         .f_oid  = FID_OID_DOT_LUSTRE_OBF,
1914                 },
1915                 .olm_namelen    = sizeof("fid") - 1,
1916         },
1917         /* .lustre/lost+found */
1918         {
1919                 .olm_name       = "lost+found",
1920                 .olm_fid        = {
1921                         .f_seq  = FID_SEQ_DOT_LUSTRE,
1922                         .f_oid  = FID_OID_DOT_LUSTRE_LPF,
1923                 },
1924                 .olm_namelen    = sizeof("lost+found") - 1,
1925         },
1926         {
1927                 .olm_name       = NULL
1928         }
1929 };
1930
1931 struct osd_ios_item {
1932         struct list_head oii_list;
1933         struct dentry   *oii_dentry;
1934         scandir_t        oii_scandir;
1935         filldir_t        oii_filldir;
1936 };
1937
1938 struct osd_ios_filldir_buf {
1939 #ifdef HAVE_DIR_CONTEXT
1940         /* please keep it as first member */
1941         struct dir_context       ctx;
1942 #endif
1943         struct osd_thread_info  *oifb_info;
1944         struct osd_device       *oifb_dev;
1945         struct dentry           *oifb_dentry;
1946         int                      oifb_items;
1947 };
1948
1949 static inline struct dentry *
1950 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1951 {
1952         struct dentry *dentry;
1953
1954         dentry = ll_lookup_one_len(name, parent, namelen);
1955         if (IS_ERR(dentry)) {
1956                 int rc = PTR_ERR(dentry);
1957
1958                 if (rc != -ENOENT)
1959                         CERROR("Fail to find %.*s in %.*s (%lu/%u): rc = %d\n",
1960                                namelen, name, parent->d_name.len,
1961                                parent->d_name.name, parent->d_inode->i_ino,
1962                                parent->d_inode->i_generation, rc);
1963
1964                 return dentry;
1965         }
1966
1967         if (dentry->d_inode == NULL) {
1968                 dput(dentry);
1969                 return ERR_PTR(-ENOENT);
1970         }
1971
1972         return dentry;
1973 }
1974
1975 static int
1976 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1977                  scandir_t scandir, filldir_t filldir)
1978 {
1979         struct osd_ios_item *item;
1980         ENTRY;
1981
1982         OBD_ALLOC_PTR(item);
1983         if (item == NULL)
1984                 RETURN(-ENOMEM);
1985
1986         INIT_LIST_HEAD(&item->oii_list);
1987         item->oii_dentry = dget(dentry);
1988         item->oii_scandir = scandir;
1989         item->oii_filldir = filldir;
1990         list_add_tail(&item->oii_list, &dev->od_ios_list);
1991
1992         RETURN(0);
1993 }
1994
1995 /**
1996  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1997  *
1998  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1999  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
2000  * reference the inode, or fixed if it is missing or references another inode.
2001  */
2002 static int
2003 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
2004                  struct inode *inode, const struct lu_fid *fid, int flags)
2005 {
2006         struct lustre_mdt_attrs *lma    = &info->oti_ost_attrs.loa_lma;
2007         struct osd_inode_id     *id     = &info->oti_id;
2008         struct osd_inode_id     *id2    = &info->oti_id2;
2009         struct osd_scrub        *scrub  = &dev->od_scrub;
2010         struct scrub_file       *sf     = &scrub->os_file;
2011         struct lu_fid            tfid;
2012         int                      rc;
2013         ENTRY;
2014
2015         rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
2016                          &info->oti_ost_attrs);
2017         if (rc != 0 && rc != -ENODATA) {
2018                 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
2019                        "rc = %d\n", osd_name(dev), rc);
2020
2021                 RETURN(rc);
2022         }
2023
2024         osd_id_gen(id, inode->i_ino, inode->i_generation);
2025         if (rc == -ENODATA) {
2026                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
2027                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
2028                 } else {
2029                         tfid = *fid;
2030                         if (flags & OLF_IDX_IN_FID) {
2031                                 LASSERT(dev->od_index >= 0);
2032
2033                                 tfid.f_oid = dev->od_index;
2034                         }
2035                 }
2036                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
2037                 if (rc != 0) {
2038                         CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
2039                               "scrub: rc = %d\n", osd_name(dev), rc);
2040
2041                         RETURN(rc);
2042                 }
2043         } else {
2044                 if (lma->lma_compat & LMAC_NOT_IN_OI)
2045                         RETURN(0);
2046
2047                 tfid = lma->lma_self_fid;
2048         }
2049
2050         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
2051         if (rc != 0) {
2052                 if (rc != -ENOENT)
2053                         RETURN(rc);
2054
2055                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
2056                                                DTO_INDEX_INSERT, true, 0, NULL);
2057                 if (rc > 0)
2058                         rc = 0;
2059
2060                 RETURN(rc);
2061         }
2062
2063         if (osd_id_eq_strict(id, id2))
2064                 RETURN(0);
2065
2066         if (!(sf->sf_flags & SF_INCONSISTENT)) {
2067                 osd_scrub_file_reset(scrub,
2068                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2069                                      SF_INCONSISTENT);
2070                 rc = osd_scrub_file_store(scrub);
2071                 if (rc != 0)
2072                         RETURN(rc);
2073         }
2074
2075         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
2076                                        DTO_INDEX_UPDATE, true, 0, NULL);
2077         if (rc > 0)
2078                 rc = 0;
2079
2080         RETURN(rc);
2081 }
2082
2083 /**
2084  * It scans the /lost+found, and for the OST-object (with filter_fid
2085  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
2086  */
2087 #ifdef HAVE_FILLDIR_USE_CTX
2088 static int osd_ios_lf_fill(struct dir_context *buf,
2089 #else
2090 static int osd_ios_lf_fill(void *buf,
2091 #endif
2092                            const char *name, int namelen,
2093                            loff_t offset, __u64 ino, unsigned d_type)
2094 {
2095         struct osd_ios_filldir_buf *fill_buf =
2096                 (struct osd_ios_filldir_buf *)buf;
2097         struct osd_thread_info     *info     = fill_buf->oifb_info;
2098         struct osd_device          *dev      = fill_buf->oifb_dev;
2099         struct lu_fid              *fid      = &info->oti_fid;
2100         struct osd_scrub           *scrub    = &dev->od_scrub;
2101         struct dentry              *parent   = fill_buf->oifb_dentry;
2102         struct dentry              *child;
2103         struct inode               *dir      = parent->d_inode;
2104         struct inode               *inode;
2105         int                         rc;
2106         ENTRY;
2107
2108         fill_buf->oifb_items++;
2109
2110         /* skip any '.' started names */
2111         if (name[0] == '.')
