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                     removed = false;
 381         bool                     reset   = true;
 382         ENTRY;
 383
 384         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
 385                 RETURN(0);
 386
 387         if (fid_is_idif(fid) && dev->od_index_in_idif == 0) {
 388                 struct ost_id *oi = &info->oti_ostid;
 389
 390                 fid_to_ostid(fid, oi);
 391                 ostid_to_fid(tfid, oi, 0);
 392         } else {
 393                 *tfid = *fid;
 394         }
 395
 396         /* We want the LMA to fit into the 256-byte OST inode, so operate
 397          * as following:
 398          * 1) read old XATTR_NAME_FID and save the parent FID;
 399          * 2) delete the old XATTR_NAME_FID;
 400          * 3) make new LMA and add it;
 401          * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
 402          *
 403          * Making the LMA to fit into the 256-byte OST inode can save time for
 404          * normal osd_check_lma() and for other OI scrub scanning in future.
 405          * So it is worth to make some slow conversion here. */
 406         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 407                                 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
 408         if (IS_ERR(jh)) {
 409                 rc = PTR_ERR(jh);
 410                 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
 411                        DFID": rc = %d\n", osd_name(dev), PFID(tfid), rc);
 412                 RETURN(rc);
 413         }
 414
 415         /* 1) read old XATTR_NAME_FID and save the parent FID */
 416         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 417         if (rc == sizeof(*ff)) {
 418                 /* 2) delete the old XATTR_NAME_FID */
 419                 ll_vfs_dq_init(inode);
 420                 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
 421                 if (rc != 0)
 422                         GOTO(stop, rc);
 423
 424                 removed = true;
 425         } else if (unlikely(rc == -ENODATA)) {
 426                 reset = false;
 427         } else if (rc != sizeof(struct filter_fid)) {
 428                 GOTO(stop, rc = -EINVAL);
 429         }
 430
 431         /* 3) make new LMA and add it */
 432         rc = osd_ea_fid_set(info, inode, tfid, LMAC_FID_ON_OST, 0);
 433         if (rc == 0 && reset)
 434                 size = sizeof(struct filter_fid);
 435         else if (rc != 0 && removed)
 436                 /* If failed, we should try to add the old back. */
 437                 size = sizeof(struct filter_fid_old);
 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 };
 818
 819 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 820                                    struct osd_device *dev,
 821                                    struct osd_iit_param *param,
 822                                    struct osd_idmap_cache **oic,
 823                                    const bool noslot);
 824
 825 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 826                                    struct osd_device *dev,
 827                                    struct osd_iit_param *param,
 828                                    struct osd_idmap_cache *oic,
 829                                    bool *noslot, int rc);
 830
 831 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 832 {
 833         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 834                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 835         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 836                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 837                 return SCRUB_NEXT_BREAK;
 838         } else {
 839                 *pos = param->gbase + param->offset;
 840                 return 0;
 841         }
 842 }
 843
 844 /**
 845  * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
 846  * \retval 0: FID-on-MDT
 847  */
 848 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
 849                                       struct osd_device *dev,
 850                                       struct lu_fid *fid)
 851 {
 852         /* XXX: The initial OI scrub will scan the top level /O to generate
 853          *      a small local FLDB according to the <seq>. If the given FID
 854          *      is in the local FLDB, then it is FID-on-OST; otherwise it's
 855          *      quite possible for FID-on-MDT. */
 856         if (dev->od_is_ost)
 857                 return SCRUB_NEXT_OSTOBJ_OLD;
 858         else
 859                 return 0;
 860 }
 861
 862 static int osd_scrub_get_fid(struct osd_thread_info *info,
 863                              struct osd_device *dev, struct inode *inode,
 864                              struct lu_fid *fid, bool scrub)
 865 {
 866         struct lustre_mdt_attrs *lma     = &info->oti_mdt_attrs;
 867         int                      rc;
 868         bool                     has_lma = false;
 869
 870         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 871         if (rc == 0) {
 872                 has_lma = true;
 873                 if (lma->lma_compat & LMAC_NOT_IN_OI ||
 874                     lma->lma_incompat & LMAI_AGENT)
 875                         return SCRUB_NEXT_CONTINUE;
 876
 877                 *fid = lma->lma_self_fid;
 878                 if (!scrub)
 879                         return 0;
 880
 881                 if (lma->lma_compat & LMAC_FID_ON_OST)
 882                         return SCRUB_NEXT_OSTOBJ;
 883
 884                 if (fid_is_idif(fid))
 885                         return SCRUB_NEXT_OSTOBJ_OLD;
 886
 887                 /* For local object. */
 888                 if (fid_is_internal(fid))
 889                         return 0;
 890
 891                 /* For external visible MDT-object with non-normal FID. */
 892                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
 893                         return 0;
 894
 895                 /* For the object with normal FID, it may be MDT-object,
 896                  * or may be 2.4 OST-object, need further distinguish.
 897                  * Fall through to next section. */
 898         }
 899
 900         if (rc == -ENODATA || rc == 0) {
 901                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
 902                 if (rc == 0) {
 903                         if (scrub)
 904                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
 905                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 906                         return rc;
 907                 }
 908
 909                 if (rc > 0) {
 910                         if (!has_lma)
 911                                 /* It is FID-on-OST, but we do not know how
 912                                  * to generate its FID, ignore it directly. */
 913                                 rc = SCRUB_NEXT_CONTINUE;
 914                         else
 915                                 /* It is 2.4 OST-object. */
 916                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 917                         return rc;
 918                 }
 919
 920                 if (rc != -ENODATA)
 921                         return rc;
 922
 923                 if (!has_lma) {
 924                         if (dev->od_scrub.os_convert_igif) {
 925                                 lu_igif_build(fid, inode->i_ino,
 926                                               inode->i_generation);
 927                                 if (scrub)
 928                                         rc = SCRUB_NEXT_NOLMA;
 929                                 else
 930                                         rc = 0;
 931                         } else {
 932                                 /* It may be FID-on-OST, or may be FID for
 933                                  * non-MDT0, anyway, we do not know how to
 934                                  * generate its FID, ignore it directly. */
 935                                 rc = SCRUB_NEXT_CONTINUE;
 936                         }
 937                         return rc;
 938                 }
 939
 940                 /* For OI scrub case only: the object has LMA but has no ff
 941                  * (or ff crashed). It may be MDT-object, may be OST-object
 942                  * with crashed ff. The last check is local FLDB. */
 943                 rc = osd_scrub_check_local_fldb(info, dev, fid);
 944         }
 945
 946         return rc;
 947 }
 948
 949 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 950                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 951                         struct super_block *sb, bool scrub)
 952 {
 953         struct inode *inode;
 954         int           rc;
 955         ENTRY;
 956
 957         /* Not handle the backend root object and agent parent object.
 958          * They are neither visible to namespace nor have OI mappings. */
 959         if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
 960                      pos == osd_remote_parent_ino(dev)))
 961                 RETURN(SCRUB_NEXT_CONTINUE);
 962
 963         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 964         inode = osd_iget(info, dev, lid);
 965         if (IS_ERR(inode)) {
 966                 rc = PTR_ERR(inode);
 967                 /* The inode may be removed after bitmap searching, or the
 968                  * file is new created without inode initialized yet. */
 969                 if (rc == -ENOENT || rc == -ESTALE)
 970                         RETURN(SCRUB_NEXT_CONTINUE);
 971
 972                 CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: "
 973                        "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name,
 974                        pos, rc);
 975                 RETURN(rc);
 976         }
 977
 978         /* It is an EA inode, no OI mapping for it, skip it. */
 979         if (osd_is_ea_inode(inode))
 980                 GOTO(put, rc = SCRUB_NEXT_CONTINUE);
 981
 982         if (scrub &&
 983             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 984                 /* Only skip it for the first OI scrub accessing. */
 985                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 986                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
 987         }
 988
 989         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
 990
 991         GOTO(put, rc);
 992
 993 put:
 994         iput(inode);
 995         return rc;
 996 }
 997
 998 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 999                           struct osd_iit_param *param,
1000                           struct osd_idmap_cache **oic, const bool noslot)
1001 {
1002         struct osd_scrub     *scrub  = &dev->od_scrub;
1003         struct ptlrpc_thread *thread = &scrub->os_thread;
1004         struct lu_fid        *fid;
1005         struct osd_inode_id  *lid;
1006         int                   rc;
1007
1008         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
1009                 struct l_wait_info lwi;
1010
1011                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
1012                 if (likely(lwi.lwi_timeout > 0))
1013                         l_wait_event(thread->t_ctl_waitq,
1014                                 !list_empty(&scrub->os_inconsistent_items) ||
1015                                 !thread_is_running(thread),
1016                                 &lwi);
1017         }
1018
1019         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
1020                 spin_lock(&scrub->os_lock);
1021                 thread_set_flags(thread, SVC_STOPPING);
1022                 spin_unlock(&scrub->os_lock);
1023                 return SCRUB_NEXT_CRASH;
1024         }
1025
1026         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
1027                 return SCRUB_NEXT_FATAL;
1028
1029         if (unlikely(!thread_is_running(thread)))
1030                 return SCRUB_NEXT_EXIT;
1031
1032         if (!list_empty(&scrub->os_inconsistent_items)) {
1033                 spin_lock(&scrub->os_lock);
1034                 if (likely(!list_empty(&scrub->os_inconsistent_items))) {
1035                         struct osd_inconsistent_item *oii;
1036
1037                         oii = list_entry(scrub->os_inconsistent_items.next,
1038                                 struct osd_inconsistent_item, oii_list);
1039                         spin_unlock(&scrub->os_lock);
1040
1041                         *oic = &oii->oii_cache;
1042                         scrub->os_in_prior = 1;
