lustre/osd-ldiskfs/osd_scrub.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2012, 2014, Intel Corporation.
  24  */
  25 /*
  26  * lustre/osd-ldiskfs/osd_scrub.c
  27  *
  28  * Top-level entry points into osd module
  29  *
  30  * The OI scrub is used for rebuilding Object Index files when restores MDT from
  31  * file-level backup.
  32  *
  33  * The otable based iterator scans ldiskfs inode table to feed up layer LFSCK.
  34  *
  35  * Author: Fan Yong <yong.fan@whamcloud.com>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_LFSCK
  39
  40 #include <lustre/lustre_idl.h>
  41 #include <lustre_disk.h>
  42 #include <dt_object.h>
  43 #include <linux/xattr.h>
  44
  45 #include "osd_internal.h"
  46 #include "osd_oi.h"
  47 #include "osd_scrub.h"
  48
  49 #define HALF_SEC        msecs_to_jiffies(MSEC_PER_SEC >> 1)
  50
  51 #define OSD_OTABLE_MAX_HASH             0x00000000ffffffffULL
  52
  53 #define SCRUB_NEXT_BREAK        1 /* exit current loop and process next group */
  54 #define SCRUB_NEXT_CONTINUE     2 /* skip current object and process next bit */
  55 #define SCRUB_NEXT_EXIT         3 /* exit all the loops */
  56 #define SCRUB_NEXT_WAIT         4 /* wait for free cache slot */
  57 #define SCRUB_NEXT_CRASH        5 /* simulate system crash during OI scrub */
  58 #define SCRUB_NEXT_FATAL        6 /* simulate failure during OI scrub */
  59 #define SCRUB_NEXT_NOSCRUB      7 /* new created object, no scrub on it */
  60 #define SCRUB_NEXT_NOLMA        8 /* the inode has no FID-in-LMA */
  61 #define SCRUB_NEXT_OSTOBJ       9 /* for OST-object */
  62 #define SCRUB_NEXT_OSTOBJ_OLD   10 /* old OST-object, no LMA or no FID-on-OST
  63                                     * flags in LMA */
  64
  65 /* misc functions */
  66
  67 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
  68 {
  69         return container_of0(scrub, struct osd_device, od_scrub);
  70 }
  71
  72 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
  73 {
  74         return osd_sb(osd_scrub2dev(scrub));
  75 }
  76
  77 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
  78                                        struct osd_otable_cache *ooc)
  79 {
  80         return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
  81 }
  82
  83 static inline const char *osd_scrub2name(struct osd_scrub *scrub)
  84 {
  85         return LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
  86 }
  87
  88 /**
  89  * update/insert/delete the specified OI mapping (@fid @id) according to the ops
  90  *
  91  * \retval   1, changed nothing
  92  * \retval   0, changed successfully
  93  * \retval -ve, on error
  94  */
  95 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
  96                                      struct osd_device *dev,
  97                                      const struct lu_fid *fid,
  98                                      const struct osd_inode_id *id,
  99                                      int ops, bool force,
 100                                      enum oi_check_flags flags)
 101 {
 102         handle_t *th;
 103         int       rc;
 104         ENTRY;
 105
 106         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
 107                 RETURN(0);
 108
 109         /* DTO_INDEX_INSERT is enough for other two ops:
 110          * delete/update, but save stack. */
 111         th = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 112                                 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
 113         if (IS_ERR(th)) {
 114                 rc = PTR_ERR(th);
 115                 CDEBUG(D_LFSCK, "%s: fail to start trans for scrub op %d "
 116                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 117                        PFID(fid), id->oii_ino, id->oii_gen, rc);
 118                 RETURN(rc);
 119         }
 120
 121         switch (ops) {
 122         case DTO_INDEX_UPDATE:
 123                 rc = osd_oi_update(info, dev, fid, id, th, flags);
 124                 if (unlikely(rc == -ENOENT)) {
 125                         /* Some unlink thread may removed the OI mapping. */
 126                         rc = 1;
 127                 }
 128                 break;
 129         case DTO_INDEX_INSERT:
 130                 rc = osd_oi_insert(info, dev, fid, id, th, flags);
 131                 if (unlikely(rc == -EEXIST)) {
 132                         rc = 1;
 133                         /* XXX: There are trouble things when adding OI
 134                          *      mapping for IGIF object, which may cause
 135                          *      multiple objects to be mapped to the same
 136                          *      IGIF formatted FID. Consider the following
 137                          *      situations:
 138                          *
 139                          *      1) The MDT is upgrading from 1.8 device.
 140                          *      The OI scrub generates IGIF FID1 for the
 141                          *      OBJ1 and adds the OI mapping.
 142                          *
 143                          *      2) For some reason, the OI scrub does not
 144                          *      process all the IGIF objects completely.
 145                          *
 146                          *      3) The MDT is backuped and restored against
 147                          *      this device.
 148                          *
 149                          *      4) When the MDT mounts up, the OI scrub will
 150                          *      try to rebuild the OI files. For some IGIF
 151                          *      object, OBJ2, which was not processed by the
 152                          *      OI scrub before the backup/restore, and the
 153                          *      new generated IGIF formatted FID may be just
 154                          *      the FID1, the same as OBJ1.
 155                          *
 156                          *      Under such case, the OI scrub cannot know how
 157                          *      to generate new FID for the OBJ2.
 158                          *
 159                          *      Currently, we do nothing for that. One possible
 160                          *      solution is to generate new normal FID for the
 161                          *      conflict object.
 162                          *
 163                          *      Anyway, it is rare, only exists in theory. */
 164                 }
 165                 break;
 166         case DTO_INDEX_DELETE:
 167                 rc = osd_oi_delete(info, dev, fid, th, flags);
 168                 if (rc == -ENOENT) {
 169                         /* It is normal that the unlink thread has removed the
 170                          * OI mapping already. */
 171                         rc = 1;
 172                 }
 173                 break;
 174         default:
 175                 LASSERTF(0, "Unexpected ops %d\n", ops);
 176                 break;
 177         }
 178
 179         ldiskfs_journal_stop(th);
 180         if (rc < 0)
 181                 CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d "
 182                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 183                        PFID(fid), id->oii_ino, id->oii_gen, rc);
 184
 185         RETURN(rc);
 186 }
 187
 188 /* OI_scrub file ops */
 189
 190 static void osd_scrub_file_to_cpu(struct scrub_file *des,
 191                                   struct scrub_file *src)
 192 {
 193         memcpy(des->sf_uuid, src->sf_uuid, 16);
 194         des->sf_flags   = le64_to_cpu(src->sf_flags);
 195         des->sf_magic   = le32_to_cpu(src->sf_magic);
 196         des->sf_status  = le16_to_cpu(src->sf_status);
 197         des->sf_param   = le16_to_cpu(src->sf_param);
 198         des->sf_time_last_complete      =
 199                                 le64_to_cpu(src->sf_time_last_complete);
 200         des->sf_time_latest_start       =
 201                                 le64_to_cpu(src->sf_time_latest_start);
 202         des->sf_time_last_checkpoint    =
 203                                 le64_to_cpu(src->sf_time_last_checkpoint);
 204         des->sf_pos_latest_start        =
 205                                 le64_to_cpu(src->sf_pos_latest_start);
 206         des->sf_pos_last_checkpoint     =
 207                                 le64_to_cpu(src->sf_pos_last_checkpoint);
 208         des->sf_pos_first_inconsistent  =
 209                                 le64_to_cpu(src->sf_pos_first_inconsistent);
 210         des->sf_items_checked           =
 211                                 le64_to_cpu(src->sf_items_checked);
 212         des->sf_items_updated           =
 213                                 le64_to_cpu(src->sf_items_updated);
 214         des->sf_items_failed            =
 215                                 le64_to_cpu(src->sf_items_failed);
 216         des->sf_items_updated_prior     =
 217                                 le64_to_cpu(src->sf_items_updated_prior);
 218         des->sf_run_time        = le32_to_cpu(src->sf_run_time);
 219         des->sf_success_count   = le32_to_cpu(src->sf_success_count);
 220         des->sf_oi_count        = le16_to_cpu(src->sf_oi_count);
 221         des->sf_internal_flags  = le16_to_cpu(src->sf_internal_flags);
 222         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 223 }
 224
 225 static void osd_scrub_file_to_le(struct scrub_file *des,
 226                                  struct scrub_file *src)
 227 {
 228         memcpy(des->sf_uuid, src->sf_uuid, 16);
 229         des->sf_flags   = cpu_to_le64(src->sf_flags);
 230         des->sf_magic   = cpu_to_le32(src->sf_magic);
 231         des->sf_status  = cpu_to_le16(src->sf_status);
 232         des->sf_param   = cpu_to_le16(src->sf_param);
 233         des->sf_time_last_complete      =
 234                                 cpu_to_le64(src->sf_time_last_complete);
 235         des->sf_time_latest_start       =
 236                                 cpu_to_le64(src->sf_time_latest_start);
 237         des->sf_time_last_checkpoint    =
 238                                 cpu_to_le64(src->sf_time_last_checkpoint);
 239         des->sf_pos_latest_start        =
 240                                 cpu_to_le64(src->sf_pos_latest_start);
 241         des->sf_pos_last_checkpoint     =
 242                                 cpu_to_le64(src->sf_pos_last_checkpoint);
 243         des->sf_pos_first_inconsistent  =
 244                                 cpu_to_le64(src->sf_pos_first_inconsistent);
 245         des->sf_items_checked           =
 246                                 cpu_to_le64(src->sf_items_checked);
 247         des->sf_items_updated           =
 248                                 cpu_to_le64(src->sf_items_updated);
 249         des->sf_items_failed            =
 250                                 cpu_to_le64(src->sf_items_failed);
 251         des->sf_items_updated_prior     =
 252                                 cpu_to_le64(src->sf_items_updated_prior);
 253         des->sf_run_time        = cpu_to_le32(src->sf_run_time);
 254         des->sf_success_count   = cpu_to_le32(src->sf_success_count);
 255         des->sf_oi_count        = cpu_to_le16(src->sf_oi_count);
 256         des->sf_internal_flags  = cpu_to_le16(src->sf_internal_flags);
 257         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 258 }
 259
 260 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
 261 {
 262         struct scrub_file *sf = &scrub->os_file;
 263
 264         memset(sf, 0, sizeof(*sf));
 265         memcpy(sf->sf_uuid, uuid, 16);
 266         sf->sf_magic = SCRUB_MAGIC_V1;
 267         sf->sf_status = SS_INIT;
 268 }
 269
 270 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
 271 {
 272         struct scrub_file *sf = &scrub->os_file;
 273
 274         CDEBUG(D_LFSCK, "%.16s: reset OI scrub file, old flags = "
 275                LPX64", add flags = "LPX64"\n",
 276                osd_scrub2name(scrub), sf->sf_flags, flags);
 277
 278         memcpy(sf->sf_uuid, uuid, 16);
 279         sf->sf_status = SS_INIT;
 280         sf->sf_flags |= flags;
 281         sf->sf_flags &= ~SF_AUTO;
 282         sf->sf_run_time = 0;
 283         sf->sf_time_latest_start = 0;
 284         sf->sf_time_last_checkpoint = 0;
 285         sf->sf_pos_latest_start = 0;
 286         sf->sf_pos_last_checkpoint = 0;
 287         sf->sf_pos_first_inconsistent = 0;
 288         sf->sf_items_checked = 0;
 289         sf->sf_items_updated = 0;
 290         sf->sf_items_failed = 0;
 291         if (!scrub->os_in_join)
 292                 sf->sf_items_updated_prior = 0;
 293
 294         sf->sf_items_noscrub = 0;
 295         sf->sf_items_igif = 0;
 296 }
 297
 298 static int osd_scrub_file_load(struct osd_scrub *scrub)
 299 {
 300         loff_t  pos  = 0;
 301         int     len  = sizeof(scrub->os_file_disk);
 302         int     rc;
 303
 304         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
 305         if (rc == len) {
 306                 struct scrub_file *sf = &scrub->os_file;
 307
 308                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
 309                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
 310                         CDEBUG(D_LFSCK, "%.16s: invalid scrub magic "
 311                                "0x%x != 0x%x\n", osd_scrub2name(scrub),
 312                                sf->sf_magic, SCRUB_MAGIC_V1);
 313                         /* Process it as new scrub file. */
 314                         rc = -ENOENT;
 315                 } else {
 316                         rc = 0;
 317                 }
 318         } else if (rc != 0) {
 319                 CDEBUG(D_LFSCK, "%.16s: fail to load scrub file, "
 320                        "expected = %d: rc = %d\n",
 321                        osd_scrub2name(scrub), len, rc);
 322                 if (rc > 0)
 323                         rc = -EFAULT;
 324         } else {
 325                 /* return -ENOENT for empty scrub file case. */
 326                 rc = -ENOENT;
 327         }
 328
 329         return rc;
 330 }
 331
 332 int osd_scrub_file_store(struct osd_scrub *scrub)
 333 {
 334         struct osd_device *dev;
 335         handle_t          *jh;
 336         loff_t             pos     = 0;
 337         int                len     = sizeof(scrub->os_file_disk);
 338         int                credits;
 339         int                rc;
 340
 341         dev = container_of0(scrub, struct osd_device, od_scrub);
 342         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
 343                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
 344         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits);
 345         if (IS_ERR(jh)) {
 346                 rc = PTR_ERR(jh);
 347                 CDEBUG(D_LFSCK, "%.16s: fail to start trans for scrub store: "
 348                        "rc = %d\n", osd_scrub2name(scrub), rc);
 349                 return rc;
 350         }
 351
 352         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
 353         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
 354                                       len, 0, &pos, jh);
 355         ldiskfs_journal_stop(jh);
 356         if (rc != 0)
 357                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file, "
 358                        "expected = %d: rc = %d\n",
 359                        osd_scrub2name(scrub), len, rc);
 360
 361         scrub->os_time_last_checkpoint = cfs_time_current();
 362         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
 363                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
 364         return rc;
 365 }
 366
 367 static int
 368 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
 369                      struct inode *inode, const struct lu_fid *fid)
 370 {
 371         struct filter_fid_old   *ff      = &info->oti_ff;
 372         struct dentry           *dentry  = &info->oti_obj_dentry;
 373         handle_t                *jh;
 374         int                      size    = 0;
 375         int                      rc;
 376         bool                     removed = false;
 377         bool                     reset   = true;
 378         ENTRY;
 379
 380         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
 381                 RETURN(0);
 382
 383         /* We want the LMA to fit into the 256-byte OST inode, so operate
 384          * as following:
 385          * 1) read old XATTR_NAME_FID and save the parent FID;
 386          * 2) delete the old XATTR_NAME_FID;
 387          * 3) make new LMA and add it;
 388          * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
 389          *
 390          * Making the LMA to fit into the 256-byte OST inode can save time for
 391          * normal osd_check_lma() and for other OI scrub scanning in future.
