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_18_23 *ff = &info->oti_ff_old;
 375         struct dentry *dentry = &info->oti_obj_dentry;
 376         struct lu_fid *tfid = &info->oti_fid;
 377         bool fid_18_23 = false;
 378         handle_t *jh;
 379         int size = 0;
 380         int rc;
 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 = osd_removexattr(dentry, inode, XATTR_NAME_FID);
 420                 if (rc)
 421                         GOTO(stop, rc);
 422
 423                 fid_18_23 = true;
 424         } else if (rc != -ENODATA && rc < (int)sizeof(struct filter_fid_24_29)) {
 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 (fid_18_23) {
 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 2.3 or older 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 or newer 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          /* Skip project quota inode since it is greater than s_first_ino. */
 970 #ifdef HAVE_PROJECT_QUOTA
 971         if (ldiskfs_has_feature_project(sb) &&
 972             pos == le32_to_cpu(LDISKFS_SB(sb)->s_es->s_prj_quota_inum))
 973                 RETURN(SCRUB_NEXT_CONTINUE);
 974 #endif
 975
 976         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 977         inode = osd_iget(info, dev, lid);
 978         if (IS_ERR(inode)) {
 979                 rc = PTR_ERR(inode);
 980                 /* The inode may be removed after bitmap searching, or the
 981                  * file is new created without inode initialized yet. */
 982                 if (rc == -ENOENT || rc == -ESTALE)
 983                         RETURN(SCRUB_NEXT_CONTINUE);
 984
 985                 CDEBUG(D_LFSCK, "%s: fail to read inode, ino# = %u: "
 986                        "rc = %d\n", osd_dev2name(dev), pos, rc);
 987                 RETURN(rc);
 988         }
 989
 990         /* It is an EA inode, no OI mapping for it, skip it. */
 991         if (osd_is_ea_inode(inode))
 992                 GOTO(put, rc = SCRUB_NEXT_CONTINUE);
 993
 994         if (scrub &&
 995             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 996                 /* Only skip it for the first OI scrub accessing. */
 997                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 998                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
 999         }
1000
1001         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
1002
1003         GOTO(put, rc);
1004
1005 put:
1006         iput(inode);
1007         return rc;
1008 }
1009
1010 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
1011                           struct osd_iit_param *param,
1012                           struct osd_idmap_cache **oic, const bool noslot)
1013 {
1014         struct osd_scrub     *scrub  = &dev->od_scrub;
1015         struct ptlrpc_thread *thread = &scrub->os_thread;
1016         struct lu_fid        *fid;
1017         struct osd_inode_id  *lid;
1018         int                   rc;
1019
1020         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
1021                 struct l_wait_info lwi;
1022
1023                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
1024                 if (likely(lwi.lwi_timeout > 0))
1025                         l_wait_event(thread->t_ctl_waitq,
1026                                 !list_empty(&scrub->os_inconsistent_items) ||
1027                                 !thread_is_running(thread),
1028                                 &lwi);
1029         }
1030
1031         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
1032                 spin_lock(&scrub->os_lock);
1033                 thread_set_flags(thread, SVC_STOPPING);
1034                 spin_unlock(&scrub->os_lock);
1035                 return SCRUB_NEXT_CRASH;
1036         }
1037
1038         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
1039                 return SCRUB_NEXT_FATAL;
1040
1041         if (unlikely(!thread_is_running(thread)))
1042                 return SCRUB_NEXT_EXIT;
1043
1044         if (!list_empty(&scrub->os_inconsistent_items)) {
1045                 spin_lock(&scrub->os_lock);
1046                 if (likely(!list_empty(&scrub->os_inconsistent_items))) {
1047                         struct osd_inconsistent_item *oii;
1048
1049                         oii = list_entry(scrub->os_inconsistent_items.next,
1050                                 struct osd_inconsistent_item, oii_list);
1051                         spin_unlock(&scrub->os_lock);
1052
1053                         *oic = &oii->oii_cache;
1054                         scrub->os_in_prior = 1;
1055
1056                         return 0;
1057                 }
1058                 spin_unlock(&scrub->os_lock);
1059         }
1060
1061         if (noslot)
1062                 return SCRUB_NEXT_WAIT;
1063
1064         rc = osd_iit_next(param, &scrub->os_pos_current);
1065         if (rc != 0)
1066                 return rc;
1067
1068         *oic = &scrub->os_oic;
1069         fid = &(*oic)->oic_fid;
1070         lid = &(*oic)->oic_lid;
1071         rc = osd_iit_iget(info, dev, fid, lid,
1072                           scrub->os_pos_current, param->sb, true);
1073         return rc;
1074 }
1075
1076 static int osd_preload_next(struct osd_thread_info *info,
1077                             struct osd_device *dev, struct osd_iit_param *param,
1078                             struct osd_idmap_cache **oic, const bool noslot)
1079 {
1080         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1081         struct osd_scrub *scrub = &dev->od_scrub;
1082         struct ptlrpc_thread *thread = &scrub->os_thread;
1083         int rc;
1084
1085         if (thread_is_running(thread) &&
1086             ooc->ooc_pos_preload >= scrub->os_pos_current)
1087                 return SCRUB_NEXT_EXIT;
1088
1089         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1090         if (rc)
1091                 return rc;
1092
1093         rc = osd_iit_iget(info, dev,
1094                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1095                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1096                           ooc->ooc_pos_preload, param->sb, false);
1097         return rc;
1098 }
1099
1100 static inline int
1101 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1102 {
1103         spin_lock(&scrub->os_lock);
1104         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1105             !list_empty(&scrub->os_inconsistent_items) ||
1106             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1107                 scrub->os_waiting = 0;
1108         else
1109                 scrub->os_waiting = 1;
1110         spin_unlock(&scrub->os_lock);
1111
1112         return !scrub->os_waiting;
1113 }
1114
1115 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1116                           struct osd_iit_param *param,
1117                           struct osd_idmap_cache *oic, bool *noslot, int rc)
1118 {
1119         struct l_wait_info       lwi    = { 0 };
1120         struct osd_scrub        *scrub  = &dev->od_scrub;
1121         struct scrub_file       *sf     = &scrub->os_file;
1122         struct ptlrpc_thread    *thread = &scrub->os_thread;
1123         struct osd_otable_it    *it     = dev->od_otable_it;
1124         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
1125
1126         switch (rc) {
1127         case SCRUB_NEXT_NOSCRUB:
1128                 down_write(&scrub->os_rwsem);
1129                 scrub->os_new_checked++;
1130                 sf->sf_items_noscrub++;
1131                 up_write(&scrub->os_rwsem);
1132         case SCRUB_NEXT_CONTINUE:
1133         case SCRUB_NEXT_WAIT:
1134                 goto wait;
1135         }
1136
1137         rc = osd_scrub_check_update(info, dev, oic, rc);
1138         if (rc != 0) {
1139                 scrub->os_in_prior = 0;
1140                 return rc;
1141         }
1142
1143         rc = osd_scrub_checkpoint(scrub);
1144         if (rc != 0) {
1145                 CDEBUG(D_LFSCK, "%s: fail to checkpoint, pos = %u: "
1146                        "rc = %d\n", osd_scrub2name(scrub),
1147                        scrub->os_pos_current, rc);
1148                 /* Continue, as long as the scrub itself can go ahead. */
1149         }
1150
1151         if (scrub->os_in_prior) {
1152                 scrub->os_in_prior = 0;
1153                 return 0;
1154         }
1155
1156 wait:
1157         if (it != NULL && it->ooi_waiting && ooc != NULL &&
1158             ooc->ooc_pos_preload < scrub->os_pos_current) {
1159                 spin_lock(&scrub->os_lock);
1160                 it->ooi_waiting = 0;
1161                 wake_up_all(&thread->t_ctl_waitq);
1162                 spin_unlock(&scrub->os_lock);
1163         }
1164
1165         if (rc == SCRUB_NEXT_CONTINUE)
1166                 return 0;
1167
1168         if (scrub->os_full_speed || !ooc || osd_scrub_has_window(scrub, ooc)) {
1169                 *noslot = false;
1170                 return 0;
1171         }
1172
1173         if (it != NULL)
1174                 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1175                              &lwi);
1176
1177         if (!ooc || osd_scrub_has_window(scrub, ooc))
1178                 *noslot = false;
1179         else
1180                 *noslot = true;
1181         return 0;
1182 }
1183
1184 static int osd_preload_exec(struct osd_thread_info *info,
1185                             struct osd_device *dev, struct osd_iit_param *param,
1186                             struct osd_idmap_cache *oic, bool *noslot, int rc)
1187 {
1188         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1189
1190         if (rc == 0) {
1191                 ooc->ooc_cached_items++;
1192                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1193                                         ~OSD_OTABLE_IT_CACHE_MASK;
1194         }
1195         return rc > 0 ? 0 : rc;
1196 }
1197
1198 #define SCRUB_IT_ALL    1
1199 #define SCRUB_IT_CRASH  2
1200
1201 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1202                            bool inconsistent)
1203 {
1204         struct osd_scrub     *scrub  = &dev->od_scrub;
1205         struct ptlrpc_thread *thread = &scrub->os_thread;
1206         struct scrub_file    *sf     = &scrub->os_file;
1207         int                   rc;
1208         ENTRY;
1209
1210         LASSERT(!(flags & SS_AUTO_PARTIAL));
1211
1212         down_write(&scrub->os_rwsem);
1213         scrub->os_in_join = 1;
1214         if (flags & SS_SET_FAILOUT)
1215                 sf->sf_param |= SP_FAILOUT;
1216         else if (flags & SS_CLEAR_FAILOUT)
1217                 sf->sf_param &= ~SP_FAILOUT;
1218
1219         if (flags & SS_SET_DRYRUN)
1220                 sf->sf_param |= SP_DRYRUN;
1221         else if (flags & SS_CLEAR_DRYRUN)
1222                 sf->sf_param &= ~SP_DRYRUN;
1223
1224         if (flags & SS_RESET) {
1225                 osd_scrub_file_reset(scrub,
1226                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1227                         inconsistent ? SF_INCONSISTENT : 0);
1228                 sf->sf_status = SS_SCANNING;
1229         }
1230
1231         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1232                 scrub->os_full_speed = 1;
1233         else
1234                 scrub->os_full_speed = 0;
1235
1236         if (flags & SS_AUTO_FULL) {
1237                 sf->sf_flags |= SF_AUTO;
1238                 scrub->os_full_speed = 1;
1239         }
1240
1241         scrub->os_new_checked = 0;
1242         if (sf->sf_pos_last_checkpoint != 0)
1243                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1244         else
