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                 if (likely(lwi.lwi_timeout > 0))
 979                         l_wait_event(thread->t_ctl_waitq,
 980                                 !list_empty(&scrub->os_inconsistent_items) ||
 981                                 !thread_is_running(thread),
 982                                 &lwi);
 983         }
 984
 985         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 986                 spin_lock(&scrub->os_lock);
 987                 thread_set_flags(thread, SVC_STOPPING);
 988                 spin_unlock(&scrub->os_lock);
 989                 return SCRUB_NEXT_CRASH;
 990         }
 991
 992         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
 993                 return SCRUB_NEXT_FATAL;
 994
 995         if (unlikely(!thread_is_running(thread)))
 996                 return SCRUB_NEXT_EXIT;
 997
 998         if (!list_empty(&scrub->os_inconsistent_items)) {
 999                 struct osd_inconsistent_item *oii;
1000
1001                 oii = list_entry(scrub->os_inconsistent_items.next,
1002                                  struct osd_inconsistent_item, oii_list);
1003                 *oic = &oii->oii_cache;
1004                 scrub->os_in_prior = 1;
1005                 return 0;
1006         }
1007
1008         if (noslot)
1009                 return SCRUB_NEXT_WAIT;
1010
1011         rc = osd_iit_next(param, &scrub->os_pos_current);
1012         if (rc != 0)
1013                 return rc;
1014
1015         *oic = &scrub->os_oic;
1016         fid = &(*oic)->oic_fid;
1017         lid = &(*oic)->oic_lid;
1018         rc = osd_iit_iget(info, dev, fid, lid,
1019                           scrub->os_pos_current, param->sb, true);
1020         return rc;
1021 }
1022
1023 static int osd_preload_next(struct osd_thread_info *info,
1024                             struct osd_device *dev, struct osd_iit_param *param,
1025                             struct osd_idmap_cache **oic, const bool noslot)
1026 {
1027         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
1028         struct osd_scrub        *scrub;
1029         struct ptlrpc_thread    *thread;
1030         int                      rc;
1031
1032         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1033         if (rc != 0)
1034                 return rc;
1035
1036         scrub = &dev->od_scrub;
1037         thread = &scrub->os_thread;
1038         if (thread_is_running(thread) &&
1039             ooc->ooc_pos_preload >= scrub->os_pos_current)
1040                 return SCRUB_NEXT_EXIT;
1041
1042         rc = osd_iit_iget(info, dev,
1043                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1044                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1045                           ooc->ooc_pos_preload, param->sb, false);
1046         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1047          * ignore the failure, so it still need to skip the inode next time. */
1048         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1049         return rc;
1050 }
1051
1052 static inline int
1053 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1054 {
1055         spin_lock(&scrub->os_lock);
1056         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1057             !list_empty(&scrub->os_inconsistent_items) ||
1058             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1059                 scrub->os_waiting = 0;
1060         else
1061                 scrub->os_waiting = 1;
1062         spin_unlock(&scrub->os_lock);
1063
1064         return !scrub->os_waiting;
1065 }
1066
1067 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1068                           struct osd_iit_param *param,
1069                           struct osd_idmap_cache *oic, bool *noslot, int rc)
1070 {
1071         struct l_wait_info       lwi    = { 0 };
1072         struct osd_scrub        *scrub  = &dev->od_scrub;
1073         struct scrub_file       *sf     = &scrub->os_file;
1074         struct ptlrpc_thread    *thread = &scrub->os_thread;
1075         struct osd_otable_it    *it     = dev->od_otable_it;
1076         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
1077
1078         switch (rc) {
1079         case SCRUB_NEXT_CONTINUE:
1080                 goto next;
1081         case SCRUB_NEXT_WAIT:
1082                 goto wait;
1083         case SCRUB_NEXT_NOSCRUB:
1084                 down_write(&scrub->os_rwsem);
1085                 scrub->os_new_checked++;
1086                 sf->sf_items_noscrub++;
1087                 up_write(&scrub->os_rwsem);
1088                 goto next;
1089         }
1090
1091         rc = osd_scrub_check_update(info, dev, oic, rc);
1092         if (rc != 0)
1093                 return rc;
1094
1095         rc = osd_scrub_checkpoint(scrub);
1096         if (rc != 0) {
1097                 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1098                        "rc = %d\n", osd_scrub2name(scrub),
1099                        scrub->os_pos_current, rc);
1100                 /* Continue, as long as the scrub itself can go ahead. */
1101         }
1102
1103         if (scrub->os_in_prior) {
1104                 scrub->os_in_prior = 0;
1105                 return 0;
1106         }
1107
1108 next:
1109         scrub->os_pos_current = param->gbase + ++(param->offset);
1110
1111 wait:
1112         if (it != NULL && it->ooi_waiting && ooc != NULL &&
1113             ooc->ooc_pos_preload < scrub->os_pos_current) {
1114                 spin_lock(&scrub->os_lock);
1115                 it->ooi_waiting = 0;
1116                 wake_up_all(&thread->t_ctl_waitq);
1117                 spin_unlock(&scrub->os_lock);
1118         }
1119
1120         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1121                 return 0;
1122
1123         if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1124                 *noslot = false;
1125                 return 0;
1126         }
1127
1128         if (it != NULL)
1129                 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1130                              &lwi);
1131
1132         if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1133                 *noslot = false;
1134         else
1135                 *noslot = true;
1136         return 0;
1137 }
1138
1139 static int osd_preload_exec(struct osd_thread_info *info,
1140                             struct osd_device *dev, struct osd_iit_param *param,
1141                             struct osd_idmap_cache *oic, bool *noslot, int rc)
1142 {
1143         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1144
1145         if (rc == 0) {
1146                 ooc->ooc_cached_items++;
1147                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1148                                         ~OSD_OTABLE_IT_CACHE_MASK;
1149         }
1150         return rc > 0 ? 0 : rc;
1151 }
1152
1153 #define SCRUB_IT_ALL    1
1154 #define SCRUB_IT_CRASH  2
1155
1156 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1157                            bool inconsistent)
1158 {
1159         struct osd_scrub     *scrub  = &dev->od_scrub;
1160         struct ptlrpc_thread *thread = &scrub->os_thread;
1161         struct scrub_file    *sf     = &scrub->os_file;
1162         int                   rc;
1163         ENTRY;
1164
1165         LASSERT(!(flags & SS_AUTO_PARTIAL));
1166
1167         down_write(&scrub->os_rwsem);
1168         scrub->os_in_join = 1;
1169         if (flags & SS_SET_FAILOUT)
1170                 sf->sf_param |= SP_FAILOUT;
1171         else if (flags & SS_CLEAR_FAILOUT)
1172                 sf->sf_param &= ~SP_FAILOUT;
1173
1174         if (flags & SS_SET_DRYRUN)
1175                 sf->sf_param |= SP_DRYRUN;
1176         else if (flags & SS_CLEAR_DRYRUN)
1177                 sf->sf_param &= ~SP_DRYRUN;
1178
1179         if (flags & SS_RESET) {
1180                 osd_scrub_file_reset(scrub,
1181                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1182                         inconsistent ? SF_INCONSISTENT : 0);
1183                 sf->sf_status = SS_SCANNING;
1184         }
1185
1186         if (flags & SS_AUTO_FULL) {
1187                 sf->sf_flags |= SF_AUTO;
1188                 scrub->os_full_speed = 1;
1189         }
1190
1191         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1192                 scrub->os_full_speed = 1;
1193         else
1194                 scrub->os_full_speed = 0;
1195
1196         scrub->os_new_checked = 0;
1197         if (sf->sf_pos_last_checkpoint != 0)
1198                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1199         else
1200                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1201
1202         scrub->os_pos_current = sf->sf_pos_latest_start;
1203         sf->sf_time_latest_start = cfs_time_current_sec();
1204         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1205         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1206         rc = osd_scrub_file_store(scrub);
1207         if (rc != 0)
1208                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
1209                        "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1210
1211         spin_lock(&scrub->os_lock);
1212         scrub->os_waiting = 0;
1213         scrub->os_paused = 0;
1214         scrub->os_partial_scan = 0;
1215         scrub->os_in_join = 0;
1216         scrub->os_full_scrub = 0;
1217         spin_unlock(&scrub->os_lock);
1218         wake_up_all(&thread->t_ctl_waitq);
1219         up_write(&scrub->os_rwsem);
1220
1221         EXIT;
1222 }
1223
1224 static int osd_inode_iteration(struct osd_thread_info *info,
1225                                struct osd_device *dev, __u32 max, bool preload)
1226 {
