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