1245                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1246
1247         scrub->os_pos_current = sf->sf_pos_latest_start;
1248         sf->sf_time_latest_start = cfs_time_current_sec();
1249         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1250         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1251         rc = osd_scrub_file_store(scrub);
1252         if (rc != 0)
1253                 CDEBUG(D_LFSCK, "%s: fail to store scrub file when join "
1254                        "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1255
1256         spin_lock(&scrub->os_lock);
1257         scrub->os_waiting = 0;
1258         scrub->os_paused = 0;
1259         scrub->os_partial_scan = 0;
1260         scrub->os_in_join = 0;
1261         scrub->os_full_scrub = 0;
1262         spin_unlock(&scrub->os_lock);
1263         wake_up_all(&thread->t_ctl_waitq);
1264         up_write(&scrub->os_rwsem);
1265
1266         EXIT;
1267 }
1268
1269 static int osd_inode_iteration(struct osd_thread_info *info,
1270                                struct osd_device *dev, __u32 max, bool preload)
1271 {
1272         struct osd_scrub     *scrub  = &dev->od_scrub;
1273         struct ptlrpc_thread *thread = &scrub->os_thread;
1274         struct scrub_file    *sf     = &scrub->os_file;
1275         osd_iit_next_policy   next;
1276         osd_iit_exec_policy   exec;
1277         __u32                *pos;
1278         __u32                *count;
1279         struct osd_iit_param *param;
1280         struct l_wait_info    lwi    = { 0 };
1281         __u32                 limit =
1282                 le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count);
1283         int                   rc;
1284         bool                  noslot = true;
1285         ENTRY;
1286
1287         if (preload)
1288                 goto full;
1289
1290         param = &scrub->os_iit_param;
1291         memset(param, 0, sizeof(*param));
1292         param->sb = osd_sb(dev);
1293
1294         while (scrub->os_partial_scan && !scrub->os_in_join) {
1295                 struct osd_idmap_cache *oic = NULL;
1296
1297                 rc = osd_scrub_next(info, dev, param, &oic, noslot);
1298                 switch (rc) {
1299                 case SCRUB_NEXT_EXIT:
1300                         RETURN(0);
1301                 case SCRUB_NEXT_CRASH:
1302                         RETURN(SCRUB_IT_CRASH);
1303                 case SCRUB_NEXT_FATAL:
1304                         RETURN(-EINVAL);
1305                 case SCRUB_NEXT_WAIT: {
1306                         struct kstatfs *ksfs = &info->oti_ksfs;
1307                         __u64 saved_flags;
1308
1309                         if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1310                             unlikely(sf->sf_items_updated_prior == 0))
1311                                 goto wait;
1312
1313                         if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1314                             scrub->os_full_scrub) {
1315                                 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1316                                                true);
1317                                 goto full;
1318                         }
1319
1320                         rc = param->sb->s_op->statfs(param->sb->s_root, ksfs);
1321                         if (rc == 0) {
1322                                 __u64 used = ksfs->f_files - ksfs->f_ffree;
1323
1324                                 do_div(used, sf->sf_items_updated_prior);
1325                                 /* If we hit too much inconsistent OI
1326                                  * mappings during the partial scan,
1327                                  * then scan the device completely. */
1328                                 if (used < dev->od_full_scrub_ratio) {
1329                                         osd_scrub_join(dev,
1330                                                 SS_AUTO_FULL | SS_RESET, true);
1331                                         goto full;
1332                                 }
1333                         }
1334
1335 wait:
1336                         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1337                             cfs_fail_val > 0)
1338                                 continue;
1339
1340                         saved_flags = sf->sf_flags;
1341                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1342                                           SF_UPGRADE | SF_AUTO);
1343                         sf->sf_status = SS_COMPLETED;
1344                         l_wait_event(thread->t_ctl_waitq,
1345                                      !thread_is_running(thread) ||
1346                                      !scrub->os_partial_scan ||
1347                                      scrub->os_in_join ||
1348                                      !list_empty(&scrub->os_inconsistent_items),
1349                                      &lwi);
1350                         sf->sf_flags = saved_flags;
1351                         sf->sf_status = SS_SCANNING;
1352
1353                         if (unlikely(!thread_is_running(thread)))
1354                                 RETURN(0);
1355
1356                         if (!scrub->os_partial_scan || scrub->os_in_join)
1357                                 goto full;
1358
1359                         continue;
1360                 }
1361                 default:
1362                         LASSERTF(rc == 0, "rc = %d\n", rc);
1363
1364                         osd_scrub_exec(info, dev, param, oic, &noslot, rc);
1365                         break;
1366                 }
1367         }
1368
1369 full:
1370         if (!preload) {
1371                 l_wait_event(thread->t_ctl_waitq,
1372                              !thread_is_running(thread) || !scrub->os_in_join,
1373                              &lwi);
1374
1375                 if (unlikely(!thread_is_running(thread)))
1376                         RETURN(0);
1377         }
1378
1379         noslot = false;
1380         if (!preload) {
1381                 next = osd_scrub_next;
1382                 exec = osd_scrub_exec;
1383                 pos = &scrub->os_pos_current;
1384                 count = &scrub->os_new_checked;
1385                 param->start = *pos;
1386                 param->bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param->sb);
1387                 param->offset =
1388                         (*pos - 1) % LDISKFS_INODES_PER_GROUP(param->sb);
1389                 param->gbase =
1390                         1 + param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
1391         } else {
1392                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1393
1394                 next = osd_preload_next;
1395                 exec = osd_preload_exec;
1396                 pos = &ooc->ooc_pos_preload;
1397                 count = &ooc->ooc_cached_items;
1398                 param = &dev->od_otable_it->ooi_iit_param;
1399         }
1400
1401         rc = 0;
1402         while (*pos <= limit && *count < max) {
1403                 struct ldiskfs_group_desc *desc;
1404                 bool next_group = false;
1405
1406                 desc = ldiskfs_get_group_desc(param->sb, param->bg, NULL);
1407                 if (!desc)
1408                         RETURN(-EIO);
1409
1410                 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1411                         next_group = true;
1412                         goto next_group;
1413                 }
1414
1415                 param->bitmap = ldiskfs_read_inode_bitmap(param->sb, param->bg);
1416                 if (!param->bitmap) {
1417                         CERROR("%s: fail to read bitmap for %u, "
1418                                "scrub will stop, urgent mode\n",
1419                                osd_scrub2name(scrub), (__u32)param->bg);
1420                         RETURN(-EIO);
1421                 }
1422
1423                 do {
1424                         struct osd_idmap_cache *oic = NULL;
1425
1426                         if (param->offset +
1427                                 ldiskfs_itable_unused_count(param->sb, desc) >=
1428                             LDISKFS_INODES_PER_GROUP(param->sb)) {
1429                                 next_group = true;
1430                                 goto next_group;
1431                         }
1432
1433                         rc = next(info, dev, param, &oic, noslot);
1434                         switch (rc) {
1435                         case SCRUB_NEXT_BREAK:
1436                                 next_group = true;
1437                                 goto next_group;
1438                         case SCRUB_NEXT_EXIT:
1439                                 brelse(param->bitmap);
1440                                 RETURN(0);
1441                         case SCRUB_NEXT_CRASH:
1442                                 brelse(param->bitmap);
1443                                 RETURN(SCRUB_IT_CRASH);
1444                         case SCRUB_NEXT_FATAL:
1445                                 brelse(param->bitmap);
1446                                 RETURN(-EINVAL);
1447                         }
1448
1449                         rc = exec(info, dev, param, oic, &noslot, rc);
1450                 } while (!rc && *pos <= limit && *count < max);
1451
1452 next_group:
1453                 if (param->bitmap) {
1454                         brelse(param->bitmap);
1455                         param->bitmap = NULL;
1456                 }
1457
1458                 if (rc < 0)
1459                         GOTO(out, rc);
1460
1461                 if (next_group) {
1462                         param->bg++;
1463                         param->offset = 0;
1464                         param->gbase = 1 +
1465                                 param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
1466                         *pos = param->gbase;
1467                         param->start = *pos;
1468                 }
1469         }
1470
1471         if (*pos > limit)
1472                 RETURN(SCRUB_IT_ALL);
1473
1474 out:
1475         RETURN(rc);
1476 }
1477
1478 static int osd_otable_it_preload(const struct lu_env *env,
1479                                  struct osd_otable_it *it)
1480 {
1481         struct osd_device       *dev   = it->ooi_dev;
1482         struct osd_scrub        *scrub = &dev->od_scrub;
1483         struct osd_otable_cache *ooc   = &it->ooi_cache;
1484         int                      rc;
1485         ENTRY;
1486
1487         rc = osd_inode_iteration(osd_oti_get(env), dev,
1488                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1489         if (rc == SCRUB_IT_ALL)
1490                 it->ooi_all_cached = 1;
1491
1492         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1493                 spin_lock(&scrub->os_lock);
1494                 scrub->os_waiting = 0;
1495                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1496                 spin_unlock(&scrub->os_lock);
1497         }
1498
1499         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1500 }
1501
1502 static int osd_scrub_main(void *args)
1503 {
1504         struct lu_env         env;
1505         struct osd_device    *dev    = (struct osd_device *)args;
1506         struct osd_scrub     *scrub  = &dev->od_scrub;
1507         struct ptlrpc_thread *thread = &scrub->os_thread;
1508         int                   rc;
1509         ENTRY;
1510
1511         rc = lu_env_init(&env, LCT_LOCAL);
1512         if (rc != 0) {
1513                 CDEBUG(D_LFSCK, "%s: OI scrub fail to init env: rc = %d\n",