1227         struct osd_scrub     *scrub  = &dev->od_scrub;
1228         struct ptlrpc_thread *thread = &scrub->os_thread;
1229         struct scrub_file    *sf     = &scrub->os_file;
1230         osd_iit_next_policy   next;
1231         osd_iit_exec_policy   exec;
1232         __u32                *pos;
1233         __u32                *count;
1234         struct osd_iit_param  param  = { NULL };
1235         struct l_wait_info    lwi    = { 0 };
1236         __u32                 limit;
1237         int                   rc;
1238         bool                  noslot = true;
1239         ENTRY;
1240
1241         param.sb = osd_sb(dev);
1242         if (preload)
1243                 goto full;
1244
1245         while (scrub->os_partial_scan && !scrub->os_in_join) {
1246                 struct osd_idmap_cache *oic = NULL;
1247
1248                 rc = osd_scrub_next(info, dev, &param, &oic, noslot);
1249                 switch (rc) {
1250                 case SCRUB_NEXT_EXIT:
1251                         RETURN(0);
1252                 case SCRUB_NEXT_CRASH:
1253                         RETURN(SCRUB_IT_CRASH);
1254                 case SCRUB_NEXT_FATAL:
1255                         RETURN(-EINVAL);
1256                 case SCRUB_NEXT_WAIT: {
1257                         struct kstatfs *ksfs = &info->oti_ksfs;
1258                         __u64 saved_flags;
1259
1260                         if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1261                             unlikely(sf->sf_items_updated_prior == 0))
1262                                 goto wait;
1263
1264                         if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1265                             scrub->os_full_scrub) {
1266                                 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1267                                                true);
1268                                 goto full;
1269                         }
1270
1271                         rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1272                         if (rc == 0) {
1273                                 __u64 used = ksfs->f_files - ksfs->f_ffree;
1274
1275                                 do_div(used, sf->sf_items_updated_prior);
1276                                 /* If we hit too much inconsistent OI
1277                                  * mappings during the partial scan,
1278                                  * then scan the device completely. */
1279                                 if (used < dev->od_full_scrub_ratio) {
1280                                         osd_scrub_join(dev,
1281                                                 SS_AUTO_FULL | SS_RESET, true);
1282                                         goto full;
1283                                 }
1284                         }
1285
1286 wait:
1287                         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1288                             cfs_fail_val > 0)
1289                                 continue;
1290
1291                         saved_flags = sf->sf_flags;
1292                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1293                                           SF_UPGRADE | SF_AUTO);
1294                         sf->sf_status = SS_COMPLETED;
1295                         l_wait_event(thread->t_ctl_waitq,
1296                                      !thread_is_running(thread) ||
1297                                      !scrub->os_partial_scan ||
1298                                      scrub->os_in_join ||
1299                                      !list_empty(&scrub->os_inconsistent_items),
1300                                      &lwi);
1301                         sf->sf_flags = saved_flags;
1302                         sf->sf_status = SS_SCANNING;
1303
1304                         if (unlikely(!thread_is_running(thread)))
1305                                 RETURN(0);
1306
1307                         if (!scrub->os_partial_scan || scrub->os_in_join)
1308                                 goto full;
1309
1310                         continue;
1311                 }
1312                 default:
1313                         LASSERTF(rc == 0, "rc = %d\n", rc);
1314
1315                         osd_scrub_exec(info, dev, &param, oic, &noslot, rc);
1316                         break;
1317                 }
1318         }
1319
1320 full:
1321         if (!preload) {
1322                 l_wait_event(thread->t_ctl_waitq,
1323                              !thread_is_running(thread) || !scrub->os_in_join,
1324                              &lwi);
1325
1326                 if (unlikely(!thread_is_running(thread)))
1327                         RETURN(0);
1328         }
1329
1330         noslot = false;
1331         if (!preload) {
1332                 next = osd_scrub_next;
1333                 exec = osd_scrub_exec;
1334                 pos = &scrub->os_pos_current;
1335                 count = &scrub->os_new_checked;
1336         } else {
1337                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1338
1339                 next = osd_preload_next;
1340                 exec = osd_preload_exec;
1341                 pos = &ooc->ooc_pos_preload;
1342                 count = &ooc->ooc_cached_items;
1343         }
1344         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1345
1346         while (*pos <= limit && *count < max) {
1347                 struct osd_idmap_cache *oic = NULL;
1348                 struct ldiskfs_group_desc *desc;
1349
1350                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1351                 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1352                 if (desc == NULL)
1353                         RETURN(-EIO);
1354
1355                 ldiskfs_lock_group(param.sb, param.bg);
1356                 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1357                         ldiskfs_unlock_group(param.sb, param.bg);
1358                         *pos = 1 + (param.bg + 1) *
1359                                 LDISKFS_INODES_PER_GROUP(param.sb);
1360                         continue;
1361                 }
1362                 ldiskfs_unlock_group(param.sb, param.bg);
1363
1364                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1365                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1366                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1367                 if (param.bitmap == NULL) {
1368                         CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1369                                "scrub will stop, urgent mode\n",
1370                                osd_scrub2name(scrub), (__u32)param.bg);
1371                         RETURN(-EIO);
1372                 }
1373
1374                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1375                        *count < max) {
1376                         if (param.offset +
1377                                 ldiskfs_itable_unused_count(param.sb, desc) >
1378                             LDISKFS_INODES_PER_GROUP(param.sb))
1379                                 goto next_group;
1380
1381                         rc = next(info, dev, &param, &oic, noslot);
1382                         switch (rc) {
1383                         case SCRUB_NEXT_BREAK:
1384                                 goto next_group;
1385                         case SCRUB_NEXT_EXIT:
1386                                 brelse(param.bitmap);
1387                                 RETURN(0);
1388                         case SCRUB_NEXT_CRASH:
1389                                 brelse(param.bitmap);
1390                                 RETURN(SCRUB_IT_CRASH);
1391                         case SCRUB_NEXT_FATAL:
1392                                 brelse(param.bitmap);
1393                                 RETURN(-EINVAL);
1394                         }
1395
1396                         rc = exec(info, dev, &param, oic, &noslot, rc);
1397                         if (rc != 0) {
1398                                 brelse(param.bitmap);
1399                                 RETURN(rc);
1400                         }
1401                 }
1402
1403 next_group:
1404                 brelse(param.bitmap);
1405         }
1406
1407         if (*pos > limit)
1408                 RETURN(SCRUB_IT_ALL);
1409         RETURN(0);
1410 }
1411
1412 static int osd_otable_it_preload(const struct lu_env *env,
1413                                  struct osd_otable_it *it)
1414 {
1415         struct osd_device       *dev   = it->ooi_dev;
1416         struct osd_scrub        *scrub = &dev->od_scrub;
1417         struct osd_otable_cache *ooc   = &it->ooi_cache;
1418         int                      rc;
1419         ENTRY;
1420
1421         rc = osd_inode_iteration(osd_oti_get(env), dev,
1422                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1423         if (rc == SCRUB_IT_ALL)
1424                 it->ooi_all_cached = 1;
1425
1426         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1427                 spin_lock(&scrub->os_lock);
1428                 scrub->os_waiting = 0;
1429                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1430                 spin_unlock(&scrub->os_lock);
1431         }
1432
1433         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1434 }
1435
1436 static int osd_scrub_main(void *args)
1437 {
1438         struct lu_env         env;
1439         struct osd_device    *dev    = (struct osd_device *)args;
1440         struct osd_scrub     *scrub  = &dev->od_scrub;
1441         struct ptlrpc_thread *thread = &scrub->os_thread;
1442         int                   rc;
1443         ENTRY;
1444
1445         rc = lu_env_init(&env, LCT_LOCAL);
1446         if (rc != 0) {