 392          * So it is worth to make some slow conversion here. */
 393         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 394                                 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
 395         if (IS_ERR(jh)) {
 396                 rc = PTR_ERR(jh);
 397                 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
 398                        DFID": rc = %d\n", osd_name(dev), PFID(fid), rc);
 399                 RETURN(rc);
 400         }
 401
 402         /* 1) read old XATTR_NAME_FID and save the parent FID */
 403         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 404         if (rc == sizeof(*ff)) {
 405                 /* 2) delete the old XATTR_NAME_FID */
 406                 ll_vfs_dq_init(inode);
 407                 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
 408                 if (rc != 0)
 409                         GOTO(stop, rc);
 410
 411                 removed = true;
 412         } else if (unlikely(rc == -ENODATA)) {
 413                 reset = false;
 414         } else if (rc != sizeof(struct filter_fid)) {
 415                 GOTO(stop, rc = -EINVAL);
 416         }
 417
 418         /* 3) make new LMA and add it */
 419         rc = osd_ea_fid_set(info, inode, fid, LMAC_FID_ON_OST, 0);
 420         if (rc == 0 && reset)
 421                 size = sizeof(struct filter_fid);
 422         else if (rc != 0 && removed)
 423                 /* If failed, we should try to add the old back. */
 424                 size = sizeof(struct filter_fid_old);
 425
 426         /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
 427         if (size > 0) {
 428                 int rc1;
 429
 430                 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
 431                                       XATTR_CREATE);
 432                 if (rc1 != 0 && rc == 0)
 433                         rc = rc1;
 434         }
 435
 436         GOTO(stop, rc);
 437
 438 stop:
 439         ldiskfs_journal_stop(jh);
 440         if (rc < 0)
 441                 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
 442                        osd_name(dev), PFID(fid), rc);
 443         return rc;
 444 }
 445
 446 static int
 447 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
 448                        struct osd_idmap_cache *oic, int val)
 449 {
 450         struct osd_scrub             *scrub  = &dev->od_scrub;
 451         struct scrub_file            *sf     = &scrub->os_file;
 452         struct lu_fid                *fid    = &oic->oic_fid;
 453         struct osd_inode_id          *lid    = &oic->oic_lid;
 454         struct osd_inode_id          *lid2   = &info->oti_id;
 455         struct osd_inconsistent_item *oii    = NULL;
 456         struct inode                 *inode  = NULL;
 457         int                           ops    = DTO_INDEX_UPDATE;
 458         int                           idx;
 459         int                           rc;
 460         bool                          converted = false;
 461         ENTRY;
 462
 463         down_write(&scrub->os_rwsem);
 464         scrub->os_new_checked++;
 465         if (val < 0)
 466                 GOTO(out, rc = val);
 467
 468         if (scrub->os_in_prior)
 469                 oii = list_entry(oic, struct osd_inconsistent_item,
 470                                  oii_cache);
 471
 472         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
 473                 GOTO(out, rc = 0);
 474
 475         if (fid_is_igif(fid))
 476                 sf->sf_items_igif++;
 477
 478         if (val == SCRUB_NEXT_OSTOBJ_OLD) {
 479                 inode = osd_iget(info, dev, lid);
 480                 if (IS_ERR(inode)) {
 481                         rc = PTR_ERR(inode);
 482                         /* Someone removed the inode. */
 483                         if (rc == -ENOENT || rc == -ESTALE)
 484                                 rc = 0;
 485                         GOTO(out, rc);
 486                 }
 487
 488                 sf->sf_flags |= SF_UPGRADE;
 489                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
 490                 dev->od_check_ff = 1;
 491                 rc = osd_scrub_convert_ff(info, dev, inode, fid);
 492                 if (rc != 0)
 493                         GOTO(out, rc);
 494
 495                 converted = true;
 496         }
 497
 498         if ((val == SCRUB_NEXT_NOLMA) &&
 499             (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
 500                 GOTO(out, rc = 0);
 501
 502         if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA))
 503                 goto iget;
 504
 505         rc = osd_oi_lookup(info, dev, fid, lid2,
 506                 (val == SCRUB_NEXT_OSTOBJ ||
 507                  val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 508         if (rc != 0) {
 509                 if (rc != -ENOENT && rc != -ESTALE)
 510                         GOTO(out, rc);
 511
 512 iget:
 513                 if (inode == NULL) {
 514                         inode = osd_iget(info, dev, lid);
 515                         if (IS_ERR(inode)) {
 516                                 rc = PTR_ERR(inode);
 517                                 /* Someone removed the inode. */
 518                                 if (rc == -ENOENT || rc == -ESTALE)
 519                                         rc = 0;
 520                                 GOTO(out, rc);
 521                         }
 522                 }
 523
 524                 if (!scrub->os_partial_scan)
 525                         scrub->os_full_speed = 1;
 526
 527                 ops = DTO_INDEX_INSERT;
 528                 idx = osd_oi_fid2idx(dev, fid);
 529                 switch (val) {
 530                 case SCRUB_NEXT_NOLMA:
 531                         sf->sf_flags |= SF_UPGRADE;
 532                         if (!(sf->sf_param & SP_DRYRUN)) {
 533                                 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
 534                                 if (rc != 0)
 535                                         GOTO(out, rc);
 536                         }
 537
 538                         if (!(sf->sf_flags & SF_INCONSISTENT))
 539                                 dev->od_igif_inoi = 0;
 540                         break;
 541                 case SCRUB_NEXT_OSTOBJ:
 542                         sf->sf_flags |= SF_INCONSISTENT;
 543                 case SCRUB_NEXT_OSTOBJ_OLD:
 544                         break;
 545                 default:
 546                         sf->sf_flags |= SF_RECREATED;
 547                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 548                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 549                         break;
 550                 }
 551         } else if (osd_id_eq(lid, lid2)) {
 552                 if (converted)
 553                         sf->sf_items_updated++;
 554
 555                 GOTO(out, rc = 0);
 556         } else {
 557                 if (!scrub->os_partial_scan)
 558                         scrub->os_full_speed = 1;
 559
 560                 sf->sf_flags |= SF_INCONSISTENT;
 561
 562                 /* XXX: If the device is restored from file-level backup, then
 563                  *      some IGIFs may have been already in OI files, and some
 564                  *      may be not yet. Means upgrading from 1.8 may be partly
 565                  *      processed, but some clients may hold some immobilized
 566                  *      IGIFs, and use them to access related objects. Under
 567                  *      such case, OSD does not know whether an given IGIF has
 568                  *      been processed or to be processed, and it also cannot
 569                  *      generate local ino#/gen# directly from the immobilized
 570                  *      IGIF because of the backup/restore. Then force OSD to
 571                  *      lookup the given IGIF in OI files, and if no entry,
 572                  *      then ask the client to retry after upgrading completed.
 573                  *      No better choice. */
 574                 dev->od_igif_inoi = 1;
 575         }
 576
 577         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
 578                         (val == SCRUB_NEXT_OSTOBJ ||
 579                          val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 580         if (rc == 0) {
 581                 if (scrub->os_in_prior)
 582                         sf->sf_items_updated_prior++;
 583                 else
 584                         sf->sf_items_updated++;
 585
 586                 /* The target has been changed, need to be re-loaded. */
 587                 lu_object_purge(info->oti_env, osd2lu_dev(dev), fid);
 588         }
 589
 590         GOTO(out, rc);
 591
 592 out:
 593         if (rc < 0) {
 594                 sf->sf_items_failed++;
 595                 if (sf->sf_pos_first_inconsistent == 0 ||
 596                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 597                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 598         } else {
 599                 rc = 0;
 600         }
 601
 602         /* There may be conflict unlink during the OI scrub,
 603          * if happend, then remove the new added OI mapping. */
 604         if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
 605             unlikely(inode->i_nlink == 0))
 606                 osd_scrub_refresh_mapping(info, dev, fid, lid,
 607                                 DTO_INDEX_DELETE, false,
 608                                 (val == SCRUB_NEXT_OSTOBJ ||
 609                                  val == SCRUB_NEXT_OSTOBJ_OLD) ?