1514                        osd_scrub2name(scrub), rc);
1515                 GOTO(noenv, rc);
1516         }
1517
1518         rc = osd_scrub_prep(dev);
1519         if (rc != 0) {
1520                 CDEBUG(D_LFSCK, "%s: OI scrub fail to scrub prep: rc = %d\n",
1521                        osd_scrub2name(scrub), rc);
1522                 GOTO(out, rc);
1523         }
1524
1525         if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1526                 struct l_wait_info lwi = { 0 };
1527                 struct osd_otable_it *it = dev->od_otable_it;
1528                 struct osd_otable_cache *ooc = &it->ooi_cache;
1529
1530                 l_wait_event(thread->t_ctl_waitq,
1531                              it->ooi_user_ready || !thread_is_running(thread),
1532                              &lwi);
1533                 if (unlikely(!thread_is_running(thread)))
1534                         GOTO(post, rc = 0);
1535
1536                 scrub->os_pos_current = ooc->ooc_pos_preload;
1537         }
1538
1539         CDEBUG(D_LFSCK, "%s: OI scrub start, flags = 0x%x, pos = %u\n",
1540                osd_scrub2name(scrub), scrub->os_start_flags,
1541                scrub->os_pos_current);
1542
1543         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1544         if (unlikely(rc == SCRUB_IT_CRASH))
1545                 GOTO(out, rc = -EINVAL);
1546         GOTO(post, rc);
1547
1548 post:
1549         rc = osd_scrub_post(scrub, rc);
1550         CDEBUG(D_LFSCK, "%s: OI scrub: stop, pos = %u: rc = %d\n",
1551                osd_scrub2name(scrub), scrub->os_pos_current, rc);
1552
1553 out:
1554         while (!list_empty(&scrub->os_inconsistent_items)) {
1555                 struct osd_inconsistent_item *oii;
1556
1557                 oii = list_entry(scrub->os_inconsistent_items.next,
1558                                      struct osd_inconsistent_item, oii_list);
1559                 list_del_init(&oii->oii_list);
1560                 OBD_FREE_PTR(oii);
1561         }
1562         lu_env_fini(&env);
1563
1564 noenv:
1565         spin_lock(&scrub->os_lock);
1566         thread_set_flags(thread, SVC_STOPPED);
1567         wake_up_all(&thread->t_ctl_waitq);
1568         spin_unlock(&scrub->os_lock);
1569         return rc;
1570 }
1571
1572 /* initial OI scrub */
1573
1574 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1575                          struct dentry *, filldir_t filldir);
1576
1577 #ifdef HAVE_FILLDIR_USE_CTX
1578 static int osd_ios_varfid_fill(struct dir_context *buf, const char *name,
1579                                int namelen, loff_t offset, __u64 ino,
1580                                unsigned d_type);
1581 static int osd_ios_lf_fill(struct dir_context *buf, const char *name,
1582                            int namelen, loff_t offset, __u64 ino,
1583                            unsigned d_type);
1584 static int osd_ios_dl_fill(struct dir_context *buf, const char *name,
1585                            int namelen, loff_t offset, __u64 ino,
1586                            unsigned d_type);
1587 static int osd_ios_uld_fill(struct dir_context *buf, const char *name,
1588                             int namelen, loff_t offset, __u64 ino,
1589                             unsigned d_type);
1590 #else
1591 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1592                                loff_t offset, __u64 ino, unsigned d_type);
1593 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1594                            loff_t offset, __u64 ino, unsigned d_type);
1595 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1596                            loff_t offset, __u64 ino, unsigned d_type);
1597 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1598                             loff_t offset, __u64 ino, unsigned d_type);
1599 #endif
1600
1601 static int
1602 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1603                      struct dentry *dentry, filldir_t filldir);
1604 static int
1605 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1606                   struct dentry *dentry, filldir_t filldir);
1607
1608 static int
1609 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1610                      struct dentry *dentry, filldir_t filldir);
1611
1612 enum osd_lf_flags {
1613         OLF_SCAN_SUBITEMS       = 0x0001,
1614         OLF_HIDE_FID            = 0x0002,
1615         OLF_SHOW_NAME           = 0x0004,
1616         OLF_NO_OI               = 0x0008,
1617         OLF_IDX_IN_FID          = 0x0010,
1618 };
1619
1620 struct osd_lf_map {
1621         char            *olm_name;
1622         struct lu_fid    olm_fid;
1623         __u16            olm_flags;
1624         __u16            olm_namelen;
1625         scandir_t        olm_scandir;
1626         filldir_t        olm_filldir;
1627 };
1628
1629 /* Add the new introduced local files in the list in the future. */
1630 static const struct osd_lf_map osd_lf_maps[] = {
1631         /* CATALOGS */
1632         {
1633                 .olm_name       = CATLIST,
1634                 .olm_fid        = {
1635                         .f_seq  = FID_SEQ_LOCAL_FILE,
1636                         .f_oid  = LLOG_CATALOGS_OID,
1637                 },
1638                 .olm_flags      = OLF_SHOW_NAME,
1639                 .olm_namelen    = sizeof(CATLIST) - 1,
1640         },
1641
1642         /* CONFIGS */
1643         {
1644                 .olm_name       = MOUNT_CONFIGS_DIR,
1645                 .olm_fid        = {
1646                         .f_seq  = FID_SEQ_LOCAL_FILE,
1647                         .f_oid  = MGS_CONFIGS_OID,
1648                 },
1649                 .olm_flags      = OLF_SCAN_SUBITEMS,
1650                 .olm_namelen    = sizeof(MOUNT_CONFIGS_DIR) - 1,
1651                 .olm_scandir    = osd_ios_general_scan,
1652                 .olm_filldir    = osd_ios_varfid_fill,
1653         },
1654
1655         /* NIDTBL_VERSIONS */
1656         {
1657                 .olm_name       = MGS_NIDTBL_DIR,
1658                 .olm_flags      = OLF_SCAN_SUBITEMS,
1659                 .olm_namelen    = sizeof(MGS_NIDTBL_DIR) - 1,
1660                 .olm_scandir    = osd_ios_general_scan,
1661                 .olm_filldir    = osd_ios_varfid_fill,
1662         },
1663
1664         /* PENDING */
1665         {
1666                 .olm_name       = "PENDING",
1667                 .olm_namelen    = sizeof("PENDING") - 1,
1668         },
1669
1670         /* ROOT */
1671         {
1672                 .olm_name       = "ROOT",
1673                 .olm_fid        = {
1674                         .f_seq  = FID_SEQ_ROOT,
1675                         .f_oid  = FID_OID_ROOT,
1676                 },
1677                 .olm_flags      = OLF_SCAN_SUBITEMS | OLF_HIDE_FID,
1678                 .olm_namelen    = sizeof("ROOT") - 1,
1679                 .olm_scandir    = osd_ios_ROOT_scan,
1680         },
1681
1682         /* changelog_catalog */
1683         {
1684                 .olm_name       = CHANGELOG_CATALOG,
1685                 .olm_namelen    = sizeof(CHANGELOG_CATALOG) - 1,
1686         },
1687
1688         /* changelog_users */
1689         {
1690                 .olm_name       = CHANGELOG_USERS,
1691                 .olm_namelen    = sizeof(CHANGELOG_USERS) - 1,
1692         },
1693
1694         /* fld */
1695         {
1696                 .olm_name       = "fld",
1697                 .olm_fid        = {
1698                         .f_seq  = FID_SEQ_LOCAL_FILE,
1699                         .f_oid  = FLD_INDEX_OID,
1700                 },
1701                 .olm_flags      = OLF_SHOW_NAME,
1702                 .olm_namelen    = sizeof("fld") - 1,
1703         },
1704
1705         /* last_rcvd */
1706         {
1707                 .olm_name       = LAST_RCVD,
1708                 .olm_fid        = {
1709                         .f_seq  = FID_SEQ_LOCAL_FILE,
1710                         .f_oid  = LAST_RECV_OID,
1711                 },
1712                 .olm_flags      = OLF_SHOW_NAME,
1713                 .olm_namelen    = sizeof(LAST_RCVD) - 1,
1714         },
1715
1716         /* reply_data */
1717         {
1718                 .olm_name       = REPLY_DATA,
1719                 .olm_fid        = {
1720                         .f_seq  = FID_SEQ_LOCAL_FILE,
1721                         .f_oid  = REPLY_DATA_OID,
1722                 },
1723                 .olm_flags      = OLF_SHOW_NAME,
1724                 .olm_namelen    = sizeof(REPLY_DATA) - 1,
1725         },
1726
1727         /* lov_objid */
1728         {
1729                 .olm_name       = LOV_OBJID,
1730                 .olm_fid        = {
1731                         .f_seq  = FID_SEQ_LOCAL_FILE,
1732                         .f_oid  = MDD_LOV_OBJ_OID,
1733                 },
1734                 .olm_flags      = OLF_SHOW_NAME,
1735                 .olm_namelen    = sizeof(LOV_OBJID) - 1,
1736         },
1737
1738         /* lov_objseq */
1739         {
1740                 .olm_name       = LOV_OBJSEQ,
1741                 .olm_fid        = {
1742                         .f_seq  = FID_SEQ_LOCAL_FILE,
1743                         .f_oid  = MDD_LOV_OBJ_OSEQ,
1744                 },
1745                 .olm_flags      = OLF_SHOW_NAME,
1746                 .olm_namelen    = sizeof(LOV_OBJSEQ) - 1,
1747         },
1748
1749         /* quota_master */
1750         {
1751                 .olm_name       = QMT_DIR,
1752                 .olm_flags      = OLF_SCAN_SUBITEMS,
1753                 .olm_namelen    = sizeof(QMT_DIR) - 1,
1754                 .olm_scandir    = osd_ios_general_scan,
1755                 .olm_filldir    = osd_ios_varfid_fill,
1756         },
1757
1758         /* quota_slave */
1759         {
1760                 .olm_name       = QSD_DIR,
1761                 .olm_flags      = OLF_SCAN_SUBITEMS,
1762                 .olm_namelen    = sizeof(QSD_DIR) - 1,
1763                 .olm_scandir    = osd_ios_general_scan,
1764                 .olm_filldir    = osd_ios_varfid_fill,
1765         },
1766
1767         /* seq_ctl */
1768         {
1769                 .olm_name       = "seq_ctl",
1770                 .olm_fid        = {
1771                         .f_seq  = FID_SEQ_LOCAL_FILE,
1772                         .f_oid  = FID_SEQ_CTL_OID,
1773                 },
1774                 .olm_flags      = OLF_SHOW_NAME,
1775                 .olm_namelen    = sizeof("seq_ctl") - 1,
1776         },
1777
1778         /* seq_srv */
1779         {
1780                 .olm_name       = "seq_srv",
1781                 .olm_fid        = {
1782                         .f_seq  = FID_SEQ_LOCAL_FILE,
1783                         .f_oid  = FID_SEQ_SRV_OID,
1784                 },
1785                 .olm_flags      = OLF_SHOW_NAME,
1786                 .olm_namelen    = sizeof("seq_srv") - 1,
1787         },
1788
1789         /* health_check */
1790         {
1791                 .olm_name       = HEALTH_CHECK,
1792                 .olm_fid        = {
1793                         .f_seq  = FID_SEQ_LOCAL_FILE,
1794                         .f_oid  = OFD_HEALTH_CHECK_OID,
1795                 },
1796                 .olm_flags      = OLF_SHOW_NAME,
1797                 .olm_namelen    = sizeof(HEALTH_CHECK) - 1,
1798         },
1799
1800         /* LFSCK */
1801         {
1802                 .olm_name       = LFSCK_DIR,
1803                 .olm_namelen    = sizeof(LFSCK_DIR) - 1,
1804                 .olm_scandir    = osd_ios_general_scan,
1805                 .olm_filldir    = osd_ios_varfid_fill,
1806         },
1807
1808         /* lfsck_bookmark */
1809         {
1810                 .olm_name       = LFSCK_BOOKMARK,