1447                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1448                        osd_scrub2name(scrub), rc);
1449                 GOTO(noenv, rc);
1450         }
1451
1452         rc = osd_scrub_prep(dev);
1453         if (rc != 0) {
1454                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1455                        osd_scrub2name(scrub), rc);
1456                 GOTO(out, rc);
1457         }
1458
1459         if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1460                 struct l_wait_info lwi = { 0 };
1461                 struct osd_otable_it *it = dev->od_otable_it;
1462                 struct osd_otable_cache *ooc = &it->ooi_cache;
1463
1464                 l_wait_event(thread->t_ctl_waitq,
1465                              it->ooi_user_ready || !thread_is_running(thread),
1466                              &lwi);
1467                 if (unlikely(!thread_is_running(thread)))
1468                         GOTO(post, rc = 0);
1469
1470                 scrub->os_pos_current = ooc->ooc_pos_preload;
1471         }
1472
1473         CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1474                osd_scrub2name(scrub), scrub->os_start_flags,
1475                scrub->os_pos_current);
1476
1477         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1478         if (unlikely(rc == SCRUB_IT_CRASH))
1479                 GOTO(out, rc = -EINVAL);
1480         GOTO(post, rc);
1481
1482 post:
1483         osd_scrub_post(scrub, rc);
1484         CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1485                osd_scrub2name(scrub), scrub->os_pos_current, rc);
1486
1487 out:
1488         while (!list_empty(&scrub->os_inconsistent_items)) {
1489                 struct osd_inconsistent_item *oii;
1490
1491                 oii = list_entry(scrub->os_inconsistent_items.next,
1492                                      struct osd_inconsistent_item, oii_list);
1493                 list_del_init(&oii->oii_list);
1494                 OBD_FREE_PTR(oii);
1495         }
1496         lu_env_fini(&env);
1497
1498 noenv:
1499         spin_lock(&scrub->os_lock);
1500         thread_set_flags(thread, SVC_STOPPED);
1501         wake_up_all(&thread->t_ctl_waitq);
1502         spin_unlock(&scrub->os_lock);
1503         return rc;
1504 }
1505
1506 /* initial OI scrub */
1507
1508 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1509                          struct dentry *, filldir_t filldir);
1510
1511 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1512                                loff_t offset, __u64 ino, unsigned d_type);
1513 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1514                            loff_t offset, __u64 ino, unsigned d_type);
1515 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1516                            loff_t offset, __u64 ino, unsigned d_type);
1517
1518 static int
1519 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1520                      struct dentry *dentry, filldir_t filldir);
1521 static int
1522 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1523                   struct dentry *dentry, filldir_t filldir);
1524
1525 static int
1526 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1527                      struct dentry *dentry, filldir_t filldir);
1528
1529 enum osd_lf_flags {
1530         OLF_SCAN_SUBITEMS       = 0x0001,
1531         OLF_HIDE_FID            = 0x0002,
1532         OLF_SHOW_NAME           = 0x0004,
1533         OLF_NO_OI               = 0x0008,
1534 };
1535
1536 struct osd_lf_map {
1537         char            *olm_name;
1538         struct lu_fid    olm_fid;
1539         __u16            olm_flags;
1540         scandir_t        olm_scandir;
1541         filldir_t        olm_filldir;
1542 };
1543
1544 /* Add the new introduced local files in the list in the future. */
1545 static const struct osd_lf_map osd_lf_maps[] = {
1546         /* CATALOGS */
1547         { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1548                 NULL, NULL },
1549
1550         /* CONFIGS */
1551         { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1552                 OLF_SCAN_SUBITEMS, osd_ios_general_scan,
1553                 osd_ios_varfid_fill },
1554
1555         /* NIDTBL_VERSIONS */
1556         { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1557                 osd_ios_general_scan, osd_ios_varfid_fill },
1558
1559         /* PENDING */
1560         { "PENDING", { 0, 0, 0 }, 0, NULL, NULL },
1561
1562         /* ROOT */
1563         { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 },
1564                 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, osd_ios_ROOT_scan, NULL },
1565
1566         /* changelog_catalog */
1567         { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, NULL, NULL },
1568
1569         /* changelog_users */
1570         { CHANGELOG_USERS, { 0, 0, 0 }, 0, NULL, NULL },
1571
1572         /* fld */
1573         { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1574                 NULL, NULL },
1575
1576         /* last_rcvd */
1577         { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1578                 NULL, NULL },
1579
1580         /* lfsck_bookmark */
1581         { "lfsck_bookmark", { 0, 0, 0 }, 0, NULL, NULL },
1582
1583         /* lov_objid */
1584         { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1585                 NULL, NULL },
1586
1587         /* lov_objseq */
1588         { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1589                 OLF_SHOW_NAME, NULL, NULL },
1590
1591         /* quota_master */
1592         { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1593                 osd_ios_general_scan, osd_ios_varfid_fill },
1594
1595         /* quota_slave */
1596         { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1597                 osd_ios_general_scan, osd_ios_varfid_fill },
1598
1599         /* seq_ctl */
1600         { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1601                 OLF_SHOW_NAME, NULL, NULL },
1602
1603         /* seq_srv */
1604         { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1605                 OLF_SHOW_NAME, NULL, NULL },
1606
1607         /* health_check */
1608         { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1609                 OLF_SHOW_NAME, NULL, NULL },
1610
1611         /* lfsck_namespace */
1612         { "lfsck_namespace", { 0, 0, 0 }, 0, NULL, NULL },
1613
1614         /* OBJECTS, upgrade from old device */
1615         { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, osd_ios_OBJECTS_scan, NULL },
1616
1617         /* lquota_v2.user, upgrade from old device */
1618         { "lquota_v2.user", { 0, 0, 0 }, 0, NULL, NULL },
1619
1620         /* lquota_v2.group, upgrade from old device */
1621         { "lquota_v2.group", { 0, 0, 0 }, 0, NULL, NULL },
1622
1623         /* LAST_GROUP, upgrade from old device */
1624         { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1625                 OLF_SHOW_NAME, NULL, NULL },
1626
1627         /* SLAVE_LOG, llog for destroy slave stripes of striped dir */
1628         { "SLAVE_LOG", { FID_SEQ_LOCAL_FILE, SLAVE_LLOG_CATALOGS_OID, 0 },
1629                OLF_SHOW_NAME, NULL, NULL },
1630
1631         /* lost+found */
1632         { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 },
1633                 OLF_SCAN_SUBITEMS, osd_ios_general_scan, osd_ios_lf_fill },
1634
1635         { NULL, { 0, 0, 0 }, 0, NULL, NULL }
1636 };
1637
1638 /* Add the new introduced files under .lustre/ in the list in the future. */
1639 static const struct osd_lf_map osd_dl_maps[] = {
1640         /* .lustre/fid */
1641         { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1642                 NULL, NULL },
1643
1644         /* .lustre/lost+found */
1645         { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1646                 NULL, NULL },
1647
1648         { NULL, { 0, 0, 0 }, 0, NULL, NULL }
1649 };
1650
1651 struct osd_ios_item {
1652         struct list_head oii_list;
1653         struct dentry   *oii_dentry;
1654         scandir_t        oii_scandir;
1655         filldir_t        oii_filldir;
1656 };
1657
1658 struct osd_ios_filldir_buf {
1659 #ifdef HAVE_DIR_CONTEXT
1660         /* please keep it as first member */
1661         struct dir_context       ctx;
1662 #endif
1663         struct osd_thread_info  *oifb_info;
1664         struct osd_device       *oifb_dev;
1665         struct dentry           *oifb_dentry;
1666 };
1667
1668 static inline struct dentry *
1669 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1670 {
1671         struct dentry *dentry;
1672
1673         dentry = ll_lookup_one_len(name, parent, namelen);
1674         if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1675                 dput(dentry);
1676                 return ERR_PTR(-ENOENT);
1677         }
1678
1679         return dentry;
1680 }
1681
1682 static int
1683 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1684                  scandir_t scandir, filldir_t filldir)
1685 {
1686         struct osd_ios_item *item;
1687         ENTRY;
1688
1689         OBD_ALLOC_PTR(item);
1690         if (item == NULL)
1691                 RETURN(-ENOMEM);
1692
1693         INIT_LIST_HEAD(&item->oii_list);
1694         item->oii_dentry = dget(dentry);
1695         item->oii_scandir = scandir;
1696         item->oii_filldir = filldir;
1697         list_add_tail(&item->oii_list, &dev->od_ios_list);
1698
1699         RETURN(0);
1700 }
1701
1702 /**
1703  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1704  *
1705  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1706  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1707  * reference the inode, or fixed if it is missing or references another inode.