 610                                 OI_KNOWN_ON_OST : 0);
 611         up_write(&scrub->os_rwsem);
 612
 613         if (inode != NULL && !IS_ERR(inode))
 614                 iput(inode);
 615
 616         if (oii != NULL) {
 617                 LASSERT(!list_empty(&oii->oii_list));
 618
 619                 spin_lock(&scrub->os_lock);
 620                 list_del_init(&oii->oii_list);
 621                 spin_unlock(&scrub->os_lock);
 622                 OBD_FREE_PTR(oii);
 623         }
 624         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 625 }
 626
 627 /* OI scrub APIs */
 628
 629 static int osd_scrub_prep(struct osd_device *dev)
 630 {
 631         struct osd_scrub     *scrub  = &dev->od_scrub;
 632         struct ptlrpc_thread *thread = &scrub->os_thread;
 633         struct scrub_file    *sf     = &scrub->os_file;
 634         __u32                 flags  = scrub->os_start_flags;
 635         int                   rc;
 636         bool                  drop_dryrun = false;
 637         ENTRY;
 638
 639         CDEBUG(D_LFSCK, "%.16s: OI scrub prep, flags = 0x%x\n",
 640                osd_scrub2name(scrub), flags);
 641
 642         down_write(&scrub->os_rwsem);
 643         if (flags & SS_SET_FAILOUT)
 644                 sf->sf_param |= SP_FAILOUT;
 645         else if (flags & SS_CLEAR_FAILOUT)
 646                 sf->sf_param &= ~SP_FAILOUT;
 647
 648         if (flags & SS_SET_DRYRUN) {
 649                 sf->sf_param |= SP_DRYRUN;
 650         } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
 651                 sf->sf_param &= ~SP_DRYRUN;
 652                 drop_dryrun = true;
 653         }
 654
 655         if (flags & SS_RESET)
 656                 osd_scrub_file_reset(scrub,
 657                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
 658
 659         if (flags & SS_AUTO_FULL) {
 660                 scrub->os_full_speed = 1;
 661                 scrub->os_partial_scan = 0;
 662                 sf->sf_flags |= SF_AUTO;
 663         } else if (flags & SS_AUTO_PARTIAL) {
 664                 scrub->os_full_speed = 0;
 665                 scrub->os_partial_scan = 1;
 666                 sf->sf_flags |= SF_AUTO;
 667         } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
 668                                    SF_UPGRADE)) {
 669                 scrub->os_full_speed = 1;
 670                 scrub->os_partial_scan = 0;
 671         } else {
 672                 scrub->os_full_speed = 0;
 673                 scrub->os_partial_scan = 0;
 674         }
 675
 676         spin_lock(&scrub->os_lock);
 677         scrub->os_in_prior = 0;
 678         scrub->os_waiting = 0;
 679         scrub->os_paused = 0;
 680         scrub->os_in_join = 0;
 681         scrub->os_full_scrub = 0;
 682         spin_unlock(&scrub->os_lock);
 683         scrub->os_new_checked = 0;
 684         if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
 685                 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
 686         else if (sf->sf_pos_last_checkpoint != 0)
 687                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 688         else
 689                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 690
 691         scrub->os_pos_current = sf->sf_pos_latest_start;
 692         sf->sf_status = SS_SCANNING;
 693         sf->sf_time_latest_start = cfs_time_current_sec();
 694         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 695         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
 696         rc = osd_scrub_file_store(scrub);
 697         if (rc == 0) {
 698                 spin_lock(&scrub->os_lock);
 699                 thread_set_flags(thread, SVC_RUNNING);
 700                 spin_unlock(&scrub->os_lock);
 701                 wake_up_all(&thread->t_ctl_waitq);
 702         }
 703         up_write(&scrub->os_rwsem);
 704
 705         RETURN(rc);
 706 }
 707
 708 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
 709 {
 710         struct scrub_file *sf = &scrub->os_file;
 711         int                rc;
 712
 713         if (likely(cfs_time_before(cfs_time_current(),
 714                                    scrub->os_time_next_checkpoint) ||
 715                    scrub->os_new_checked == 0))
 716                 return 0;
 717
 718         down_write(&scrub->os_rwsem);
 719         sf->sf_items_checked += scrub->os_new_checked;
 720         scrub->os_new_checked = 0;
 721         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 722         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 723         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 724                                             scrub->os_time_last_checkpoint);
 725         rc = osd_scrub_file_store(scrub);
 726         up_write(&scrub->os_rwsem);
 727
 728         return rc;
 729 }
 730
 731 static void osd_scrub_post(struct osd_scrub *scrub, int result)
 732 {
 733         struct scrub_file *sf = &scrub->os_file;
 734         ENTRY;
 735
 736         CDEBUG(D_LFSCK, "%.16s: OI scrub post, result = %d\n",
 737                osd_scrub2name(scrub), result);
 738
 739         down_write(&scrub->os_rwsem);
 740         spin_lock(&scrub->os_lock);
 741         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 742         spin_unlock(&scrub->os_lock);
 743         if (scrub->os_new_checked > 0) {
 744                 sf->sf_items_checked += scrub->os_new_checked;
 745                 scrub->os_new_checked = 0;
 746                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 747         }
 748         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 749         if (result > 0) {
 750                 struct osd_device *dev =
 751                         container_of0(scrub, struct osd_device, od_scrub);
 752
 753                 dev->od_igif_inoi = 1;
 754                 dev->od_check_ff = 0;
 755                 sf->sf_status = SS_COMPLETED;
 756                 if (!(sf->sf_param & SP_DRYRUN)) {
 757                         memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 758                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 759                                           SF_UPGRADE | SF_AUTO);
 760                 }
 761                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 762                 sf->sf_success_count++;
 763         } else if (result == 0) {
 764                 if (scrub->os_paused)
 765                         sf->sf_status = SS_PAUSED;
 766                 else
 767                         sf->sf_status = SS_STOPPED;
 768         } else {
 769                 sf->sf_status = SS_FAILED;
 770         }
 771         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 772                                             scrub->os_time_last_checkpoint);
 773         result = osd_scrub_file_store(scrub);
 774         up_write(&scrub->os_rwsem);
 775
 776         EXIT;
 777 }
 778
 779 /* iteration engine */
 780
 781 struct osd_iit_param {
 782         struct super_block *sb;
 783         struct buffer_head *bitmap;
 784         ldiskfs_group_t bg;
 785         __u32 gbase;
 786         __u32 offset;
 787 };
 788
 789 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 790                                    struct osd_device *dev,
 791                                    struct osd_iit_param *param,
 792                                    struct osd_idmap_cache **oic,
 793                                    const bool noslot);
 794
 795 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 796                                    struct osd_device *dev,
 797                                    struct osd_iit_param *param,
 798                                    struct osd_idmap_cache *oic,
 799                                    bool *noslot, int rc);
 800
 801 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 802 {
 803         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 804                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 805         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 806                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 807                 return SCRUB_NEXT_BREAK;
 808         } else {
 809                 *pos = param->gbase + param->offset;
 810                 return 0;
 811         }
 812 }
 813
 814 /**
 815  * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
 816  * \retval 0: FID-on-MDT
 817  */
 818 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
 819                                       struct osd_device *dev,
 820                                       struct lu_fid *fid)
 821 {
 822         /* XXX: The initial OI scrub will scan the top level /O to generate
 823          *      a small local FLDB according to the <seq>. If the given FID
 824          *      is in the local FLDB, then it is FID-on-OST; otherwise it's
 825          *      quite possible for FID-on-MDT. */
 826         if (dev->od_is_ost)
 827                 return SCRUB_NEXT_OSTOBJ_OLD;
 828         else
 829                 return 0;
 830 }
 831
 832 static int osd_scrub_get_fid(struct osd_thread_info *info,
 833                              struct osd_device *dev, struct inode *inode,
 834                              struct lu_fid *fid, bool scrub)
 835 {
 836         struct lustre_mdt_attrs *lma     = &info->oti_mdt_attrs;
 837         int                      rc;
 838         bool                     has_lma = false;
 839
 840         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 841         if (rc == 0) {
 842                 has_lma = true;
 843                 if (lma->lma_compat & LMAC_NOT_IN_OI ||
 844                     lma->lma_incompat & LMAI_AGENT)
 845                         return SCRUB_NEXT_CONTINUE;
 846
 847                 *fid = lma->lma_self_fid;
 848                 if (!scrub)
 849                         return 0;
 850
 851                 if (lma->lma_compat & LMAC_FID_ON_OST)
 852                         return SCRUB_NEXT_OSTOBJ;
 853
 854                 if (fid_is_idif(fid))
 855                         return SCRUB_NEXT_OSTOBJ_OLD;
 856
 857                 /* For local object. */
 858                 if (fid_is_internal(fid))
 859                         return 0;
 860
 861                 /* For external visible MDT-object with non-normal FID. */
 862                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
 863                         return 0;
 864
 865                 /* For the object with normal FID, it may be MDT-object,
 866                  * or may be 2.4 OST-object, need further distinguish.
 867                  * Fall through to next section. */
 868         }
 869
 870         if (rc == -ENODATA || rc == 0) {
 871                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
 872                 if (rc == 0) {
 873                         if (scrub)
 874                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
 875                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 876                         return rc;
 877                 }
 878
 879                 if (rc > 0) {
 880                         if (!has_lma)
 881                                 /* It is FID-on-OST, but we do not know how
 882                                  * to generate its FID, ignore it directly. */
 883                                 rc = SCRUB_NEXT_CONTINUE;
 884                         else
 885                                 /* It is 2.4 OST-object. */
 886                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 887                         return rc;
 888                 }
 889
 890                 if (rc != -ENODATA)
 891                         return rc;
 892
 893                 if (!has_lma) {
 894                         if (dev->od_scrub.os_convert_igif) {
 895                                 lu_igif_build(fid, inode->i_ino,
 896                                               inode->i_generation);
 897                                 if (scrub)
 898                                         rc = SCRUB_NEXT_NOLMA;
 899                                 else
 900                                         rc = 0;
 901                         } else {
 902                                 /* It may be FID-on-OST, or may be FID for
 903                                  * non-MDT0, anyway, we do not know how to
 904                                  * generate its FID, ignore it directly. */
 905                                 rc = SCRUB_NEXT_CONTINUE;
 906                         }
 907                         return rc;
 908                 }
 909
 910                 /* For OI scrub case only: the object has LMA but has no ff
 911                  * (or ff crashed). It may be MDT-object, may be OST-object
 912                  * with crashed ff. The last check is local FLDB. */
 913                 rc = osd_scrub_check_local_fldb(info, dev, fid);
 914         }
 915
 916         return rc;
 917 }
 918
 919 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 920                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 921                         struct super_block *sb, bool scrub)
 922 {
 923         struct inode *inode;
 924         int           rc;
 925         ENTRY;
 926
 927         /* Not handle the backend root object and agent parent object.
 928          * They are neither visible to namespace nor have OI mappings. */
 929         if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
 930                      pos == osd_remote_parent_ino(dev)))
 931                 RETURN(SCRUB_NEXT_CONTINUE);
 932
 933         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 934         inode = osd_iget(info, dev, lid);
 935         if (IS_ERR(inode)) {
 936                 rc = PTR_ERR(inode);
 937                 /* The inode may be removed after bitmap searching, or the
 938                  * file is new created without inode initialized yet. */
 939                 if (rc == -ENOENT || rc == -ESTALE)
 940                         RETURN(SCRUB_NEXT_CONTINUE);
 941
 942                 CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: "
 943                        "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name,
 944                        pos, rc);
 945                 RETURN(rc);
 946         }
 947
 948         if (scrub &&
 949             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 950                 /* Only skip it for the first OI scrub accessing. */
 951                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 952                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
 953         }
 954
 955         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
 956
 957         GOTO(put, rc);
 958
 959 put:
 960         iput(inode);
 961         return rc;
 962 }
 963
 964 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 965                           struct osd_iit_param *param,
 966                           struct osd_idmap_cache **oic, const bool noslot)
 967 {
 968         struct osd_scrub     *scrub  = &dev->od_scrub;
 969         struct ptlrpc_thread *thread = &scrub->os_thread;
 970         struct lu_fid        *fid;
 971         struct osd_inode_id  *lid;
 972         int                   rc;
 973
 974         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
 975                 struct l_wait_info lwi;
 976
 977                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
 978                 l_wait_event(thread->t_ctl_waitq,
 979                              !list_empty(&scrub->os_inconsistent_items) ||
 980                              !thread_is_running(thread),
 981                              &lwi);
 982         }
 983
 984         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 985                 spin_lock(&scrub->os_lock);
 986                 thread_set_flags(thread, SVC_STOPPING);
 987                 spin_unlock(&scrub->os_lock);
 988                 return SCRUB_NEXT_CRASH;
 989         }
 990
 991         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
 992                 return SCRUB_NEXT_FATAL;
 993
 994         if (unlikely(!thread_is_running(thread)))
 995                 return SCRUB_NEXT_EXIT;
 996
 997         if (!list_empty(&scrub->os_inconsistent_items)) {
 998                 struct osd_inconsistent_item *oii;
 999
1000                 oii = list_entry(scrub->os_inconsistent_items.next,
1001                                  struct osd_inconsistent_item, oii_list);
1002                 *oic = &oii->oii_cache;
1003                 scrub->os_in_prior = 1;
1004                 return 0;
1005         }
1006
1007         if (noslot)
1008                 return SCRUB_NEXT_WAIT;
1009
1010         rc = osd_iit_next(param, &scrub->os_pos_current);
1011         if (rc != 0)
1012                 return rc;
1013
1014         *oic = &scrub->os_oic;
1015         fid = &(*oic)->oic_fid;
1016         lid = &(*oic)->oic_lid;
1017         rc = osd_iit_iget(info, dev, fid, lid,
1018                           scrub->os_pos_current, param->sb, true);