1811                 .olm_namelen    = sizeof(LFSCK_BOOKMARK) - 1,
1812         },
1813
1814         /* lfsck_layout */
1815         {
1816                 .olm_name       = LFSCK_LAYOUT,
1817                 .olm_namelen    = sizeof(LFSCK_LAYOUT) - 1,
1818         },
1819
1820         /* lfsck_namespace */
1821         {
1822                 .olm_name       = LFSCK_NAMESPACE,
1823                 .olm_namelen    = sizeof(LFSCK_NAMESPACE) - 1,
1824         },
1825
1826         /* OBJECTS, upgrade from old device */
1827         {
1828                 .olm_name       = OBJECTS,
1829                 .olm_flags      = OLF_SCAN_SUBITEMS,
1830                 .olm_namelen    = sizeof(OBJECTS) - 1,
1831                 .olm_scandir    = osd_ios_OBJECTS_scan,
1832         },
1833
1834         /* lquota_v2.user, upgrade from old device */
1835         {
1836                 .olm_name       = "lquota_v2.user",
1837                 .olm_namelen    = sizeof("lquota_v2.user") - 1,
1838         },
1839
1840         /* lquota_v2.group, upgrade from old device */
1841         {
1842                 .olm_name       = "lquota_v2.group",
1843                 .olm_namelen    = sizeof("lquota_v2.group") - 1,
1844         },
1845
1846         /* LAST_GROUP, upgrade from old device */
1847         {
1848                 .olm_name       = "LAST_GROUP",
1849                 .olm_fid        = {
1850                         .f_seq  = FID_SEQ_LOCAL_FILE,
1851                         .f_oid  = OFD_LAST_GROUP_OID,
1852                 },
1853                 .olm_flags      = OLF_SHOW_NAME,
1854                 .olm_namelen    = sizeof("LAST_GROUP") - 1,
1855         },
1856
1857         /* committed batchid for cross-MDT operation */
1858         {
1859                 .olm_name       = "BATCHID",
1860                 .olm_fid        = {
1861                         .f_seq  = FID_SEQ_LOCAL_FILE,
1862                         .f_oid  = BATCHID_COMMITTED_OID,
1863                 },
1864                 .olm_flags      = OLF_SHOW_NAME,
1865                 .olm_namelen    = sizeof("BATCHID") - 1,
1866         },
1867
1868         /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1869          * and f_oid = index for their log files.  See lu_update_log{_dir}_fid()
1870          * for more details. */
1871
1872         /* update_log */
1873         {
1874                 .olm_name       = "update_log",
1875                 .olm_fid        = {
1876                         .f_seq  = FID_SEQ_UPDATE_LOG,
1877                 },
1878                 .olm_flags      = OLF_SHOW_NAME | OLF_IDX_IN_FID,
1879                 .olm_namelen    = sizeof("update_log") - 1,
1880         },
1881
1882         /* update_log_dir */
1883         {
1884                 .olm_name       = "update_log_dir",
1885                 .olm_fid        = {
1886                         .f_seq  = FID_SEQ_UPDATE_LOG_DIR,
1887                 },
1888                 .olm_flags      = OLF_SHOW_NAME | OLF_SCAN_SUBITEMS |
1889                                   OLF_IDX_IN_FID,
1890                 .olm_namelen    = sizeof("update_log_dir") - 1,
1891                 .olm_scandir    = osd_ios_general_scan,
1892                 .olm_filldir    = osd_ios_uld_fill,
1893         },
1894
1895         /* lost+found */
1896         {
1897                 .olm_name       = "lost+found",
1898                 .olm_fid        = {
1899                         .f_seq  = FID_SEQ_LOCAL_FILE,
1900                         .f_oid  = OSD_LPF_OID,
1901                 },
1902                 .olm_flags      = OLF_SCAN_SUBITEMS,
1903                 .olm_namelen    = sizeof("lost+found") - 1,
1904                 .olm_scandir    = osd_ios_general_scan,
1905                 .olm_filldir    = osd_ios_lf_fill,
1906         },
1907
1908         {
1909                 .olm_name       = NULL
1910         }
1911 };
1912
1913 /* Add the new introduced files under .lustre/ in the list in the future. */
1914 static const struct osd_lf_map osd_dl_maps[] = {
1915         /* .lustre/fid */
1916         {
1917                 .olm_name       = "fid",
1918                 .olm_fid        = {
1919                         .f_seq  = FID_SEQ_DOT_LUSTRE,
1920                         .f_oid  = FID_OID_DOT_LUSTRE_OBF,
1921                 },
1922                 .olm_namelen    = sizeof("fid") - 1,
1923         },
1924         /* .lustre/lost+found */
1925         {
1926                 .olm_name       = "lost+found",
1927                 .olm_fid        = {
1928                         .f_seq  = FID_SEQ_DOT_LUSTRE,
1929                         .f_oid  = FID_OID_DOT_LUSTRE_LPF,
1930                 },
1931                 .olm_namelen    = sizeof("lost+found") - 1,
1932         },
1933         {
1934                 .olm_name       = NULL
1935         }
1936 };
1937
1938 struct osd_ios_item {
1939         struct list_head oii_list;
1940         struct dentry   *oii_dentry;
1941         scandir_t        oii_scandir;
1942         filldir_t        oii_filldir;
1943 };
1944
1945 struct osd_ios_filldir_buf {
1946 #ifdef HAVE_DIR_CONTEXT
1947         /* please keep it as first member */
1948         struct dir_context       ctx;
1949 #endif
1950         struct osd_thread_info  *oifb_info;
1951         struct osd_device       *oifb_dev;
1952         struct dentry           *oifb_dentry;
1953         int                      oifb_items;
1954 };
1955
1956 static inline struct dentry *
1957 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1958 {
1959         struct dentry *dentry;
1960
1961         dentry = ll_lookup_one_len(name, parent, namelen);
1962         if (IS_ERR(dentry)) {
1963                 int rc = PTR_ERR(dentry);
1964
1965                 if (rc != -ENOENT)
1966                         CERROR("Fail to find %.*s in %.*s (%lu/%u): rc = %d\n",
1967                                namelen, name, parent->d_name.len,
1968                                parent->d_name.name, parent->d_inode->i_ino,
1969                                parent->d_inode->i_generation, rc);
1970
1971                 return dentry;
1972         }
1973
1974         if (dentry->d_inode == NULL) {
1975                 dput(dentry);
1976                 return ERR_PTR(-ENOENT);
1977         }
1978
1979         return dentry;
1980 }
1981
1982 static int
1983 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1984                  scandir_t scandir, filldir_t filldir)
1985 {
1986         struct osd_ios_item *item;
1987         ENTRY;
1988
1989         OBD_ALLOC_PTR(item);
1990         if (item == NULL)
1991                 RETURN(-ENOMEM);
1992
1993         INIT_LIST_HEAD(&item->oii_list);
1994         item->oii_dentry = dget(dentry);
1995         item->oii_scandir = scandir;
1996         item->oii_filldir = filldir;
1997         list_add_tail(&item->oii_list, &dev->od_ios_list);
1998
1999         RETURN(0);
2000 }
2001
2002 /**
2003  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
2004  *
2005  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
2006  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
2007  * reference the inode, or fixed if it is missing or references another inode.
2008  */
2009 static int
2010 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
2011                  struct inode *inode, const struct lu_fid *fid, int flags)
2012 {
2013         struct lustre_mdt_attrs *lma    = &info->oti_ost_attrs.loa_lma;
2014         struct osd_inode_id     *id     = &info->oti_id;
2015         struct osd_inode_id     *id2    = &info->oti_id2;
2016         struct osd_scrub        *scrub  = &dev->od_scrub;
2017         struct scrub_file       *sf     = &scrub->os_file;
2018         struct lu_fid            tfid;
2019         int                      rc;
2020         ENTRY;
2021
2022         rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
2023                          &info->oti_ost_attrs);
2024         if (rc != 0 && rc != -ENODATA) {
2025                 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
2026                        "rc = %d\n", osd_name(dev), rc);
2027
2028                 RETURN(rc);
2029         }
2030
2031         osd_id_gen(id, inode->i_ino, inode->i_generation);
2032         if (rc == -ENODATA) {
2033                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
2034                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
2035                 } else {
2036                         tfid = *fid;
2037                         if (flags & OLF_IDX_IN_FID) {
2038                                 LASSERT(dev->od_index >= 0);
2039
2040                                 tfid.f_oid = dev->od_index;
2041                         }
2042                 }
2043                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
2044                 if (rc != 0) {
2045                         CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
2046                               "scrub: rc = %d\n", osd_name(dev), rc);
2047
2048                         RETURN(rc);
2049                 }
2050         } else {
2051                 if (lma->lma_compat & LMAC_NOT_IN_OI)
2052                         RETURN(0);
2053
2054                 tfid = lma->lma_self_fid;
2055         }
2056
2057         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
2058         if (rc != 0) {
2059                 if (rc != -ENOENT)
2060                         RETURN(rc);
2061
2062                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
2063                                                DTO_INDEX_INSERT, true, 0, NULL);
2064                 if (rc > 0)
2065                         rc = 0;
2066
2067                 RETURN(rc);
2068         }
2069
2070         if (osd_id_eq_strict(id, id2))
2071                 RETURN(0);
2072
2073         if (!(sf->sf_flags & SF_INCONSISTENT)) {
2074                 osd_scrub_file_reset(scrub,
2075                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2076                                      SF_INCONSISTENT);
2077                 rc = osd_scrub_file_store(scrub);
2078                 if (rc != 0)
2079                         RETURN(rc);
2080         }
2081
2082         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
2083                                        DTO_INDEX_UPDATE, true, 0, NULL);
2084         if (rc > 0)
2085                 rc = 0;
2086
2087         RETURN(rc);
2088 }
2089
2090 /**
2091  * It scans the /lost+found, and for the OST-object (with filter_fid
2092  * or filter_fid_18_23), move them back to its proper /O/<seq>/d<x>.
2093  */
2094 #ifdef HAVE_FILLDIR_USE_CTX
2095 static int osd_ios_lf_fill(struct dir_context *buf,
2096 #else
2097 static int osd_ios_lf_fill(void *buf,
2098 #endif
2099                            const char *name, int namelen,
2100                            loff_t offset, __u64 ino, unsigned d_type)
2101 {
2102         struct osd_ios_filldir_buf *fill_buf =
2103                 (struct osd_ios_filldir_buf *)buf;
2104         struct osd_thread_info     *info     = fill_buf->oifb_info;
2105         struct osd_device          *dev      = fill_buf->oifb_dev;
2106         struct lu_fid              *fid      = &info->oti_fid;
2107         struct osd_scrub           *scrub    = &dev->od_scrub;
2108         struct dentry              *parent   = fill_buf->oifb_dentry;
2109         struct dentry              *child;
2110         struct inode               *dir      = parent->d_inode;
2111         struct inode               *inode;
2112         int                         rc;
2113         ENTRY;
2114
2115         fill_buf->oifb_items++;
2116
2117         /* skip any '.' started names */
2118         if (name[0] == '.')