2112                 RETURN(0);
2113
2114         scrub->os_lf_scanned++;
2115         child = osd_ios_lookup_one_len(name, parent, namelen);
2116         if (IS_ERR(child)) {
2117                 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
2118                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
2119                 RETURN(0);
2120         }
2121
2122         inode = child->d_inode;
2123         if (S_ISDIR(inode->i_mode)) {
2124                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2125                                       osd_ios_lf_fill);
2126                 if (rc != 0)
2127                         CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
2128                               "rc = %d\n", osd_name(dev), namelen, name, rc);
2129                 GOTO(put, rc);
2130         }
2131
2132         if (!S_ISREG(inode->i_mode))
2133                 GOTO(put, rc = 0);
2134
2135         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
2136         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
2137                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
2138                 if (rc == 0) {
2139                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
2140                                "/lost+found.\n", namelen, name, PFID(fid));
2141                         scrub->os_lf_repaired++;
2142                 } else {
2143                         CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
2144                                DFID": rc = %d\n",
2145                                osd_name(dev), namelen, name, PFID(fid), rc);
2146                 }
2147         }
2148
2149         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
2150          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
2151          *      can process them in furtuer. */
2152
2153         GOTO(put, rc);
2154
2155 put:
2156         if (rc < 0)
2157                 scrub->os_lf_failed++;
2158         dput(child);
2159         /* skip the failure to make the scanning to continue. */
2160         return 0;
2161 }
2162
2163 #ifdef HAVE_FILLDIR_USE_CTX
2164 static int osd_ios_varfid_fill(struct dir_context *buf,
2165 #else
2166 static int osd_ios_varfid_fill(void *buf,
2167 #endif
2168                                const char *name, int namelen,
2169                                loff_t offset, __u64 ino, unsigned d_type)
2170 {
2171         struct osd_ios_filldir_buf *fill_buf =
2172                 (struct osd_ios_filldir_buf *)buf;
2173         struct osd_device          *dev      = fill_buf->oifb_dev;
2174         struct dentry              *child;
2175         int                         rc;
2176         ENTRY;
2177
2178         fill_buf->oifb_items++;
2179
2180         /* skip any '.' started names */
2181         if (name[0] == '.')
2182                 RETURN(0);
2183
2184         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2185         if (IS_ERR(child))
2186                 RETURN(PTR_ERR(child));
2187
2188         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2189                               NULL, 0);
2190         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
2191                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2192                                       osd_ios_varfid_fill);
2193         dput(child);
2194
2195         RETURN(rc);
2196 }
2197
2198 #ifdef HAVE_FILLDIR_USE_CTX
2199 static int osd_ios_dl_fill(struct dir_context *buf,
2200 #else
2201 static int osd_ios_dl_fill(void *buf,
2202 #endif
2203                            const char *name, int namelen,
2204                            loff_t offset, __u64 ino, unsigned d_type)
2205 {
2206         struct osd_ios_filldir_buf *fill_buf =
2207                 (struct osd_ios_filldir_buf *)buf;
2208         struct osd_device          *dev      = fill_buf->oifb_dev;
2209         const struct osd_lf_map    *map;
2210         struct dentry              *child;
2211         int                         rc       = 0;
2212         ENTRY;
2213
2214         fill_buf->oifb_items++;
2215
2216         /* skip any '.' started names */
2217         if (name[0] == '.')
2218                 RETURN(0);
2219
2220         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
2221                 if (map->olm_namelen != namelen)
2222                         continue;
2223
2224                 if (strncmp(map->olm_name, name, namelen) == 0)
2225                         break;
2226         }
2227
2228         if (map->olm_name == NULL)
2229                 RETURN(0);
2230
2231         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2232         if (IS_ERR(child))
2233                 RETURN(PTR_ERR(child));
2234
2235         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2236                               &map->olm_fid, map->olm_flags);
2237         dput(child);
2238
2239         RETURN(rc);
2240 }
2241
2242 #ifdef HAVE_FILLDIR_USE_CTX
2243 static int osd_ios_uld_fill(struct dir_context *buf,
2244 #else
2245 static int osd_ios_uld_fill(void *buf,
2246 #endif
2247                             const char *name, int namelen,
2248                             loff_t offset, __u64 ino, unsigned d_type)
2249 {
2250         struct osd_ios_filldir_buf *fill_buf =
2251                 (struct osd_ios_filldir_buf *)buf;
2252         struct dentry              *child;
2253         struct lu_fid               tfid;
2254         int                         rc       = 0;
2255         ENTRY;
2256
2257         fill_buf->oifb_items++;
2258
2259         /* skip any non-DFID format name */
2260         if (name[0] != '[')
2261                 RETURN(0);
2262
2263         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2264         if (IS_ERR(child))
2265                 RETURN(PTR_ERR(child));
2266
2267         /* skip the start '[' */
2268         sscanf(&name[1], SFID, RFID(&tfid));
2269         if (fid_is_sane(&tfid))
2270                 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
2271                                       child->d_inode, &tfid, 0);
2272         else
2273                 rc = -EIO;
2274         dput(child);
2275
2276         RETURN(rc);
2277 }
2278
2279 #ifdef HAVE_FILLDIR_USE_CTX
2280 static int osd_ios_root_fill(struct dir_context *buf,
2281 #else
2282 static int osd_ios_root_fill(void *buf,
2283 #endif
2284                              const char *name, int namelen,
2285                              loff_t offset, __u64 ino, unsigned d_type)
2286 {
2287         struct osd_ios_filldir_buf *fill_buf =
2288                 (struct osd_ios_filldir_buf *)buf;
2289         struct osd_device          *dev      = fill_buf->oifb_dev;
2290         const struct osd_lf_map    *map;
2291         struct dentry              *child;
2292         int                         rc       = 0;
2293         ENTRY;
2294
2295         fill_buf->oifb_items++;
2296
2297         /* skip any '.' started names */
2298         if (name[0] == '.')