1019         return rc;
1020 }
1021
1022 static int osd_preload_next(struct osd_thread_info *info,
1023                             struct osd_device *dev, struct osd_iit_param *param,
1024                             struct osd_idmap_cache **oic, const bool noslot)
1025 {
1026         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
1027         struct osd_scrub        *scrub;
1028         struct ptlrpc_thread    *thread;
1029         int                      rc;
1030
1031         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1032         if (rc != 0)
1033                 return rc;
1034
1035         scrub = &dev->od_scrub;
1036         thread = &scrub->os_thread;
1037         if (thread_is_running(thread) &&
1038             ooc->ooc_pos_preload >= scrub->os_pos_current)
1039                 return SCRUB_NEXT_EXIT;
1040
1041         rc = osd_iit_iget(info, dev,
1042                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1043                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1044                           ooc->ooc_pos_preload, param->sb, false);
1045         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1046          * ignore the failure, so it still need to skip the inode next time. */
1047         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1048         return rc;
1049 }
1050
1051 static inline int
1052 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1053 {
1054         spin_lock(&scrub->os_lock);
1055         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1056             !list_empty(&scrub->os_inconsistent_items) ||
1057             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1058                 scrub->os_waiting = 0;
1059         else
1060                 scrub->os_waiting = 1;
1061         spin_unlock(&scrub->os_lock);
1062
1063         return !scrub->os_waiting;
1064 }
1065
1066 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1067                           struct osd_iit_param *param,
1068                           struct osd_idmap_cache *oic, bool *noslot, int rc)
1069 {
1070         struct l_wait_info       lwi    = { 0 };
1071         struct osd_scrub        *scrub  = &dev->od_scrub;
1072         struct scrub_file       *sf     = &scrub->os_file;
1073         struct ptlrpc_thread    *thread = &scrub->os_thread;
1074         struct osd_otable_it    *it     = dev->od_otable_it;
1075         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
1076
1077         switch (rc) {
1078         case SCRUB_NEXT_CONTINUE:
1079                 goto next;
1080         case SCRUB_NEXT_WAIT:
1081                 goto wait;
1082         case SCRUB_NEXT_NOSCRUB:
1083                 down_write(&scrub->os_rwsem);
1084                 scrub->os_new_checked++;
1085                 sf->sf_items_noscrub++;
1086                 up_write(&scrub->os_rwsem);
1087                 goto next;
1088         }
1089
1090         rc = osd_scrub_check_update(info, dev, oic, rc);
1091         if (rc != 0)
1092                 return rc;
1093
1094         rc = osd_scrub_checkpoint(scrub);
1095         if (rc != 0) {
1096                 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1097                        "rc = %d\n", osd_scrub2name(scrub),
1098                        scrub->os_pos_current, rc);
1099                 /* Continue, as long as the scrub itself can go ahead. */
1100         }
1101
1102         if (scrub->os_in_prior) {
1103                 scrub->os_in_prior = 0;
1104                 return 0;
1105         }
1106
1107 next:
1108         scrub->os_pos_current = param->gbase + ++(param->offset);
1109
1110 wait:
1111         if (it != NULL && it->ooi_waiting && ooc != NULL &&
1112             ooc->ooc_pos_preload < scrub->os_pos_current) {
1113                 spin_lock(&scrub->os_lock);
1114                 it->ooi_waiting = 0;
1115                 wake_up_all(&thread->t_ctl_waitq);
1116                 spin_unlock(&scrub->os_lock);
1117         }
1118
1119         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1120                 return 0;
1121
1122         if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1123                 *noslot = false;
1124                 return 0;
1125         }
1126
1127         if (it != NULL)
1128                 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1129                              &lwi);
1130
1131         if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1132                 *noslot = false;
1133         else
1134                 *noslot = true;
1135         return 0;
1136 }
1137
1138 static int osd_preload_exec(struct osd_thread_info *info,
1139                             struct osd_device *dev, struct osd_iit_param *param,
1140                             struct osd_idmap_cache *oic, bool *noslot, int rc)
1141 {
1142         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1143
1144         if (rc == 0) {
1145                 ooc->ooc_cached_items++;
1146                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1147                                         ~OSD_OTABLE_IT_CACHE_MASK;
1148         }
1149         return rc > 0 ? 0 : rc;
1150 }
1151
1152 #define SCRUB_IT_ALL    1
1153 #define SCRUB_IT_CRASH  2
1154
1155 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1156                            bool inconsistent)
1157 {
1158         struct osd_scrub     *scrub  = &dev->od_scrub;
1159         struct ptlrpc_thread *thread = &scrub->os_thread;
1160         struct scrub_file    *sf     = &scrub->os_file;
1161         int                   rc;
1162         ENTRY;
1163
1164         LASSERT(!(flags & SS_AUTO_PARTIAL));
1165
1166         down_write(&scrub->os_rwsem);
1167         scrub->os_in_join = 1;
1168         if (flags & SS_SET_FAILOUT)
1169                 sf->sf_param |= SP_FAILOUT;
1170         else if (flags & SS_CLEAR_FAILOUT)
1171                 sf->sf_param &= ~SP_FAILOUT;
1172
1173         if (flags & SS_SET_DRYRUN)
1174                 sf->sf_param |= SP_DRYRUN;
1175         else if (flags & SS_CLEAR_DRYRUN)
1176                 sf->sf_param &= ~SP_DRYRUN;
1177
1178         if (flags & SS_RESET) {
1179                 osd_scrub_file_reset(scrub,
1180                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1181                         inconsistent ? SF_INCONSISTENT : 0);
1182                 sf->sf_status = SS_SCANNING;
1183         }
1184
1185         if (flags & SS_AUTO_FULL) {
1186                 sf->sf_flags |= SF_AUTO;
1187                 scrub->os_full_speed = 1;
1188         }
1189
1190         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1191                 scrub->os_full_speed = 1;
1192         else
1193                 scrub->os_full_speed = 0;
1194
1195         scrub->os_new_checked = 0;
1196         if (sf->sf_pos_last_checkpoint != 0)
1197                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1198         else
1199                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1200
1201         scrub->os_pos_current = sf->sf_pos_latest_start;
1202         sf->sf_time_latest_start = cfs_time_current_sec();
1203         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1204         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1205         rc = osd_scrub_file_store(scrub);
1206         if (rc != 0)
1207                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
1208                        "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1209
1210         spin_lock(&scrub->os_lock);
1211         scrub->os_waiting = 0;
1212         scrub->os_paused = 0;
1213         scrub->os_partial_scan = 0;
1214         scrub->os_in_join = 0;
1215         scrub->os_full_scrub = 0;
1216         spin_unlock(&scrub->os_lock);
1217         wake_up_all(&thread->t_ctl_waitq);
1218         up_write(&scrub->os_rwsem);
1219
1220         EXIT;
1221 }
1222
1223 static int osd_inode_iteration(struct osd_thread_info *info,
1224                                struct osd_device *dev, __u32 max, bool preload)
1225 {
1226         struct osd_scrub     *scrub  = &dev->od_scrub;
1227         struct ptlrpc_thread *thread = &scrub->os_thread;
1228         struct scrub_file    *sf     = &scrub->os_file;
1229         osd_iit_next_policy   next;
1230         osd_iit_exec_policy   exec;
1231         __u32                *pos;
1232         __u32                *count;
1233         struct osd_iit_param  param  = { 0 };
1234         struct l_wait_info    lwi    = { 0 };
1235         __u32                 limit;
1236         int                   rc;
1237         bool                  noslot = true;
1238         ENTRY;
1239
1240         param.sb = osd_sb(dev);
1241         if (preload)
1242                 goto full;
1243
1244         while (scrub->os_partial_scan && !scrub->os_in_join) {
1245                 struct osd_idmap_cache *oic = NULL;
1246
1247                 rc = osd_scrub_next(info, dev, &param, &oic, noslot);
1248                 switch (rc) {
1249                 case SCRUB_NEXT_EXIT:
1250                         RETURN(0);
1251                 case SCRUB_NEXT_CRASH:
1252                         RETURN(SCRUB_IT_CRASH);
1253                 case SCRUB_NEXT_FATAL:
1254                         RETURN(-EINVAL);
1255                 case SCRUB_NEXT_WAIT: {
1256                         struct kstatfs *ksfs = &info->oti_ksfs;
1257                         __u64 saved_flags;
1258
1259                         if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1260                             unlikely(sf->sf_items_updated_prior == 0))
1261                                 goto wait;
1262
1263                         if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1264                             scrub->os_full_scrub) {
1265                                 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1266                                                true);
1267                                 goto full;
1268                         }
1269
1270                         rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1271                         if (rc == 0) {
1272                                 __u64 used = ksfs->f_files - ksfs->f_ffree;
1273
1274                                 do_div(used, sf->sf_items_updated_prior);
1275                                 /* If we hit too much inconsistent OI
1276                                  * mappings during the partial scan,
1277                                  * then scan the device completely. */
1278                                 if (used < dev->od_full_scrub_ratio) {
1279                                         osd_scrub_join(dev,
1280                                                 SS_AUTO_FULL | SS_RESET, true);
1281                                         goto full;
1282                                 }
1283                         }
1284
1285 wait:
1286                         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1287                             cfs_fail_val > 0)
1288                                 continue;
1289
1290                         saved_flags = sf->sf_flags;
1291                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1292                                           SF_UPGRADE | SF_AUTO);
1293                         sf->sf_status = SS_COMPLETED;
1294                         l_wait_event(thread->t_ctl_waitq,
1295                                      !thread_is_running(thread) ||
1296                                      !scrub->os_partial_scan ||
1297                                      scrub->os_in_join ||
1298                                      !list_empty(&scrub->os_inconsistent_items),
1299                                      &lwi);
1300                         sf->sf_flags = saved_flags;
1301                         sf->sf_status = SS_SCANNING;
1302
1303                         if (unlikely(!thread_is_running(thread)))
1304                                 RETURN(0);
1305
1306                         if (!scrub->os_partial_scan || scrub->os_in_join)
1307                                 goto full;
1308
1309                         continue;
1310                 }
1311                 default:
1312                         LASSERTF(rc == 0, "rc = %d\n", rc);
1313
1314                         osd_scrub_exec(info, dev, &param, oic, &noslot, rc);
1315                         break;
1316                 }
1317         }
1318
1319 full:
1320         if (!preload) {
1321                 l_wait_event(thread->t_ctl_waitq,
1322                              !thread_is_running(thread) || !scrub->os_in_join,
1323                              &lwi);
1324
1325                 if (unlikely(!thread_is_running(thread)))
1326                         RETURN(0);
1327         }
1328
1329         noslot = false;
1330         if (!preload) {
1331                 next = osd_scrub_next;
1332                 exec = osd_scrub_exec;
1333                 pos = &scrub->os_pos_current;
1334                 count = &scrub->os_new_checked;
1335         } else {
1336                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1337
1338                 next = osd_preload_next;
1339                 exec = osd_preload_exec;
1340                 pos = &ooc->ooc_pos_preload;
1341                 count = &ooc->ooc_cached_items;
1342         }
1343         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1344
1345         while (*pos <= limit && *count < max) {
1346                 struct osd_idmap_cache *oic = NULL;
1347                 struct ldiskfs_group_desc *desc;
1348
1349                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1350                 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1351                 if (desc == NULL)
1352                         RETURN(-EIO);
1353
1354                 ldiskfs_lock_group(param.sb, param.bg);
1355                 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1356                         ldiskfs_unlock_group(param.sb, param.bg);
1357                         *pos = 1 + (param.bg + 1) *
1358                                 LDISKFS_INODES_PER_GROUP(param.sb);
1359                         continue;
1360                 }
1361                 ldiskfs_unlock_group(param.sb, param.bg);
1362
1363                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1364                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1365                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1366                 if (param.bitmap == NULL) {
1367                         CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1368                                "scrub will stop, urgent mode\n",
1369                                osd_scrub2name(scrub), (__u32)param.bg);
1370                         RETURN(-EIO);
1371                 }
1372
1373                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1374                        *count < max) {
1375                         if (param.offset +
1376                                 ldiskfs_itable_unused_count(param.sb, desc) >
1377                             LDISKFS_INODES_PER_GROUP(param.sb))
1378                                 goto next_group;
1379
1380                         rc = next(info, dev, &param, &oic, noslot);
1381                         switch (rc) {
1382                         case SCRUB_NEXT_BREAK:
1383                                 goto next_group;
1384                         case SCRUB_NEXT_EXIT:
1385                                 brelse(param.bitmap);
1386                                 RETURN(0);
1387                         case SCRUB_NEXT_CRASH:
1388                                 brelse(param.bitmap);
1389                                 RETURN(SCRUB_IT_CRASH);
1390                         case SCRUB_NEXT_FATAL:
1391                                 brelse(param.bitmap);
1392                                 RETURN(-EINVAL);
1393                         }
1394
1395                         rc = exec(info, dev, &param, oic, &noslot, rc);
1396                         if (rc != 0) {
1397                                 brelse(param.bitmap);
1398                                 RETURN(rc);
1399                         }
1400                 }
1401
1402 next_group:
1403                 brelse(param.bitmap);
1404         }
1405
1406         if (*pos > limit)
1407                 RETURN(SCRUB_IT_ALL);
1408         RETURN(0);
1409 }
1410
1411 static int osd_otable_it_preload(const struct lu_env *env,
1412                                  struct osd_otable_it *it)
1413 {
1414         struct osd_device       *dev   = it->ooi_dev;
1415         struct osd_scrub        *scrub = &dev->od_scrub;
1416         struct osd_otable_cache *ooc   = &it->ooi_cache;
1417         int                      rc;
1418         ENTRY;
1419
1420         rc = osd_inode_iteration(osd_oti_get(env), dev,
1421                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1422         if (rc == SCRUB_IT_ALL)
1423                 it->ooi_all_cached = 1;
1424