2119                 RETURN(0);
2120
2121         scrub->os_lf_scanned++;
2122         child = osd_ios_lookup_one_len(name, parent, namelen);
2123         if (IS_ERR(child)) {
2124                 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
2125                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
2126                 RETURN(0);
2127         }
2128
2129         inode = child->d_inode;
2130         if (S_ISDIR(inode->i_mode)) {
2131                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2132                                       osd_ios_lf_fill);
2133                 if (rc != 0)
2134                         CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
2135                               "rc = %d\n", osd_name(dev), namelen, name, rc);
2136                 GOTO(put, rc);
2137         }
2138
2139         if (!S_ISREG(inode->i_mode))
2140                 GOTO(put, rc = 0);
2141
2142         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
2143         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
2144                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
2145                 if (rc == 0) {
2146                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
2147                                "/lost+found.\n", namelen, name, PFID(fid));
2148                         scrub->os_lf_repaired++;
2149                 } else {
2150                         CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
2151                                DFID": rc = %d\n",
2152                                osd_name(dev), namelen, name, PFID(fid), rc);
2153                 }
2154         }
2155
2156         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
2157          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
2158          *      can process them in furtuer. */
2159
2160         GOTO(put, rc);
2161
2162 put:
2163         if (rc < 0)
2164                 scrub->os_lf_failed++;
2165         dput(child);
2166         /* skip the failure to make the scanning to continue. */
2167         return 0;
2168 }
2169
2170 #ifdef HAVE_FILLDIR_USE_CTX
2171 static int osd_ios_varfid_fill(struct dir_context *buf,
2172 #else
2173 static int osd_ios_varfid_fill(void *buf,
2174 #endif
2175                                const char *name, int namelen,
2176                                loff_t offset, __u64 ino, unsigned d_type)
2177 {
2178         struct osd_ios_filldir_buf *fill_buf =
2179                 (struct osd_ios_filldir_buf *)buf;
2180         struct osd_device          *dev      = fill_buf->oifb_dev;
2181         struct dentry              *child;
2182         int                         rc;
2183         ENTRY;
2184
2185         fill_buf->oifb_items++;
2186
2187         /* skip any '.' started names */
2188         if (name[0] == '.')
2189                 RETURN(0);
2190
2191         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2192         if (IS_ERR(child))
2193                 RETURN(PTR_ERR(child));
2194
2195         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2196                               NULL, 0);
2197         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
2198                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2199                                       osd_ios_varfid_fill);
2200         dput(child);
2201
2202         RETURN(rc);
2203 }
2204
2205 #ifdef HAVE_FILLDIR_USE_CTX
2206 static int osd_ios_dl_fill(struct dir_context *buf,
2207 #else
2208 static int osd_ios_dl_fill(void *buf,
2209 #endif
2210                            const char *name, int namelen,
2211                            loff_t offset, __u64 ino, unsigned d_type)
2212 {
2213         struct osd_ios_filldir_buf *fill_buf =
2214                 (struct osd_ios_filldir_buf *)buf;
2215         struct osd_device          *dev      = fill_buf->oifb_dev;
2216         const struct osd_lf_map    *map;
2217         struct dentry              *child;
2218         int                         rc       = 0;
2219         ENTRY;
2220
2221         fill_buf->oifb_items++;
2222
2223         /* skip any '.' started names */
2224         if (name[0] == '.')
2225                 RETURN(0);
2226
2227         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
2228                 if (map->olm_namelen != namelen)
2229                         continue;
2230
2231                 if (strncmp(map->olm_name, name, namelen) == 0)
2232                         break;
2233         }
2234
2235         if (map->olm_name == NULL)
2236                 RETURN(0);
2237
2238         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2239         if (IS_ERR(child))
2240                 RETURN(PTR_ERR(child));
2241
2242         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2243                               &map->olm_fid, map->olm_flags);
2244         dput(child);
2245
2246         RETURN(rc);
2247 }
2248
2249 #ifdef HAVE_FILLDIR_USE_CTX
2250 static int osd_ios_uld_fill(struct dir_context *buf,
2251 #else
2252 static int osd_ios_uld_fill(void *buf,
2253 #endif
2254                             const char *name, int namelen,
2255                             loff_t offset, __u64 ino, unsigned d_type)
2256 {
2257         struct osd_ios_filldir_buf *fill_buf =
2258                 (struct osd_ios_filldir_buf *)buf;
2259         struct dentry              *child;
2260         struct lu_fid               tfid;
2261         int                         rc       = 0;
2262         ENTRY;
2263
2264         fill_buf->oifb_items++;
2265
2266         /* skip any non-DFID format name */
2267         if (name[0] != '[')
2268                 RETURN(0);
2269
2270         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2271         if (IS_ERR(child))
2272                 RETURN(PTR_ERR(child));
2273
2274         /* skip the start '[' */
2275         sscanf(&name[1], SFID, RFID(&tfid));
2276         if (fid_is_sane(&tfid))
2277                 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
2278                                       child->d_inode, &tfid, 0);
2279         else
2280                 rc = -EIO;
2281         dput(child);
2282
2283         RETURN(rc);
2284 }
2285
2286 #ifdef HAVE_FILLDIR_USE_CTX
2287 static int osd_ios_root_fill(struct dir_context *buf,
2288 #else
2289 static int osd_ios_root_fill(void *buf,
2290 #endif
2291                              const char *name, int namelen,
2292                              loff_t offset, __u64 ino, unsigned d_type)
2293 {
2294         struct osd_ios_filldir_buf *fill_buf =
2295                 (struct osd_ios_filldir_buf *)buf;
2296         struct osd_device          *dev      = fill_buf->oifb_dev;
2297         const struct osd_lf_map    *map;
2298         struct dentry              *child;
2299         int                         rc       = 0;
2300         ENTRY;
2301
2302         fill_buf->oifb_items++;
2303
2304         /* skip any '.' started names */
2305         if (name[0] == '.')
2306                 RETURN(0);
2307
2308         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2309                 if (map->olm_namelen != namelen)
2310                         continue;
2311
2312                 if (strncmp(map->olm_name, name, namelen) == 0)
2313                         break;
2314         }
2315
2316         if (map->olm_name == NULL)
2317                 RETURN(0);
2318
2319         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2320         if (IS_ERR(child))
2321                 RETURN(PTR_ERR(child));
2322
2323         if (!(map->olm_flags & OLF_NO_OI))
2324                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2325                                       &map->olm_fid, map->olm_flags);
2326         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2327                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2328                                       map->olm_filldir);
2329         dput(child);
2330
2331         RETURN(rc);
2332 }
2333
2334 static int
2335 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2336                      struct dentry *dentry, filldir_t filldir)
2337 {
2338         struct osd_ios_filldir_buf    buf   = {
2339 #ifdef HAVE_DIR_CONTEXT
2340                                                 .ctx.actor = filldir,
2341 #endif
2342                                                 .oifb_info = info,
2343                                                 .oifb_dev = dev,
2344                                                 .oifb_dentry = dentry };
2345         struct file                  *filp  = &info->oti_file;
2346         struct inode                 *inode = dentry->d_inode;
2347         const struct file_operations *fops  = inode->i_fop;
2348         int                           rc;
2349         ENTRY;
2350
2351         LASSERT(filldir != NULL);
2352
2353         filp->f_pos = 0;
2354         filp->f_path.dentry = dentry;
2355         filp->f_mode = FMODE_64BITHASH;
2356         filp->f_mapping = inode->i_mapping;
2357         filp->f_op = fops;
2358         filp->private_data = NULL;
2359         set_file_inode(filp, inode);
2360
2361         do {
2362                 buf.oifb_items = 0;
2363 #ifdef HAVE_DIR_CONTEXT
2364                 buf.ctx.pos = filp->f_pos;
2365                 rc = fops->iterate(filp, &buf.ctx);
2366                 filp->f_pos = buf.ctx.pos;
2367 #else
2368                 rc = fops->readdir(filp, &buf, filldir);
2369 #endif
2370         } while (rc >= 0 && buf.oifb_items > 0 &&
2371                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
2372         fops->release(inode, filp);
2373
2374         RETURN(rc);
2375 }
2376
2377 static int
2378 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2379                   struct dentry *dentry, filldir_t filldir)
2380 {
2381         struct osd_scrub  *scrub  = &dev->od_scrub;
2382         struct scrub_file *sf     = &scrub->os_file;
2383         struct dentry     *child;
2384         int                rc;
2385         ENTRY;
2386
2387         /* It is existing MDT0 device. We only allow the case of object without
2388          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2389          * can generate IGIF mode FID for the object and related OI mapping. If
2390          * it is on other MDTs, then becuase file-level backup/restore, related
2391          * OI mapping may be invalid already, we do not know which is the right
2392          * FID for the object. We only allow IGIF objects to reside on the MDT0.
2393          *
2394          * XXX: For the case of object on non-MDT0 device with neither LMA nor
2395          *      "fid" xattr, then something crashed. We cannot re-generate the
2396          *      FID directly, instead, the OI scrub will scan the OI structure
2397          *      and try to re-generate the LMA from the OI mapping. But if the
2398          *      OI mapping crashed or lost also, then we have to give up under
2399          *      double failure cases. */
2400         scrub->os_convert_igif = 1;
2401         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2402                                        strlen(dot_lustre_name));
2403         if (IS_ERR(child)) {
2404                 rc = PTR_ERR(child);
2405                 if (rc == -ENOENT) {
2406                         /* It is 1.8 MDT device. */
2407                         if (!(sf->sf_flags & SF_UPGRADE)) {
2408                                 osd_scrub_file_reset(scrub,
2409                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2410                                         SF_UPGRADE);
2411                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2412                                 rc = osd_scrub_file_store(scrub);
2413                         } else {
2414                                 rc = 0;
2415                         }
2416                 }
2417         } else {
2418                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2419                  * so the client will get IGIF for the ".lustre" object when
2420                  * the MDT restart.
2421                  *
2422                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2423                  * it does not know whether there are some old clients cached
2424                  * the ".lustre" IGIF during the upgrading. Two choices:
2425                  *
2426                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2427                  *    It will allow the old connected clients to access the
2428                  *    ".lustre" with cached IGIF. But it will cause others
2429                  *    on the MDT failed to check "fid_is_dot_lustre()".
2430                  *
2431                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2432                  *    for ".lustre" in spite of whether there are some clients
2433                  *    cached the ".lustre" IGIF or not. It enables the check
2434                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
2435                  *    that the old connected clients cannot access the ".lustre"
2436                  *    with the cached IGIF.