1708  */
1709 static int
1710 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1711                  struct inode *inode, const struct lu_fid *fid, int flags)
1712 {
1713         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
1714         struct osd_inode_id     *id     = &info->oti_id;
1715         struct osd_inode_id     *id2    = &info->oti_id2;
1716         struct osd_scrub        *scrub  = &dev->od_scrub;
1717         struct scrub_file       *sf     = &scrub->os_file;
1718         struct lu_fid            tfid;
1719         int                      rc;
1720         ENTRY;
1721
1722         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1723         if (rc != 0 && rc != -ENODATA) {
1724                 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1725                        "rc = %d\n", osd_name(dev), rc);
1726
1727                 RETURN(rc);
1728         }
1729
1730         osd_id_gen(id, inode->i_ino, inode->i_generation);
1731         if (rc == -ENODATA) {
1732                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID)
1733                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1734                 else
1735                         tfid = *fid;
1736                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1737                 if (rc != 0) {
1738                         CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1739                               "scrub: rc = %d\n", osd_name(dev), rc);
1740
1741                         RETURN(rc);
1742                 }
1743         } else {
1744                 if (lma->lma_compat & LMAC_NOT_IN_OI)
1745                         RETURN(0);
1746
1747                 tfid = lma->lma_self_fid;
1748         }
1749
1750         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1751         if (rc != 0) {
1752                 if (rc != -ENOENT)
1753                         RETURN(rc);
1754
1755                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1756                                                DTO_INDEX_INSERT, true, 0);
1757                 if (rc > 0)
1758                         rc = 0;
1759
1760                 RETURN(rc);
1761         }
1762
1763         if (osd_id_eq_strict(id, id2))
1764                 RETURN(0);
1765
1766         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1767                 osd_scrub_file_reset(scrub,
1768                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1769                                      SF_INCONSISTENT);
1770                 rc = osd_scrub_file_store(scrub);
1771                 if (rc != 0)
1772                         RETURN(rc);
1773         }
1774
1775         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1776                                        DTO_INDEX_UPDATE, true, 0);
1777         if (rc > 0)
1778                 rc = 0;
1779
1780         RETURN(rc);
1781 }
1782
1783 /**
1784  * It scans the /lost+found, and for the OST-object (with filter_fid
1785  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1786  */
1787 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1788                            loff_t offset, __u64 ino, unsigned d_type)
1789 {
1790         struct osd_ios_filldir_buf *fill_buf = buf;
1791         struct osd_thread_info     *info     = fill_buf->oifb_info;
1792         struct osd_device          *dev      = fill_buf->oifb_dev;
1793         struct lu_fid              *fid      = &info->oti_fid;
1794         struct osd_scrub           *scrub    = &dev->od_scrub;
1795         struct dentry              *parent   = fill_buf->oifb_dentry;
1796         struct dentry              *child;
1797         struct inode               *dir      = parent->d_inode;
1798         struct inode               *inode;
1799         int                         rc;
1800         ENTRY;
1801
1802         /* skip any '.' started names */
1803         if (name[0] == '.')
1804                 RETURN(0);
1805
1806         scrub->os_lf_scanned++;
1807         child = osd_ios_lookup_one_len(name, parent, namelen);
1808         if (IS_ERR(child)) {
1809                 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1810                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
1811                 RETURN(0);
1812         }
1813
1814         inode = child->d_inode;
1815         if (S_ISDIR(inode->i_mode)) {
1816                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1817                                       osd_ios_lf_fill);
1818                 if (rc != 0)
1819                         CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1820                               "rc = %d\n", osd_name(dev), namelen, name, rc);
1821                 GOTO(put, rc);
1822         }
1823
1824         if (!S_ISREG(inode->i_mode))
1825                 GOTO(put, rc = 0);
1826
1827         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1828         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1829                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1830                 if (rc == 0) {
1831                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1832                                "/lost+found.\n", namelen, name, PFID(fid));
1833                         scrub->os_lf_repaired++;
1834                 } else {
1835                         CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1836                                DFID": rc = %d\n",
1837                                osd_name(dev), namelen, name, PFID(fid), rc);
1838                 }
1839         }
1840
1841         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1842          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1843          *      can process them in furtuer. */
1844
1845         GOTO(put, rc);
1846
1847 put:
1848         if (rc < 0)
1849                 scrub->os_lf_failed++;
1850         dput(child);
1851         /* skip the failure to make the scanning to continue. */
1852         return 0;
1853 }
1854
1855 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1856                                loff_t offset, __u64 ino, unsigned d_type)
1857 {
1858         struct osd_ios_filldir_buf *fill_buf = buf;
1859         struct osd_device          *dev      = fill_buf->oifb_dev;
1860         struct dentry              *child;
1861         int                         rc;
1862         ENTRY;
1863
1864         /* skip any '.' started names */
1865         if (name[0] == '.')
1866                 RETURN(0);
1867
1868         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1869         if (IS_ERR(child))
1870                 RETURN(PTR_ERR(child));
1871
1872         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1873                               NULL, 0);
1874         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1875                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1876                                       osd_ios_varfid_fill);
1877         dput(child);
1878
1879         RETURN(rc);
1880 }
1881
1882 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1883                            loff_t offset, __u64 ino, unsigned d_type)
1884 {
1885         struct osd_ios_filldir_buf *fill_buf = buf;
1886         struct osd_device          *dev      = fill_buf->oifb_dev;
1887         const struct osd_lf_map    *map;
1888         struct dentry              *child;
1889         int                         rc       = 0;
1890         ENTRY;
1891
1892         /* skip any '.' started names */
1893         if (name[0] == '.')
1894                 RETURN(0);
1895
1896         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
1897                 if (strlen(map->olm_name) != namelen)
1898                         continue;
1899
1900                 if (strncmp(map->olm_name, name, namelen) == 0)
1901                         break;
1902         }
1903
1904         if (map->olm_name == NULL)
1905                 RETURN(0);
1906
1907         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1908         if (IS_ERR(child))
1909                 RETURN(PTR_ERR(child));
1910
1911         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1912                               &map->olm_fid, map->olm_flags);
1913         dput(child);
1914
1915         RETURN(rc);
1916 }
1917
1918 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
1919                              loff_t offset, __u64 ino, unsigned d_type)
1920 {
1921         struct osd_ios_filldir_buf *fill_buf = buf;
1922         struct osd_device          *dev      = fill_buf->oifb_dev;
1923         const struct osd_lf_map    *map;
1924         struct dentry              *child;
1925         int                         rc       = 0;
1926         ENTRY;
1927
1928         /* skip any '.' started names */
1929         if (name[0] == '.')
1930                 RETURN(0);
1931
1932         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
1933                 if (strlen(map->olm_name) != namelen)
1934                         continue;
1935
1936                 if (strncmp(map->olm_name, name, namelen) == 0)
1937                         break;
1938         }
1939
1940         if (map->olm_name == NULL)
1941                 RETURN(0);
1942
1943         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1944         if (IS_ERR(child))
1945                 RETURN(PTR_ERR(child));
1946
1947         if (!(map->olm_flags & OLF_NO_OI))
1948                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1949                                       &map->olm_fid, map->olm_flags);
1950         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
1951                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
1952                                       map->olm_filldir);
1953         dput(child);
1954
1955         RETURN(rc);
1956 }
1957
1958 static int
1959 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1960                      struct dentry *dentry, filldir_t filldir)
1961 {
1962         struct osd_ios_filldir_buf    buf   = {
1963 #ifdef HAVE_DIR_CONTEXT
1964                                                 .ctx.actor = filldir,
1965 #endif
1966                                                 .oifb_info = info,
1967                                                 .oifb_dev = dev,
1968                                                 .oifb_dentry = dentry };
1969         struct file                  *filp  = &info->oti_file;
1970         struct inode                 *inode = dentry->d_inode;
1971         const struct file_operations *fops  = inode->i_fop;
1972         int                           rc;
1973         ENTRY;
1974
1975         LASSERT(filldir != NULL);
1976
1977         filp->f_pos = 0;
1978         filp->f_dentry = dentry;
1979         filp->f_mode = FMODE_64BITHASH;
1980         filp->f_mapping = inode->i_mapping;
1981         filp->f_op = fops;
1982         filp->private_data = NULL;
1983         set_file_inode(filp, inode);
1984
1985 #ifdef HAVE_DIR_CONTEXT
1986         buf.ctx.pos = filp->f_pos;
1987         rc = fops->iterate(filp, &buf.ctx);
1988         filp->f_pos = buf.ctx.pos;
1989 #else
1990         rc = fops->readdir(filp, &buf, filldir);
1991 #endif
1992         fops->release(inode, filp);
1993
1994         RETURN(rc);
1995 }
1996
1997 static int
1998 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1999                   struct dentry *dentry, filldir_t filldir)
2000 {
2001         struct osd_scrub  *scrub  = &dev->od_scrub;
2002         struct scrub_file *sf     = &scrub->os_file;
2003         struct dentry     *child;
2004         int                rc;
2005         ENTRY;
2006
2007         /* It is existing MDT0 device. We only allow the case of object without
2008          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2009          * can generate IGIF mode FID for the object and related OI mapping. If
2010          * it is on other MDTs, then becuase file-level backup/restore, related
2011          * OI mapping may be invalid already, we do not know which is the right
2012          * FID for the object. We only allow IGIF objects to reside on the MDT0.