1425         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1426                 spin_lock(&scrub->os_lock);
1427                 scrub->os_waiting = 0;
1428                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1429                 spin_unlock(&scrub->os_lock);
1430         }
1431
1432         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1433 }
1434
1435 static int osd_scrub_main(void *args)
1436 {
1437         struct lu_env         env;
1438         struct osd_device    *dev    = (struct osd_device *)args;
1439         struct osd_scrub     *scrub  = &dev->od_scrub;
1440         struct ptlrpc_thread *thread = &scrub->os_thread;
1441         int                   rc;
1442         ENTRY;
1443
1444         rc = lu_env_init(&env, LCT_LOCAL);
1445         if (rc != 0) {
1446                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1447                        osd_scrub2name(scrub), rc);
1448                 GOTO(noenv, rc);
1449         }
1450
1451         rc = osd_scrub_prep(dev);
1452         if (rc != 0) {
1453                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1454                        osd_scrub2name(scrub), rc);
1455                 GOTO(out, rc);
1456         }
1457
1458         if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1459                 struct l_wait_info lwi = { 0 };
1460                 struct osd_otable_it *it = dev->od_otable_it;
1461                 struct osd_otable_cache *ooc = &it->ooi_cache;
1462
1463                 l_wait_event(thread->t_ctl_waitq,
1464                              it->ooi_user_ready || !thread_is_running(thread),
1465                              &lwi);
1466                 if (unlikely(!thread_is_running(thread)))
1467                         GOTO(post, rc = 0);
1468
1469                 scrub->os_pos_current = ooc->ooc_pos_preload;
1470         }
1471
1472         CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1473                osd_scrub2name(scrub), scrub->os_start_flags,
1474                scrub->os_pos_current);
1475
1476         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1477         if (unlikely(rc == SCRUB_IT_CRASH))
1478                 GOTO(out, rc = -EINVAL);
1479         GOTO(post, rc);
1480
1481 post:
1482         osd_scrub_post(scrub, rc);
1483         CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1484                osd_scrub2name(scrub), scrub->os_pos_current, rc);
1485
1486 out:
1487         while (!list_empty(&scrub->os_inconsistent_items)) {
1488                 struct osd_inconsistent_item *oii;
1489
1490                 oii = list_entry(scrub->os_inconsistent_items.next,
1491                                      struct osd_inconsistent_item, oii_list);
1492                 list_del_init(&oii->oii_list);
1493                 OBD_FREE_PTR(oii);
1494         }
1495         lu_env_fini(&env);
1496
1497 noenv:
1498         spin_lock(&scrub->os_lock);
1499         thread_set_flags(thread, SVC_STOPPED);
1500         wake_up_all(&thread->t_ctl_waitq);
1501         spin_unlock(&scrub->os_lock);
1502         return rc;
1503 }
1504
1505 /* initial OI scrub */
1506
1507 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1508                          struct dentry *, filldir_t filldir);
1509
1510 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1511                                loff_t offset, __u64 ino, unsigned d_type);
1512 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1513                            loff_t offset, __u64 ino, unsigned d_type);
1514 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1515                            loff_t offset, __u64 ino, unsigned d_type);
1516
1517 static int
1518 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1519                      struct dentry *dentry, filldir_t filldir);
1520 static int
1521 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1522                   struct dentry *dentry, filldir_t filldir);
1523
1524 static int
1525 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1526                      struct dentry *dentry, filldir_t filldir);
1527
1528 enum osd_lf_flags {
1529         OLF_SCAN_SUBITEMS       = 0x0001,
1530         OLF_HIDE_FID            = 0x0002,
1531         OLF_SHOW_NAME           = 0x0004,
1532         OLF_NO_OI               = 0x0008,
1533 };
1534
1535 struct osd_lf_map {
1536         char            *olm_name;
1537         struct lu_fid    olm_fid;
1538         __u16            olm_flags;
1539         scandir_t        olm_scandir;
1540         filldir_t        olm_filldir;
1541 };
1542
1543 /* Add the new introduced local files in the list in the future. */
1544 static const struct osd_lf_map osd_lf_maps[] = {
1545         /* CATALOGS */
1546         { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1547                 NULL, NULL },
1548
1549         /* CONFIGS */
1550         { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1551                 OLF_SCAN_SUBITEMS, osd_ios_general_scan,
1552                 osd_ios_varfid_fill },
1553
1554         /* NIDTBL_VERSIONS */
1555         { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1556                 osd_ios_general_scan, osd_ios_varfid_fill },
1557
1558         /* PENDING */
1559         { "PENDING", { 0, 0, 0 }, 0, NULL, NULL },
1560
1561         /* ROOT */
1562         { "ROOT", { FID_SEQ_ROOT, 1, 0 },
1563                 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, osd_ios_ROOT_scan, NULL },
1564
1565         /* changelog_catalog */
1566         { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, NULL, NULL },
1567
1568         /* changelog_users */
1569         { CHANGELOG_USERS, { 0, 0, 0 }, 0, NULL, NULL },
1570
1571         /* fld */
1572         { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1573                 NULL, NULL },
1574
1575         /* last_rcvd */
1576         { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1577                 NULL, NULL },
1578
1579         /* lfsck_bookmark */
1580         { "lfsck_bookmark", { 0, 0, 0 }, 0, NULL, NULL },
1581
1582         /* lov_objid */
1583         { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1584                 NULL, NULL },
1585
1586         /* lov_objseq */
1587         { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1588                 OLF_SHOW_NAME, NULL, NULL },
1589
1590         /* quota_master */
1591         { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1592                 osd_ios_general_scan, osd_ios_varfid_fill },
1593
1594         /* quota_slave */
1595         { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1596                 osd_ios_general_scan, osd_ios_varfid_fill },
1597
1598         /* seq_ctl */
1599         { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1600                 OLF_SHOW_NAME, NULL, NULL },
1601
1602         /* seq_srv */
1603         { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1604                 OLF_SHOW_NAME, NULL, NULL },
1605
1606         /* health_check */
1607         { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1608                 OLF_SHOW_NAME, NULL, NULL },
1609
1610         /* lfsck_namespace */
1611         { "lfsck_namespace", { 0, 0, 0 }, 0, NULL, NULL },
1612
1613         /* OBJECTS, upgrade from old device */
1614         { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, osd_ios_OBJECTS_scan, NULL },
1615
1616         /* lquota_v2.user, upgrade from old device */
1617         { "lquota_v2.user", { 0, 0, 0 }, 0, NULL, NULL },
1618
1619         /* lquota_v2.group, upgrade from old device */
1620         { "lquota_v2.group", { 0, 0, 0 }, 0, NULL, NULL },
1621
1622         /* LAST_GROUP, upgrade from old device */
1623         { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1624                 OLF_SHOW_NAME, NULL, NULL },
1625
1626         /* SLAVE_LOG, llog for destroy slave stripes of striped dir */
1627         { "SLAVE_LOG", { FID_SEQ_LOCAL_FILE, SLAVE_LLOG_CATALOGS_OID, 0 },
1628                OLF_SHOW_NAME, NULL, NULL },
1629
1630         /* lost+found */
1631         { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 },
1632                 OLF_SCAN_SUBITEMS, osd_ios_general_scan, osd_ios_lf_fill },
1633
1634         { NULL, { 0, 0, 0 }, 0, NULL, NULL }
1635 };
1636
1637 /* Add the new introduced files under .lustre/ in the list in the future. */
1638 static const struct osd_lf_map osd_dl_maps[] = {
1639         /* .lustre/fid */
1640         { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1641                 NULL, NULL },
1642
1643         /* .lustre/lost+found */
1644         { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1645                 NULL, NULL },
1646
1647         { NULL, { 0, 0, 0 }, 0, NULL, NULL }
1648 };
1649
1650 struct osd_ios_item {
1651         struct list_head oii_list;
1652         struct dentry   *oii_dentry;
1653         scandir_t        oii_scandir;
1654         filldir_t        oii_filldir;
1655 };
1656
1657 struct osd_ios_filldir_buf {
1658 #ifdef HAVE_DIR_CONTEXT
1659         /* please keep it as first member */
1660         struct dir_context       ctx;
1661 #endif
1662         struct osd_thread_info  *oifb_info;
1663         struct osd_device       *oifb_dev;
1664         struct dentry           *oifb_dentry;
1665 };
1666
1667 static inline struct dentry *
1668 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1669 {
1670         struct dentry *dentry;
1671
1672         dentry = ll_lookup_one_len(name, parent, namelen);
1673         if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1674                 dput(dentry);
1675                 return ERR_PTR(-ENOENT);
1676         }
1677
1678         return dentry;
1679 }
1680
1681 static int
1682 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1683                  scandir_t scandir, filldir_t filldir)
1684 {
1685         struct osd_ios_item *item;
1686         ENTRY;
1687
1688         OBD_ALLOC_PTR(item);
1689         if (item == NULL)
1690                 RETURN(-ENOMEM);
1691
1692         INIT_LIST_HEAD(&item->oii_list);
1693         item->oii_dentry = dget(dentry);
1694         item->oii_scandir = scandir;
1695         item->oii_filldir = filldir;
1696         list_add_tail(&item->oii_list, &dev->od_ios_list);
1697
1698         RETURN(0);
1699 }
1700
1701 /**
1702  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1703  *
1704  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1705  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1706  * reference the inode, or fixed if it is missing or references another inode.
1707  */
1708 static int
1709 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1710                  struct inode *inode, const struct lu_fid *fid, int flags)
1711 {
1712         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
1713         struct osd_inode_id     *id     = &info->oti_id;
1714         struct osd_inode_id     *id2    = &info->oti_id2;
1715         struct osd_scrub        *scrub  = &dev->od_scrub;
1716         struct scrub_file       *sf     = &scrub->os_file;
1717         struct lu_fid            tfid;
1718         int                      rc;
1719         ENTRY;
1720
1721         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1722         if (rc != 0 && rc != -ENODATA) {
1723                 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1724                        "rc = %d\n", osd_name(dev), rc);
1725
1726                 RETURN(rc);
1727         }
1728
1729         osd_id_gen(id, inode->i_ino, inode->i_generation);
1730         if (rc == -ENODATA) {
1731                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID)
1732                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1733                 else
1734                         tfid = *fid;
1735                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1736                 if (rc != 0) {
1737                         CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1738                               "scrub: rc = %d\n", osd_name(dev), rc);
1739
1740                         RETURN(rc);
1741                 }
1742         } else {
1743                 if (lma->lma_compat & LMAC_NOT_IN_OI)
1744                         RETURN(0);
1745
1746                 tfid = lma->lma_self_fid;
1747         }
1748
1749         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1750         if (rc != 0) {
1751                 if (rc != -ENOENT)
1752                         RETURN(rc);
1753
1754                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1755                                                DTO_INDEX_INSERT, true, 0);
1756                 if (rc > 0)
1757                         rc = 0;
1758
1759                 RETURN(rc);
1760         }
1761
1762         if (osd_id_eq_strict(id, id2))
1763                 RETURN(0);
1764
1765         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1766                 osd_scrub_file_reset(scrub,
1767                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1768                                      SF_INCONSISTENT);
1769                 rc = osd_scrub_file_store(scrub);
1770                 if (rc != 0)
1771                         RETURN(rc);
1772         }
1773
1774         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1775                                        DTO_INDEX_UPDATE, true, 0);
1776         if (rc > 0)
1777                 rc = 0;
1778
1779         RETURN(rc);
1780 }
1781
1782 /**
1783  * It scans the /lost+found, and for the OST-object (with filter_fid
1784  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1785  */
1786 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1787                            loff_t offset, __u64 ino, unsigned d_type)
1788 {
1789         struct osd_ios_filldir_buf *fill_buf = buf;
1790         struct osd_thread_info     *info     = fill_buf->oifb_info;
1791         struct osd_device          *dev      = fill_buf->oifb_dev;
1792         struct lu_fid              *fid      = &info->oti_fid;
1793         struct osd_scrub           *scrub    = &dev->od_scrub;
1794         struct dentry              *parent   = fill_buf->oifb_dentry;
1795         struct dentry              *child;
1796         struct inode               *dir      = parent->d_inode;
1797         struct inode               *inode;
1798         int                         rc;
1799         ENTRY;
1800
1801         /* skip any '.' started names */
1802         if (name[0] == '.')
1803                 RETURN(0);
1804
1805         scrub->os_lf_scanned++;
1806         child = osd_ios_lookup_one_len(name, parent, namelen);
1807         if (IS_ERR(child)) {
1808                 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1809                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
1810                 RETURN(0);
1811         }
1812
1813         inode = child->d_inode;
1814         if (S_ISDIR(inode->i_mode)) {
1815                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1816                                       osd_ios_lf_fill);
1817                 if (rc != 0)
1818                         CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1819                               "rc = %d\n", osd_name(dev), namelen, name, rc);
1820                 GOTO(put, rc);
1821         }
1822
1823         if (!S_ISREG(inode->i_mode))
1824                 GOTO(put, rc = 0);
1825
1826         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1827         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1828                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1829                 if (rc == 0) {
1830                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1831                                "/lost+found.\n", namelen, name, PFID(fid));
1832                         scrub->os_lf_repaired++;
1833                 } else {
1834                         CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1835                                DFID": rc = %d\n",
1836                                osd_name(dev), namelen, name, PFID(fid), rc);
1837                 }
1838         }
1839
1840         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1841          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1842          *      can process them in furtuer. */
1843
1844         GOTO(put, rc);
1845
1846 put:
1847         if (rc < 0)
1848                 scrub->os_lf_failed++;
1849         dput(child);
1850         /* skip the failure to make the scanning to continue. */
1851         return 0;
1852 }
1853
1854 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1855                                loff_t offset, __u64 ino, unsigned d_type)
1856 {
1857         struct osd_ios_filldir_buf *fill_buf = buf;
1858         struct osd_device          *dev      = fill_buf->oifb_dev;
1859         struct dentry              *child;
1860         int                         rc;
1861         ENTRY;
1862
1863         /* skip any '.' started names */
1864         if (name[0] == '.')