2437                  *
2438                  * Usually, it is rare case for the old connected clients
2439                  * to access the ".lustre" with cached IGIF. So we prefer
2440                  * to the solution 2). */
2441                 rc = osd_ios_scan_one(info, dev, child->d_inode,
2442                                       &LU_DOT_LUSTRE_FID, 0);
2443                 if (rc == 0)
2444                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2445                                               osd_ios_dl_fill);
2446                 dput(child);
2447         }
2448
2449         RETURN(rc);
2450 }
2451
2452 static int
2453 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2454                      struct dentry *dentry, filldir_t filldir)
2455 {
2456         struct osd_scrub  *scrub  = &dev->od_scrub;
2457         struct scrub_file *sf     = &scrub->os_file;
2458         struct dentry     *child;
2459         int                rc;
2460         ENTRY;
2461
2462         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2463                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2464                 rc = osd_scrub_file_store(scrub);
2465                 if (rc != 0)
2466                         RETURN(rc);
2467         }
2468
2469         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2470         if (!IS_ERR(child)) {
2471                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2472                 dput(child);
2473         } else {
2474                 rc = PTR_ERR(child);
2475         }
2476
2477         if (rc != 0 && rc != -ENOENT)
2478                 RETURN(rc);
2479
2480         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2481         if (!IS_ERR(child)) {
2482                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2483                 dput(child);
2484         } else {
2485                 rc = PTR_ERR(child);
2486         }
2487
2488         if (rc == -ENOENT)
2489                 rc = 0;
2490
2491         RETURN(rc);
2492 }
2493
2494 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2495                                 struct osd_device *dev)
2496 {
2497         struct osd_ios_item     *item    = NULL;
2498         scandir_t                scandir = osd_ios_general_scan;
2499         filldir_t                filldir = osd_ios_root_fill;
2500         struct dentry           *dentry  = osd_sb(dev)->s_root;
2501         const struct osd_lf_map *map     = osd_lf_maps;
2502         int                      rc;
2503         ENTRY;
2504
2505         /* Lookup IGIF in OI by force for initial OI scrub. */
2506         dev->od_igif_inoi = 1;
2507
2508         while (1) {
2509                 rc = scandir(info, dev, dentry, filldir);
2510                 if (item != NULL) {
2511                         dput(item->oii_dentry);
2512                         OBD_FREE_PTR(item);
2513                 }
2514
2515                 if (rc != 0)
2516                         break;
2517
2518                 if (list_empty(&dev->od_ios_list))
2519                         break;
2520
2521                 item = list_entry(dev->od_ios_list.next,
2522                                   struct osd_ios_item, oii_list);
2523                 list_del_init(&item->oii_list);
2524
2525                 LASSERT(item->oii_scandir != NULL);
2526                 scandir = item->oii_scandir;
2527                 filldir = item->oii_filldir;
2528                 dentry = item->oii_dentry;
2529         }
2530
2531         while (!list_empty(&dev->od_ios_list)) {
2532                 item = list_entry(dev->od_ios_list.next,
2533                                   struct osd_ios_item, oii_list);
2534                 list_del_init(&item->oii_list);
2535                 dput(item->oii_dentry);
2536                 OBD_FREE_PTR(item);
2537         }
2538
2539         if (rc != 0)
2540                 RETURN(rc);
2541
2542         /* There maybe the case that the object has been removed, but its OI
2543          * mapping is still in the OI file, such as the "CATALOGS" after MDT
2544          * file-level backup/restore. So here cleanup the stale OI mappings. */
2545         while (map->olm_name != NULL) {
2546                 struct dentry *child;
2547
2548                 if (fid_is_zero(&map->olm_fid)) {
2549                         map++;
2550                         continue;
2551                 }
2552
2553                 child = osd_ios_lookup_one_len(map->olm_name,
2554                                                osd_sb(dev)->s_root,
2555                                                map->olm_namelen);
2556                 if (!IS_ERR(child))
2557                         dput(child);
2558                 else if (PTR_ERR(child) == -ENOENT)
2559                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2560                                                   NULL, DTO_INDEX_DELETE,
2561                                                   true, 0, NULL);
2562                 map++;
2563         }
2564
2565         RETURN(0);
2566 }
2567
2568 char *osd_lf_fid2name(const struct lu_fid *fid)
2569 {
2570         const struct osd_lf_map *map = osd_lf_maps;
2571
2572         while (map->olm_name != NULL) {
2573                 if (!lu_fid_eq(fid, &map->olm_fid)) {
2574                         map++;
2575                         continue;
2576                 }
2577
2578                 if (map->olm_flags & OLF_SHOW_NAME)
2579                         return map->olm_name;
2580                 else
2581                         return "";
2582         }
2583
2584         return NULL;
2585 }
2586
2587 /* OI scrub start/stop */
2588
2589 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2590 {
2591         struct osd_scrub     *scrub  = &dev->od_scrub;
2592         struct ptlrpc_thread *thread = &scrub->os_thread;
2593         struct l_wait_info    lwi    = { 0 };
2594         struct task_struct   *task;
2595         int                   rc;
2596         ENTRY;
2597
2598         if (dev->od_dt_dev.dd_rdonly)
2599                 RETURN(-EROFS);
2600
2601         /* os_lock: sync status between stop and scrub thread */
2602         spin_lock(&scrub->os_lock);
2603
2604 again:
2605         if (thread_is_running(thread)) {
2606                 spin_unlock(&scrub->os_lock);
2607                 if (!(scrub->os_file.sf_flags & SF_AUTO ||
2608                       scrub->os_partial_scan) ||
2609                      (flags & SS_AUTO_PARTIAL))
2610                         RETURN(-EALREADY);
2611
2612                 osd_scrub_join(dev, flags, false);
2613                 spin_lock(&scrub->os_lock);
2614                 if (!thread_is_running(thread))
2615                         goto again;
2616
2617                 spin_unlock(&scrub->os_lock);
2618                 RETURN(0);
2619         }
2620
2621         if (unlikely(thread_is_stopping(thread))) {
2622                 spin_unlock(&scrub->os_lock);
2623                 l_wait_event(thread->t_ctl_waitq,
2624                              thread_is_stopped(thread),
2625                              &lwi);
2626                 spin_lock(&scrub->os_lock);
2627                 goto again;
2628         }
2629         spin_unlock(&scrub->os_lock);
2630
2631         if (scrub->os_file.sf_status == SS_COMPLETED) {
2632                 if (!(flags & SS_SET_FAILOUT))
2633                         flags |= SS_CLEAR_FAILOUT;
2634
2635                 if (!(flags & SS_SET_DRYRUN))
2636                         flags |= SS_CLEAR_DRYRUN;
2637
2638                 flags |= SS_RESET;
2639         }
2640
2641         scrub->os_start_flags = flags;
2642         thread_set_flags(thread, 0);
2643         task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2644         if (IS_ERR(task)) {
2645                 rc = PTR_ERR(task);
2646                 CERROR("%s: cannot start iteration thread: rc = %d\n",
2647                        osd_scrub2name(scrub), rc);
2648                 RETURN(rc);
2649         }
2650
2651         l_wait_event(thread->t_ctl_waitq,
2652                      thread_is_running(thread) || thread_is_stopped(thread),
2653                      &lwi);
2654
2655         RETURN(0);
2656 }
2657
2658 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2659 {
2660         int rc;
2661         ENTRY;
2662
2663         /* od_otable_mutex: prevent curcurrent start/stop */
2664         mutex_lock(&dev->od_otable_mutex);
2665         rc = do_osd_scrub_start(dev, flags);
2666         mutex_unlock(&dev->od_otable_mutex);
2667
2668         RETURN(rc == -EALREADY ? 0 : rc);
2669 }
2670
2671 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2672 {
2673         struct ptlrpc_thread *thread = &scrub->os_thread;
2674         struct l_wait_info    lwi    = { 0 };
2675
2676         /* os_lock: sync status between stop and scrub thread */
2677         spin_lock(&scrub->os_lock);
2678         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2679                 thread_set_flags(thread, SVC_STOPPING);
2680                 spin_unlock(&scrub->os_lock);
2681                 wake_up_all(&thread->t_ctl_waitq);
2682                 l_wait_event(thread->t_ctl_waitq,
2683                              thread_is_stopped(thread),
2684                              &lwi);
2685                 /* Do not skip the last lock/unlock, which can guarantee that
2686                  * the caller cannot return until the OI scrub thread exit. */
2687                 spin_lock(&scrub->os_lock);
2688         }
2689         spin_unlock(&scrub->os_lock);
2690 }
2691
2692 static void osd_scrub_stop(struct osd_device *dev)
2693 {
2694         /* od_otable_mutex: prevent curcurrent start/stop */
2695         mutex_lock(&dev->od_otable_mutex);
2696         dev->od_scrub.os_paused = 1;
2697         do_osd_scrub_stop(&dev->od_scrub);
2698         mutex_unlock(&dev->od_otable_mutex);
2699 }
2700
2701 /* OI scrub setup/cleanup */
2702
2703 static const char osd_scrub_name[] = "OI_scrub";
2704
2705 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2706 {
2707         struct osd_thread_info     *info   = osd_oti_get(env);
2708         struct osd_scrub           *scrub  = &dev->od_scrub;
2709         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
2710         struct scrub_file          *sf     = &scrub->os_file;
2711         struct super_block         *sb     = osd_sb(dev);
2712         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
2713         struct lvfs_run_ctxt        saved;
2714         struct file                *filp;
2715         struct inode               *inode;
2716         struct lu_fid              *fid    = &info->oti_fid;
2717         bool                        dirty  = false;
2718         bool                        restored = false;
2719         int                         rc     = 0;
2720         ENTRY;
2721
2722         memset(scrub, 0, sizeof(*scrub));
2723         OBD_SET_CTXT_MAGIC(ctxt);
2724         ctxt->pwdmnt = dev->od_mnt;
2725         ctxt->pwd = dev->od_mnt->mnt_root;
2726         ctxt->fs = get_ds();
2727
2728         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2729         init_rwsem(&scrub->os_rwsem);
2730         spin_lock_init(&scrub->os_lock);
2731         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2732
2733         push_ctxt(&saved, ctxt);
2734         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2735         if (IS_ERR(filp)) {
2736                 pop_ctxt(&saved, ctxt);
2737                 RETURN(PTR_ERR(filp));
2738         }
2739
2740         inode = file_inode(filp);
2741         /* 'What the @fid is' is not imporatant, because the object
2742          * has no OI mapping, and only is visible inside the OSD.*/
2743         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2744         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2745         if (rc != 0) {
2746                 filp_close(filp, NULL);
2747                 pop_ctxt(&saved, ctxt);
2748                 RETURN(rc);
2749         }
2750
2751         scrub->os_inode = igrab(inode);
2752         filp_close(filp, NULL);
2753         pop_ctxt(&saved, ctxt);
2754
2755         rc = osd_scrub_file_load(scrub);
2756         if (rc == -ENOENT) {
2757                 osd_scrub_file_init(scrub, es->s_uuid);
2758                 /* If the "/O" dir does not exist when mount (indicated by
2759                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2760                  * then it is quite probably that the device is a new one,
2761                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2762                  *
2763                  * For the rare case that "/O" and "OI_scrub" both lost on
2764                  * an old device, it can be found and cleared later.