1043
1044                         return 0;
1045                 }
1046                 spin_unlock(&scrub->os_lock);
1047         }
1048
1049         if (noslot)
1050                 return SCRUB_NEXT_WAIT;
1051
1052         rc = osd_iit_next(param, &scrub->os_pos_current);
1053         if (rc != 0)
1054                 return rc;
1055
1056         *oic = &scrub->os_oic;
1057         fid = &(*oic)->oic_fid;
1058         lid = &(*oic)->oic_lid;
1059         rc = osd_iit_iget(info, dev, fid, lid,
1060                           scrub->os_pos_current, param->sb, true);
1061         return rc;
1062 }
1063
1064 static int osd_preload_next(struct osd_thread_info *info,
1065                             struct osd_device *dev, struct osd_iit_param *param,
1066                             struct osd_idmap_cache **oic, const bool noslot)
1067 {
1068         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
1069         struct osd_scrub        *scrub;
1070         struct ptlrpc_thread    *thread;
1071         int                      rc;
1072
1073         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1074         if (rc != 0)
1075                 return rc;
1076
1077         scrub = &dev->od_scrub;
1078         thread = &scrub->os_thread;
1079         if (thread_is_running(thread) &&
1080             ooc->ooc_pos_preload >= scrub->os_pos_current)
1081                 return SCRUB_NEXT_EXIT;
1082
1083         rc = osd_iit_iget(info, dev,
1084                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1085                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1086                           ooc->ooc_pos_preload, param->sb, false);
1087         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1088          * ignore the failure, so it still need to skip the inode next time. */
1089         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1090         return rc;
1091 }
1092
1093 static inline int
1094 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1095 {
1096         spin_lock(&scrub->os_lock);
1097         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1098             !list_empty(&scrub->os_inconsistent_items) ||
1099             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1100                 scrub->os_waiting = 0;
1101         else
1102                 scrub->os_waiting = 1;
1103         spin_unlock(&scrub->os_lock);
1104
1105         return !scrub->os_waiting;
1106 }
1107
1108 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1109                           struct osd_iit_param *param,
1110                           struct osd_idmap_cache *oic, bool *noslot, int rc)
1111 {
1112         struct l_wait_info       lwi    = { 0 };
1113         struct osd_scrub        *scrub  = &dev->od_scrub;
1114         struct scrub_file       *sf     = &scrub->os_file;
1115         struct ptlrpc_thread    *thread = &scrub->os_thread;
1116         struct osd_otable_it    *it     = dev->od_otable_it;
1117         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
1118
1119         switch (rc) {
1120         case SCRUB_NEXT_CONTINUE:
1121                 goto next;
1122         case SCRUB_NEXT_WAIT:
1123                 goto wait;
1124         case SCRUB_NEXT_NOSCRUB:
1125                 down_write(&scrub->os_rwsem);
1126                 scrub->os_new_checked++;
1127                 sf->sf_items_noscrub++;
1128                 up_write(&scrub->os_rwsem);
1129                 goto next;
1130         }
1131
1132         rc = osd_scrub_check_update(info, dev, oic, rc);
1133         if (rc != 0) {
1134                 scrub->os_in_prior = 0;
1135                 return rc;
1136         }
1137
1138         rc = osd_scrub_checkpoint(scrub);
1139         if (rc != 0) {
1140                 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1141                        "rc = %d\n", osd_scrub2name(scrub),
1142                        scrub->os_pos_current, rc);
1143                 /* Continue, as long as the scrub itself can go ahead. */
1144         }
1145
1146         if (scrub->os_in_prior) {
1147                 scrub->os_in_prior = 0;
1148                 return 0;
1149         }
1150
1151 next:
1152         scrub->os_pos_current = param->gbase + ++(param->offset);
1153
1154 wait:
1155         if (it != NULL && it->ooi_waiting && ooc != NULL &&
1156             ooc->ooc_pos_preload < scrub->os_pos_current) {
1157                 spin_lock(&scrub->os_lock);
1158                 it->ooi_waiting = 0;
1159                 wake_up_all(&thread->t_ctl_waitq);
1160                 spin_unlock(&scrub->os_lock);
1161         }
1162
1163         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1164                 return 0;
1165
1166         if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1167                 *noslot = false;
1168                 return 0;
1169         }
1170
1171         if (it != NULL)
1172                 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1173                              &lwi);
1174
1175         if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1176                 *noslot = false;
1177         else
1178                 *noslot = true;
1179         return 0;
1180 }
1181
1182 static int osd_preload_exec(struct osd_thread_info *info,
1183                             struct osd_device *dev, struct osd_iit_param *param,
1184                             struct osd_idmap_cache *oic, bool *noslot, int rc)
1185 {
1186         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1187
1188         if (rc == 0) {
1189                 ooc->ooc_cached_items++;
1190                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1191                                         ~OSD_OTABLE_IT_CACHE_MASK;
1192         }
1193         return rc > 0 ? 0 : rc;
1194 }
1195
1196 #define SCRUB_IT_ALL    1
1197 #define SCRUB_IT_CRASH  2
1198
1199 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1200                            bool inconsistent)
1201 {
1202         struct osd_scrub     *scrub  = &dev->od_scrub;
1203         struct ptlrpc_thread *thread = &scrub->os_thread;
1204         struct scrub_file    *sf     = &scrub->os_file;
1205         int                   rc;
1206         ENTRY;
1207
1208         LASSERT(!(flags & SS_AUTO_PARTIAL));
1209
1210         down_write(&scrub->os_rwsem);
1211         scrub->os_in_join = 1;
1212         if (flags & SS_SET_FAILOUT)
1213                 sf->sf_param |= SP_FAILOUT;
1214         else if (flags & SS_CLEAR_FAILOUT)
1215                 sf->sf_param &= ~SP_FAILOUT;
1216
1217         if (flags & SS_SET_DRYRUN)
1218                 sf->sf_param |= SP_DRYRUN;
1219         else if (flags & SS_CLEAR_DRYRUN)
1220                 sf->sf_param &= ~SP_DRYRUN;
1221
1222         if (flags & SS_RESET) {
1223                 osd_scrub_file_reset(scrub,
1224                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1225                         inconsistent ? SF_INCONSISTENT : 0);
1226                 sf->sf_status = SS_SCANNING;
1227         }
1228
1229         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1230                 scrub->os_full_speed = 1;
1231         else
1232                 scrub->os_full_speed = 0;
1233
1234         if (flags & SS_AUTO_FULL) {
1235                 sf->sf_flags |= SF_AUTO;
1236                 scrub->os_full_speed = 1;
1237         }
1238
1239         scrub->os_new_checked = 0;
1240         if (sf->sf_pos_last_checkpoint != 0)
1241                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1242         else
1243                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1244
1245         scrub->os_pos_current = sf->sf_pos_latest_start;
1246         sf->sf_time_latest_start = cfs_time_current_sec();
1247         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1248         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1249         rc = osd_scrub_file_store(scrub);
1250         if (rc != 0)
1251                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
1252                        "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1253
1254         spin_lock(&scrub->os_lock);
1255         scrub->os_waiting = 0;
1256         scrub->os_paused = 0;
1257         scrub->os_partial_scan = 0;
1258         scrub->os_in_join = 0;
1259         scrub->os_full_scrub = 0;
1260         spin_unlock(&scrub->os_lock);
1261         wake_up_all(&thread->t_ctl_waitq);
1262         up_write(&scrub->os_rwsem);
1263
1264         EXIT;
1265 }
1266
1267 static int osd_inode_iteration(struct osd_thread_info *info,
1268                                struct osd_device *dev, __u32 max, bool preload)
1269 {
1270         struct osd_scrub     *scrub  = &dev->od_scrub;
1271         struct ptlrpc_thread *thread = &scrub->os_thread;
1272         struct scrub_file    *sf     = &scrub->os_file;
1273         osd_iit_next_policy   next;
1274         osd_iit_exec_policy   exec;
1275         __u32                *pos;
1276         __u32                *count;
1277         struct osd_iit_param  param  = { NULL };
1278         struct l_wait_info    lwi    = { 0 };
1279         __u32                 limit;
1280         int                   rc;
1281         bool                  noslot = true;
1282         ENTRY;
1283
1284         param.sb = osd_sb(dev);
1285         if (preload)
1286                 goto full;
1287
1288         while (scrub->os_partial_scan && !scrub->os_in_join) {
1289                 struct osd_idmap_cache *oic = NULL;
1290
1291                 rc = osd_scrub_next(info, dev, &param, &oic, noslot);
1292                 switch (rc) {
1293                 case SCRUB_NEXT_EXIT:
1294                         RETURN(0);
1295                 case SCRUB_NEXT_CRASH:
1296                         RETURN(SCRUB_IT_CRASH);
1297                 case SCRUB_NEXT_FATAL:
1298                         RETURN(-EINVAL);
1299                 case SCRUB_NEXT_WAIT: {
1300                         struct kstatfs *ksfs = &info->oti_ksfs;
1301                         __u64 saved_flags;
1302
1303                         if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1304                             unlikely(sf->sf_items_updated_prior == 0))
1305                                 goto wait;
1306
1307                         if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1308                             scrub->os_full_scrub) {
1309                                 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1310                                                true);
1311                                 goto full;
1312                         }
1313
1314                         rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1315                         if (rc == 0) {
1316                                 __u64 used = ksfs->f_files - ksfs->f_ffree;
1317
1318                                 do_div(used, sf->sf_items_updated_prior);
1319                                 /* If we hit too much inconsistent OI
1320                                  * mappings during the partial scan,
1321                                  * then scan the device completely. */
1322                                 if (used < dev->od_full_scrub_ratio) {
1323                                         osd_scrub_join(dev,
1324                                                 SS_AUTO_FULL | SS_RESET, true);
1325                                         goto full;
1326                                 }
1327                         }
1328
1329 wait:
1330                         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1331                             cfs_fail_val > 0)
1332                                 continue;
1333
1334                         saved_flags = sf->sf_flags;
1335                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1336                                           SF_UPGRADE | SF_AUTO);
1337                         sf->sf_status = SS_COMPLETED;
1338                         l_wait_event(thread->t_ctl_waitq,
1339                                      !thread_is_running(thread) ||
1340                                      !scrub->os_partial_scan ||
1341                                      scrub->os_in_join ||
1342                                      !list_empty(&scrub->os_inconsistent_items),
1343                                      &lwi);
1344                         sf->sf_flags = saved_flags;
1345                         sf->sf_status = SS_SCANNING;
1346
1347                         if (unlikely(!thread_is_running(thread)))
1348                                 RETURN(0);
1349