2013          *
2014          * XXX: For the case of object on non-MDT0 device with neither LMA nor
2015          *      "fid" xattr, then something crashed. We cannot re-generate the
2016          *      FID directly, instead, the OI scrub will scan the OI structure
2017          *      and try to re-generate the LMA from the OI mapping. But if the
2018          *      OI mapping crashed or lost also, then we have to give up under
2019          *      double failure cases. */
2020         scrub->os_convert_igif = 1;
2021         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2022                                        strlen(dot_lustre_name));
2023         if (IS_ERR(child)) {
2024                 rc = PTR_ERR(child);
2025                 if (rc == -ENOENT) {
2026                         /* It is 1.8 MDT device. */
2027                         if (!(sf->sf_flags & SF_UPGRADE)) {
2028                                 osd_scrub_file_reset(scrub,
2029                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2030                                         SF_UPGRADE);
2031                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2032                                 rc = osd_scrub_file_store(scrub);
2033                         } else {
2034                                 rc = 0;
2035                         }
2036                 }
2037         } else {
2038                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2039                  * so the client will get IGIF for the ".lustre" object when
2040                  * the MDT restart.
2041                  *
2042                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2043                  * it does not know whether there are some old clients cached
2044                  * the ".lustre" IGIF during the upgrading. Two choices:
2045                  *
2046                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2047                  *    It will allow the old connected clients to access the
2048                  *    ".lustre" with cached IGIF. But it will cause others
2049                  *    on the MDT failed to check "fid_is_dot_lustre()".
2050                  *
2051                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2052                  *    for ".lustre" in spite of whether there are some clients
2053                  *    cached the ".lustre" IGIF or not. It enables the check
2054                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
2055                  *    that the old connected clients cannot access the ".lustre"
2056                  *    with the cached IGIF.
2057                  *
2058                  * Usually, it is rare case for the old connected clients
2059                  * to access the ".lustre" with cached IGIF. So we prefer
2060                  * to the solution 2). */
2061                 rc = osd_ios_scan_one(info, dev, child->d_inode,
2062                                       &LU_DOT_LUSTRE_FID, 0);
2063                 if (rc == 0)
2064                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2065                                               osd_ios_dl_fill);
2066                 dput(child);
2067         }
2068
2069         RETURN(rc);
2070 }
2071
2072 static int
2073 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2074                      struct dentry *dentry, filldir_t filldir)
2075 {
2076         struct osd_scrub  *scrub  = &dev->od_scrub;
2077         struct scrub_file *sf     = &scrub->os_file;
2078         struct dentry     *child;
2079         int                rc;
2080         ENTRY;
2081
2082         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2083                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2084                 rc = osd_scrub_file_store(scrub);
2085                 if (rc != 0)
2086                         RETURN(rc);
2087         }
2088
2089         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2090         if (!IS_ERR(child)) {
2091                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2092                 dput(child);
2093         } else {
2094                 rc = PTR_ERR(child);
2095         }
2096
2097         if (rc != 0 && rc != -ENOENT)
2098                 RETURN(rc);
2099
2100         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2101         if (!IS_ERR(child)) {
2102                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2103                 dput(child);
2104         } else {
2105                 rc = PTR_ERR(child);
2106         }
2107
2108         if (rc == -ENOENT)
2109                 rc = 0;
2110
2111         RETURN(rc);
2112 }
2113
2114 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2115                                 struct osd_device *dev)
2116 {
2117         struct osd_ios_item     *item    = NULL;
2118         scandir_t                scandir = osd_ios_general_scan;
2119         filldir_t                filldir = osd_ios_root_fill;
2120         struct dentry           *dentry  = osd_sb(dev)->s_root;
2121         const struct osd_lf_map *map     = osd_lf_maps;
2122         int                      rc;
2123         ENTRY;
2124
2125         /* Lookup IGIF in OI by force for initial OI scrub. */
2126         dev->od_igif_inoi = 1;
2127
2128         while (1) {
2129                 rc = scandir(info, dev, dentry, filldir);
2130                 if (item != NULL) {
2131                         dput(item->oii_dentry);
2132                         OBD_FREE_PTR(item);
2133                 }
2134
2135                 if (rc != 0)
2136                         break;
2137
2138                 if (list_empty(&dev->od_ios_list))
2139                         break;
2140
2141                 item = list_entry(dev->od_ios_list.next,
2142                                   struct osd_ios_item, oii_list);
2143                 list_del_init(&item->oii_list);
2144
2145                 LASSERT(item->oii_scandir != NULL);
2146                 scandir = item->oii_scandir;
2147                 filldir = item->oii_filldir;
2148                 dentry = item->oii_dentry;
2149         }
2150
2151         while (!list_empty(&dev->od_ios_list)) {
2152                 item = list_entry(dev->od_ios_list.next,
2153                                   struct osd_ios_item, oii_list);
2154                 list_del_init(&item->oii_list);
2155                 dput(item->oii_dentry);
2156                 OBD_FREE_PTR(item);
2157         }
2158
2159         if (rc != 0)
2160                 RETURN(rc);
2161
2162         /* There maybe the case that the object has been removed, but its OI
2163          * mapping is still in the OI file, such as the "CATALOGS" after MDT
2164          * file-level backup/restore. So here cleanup the stale OI mappings. */
2165         while (map->olm_name != NULL) {
2166                 struct dentry *child;
2167
2168                 if (fid_is_zero(&map->olm_fid)) {
2169                         map++;
2170                         continue;
2171                 }
2172
2173                 child = osd_ios_lookup_one_len(map->olm_name,
2174                                                osd_sb(dev)->s_root,
2175                                                strlen(map->olm_name));
2176                 if (!IS_ERR(child))
2177                         dput(child);
2178                 else if (PTR_ERR(child) == -ENOENT)
2179                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2180                                                   NULL, DTO_INDEX_DELETE,
2181                                                   true, 0);
2182                 map++;
2183         }
2184
2185         RETURN(0);
2186 }
2187
2188 char *osd_lf_fid2name(const struct lu_fid *fid)
2189 {
2190         const struct osd_lf_map *map = osd_lf_maps;
2191
2192         while (map->olm_name != NULL) {
2193                 if (!lu_fid_eq(fid, &map->olm_fid)) {
2194                         map++;
2195                         continue;
2196                 }
2197
2198                 if (map->olm_flags & OLF_SHOW_NAME)
2199                         return map->olm_name;
2200                 else
2201                         return "";
2202         }
2203
2204         return NULL;
2205 }
2206
2207 /* OI scrub start/stop */
2208
2209 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2210 {
2211         struct osd_scrub     *scrub  = &dev->od_scrub;
2212         struct ptlrpc_thread *thread = &scrub->os_thread;
2213         struct l_wait_info    lwi    = { 0 };
2214         struct task_struct   *task;
2215         int                   rc;
2216         ENTRY;
2217
2218         /* os_lock: sync status between stop and scrub thread */
2219         spin_lock(&scrub->os_lock);
2220