2765                  *
2766                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2767                  * need to check "filter_fid_18_23" and to convert it to
2768                  * "filter_fid" for each object, and all the IGIF should
2769                  * have their FID mapping in OI files already. */
2770                 if (dev->od_maybe_new)
2771                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2772                 dirty = true;
2773         } else if (rc != 0) {
2774                 GOTO(cleanup_inode, rc);
2775         } else {
2776                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2777                         struct obd_uuid *old_uuid;
2778                         struct obd_uuid *new_uuid;
2779
2780                         OBD_ALLOC_PTR(old_uuid);
2781                         OBD_ALLOC_PTR(new_uuid);
2782                         if (old_uuid == NULL || new_uuid == NULL) {
2783                                 CERROR("%s: UUID has been changed, but"
2784                                        "failed to allocate RAM for report\n",
2785                                        osd_dev2name(dev));
2786                         } else {
2787                                 class_uuid_unparse(sf->sf_uuid, old_uuid);
2788                                 class_uuid_unparse(es->s_uuid, new_uuid);
2789                                 CERROR("%s: UUID has been changed from "
2790                                        "%s to %s\n", osd_dev2name(dev),
2791                                        old_uuid->uuid, new_uuid->uuid);
2792                         }
2793                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2794                         dirty = true;
2795                         restored = true;
2796                         if (old_uuid != NULL)
2797                                 OBD_FREE_PTR(old_uuid);
2798                         if (new_uuid != NULL)
2799                                 OBD_FREE_PTR(new_uuid);
2800                 } else if (sf->sf_status == SS_SCANNING) {
2801                         sf->sf_status = SS_CRASHED;
2802                         dirty = true;
2803                 }
2804         }
2805
2806         if (sf->sf_pos_last_checkpoint != 0)
2807                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2808         else
2809                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2810
2811         if (dirty) {
2812                 rc = osd_scrub_file_store(scrub);
2813                 if (rc != 0)
2814                         GOTO(cleanup_inode, rc);
2815         }
2816
2817         /* Initialize OI files. */
2818         rc = osd_oi_init(info, dev, restored);
2819         if (rc < 0)
2820                 GOTO(cleanup_inode, rc);
2821
2822         rc = osd_initial_OI_scrub(info, dev);
2823         if (rc != 0)
2824                 GOTO(cleanup_oi, rc);
2825
2826         if (sf->sf_flags & SF_UPGRADE ||
2827             !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2828               sf->sf_success_count > 0)) {
2829                 dev->od_igif_inoi = 0;
2830                 dev->od_check_ff = dev->od_is_ost;
2831         } else {
2832                 dev->od_igif_inoi = 1;
2833                 dev->od_check_ff = 0;
2834         }
2835
2836         if (sf->sf_flags & SF_INCONSISTENT)
2837                 /* The 'od_igif_inoi' will be set under the
2838                  * following cases:
2839                  * 1) new created system, or
2840                  * 2) restored from file-level backup, or
2841                  * 3) the upgrading completed.
2842                  *
2843                  * The 'od_igif_inoi' may be cleared by OI scrub
2844                  * later if found that the system is upgrading. */
2845                 dev->od_igif_inoi = 1;
2846
2847         if (!dev->od_dt_dev.dd_rdonly && !dev->od_noscrub &&
2848             ((sf->sf_status == SS_PAUSED) ||
2849              (sf->sf_status == SS_CRASHED &&
2850               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2851                               SF_UPGRADE | SF_AUTO)) ||
2852              (sf->sf_status == SS_INIT &&
2853               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2854                               SF_UPGRADE))))
2855                 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2856
2857         if (rc != 0)
2858                 GOTO(cleanup_oi, rc);
2859
2860         /* it is possible that dcache entries may keep objects after they are
2861          * deleted by OSD. While it looks safe this can cause object data to
2862          * stay until umount causing failures in tests calculating free space,
2863          * e.g. replay-ost-single. Since those dcache entries are not used
2864          * anymore let's just free them after use here */
2865         shrink_dcache_sb(sb);
2866
2867         RETURN(0);
2868 cleanup_oi:
2869         osd_oi_fini(info, dev);
2870 cleanup_inode:
2871         iput(scrub->os_inode);
2872         scrub->os_inode = NULL;
2873
2874         return rc;
2875 }
2876
2877 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2878 {
2879         struct osd_scrub *scrub = &dev->od_scrub;
2880
2881         LASSERT(dev->od_otable_it == NULL);
2882
2883         if (scrub->os_inode != NULL) {
2884                 osd_scrub_stop(dev);
2885                 iput(scrub->os_inode);
2886                 scrub->os_inode = NULL;
2887         }
2888         if (dev->od_oi_table != NULL)
2889                 osd_oi_fini(osd_oti_get(env), dev);
2890 }
2891
2892 /* object table based iteration APIs */
2893
2894 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2895                                        struct dt_object *dt, __u32 attr)
2896 {
2897         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2898         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2899         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2900         struct osd_scrub       *scrub = &dev->od_scrub;
2901         struct osd_otable_it   *it;
2902         __u32                   start = 0;
2903         int                     rc;
2904         ENTRY;
2905
2906         /* od_otable_mutex: prevent curcurrent init/fini */
2907         mutex_lock(&dev->od_otable_mutex);
2908         if (dev->od_otable_it != NULL)
2909                 GOTO(out, it = ERR_PTR(-EALREADY));
2910
2911         OBD_ALLOC_PTR(it);
2912         if (it == NULL)
2913                 GOTO(out, it = ERR_PTR(-ENOMEM));
2914
2915         dev->od_otable_it = it;
2916         it->ooi_dev = dev;
2917         it->ooi_cache.ooc_consumer_idx = -1;
2918         if (flags & DOIF_OUTUSED)
2919                 it->ooi_used_outside = 1;
2920
2921         if (flags & DOIF_RESET)
2922                 start |= SS_RESET;
2923
2924         if (valid & DOIV_ERROR_HANDLE) {
2925                 if (flags & DOIF_FAILOUT)
2926                         start |= SS_SET_FAILOUT;
2927                 else
2928                         start |= SS_CLEAR_FAILOUT;
2929         }
2930
2931         if (valid & DOIV_DRYRUN) {
2932                 if (flags & DOIF_DRYRUN)
2933                         start |= SS_SET_DRYRUN;
2934                 else
2935                         start |= SS_CLEAR_DRYRUN;
2936         }
2937
2938         rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2939         if (rc < 0 && rc != -EALREADY) {
2940                 dev->od_otable_it = NULL;
2941                 OBD_FREE_PTR(it);
2942                 GOTO(out, it = ERR_PTR(rc));
2943         }
2944
2945         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2946
2947         GOTO(out, it);
2948
2949 out:
2950         mutex_unlock(&dev->od_otable_mutex);
2951         return (struct dt_it *)it;
2952 }
2953
2954 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2955 {
2956         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2957         struct osd_device    *dev = it->ooi_dev;
2958
2959         /* od_otable_mutex: prevent curcurrent init/fini */
2960         mutex_lock(&dev->od_otable_mutex);
2961         do_osd_scrub_stop(&dev->od_scrub);
2962         LASSERT(dev->od_otable_it == it);
2963
2964         dev->od_otable_it = NULL;
2965         mutex_unlock(&dev->od_otable_mutex);
2966         OBD_FREE_PTR(it);
2967 }
2968
2969 static int osd_otable_it_get(const struct lu_env *env,
2970                              struct dt_it *di, const struct dt_key *key)
2971 {
2972         return 0;
2973 }
2974
2975 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2976 {
2977 }
2978
2979 static inline int
2980 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2981 {
2982         spin_lock(&scrub->os_lock);
2983         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2984             scrub->os_waiting ||
2985             !thread_is_running(&scrub->os_thread))
2986                 it->ooi_waiting = 0;
2987         else
2988                 it->ooi_waiting = 1;
2989         spin_unlock(&scrub->os_lock);
2990
2991         return !it->ooi_waiting;
2992 }
2993
2994 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2995 {
2996         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2997         struct osd_device       *dev    = it->ooi_dev;
2998         struct osd_scrub        *scrub  = &dev->od_scrub;
2999         struct osd_otable_cache *ooc    = &it->ooi_cache;
3000         struct ptlrpc_thread    *thread = &scrub->os_thread;
3001         struct l_wait_info       lwi    = { 0 };
3002         int                      rc;
3003         ENTRY;
3004
3005         LASSERT(it->ooi_user_ready);
3006
3007 again:
3008         if (!thread_is_running(thread) && !it->ooi_used_outside)
3009                 RETURN(1);
3010
3011         if (ooc->ooc_cached_items > 0) {
3012                 ooc->ooc_cached_items--;
3013                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
3014                                         ~OSD_OTABLE_IT_CACHE_MASK;
3015                 RETURN(0);
3016         }
3017
3018         if (it->ooi_all_cached) {
3019                 l_wait_event(thread->t_ctl_waitq,
3020                              !thread_is_running(thread),
3021                              &lwi);
3022                 RETURN(1);
3023         }
3024
3025         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
3026                 spin_lock(&scrub->os_lock);
3027                 scrub->os_waiting = 0;
3028                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
3029                 spin_unlock(&scrub->os_lock);
3030         }
3031
3032         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
3033                 l_wait_event(thread->t_ctl_waitq,
3034                              osd_otable_it_wakeup(scrub, it),
3035                              &lwi);
3036
3037         if (!thread_is_running(thread) && !it->ooi_used_outside)
3038                 RETURN(1);
3039
3040         rc = osd_otable_it_preload(env, it);
3041         if (rc >= 0)
3042                 goto again;
3043
3044         RETURN(rc);
3045 }
3046
3047 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
3048                                         const struct dt_it *di)
3049 {
3050         return NULL;
3051 }
3052
3053 static int osd_otable_it_key_size(const struct lu_env *env,
3054                                   const struct dt_it *di)
3055 {
3056         return sizeof(__u64);
3057 }
3058
3059 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
3060                              struct dt_rec *rec, __u32 attr)
3061 {
3062         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
3063         struct osd_otable_cache *ooc = &it->ooi_cache;
3064
3065         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
3066
3067         /* Filter out Invald FID already. */
3068         LASSERTF(fid_is_sane((struct lu_fid *)rec),
3069                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
3070                  PFID((struct lu_fid *)rec),
3071                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
3072
3073         return 0;
3074 }
3075
3076 static __u64 osd_otable_it_store(const struct lu_env *env,
3077                                  const struct dt_it *di)
3078 {
3079         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
3080         struct osd_otable_cache *ooc = &it->ooi_cache;
3081         __u64                    hash;
3082
3083         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
3084                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
3085         else
3086                 hash = ooc->ooc_pos_preload;
3087         return hash;
3088 }
3089
3090 /**
3091  * Set the OSD layer iteration start position as the specified hash.