1350                         if (!scrub->os_partial_scan || scrub->os_in_join)
1351                                 goto full;
1352
1353                         continue;
1354                 }
1355                 default:
1356                         LASSERTF(rc == 0, "rc = %d\n", rc);
1357
1358                         osd_scrub_exec(info, dev, &param, oic, &noslot, rc);
1359                         break;
1360                 }
1361         }
1362
1363 full:
1364         if (!preload) {
1365                 l_wait_event(thread->t_ctl_waitq,
1366                              !thread_is_running(thread) || !scrub->os_in_join,
1367                              &lwi);
1368
1369                 if (unlikely(!thread_is_running(thread)))
1370                         RETURN(0);
1371         }
1372
1373         noslot = false;
1374         if (!preload) {
1375                 next = osd_scrub_next;
1376                 exec = osd_scrub_exec;
1377                 pos = &scrub->os_pos_current;
1378                 count = &scrub->os_new_checked;
1379         } else {
1380                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1381
1382                 next = osd_preload_next;
1383                 exec = osd_preload_exec;
1384                 pos = &ooc->ooc_pos_preload;
1385                 count = &ooc->ooc_cached_items;
1386         }
1387
1388         param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1389         param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1390         param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1391         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1392
1393         while (*pos <= limit && *count < max) {
1394                 struct ldiskfs_group_desc *desc;
1395                 bool next_group = false;
1396
1397                 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1398                 if (!desc)
1399                         RETURN(-EIO);
1400
1401                 ldiskfs_lock_group(param.sb, param.bg);
1402                 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1403                         ldiskfs_unlock_group(param.sb, param.bg);
1404                         next_group = true;
1405                         goto next_group;
1406                 }
1407                 ldiskfs_unlock_group(param.sb, param.bg);
1408
1409                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1410                 if (!param.bitmap) {
1411                         CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1412                                "scrub will stop, urgent mode\n",
1413                                osd_scrub2name(scrub), (__u32)param.bg);
1414                         RETURN(-EIO);
1415                 }
1416
1417                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1418                        *count < max) {
1419                         struct osd_idmap_cache *oic = NULL;
1420
1421                         if (param.offset +
1422                                 ldiskfs_itable_unused_count(param.sb, desc) >
1423                             LDISKFS_INODES_PER_GROUP(param.sb)) {
1424                                 next_group = true;
1425                                 goto next_group;
1426                         }
1427
1428                         rc = next(info, dev, &param, &oic, noslot);
1429                         switch (rc) {
1430                         case SCRUB_NEXT_BREAK:
1431                                 next_group = true;
1432                                 goto next_group;
1433                         case SCRUB_NEXT_EXIT:
1434                                 brelse(param.bitmap);
1435                                 RETURN(0);
1436                         case SCRUB_NEXT_CRASH:
1437                                 brelse(param.bitmap);
1438                                 RETURN(SCRUB_IT_CRASH);
1439                         case SCRUB_NEXT_FATAL:
1440                                 brelse(param.bitmap);
1441                                 RETURN(-EINVAL);
1442                         }
1443
1444                         rc = exec(info, dev, &param, oic, &noslot, rc);
1445                         if (rc != 0) {
1446                                 brelse(param.bitmap);
1447                                 RETURN(rc);
1448                         }
1449                 }
1450
1451 next_group:
1452                 if (param.bitmap) {
1453                         brelse(param.bitmap);
1454                         param.bitmap = NULL;
1455                 }
1456
1457                 if (next_group) {
1458                         param.bg++;
1459                         param.offset = 0;
1460                         param.gbase = 1 +
1461                                 param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1462                         *pos = param.gbase;
1463                 }
1464         }
1465
1466         if (*pos > limit)
1467                 RETURN(SCRUB_IT_ALL);
1468         RETURN(0);
1469 }
1470
1471 static int osd_otable_it_preload(const struct lu_env *env,
1472                                  struct osd_otable_it *it)
1473 {
1474         struct osd_device       *dev   = it->ooi_dev;
1475         struct osd_scrub        *scrub = &dev->od_scrub;
1476         struct osd_otable_cache *ooc   = &it->ooi_cache;
1477         int                      rc;
1478         ENTRY;
1479
1480         rc = osd_inode_iteration(osd_oti_get(env), dev,
1481                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1482         if (rc == SCRUB_IT_ALL)
1483                 it->ooi_all_cached = 1;
1484
1485         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1486                 spin_lock(&scrub->os_lock);
1487                 scrub->os_waiting = 0;
1488                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1489                 spin_unlock(&scrub->os_lock);
1490         }
1491
1492         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1493 }
1494
1495 static int osd_scrub_main(void *args)
1496 {
1497         struct lu_env         env;
1498         struct osd_device    *dev    = (struct osd_device *)args;
1499         struct osd_scrub     *scrub  = &dev->od_scrub;
1500         struct ptlrpc_thread *thread = &scrub->os_thread;
1501         int                   rc;
1502         ENTRY;
1503
1504         rc = lu_env_init(&env, LCT_LOCAL);
1505         if (rc != 0) {
1506                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1507                        osd_scrub2name(scrub), rc);
1508                 GOTO(noenv, rc);
1509         }
1510
1511         rc = osd_scrub_prep(dev);
1512         if (rc != 0) {
1513                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1514                        osd_scrub2name(scrub), rc);
1515                 GOTO(out, rc);
1516         }
1517
1518         if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1519                 struct l_wait_info lwi = { 0 };
1520                 struct osd_otable_it *it = dev->od_otable_it;
1521                 struct osd_otable_cache *ooc = &it->ooi_cache;
1522
1523                 l_wait_event(thread->t_ctl_waitq,
1524                              it->ooi_user_ready || !thread_is_running(thread),
1525                              &lwi);
1526                 if (unlikely(!thread_is_running(thread)))
1527                         GOTO(post, rc = 0);
1528
1529                 scrub->os_pos_current = ooc->ooc_pos_preload;
1530         }
1531
1532         CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1533                osd_scrub2name(scrub), scrub->os_start_flags,
1534                scrub->os_pos_current);
1535
1536         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1537         if (unlikely(rc == SCRUB_IT_CRASH))
1538                 GOTO(out, rc = -EINVAL);
1539         GOTO(post, rc);
1540
1541 post:
1542         rc = osd_scrub_post(scrub, rc);
1543         CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1544                osd_scrub2name(scrub), scrub->os_pos_current, rc);
1545
1546 out:
1547         while (!list_empty(&scrub->os_inconsistent_items)) {
1548                 struct osd_inconsistent_item *oii;
1549
1550                 oii = list_entry(scrub->os_inconsistent_items.next,
1551                                      struct osd_inconsistent_item, oii_list);
1552                 list_del_init(&oii->oii_list);
1553                 OBD_FREE_PTR(oii);
1554         }
1555         lu_env_fini(&env);
1556
1557 noenv:
1558         spin_lock(&scrub->os_lock);
1559         thread_set_flags(thread, SVC_STOPPED);
1560         wake_up_all(&thread->t_ctl_waitq);
1561         spin_unlock(&scrub->os_lock);
1562         return rc;
1563 }
1564
1565 /* initial OI scrub */
1566
1567 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1568                          struct dentry *, filldir_t filldir);
1569
1570 #ifdef HAVE_FILLDIR_USE_CTX
1571 static int osd_ios_varfid_fill(struct dir_context *buf, const char *name,
1572                                int namelen, loff_t offset, __u64 ino,
1573                                unsigned d_type);
1574 static int osd_ios_lf_fill(struct dir_context *buf, const char *name,
1575                            int namelen, loff_t offset, __u64 ino,
1576                            unsigned d_type);
1577 static int osd_ios_dl_fill(struct dir_context *buf, const char *name,
1578                            int namelen, loff_t offset, __u64 ino,
1579                            unsigned d_type);
1580 static int osd_ios_uld_fill(struct dir_context *buf, const char *name,
1581                             int namelen, loff_t offset, __u64 ino,
1582                             unsigned d_type);
1583 #else
1584 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1585                                loff_t offset, __u64 ino, unsigned d_type);
1586 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1587                            loff_t offset, __u64 ino, unsigned d_type);
1588 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1589                            loff_t offset, __u64 ino, unsigned d_type);
1590 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1591                             loff_t offset, __u64 ino, unsigned d_type);
1592 #endif
1593
1594 static int
1595 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1596                      struct dentry *dentry, filldir_t filldir);
1597 static int
1598 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1599                   struct dentry *dentry, filldir_t filldir);
1600
1601 static int
1602 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1603                      struct dentry *dentry, filldir_t filldir);
1604
1605 enum osd_lf_flags {
1606         OLF_SCAN_SUBITEMS       = 0x0001,
1607         OLF_HIDE_FID            = 0x0002,
1608         OLF_SHOW_NAME           = 0x0004,
1609         OLF_NO_OI               = 0x0008,
1610         OLF_IDX_IN_FID          = 0x0010,
1611 };
1612
1613 struct osd_lf_map {
1614         char            *olm_name;
1615         struct lu_fid    olm_fid;
1616         __u16            olm_flags;
1617         __u16            olm_namelen;
1618         scandir_t        olm_scandir;
1619         filldir_t        olm_filldir;
1620 };
1621
1622 /* Add the new introduced local files in the list in the future. */
1623 static const struct osd_lf_map osd_lf_maps[] = {
1624         /* CATALOGS */
1625         {
1626                 .olm_name       = CATLIST,
1627                 .olm_fid        = {
1628                         .f_seq  = FID_SEQ_LOCAL_FILE,
1629                         .f_oid  = LLOG_CATALOGS_OID,
1630                 },
1631                 .olm_flags      = OLF_SHOW_NAME,
1632                 .olm_namelen    = sizeof(CATLIST) - 1,
1633         },
1634
1635         /* CONFIGS */
1636         {
1637                 .olm_name       = MOUNT_CONFIGS_DIR,
1638                 .olm_fid        = {
1639                         .f_seq  = FID_SEQ_LOCAL_FILE,
1640                         .f_oid  = MGS_CONFIGS_OID,
1641                 },
1642                 .olm_flags      = OLF_SCAN_SUBITEMS,
1643                 .olm_namelen    = sizeof(MOUNT_CONFIGS_DIR) - 1,
1644                 .olm_scandir    = osd_ios_general_scan,
1645                 .olm_filldir    = osd_ios_varfid_fill,
1646         },
1647
1648         /* NIDTBL_VERSIONS */
1649         {
1650                 .olm_name       = MGS_NIDTBL_DIR,
1651                 .olm_flags      = OLF_SCAN_SUBITEMS,
1652                 .olm_namelen    = sizeof(MGS_NIDTBL_DIR) - 1,