1865                 RETURN(0);
1866
1867         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1868         if (IS_ERR(child))
1869                 RETURN(PTR_ERR(child));
1870
1871         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1872                               NULL, 0);
1873         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1874                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1875                                       osd_ios_varfid_fill);
1876         dput(child);
1877
1878         RETURN(rc);
1879 }
1880
1881 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1882                            loff_t offset, __u64 ino, unsigned d_type)
1883 {
1884         struct osd_ios_filldir_buf *fill_buf = buf;
1885         struct osd_device          *dev      = fill_buf->oifb_dev;
1886         const struct osd_lf_map    *map;
1887         struct dentry              *child;
1888         int                         rc       = 0;
1889         ENTRY;
1890
1891         /* skip any '.' started names */
1892         if (name[0] == '.')
1893                 RETURN(0);
1894
1895         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
1896                 if (strlen(map->olm_name) != namelen)
1897                         continue;
1898
1899                 if (strncmp(map->olm_name, name, namelen) == 0)
1900                         break;
1901         }
1902
1903         if (map->olm_name == NULL)
1904                 RETURN(0);
1905
1906         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1907         if (IS_ERR(child))
1908                 RETURN(PTR_ERR(child));
1909
1910         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1911                               &map->olm_fid, map->olm_flags);
1912         dput(child);
1913
1914         RETURN(rc);
1915 }
1916
1917 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
1918                              loff_t offset, __u64 ino, unsigned d_type)
1919 {
1920         struct osd_ios_filldir_buf *fill_buf = buf;
1921         struct osd_device          *dev      = fill_buf->oifb_dev;
1922         const struct osd_lf_map    *map;
1923         struct dentry              *child;
1924         int                         rc       = 0;
1925         ENTRY;
1926
1927         /* skip any '.' started names */
1928         if (name[0] == '.')
1929                 RETURN(0);
1930
1931         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
1932                 if (strlen(map->olm_name) != namelen)
1933                         continue;
1934
1935                 if (strncmp(map->olm_name, name, namelen) == 0)
1936                         break;
1937         }
1938
1939         if (map->olm_name == NULL)
1940                 RETURN(0);
1941
1942         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1943         if (IS_ERR(child))
1944                 RETURN(PTR_ERR(child));
1945
1946         if (!(map->olm_flags & OLF_NO_OI))
1947                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1948                                       &map->olm_fid, map->olm_flags);
1949         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
1950                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
1951                                       map->olm_filldir);
1952         dput(child);
1953
1954         RETURN(rc);
1955 }
1956
1957 static int
1958 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1959                      struct dentry *dentry, filldir_t filldir)
1960 {
1961         struct osd_ios_filldir_buf    buf   = {
1962 #ifdef HAVE_DIR_CONTEXT
1963                                                 .ctx.actor = filldir,
1964 #endif
1965                                                 .oifb_info = info,
1966                                                 .oifb_dev = dev,
1967                                                 .oifb_dentry = dentry };
1968         struct file                  *filp  = &info->oti_it_ea.oie_file;
1969         struct inode                 *inode = dentry->d_inode;
1970         const struct file_operations *fops  = inode->i_fop;
1971         int                           rc;
1972         ENTRY;
1973
1974         LASSERT(filldir != NULL);
1975
1976         filp->f_pos = 0;
1977         filp->f_dentry = dentry;
1978         filp->f_mode = FMODE_64BITHASH;
1979         filp->f_mapping = inode->i_mapping;
1980         filp->f_op = fops;
1981         filp->private_data = NULL;
1982         set_file_inode(filp, inode);
1983
1984 #ifdef HAVE_DIR_CONTEXT
1985         buf.ctx.pos = filp->f_pos;
1986         rc = fops->iterate(filp, &buf.ctx);
1987         filp->f_pos = buf.ctx.pos;
1988 #else
1989         rc = fops->readdir(filp, &buf, filldir);
1990 #endif
1991         fops->release(inode, filp);
1992
1993         RETURN(rc);
1994 }
1995
1996 static int
1997 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1998                   struct dentry *dentry, filldir_t filldir)
1999 {
2000         struct osd_scrub  *scrub  = &dev->od_scrub;
2001         struct scrub_file *sf     = &scrub->os_file;
2002         struct dentry     *child;
2003         int                rc;
2004         ENTRY;
2005
2006         /* It is existing MDT0 device. We only allow the case of object without
2007          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2008          * can generate IGIF mode FID for the object and related OI mapping. If
2009          * it is on other MDTs, then becuase file-level backup/restore, related
2010          * OI mapping may be invalid already, we do not know which is the right
2011          * FID for the object. We only allow IGIF objects to reside on the MDT0.
2012          *
2013          * XXX: For the case of object on non-MDT0 device with neither LMA nor
2014          *      "fid" xattr, then something crashed. We cannot re-generate the
2015          *      FID directly, instead, the OI scrub will scan the OI structure
2016          *      and try to re-generate the LMA from the OI mapping. But if the
2017          *      OI mapping crashed or lost also, then we have to give up under
2018          *      double failure cases. */
2019         scrub->os_convert_igif = 1;
2020         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2021                                        strlen(dot_lustre_name));
2022         if (IS_ERR(child)) {
2023                 rc = PTR_ERR(child);
2024                 if (rc == -ENOENT) {
2025                         /* It is 1.8 MDT device. */
2026                         if (!(sf->sf_flags & SF_UPGRADE)) {
2027                                 osd_scrub_file_reset(scrub,
2028                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2029                                         SF_UPGRADE);
2030                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2031                                 rc = osd_scrub_file_store(scrub);
2032                         } else {
2033                                 rc = 0;
2034                         }
2035                 }
2036         } else {
2037                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2038                  * so the client will get IGIF for the ".lustre" object when
2039                  * the MDT restart.
2040                  *
2041                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2042                  * it does not know whether there are some old clients cached
2043                  * the ".lustre" IGIF during the upgrading. Two choices:
2044                  *
2045                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2046                  *    It will allow the old connected clients to access the
2047                  *    ".lustre" with cached IGIF. But it will cause others
2048                  *    on the MDT failed to check "fid_is_dot_lustre()".
2049                  *
2050                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2051                  *    for ".lustre" in spite of whether there are some clients
2052                  *    cached the ".lustre" IGIF or not. It enables the check
2053                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
2054                  *    that the old connected clients cannot access the ".lustre"
2055                  *    with the cached IGIF.
2056                  *
2057                  * Usually, it is rare case for the old connected clients
2058                  * to access the ".lustre" with cached IGIF. So we prefer
2059                  * to the solution 2). */
2060                 rc = osd_ios_scan_one(info, dev, child->d_inode,
2061                                       &LU_DOT_LUSTRE_FID, 0);
2062                 if (rc == 0)
2063                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2064                                               osd_ios_dl_fill);
2065                 dput(child);
2066         }
2067
2068         RETURN(rc);
2069 }
2070
2071 static int
2072 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2073                      struct dentry *dentry, filldir_t filldir)
2074 {
2075         struct osd_scrub  *scrub  = &dev->od_scrub;
2076         struct scrub_file *sf     = &scrub->os_file;
2077         struct dentry     *child;
2078         int                rc;
2079         ENTRY;
2080
2081         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2082                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2083                 rc = osd_scrub_file_store(scrub);
2084                 if (rc != 0)
2085                         RETURN(rc);
2086         }
2087
2088         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2089         if (!IS_ERR(child)) {
2090                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2091                 dput(child);
2092         } else {
2093                 rc = PTR_ERR(child);
2094         }
2095
2096         if (rc != 0 && rc != -ENOENT)
2097                 RETURN(rc);
2098
2099         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2100         if (!IS_ERR(child)) {
2101                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2102                 dput(child);
2103         } else {
2104                 rc = PTR_ERR(child);
2105         }
2106
2107         if (rc == -ENOENT)
2108                 rc = 0;
2109
2110         RETURN(rc);
2111 }
2112
2113 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2114                                 struct osd_device *dev)
2115 {
2116         struct osd_ios_item     *item    = NULL;
2117         scandir_t                scandir = osd_ios_general_scan;
2118         filldir_t                filldir = osd_ios_root_fill;
2119         struct dentry           *dentry  = osd_sb(dev)->s_root;
2120         const struct osd_lf_map *map     = osd_lf_maps;
2121         int                      rc;
2122         ENTRY;
2123
2124         /* Lookup IGIF in OI by force for initial OI scrub. */
2125         dev->od_igif_inoi = 1;
2126
2127         while (1) {
2128                 rc = scandir(info, dev, dentry, filldir);
2129                 if (item != NULL) {
2130                         dput(item->oii_dentry);
2131                         OBD_FREE_PTR(item);
2132                 }
2133
2134                 if (rc != 0)
2135                         break;
2136
2137                 if (list_empty(&dev->od_ios_list))
2138                         break;
2139
2140                 item = list_entry(dev->od_ios_list.next,
2141                                   struct osd_ios_item, oii_list);
2142                 list_del_init(&item->oii_list);
2143
2144                 LASSERT(item->oii_scandir != NULL);
2145                 scandir = item->oii_scandir;
2146                 filldir = item->oii_filldir;
2147                 dentry = item->oii_dentry;
2148         }
2149
2150         while (!list_empty(&dev->od_ios_list)) {
2151                 item = list_entry(dev->od_ios_list.next,
2152                                   struct osd_ios_item, oii_list);
2153                 list_del_init(&item->oii_list);
2154                 dput(item->oii_dentry);
2155                 OBD_FREE_PTR(item);
2156         }
2157
2158         if (rc != 0)
2159                 RETURN(rc);
2160
2161         /* There maybe the case that the object has been removed, but its OI
2162          * mapping is still in the OI file, such as the "CATALOGS" after MDT
2163          * file-level backup/restore. So here cleanup the stale OI mappings. */
2164         while (map->olm_name != NULL) {
2165                 struct dentry *child;
2166
2167                 if (fid_is_zero(&map->olm_fid)) {
2168                         map++;
2169                         continue;
2170                 }
2171
2172                 child = osd_ios_lookup_one_len(map->olm_name,
2173                                                osd_sb(dev)->s_root,
2174                                                strlen(map->olm_name));
2175                 if (!IS_ERR(child))
2176                         dput(child);
2177                 else if (PTR_ERR(child) == -ENOENT)
2178                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2179                                                   NULL, DTO_INDEX_DELETE,
2180                                                   true, 0);
2181                 map++;
2182         }
2183
2184         RETURN(0);
2185 }
2186
2187 char *osd_lf_fid2name(const struct lu_fid *fid)
2188 {
2189         const struct osd_lf_map *map = osd_lf_maps;
2190
2191         while (map->olm_name != NULL) {
2192                 if (!lu_fid_eq(fid, &map->olm_fid)) {
2193                         map++;
2194                         continue;
2195                 }
2196
2197                 if (map->olm_flags & OLF_SHOW_NAME)
2198                         return map->olm_name;
2199                 else
2200                         return "";
2201         }
2202
2203         return NULL;
2204 }
2205
2206 /* OI scrub start/stop */
2207
2208 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2209 {
2210         struct osd_scrub     *scrub  = &dev->od_scrub;
2211         struct ptlrpc_thread *thread = &scrub->os_thread;
2212         struct l_wait_info    lwi    = { 0 };
2213         struct task_struct   *task;
2214         int                   rc;
2215         ENTRY;
2216
2217         /* os_lock: sync status between stop and scrub thread */
2218         spin_lock(&scrub->os_lock);
2219
2220 again:
2221         if (thread_is_running(thread)) {
2222                 spin_unlock(&scrub->os_lock);
2223                 if (!scrub->os_partial_scan || flags & SS_AUTO_PARTIAL)
2224                         RETURN(-EALREADY);
2225
2226                 osd_scrub_join(dev, flags, false);
2227                 spin_lock(&scrub->os_lock);
2228                 if (!thread_is_running(thread))
2229                         goto again;
2230
2231                 spin_unlock(&scrub->os_lock);
2232                 RETURN(0);
2233         }
2234
2235         if (unlikely(thread_is_stopping(thread))) {
2236                 spin_unlock(&scrub->os_lock);
2237                 l_wait_event(thread->t_ctl_waitq,
2238                              thread_is_stopped(thread),
2239                              &lwi);
2240                 spin_lock(&scrub->os_lock);
2241                 goto again;
2242         }
2243         spin_unlock(&scrub->os_lock);
2244
2245         if (scrub->os_file.sf_status == SS_COMPLETED) {
2246                 if (!(flags & SS_SET_FAILOUT))
2247                         flags |= SS_CLEAR_FAILOUT;
2248
2249                 if (!(flags & SS_SET_DRYRUN))
2250                         flags |= SS_CLEAR_DRYRUN;
2251
2252                 flags |= SS_RESET;
2253         }
2254
2255         scrub->os_start_flags = flags;
2256         thread_set_flags(thread, 0);
2257         task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2258         if (IS_ERR(task)) {
2259                 rc = PTR_ERR(task);
2260                 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2261                        osd_scrub2name(scrub), rc);
2262                 RETURN(rc);
2263         }
2264