2299                 RETURN(0);
2300
2301         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2302                 if (map->olm_namelen != namelen)
2303                         continue;
2304
2305                 if (strncmp(map->olm_name, name, namelen) == 0)
2306                         break;
2307         }
2308
2309         if (map->olm_name == NULL)
2310                 RETURN(0);
2311
2312         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2313         if (IS_ERR(child))
2314                 RETURN(PTR_ERR(child));
2315
2316         if (!(map->olm_flags & OLF_NO_OI))
2317                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2318                                       &map->olm_fid, map->olm_flags);
2319         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2320                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2321                                       map->olm_filldir);
2322         dput(child);
2323
2324         RETURN(rc);
2325 }
2326
2327 static int
2328 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2329                      struct dentry *dentry, filldir_t filldir)
2330 {
2331         struct osd_ios_filldir_buf    buf   = {
2332 #ifdef HAVE_DIR_CONTEXT
2333                                                 .ctx.actor = filldir,
2334 #endif
2335                                                 .oifb_info = info,
2336                                                 .oifb_dev = dev,
2337                                                 .oifb_dentry = dentry };
2338         struct file                  *filp  = &info->oti_file;
2339         struct inode                 *inode = dentry->d_inode;
2340         const struct file_operations *fops  = inode->i_fop;
2341         int                           rc;
2342         ENTRY;
2343
2344         LASSERT(filldir != NULL);
2345
2346         filp->f_pos = 0;
2347         filp->f_path.dentry = dentry;
2348         filp->f_mode = FMODE_64BITHASH;
2349         filp->f_mapping = inode->i_mapping;
2350         filp->f_op = fops;
2351         filp->private_data = NULL;
2352         set_file_inode(filp, inode);
2353
2354         do {
2355                 buf.oifb_items = 0;
2356 #ifdef HAVE_DIR_CONTEXT
2357                 buf.ctx.pos = filp->f_pos;
2358                 rc = fops->iterate(filp, &buf.ctx);
2359                 filp->f_pos = buf.ctx.pos;
2360 #else
2361                 rc = fops->readdir(filp, &buf, filldir);
2362 #endif
2363         } while (rc >= 0 && buf.oifb_items > 0 &&
2364                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
2365         fops->release(inode, filp);
2366
2367         RETURN(rc);
2368 }
2369
2370 static int
2371 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2372                   struct dentry *dentry, filldir_t filldir)
2373 {
2374         struct osd_scrub  *scrub  = &dev->od_scrub;
2375         struct scrub_file *sf     = &scrub->os_file;
2376         struct dentry     *child;
2377         int                rc;
2378         ENTRY;
2379
2380         /* It is existing MDT0 device. We only allow the case of object without
2381          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2382          * can generate IGIF mode FID for the object and related OI mapping. If
2383          * it is on other MDTs, then becuase file-level backup/restore, related
2384          * OI mapping may be invalid already, we do not know which is the right
2385          * FID for the object. We only allow IGIF objects to reside on the MDT0.
2386          *
2387          * XXX: For the case of object on non-MDT0 device with neither LMA nor
2388          *      "fid" xattr, then something crashed. We cannot re-generate the
2389          *      FID directly, instead, the OI scrub will scan the OI structure
2390          *      and try to re-generate the LMA from the OI mapping. But if the
2391          *      OI mapping crashed or lost also, then we have to give up under
2392          *      double failure cases. */
2393         scrub->os_convert_igif = 1;
2394         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2395                                        strlen(dot_lustre_name));
2396         if (IS_ERR(child)) {
2397                 rc = PTR_ERR(child);
2398                 if (rc == -ENOENT) {
2399                         /* It is 1.8 MDT device. */
2400                         if (!(sf->sf_flags & SF_UPGRADE)) {
2401                                 osd_scrub_file_reset(scrub,
2402                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2403                                         SF_UPGRADE);
2404                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2405                                 rc = osd_scrub_file_store(scrub);
2406                         } else {
2407                                 rc = 0;
2408                         }
2409                 }
2410         } else {
2411                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2412                  * so the client will get IGIF for the ".lustre" object when
2413                  * the MDT restart.
2414                  *
2415                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2416                  * it does not know whether there are some old clients cached
2417                  * the ".lustre" IGIF during the upgrading. Two choices:
2418                  *
2419                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2420                  *    It will allow the old connected clients to access the
2421                  *    ".lustre" with cached IGIF. But it will cause others
2422                  *    on the MDT failed to check "fid_is_dot_lustre()".
2423                  *
2424                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2425                  *    for ".lustre" in spite of whether there are some clients
2426                  *    cached the ".lustre" IGIF or not. It enables the check
2427                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
2428                  *    that the old connected clients cannot access the ".lustre"
2429                  *    with the cached IGIF.
2430                  *
2431                  * Usually, it is rare case for the old connected clients
2432                  * to access the ".lustre" with cached IGIF. So we prefer
2433                  * to the solution 2). */
2434                 rc = osd_ios_scan_one(info, dev, child->d_inode,
2435                                       &LU_DOT_LUSTRE_FID, 0);
2436                 if (rc == 0)
2437                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2438                                               osd_ios_dl_fill);
2439                 dput(child);
2440         }
2441
2442         RETURN(rc);
2443 }
2444
2445 static int
2446 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2447                      struct dentry *dentry, filldir_t filldir)
2448 {
2449         struct osd_scrub  *scrub  = &dev->od_scrub;
2450         struct scrub_file *sf     = &scrub->os_file;
2451         struct dentry     *child;
2452         int                rc;
2453         ENTRY;
2454
2455         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2456                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2457                 rc = osd_scrub_file_store(scrub);
2458                 if (rc != 0)
2459                         RETURN(rc);
2460         }
2461
2462         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2463         if (!IS_ERR(child)) {