1653                 .olm_scandir    = osd_ios_general_scan,
1654                 .olm_filldir    = osd_ios_varfid_fill,
1655         },
1656
1657         /* PENDING */
1658         {
1659                 .olm_name       = "PENDING",
1660                 .olm_namelen    = sizeof("PENDING") - 1,
1661         },
1662
1663         /* ROOT */
1664         {
1665                 .olm_name       = "ROOT",
1666                 .olm_fid        = {
1667                         .f_seq  = FID_SEQ_ROOT,
1668                         .f_oid  = FID_OID_ROOT,
1669                 },
1670                 .olm_flags      = OLF_SCAN_SUBITEMS | OLF_HIDE_FID,
1671                 .olm_namelen    = sizeof("ROOT") - 1,
1672                 .olm_scandir    = osd_ios_ROOT_scan,
1673         },
1674
1675         /* changelog_catalog */
1676         {
1677                 .olm_name       = CHANGELOG_CATALOG,
1678                 .olm_namelen    = sizeof(CHANGELOG_CATALOG) - 1,
1679         },
1680
1681         /* changelog_users */
1682         {
1683                 .olm_name       = CHANGELOG_USERS,
1684                 .olm_namelen    = sizeof(CHANGELOG_USERS) - 1,
1685         },
1686
1687         /* fld */
1688         {
1689                 .olm_name       = "fld",
1690                 .olm_fid        = {
1691                         .f_seq  = FID_SEQ_LOCAL_FILE,
1692                         .f_oid  = FLD_INDEX_OID,
1693                 },
1694                 .olm_flags      = OLF_SHOW_NAME,
1695                 .olm_namelen    = sizeof("fld") - 1,
1696         },
1697
1698         /* last_rcvd */
1699         {
1700                 .olm_name       = LAST_RCVD,
1701                 .olm_fid        = {
1702                         .f_seq  = FID_SEQ_LOCAL_FILE,
1703                         .f_oid  = LAST_RECV_OID,
1704                 },
1705                 .olm_flags      = OLF_SHOW_NAME,
1706                 .olm_namelen    = sizeof(LAST_RCVD) - 1,
1707         },
1708
1709         /* reply_data */
1710         {
1711                 .olm_name       = REPLY_DATA,
1712                 .olm_fid        = {
1713                         .f_seq  = FID_SEQ_LOCAL_FILE,
1714                         .f_oid  = REPLY_DATA_OID,
1715                 },
1716                 .olm_flags      = OLF_SHOW_NAME,
1717                 .olm_namelen    = sizeof(REPLY_DATA) - 1,
1718         },
1719
1720         /* lov_objid */
1721         {
1722                 .olm_name       = LOV_OBJID,
1723                 .olm_fid        = {
1724                         .f_seq  = FID_SEQ_LOCAL_FILE,
1725                         .f_oid  = MDD_LOV_OBJ_OID,
1726                 },
1727                 .olm_flags      = OLF_SHOW_NAME,
1728                 .olm_namelen    = sizeof(LOV_OBJID) - 1,
1729         },
1730
1731         /* lov_objseq */
1732         {
1733                 .olm_name       = LOV_OBJSEQ,
1734                 .olm_fid        = {
1735                         .f_seq  = FID_SEQ_LOCAL_FILE,
1736                         .f_oid  = MDD_LOV_OBJ_OSEQ,
1737                 },
1738                 .olm_flags      = OLF_SHOW_NAME,
1739                 .olm_namelen    = sizeof(LOV_OBJSEQ) - 1,
1740         },
1741
1742         /* quota_master */
1743         {
1744                 .olm_name       = QMT_DIR,
1745                 .olm_flags      = OLF_SCAN_SUBITEMS,
1746                 .olm_namelen    = sizeof(QMT_DIR) - 1,
1747                 .olm_scandir    = osd_ios_general_scan,
1748                 .olm_filldir    = osd_ios_varfid_fill,
1749         },
1750
1751         /* quota_slave */
1752         {
1753                 .olm_name       = QSD_DIR,
1754                 .olm_flags      = OLF_SCAN_SUBITEMS,
1755                 .olm_namelen    = sizeof(QSD_DIR) - 1,
1756                 .olm_scandir    = osd_ios_general_scan,
1757                 .olm_filldir    = osd_ios_varfid_fill,
1758         },
1759
1760         /* seq_ctl */
1761         {
1762                 .olm_name       = "seq_ctl",
1763                 .olm_fid        = {
1764                         .f_seq  = FID_SEQ_LOCAL_FILE,
1765                         .f_oid  = FID_SEQ_CTL_OID,
1766                 },
1767                 .olm_flags      = OLF_SHOW_NAME,
1768                 .olm_namelen    = sizeof("seq_ctl") - 1,
1769         },
1770
1771         /* seq_srv */
1772         {
1773                 .olm_name       = "seq_srv",
1774                 .olm_fid        = {
1775                         .f_seq  = FID_SEQ_LOCAL_FILE,
1776                         .f_oid  = FID_SEQ_SRV_OID,
1777                 },
1778                 .olm_flags      = OLF_SHOW_NAME,
1779                 .olm_namelen    = sizeof("seq_srv") - 1,
1780         },
1781
1782         /* health_check */
1783         {
1784                 .olm_name       = HEALTH_CHECK,
1785                 .olm_fid        = {
1786                         .f_seq  = FID_SEQ_LOCAL_FILE,
1787                         .f_oid  = OFD_HEALTH_CHECK_OID,
1788                 },
1789                 .olm_flags      = OLF_SHOW_NAME,
1790                 .olm_namelen    = sizeof(HEALTH_CHECK) - 1,
1791         },
1792
1793         /* LFSCK */
1794         {
1795                 .olm_name       = LFSCK_DIR,
1796                 .olm_namelen    = sizeof(LFSCK_DIR) - 1,
1797                 .olm_scandir    = osd_ios_general_scan,
1798                 .olm_filldir    = osd_ios_varfid_fill,
1799         },
1800
1801         /* lfsck_bookmark */
1802         {
1803                 .olm_name       = LFSCK_BOOKMARK,
1804                 .olm_namelen    = sizeof(LFSCK_BOOKMARK) - 1,
1805         },
1806
1807         /* lfsck_layout */
1808         {
1809                 .olm_name       = LFSCK_LAYOUT,
1810                 .olm_namelen    = sizeof(LFSCK_LAYOUT) - 1,
1811         },
1812
1813         /* lfsck_namespace */
1814         {
1815                 .olm_name       = LFSCK_NAMESPACE,
1816                 .olm_namelen    = sizeof(LFSCK_NAMESPACE) - 1,
1817         },
1818
1819         /* OBJECTS, upgrade from old device */
1820         {
1821                 .olm_name       = OBJECTS,
1822                 .olm_flags      = OLF_SCAN_SUBITEMS,
1823                 .olm_namelen    = sizeof(OBJECTS) - 1,
1824                 .olm_scandir    = osd_ios_OBJECTS_scan,
1825         },
1826
1827         /* lquota_v2.user, upgrade from old device */
1828         {
1829                 .olm_name       = "lquota_v2.user",
1830                 .olm_namelen    = sizeof("lquota_v2.user") - 1,
1831         },
1832
1833         /* lquota_v2.group, upgrade from old device */
1834         {
1835                 .olm_name       = "lquota_v2.group",
1836                 .olm_namelen    = sizeof("lquota_v2.group") - 1,
1837         },
1838
1839         /* LAST_GROUP, upgrade from old device */
1840         {
1841                 .olm_name       = "LAST_GROUP",
1842                 .olm_fid        = {
1843                         .f_seq  = FID_SEQ_LOCAL_FILE,
1844                         .f_oid  = OFD_LAST_GROUP_OID,
1845                 },
1846                 .olm_flags      = OLF_SHOW_NAME,
1847                 .olm_namelen    = sizeof("LAST_GROUP") - 1,
1848         },
1849
1850         /* committed batchid for cross-MDT operation */
1851         {
1852                 .olm_name       = "BATCHID",
1853                 .olm_fid        = {
1854                         .f_seq  = FID_SEQ_LOCAL_FILE,
1855                         .f_oid  = BATCHID_COMMITTED_OID,
1856                 },
1857                 .olm_flags      = OLF_SHOW_NAME,
1858                 .olm_namelen    = sizeof("BATCHID") - 1,
1859         },
1860
1861         /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1862          * and f_oid = index for their log files.  See lu_update_log{_dir}_fid()
1863          * for more details. */
1864
1865         /* update_log */
1866         {
1867                 .olm_name       = "update_log",
1868                 .olm_fid        = {
1869                         .f_seq  = FID_SEQ_UPDATE_LOG,
1870                 },
1871                 .olm_flags      = OLF_SHOW_NAME | OLF_IDX_IN_FID,
1872                 .olm_namelen    = sizeof("update_log") - 1,
1873         },
1874
1875         /* update_log_dir */
1876         {
1877                 .olm_name       = "update_log_dir",
1878                 .olm_fid        = {
1879                         .f_seq  = FID_SEQ_UPDATE_LOG_DIR,
1880                 },
1881                 .olm_flags      = OLF_SHOW_NAME | OLF_SCAN_SUBITEMS |
1882                                   OLF_IDX_IN_FID,
1883                 .olm_namelen    = sizeof("update_log_dir") - 1,
1884                 .olm_scandir    = osd_ios_general_scan,
1885                 .olm_filldir    = osd_ios_uld_fill,
1886         },
1887
1888         /* lost+found */
1889         {
1890                 .olm_name       = "lost+found",
1891                 .olm_fid        = {
1892                         .f_seq  = FID_SEQ_LOCAL_FILE,
1893                         .f_oid  = OSD_LPF_OID,
1894                 },
1895                 .olm_flags      = OLF_SCAN_SUBITEMS,
1896                 .olm_namelen    = sizeof("lost+found") - 1,
1897                 .olm_scandir    = osd_ios_general_scan,
1898                 .olm_filldir    = osd_ios_lf_fill,
1899         },
1900
1901         {
1902                 .olm_name       = NULL
1903         }
1904 };
1905
1906 /* Add the new introduced files under .lustre/ in the list in the future. */
1907 static const struct osd_lf_map osd_dl_maps[] = {
1908         /* .lustre/fid */
1909         {
1910                 .olm_name       = "fid",
1911                 .olm_fid        = {
1912                         .f_seq  = FID_SEQ_DOT_LUSTRE,
1913                         .f_oid  = FID_OID_DOT_LUSTRE_OBF,
1914                 },
1915                 .olm_namelen    = sizeof("fid") - 1,
1916         },
1917         /* .lustre/lost+found */
1918         {
1919                 .olm_name       = "lost+found",
1920                 .olm_fid        = {
1921                         .f_seq  = FID_SEQ_DOT_LUSTRE,
1922                         .f_oid  = FID_OID_DOT_LUSTRE_LPF,
1923                 },
1924                 .olm_namelen    = sizeof("lost+found") - 1,
1925         },
1926         {
1927                 .olm_name       = NULL
1928         }
1929 };
1930
1931 struct osd_ios_item {
1932         struct list_head oii_list;
1933         struct dentry   *oii_dentry;
1934         scandir_t        oii_scandir;
1935         filldir_t        oii_filldir;
1936 };
1937
1938 struct osd_ios_filldir_buf {
1939 #ifdef HAVE_DIR_CONTEXT
1940         /* please keep it as first member */
1941         struct dir_context       ctx;
1942 #endif
1943         struct osd_thread_info  *oifb_info;
1944         struct osd_device       *oifb_dev;
1945         struct dentry           *oifb_dentry;
1946 };
1947
1948 static inline struct dentry *
1949 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1950 {
1951         struct dentry *dentry;
1952
1953         dentry = ll_lookup_one_len(name, parent, namelen);
1954         if (IS_ERR(dentry)) {
1955                 int rc = PTR_ERR(dentry);
1956
1957                 if (rc != -ENOENT)
1958                         CERROR("Fail to find %.*s in %.*s (%lu/%u): rc = %d\n",
1959                                namelen, name, parent->d_name.len,
1960                                parent->d_name.name, parent->d_inode->i_ino,
1961                                parent->d_inode->i_generation, rc);
1962
1963                 return dentry;
1964         }
1965
1966         if (dentry->d_inode == NULL) {
1967                 dput(dentry);
1968                 return ERR_PTR(-ENOENT);
1969         }
1970
1971         return dentry;
1972 }
1973
1974 static int
1975 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1976                  scandir_t scandir, filldir_t filldir)
1977 {
1978         struct osd_ios_item *item;
1979         ENTRY;
1980
1981         OBD_ALLOC_PTR(item);
1982         if (item == NULL)
1983                 RETURN(-ENOMEM);
1984
1985         INIT_LIST_HEAD(&item->oii_list);
1986         item->oii_dentry = dget(dentry);
1987         item->oii_scandir = scandir;
1988         item->oii_filldir = filldir;
1989         list_add_tail(&item->oii_list, &dev->od_ios_list);
1990
1991         RETURN(0);
1992 }
1993
1994 /**
1995  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1996  *
1997  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1998  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1999  * reference the inode, or fixed if it is missing or references another inode.
