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         struct lu_fid           *tfid    = &info->oti_fid;
 374         handle_t                *jh;
 375         int                      size    = 0;
 376         int                      rc;
 377         bool                     removed = false;
 378         bool                     reset   = true;
 379         ENTRY;
 380
 381         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
 382                 RETURN(0);
 383
 384         if (fid_is_idif(fid) && dev->od_index_in_idif == 0) {
 385                 struct ost_id *oi = &info->oti_ostid;
 386
 387                 fid_to_ostid(fid, oi);
 388                 ostid_to_fid(tfid, oi, 0);
 389         } else {
 390                 *tfid = *fid;
 391         }
 392
 393         /* We want the LMA to fit into the 256-byte OST inode, so operate
 394          * as following:
 395          * 1) read old XATTR_NAME_FID and save the parent FID;
 396          * 2) delete the old XATTR_NAME_FID;
 397          * 3) make new LMA and add it;
 398          * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
 399          *
 400          * Making the LMA to fit into the 256-byte OST inode can save time for
 401          * normal osd_check_lma() and for other OI scrub scanning in future.
 402          * So it is worth to make some slow conversion here. */
 403         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 404                                 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
 405         if (IS_ERR(jh)) {
 406                 rc = PTR_ERR(jh);
 407                 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
 408                        DFID": rc = %d\n", osd_name(dev), PFID(tfid), rc);
 409                 RETURN(rc);
 410         }
 411
 412         /* 1) read old XATTR_NAME_FID and save the parent FID */
 413         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 414         if (rc == sizeof(*ff)) {
 415                 /* 2) delete the old XATTR_NAME_FID */
 416                 ll_vfs_dq_init(inode);
 417                 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
 418                 if (rc != 0)
 419                         GOTO(stop, rc);
 420
 421                 removed = true;
 422         } else if (unlikely(rc == -ENODATA)) {
 423                 reset = false;
 424         } else if (rc != sizeof(struct filter_fid)) {
 425                 GOTO(stop, rc = -EINVAL);
 426         }
 427
 428         /* 3) make new LMA and add it */
 429         rc = osd_ea_fid_set(info, inode, tfid, LMAC_FID_ON_OST, 0);
 430         if (rc == 0 && reset)
 431                 size = sizeof(struct filter_fid);
 432         else if (rc != 0 && removed)
 433                 /* If failed, we should try to add the old back. */
 434                 size = sizeof(struct filter_fid_old);
 435
 436         /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
 437         if (size > 0) {
 438                 int rc1;
 439
 440                 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
 441                                       XATTR_CREATE);
 442                 if (rc1 != 0 && rc == 0)
 443                         rc = rc1;
 444         }
 445
 446         GOTO(stop, rc);
 447
 448 stop:
 449         ldiskfs_journal_stop(jh);
 450         if (rc < 0)
 451                 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
 452                        osd_name(dev), PFID(tfid), rc);
 453         return rc;
 454 }
 455
 456 static int
 457 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
 458                        struct osd_idmap_cache *oic, int val)
 459 {
 460         struct osd_scrub             *scrub  = &dev->od_scrub;
 461         struct scrub_file            *sf     = &scrub->os_file;
 462         struct lu_fid                *fid    = &oic->oic_fid;
 463         struct osd_inode_id          *lid    = &oic->oic_lid;
 464         struct osd_inode_id          *lid2   = &info->oti_id;
 465         struct osd_inconsistent_item *oii    = NULL;
 466         struct inode                 *inode  = NULL;
 467         int                           ops    = DTO_INDEX_UPDATE;
 468         int                           idx;
 469         int                           rc;
 470         bool                          converted = false;
 471         ENTRY;
 472
 473         down_write(&scrub->os_rwsem);
 474         scrub->os_new_checked++;
 475         if (val < 0)
 476                 GOTO(out, rc = val);
 477
 478         if (scrub->os_in_prior)
 479                 oii = list_entry(oic, struct osd_inconsistent_item,
 480                                  oii_cache);
 481
 482         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