2464                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2465                 dput(child);
2466         } else {
2467                 rc = PTR_ERR(child);
2468         }
2469
2470         if (rc != 0 && rc != -ENOENT)
2471                 RETURN(rc);
2472
2473         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2474         if (!IS_ERR(child)) {
2475                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2476                 dput(child);
2477         } else {
2478                 rc = PTR_ERR(child);
2479         }
2480
2481         if (rc == -ENOENT)
2482                 rc = 0;
2483
2484         RETURN(rc);
2485 }
2486
2487 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2488                                 struct osd_device *dev)
2489 {
2490         struct osd_ios_item     *item    = NULL;
2491         scandir_t                scandir = osd_ios_general_scan;
2492         filldir_t                filldir = osd_ios_root_fill;
2493         struct dentry           *dentry  = osd_sb(dev)->s_root;
2494         const struct osd_lf_map *map     = osd_lf_maps;
2495         int                      rc;
2496         ENTRY;
2497
2498         /* Lookup IGIF in OI by force for initial OI scrub. */
2499         dev->od_igif_inoi = 1;
2500
2501         while (1) {
2502                 rc = scandir(info, dev, dentry, filldir);
2503                 if (item != NULL) {
2504                         dput(item->oii_dentry);
2505                         OBD_FREE_PTR(item);
2506                 }
2507
2508                 if (rc != 0)
2509                         break;
2510
2511                 if (list_empty(&dev->od_ios_list))
2512                         break;
2513
2514                 item = list_entry(dev->od_ios_list.next,
2515                                   struct osd_ios_item, oii_list);
2516                 list_del_init(&item->oii_list);
2517
2518                 LASSERT(item->oii_scandir != NULL);
2519                 scandir = item->oii_scandir;
2520                 filldir = item->oii_filldir;
2521                 dentry = item->oii_dentry;
2522         }
2523
2524         while (!list_empty(&dev->od_ios_list)) {
2525                 item = list_entry(dev->od_ios_list.next,
2526                                   struct osd_ios_item, oii_list);
2527                 list_del_init(&item->oii_list);
2528                 dput(item->oii_dentry);
2529                 OBD_FREE_PTR(item);
2530         }
2531
2532         if (rc != 0)
2533                 RETURN(rc);
2534
2535         /* There maybe the case that the object has been removed, but its OI
2536          * mapping is still in the OI file, such as the "CATALOGS" after MDT
2537          * file-level backup/restore. So here cleanup the stale OI mappings. */
2538         while (map->olm_name != NULL) {
2539                 struct dentry *child;
2540
2541                 if (fid_is_zero(&map->olm_fid)) {
2542                         map++;
2543                         continue;
2544                 }
2545
2546                 child = osd_ios_lookup_one_len(map->olm_name,
2547                                                osd_sb(dev)->s_root,
2548                                                map->olm_namelen);
2549                 if (!IS_ERR(child))
2550                         dput(child);
2551                 else if (PTR_ERR(child) == -ENOENT)
2552                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2553                                                   NULL, DTO_INDEX_DELETE,
2554                                                   true, 0, NULL);
2555                 map++;
2556         }
2557
2558         RETURN(0);
2559 }
2560
2561 char *osd_lf_fid2name(const struct lu_fid *fid)
2562 {
2563         const struct osd_lf_map *map = osd_lf_maps;
2564
2565         while (map->olm_name != NULL) {
2566                 if (!lu_fid_eq(fid, &map->olm_fid)) {
2567                         map++;
2568                         continue;
2569                 }
2570
2571                 if (map->olm_flags & OLF_SHOW_NAME)
2572                         return map->olm_name;
2573                 else
2574                         return "";
2575         }
2576
2577         return NULL;
2578 }
2579
2580 /* OI scrub start/stop */
2581
2582 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2583 {
2584         struct osd_scrub     *scrub  = &dev->od_scrub;
2585         struct ptlrpc_thread *thread = &scrub->os_thread;
2586         struct l_wait_info    lwi    = { 0 };
2587         struct task_struct   *task;
2588         int                   rc;
2589         ENTRY;
2590
2591         if (dev->od_dt_dev.dd_rdonly)
2592                 RETURN(-EROFS);
2593
2594         /* os_lock: sync status between stop and scrub thread */
2595         spin_lock(&scrub->os_lock);
2596
2597 again:
2598         if (thread_is_running(thread)) {
2599                 spin_unlock(&scrub->os_lock);
2600                 if (!(scrub->os_file.sf_flags & SF_AUTO ||
2601                       scrub->os_partial_scan) ||
2602                      (flags & SS_AUTO_PARTIAL))
2603                         RETURN(-EALREADY);
2604
2605                 osd_scrub_join(dev, flags, false);
2606                 spin_lock(&scrub->os_lock);
2607                 if (!thread_is_running(thread))
2608                         goto again;
2609
2610                 spin_unlock(&scrub->os_lock);
2611                 RETURN(0);
2612         }
2613
2614         if (unlikely(thread_is_stopping(thread))) {
2615                 spin_unlock(&scrub->os_lock);
2616                 l_wait_event(thread->t_ctl_waitq,
2617                              thread_is_stopped(thread),
2618                              &lwi);
2619                 spin_lock(&scrub->os_lock);
2620                 goto again;
2621         }
2622         spin_unlock(&scrub->os_lock);
2623
2624         if (scrub->os_file.sf_status == SS_COMPLETED) {
2625                 if (!(flags & SS_SET_FAILOUT))
2626                         flags |= SS_CLEAR_FAILOUT;
2627
2628                 if (!(flags & SS_SET_DRYRUN))
2629                         flags |= SS_CLEAR_DRYRUN;
2630
2631                 flags |= SS_RESET;
2632         }
2633
2634         scrub->os_start_flags = flags;
2635         thread_set_flags(thread, 0);
2636         task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2637         if (IS_ERR(task)) {
2638                 rc = PTR_ERR(task);
2639                 CERROR("%s: cannot start iteration thread: rc = %d\n",
2640                        osd_scrub2name(scrub), rc);
2641                 RETURN(rc);
2642         }
2643
2644         l_wait_event(thread->t_ctl_waitq,
2645                      thread_is_running(thread) || thread_is_stopped(thread),
2646                      &lwi);
2647
2648         RETURN(0);
2649 }
2650
2651 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2652 {
2653         int rc;
2654         ENTRY;
2655
2656         /* od_otable_mutex: prevent curcurrent start/stop */
2657         mutex_lock(&dev->od_otable_mutex);
2658         rc = do_osd_scrub_start(dev, flags);
2659         mutex_unlock(&dev->od_otable_mutex);
2660
2661         RETURN(rc == -EALREADY ? 0 : rc);
2662 }
2663
2664 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2665 {
2666         struct ptlrpc_thread *thread = &scrub->os_thread;
2667         struct l_wait_info    lwi    = { 0 };
2668
2669         /* os_lock: sync status between stop and scrub thread */