2000  */
2001 static int
2002 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
2003                  struct inode *inode, const struct lu_fid *fid, int flags)
2004 {
2005         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
2006         struct osd_inode_id     *id     = &info->oti_id;
2007         struct osd_inode_id     *id2    = &info->oti_id2;
2008         struct osd_scrub        *scrub  = &dev->od_scrub;
2009         struct scrub_file       *sf     = &scrub->os_file;
2010         struct lu_fid            tfid;
2011         int                      rc;
2012         ENTRY;
2013
2014         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
2015         if (rc != 0 && rc != -ENODATA) {
2016                 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
2017                        "rc = %d\n", osd_name(dev), rc);
2018
2019                 RETURN(rc);
2020         }
2021
2022         osd_id_gen(id, inode->i_ino, inode->i_generation);
2023         if (rc == -ENODATA) {
2024                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
2025                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
2026                 } else {
2027                         tfid = *fid;
2028                         if (flags & OLF_IDX_IN_FID) {
2029                                 LASSERT(dev->od_index >= 0);
2030
2031                                 tfid.f_oid = dev->od_index;
2032                         }
2033                 }
2034                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
2035                 if (rc != 0) {
2036                         CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
2037                               "scrub: rc = %d\n", osd_name(dev), rc);
2038
2039                         RETURN(rc);
2040                 }
2041         } else {
2042                 if (lma->lma_compat & LMAC_NOT_IN_OI)
2043                         RETURN(0);
2044
2045                 tfid = lma->lma_self_fid;
2046         }
2047
2048         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
2049         if (rc != 0) {
2050                 if (rc != -ENOENT)
2051                         RETURN(rc);
2052
2053                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
2054                                                DTO_INDEX_INSERT, true, 0, NULL);
2055                 if (rc > 0)
2056                         rc = 0;
2057
2058                 RETURN(rc);
2059         }
2060
2061         if (osd_id_eq_strict(id, id2))
2062                 RETURN(0);
2063
2064         if (!(sf->sf_flags & SF_INCONSISTENT)) {
2065                 osd_scrub_file_reset(scrub,
2066                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2067                                      SF_INCONSISTENT);
2068                 rc = osd_scrub_file_store(scrub);
2069                 if (rc != 0)
2070                         RETURN(rc);
2071         }
2072
2073         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
2074                                        DTO_INDEX_UPDATE, true, 0, NULL);
2075         if (rc > 0)
2076                 rc = 0;
2077
2078         RETURN(rc);
2079 }
2080
2081 /**
2082  * It scans the /lost+found, and for the OST-object (with filter_fid
2083  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
2084  */
2085 #ifdef HAVE_FILLDIR_USE_CTX
2086 static int osd_ios_lf_fill(struct dir_context *buf,
2087 #else
2088 static int osd_ios_lf_fill(void *buf,
2089 #endif
2090                            const char *name, int namelen,
2091                            loff_t offset, __u64 ino, unsigned d_type)
2092 {
2093         struct osd_ios_filldir_buf *fill_buf =
2094                 (struct osd_ios_filldir_buf *)buf;
2095         struct osd_thread_info     *info     = fill_buf->oifb_info;
2096         struct osd_device          *dev      = fill_buf->oifb_dev;
2097         struct lu_fid              *fid      = &info->oti_fid;
2098         struct osd_scrub           *scrub    = &dev->od_scrub;
2099         struct dentry              *parent   = fill_buf->oifb_dentry;
2100         struct dentry              *child;
2101         struct inode               *dir      = parent->d_inode;
2102         struct inode               *inode;
2103         int                         rc;
2104         ENTRY;
2105
2106         /* skip any '.' started names */
2107         if (name[0] == '.')
2108                 RETURN(0);
2109
2110         scrub->os_lf_scanned++;
2111         child = osd_ios_lookup_one_len(name, parent, namelen);
2112         if (IS_ERR(child)) {
2113                 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
2114                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
2115                 RETURN(0);
2116         }
2117
2118         inode = child->d_inode;
2119         if (S_ISDIR(inode->i_mode)) {
2120                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2121                                       osd_ios_lf_fill);
2122                 if (rc != 0)
2123                         CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
2124                               "rc = %d\n", osd_name(dev), namelen, name, rc);
2125                 GOTO(put, rc);
2126         }
2127
2128         if (!S_ISREG(inode->i_mode))
2129                 GOTO(put, rc = 0);
2130
2131         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
2132         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
2133                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
2134                 if (rc == 0) {
2135                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
2136                                "/lost+found.\n", namelen, name, PFID(fid));
2137                         scrub->os_lf_repaired++;
2138                 } else {
2139                         CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
2140                                DFID": rc = %d\n",
2141                                osd_name(dev), namelen, name, PFID(fid), rc);
2142                 }
2143         }
2144
2145         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
2146          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
2147          *      can process them in furtuer. */
2148
2149         GOTO(put, rc);
2150
2151 put:
2152         if (rc < 0)
2153                 scrub->os_lf_failed++;
2154         dput(child);
2155         /* skip the failure to make the scanning to continue. */
2156         return 0;
2157 }
2158
2159 #ifdef HAVE_FILLDIR_USE_CTX
2160 static int osd_ios_varfid_fill(struct dir_context *buf,
2161 #else
2162 static int osd_ios_varfid_fill(void *buf,
2163 #endif
2164                                const char *name, int namelen,
2165                                loff_t offset, __u64 ino, unsigned d_type)
2166 {
2167         struct osd_ios_filldir_buf *fill_buf =
2168                 (struct osd_ios_filldir_buf *)buf;
2169         struct osd_device          *dev      = fill_buf->oifb_dev;
2170         struct dentry              *child;
2171         int                         rc;
2172         ENTRY;
2173
2174         /* skip any '.' started names */
2175         if (name[0] == '.')
2176                 RETURN(0);
2177
2178         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2179         if (IS_ERR(child))
2180                 RETURN(PTR_ERR(child));
2181
2182         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2183                               NULL, 0);
2184         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
2185                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2186                                       osd_ios_varfid_fill);
2187         dput(child);
2188
2189         RETURN(rc);
2190 }
2191
2192 #ifdef HAVE_FILLDIR_USE_CTX
2193 static int osd_ios_dl_fill(struct dir_context *buf,
2194 #else
2195 static int osd_ios_dl_fill(void *buf,
2196 #endif
2197                            const char *name, int namelen,
2198                            loff_t offset, __u64 ino, unsigned d_type)
2199 {
2200         struct osd_ios_filldir_buf *fill_buf =
2201                 (struct osd_ios_filldir_buf *)buf;
2202         struct osd_device          *dev      = fill_buf->oifb_dev;
2203         const struct osd_lf_map    *map;
2204         struct dentry              *child;
2205         int                         rc       = 0;
2206         ENTRY;
2207
2208         /* skip any '.' started names */
2209         if (name[0] == '.')
2210                 RETURN(0);
2211
2212         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
2213                 if (map->olm_namelen != namelen)
2214                         continue;
2215
2216                 if (strncmp(map->olm_name, name, namelen) == 0)
2217                         break;
2218         }
2219
2220         if (map->olm_name == NULL)
2221                 RETURN(0);
2222
2223         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2224         if (IS_ERR(child))
2225                 RETURN(PTR_ERR(child));
2226
2227         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2228                               &map->olm_fid, map->olm_flags);
2229         dput(child);
2230
2231         RETURN(rc);
2232 }
2233
2234 #ifdef HAVE_FILLDIR_USE_CTX
2235 static int osd_ios_uld_fill(struct dir_context *buf,
2236 #else
2237 static int osd_ios_uld_fill(void *buf,
2238 #endif
2239                             const char *name, int namelen,
2240                             loff_t offset, __u64 ino, unsigned d_type)
2241 {
2242         struct osd_ios_filldir_buf *fill_buf =
2243                 (struct osd_ios_filldir_buf *)buf;
2244         struct dentry              *child;
2245         struct lu_fid               tfid;
2246         int                         rc       = 0;
2247         ENTRY;
2248
2249         /* skip any non-DFID format name */
2250         if (name[0] != '[')
2251                 RETURN(0);
2252
2253         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2254         if (IS_ERR(child))
2255                 RETURN(PTR_ERR(child));
2256
2257         /* skip the start '[' */
2258         sscanf(&name[1], SFID, RFID(&tfid));
2259         if (fid_is_sane(&tfid))
2260                 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
2261                                       child->d_inode, &tfid, 0);
2262         else
2263                 rc = -EIO;
2264         dput(child);
2265
2266         RETURN(rc);
2267 }
2268
2269 #ifdef HAVE_FILLDIR_USE_CTX
2270 static int osd_ios_root_fill(struct dir_context *buf,
2271 #else
2272 static int osd_ios_root_fill(void *buf,
2273 #endif
2274                              const char *name, int namelen,
2275                              loff_t offset, __u64 ino, unsigned d_type)
2276 {
2277         struct osd_ios_filldir_buf *fill_buf =
2278                 (struct osd_ios_filldir_buf *)buf;
2279         struct osd_device          *dev      = fill_buf->oifb_dev;
2280         const struct osd_lf_map    *map;
2281         struct dentry              *child;
2282         int                         rc       = 0;
2283         ENTRY;
2284
2285         /* skip any '.' started names */
2286         if (name[0] == '.')
2287                 RETURN(0);
2288
2289         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2290                 if (map->olm_namelen != namelen)
2291                         continue;
2292
2293                 if (strncmp(map->olm_name, name, namelen) == 0)
2294                         break;
2295         }
2296
2297         if (map->olm_name == NULL)
2298                 RETURN(0);
2299
2300         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2301         if (IS_ERR(child))
2302                 RETURN(PTR_ERR(child));
2303
2304         if (!(map->olm_flags & OLF_NO_OI))
2305                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2306                                       &map->olm_fid, map->olm_flags);
2307         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2308                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2309                                       map->olm_filldir);
2310         dput(child);
2311
2312         RETURN(rc);
2313 }
2314
2315 static int
2316 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2317                      struct dentry *dentry, filldir_t filldir)
2318 {
2319         struct osd_ios_filldir_buf    buf   = {
2320 #ifdef HAVE_DIR_CONTEXT
2321                                                 .ctx.actor = filldir,
2322 #endif
2323                                                 .oifb_info = info,
2324                                                 .oifb_dev = dev,
2325                                                 .oifb_dentry = dentry };
2326         struct file                  *filp  = &info->oti_file;
2327         struct inode                 *inode = dentry->d_inode;
2328         const struct file_operations *fops  = inode->i_fop;
2329         int                           rc;
2330         ENTRY;
2331
2332         LASSERT(filldir != NULL);
2333
2334         filp->f_pos = 0;
2335         filp->f_path.dentry = dentry;
2336         filp->f_mode = FMODE_64BITHASH;
2337         filp->f_mapping = inode->i_mapping;
2338         filp->f_op = fops;
2339         filp->private_data = NULL;
2340         set_file_inode(filp, inode);
2341
2342 #ifdef HAVE_DIR_CONTEXT
2343         buf.ctx.pos = filp->f_pos;
2344         rc = fops->iterate(filp, &buf.ctx);
2345         filp->f_pos = buf.ctx.pos;
2346 #else
2347         rc = fops->readdir(filp, &buf, filldir);
2348 #endif
2349         fops->release(inode, filp);
2350
2351         RETURN(rc);
2352 }
2353
2354 static int
2355 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2356                   struct dentry *dentry, filldir_t filldir)
2357 {
2358         struct osd_scrub  *scrub  = &dev->od_scrub;
2359         struct scrub_file *sf     = &scrub->os_file;
2360         struct dentry     *child;
2361         int                rc;
2362         ENTRY;
2363
2364         /* It is existing MDT0 device. We only allow the case of object without
2365          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2366          * can generate IGIF mode FID for the object and related OI mapping. If
2367          * it is on other MDTs, then becuase file-level backup/restore, related
2368          * OI mapping may be invalid already, we do not know which is the right
2369          * FID for the object. We only allow IGIF objects to reside on the MDT0.