 483                 GOTO(out, rc = 0);
 484
 485         if (fid_is_igif(fid))
 486                 sf->sf_items_igif++;
 487
 488         if (val == SCRUB_NEXT_OSTOBJ_OLD) {
 489                 inode = osd_iget(info, dev, lid);
 490                 if (IS_ERR(inode)) {
 491                         rc = PTR_ERR(inode);
 492                         /* Someone removed the inode. */
 493                         if (rc == -ENOENT || rc == -ESTALE)
 494                                 rc = 0;
 495                         GOTO(out, rc);
 496                 }
 497
 498                 sf->sf_flags |= SF_UPGRADE;
 499                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
 500                 dev->od_check_ff = 1;
 501                 rc = osd_scrub_convert_ff(info, dev, inode, fid);
 502                 if (rc != 0)
 503                         GOTO(out, rc);
 504
 505                 converted = true;
 506         }
 507
 508         if ((val == SCRUB_NEXT_NOLMA) &&
 509             (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
 510                 GOTO(out, rc = 0);
 511
 512         if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA)) {
 513                 ops = DTO_INDEX_INSERT;
 514
 515                 goto iget;
 516         }
 517
 518         rc = osd_oi_lookup(info, dev, fid, lid2,
 519                 (val == SCRUB_NEXT_OSTOBJ ||
 520                  val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 521         if (rc != 0) {
 522                 if (rc == -ENOENT)
 523                         ops = DTO_INDEX_INSERT;
 524                 else if (rc != -ESTALE)
 525                         GOTO(out, rc);
 526
 527 iget:
 528                 if (inode == NULL) {
 529                         inode = osd_iget(info, dev, lid);
 530                         if (IS_ERR(inode)) {
 531                                 rc = PTR_ERR(inode);
 532                                 /* Someone removed the inode. */
 533                                 if (rc == -ENOENT || rc == -ESTALE)
 534                                         rc = 0;
 535                                 GOTO(out, rc);
 536                         }
 537                 }
 538
 539                 if (!scrub->os_partial_scan)
 540                         scrub->os_full_speed = 1;
 541
 542                 idx = osd_oi_fid2idx(dev, fid);
 543                 switch (val) {
 544                 case SCRUB_NEXT_NOLMA:
 545                         sf->sf_flags |= SF_UPGRADE;
 546                         if (!(sf->sf_param & SP_DRYRUN)) {
 547                                 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
 548                                 if (rc != 0)
 549                                         GOTO(out, rc);
 550                         }
 551
 552                         if (!(sf->sf_flags & SF_INCONSISTENT))
 553                                 dev->od_igif_inoi = 0;
 554                         break;
 555                 case SCRUB_NEXT_OSTOBJ:
 556                         sf->sf_flags |= SF_INCONSISTENT;
 557                 case SCRUB_NEXT_OSTOBJ_OLD:
 558                         break;
 559                 default:
 560                         sf->sf_flags |= SF_RECREATED;
 561                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 562                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 563                         break;
 564                 }
 565         } else if (osd_id_eq(lid, lid2)) {
 566                 if (converted)
 567                         sf->sf_items_updated++;
 568
 569                 GOTO(out, rc = 0);
 570         } else {
 571                 if (!scrub->os_partial_scan)
 572                         scrub->os_full_speed = 1;
 573
 574                 sf->sf_flags |= SF_INCONSISTENT;
 575
 576                 /* XXX: If the device is restored from file-level backup, then
 577                  *      some IGIFs may have been already in OI files, and some
 578                  *      may be not yet. Means upgrading from 1.8 may be partly
 579                  *      processed, but some clients may hold some immobilized
 580                  *      IGIFs, and use them to access related objects. Under
 581                  *      such case, OSD does not know whether an given IGIF has
 582                  *      been processed or to be processed, and it also cannot
 583                  *      generate local ino#/gen# directly from the immobilized
 584                  *      IGIF because of the backup/restore. Then force OSD to
 585                  *      lookup the given IGIF in OI files, and if no entry,
 586                  *      then ask the client to retry after upgrading completed.