2670         spin_lock(&scrub->os_lock);
2671         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2672                 thread_set_flags(thread, SVC_STOPPING);
2673                 spin_unlock(&scrub->os_lock);
2674                 wake_up_all(&thread->t_ctl_waitq);
2675                 l_wait_event(thread->t_ctl_waitq,
2676                              thread_is_stopped(thread),
2677                              &lwi);
2678                 /* Do not skip the last lock/unlock, which can guarantee that
2679                  * the caller cannot return until the OI scrub thread exit. */
2680                 spin_lock(&scrub->os_lock);
2681         }
2682         spin_unlock(&scrub->os_lock);
2683 }
2684
2685 static void osd_scrub_stop(struct osd_device *dev)
2686 {
2687         /* od_otable_mutex: prevent curcurrent start/stop */
2688         mutex_lock(&dev->od_otable_mutex);
2689         dev->od_scrub.os_paused = 1;
2690         do_osd_scrub_stop(&dev->od_scrub);
2691         mutex_unlock(&dev->od_otable_mutex);
2692 }
2693
2694 /* OI scrub setup/cleanup */
2695
2696 static const char osd_scrub_name[] = "OI_scrub";
2697
2698 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2699 {
2700         struct osd_thread_info     *info   = osd_oti_get(env);
2701         struct osd_scrub           *scrub  = &dev->od_scrub;
2702         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
2703         struct scrub_file          *sf     = &scrub->os_file;
2704         struct super_block         *sb     = osd_sb(dev);
2705         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
2706         struct lvfs_run_ctxt        saved;
2707         struct file                *filp;
2708         struct inode               *inode;
2709         struct lu_fid              *fid    = &info->oti_fid;
2710         bool                        dirty  = false;
2711         bool                        restored = false;
2712         int                         rc     = 0;
2713         ENTRY;
2714
2715         memset(scrub, 0, sizeof(*scrub));
2716         OBD_SET_CTXT_MAGIC(ctxt);
2717         ctxt->pwdmnt = dev->od_mnt;
2718         ctxt->pwd = dev->od_mnt->mnt_root;
2719         ctxt->fs = get_ds();
2720
2721         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2722         init_rwsem(&scrub->os_rwsem);
2723         spin_lock_init(&scrub->os_lock);
2724         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2725
2726         push_ctxt(&saved, ctxt);
2727         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2728         if (IS_ERR(filp)) {
2729                 pop_ctxt(&saved, ctxt);
2730                 RETURN(PTR_ERR(filp));
2731         }
2732
2733         inode = file_inode(filp);
2734         /* 'What the @fid is' is not imporatant, because the object
2735          * has no OI mapping, and only is visible inside the OSD.*/
2736         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2737         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2738         if (rc != 0) {
2739                 filp_close(filp, NULL);
2740                 pop_ctxt(&saved, ctxt);
2741                 RETURN(rc);
2742         }
2743
2744         scrub->os_inode = igrab(inode);
2745         filp_close(filp, NULL);
2746         pop_ctxt(&saved, ctxt);
2747
2748         rc = osd_scrub_file_load(scrub);
2749         if (rc == -ENOENT) {
2750                 osd_scrub_file_init(scrub, es->s_uuid);
2751                 /* If the "/O" dir does not exist when mount (indicated by
2752                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2753                  * then it is quite probably that the device is a new one,
2754                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2755                  *
2756                  * For the rare case that "/O" and "OI_scrub" both lost on
2757                  * an old device, it can be found and cleared later.
2758                  *
2759                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2760                  * need to check "filter_fid_old" and to convert it to
2761                  * "filter_fid" for each object, and all the IGIF should
2762                  * have their FID mapping in OI files already. */
2763                 if (dev->od_maybe_new)
2764                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2765                 dirty = true;
2766         } else if (rc != 0) {
2767                 GOTO(cleanup_inode, rc);
2768         } else {
2769                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2770                         struct obd_uuid *old_uuid;
2771                         struct obd_uuid *new_uuid;
2772
2773                         OBD_ALLOC_PTR(old_uuid);
2774                         OBD_ALLOC_PTR(new_uuid);
2775                         if (old_uuid == NULL || new_uuid == NULL) {
2776                                 CERROR("%s: UUID has been changed, but"
2777                                        "failed to allocate RAM for report\n",
2778                                        osd_dev2name(dev));
2779                         } else {
2780                                 class_uuid_unparse(sf->sf_uuid, old_uuid);
2781                                 class_uuid_unparse(es->s_uuid, new_uuid);
2782                                 CERROR("%s: UUID has been changed from "
2783                                        "%s to %s\n", osd_dev2name(dev),
2784                                        old_uuid->uuid, new_uuid->uuid);
2785                         }
2786                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2787                         dirty = true;
2788                         restored = true;
2789                         if (old_uuid != NULL)
2790                                 OBD_FREE_PTR(old_uuid);
2791                         if (new_uuid != NULL)
2792                                 OBD_FREE_PTR(new_uuid);
2793                 } else if (sf->sf_status == SS_SCANNING) {
2794                         sf->sf_status = SS_CRASHED;
2795                         dirty = true;
2796                 }
2797         }
2798
2799         if (sf->sf_pos_last_checkpoint != 0)
2800                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2801         else
2802                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2803
2804         if (dirty) {
2805                 rc = osd_scrub_file_store(scrub);
2806                 if (rc != 0)
2807                         GOTO(cleanup_inode, rc);
2808         }
2809
2810         /* Initialize OI files. */
2811         rc = osd_oi_init(info, dev, restored);
2812         if (rc < 0)
2813                 GOTO(cleanup_inode, rc);
2814
2815         rc = osd_initial_OI_scrub(info, dev);
2816         if (rc != 0)
2817                 GOTO(cleanup_oi, rc);
2818
2819         if (sf->sf_flags & SF_UPGRADE ||
2820             !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2821               sf->sf_success_count > 0)) {
2822                 dev->od_igif_inoi = 0;
2823                 dev->od_check_ff = dev->od_is_ost;
2824         } else {
2825                 dev->od_igif_inoi = 1;
2826                 dev->od_check_ff = 0;
2827         }
2828
2829         if (sf->sf_flags & SF_INCONSISTENT)
2830                 /* The 'od_igif_inoi' will be set under the
2831                  * following cases:
2832                  * 1) new created system, or
2833                  * 2) restored from file-level backup, or
2834                  * 3) the upgrading completed.