2370          *
2371          * XXX: For the case of object on non-MDT0 device with neither LMA nor
2372          *      "fid" xattr, then something crashed. We cannot re-generate the
2373          *      FID directly, instead, the OI scrub will scan the OI structure
2374          *      and try to re-generate the LMA from the OI mapping. But if the
2375          *      OI mapping crashed or lost also, then we have to give up under
2376          *      double failure cases. */
2377         scrub->os_convert_igif = 1;
2378         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2379                                        strlen(dot_lustre_name));
2380         if (IS_ERR(child)) {
2381                 rc = PTR_ERR(child);
2382                 if (rc == -ENOENT) {
2383                         /* It is 1.8 MDT device. */
2384                         if (!(sf->sf_flags & SF_UPGRADE)) {
2385                                 osd_scrub_file_reset(scrub,
2386                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2387                                         SF_UPGRADE);
2388                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2389                                 rc = osd_scrub_file_store(scrub);
2390                         } else {
2391                                 rc = 0;
2392                         }
2393                 }
2394         } else {
2395                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2396                  * so the client will get IGIF for the ".lustre" object when
2397                  * the MDT restart.
2398                  *
2399                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2400                  * it does not know whether there are some old clients cached
2401                  * the ".lustre" IGIF during the upgrading. Two choices:
2402                  *
2403                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2404                  *    It will allow the old connected clients to access the
2405                  *    ".lustre" with cached IGIF. But it will cause others
2406                  *    on the MDT failed to check "fid_is_dot_lustre()".
2407                  *
2408                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2409                  *    for ".lustre" in spite of whether there are some clients
2410                  *    cached the ".lustre" IGIF or not. It enables the check
2411                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
2412                  *    that the old connected clients cannot access the ".lustre"
2413                  *    with the cached IGIF.
2414                  *
2415                  * Usually, it is rare case for the old connected clients
2416                  * to access the ".lustre" with cached IGIF. So we prefer
2417                  * to the solution 2). */
2418                 rc = osd_ios_scan_one(info, dev, child->d_inode,
2419                                       &LU_DOT_LUSTRE_FID, 0);
2420                 if (rc == 0)
2421                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2422                                               osd_ios_dl_fill);
2423                 dput(child);
2424         }
2425
2426         RETURN(rc);
2427 }
2428
2429 static int
2430 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2431                      struct dentry *dentry, filldir_t filldir)
2432 {
2433         struct osd_scrub  *scrub  = &dev->od_scrub;
2434         struct scrub_file *sf     = &scrub->os_file;
2435         struct dentry     *child;
2436         int                rc;
2437         ENTRY;
2438
2439         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2440                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2441                 rc = osd_scrub_file_store(scrub);
2442                 if (rc != 0)
2443                         RETURN(rc);
2444         }
2445
2446         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2447         if (!IS_ERR(child)) {
2448                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2449                 dput(child);
2450         } else {
2451                 rc = PTR_ERR(child);
2452         }
2453
2454         if (rc != 0 && rc != -ENOENT)
2455                 RETURN(rc);
2456
2457         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2458         if (!IS_ERR(child)) {
2459                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2460                 dput(child);
2461         } else {
2462                 rc = PTR_ERR(child);
2463         }
2464
2465         if (rc == -ENOENT)
2466                 rc = 0;
2467
2468         RETURN(rc);
2469 }
2470
2471 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2472                                 struct osd_device *dev)
2473 {
2474         struct osd_ios_item     *item    = NULL;
2475         scandir_t                scandir = osd_ios_general_scan;
2476         filldir_t                filldir = osd_ios_root_fill;
2477         struct dentry           *dentry  = osd_sb(dev)->s_root;
2478         const struct osd_lf_map *map     = osd_lf_maps;
2479         int                      rc;
2480         ENTRY;
2481
2482         /* Lookup IGIF in OI by force for initial OI scrub. */
2483         dev->od_igif_inoi = 1;
2484
2485         while (1) {
2486                 rc = scandir(info, dev, dentry, filldir);
2487                 if (item != NULL) {
2488                         dput(item->oii_dentry);
2489                         OBD_FREE_PTR(item);
2490                 }
2491
2492                 if (rc != 0)
2493                         break;
2494
2495                 if (list_empty(&dev->od_ios_list))
2496                         break;
2497
2498                 item = list_entry(dev->od_ios_list.next,
2499                                   struct osd_ios_item, oii_list);
2500                 list_del_init(&item->oii_list);
2501
2502                 LASSERT(item->oii_scandir != NULL);
2503                 scandir = item->oii_scandir;
2504                 filldir = item->oii_filldir;
2505                 dentry = item->oii_dentry;
2506         }
2507
2508         while (!list_empty(&dev->od_ios_list)) {
2509                 item = list_entry(dev->od_ios_list.next,
2510                                   struct osd_ios_item, oii_list);
2511                 list_del_init(&item->oii_list);
2512                 dput(item->oii_dentry);
2513                 OBD_FREE_PTR(item);
2514         }
2515
2516         if (rc != 0)
2517                 RETURN(rc);
2518
2519         /* There maybe the case that the object has been removed, but its OI
2520          * mapping is still in the OI file, such as the "CATALOGS" after MDT
2521          * file-level backup/restore. So here cleanup the stale OI mappings. */
2522         while (map->olm_name != NULL) {
2523                 struct dentry *child;
2524
2525                 if (fid_is_zero(&map->olm_fid)) {
2526                         map++;
2527                         continue;
2528                 }
2529
2530                 child = osd_ios_lookup_one_len(map->olm_name,
2531                                                osd_sb(dev)->s_root,
2532                                                map->olm_namelen);
2533                 if (!IS_ERR(child))
2534                         dput(child);
2535                 else if (PTR_ERR(child) == -ENOENT)
2536                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2537                                                   NULL, DTO_INDEX_DELETE,
2538                                                   true, 0, NULL);
2539                 map++;
2540         }
2541
2542         RETURN(0);
2543 }
2544
2545 char *osd_lf_fid2name(const struct lu_fid *fid)
2546 {
2547         const struct osd_lf_map *map = osd_lf_maps;
2548
2549         while (map->olm_name != NULL) {
2550                 if (!lu_fid_eq(fid, &map->olm_fid)) {
2551                         map++;
2552                         continue;
2553                 }
2554
2555                 if (map->olm_flags & OLF_SHOW_NAME)
2556                         return map->olm_name;
2557                 else
2558                         return "";
2559         }
2560
2561         return NULL;
2562 }
2563
2564 /* OI scrub start/stop */
2565
2566 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2567 {
2568         struct osd_scrub     *scrub  = &dev->od_scrub;
2569         struct ptlrpc_thread *thread = &scrub->os_thread;
2570         struct l_wait_info    lwi    = { 0 };
2571         struct task_struct   *task;
2572         int                   rc;
2573         ENTRY;
2574
2575         if (dev->od_dt_dev.dd_rdonly)
2576                 RETURN(-EROFS);
2577
2578         /* os_lock: sync status between stop and scrub thread */
2579         spin_lock(&scrub->os_lock);
2580
2581 again:
2582         if (thread_is_running(thread)) {
2583                 spin_unlock(&scrub->os_lock);
2584                 if (!(scrub->os_file.sf_flags & SF_AUTO ||
2585                       scrub->os_partial_scan) ||
2586                      (flags & SS_AUTO_PARTIAL))
2587                         RETURN(-EALREADY);
2588
2589                 osd_scrub_join(dev, flags, false);
2590                 spin_lock(&scrub->os_lock);
2591                 if (!thread_is_running(thread))
2592                         goto again;
2593
2594                 spin_unlock(&scrub->os_lock);
2595                 RETURN(0);
2596         }
2597
2598         if (unlikely(thread_is_stopping(thread))) {
2599                 spin_unlock(&scrub->os_lock);
2600                 l_wait_event(thread->t_ctl_waitq,
2601                              thread_is_stopped(thread),
2602                              &lwi);
2603                 spin_lock(&scrub->os_lock);
2604                 goto again;
2605         }
2606         spin_unlock(&scrub->os_lock);
2607
2608         if (scrub->os_file.sf_status == SS_COMPLETED) {
2609                 if (!(flags & SS_SET_FAILOUT))
2610                         flags |= SS_CLEAR_FAILOUT;
2611
2612                 if (!(flags & SS_SET_DRYRUN))
2613                         flags |= SS_CLEAR_DRYRUN;
2614
2615                 flags |= SS_RESET;
2616         }
2617
2618         scrub->os_start_flags = flags;
2619         thread_set_flags(thread, 0);
2620         task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2621         if (IS_ERR(task)) {
2622                 rc = PTR_ERR(task);
2623                 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2624                        osd_scrub2name(scrub), rc);
2625                 RETURN(rc);
2626         }
2627
2628         l_wait_event(thread->t_ctl_waitq,
2629                      thread_is_running(thread) || thread_is_stopped(thread),
2630                      &lwi);
2631
2632         RETURN(0);
2633 }
2634
2635 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2636 {
2637         int rc;
2638         ENTRY;
2639
2640         /* od_otable_mutex: prevent curcurrent start/stop */
2641         mutex_lock(&dev->od_otable_mutex);
2642         rc = do_osd_scrub_start(dev, flags);
2643         mutex_unlock(&dev->od_otable_mutex);
2644
2645         RETURN(rc == -EALREADY ? 0 : rc);
2646 }
2647
2648 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2649 {
2650         struct ptlrpc_thread *thread = &scrub->os_thread;
2651         struct l_wait_info    lwi    = { 0 };
2652
2653         /* os_lock: sync status between stop and scrub thread */
2654         spin_lock(&scrub->os_lock);
2655         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2656                 thread_set_flags(thread, SVC_STOPPING);
2657                 spin_unlock(&scrub->os_lock);
2658                 wake_up_all(&thread->t_ctl_waitq);
2659                 l_wait_event(thread->t_ctl_waitq,
2660                              thread_is_stopped(thread),
2661                              &lwi);
2662                 /* Do not skip the last lock/unlock, which can guarantee that
2663                  * the caller cannot return until the OI scrub thread exit. */
2664                 spin_lock(&scrub->os_lock);
2665         }
2666         spin_unlock(&scrub->os_lock);
2667 }
2668
2669 static void osd_scrub_stop(struct osd_device *dev)
2670 {
2671         /* od_otable_mutex: prevent curcurrent start/stop */
2672         mutex_lock(&dev->od_otable_mutex);
2673         dev->od_scrub.os_paused = 1;
2674         do_osd_scrub_stop(&dev->od_scrub);
2675         mutex_unlock(&dev->od_otable_mutex);
2676 }
2677
2678 /* OI scrub setup/cleanup */
2679
2680 static const char osd_scrub_name[] = "OI_scrub";
2681
2682 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2683 {
2684         struct osd_thread_info     *info   = osd_oti_get(env);
2685         struct osd_scrub           *scrub  = &dev->od_scrub;
2686         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
2687         struct scrub_file          *sf     = &scrub->os_file;
2688         struct super_block         *sb     = osd_sb(dev);
2689         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
2690         struct lvfs_run_ctxt        saved;
2691         struct file                *filp;
2692         struct inode               *inode;
2693         struct lu_fid              *fid    = &info->oti_fid;
2694         bool                        dirty  = false;
2695         bool                        restored = false;
2696         int                         rc     = 0;
2697         ENTRY;
2698
2699         memset(scrub, 0, sizeof(*scrub));
2700         OBD_SET_CTXT_MAGIC(ctxt);
2701         ctxt->pwdmnt = dev->od_mnt;
2702         ctxt->pwd = dev->od_mnt->mnt_root;
2703         ctxt->fs = get_ds();
2704
2705         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2706         init_rwsem(&scrub->os_rwsem);
2707         spin_lock_init(&scrub->os_lock);
2708         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2709
2710         push_ctxt(&saved, ctxt);
2711         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2712         if (IS_ERR(filp)) {
2713                 pop_ctxt(&saved, ctxt);
2714                 RETURN(PTR_ERR(filp));
2715         }
2716
2717         inode = file_inode(filp);
2718         /* 'What the @fid is' is not imporatant, because the object
2719          * has no OI mapping, and only is visible inside the OSD.*/
2720         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2721         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2722         if (rc != 0) {
2723                 filp_close(filp, NULL);
2724                 pop_ctxt(&saved, ctxt);
2725                 RETURN(rc);
2726         }
2727
2728         scrub->os_inode = igrab(inode);
2729         filp_close(filp, NULL);
2730         pop_ctxt(&saved, ctxt);
2731
2732         rc = osd_scrub_file_load(scrub);
2733         if (rc == -ENOENT) {
2734                 osd_scrub_file_init(scrub, es->s_uuid);
2735                 /* If the "/O" dir does not exist when mount (indicated by
2736                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2737                  * then it is quite probably that the device is a new one,
2738                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2739                  *
2740                  * For the rare case that "/O" and "OI_scrub" both lost on
2741                  * an old device, it can be found and cleared later.