2265         l_wait_event(thread->t_ctl_waitq,
2266                      thread_is_running(thread) || thread_is_stopped(thread),
2267                      &lwi);
2268
2269         RETURN(0);
2270 }
2271
2272 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2273 {
2274         int rc;
2275         ENTRY;
2276
2277         /* od_otable_mutex: prevent curcurrent start/stop */
2278         mutex_lock(&dev->od_otable_mutex);
2279         rc = do_osd_scrub_start(dev, flags);
2280         mutex_unlock(&dev->od_otable_mutex);
2281
2282         RETURN(rc == -EALREADY ? 0 : rc);
2283 }
2284
2285 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2286 {
2287         struct ptlrpc_thread *thread = &scrub->os_thread;
2288         struct l_wait_info    lwi    = { 0 };
2289
2290         /* os_lock: sync status between stop and scrub thread */
2291         spin_lock(&scrub->os_lock);
2292         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2293                 thread_set_flags(thread, SVC_STOPPING);
2294                 spin_unlock(&scrub->os_lock);
2295                 wake_up_all(&thread->t_ctl_waitq);
2296                 l_wait_event(thread->t_ctl_waitq,
2297                              thread_is_stopped(thread),
2298                              &lwi);
2299                 /* Do not skip the last lock/unlock, which can guarantee that
2300                  * the caller cannot return until the OI scrub thread exit. */
2301                 spin_lock(&scrub->os_lock);
2302         }
2303         spin_unlock(&scrub->os_lock);
2304 }
2305
2306 static void osd_scrub_stop(struct osd_device *dev)
2307 {
2308         /* od_otable_mutex: prevent curcurrent start/stop */
2309         mutex_lock(&dev->od_otable_mutex);
2310         dev->od_scrub.os_paused = 1;
2311         do_osd_scrub_stop(&dev->od_scrub);
2312         mutex_unlock(&dev->od_otable_mutex);
2313 }
2314
2315 /* OI scrub setup/cleanup */
2316
2317 static const char osd_scrub_name[] = "OI_scrub";
2318
2319 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2320 {
2321         struct osd_thread_info     *info   = osd_oti_get(env);
2322         struct osd_scrub           *scrub  = &dev->od_scrub;
2323         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
2324         struct scrub_file          *sf     = &scrub->os_file;
2325         struct super_block         *sb     = osd_sb(dev);
2326         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
2327         struct lvfs_run_ctxt        saved;
2328         struct file                *filp;
2329         struct inode               *inode;
2330         struct lu_fid              *fid    = &info->oti_fid;
2331         int                         dirty  = 0;
2332         int                         rc     = 0;
2333         ENTRY;
2334
2335         memset(scrub, 0, sizeof(*scrub));
2336         OBD_SET_CTXT_MAGIC(ctxt);
2337         ctxt->pwdmnt = dev->od_mnt;
2338         ctxt->pwd = dev->od_mnt->mnt_root;
2339         ctxt->fs = get_ds();
2340
2341         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2342         init_rwsem(&scrub->os_rwsem);
2343         spin_lock_init(&scrub->os_lock);
2344         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2345
2346         push_ctxt(&saved, ctxt);
2347         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2348         if (IS_ERR(filp)) {
2349                 pop_ctxt(&saved, ctxt);
2350                 RETURN(PTR_ERR(filp));
2351         }
2352
2353         inode = filp->f_dentry->d_inode;
2354         /* 'What the @fid is' is not imporatant, because the object
2355          * has no OI mapping, and only is visible inside the OSD.*/
2356         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2357         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2358         if (rc != 0) {
2359                 filp_close(filp, 0);
2360                 pop_ctxt(&saved, ctxt);
2361                 RETURN(rc);
2362         }
2363
2364         scrub->os_inode = igrab(inode);
2365         filp_close(filp, 0);
2366         pop_ctxt(&saved, ctxt);
2367
2368         rc = osd_scrub_file_load(scrub);
2369         if (rc == -ENOENT) {
2370                 osd_scrub_file_init(scrub, es->s_uuid);
2371                 /* If the "/O" dir does not exist when mount (indicated by
2372                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2373                  * then it is quite probably that the device is a new one,
2374                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2375                  *
2376                  * For the rare case that "/O" and "OI_scrub" both lost on
2377                  * an old device, it can be found and cleared later.
2378                  *
2379                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2380                  * need to check "filter_fid_old" and to convert it to
2381                  * "filter_fid" for each object, and all the IGIF should
2382                  * have their FID mapping in OI files already. */
2383                 if (dev->od_maybe_new)
2384                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2385                 dirty = 1;
2386         } else if (rc != 0) {
2387                 GOTO(cleanup_inode, rc);
2388         } else {
2389                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2390                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2391                         dirty = 1;
2392                 } else if (sf->sf_status == SS_SCANNING) {
2393                         sf->sf_status = SS_CRASHED;
2394                         dirty = 1;
2395                 }
2396         }
2397
2398         if (sf->sf_pos_last_checkpoint != 0)
2399                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2400         else
2401                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2402
2403         if (dirty != 0) {
2404                 rc = osd_scrub_file_store(scrub);
2405                 if (rc != 0)
2406                         GOTO(cleanup_inode, rc);
2407         }
2408
2409         /* Initialize OI files. */
2410         rc = osd_oi_init(info, dev);
2411         if (rc < 0)
2412                 GOTO(cleanup_inode, rc);
2413
2414         rc = osd_initial_OI_scrub(info, dev);
2415         if (rc != 0)
2416                 GOTO(cleanup_oi, rc);
2417
2418         if (sf->sf_flags & SF_UPGRADE ||
2419             !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2420               sf->sf_success_count > 0)) {
2421                 dev->od_igif_inoi = 0;
2422                 dev->od_check_ff = dev->od_is_ost;
2423         } else {
2424                 dev->od_igif_inoi = 1;
2425                 dev->od_check_ff = 0;
2426         }
2427
2428         if (sf->sf_flags & SF_INCONSISTENT)
2429                 /* The 'od_igif_inoi' will be set under the
2430                  * following cases:
2431                  * 1) new created system, or
2432                  * 2) restored from file-level backup, or
2433                  * 3) the upgrading completed.
2434                  *
2435                  * The 'od_igif_inoi' may be cleared by OI scrub
2436                  * later if found that the system is upgrading. */
2437                 dev->od_igif_inoi = 1;
2438
2439         if (!dev->od_noscrub &&
2440             ((sf->sf_status == SS_PAUSED) ||
2441              (sf->sf_status == SS_CRASHED &&
2442               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2443                               SF_UPGRADE | SF_AUTO)) ||
2444              (sf->sf_status == SS_INIT &&
2445               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2446                               SF_UPGRADE))))
2447                 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2448
2449         if (rc != 0)
2450                 GOTO(cleanup_oi, rc);
2451
2452         /* it is possible that dcache entries may keep objects after they are
2453          * deleted by OSD. While it looks safe this can cause object data to
2454          * stay until umount causing failures in tests calculating free space,
2455          * e.g. replay-ost-single. Since those dcache entries are not used
2456          * anymore let's just free them after use here */
2457         shrink_dcache_sb(sb);
2458
2459         RETURN(0);
2460 cleanup_oi:
2461         osd_oi_fini(info, dev);
2462 cleanup_inode:
2463         iput(scrub->os_inode);
2464         scrub->os_inode = NULL;
2465
2466         return rc;
2467 }
2468
2469 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2470 {
2471         struct osd_scrub *scrub = &dev->od_scrub;
2472
2473         LASSERT(dev->od_otable_it == NULL);
2474
2475         if (scrub->os_inode != NULL) {
2476                 osd_scrub_stop(dev);
2477                 iput(scrub->os_inode);
2478                 scrub->os_inode = NULL;
2479         }
2480         if (dev->od_oi_table != NULL)
2481                 osd_oi_fini(osd_oti_get(env), dev);
2482 }
2483
2484 /* object table based iteration APIs */
2485
2486 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2487                                        struct dt_object *dt, __u32 attr,
2488                                        struct lustre_capa *capa)
2489 {
2490         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2491         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2492         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2493         struct osd_scrub       *scrub = &dev->od_scrub;
2494         struct osd_otable_it   *it;
2495         __u32                   start = 0;
2496         int                     rc;
2497         ENTRY;
2498
2499         /* od_otable_mutex: prevent curcurrent init/fini */
2500         mutex_lock(&dev->od_otable_mutex);
2501         if (dev->od_otable_it != NULL)
2502                 GOTO(out, it = ERR_PTR(-EALREADY));
2503
2504         OBD_ALLOC_PTR(it);
2505         if (it == NULL)
2506                 GOTO(out, it = ERR_PTR(-ENOMEM));
2507
2508         dev->od_otable_it = it;
2509         it->ooi_dev = dev;
2510         it->ooi_cache.ooc_consumer_idx = -1;
2511         if (flags & DOIF_OUTUSED)
2512                 it->ooi_used_outside = 1;
2513
2514         if (flags & DOIF_RESET)
2515                 start |= SS_RESET;
2516
2517         if (valid & DOIV_ERROR_HANDLE) {
2518                 if (flags & DOIF_FAILOUT)
2519                         start |= SS_SET_FAILOUT;
2520                 else
2521                         start |= SS_CLEAR_FAILOUT;
2522         }
2523
2524         if (valid & DOIV_DRYRUN) {
2525                 if (flags & DOIF_DRYRUN)
2526                         start |= SS_SET_DRYRUN;
2527                 else
2528                         start |= SS_CLEAR_DRYRUN;
2529         }
2530
2531         rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2532         if (rc < 0 && rc != -EALREADY) {
2533                 dev->od_otable_it = NULL;
2534                 OBD_FREE_PTR(it);
2535                 GOTO(out, it = ERR_PTR(rc));
2536         }
2537
2538         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2539
2540         GOTO(out, it);
2541
2542 out:
2543         mutex_unlock(&dev->od_otable_mutex);
2544         return (struct dt_it *)it;
2545 }
2546
2547 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2548 {
2549         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2550         struct osd_device    *dev = it->ooi_dev;
2551
2552         /* od_otable_mutex: prevent curcurrent init/fini */
2553         mutex_lock(&dev->od_otable_mutex);
2554         do_osd_scrub_stop(&dev->od_scrub);
2555         LASSERT(dev->od_otable_it == it);
2556
2557         dev->od_otable_it = NULL;
2558         mutex_unlock(&dev->od_otable_mutex);
2559         OBD_FREE_PTR(it);
2560 }
2561
2562 static int osd_otable_it_get(const struct lu_env *env,
2563                              struct dt_it *di, const struct dt_key *key)
2564 {
2565         return 0;
2566 }
2567
2568 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2569 {
2570 }
2571
2572 static inline int
2573 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2574 {
2575         spin_lock(&scrub->os_lock);
2576         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2577             scrub->os_waiting ||
2578             !thread_is_running(&scrub->os_thread))
2579                 it->ooi_waiting = 0;
2580         else
2581                 it->ooi_waiting = 1;
2582         spin_unlock(&scrub->os_lock);
2583
2584         return !it->ooi_waiting;
2585 }
2586
2587 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2588 {
2589         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2590         struct osd_device       *dev    = it->ooi_dev;
2591         struct osd_scrub        *scrub  = &dev->od_scrub;
2592         struct osd_otable_cache *ooc    = &it->ooi_cache;
2593         struct ptlrpc_thread    *thread = &scrub->os_thread;
2594         struct l_wait_info       lwi    = { 0 };
2595         int                      rc;
2596         ENTRY;
2597
2598         LASSERT(it->ooi_user_ready);
2599
2600 again:
2601         if (!thread_is_running(thread) && !it->ooi_used_outside)
2602                 RETURN(1);
2603
2604         if (ooc->ooc_cached_items > 0) {
2605                 ooc->ooc_cached_items--;
2606                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2607                                         ~OSD_OTABLE_IT_CACHE_MASK;
2608                 RETURN(0);
2609         }
2610
2611         if (it->ooi_all_cached) {
2612                 l_wait_event(thread->t_ctl_waitq,
2613                              !thread_is_running(thread),
2614                              &lwi);
2615                 RETURN(1);
2616         }
2617
2618         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2619                 spin_lock(&scrub->os_lock);
2620                 scrub->os_waiting = 0;
2621                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2622                 spin_unlock(&scrub->os_lock);
2623         }
2624
2625         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2626                 l_wait_event(thread->t_ctl_waitq,
2627                              osd_otable_it_wakeup(scrub, it),
2628                              &lwi);
2629
2630         if (!thread_is_running(thread) && !it->ooi_used_outside)
2631                 RETURN(1);
2632
2633         rc = osd_otable_it_preload(env, it);
2634         if (rc >= 0)
2635                 goto again;
2636
2637         RETURN(rc);
2638 }
2639
2640 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2641                                         const struct dt_it *di)
2642 {
2643         return NULL;
2644 }
2645
2646 static int osd_otable_it_key_size(const struct lu_env *env,
2647                                   const struct dt_it *di)
2648 {
2649         return sizeof(__u64);
2650 }
2651
2652 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2653                              struct dt_rec *rec, __u32 attr)
2654 {
2655         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2656         struct osd_otable_cache *ooc = &it->ooi_cache;
2657
2658         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2659
2660         /* Filter out Invald FID already. */
2661         LASSERTF(fid_is_sane((struct lu_fid *)rec),
2662                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2663                  PFID((struct lu_fid *)rec),
2664                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2665
2666         return 0;
2667 }
2668
2669 static __u64 osd_otable_it_store(const struct lu_env *env,
2670                                  const struct dt_it *di)
2671 {
2672         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2673         struct osd_otable_cache *ooc = &it->ooi_cache;
2674         __u64                    hash;
2675
2676         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2677                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2678         else
2679                 hash = ooc->ooc_pos_preload;
2680         return hash;
2681 }
2682
2683 /**
2684  * Set the OSD layer iteration start position as the specified hash.