2835                  *
2836                  * The 'od_igif_inoi' may be cleared by OI scrub
2837                  * later if found that the system is upgrading. */
2838                 dev->od_igif_inoi = 1;
2839
2840         if (!dev->od_dt_dev.dd_rdonly && !dev->od_noscrub &&
2841             ((sf->sf_status == SS_PAUSED) ||
2842              (sf->sf_status == SS_CRASHED &&
2843               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2844                               SF_UPGRADE | SF_AUTO)) ||
2845              (sf->sf_status == SS_INIT &&
2846               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2847                               SF_UPGRADE))))
2848                 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2849
2850         if (rc != 0)
2851                 GOTO(cleanup_oi, rc);
2852
2853         /* it is possible that dcache entries may keep objects after they are
2854          * deleted by OSD. While it looks safe this can cause object data to
2855          * stay until umount causing failures in tests calculating free space,
2856          * e.g. replay-ost-single. Since those dcache entries are not used
2857          * anymore let's just free them after use here */
2858         shrink_dcache_sb(sb);
2859
2860         RETURN(0);
2861 cleanup_oi:
2862         osd_oi_fini(info, dev);
2863 cleanup_inode:
2864         iput(scrub->os_inode);
2865         scrub->os_inode = NULL;
2866
2867         return rc;
2868 }
2869
2870 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2871 {
2872         struct osd_scrub *scrub = &dev->od_scrub;
2873
2874         LASSERT(dev->od_otable_it == NULL);
2875
2876         if (scrub->os_inode != NULL) {
2877                 osd_scrub_stop(dev);
2878                 iput(scrub->os_inode);
2879                 scrub->os_inode = NULL;
2880         }
2881         if (dev->od_oi_table != NULL)
2882                 osd_oi_fini(osd_oti_get(env), dev);
2883 }
2884
2885 /* object table based iteration APIs */
2886
2887 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2888                                        struct dt_object *dt, __u32 attr)
2889 {
2890         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2891         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2892         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2893         struct osd_scrub       *scrub = &dev->od_scrub;
2894         struct osd_otable_it   *it;
2895         __u32                   start = 0;
2896         int                     rc;
2897         ENTRY;
2898
2899         /* od_otable_mutex: prevent curcurrent init/fini */
2900         mutex_lock(&dev->od_otable_mutex);
2901         if (dev->od_otable_it != NULL)
2902                 GOTO(out, it = ERR_PTR(-EALREADY));
2903
2904         OBD_ALLOC_PTR(it);
2905         if (it == NULL)
2906                 GOTO(out, it = ERR_PTR(-ENOMEM));
2907
2908         dev->od_otable_it = it;
2909         it->ooi_dev = dev;
2910         it->ooi_cache.ooc_consumer_idx = -1;
2911         if (flags & DOIF_OUTUSED)
2912                 it->ooi_used_outside = 1;
2913
2914         if (flags & DOIF_RESET)
2915                 start |= SS_RESET;
2916
2917         if (valid & DOIV_ERROR_HANDLE) {
2918                 if (flags & DOIF_FAILOUT)
2919                         start |= SS_SET_FAILOUT;
2920                 else
2921                         start |= SS_CLEAR_FAILOUT;
2922         }
2923
2924         if (valid & DOIV_DRYRUN) {
2925                 if (flags & DOIF_DRYRUN)
2926                         start |= SS_SET_DRYRUN;
2927                 else
2928                         start |= SS_CLEAR_DRYRUN;
2929         }
2930
2931         rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2932         if (rc < 0 && rc != -EALREADY) {
2933                 dev->od_otable_it = NULL;
2934                 OBD_FREE_PTR(it);
2935                 GOTO(out, it = ERR_PTR(rc));
2936         }
2937
2938         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2939
2940         GOTO(out, it);
2941
2942 out:
2943         mutex_unlock(&dev->od_otable_mutex);
2944         return (struct dt_it *)it;
2945 }
2946
2947 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2948 {
2949         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2950         struct osd_device    *dev = it->ooi_dev;
2951
2952         /* od_otable_mutex: prevent curcurrent init/fini */
2953         mutex_lock(&dev->od_otable_mutex);
2954         do_osd_scrub_stop(&dev->od_scrub);
2955         LASSERT(dev->od_otable_it == it);
2956
2957         dev->od_otable_it = NULL;
2958         mutex_unlock(&dev->od_otable_mutex);
2959         OBD_FREE_PTR(it);
2960 }
2961
2962 static int osd_otable_it_get(const struct lu_env *env,
2963                              struct dt_it *di, const struct dt_key *key)
2964 {
2965         return 0;
2966 }
2967
2968 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2969 {
2970 }
2971
2972 static inline int
2973 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2974 {
2975         spin_lock(&scrub->os_lock);
2976         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2977             scrub->os_waiting ||
2978             !thread_is_running(&scrub->os_thread))
2979                 it->ooi_waiting = 0;
2980         else
2981                 it->ooi_waiting = 1;
2982         spin_unlock(&scrub->os_lock);
2983
2984         return !it->ooi_waiting;
2985 }
2986
2987 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2988 {
2989         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2990         struct osd_device       *dev    = it->ooi_dev;
2991         struct osd_scrub        *scrub  = &dev->od_scrub;
2992         struct osd_otable_cache *ooc    = &it->ooi_cache;
2993         struct ptlrpc_thread    *thread = &scrub->os_thread;
2994         struct l_wait_info       lwi    = { 0 };
2995         int                      rc;
2996         ENTRY;
2997
2998         LASSERT(it->ooi_user_ready);
2999
3000 again:
3001         if (!thread_is_running(thread) && !it->ooi_used_outside)
3002                 RETURN(1);
3003
3004         if (ooc->ooc_cached_items > 0) {
3005                 ooc->ooc_cached_items--;
3006                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
3007                                         ~OSD_OTABLE_IT_CACHE_MASK;
3008                 RETURN(0);
3009         }
3010
3011         if (it->ooi_all_cached) {
3012                 l_wait_event(thread->t_ctl_waitq,
3013                              !thread_is_running(thread),
3014                              &lwi);
3015                 RETURN(1);
3016         }
3017
3018         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
3019                 spin_lock(&scrub->os_lock);
3020                 scrub->os_waiting = 0;
3021                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
3022                 spin_unlock(&scrub->os_lock);
3023         }
3024
3025         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
3026                 l_wait_event(thread->t_ctl_waitq,
3027                              osd_otable_it_wakeup(scrub, it),
3028                              &lwi);
3029
3030         if (!thread_is_running(thread) && !it->ooi_used_outside)
3031                 RETURN(1);
3032
3033         rc = osd_otable_it_preload(env, it);
3034         if (rc >= 0)
3035                 goto again;
3036
3037         RETURN(rc);
3038 }
3039
3040 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
3041                                         const struct dt_it *di)
3042 {
3043         return NULL;
3044 }
3045
3046 static int osd_otable_it_key_size(const struct lu_env *env,
3047                                   const struct dt_it *di)
3048 {
3049         return sizeof(__u64);
3050 }
3051
3052 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
3053                              struct dt_rec *rec, __u32 attr)
3054 {
3055         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
3056         struct osd_otable_cache *ooc = &it->ooi_cache;
3057
3058         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
3059
3060         /* Filter out Invald FID already. */
3061         LASSERTF(fid_is_sane((struct lu_fid *)rec),
3062                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
3063                  PFID((struct lu_fid *)rec),
3064                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
3065
3066         return 0;
3067 }
3068
3069 static __u64 osd_otable_it_store(const struct lu_env *env,
3070                                  const struct dt_it *di)
3071 {
3072         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
3073         struct osd_otable_cache *ooc = &it->ooi_cache;
3074         __u64                    hash;
3075
3076         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
3077                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
3078         else
3079                 hash = ooc->ooc_pos_preload;
3080         return hash;
3081 }
3082
3083 /**
3084  * Set the OSD layer iteration start position as the specified hash.