3092  */
3093 static int osd_otable_it_load(const struct lu_env *env,
3094                               const struct dt_it *di, __u64 hash)
3095 {
3096         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
3097         struct osd_device       *dev   = it->ooi_dev;
3098         struct osd_otable_cache *ooc   = &it->ooi_cache;
3099         struct osd_scrub        *scrub = &dev->od_scrub;
3100         struct osd_iit_param    *param = &it->ooi_iit_param;
3101         int                      rc;
3102         ENTRY;
3103
3104         /* Forbid to set iteration position after iteration started. */
3105         if (it->ooi_user_ready)
3106                 RETURN(-EPERM);
3107
3108         LASSERT(!scrub->os_partial_scan);
3109
3110         if (hash > OSD_OTABLE_MAX_HASH)
3111                 hash = OSD_OTABLE_MAX_HASH;
3112
3113         /* The hash is the last checkpoint position,
3114          * we will start from the next one. */
3115         ooc->ooc_pos_preload = hash + 1;
3116         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
3117                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
3118
3119         it->ooi_user_ready = 1;
3120         if (!scrub->os_full_speed)
3121                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
3122
3123         memset(param, 0, sizeof(*param));
3124         param->sb = osd_sb(dev);
3125         param->start = ooc->ooc_pos_preload;
3126         param->bg = (ooc->ooc_pos_preload - 1) /
3127                     LDISKFS_INODES_PER_GROUP(param->sb);
3128         param->offset = (ooc->ooc_pos_preload - 1) %
3129                         LDISKFS_INODES_PER_GROUP(param->sb);
3130         param->gbase = 1 + param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
3131
3132         /* Unplug OSD layer iteration by the first next() call. */
3133         rc = osd_otable_it_next(env, (struct dt_it *)it);
3134
3135         RETURN(rc);
3136 }
3137
3138 static int osd_otable_it_key_rec(const struct lu_env *env,
3139                                  const struct dt_it *di, void *key_rec)
3140 {
3141         return 0;
3142 }
3143
3144 const struct dt_index_operations osd_otable_ops = {
3145         .dio_it = {
3146                 .init     = osd_otable_it_init,
3147                 .fini     = osd_otable_it_fini,
3148                 .get      = osd_otable_it_get,
3149                 .put      = osd_otable_it_put,
3150                 .next     = osd_otable_it_next,
3151                 .key      = osd_otable_it_key,
3152                 .key_size = osd_otable_it_key_size,
3153                 .rec      = osd_otable_it_rec,
3154                 .store    = osd_otable_it_store,
3155                 .load     = osd_otable_it_load,
3156                 .key_rec  = osd_otable_it_key_rec,
3157         }
3158 };
3159
3160 /* high priority inconsistent items list APIs */
3161
3162 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL  60
3163
3164 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
3165                    int insert)
3166 {
3167         struct osd_inconsistent_item *oii;
3168         struct osd_scrub             *scrub  = &dev->od_scrub;
3169         struct ptlrpc_thread         *thread = &scrub->os_thread;
3170         int                           wakeup = 0;
3171         ENTRY;
3172
3173         OBD_ALLOC_PTR(oii);
3174         if (unlikely(oii == NULL))
3175                 RETURN(-ENOMEM);
3176
3177         INIT_LIST_HEAD(&oii->oii_list);
3178         oii->oii_cache = *oic;
3179         oii->oii_insert = insert;
3180
3181         if (scrub->os_partial_scan) {
3182                 __u64 now = cfs_time_current_sec();
3183
3184                 /* If there haven't been errors in a long time,
3185                  * decay old count until either the errors are
3186                  * gone or we reach the current interval. */
3187                 while (unlikely(scrub->os_bad_oimap_count > 0 &&
3188                                 scrub->os_bad_oimap_time +
3189                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
3190                         scrub->os_bad_oimap_count >>= 1;
3191                         scrub->os_bad_oimap_time +=
3192                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
3193                 }
3194
3195                 scrub->os_bad_oimap_time = now;
3196                 if (++scrub->os_bad_oimap_count >
3197                     dev->od_full_scrub_threshold_rate)
3198                         scrub->os_full_scrub = 1;
3199         }
3200
3201         spin_lock(&scrub->os_lock);
3202         if (unlikely(!thread_is_running(thread))) {
3203                 spin_unlock(&scrub->os_lock);
3204                 OBD_FREE_PTR(oii);
3205                 RETURN(-EAGAIN);
3206         }
3207
3208         if (list_empty(&scrub->os_inconsistent_items))
3209                 wakeup = 1;
3210         list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
3211         spin_unlock(&scrub->os_lock);
3212
3213         if (wakeup != 0)
3214                 wake_up_all(&thread->t_ctl_waitq);
3215
3216         RETURN(0);
3217 }
3218
3219 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
3220                    struct osd_inode_id *id)
3221 {
3222         struct osd_scrub             *scrub = &dev->od_scrub;
3223         struct osd_inconsistent_item *oii;
3224         ENTRY;
3225
3226         spin_lock(&scrub->os_lock);
3227         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
3228                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
3229                         *id = oii->oii_cache.oic_lid;
3230                         spin_unlock(&scrub->os_lock);
3231                         RETURN(0);
3232                 }
3233         }
3234         spin_unlock(&scrub->os_lock);
3235
3236         RETURN(-ENOENT);
3237 }
3238
3239 /* OI scrub dump */
3240
3241 static const char *scrub_status_names[] = {
3242         "init",
3243         "scanning",
3244         "completed",
3245         "failed",
3246         "stopped",
3247         "paused",
3248         "crashed",
3249         NULL
3250 };
3251
3252 static const char *scrub_flags_names[] = {
3253         "recreated",
3254         "inconsistent",
3255         "auto",
3256         "upgrade",
3257         NULL
3258 };
3259
3260 static const char *scrub_param_names[] = {
3261         "failout",
3262         "dryrun",
3263         NULL
3264 };
3265
3266 static void scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
3267                             const char *prefix)
3268 {
3269         int flag;
3270         int i;
3271
3272         seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
3273
3274         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
3275                 if (flag & bits) {
3276                         bits &= ~flag;
3277                         seq_printf(m, "%s%c", names[i],
3278                                    bits != 0 ? ',' : '\n');
3279                 }
3280         }
3281 }
3282
3283 static void scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
3284 {
3285         if (time != 0)
3286                 seq_printf(m, "%s: %llu seconds\n", prefix,
3287                            cfs_time_current_sec() - time);
3288         else
3289                 seq_printf(m, "%s: N/A\n", prefix);
3290 }
3291
3292 static void scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
3293 {
3294         if (pos != 0)
3295                 seq_printf(m, "%s: %llu\n", prefix, pos);
3296         else
3297                 seq_printf(m, "%s: N/A\n", prefix);
3298 }
3299
3300 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
3301 {
3302         struct osd_scrub  *scrub   = &dev->od_scrub;
3303         struct scrub_file *sf      = &scrub->os_file;
3304         __u64              checked;
3305         __u64              speed;
3306
3307         down_read(&scrub->os_rwsem);
3308         seq_printf(m, "name: OI_scrub\n"
3309                    "magic: 0x%x\n"
3310                    "oi_files: %d\n"
3311                    "status: %s\n",
3312                    sf->sf_magic, (int)sf->sf_oi_count,
3313                    scrub_status_names[sf->sf_status]);
3314
3315         scrub_bits_dump(m, sf->sf_flags, scrub_flags_names, "flags");
3316
3317         scrub_bits_dump(m, sf->sf_param, scrub_param_names, "param");
3318
3319         scrub_time_dump(m, sf->sf_time_last_complete,
3320                         "time_since_last_completed");
3321
3322         scrub_time_dump(m, sf->sf_time_latest_start,
3323                         "time_since_latest_start");
3324
3325         scrub_time_dump(m, sf->sf_time_last_checkpoint,
3326                         "time_since_last_checkpoint");
3327
3328         scrub_pos_dump(m, sf->sf_pos_latest_start,
3329                         "latest_start_position");
3330
3331         scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
3332                         "last_checkpoint_position");
3333
3334         scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
3335                         "first_failure_position");
3336
3337         checked = sf->sf_items_checked + scrub->os_new_checked;
3338         seq_printf(m, "checked: %llu\n"
3339                    "%s: %llu\n"
3340                    "failed: %llu\n"
3341                    "prior_%s: %llu\n"
3342                    "noscrub: %llu\n"
3343                    "igif: %llu\n"
3344                    "success_count: %u\n",
3345                    checked,
3346                    sf->sf_param & SP_DRYRUN ? "inconsistent" : "updated",
3347                    sf->sf_items_updated, sf->sf_items_failed,
3348                    sf->sf_param & SP_DRYRUN ? "inconsistent" : "updated",
3349                    sf->sf_items_updated_prior, sf->sf_items_noscrub,
3350                    sf->sf_items_igif, sf->sf_success_count);
3351
3352         speed = checked;
3353         if (thread_is_running(&scrub->os_thread)) {
3354                 cfs_duration_t duration = cfs_time_current() -
3355                                           scrub->os_time_last_checkpoint;
3356                 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
3357                                                      MSEC_PER_SEC);
3358                 __u32 rtime = sf->sf_run_time +
3359                               cfs_duration_sec(duration + HALF_SEC);
3360
3361                 if (duration != 0)
3362                         do_div(new_checked, duration);
3363                 if (rtime != 0)
3364                         do_div(speed, rtime);
3365                 seq_printf(m, "run_time: %u seconds\n"
3366                            "average_speed: %llu objects/sec\n"
3367                            "real-time_speed: %llu objects/sec\n"
3368                            "current_position: %u\n"
3369                            "lf_scanned: %llu\n"
3370                            "lf_%s: %llu\n"
3371                            "lf_failed: %llu\n",
3372                            rtime, speed, new_checked, scrub->os_pos_current,
3373                            scrub->os_lf_scanned,
3374                            sf->sf_param & SP_DRYRUN ?
3375                                 "inconsistent" : "repaired",
3376                            scrub->os_lf_repaired,
3377                            scrub->os_lf_failed);
3378                 seq_printf(m, "inodes_per_group: %lu\n"
3379                            "current_iit_group: %u\n"
3380                            "current_iit_base: %u\n"
3381                            "current_iit_offset: %u\n"
3382                            "scrub_in_prior: %s\n"
3383                            "scrub_full_speed: %s\n"
3384                            "partial_scan: %s\n",
3385                            LDISKFS_INODES_PER_GROUP(osd_sb(dev)),
3386                            scrub->os_iit_param.bg,
3387                            scrub->os_iit_param.gbase,
3388                            scrub->os_iit_param.offset,
3389                            scrub->os_in_prior ? "yes" : "no",
3390                            scrub->os_full_speed ? "yes" : "no",
3391                            scrub->os_partial_scan ? "yes" : "no");
3392         } else {
3393                 if (sf->sf_run_time != 0)
3394                         do_div(speed, sf->sf_run_time);
3395                 seq_printf(m, "run_time: %u seconds\n"
3396                            "average_speed: %llu objects/sec\n"
3397                            "real-time_speed: N/A\n"
3398                            "current_position: N/A\n"
3399                            "lf_scanned: %llu\n"
3400                            "lf_%s: %llu\n"
3401                            "lf_failed: %llu\n",
3402                            sf->sf_run_time, speed, scrub->os_lf_scanned,
3403                            sf->sf_param & SP_DRYRUN ?
3404                                 "inconsistent" : "repaired",
3405                            scrub->os_lf_repaired, scrub->os_lf_failed);
3406         }
3407
3408         up_read(&scrub->os_rwsem);
3409         return 0;
3410 }