2742                  *
2743                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2744                  * need to check "filter_fid_old" and to convert it to
2745                  * "filter_fid" for each object, and all the IGIF should
2746                  * have their FID mapping in OI files already. */
2747                 if (dev->od_maybe_new)
2748                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2749                 dirty = true;
2750         } else if (rc != 0) {
2751                 GOTO(cleanup_inode, rc);
2752         } else {
2753                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2754                         struct obd_uuid *old_uuid;
2755                         struct obd_uuid *new_uuid;
2756
2757                         OBD_ALLOC_PTR(old_uuid);
2758                         OBD_ALLOC_PTR(new_uuid);
2759                         if (old_uuid == NULL || new_uuid == NULL) {
2760                                 CERROR("%.16s: UUID has been changed, but"
2761                                        "failed to allocate RAM for report\n",
2762                                        LDISKFS_SB(sb)->s_es->s_volume_name);
2763                         } else {
2764                                 class_uuid_unparse(sf->sf_uuid, old_uuid);
2765                                 class_uuid_unparse(es->s_uuid, new_uuid);
2766                                 CERROR("%.16s: UUID has been changed from "
2767                                        "%s to %s\n",
2768                                        LDISKFS_SB(sb)->s_es->s_volume_name,
2769                                        old_uuid->uuid, new_uuid->uuid);
2770                         }
2771                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2772                         dirty = true;
2773                         restored = true;
2774                         if (old_uuid != NULL)
2775                                 OBD_FREE_PTR(old_uuid);
2776                         if (new_uuid != NULL)
2777                                 OBD_FREE_PTR(new_uuid);
2778                 } else if (sf->sf_status == SS_SCANNING) {
2779                         sf->sf_status = SS_CRASHED;
2780                         dirty = true;
2781                 }
2782         }
2783
2784         if (sf->sf_pos_last_checkpoint != 0)
2785                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2786         else
2787                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2788
2789         if (dirty) {
2790                 rc = osd_scrub_file_store(scrub);
2791                 if (rc != 0)
2792                         GOTO(cleanup_inode, rc);
2793         }
2794
2795         /* Initialize OI files. */
2796         rc = osd_oi_init(info, dev, restored);
2797         if (rc < 0)
2798                 GOTO(cleanup_inode, rc);
2799
2800         rc = osd_initial_OI_scrub(info, dev);
2801         if (rc != 0)
2802                 GOTO(cleanup_oi, rc);
2803
2804         if (sf->sf_flags & SF_UPGRADE ||
2805             !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2806               sf->sf_success_count > 0)) {
2807                 dev->od_igif_inoi = 0;
2808                 dev->od_check_ff = dev->od_is_ost;
2809         } else {
2810                 dev->od_igif_inoi = 1;
2811                 dev->od_check_ff = 0;
2812         }
2813
2814         if (sf->sf_flags & SF_INCONSISTENT)
2815                 /* The 'od_igif_inoi' will be set under the
2816                  * following cases:
2817                  * 1) new created system, or
2818                  * 2) restored from file-level backup, or
2819                  * 3) the upgrading completed.
2820                  *
2821                  * The 'od_igif_inoi' may be cleared by OI scrub
2822                  * later if found that the system is upgrading. */
2823                 dev->od_igif_inoi = 1;
2824
2825         if (!dev->od_dt_dev.dd_rdonly && !dev->od_noscrub &&
2826             ((sf->sf_status == SS_PAUSED) ||
2827              (sf->sf_status == SS_CRASHED &&
2828               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2829                               SF_UPGRADE | SF_AUTO)) ||
2830              (sf->sf_status == SS_INIT &&
2831               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2832                               SF_UPGRADE))))
2833                 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2834
2835         if (rc != 0)
2836                 GOTO(cleanup_oi, rc);
2837
2838         /* it is possible that dcache entries may keep objects after they are
2839          * deleted by OSD. While it looks safe this can cause object data to
2840          * stay until umount causing failures in tests calculating free space,
2841          * e.g. replay-ost-single. Since those dcache entries are not used
2842          * anymore let's just free them after use here */
2843         shrink_dcache_sb(sb);
2844
2845         RETURN(0);
2846 cleanup_oi:
2847         osd_oi_fini(info, dev);
2848 cleanup_inode:
2849         iput(scrub->os_inode);
2850         scrub->os_inode = NULL;
2851
2852         return rc;
2853 }
2854
2855 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2856 {
2857         struct osd_scrub *scrub = &dev->od_scrub;
2858
2859         LASSERT(dev->od_otable_it == NULL);
2860
2861         if (scrub->os_inode != NULL) {
2862                 osd_scrub_stop(dev);
2863                 iput(scrub->os_inode);
2864                 scrub->os_inode = NULL;
2865         }
2866         if (dev->od_oi_table != NULL)
2867                 osd_oi_fini(osd_oti_get(env), dev);
2868 }
2869
2870 /* object table based iteration APIs */
2871
2872 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2873                                        struct dt_object *dt, __u32 attr)
2874 {
2875         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2876         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2877         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2878         struct osd_scrub       *scrub = &dev->od_scrub;
2879         struct osd_otable_it   *it;
2880         __u32                   start = 0;
2881         int                     rc;
2882         ENTRY;
2883
2884         /* od_otable_mutex: prevent curcurrent init/fini */
2885         mutex_lock(&dev->od_otable_mutex);
2886         if (dev->od_otable_it != NULL)
2887                 GOTO(out, it = ERR_PTR(-EALREADY));
2888
2889         OBD_ALLOC_PTR(it);
2890         if (it == NULL)
2891                 GOTO(out, it = ERR_PTR(-ENOMEM));
2892
2893         dev->od_otable_it = it;
2894         it->ooi_dev = dev;
2895         it->ooi_cache.ooc_consumer_idx = -1;
2896         if (flags & DOIF_OUTUSED)
2897                 it->ooi_used_outside = 1;
2898
2899         if (flags & DOIF_RESET)
2900                 start |= SS_RESET;
2901
2902         if (valid & DOIV_ERROR_HANDLE) {
2903                 if (flags & DOIF_FAILOUT)
2904                         start |= SS_SET_FAILOUT;
2905                 else
2906                         start |= SS_CLEAR_FAILOUT;
2907         }
2908
2909         if (valid & DOIV_DRYRUN) {
2910                 if (flags & DOIF_DRYRUN)
2911                         start |= SS_SET_DRYRUN;
2912                 else
2913                         start |= SS_CLEAR_DRYRUN;
2914         }
2915
2916         rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2917         if (rc < 0 && rc != -EALREADY) {
2918                 dev->od_otable_it = NULL;
2919                 OBD_FREE_PTR(it);
2920                 GOTO(out, it = ERR_PTR(rc));
2921         }
2922
2923         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2924
2925         GOTO(out, it);
2926
2927 out:
2928         mutex_unlock(&dev->od_otable_mutex);
2929         return (struct dt_it *)it;
2930 }
2931
2932 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2933 {
2934         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2935         struct osd_device    *dev = it->ooi_dev;
2936
2937         /* od_otable_mutex: prevent curcurrent init/fini */
2938         mutex_lock(&dev->od_otable_mutex);
2939         do_osd_scrub_stop(&dev->od_scrub);
2940         LASSERT(dev->od_otable_it == it);
2941
2942         dev->od_otable_it = NULL;
2943         mutex_unlock(&dev->od_otable_mutex);
2944         OBD_FREE_PTR(it);
2945 }
2946
2947 static int osd_otable_it_get(const struct lu_env *env,
2948                              struct dt_it *di, const struct dt_key *key)
2949 {
2950         return 0;
2951 }
2952
2953 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2954 {
2955 }
2956
2957 static inline int
2958 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2959 {
2960         spin_lock(&scrub->os_lock);
2961         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2962             scrub->os_waiting ||
2963             !thread_is_running(&scrub->os_thread))
2964                 it->ooi_waiting = 0;
2965         else
2966                 it->ooi_waiting = 1;
2967         spin_unlock(&scrub->os_lock);
2968
2969         return !it->ooi_waiting;
2970 }
2971
2972 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2973 {
2974         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2975         struct osd_device       *dev    = it->ooi_dev;
2976         struct osd_scrub        *scrub  = &dev->od_scrub;
2977         struct osd_otable_cache *ooc    = &it->ooi_cache;
2978         struct ptlrpc_thread    *thread = &scrub->os_thread;
2979         struct l_wait_info       lwi    = { 0 };
2980         int                      rc;
2981         ENTRY;
2982
2983         LASSERT(it->ooi_user_ready);
2984
2985 again:
2986         if (!thread_is_running(thread) && !it->ooi_used_outside)
2987                 RETURN(1);
2988
2989         if (ooc->ooc_cached_items > 0) {
2990                 ooc->ooc_cached_items--;
2991                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2992                                         ~OSD_OTABLE_IT_CACHE_MASK;
2993                 RETURN(0);
2994         }
2995
2996         if (it->ooi_all_cached) {
2997                 l_wait_event(thread->t_ctl_waitq,
2998                              !thread_is_running(thread),
2999                              &lwi);
3000                 RETURN(1);
3001         }
3002
3003         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
3004                 spin_lock(&scrub->os_lock);
3005                 scrub->os_waiting = 0;
3006                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
3007                 spin_unlock(&scrub->os_lock);
3008         }
3009
3010         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
3011                 l_wait_event(thread->t_ctl_waitq,
3012                              osd_otable_it_wakeup(scrub, it),
3013                              &lwi);
3014
3015         if (!thread_is_running(thread) && !it->ooi_used_outside)
3016                 RETURN(1);
3017
3018         rc = osd_otable_it_preload(env, it);
3019         if (rc >= 0)
3020                 goto again;
3021
3022         RETURN(rc);
3023 }
3024
3025 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
3026                                         const struct dt_it *di)
3027 {
3028         return NULL;
3029 }
3030
3031 static int osd_otable_it_key_size(const struct lu_env *env,
3032                                   const struct dt_it *di)
3033 {
3034         return sizeof(__u64);
3035 }
3036
3037 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
3038                              struct dt_rec *rec, __u32 attr)
3039 {
3040         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
3041         struct osd_otable_cache *ooc = &it->ooi_cache;
3042
3043         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
3044
3045         /* Filter out Invald FID already. */
3046         LASSERTF(fid_is_sane((struct lu_fid *)rec),
3047                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
3048                  PFID((struct lu_fid *)rec),
3049                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
3050
3051         return 0;
3052 }
3053
3054 static __u64 osd_otable_it_store(const struct lu_env *env,
3055                                  const struct dt_it *di)
3056 {
3057         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
3058         struct osd_otable_cache *ooc = &it->ooi_cache;
3059         __u64                    hash;
3060
3061         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
3062                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
3063         else
3064                 hash = ooc->ooc_pos_preload;
3065         return hash;
3066 }
3067
3068 /**
3069  * Set the OSD layer iteration start position as the specified hash.