3085  */
3086 static int osd_otable_it_load(const struct lu_env *env,
3087                               const struct dt_it *di, __u64 hash)
3088 {
3089         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
3090         struct osd_device       *dev   = it->ooi_dev;
3091         struct osd_otable_cache *ooc   = &it->ooi_cache;
3092         struct osd_scrub        *scrub = &dev->od_scrub;
3093         struct osd_iit_param    *param = &it->ooi_iit_param;
3094         int                      rc;
3095         ENTRY;
3096
3097         /* Forbid to set iteration position after iteration started. */
3098         if (it->ooi_user_ready)
3099                 RETURN(-EPERM);
3100
3101         LASSERT(!scrub->os_partial_scan);
3102
3103         if (hash > OSD_OTABLE_MAX_HASH)
3104                 hash = OSD_OTABLE_MAX_HASH;
3105
3106         /* The hash is the last checkpoint position,
3107          * we will start from the next one. */
3108         ooc->ooc_pos_preload = hash + 1;
3109         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
3110                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
3111
3112         it->ooi_user_ready = 1;
3113         if (!scrub->os_full_speed)
3114                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
3115
3116         memset(param, 0, sizeof(*param));
3117         param->sb = osd_sb(dev);
3118         param->start = ooc->ooc_pos_preload;
3119         param->bg = (ooc->ooc_pos_preload - 1) /
3120                     LDISKFS_INODES_PER_GROUP(param->sb);
3121         param->offset = (ooc->ooc_pos_preload - 1) %
3122                         LDISKFS_INODES_PER_GROUP(param->sb);
3123         param->gbase = 1 + param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
3124
3125         /* Unplug OSD layer iteration by the first next() call. */
3126         rc = osd_otable_it_next(env, (struct dt_it *)it);
3127
3128         RETURN(rc);
3129 }
3130
3131 static int osd_otable_it_key_rec(const struct lu_env *env,
3132                                  const struct dt_it *di, void *key_rec)
3133 {
3134         return 0;
3135 }
3136
3137 const struct dt_index_operations osd_otable_ops = {
3138         .dio_it = {
3139                 .init     = osd_otable_it_init,
3140                 .fini     = osd_otable_it_fini,
3141                 .get      = osd_otable_it_get,
3142                 .put      = osd_otable_it_put,
3143                 .next     = osd_otable_it_next,
3144                 .key      = osd_otable_it_key,
3145                 .key_size = osd_otable_it_key_size,
3146                 .rec      = osd_otable_it_rec,
3147                 .store    = osd_otable_it_store,
3148                 .load     = osd_otable_it_load,
3149                 .key_rec  = osd_otable_it_key_rec,
3150         }
3151 };
3152
3153 /* high priority inconsistent items list APIs */
3154
3155 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL  60
3156
3157 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
3158                    int insert)
3159 {
3160         struct osd_inconsistent_item *oii;
3161         struct osd_scrub             *scrub  = &dev->od_scrub;
3162         struct ptlrpc_thread         *thread = &scrub->os_thread;
3163         int                           wakeup = 0;
3164         ENTRY;
3165
3166         OBD_ALLOC_PTR(oii);
3167         if (unlikely(oii == NULL))
3168                 RETURN(-ENOMEM);
3169
3170         INIT_LIST_HEAD(&oii->oii_list);
3171         oii->oii_cache = *oic;
3172         oii->oii_insert = insert;
3173
3174         if (scrub->os_partial_scan) {
3175                 __u64 now = cfs_time_current_sec();
3176
3177                 /* If there haven't been errors in a long time,
3178                  * decay old count until either the errors are
3179                  * gone or we reach the current interval. */
3180                 while (unlikely(scrub->os_bad_oimap_count > 0 &&
3181                                 scrub->os_bad_oimap_time +
3182                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
3183                         scrub->os_bad_oimap_count >>= 1;
3184                         scrub->os_bad_oimap_time +=
3185                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
3186                 }
3187
3188                 scrub->os_bad_oimap_time = now;
3189                 if (++scrub->os_bad_oimap_count >
3190                     dev->od_full_scrub_threshold_rate)
3191                         scrub->os_full_scrub = 1;
3192         }
3193
3194         spin_lock(&scrub->os_lock);
3195         if (unlikely(!thread_is_running(thread))) {
3196                 spin_unlock(&scrub->os_lock);
3197                 OBD_FREE_PTR(oii);
3198                 RETURN(-EAGAIN);
3199         }
3200
3201         if (list_empty(&scrub->os_inconsistent_items))
3202                 wakeup = 1;
3203         list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
3204         spin_unlock(&scrub->os_lock);
3205
3206         if (wakeup != 0)
3207                 wake_up_all(&thread->t_ctl_waitq);
3208
3209         RETURN(0);
3210 }
3211
3212 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
3213                    struct osd_inode_id *id)
3214 {
3215         struct osd_scrub             *scrub = &dev->od_scrub;
3216         struct osd_inconsistent_item *oii;
3217         ENTRY;
3218
3219         spin_lock(&scrub->os_lock);
3220         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
3221                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
3222                         *id = oii->oii_cache.oic_lid;
3223                         spin_unlock(&scrub->os_lock);
3224                         RETURN(0);
3225                 }
3226         }
3227         spin_unlock(&scrub->os_lock);
3228
3229         RETURN(-ENOENT);
3230 }
3231
3232 /* OI scrub dump */
3233
3234 static const char *scrub_status_names[] = {
3235         "init",
3236         "scanning",
3237         "completed",
3238         "failed",
3239         "stopped",
3240         "paused",
3241         "crashed",
3242         NULL
3243 };
3244
3245 static const char *scrub_flags_names[] = {
3246         "recreated",
3247         "inconsistent",
3248         "auto",
3249         "upgrade",
3250         NULL
3251 };
3252
3253 static const char *scrub_param_names[] = {
3254         "failout",
3255         "dryrun",
3256         NULL
3257 };
3258
3259 static void scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
3260                             const char *prefix)