2221 again:
2222         if (thread_is_running(thread)) {
2223                 spin_unlock(&scrub->os_lock);
2224                 if (!scrub->os_partial_scan || flags & SS_AUTO_PARTIAL)
2225                         RETURN(-EALREADY);
2226
2227                 osd_scrub_join(dev, flags, false);
2228                 spin_lock(&scrub->os_lock);
2229                 if (!thread_is_running(thread))
2230                         goto again;
2231
2232                 spin_unlock(&scrub->os_lock);
2233                 RETURN(0);
2234         }
2235
2236         if (unlikely(thread_is_stopping(thread))) {
2237                 spin_unlock(&scrub->os_lock);
2238                 l_wait_event(thread->t_ctl_waitq,
2239                              thread_is_stopped(thread),
2240                              &lwi);
2241                 spin_lock(&scrub->os_lock);
2242                 goto again;
2243         }
2244         spin_unlock(&scrub->os_lock);
2245
2246         if (scrub->os_file.sf_status == SS_COMPLETED) {
2247                 if (!(flags & SS_SET_FAILOUT))
2248                         flags |= SS_CLEAR_FAILOUT;
2249
2250                 if (!(flags & SS_SET_DRYRUN))
2251                         flags |= SS_CLEAR_DRYRUN;
2252
2253                 flags |= SS_RESET;
2254         }
2255
2256         scrub->os_start_flags = flags;
2257         thread_set_flags(thread, 0);
2258         task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2259         if (IS_ERR(task)) {
2260                 rc = PTR_ERR(task);
2261                 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2262                        osd_scrub2name(scrub), rc);
2263                 RETURN(rc);
2264         }
2265
2266         l_wait_event(thread->t_ctl_waitq,
2267                      thread_is_running(thread) || thread_is_stopped(thread),
2268                      &lwi);
2269
2270         RETURN(0);
2271 }
2272
2273 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2274 {
2275         int rc;
2276         ENTRY;
2277
2278         /* od_otable_mutex: prevent curcurrent start/stop */
2279         mutex_lock(&dev->od_otable_mutex);
2280         rc = do_osd_scrub_start(dev, flags);
2281         mutex_unlock(&dev->od_otable_mutex);
2282
2283         RETURN(rc == -EALREADY ? 0 : rc);
2284 }
2285
2286 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2287 {
2288         struct ptlrpc_thread *thread = &scrub->os_thread;
2289         struct l_wait_info    lwi    = { 0 };
2290
2291         /* os_lock: sync status between stop and scrub thread */
2292         spin_lock(&scrub->os_lock);
2293         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2294                 thread_set_flags(thread, SVC_STOPPING);
2295                 spin_unlock(&scrub->os_lock);
2296                 wake_up_all(&thread->t_ctl_waitq);
2297                 l_wait_event(thread->t_ctl_waitq,
2298                              thread_is_stopped(thread),
2299                              &lwi);
2300                 /* Do not skip the last lock/unlock, which can guarantee that
2301                  * the caller cannot return until the OI scrub thread exit. */
2302                 spin_lock(&scrub->os_lock);
2303         }
2304         spin_unlock(&scrub->os_lock);
2305 }
2306
2307 static void osd_scrub_stop(struct osd_device *dev)
2308 {
2309         /* od_otable_mutex: prevent curcurrent start/stop */
2310         mutex_lock(&dev->od_otable_mutex);
2311         dev->od_scrub.os_paused = 1;
2312         do_osd_scrub_stop(&dev->od_scrub);
2313         mutex_unlock(&dev->od_otable_mutex);
2314 }
2315
2316 /* OI scrub setup/cleanup */
2317
2318 static const char osd_scrub_name[] = "OI_scrub";
2319
2320 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2321 {
2322         struct osd_thread_info     *info   = osd_oti_get(env);
2323         struct osd_scrub           *scrub  = &dev->od_scrub;
2324         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
2325         struct scrub_file          *sf     = &scrub->os_file;
2326         struct super_block         *sb     = osd_sb(dev);
2327         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
2328         struct lvfs_run_ctxt        saved;
2329         struct file                *filp;
2330         struct inode               *inode;
2331         struct lu_fid              *fid    = &info->oti_fid;
2332         int                         dirty  = 0;
2333         int                         rc     = 0;
2334         ENTRY;
2335
2336         memset(scrub, 0, sizeof(*scrub));
2337         OBD_SET_CTXT_MAGIC(ctxt);
2338         ctxt->pwdmnt = dev->od_mnt;
2339         ctxt->pwd = dev->od_mnt->mnt_root;
2340         ctxt->fs = get_ds();
2341
2342         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2343         init_rwsem(&scrub->os_rwsem);
2344         spin_lock_init(&scrub->os_lock);
2345         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2346
2347         push_ctxt(&saved, ctxt);
2348         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2349         if (IS_ERR(filp)) {
2350                 pop_ctxt(&saved, ctxt);
2351                 RETURN(PTR_ERR(filp));
2352         }
2353
2354         inode = filp->f_dentry->d_inode;
2355         /* 'What the @fid is' is not imporatant, because the object
2356          * has no OI mapping, and only is visible inside the OSD.*/
2357         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2358         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2359         if (rc != 0) {
2360                 filp_close(filp, NULL);
2361                 pop_ctxt(&saved, ctxt);
2362                 RETURN(rc);
2363         }
2364
2365         scrub->os_inode = igrab(inode);
2366         filp_close(filp, NULL);
2367         pop_ctxt(&saved, ctxt);
2368
2369         rc = osd_scrub_file_load(scrub);
2370         if (rc == -ENOENT) {
2371                 osd_scrub_file_init(scrub, es->s_uuid);
2372                 /* If the "/O" dir does not exist when mount (indicated by
2373                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2374                  * then it is quite probably that the device is a new one,
2375                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2376                  *
2377                  * For the rare case that "/O" and "OI_scrub" both lost on
2378                  * an old device, it can be found and cleared later.
2379                  *
2380                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2381                  * need to check "filter_fid_old" and to convert it to
2382                  * "filter_fid" for each object, and all the IGIF should
2383                  * have their FID mapping in OI files already. */
2384                 if (dev->od_maybe_new)
2385                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2386                 dirty = 1;
2387         } else if (rc != 0) {
2388                 GOTO(cleanup_inode, rc);
2389         } else {
2390                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2391                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2392                         dirty = 1;
2393                 } else if (sf->sf_status == SS_SCANNING) {
2394                         sf->sf_status = SS_CRASHED;
2395                         dirty = 1;
2396                 }
2397         }
2398
2399         if (sf->sf_pos_last_checkpoint != 0)
2400                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2401         else
2402                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2403
2404         if (dirty != 0) {
2405                 rc = osd_scrub_file_store(scrub);
2406                 if (rc != 0)
2407                         GOTO(cleanup_inode, rc);
2408         }
2409
2410         /* Initialize OI files. */
2411         rc = osd_oi_init(info, dev);
2412         if (rc < 0)
2413                 GOTO(cleanup_inode, rc);
2414
2415         rc = osd_initial_OI_scrub(info, dev);
2416         if (rc != 0)
2417                 GOTO(cleanup_oi, rc);
2418
2419         if (sf->sf_flags & SF_UPGRADE ||
2420             !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2421               sf->sf_success_count > 0)) {
2422                 dev->od_igif_inoi = 0;
2423                 dev->od_check_ff = dev->od_is_ost;
2424         } else {
2425                 dev->od_igif_inoi = 1;
2426                 dev->od_check_ff = 0;
2427         }
2428
2429         if (sf->sf_flags & SF_INCONSISTENT)
2430                 /* The 'od_igif_inoi' will be set under the
2431                  * following cases:
2432                  * 1) new created system, or
2433                  * 2) restored from file-level backup, or
2434                  * 3) the upgrading completed.