3070  */
3071 static int osd_otable_it_load(const struct lu_env *env,
3072                               const struct dt_it *di, __u64 hash)
3073 {
3074         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
3075         struct osd_device       *dev   = it->ooi_dev;
3076         struct osd_otable_cache *ooc   = &it->ooi_cache;
3077         struct osd_scrub        *scrub = &dev->od_scrub;
3078         int                      rc;
3079         ENTRY;
3080
3081         /* Forbid to set iteration position after iteration started. */
3082         if (it->ooi_user_ready)
3083                 RETURN(-EPERM);
3084
3085         LASSERT(!scrub->os_partial_scan);
3086
3087         if (hash > OSD_OTABLE_MAX_HASH)
3088                 hash = OSD_OTABLE_MAX_HASH;
3089
3090         ooc->ooc_pos_preload = hash;
3091         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
3092                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
3093
3094         it->ooi_user_ready = 1;
3095         if (!scrub->os_full_speed)
3096                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
3097
3098         /* Unplug OSD layer iteration by the first next() call. */
3099         rc = osd_otable_it_next(env, (struct dt_it *)it);
3100
3101         RETURN(rc);
3102 }
3103
3104 static int osd_otable_it_key_rec(const struct lu_env *env,
3105                                  const struct dt_it *di, void *key_rec)
3106 {
3107         return 0;
3108 }
3109
3110 const struct dt_index_operations osd_otable_ops = {
3111         .dio_it = {
3112                 .init     = osd_otable_it_init,
3113                 .fini     = osd_otable_it_fini,
3114                 .get      = osd_otable_it_get,
3115                 .put      = osd_otable_it_put,
3116                 .next     = osd_otable_it_next,
3117                 .key      = osd_otable_it_key,
3118                 .key_size = osd_otable_it_key_size,
3119                 .rec      = osd_otable_it_rec,
3120                 .store    = osd_otable_it_store,
3121                 .load     = osd_otable_it_load,
3122                 .key_rec  = osd_otable_it_key_rec,
3123         }
3124 };
3125
3126 /* high priority inconsistent items list APIs */
3127
3128 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL  60
3129
3130 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
3131                    int insert)
3132 {
3133         struct osd_inconsistent_item *oii;
3134         struct osd_scrub             *scrub  = &dev->od_scrub;
3135         struct ptlrpc_thread         *thread = &scrub->os_thread;
3136         int                           wakeup = 0;
3137         ENTRY;
3138
3139         OBD_ALLOC_PTR(oii);
3140         if (unlikely(oii == NULL))
3141                 RETURN(-ENOMEM);
3142
3143         INIT_LIST_HEAD(&oii->oii_list);
3144         oii->oii_cache = *oic;
3145         oii->oii_insert = insert;
3146
3147         if (scrub->os_partial_scan) {
3148                 __u64 now = cfs_time_current_sec();
3149
3150                 /* If there haven't been errors in a long time,
3151                  * decay old count until either the errors are
3152                  * gone or we reach the current interval. */
3153                 while (unlikely(scrub->os_bad_oimap_count > 0 &&
3154                                 scrub->os_bad_oimap_time +
3155                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
3156                         scrub->os_bad_oimap_count >>= 1;
3157                         scrub->os_bad_oimap_time +=
3158                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
3159                 }
3160
3161                 scrub->os_bad_oimap_time = now;
3162                 if (++scrub->os_bad_oimap_count >
3163                     dev->od_full_scrub_threshold_rate)
3164                         scrub->os_full_scrub = 1;
3165         }
3166
3167         spin_lock(&scrub->os_lock);
3168         if (unlikely(!thread_is_running(thread))) {
3169                 spin_unlock(&scrub->os_lock);
3170                 OBD_FREE_PTR(oii);
3171                 RETURN(-EAGAIN);
3172         }
3173
3174         if (list_empty(&scrub->os_inconsistent_items))
3175                 wakeup = 1;
3176         list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
3177         spin_unlock(&scrub->os_lock);
3178
3179         if (wakeup != 0)
3180                 wake_up_all(&thread->t_ctl_waitq);
3181
3182         RETURN(0);
3183 }
3184
3185 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
3186                    struct osd_inode_id *id)
3187 {
3188         struct osd_scrub             *scrub = &dev->od_scrub;
3189         struct osd_inconsistent_item *oii;
3190         ENTRY;
3191
3192         spin_lock(&scrub->os_lock);
3193         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
3194                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
3195                         *id = oii->oii_cache.oic_lid;
3196                         spin_unlock(&scrub->os_lock);
3197                         RETURN(0);
3198                 }
3199         }
3200         spin_unlock(&scrub->os_lock);
3201
3202         RETURN(-ENOENT);
3203 }
3204
3205 /* OI scrub dump */
3206
3207 static const char *scrub_status_names[] = {
3208         "init",
3209         "scanning",
3210         "completed",
3211         "failed",
3212         "stopped",
3213         "paused",
3214         "crashed",
3215         NULL
3216 };
3217
3218 static const char *scrub_flags_names[] = {
3219         "recreated",
3220         "inconsistent",
3221         "auto",
3222         "upgrade",
3223         NULL
3224 };
3225
3226 static const char *scrub_param_names[] = {
3227         "failout",
3228         "dryrun",
3229         NULL
3230 };
3231
3232 static void scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
3233                             const char *prefix)
3234 {
3235         int flag;
3236         int i;
3237
3238         seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
3239
3240         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
3241                 if (flag & bits) {
3242                         bits &= ~flag;
3243                         seq_printf(m, "%s%c", names[i],
3244                                    bits != 0 ? ',' : '\n');
3245                 }
3246         }
3247 }
3248
3249 static void scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
3250 {
3251         if (time != 0)
3252                 seq_printf(m, "%s: %llu seconds\n", prefix,
3253                            cfs_time_current_sec() - time);
3254         else
3255                 seq_printf(m, "%s: N/A\n", prefix);
3256 }
3257
3258 static void scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
3259 {
3260         if (pos != 0)
3261                 seq_printf(m, "%s: %llu\n", prefix, pos);
3262         else
3263                 seq_printf(m, "%s: N/A\n", prefix);
3264 }
3265
3266 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
3267 {
3268         struct osd_scrub  *scrub   = &dev->od_scrub;
3269         struct scrub_file *sf      = &scrub->os_file;
3270         __u64              checked;
3271         __u64              speed;
3272
3273         down_read(&scrub->os_rwsem);
3274         seq_printf(m, "name: OI_scrub\n"
3275                    "magic: 0x%x\n"
3276                    "oi_files: %d\n"
3277                    "status: %s\n",
3278                    sf->sf_magic, (int)sf->sf_oi_count,
3279                    scrub_status_names[sf->sf_status]);
3280
3281         scrub_bits_dump(m, sf->sf_flags, scrub_flags_names, "flags");
3282
3283         scrub_bits_dump(m, sf->sf_param, scrub_param_names, "param");
3284
3285         scrub_time_dump(m, sf->sf_time_last_complete,
3286                         "time_since_last_completed");
3287
3288         scrub_time_dump(m, sf->sf_time_latest_start,
3289                         "time_since_latest_start");
3290
3291         scrub_time_dump(m, sf->sf_time_last_checkpoint,
3292                         "time_since_last_checkpoint");
3293
3294         scrub_pos_dump(m, sf->sf_pos_latest_start,
3295                         "latest_start_position");
3296
3297         scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
3298                         "last_checkpoint_position");
3299
3300         scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
3301                         "first_failure_position");
3302
3303         checked = sf->sf_items_checked + scrub->os_new_checked;
3304         seq_printf(m, "checked: %llu\n"
3305                    "updated: %llu\n"
3306                    "failed: %llu\n"
3307                    "prior_updated: %llu\n"
3308                    "noscrub: %llu\n"
3309                    "igif: %llu\n"
3310                    "success_count: %u\n",
3311                    checked, sf->sf_items_updated, sf->sf_items_failed,
3312                    sf->sf_items_updated_prior, sf->sf_items_noscrub,
3313                    sf->sf_items_igif, sf->sf_success_count);
3314
3315         speed = checked;
3316         if (thread_is_running(&scrub->os_thread)) {
3317                 cfs_duration_t duration = cfs_time_current() -
3318                                           scrub->os_time_last_checkpoint;
3319                 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
3320                                                      MSEC_PER_SEC);
3321                 __u32 rtime = sf->sf_run_time +
3322                               cfs_duration_sec(duration + HALF_SEC);
3323
3324                 if (duration != 0)
3325                         do_div(new_checked, duration);
3326                 if (rtime != 0)
3327                         do_div(speed, rtime);
3328                 seq_printf(m, "run_time: %u seconds\n"
3329                            "average_speed: %llu objects/sec\n"
3330                            "real-time_speed: %llu objects/sec\n"
3331                            "current_position: %u\n"
3332                            "lf_scanned: %llu\n"
3333                            "lf_repaired: %llu\n"
3334                            "lf_failed: %llu\n",
3335                            rtime, speed, new_checked, scrub->os_pos_current,
3336                            scrub->os_lf_scanned, scrub->os_lf_repaired,
3337                            scrub->os_lf_failed);
3338         } else {
3339                 if (sf->sf_run_time != 0)
3340                         do_div(speed, sf->sf_run_time);
3341                 seq_printf(m, "run_time: %u seconds\n"
3342                            "average_speed: %llu objects/sec\n"
3343                            "real-time_speed: N/A\n"
3344                            "current_position: N/A\n"
3345                            "lf_scanned: %llu\n"
3346                            "lf_repaired: %llu\n"
3347                            "lf_failed: %llu\n",
3348                            sf->sf_run_time, speed, scrub->os_lf_scanned,
3349                            scrub->os_lf_repaired, scrub->os_lf_failed);
3350         }
3351
3352         up_read(&scrub->os_rwsem);
3353         return 0;
3354 }