2685  */
2686 static int osd_otable_it_load(const struct lu_env *env,
2687                               const struct dt_it *di, __u64 hash)
2688 {
2689         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
2690         struct osd_device       *dev   = it->ooi_dev;
2691         struct osd_otable_cache *ooc   = &it->ooi_cache;
2692         struct osd_scrub        *scrub = &dev->od_scrub;
2693         int                      rc;
2694         ENTRY;
2695
2696         /* Forbid to set iteration position after iteration started. */
2697         if (it->ooi_user_ready)
2698                 RETURN(-EPERM);
2699
2700         if (hash > OSD_OTABLE_MAX_HASH)
2701                 hash = OSD_OTABLE_MAX_HASH;
2702
2703         ooc->ooc_pos_preload = hash;
2704         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2705                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2706
2707         it->ooi_user_ready = 1;
2708         if (!scrub->os_full_speed)
2709                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2710
2711         /* Unplug OSD layer iteration by the first next() call. */
2712         rc = osd_otable_it_next(env, (struct dt_it *)it);
2713
2714         RETURN(rc);
2715 }
2716
2717 static int osd_otable_it_key_rec(const struct lu_env *env,
2718                                  const struct dt_it *di, void *key_rec)
2719 {
2720         return 0;
2721 }
2722
2723 const struct dt_index_operations osd_otable_ops = {
2724         .dio_it = {
2725                 .init     = osd_otable_it_init,
2726                 .fini     = osd_otable_it_fini,
2727                 .get      = osd_otable_it_get,
2728                 .put      = osd_otable_it_put,
2729                 .next     = osd_otable_it_next,
2730                 .key      = osd_otable_it_key,
2731                 .key_size = osd_otable_it_key_size,
2732                 .rec      = osd_otable_it_rec,
2733                 .store    = osd_otable_it_store,
2734                 .load     = osd_otable_it_load,
2735                 .key_rec  = osd_otable_it_key_rec,
2736         }
2737 };
2738
2739 /* high priority inconsistent items list APIs */
2740
2741 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL  60
2742
2743 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2744                    int insert)
2745 {
2746         struct osd_inconsistent_item *oii;
2747         struct osd_scrub             *scrub  = &dev->od_scrub;
2748         struct ptlrpc_thread         *thread = &scrub->os_thread;
2749         int                           wakeup = 0;
2750         ENTRY;
2751
2752         OBD_ALLOC_PTR(oii);
2753         if (unlikely(oii == NULL))
2754                 RETURN(-ENOMEM);
2755
2756         INIT_LIST_HEAD(&oii->oii_list);
2757         oii->oii_cache = *oic;
2758         oii->oii_insert = insert;
2759
2760         if (scrub->os_partial_scan) {
2761                 __u64 now = cfs_time_current_sec();
2762
2763                 /* If there haven't been errors in a long time,
2764                  * decay old count until either the errors are
2765                  * gone or we reach the current interval. */
2766                 while (unlikely(scrub->os_bad_oimap_count > 0 &&
2767                                 scrub->os_bad_oimap_time +
2768                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
2769                         scrub->os_bad_oimap_count >>= 1;
2770                         scrub->os_bad_oimap_time +=
2771                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
2772                 }
2773
2774                 scrub->os_bad_oimap_time = now;
2775                 if (++scrub->os_bad_oimap_count > dev->od_full_scrub_speed)
2776                         scrub->os_full_scrub = 1;
2777         }
2778
2779         spin_lock(&scrub->os_lock);
2780         if (unlikely(!thread_is_running(thread))) {
2781                 spin_unlock(&scrub->os_lock);
2782                 OBD_FREE_PTR(oii);
2783                 RETURN(-EAGAIN);
2784         }
2785
2786         if (list_empty(&scrub->os_inconsistent_items))
2787                 wakeup = 1;
2788         list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2789         spin_unlock(&scrub->os_lock);
2790
2791         if (wakeup != 0)
2792                 wake_up_all(&thread->t_ctl_waitq);
2793
2794         RETURN(0);
2795 }
2796
2797 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2798                    struct osd_inode_id *id)
2799 {
2800         struct osd_scrub             *scrub = &dev->od_scrub;
2801         struct osd_inconsistent_item *oii;
2802         ENTRY;
2803
2804         spin_lock(&scrub->os_lock);
2805         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2806                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2807                         *id = oii->oii_cache.oic_lid;
2808                         spin_unlock(&scrub->os_lock);
2809                         RETURN(0);
2810                 }
2811         }
2812         spin_unlock(&scrub->os_lock);
2813
2814         RETURN(-ENOENT);
2815 }
2816
2817 /* OI scrub dump */
2818
2819 static const char *scrub_status_names[] = {
2820         "init",
2821         "scanning",
2822         "completed",
2823         "failed",
2824         "stopped",
2825         "paused",
2826         "crashed",
2827         NULL
2828 };
2829
2830 static const char *scrub_flags_names[] = {
2831         "recreated",
2832         "inconsistent",
2833         "auto",
2834         "upgrade",
2835         NULL
2836 };
2837
2838 static const char *scrub_param_names[] = {
2839         "failout",
2840         "dryrun",
2841         NULL
2842 };
2843
2844 static int scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
2845                            const char *prefix)
2846 {
2847         int flag;
2848         int rc;
2849         int i;
2850
2851         rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2852         if (rc < 0)
2853                 return rc;
2854
2855         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2856                 if (flag & bits) {
2857                         bits &= ~flag;
2858                         rc = seq_printf(m, "%s%c", names[i],
2859                                         bits != 0 ? ',' : '\n');
2860                         if (rc < 0)
2861                                 return rc;
2862                 }
2863         }
2864         return 0;
2865 }
2866
2867 static int scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
2868 {
2869         int rc;
2870
2871         if (time != 0)
2872                 rc = seq_printf(m, "%s: "LPU64" seconds\n", prefix,
2873                               cfs_time_current_sec() - time);
2874         else
2875                 rc = seq_printf(m, "%s: N/A\n", prefix);
2876         return rc;
2877 }
2878
2879 static int scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
2880 {
2881         int rc;
2882
2883         if (pos != 0)
2884                 rc = seq_printf(m, "%s: "LPU64"\n", prefix, pos);
2885         else
2886                 rc = seq_printf(m, "%s: N/A\n", prefix);
2887         return rc;
2888 }
2889
2890 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
2891 {
2892         struct osd_scrub  *scrub   = &dev->od_scrub;
2893         struct scrub_file *sf      = &scrub->os_file;
2894         __u64              checked;
2895         __u64              speed;
2896         int                rc;
2897
2898         down_read(&scrub->os_rwsem);
2899         rc = seq_printf(m, "name: OI_scrub\n"
2900                         "magic: 0x%x\n"
2901                         "oi_files: %d\n"
2902                         "status: %s\n",
2903                         sf->sf_magic, (int)sf->sf_oi_count,
2904                         scrub_status_names[sf->sf_status]);
2905         if (rc < 0)
2906                 goto out;
2907
2908         rc = scrub_bits_dump(m, sf->sf_flags, scrub_flags_names,
2909                              "flags");
2910         if (rc < 0)
2911                 goto out;
2912
2913         rc = scrub_bits_dump(m, sf->sf_param, scrub_param_names,
2914                              "param");
2915         if (rc < 0)
2916                 goto out;
2917
2918         rc = scrub_time_dump(m, sf->sf_time_last_complete,
2919                              "time_since_last_completed");
2920         if (rc < 0)
2921                 goto out;
2922
2923         rc = scrub_time_dump(m, sf->sf_time_latest_start,
2924                              "time_since_latest_start");
2925         if (rc < 0)
2926                 goto out;
2927
2928         rc = scrub_time_dump(m, sf->sf_time_last_checkpoint,
2929                              "time_since_last_checkpoint");
2930         if (rc < 0)
2931                 goto out;
2932
2933         rc = scrub_pos_dump(m, sf->sf_pos_latest_start,
2934                             "latest_start_position");
2935         if (rc < 0)
2936                 goto out;
2937
2938         rc = scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
2939                             "last_checkpoint_position");
2940         if (rc < 0)
2941                 goto out;
2942
2943         rc = scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
2944                             "first_failure_position");
2945         if (rc < 0)
2946                 goto out;
2947
2948         checked = sf->sf_items_checked + scrub->os_new_checked;
2949         rc = seq_printf(m, "checked: "LPU64"\n"
2950                       "updated: "LPU64"\n"
2951                       "failed: "LPU64"\n"
2952                       "prior_updated: "LPU64"\n"
2953                       "noscrub: "LPU64"\n"
2954                       "igif: "LPU64"\n"
2955                       "success_count: %u\n",
2956                       checked, sf->sf_items_updated, sf->sf_items_failed,
2957                       sf->sf_items_updated_prior, sf->sf_items_noscrub,
2958                       sf->sf_items_igif, sf->sf_success_count);
2959         if (rc < 0)
2960                 goto out;
2961
2962         speed = checked;
2963         if (thread_is_running(&scrub->os_thread)) {
2964                 cfs_duration_t duration = cfs_time_current() -
2965                                           scrub->os_time_last_checkpoint;
2966                 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
2967                                                      MSEC_PER_SEC);
2968                 __u32 rtime = sf->sf_run_time +
2969                               cfs_duration_sec(duration + HALF_SEC);
2970
2971                 if (duration != 0)
2972                         do_div(new_checked, duration);
2973                 if (rtime != 0)
2974                         do_div(speed, rtime);
2975                 rc = seq_printf(m, "run_time: %u seconds\n"
2976                               "average_speed: "LPU64" objects/sec\n"
2977                               "real-time_speed: "LPU64" objects/sec\n"
2978                               "current_position: %u\n"
2979                               "lf_scanned: "LPU64"\n"
2980                               "lf_reparied: "LPU64"\n"
2981                               "lf_failed: "LPU64"\n",
2982                               rtime, speed, new_checked, scrub->os_pos_current,
2983                               scrub->os_lf_scanned, scrub->os_lf_repaired,
2984                               scrub->os_lf_failed);
2985         } else {
2986                 if (sf->sf_run_time != 0)
2987                         do_div(speed, sf->sf_run_time);
2988                 rc = seq_printf(m, "run_time: %u seconds\n"
2989                               "average_speed: "LPU64" objects/sec\n"
2990                               "real-time_speed: N/A\n"
2991                               "current_position: N/A\n"
2992                               "lf_scanned: "LPU64"\n"
2993                               "lf_reparied: "LPU64"\n"
2994                               "lf_failed: "LPU64"\n",
2995                               sf->sf_run_time, speed, scrub->os_lf_scanned,
2996                               scrub->os_lf_repaired, scrub->os_lf_failed);
2997         }
2998
2999 out:
3000         up_read(&scrub->os_rwsem);
3001         return (rc < 0 ? -ENOSPC : 0);
3002 }