2435                  *
2436                  * The 'od_igif_inoi' may be cleared by OI scrub
2437                  * later if found that the system is upgrading. */
2438                 dev->od_igif_inoi = 1;
2439
2440         if (!dev->od_noscrub &&
2441             ((sf->sf_status == SS_PAUSED) ||
2442              (sf->sf_status == SS_CRASHED &&
2443               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2444                               SF_UPGRADE | SF_AUTO)) ||
2445              (sf->sf_status == SS_INIT &&
2446               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2447                               SF_UPGRADE))))
2448                 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2449
2450         if (rc != 0)
2451                 GOTO(cleanup_oi, rc);
2452
2453         /* it is possible that dcache entries may keep objects after they are
2454          * deleted by OSD. While it looks safe this can cause object data to
2455          * stay until umount causing failures in tests calculating free space,
2456          * e.g. replay-ost-single. Since those dcache entries are not used
2457          * anymore let's just free them after use here */
2458         shrink_dcache_sb(sb);
2459
2460         RETURN(0);
2461 cleanup_oi:
2462         osd_oi_fini(info, dev);
2463 cleanup_inode:
2464         iput(scrub->os_inode);
2465         scrub->os_inode = NULL;
2466
2467         return rc;
2468 }
2469
2470 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2471 {
2472         struct osd_scrub *scrub = &dev->od_scrub;
2473
2474         LASSERT(dev->od_otable_it == NULL);
2475
2476         if (scrub->os_inode != NULL) {
2477                 osd_scrub_stop(dev);
2478                 iput(scrub->os_inode);
2479                 scrub->os_inode = NULL;
2480         }
2481         if (dev->od_oi_table != NULL)
2482                 osd_oi_fini(osd_oti_get(env), dev);
2483 }
2484
2485 /* object table based iteration APIs */
2486
2487 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2488                                        struct dt_object *dt, __u32 attr,
2489                                        struct lustre_capa *capa)
2490 {
2491         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2492         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2493         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2494         struct osd_scrub       *scrub = &dev->od_scrub;
2495         struct osd_otable_it   *it;
2496         __u32                   start = 0;
2497         int                     rc;
2498         ENTRY;
2499
2500         /* od_otable_mutex: prevent curcurrent init/fini */
2501         mutex_lock(&dev->od_otable_mutex);
2502         if (dev->od_otable_it != NULL)
2503                 GOTO(out, it = ERR_PTR(-EALREADY));
2504
2505         OBD_ALLOC_PTR(it);
2506         if (it == NULL)
2507                 GOTO(out, it = ERR_PTR(-ENOMEM));
2508
2509         dev->od_otable_it = it;
2510         it->ooi_dev = dev;
2511         it->ooi_cache.ooc_consumer_idx = -1;
2512         if (flags & DOIF_OUTUSED)
2513                 it->ooi_used_outside = 1;
2514
2515         if (flags & DOIF_RESET)
2516                 start |= SS_RESET;
2517
2518         if (valid & DOIV_ERROR_HANDLE) {
2519                 if (flags & DOIF_FAILOUT)
2520                         start |= SS_SET_FAILOUT;
2521                 else
2522                         start |= SS_CLEAR_FAILOUT;
2523         }
2524
2525         if (valid & DOIV_DRYRUN) {
2526                 if (flags & DOIF_DRYRUN)
2527                         start |= SS_SET_DRYRUN;
2528                 else
2529                         start |= SS_CLEAR_DRYRUN;
2530         }
2531
2532         rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2533         if (rc < 0 && rc != -EALREADY) {
2534                 dev->od_otable_it = NULL;
2535                 OBD_FREE_PTR(it);
2536                 GOTO(out, it = ERR_PTR(rc));
2537         }
2538
2539         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2540
2541         GOTO(out, it);
2542
2543 out:
2544         mutex_unlock(&dev->od_otable_mutex);
2545         return (struct dt_it *)it;
2546 }
2547
2548 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2549 {
2550         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2551         struct osd_device    *dev = it->ooi_dev;
2552
2553         /* od_otable_mutex: prevent curcurrent init/fini */
2554         mutex_lock(&dev->od_otable_mutex);
2555         do_osd_scrub_stop(&dev->od_scrub);
2556         LASSERT(dev->od_otable_it == it);
2557
2558         dev->od_otable_it = NULL;
2559         mutex_unlock(&dev->od_otable_mutex);
2560         OBD_FREE_PTR(it);
2561 }
2562
2563 static int osd_otable_it_get(const struct lu_env *env,
2564                              struct dt_it *di, const struct dt_key *key)
2565 {
2566         return 0;
2567 }
2568
2569 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2570 {
2571 }
2572
2573 static inline int
2574 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2575 {
2576         spin_lock(&scrub->os_lock);
2577         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2578             scrub->os_waiting ||
2579             !thread_is_running(&scrub->os_thread))
2580                 it->ooi_waiting = 0;
2581         else
2582                 it->ooi_waiting = 1;
2583         spin_unlock(&scrub->os_lock);
2584
2585         return !it->ooi_waiting;
2586 }
2587
2588 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2589 {
2590         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2591         struct osd_device       *dev    = it->ooi_dev;
2592         struct osd_scrub        *scrub  = &dev->od_scrub;
2593         struct osd_otable_cache *ooc    = &it->ooi_cache;
2594         struct ptlrpc_thread    *thread = &scrub->os_thread;
2595         struct l_wait_info       lwi    = { 0 };
2596         int                      rc;
2597         ENTRY;
2598
2599         LASSERT(it->ooi_user_ready);
2600
2601 again:
2602         if (!thread_is_running(thread) && !it->ooi_used_outside)
2603                 RETURN(1);
2604
2605         if (ooc->ooc_cached_items > 0) {
2606                 ooc->ooc_cached_items--;
2607                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2608                                         ~OSD_OTABLE_IT_CACHE_MASK;
2609                 RETURN(0);
2610         }
2611
2612         if (it->ooi_all_cached) {
2613                 l_wait_event(thread->t_ctl_waitq,
2614                              !thread_is_running(thread),
2615                              &lwi);
2616                 RETURN(1);
2617         }
2618
2619         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2620                 spin_lock(&scrub->os_lock);
2621                 scrub->os_waiting = 0;
2622                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2623                 spin_unlock(&scrub->os_lock);
2624         }
2625
2626         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2627                 l_wait_event(thread->t_ctl_waitq,
2628                              osd_otable_it_wakeup(scrub, it),
2629                              &lwi);
2630
2631         if (!thread_is_running(thread) && !it->ooi_used_outside)
2632                 RETURN(1);
2633
2634         rc = osd_otable_it_preload(env, it);
2635         if (rc >= 0)
2636                 goto again;
2637
2638         RETURN(rc);
2639 }
2640
2641 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2642                                         const struct dt_it *di)
2643 {
2644         return NULL;
2645 }
2646
2647 static int osd_otable_it_key_size(const struct lu_env *env,
2648                                   const struct dt_it *di)
2649 {
2650         return sizeof(__u64);
2651 }
2652
2653 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2654                              struct dt_rec *rec, __u32 attr)
2655 {
2656         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2657         struct osd_otable_cache *ooc = &it->ooi_cache;
2658
2659         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2660
2661         /* Filter out Invald FID already. */
2662         LASSERTF(fid_is_sane((struct lu_fid *)rec),
2663                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2664                  PFID((struct lu_fid *)rec),
2665                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2666
2667         return 0;
2668 }
2669
2670 static __u64 osd_otable_it_store(const struct lu_env *env,
2671                                  const struct dt_it *di)
2672 {
2673         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2674         struct osd_otable_cache *ooc = &it->ooi_cache;
2675         __u64                    hash;
2676
2677         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2678                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2679         else
2680                 hash = ooc->ooc_pos_preload;
2681         return hash;
2682 }
2683
2684 /**
2685  * Set the OSD layer iteration start position as the specified hash.