3261 {
3262         int flag;
3263         int i;
3264
3265         seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
3266
3267         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
3268                 if (flag & bits) {
3269                         bits &= ~flag;
3270                         seq_printf(m, "%s%c", names[i],
3271                                    bits != 0 ? ',' : '\n');
3272                 }
3273         }
3274 }
3275
3276 static void scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
3277 {
3278         if (time != 0)
3279                 seq_printf(m, "%s: %llu seconds\n", prefix,
3280                            cfs_time_current_sec() - time);
3281         else
3282                 seq_printf(m, "%s: N/A\n", prefix);
3283 }
3284
3285 static void scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
3286 {
3287         if (pos != 0)
3288                 seq_printf(m, "%s: %llu\n", prefix, pos);
3289         else
3290                 seq_printf(m, "%s: N/A\n", prefix);
3291 }
3292
3293 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
3294 {
3295         struct osd_scrub  *scrub   = &dev->od_scrub;
3296         struct scrub_file *sf      = &scrub->os_file;
3297         __u64              checked;
3298         __u64              speed;
3299
3300         down_read(&scrub->os_rwsem);
3301         seq_printf(m, "name: OI_scrub\n"
3302                    "magic: 0x%x\n"
3303                    "oi_files: %d\n"
3304                    "status: %s\n",
3305                    sf->sf_magic, (int)sf->sf_oi_count,
3306                    scrub_status_names[sf->sf_status]);
3307
3308         scrub_bits_dump(m, sf->sf_flags, scrub_flags_names, "flags");
3309
3310         scrub_bits_dump(m, sf->sf_param, scrub_param_names, "param");
3311
3312         scrub_time_dump(m, sf->sf_time_last_complete,
3313                         "time_since_last_completed");
3314
3315         scrub_time_dump(m, sf->sf_time_latest_start,
3316                         "time_since_latest_start");
3317
3318         scrub_time_dump(m, sf->sf_time_last_checkpoint,
3319                         "time_since_last_checkpoint");
3320
3321         scrub_pos_dump(m, sf->sf_pos_latest_start,
3322                         "latest_start_position");
3323
3324         scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
3325                         "last_checkpoint_position");
3326
3327         scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
3328                         "first_failure_position");
3329
3330         checked = sf->sf_items_checked + scrub->os_new_checked;
3331         seq_printf(m, "checked: %llu\n"
3332                    "%s: %llu\n"
3333                    "failed: %llu\n"
3334                    "prior_%s: %llu\n"
3335                    "noscrub: %llu\n"
3336                    "igif: %llu\n"
3337                    "success_count: %u\n",
3338                    checked,
3339                    sf->sf_param & SP_DRYRUN ? "inconsistent" : "updated",
3340                    sf->sf_items_updated, sf->sf_items_failed,
3341                    sf->sf_param & SP_DRYRUN ? "inconsistent" : "updated",
3342                    sf->sf_items_updated_prior, sf->sf_items_noscrub,
3343                    sf->sf_items_igif, sf->sf_success_count);
3344
3345         speed = checked;
3346         if (thread_is_running(&scrub->os_thread)) {
3347                 cfs_duration_t duration = cfs_time_current() -
3348                                           scrub->os_time_last_checkpoint;
3349                 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
3350                                                      MSEC_PER_SEC);
3351                 __u32 rtime = sf->sf_run_time +
3352                               cfs_duration_sec(duration + HALF_SEC);
3353
3354                 if (duration != 0)
3355                         do_div(new_checked, duration);
3356                 if (rtime != 0)
3357                         do_div(speed, rtime);
3358                 seq_printf(m, "run_time: %u seconds\n"
3359                            "average_speed: %llu objects/sec\n"
3360                            "real-time_speed: %llu objects/sec\n"
3361                            "current_position: %u\n"
3362                            "lf_scanned: %llu\n"
3363                            "lf_%s: %llu\n"
3364                            "lf_failed: %llu\n",
3365                            rtime, speed, new_checked, scrub->os_pos_current,
3366                            scrub->os_lf_scanned,
3367                            sf->sf_param & SP_DRYRUN ?
3368                                 "inconsistent" : "repaired",
3369                            scrub->os_lf_repaired,
3370                            scrub->os_lf_failed);
3371                 seq_printf(m, "inodes_per_group: %lu\n"
3372                            "current_iit_group: %u\n"
3373                            "current_iit_base: %u\n"
3374                            "current_iit_offset: %u\n"
3375                            "scrub_in_prior: %s\n"
3376                            "scrub_full_speed: %s\n"
3377                            "partial_scan: %s\n",
3378                            LDISKFS_INODES_PER_GROUP(osd_sb(dev)),
3379                            scrub->os_iit_param.bg,
3380                            scrub->os_iit_param.gbase,
3381                            scrub->os_iit_param.offset,
3382                            scrub->os_in_prior ? "yes" : "no",
3383                            scrub->os_full_speed ? "yes" : "no",
3384                            scrub->os_partial_scan ? "yes" : "no");
3385         } else {
3386                 if (sf->sf_run_time != 0)
3387                         do_div(speed, sf->sf_run_time);
3388                 seq_printf(m, "run_time: %u seconds\n"
3389                            "average_speed: %llu objects/sec\n"
3390                            "real-time_speed: N/A\n"
3391                            "current_position: N/A\n"
3392                            "lf_scanned: %llu\n"
3393                            "lf_%s: %llu\n"
3394                            "lf_failed: %llu\n",
3395                            sf->sf_run_time, speed, scrub->os_lf_scanned,
3396                            sf->sf_param & SP_DRYRUN ?
3397                                 "inconsistent" : "repaired",
3398                            scrub->os_lf_repaired, scrub->os_lf_failed);
3399         }
3400
3401         up_read(&scrub->os_rwsem);
3402         return 0;
3403 }