2686  */
2687 static int osd_otable_it_load(const struct lu_env *env,
2688                               const struct dt_it *di, __u64 hash)
2689 {
2690         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
2691         struct osd_device       *dev   = it->ooi_dev;
2692         struct osd_otable_cache *ooc   = &it->ooi_cache;
2693         struct osd_scrub        *scrub = &dev->od_scrub;
2694         int                      rc;
2695         ENTRY;
2696
2697         /* Forbid to set iteration position after iteration started. */
2698         if (it->ooi_user_ready)
2699                 RETURN(-EPERM);
2700
2701         if (hash > OSD_OTABLE_MAX_HASH)
2702                 hash = OSD_OTABLE_MAX_HASH;
2703
2704         ooc->ooc_pos_preload = hash;
2705         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2706                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2707
2708         it->ooi_user_ready = 1;
2709         if (!scrub->os_full_speed)
2710                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2711
2712         /* Unplug OSD layer iteration by the first next() call. */
2713         rc = osd_otable_it_next(env, (struct dt_it *)it);
2714
2715         RETURN(rc);
2716 }
2717
2718 static int osd_otable_it_key_rec(const struct lu_env *env,
2719                                  const struct dt_it *di, void *key_rec)
2720 {
2721         return 0;
2722 }
2723
2724 const struct dt_index_operations osd_otable_ops = {
2725         .dio_it = {
2726                 .init     = osd_otable_it_init,
2727                 .fini     = osd_otable_it_fini,
2728                 .get      = osd_otable_it_get,
2729                 .put      = osd_otable_it_put,
2730                 .next     = osd_otable_it_next,
2731                 .key      = osd_otable_it_key,
2732                 .key_size = osd_otable_it_key_size,
2733                 .rec      = osd_otable_it_rec,
2734                 .store    = osd_otable_it_store,
2735                 .load     = osd_otable_it_load,
2736                 .key_rec  = osd_otable_it_key_rec,
2737         }
2738 };
2739
2740 /* high priority inconsistent items list APIs */
2741
2742 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL  60
2743
2744 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2745                    int insert)
2746 {
2747         struct osd_inconsistent_item *oii;
2748         struct osd_scrub             *scrub  = &dev->od_scrub;
2749         struct ptlrpc_thread         *thread = &scrub->os_thread;
2750         int                           wakeup = 0;
2751         ENTRY;
2752
2753         OBD_ALLOC_PTR(oii);
2754         if (unlikely(oii == NULL))
2755                 RETURN(-ENOMEM);
2756
2757         INIT_LIST_HEAD(&oii->oii_list);
2758         oii->oii_cache = *oic;
2759         oii->oii_insert = insert;
2760
2761         if (scrub->os_partial_scan) {
2762                 __u64 now = cfs_time_current_sec();
2763
2764                 /* If there haven't been errors in a long time,
2765                  * decay old count until either the errors are
2766                  * gone or we reach the current interval. */
2767                 while (unlikely(scrub->os_bad_oimap_count > 0 &&
2768                                 scrub->os_bad_oimap_time +
2769                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
2770                         scrub->os_bad_oimap_count >>= 1;
2771                         scrub->os_bad_oimap_time +=
2772                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
2773                 }
2774
2775                 scrub->os_bad_oimap_time = now;
2776                 if (++scrub->os_bad_oimap_count >
2777                     dev->od_full_scrub_threshold_rate)
2778                         scrub->os_full_scrub = 1;
2779         }
2780
2781         spin_lock(&scrub->os_lock);
2782         if (unlikely(!thread_is_running(thread))) {
2783                 spin_unlock(&scrub->os_lock);
2784                 OBD_FREE_PTR(oii);
2785                 RETURN(-EAGAIN);
2786         }
2787
2788         if (list_empty(&scrub->os_inconsistent_items))
2789                 wakeup = 1;
2790         list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2791         spin_unlock(&scrub->os_lock);
2792
2793         if (wakeup != 0)
2794                 wake_up_all(&thread->t_ctl_waitq);
2795
2796         RETURN(0);
2797 }
2798
2799 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2800                    struct osd_inode_id *id)
2801 {
2802         struct osd_scrub             *scrub = &dev->od_scrub;
2803         struct osd_inconsistent_item *oii;
2804         ENTRY;
2805
2806         spin_lock(&scrub->os_lock);
2807         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2808                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2809                         *id = oii->oii_cache.oic_lid;
2810                         spin_unlock(&scrub->os_lock);
2811                         RETURN(0);
2812                 }
2813         }
2814         spin_unlock(&scrub->os_lock);
2815
2816         RETURN(-ENOENT);
2817 }
2818
2819 /* OI scrub dump */
2820
2821 static const char *scrub_status_names[] = {
2822         "init",
2823         "scanning",
2824         "completed",
2825         "failed",
2826         "stopped",
2827         "paused",
2828         "crashed",
2829         NULL
2830 };
2831
2832 static const char *scrub_flags_names[] = {
2833         "recreated",
2834         "inconsistent",
2835         "auto",
2836         "upgrade",
2837         NULL
2838 };
2839
2840 static const char *scrub_param_names[] = {
2841         "failout",
2842         "dryrun",
2843         NULL
2844 };
2845
2846 static int scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
2847                            const char *prefix)
2848 {
2849         int flag;
2850         int rc;
2851         int i;
2852
2853         rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2854         if (rc < 0)
2855                 return rc;
2856
2857         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2858                 if (flag & bits) {
2859                         bits &= ~flag;
2860                         rc = seq_printf(m, "%s%c", names[i],
2861                                         bits != 0 ? ',' : '\n');
2862                         if (rc < 0)
2863                                 return rc;
2864                 }
2865         }
2866         return 0;
2867 }
2868
2869 static int scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
2870 {
2871         int rc;
2872
2873         if (time != 0)
2874                 rc = seq_printf(m, "%s: "LPU64" seconds\n", prefix,
2875                               cfs_time_current_sec() - time);
2876         else
2877                 rc = seq_printf(m, "%s: N/A\n", prefix);
2878         return rc;
2879 }
2880
2881 static int scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
2882 {
2883         int rc;
2884
2885         if (pos != 0)
2886                 rc = seq_printf(m, "%s: "LPU64"\n", prefix, pos);
2887         else
2888                 rc = seq_printf(m, "%s: N/A\n", prefix);
2889         return rc;
2890 }
2891
2892 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
2893 {
2894         struct osd_scrub  *scrub   = &dev->od_scrub;
2895         struct scrub_file *sf      = &scrub->os_file;
2896         __u64              checked;
2897         __u64              speed;
2898         int                rc;
2899
2900         down_read(&scrub->os_rwsem);
2901         rc = seq_printf(m, "name: OI_scrub\n"
2902                         "magic: 0x%x\n"
2903                         "oi_files: %d\n"
2904                         "status: %s\n",
2905                         sf->sf_magic, (int)sf->sf_oi_count,
2906                         scrub_status_names[sf->sf_status]);
2907         if (rc < 0)
2908                 goto out;
2909
2910         rc = scrub_bits_dump(m, sf->sf_flags, scrub_flags_names,
2911                              "flags");
2912         if (rc < 0)
2913                 goto out;
2914
2915         rc = scrub_bits_dump(m, sf->sf_param, scrub_param_names,
2916                              "param");
2917         if (rc < 0)
2918                 goto out;
2919
2920         rc = scrub_time_dump(m, sf->sf_time_last_complete,
2921                              "time_since_last_completed");
2922         if (rc < 0)
2923                 goto out;
2924
2925         rc = scrub_time_dump(m, sf->sf_time_latest_start,
2926                              "time_since_latest_start");
2927         if (rc < 0)
2928                 goto out;
2929
2930         rc = scrub_time_dump(m, sf->sf_time_last_checkpoint,
2931                              "time_since_last_checkpoint");
2932         if (rc < 0)
2933                 goto out;
2934
2935         rc = scrub_pos_dump(m, sf->sf_pos_latest_start,
2936                             "latest_start_position");
2937         if (rc < 0)
2938                 goto out;
2939
2940         rc = scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
2941                             "last_checkpoint_position");
2942         if (rc < 0)
2943                 goto out;
2944
2945         rc = scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
2946                             "first_failure_position");
2947         if (rc < 0)
2948                 goto out;
2949
2950         checked = sf->sf_items_checked + scrub->os_new_checked;
2951         rc = seq_printf(m, "checked: "LPU64"\n"
2952                       "updated: "LPU64"\n"
2953                       "failed: "LPU64"\n"
2954                       "prior_updated: "LPU64"\n"
2955                       "noscrub: "LPU64"\n"
2956                       "igif: "LPU64"\n"
2957                       "success_count: %u\n",
2958                       checked, sf->sf_items_updated, sf->sf_items_failed,
2959                       sf->sf_items_updated_prior, sf->sf_items_noscrub,
2960                       sf->sf_items_igif, sf->sf_success_count);
2961         if (rc < 0)
2962                 goto out;
2963
2964         speed = checked;
2965         if (thread_is_running(&scrub->os_thread)) {
2966                 cfs_duration_t duration = cfs_time_current() -
2967                                           scrub->os_time_last_checkpoint;
2968                 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
2969                                                      MSEC_PER_SEC);
2970                 __u32 rtime = sf->sf_run_time +
2971                               cfs_duration_sec(duration + HALF_SEC);
2972
2973                 if (duration != 0)
2974                         do_div(new_checked, duration);
2975                 if (rtime != 0)
2976                         do_div(speed, rtime);
2977                 rc = seq_printf(m, "run_time: %u seconds\n"
2978                               "average_speed: "LPU64" objects/sec\n"
2979                               "real-time_speed: "LPU64" objects/sec\n"
2980                               "current_position: %u\n"
2981                               "lf_scanned: "LPU64"\n"
2982                               "lf_reparied: "LPU64"\n"
2983                               "lf_failed: "LPU64"\n",
2984                               rtime, speed, new_checked, scrub->os_pos_current,
2985                               scrub->os_lf_scanned, scrub->os_lf_repaired,
2986                               scrub->os_lf_failed);
2987         } else {
2988                 if (sf->sf_run_time != 0)
2989                         do_div(speed, sf->sf_run_time);
2990                 rc = seq_printf(m, "run_time: %u seconds\n"
2991                               "average_speed: "LPU64" objects/sec\n"
2992                               "real-time_speed: N/A\n"
2993                               "current_position: N/A\n"
2994                               "lf_scanned: "LPU64"\n"
2995                               "lf_reparied: "LPU64"\n"
2996                               "lf_failed: "LPU64"\n",
2997                               sf->sf_run_time, speed, scrub->os_lf_scanned,
2998                               scrub->os_lf_repaired, scrub->os_lf_failed);
2999         }
3000
3001 out:
3002         up_read(&scrub->os_rwsem);
3003         return (rc < 0 ? -ENOSPC : 0);
3004 }