 587                  *      No better choice. */
 588                 dev->od_igif_inoi = 1;
 589         }
 590
 591         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
 592                         (val == SCRUB_NEXT_OSTOBJ ||
 593                          val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 594         if (rc == 0) {
 595                 if (scrub->os_in_prior)
 596                         sf->sf_items_updated_prior++;
 597                 else
 598                         sf->sf_items_updated++;
 599         }
 600
 601         GOTO(out, rc);
 602
 603 out:
 604         if (rc < 0) {
 605                 sf->sf_items_failed++;
 606                 if (sf->sf_pos_first_inconsistent == 0 ||
 607                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 608                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 609         } else {
 610                 rc = 0;
 611         }
 612
 613         /* There may be conflict unlink during the OI scrub,
 614          * if happend, then remove the new added OI mapping. */
 615         if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
 616             unlikely(inode->i_nlink == 0))
 617                 osd_scrub_refresh_mapping(info, dev, fid, lid,
 618                                 DTO_INDEX_DELETE, false,
 619                                 (val == SCRUB_NEXT_OSTOBJ ||
 620                                  val == SCRUB_NEXT_OSTOBJ_OLD) ?
 621                                 OI_KNOWN_ON_OST : 0);
 622         up_write(&scrub->os_rwsem);
 623
 624         if (inode != NULL && !IS_ERR(inode))
 625                 iput(inode);
 626
 627         if (oii != NULL) {
 628                 LASSERT(!list_empty(&oii->oii_list));
 629
 630                 spin_lock(&scrub->os_lock);
 631                 list_del_init(&oii->oii_list);
 632                 spin_unlock(&scrub->os_lock);
 633                 OBD_FREE_PTR(oii);
 634         }
 635         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 636 }
 637
 638 /* OI scrub APIs */
 639
 640 static int osd_scrub_prep(struct osd_device *dev)
 641 {
 642         struct osd_scrub     *scrub  = &dev->od_scrub;
 643         struct ptlrpc_thread *thread = &scrub->os_thread;
 644         struct scrub_file    *sf     = &scrub->os_file;
 645         __u32                 flags  = scrub->os_start_flags;
 646         int                   rc;
 647         bool                  drop_dryrun = false;
 648         ENTRY;
 649
 650         CDEBUG(D_LFSCK, "%.16s: OI scrub prep, flags = 0x%x\n",
 651                osd_scrub2name(scrub), flags);
 652
 653         down_write(&scrub->os_rwsem);
 654         if (flags & SS_SET_FAILOUT)
 655                 sf->sf_param |= SP_FAILOUT;
 656         else if (flags & SS_CLEAR_FAILOUT)
 657                 sf->sf_param &= ~SP_FAILOUT;
 658
 659         if (flags & SS_SET_DRYRUN) {
 660                 sf->sf_param |= SP_DRYRUN;
 661         } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
 662                 sf->sf_param &= ~SP_DRYRUN;
 663                 drop_dryrun = true;
 664         }
 665
 666         if (flags & SS_RESET)
 667                 osd_scrub_file_reset(scrub,
 668                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
 669
 670         if (flags & SS_AUTO_FULL) {
 671                 scrub->os_full_speed = 1;
 672                 scrub->os_partial_scan = 0;
 673                 sf->sf_flags |= SF_AUTO;
 674         } else if (flags & SS_AUTO_PARTIAL) {
 675                 scrub->os_full_speed = 0;
 676                 scrub->os_partial_scan = 1;
 677                 sf->sf_flags |= SF_AUTO;
 678         } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
 679                                    SF_UPGRADE)) {
 680                 scrub->os_full_speed = 1;
 681                 scrub->os_partial_scan = 0;
 682         } else {
 683                 scrub->os_full_speed = 0;
 684                 scrub->os_partial_scan = 0;
 685         }
 686
 687         spin_lock(&scrub->os_lock);
 688         scrub->os_in_prior = 0;
 689         scrub->os_waiting = 0;
 690         scrub->os_paused = 0;
 691         scrub->os_in_join = 0;
 692         scrub->os_full_scrub = 0;
 693         spin_unlock(&scrub->os_lock);
 694         scrub->os_new_checked = 0;
 695         if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
 696                 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
 697         else if (sf->sf_pos_last_checkpoint != 0)
 698                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 699         else
 700                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 701
 702         scrub->os_pos_current = sf->sf_pos_latest_start;
 703         sf->sf_status = SS_SCANNING;
 704         sf->sf_time_latest_start = cfs_time_current_sec();
 705         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 706         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
 707         rc = osd_scrub_file_store(scrub);
 708         if (rc == 0) {
 709                 spin_lock(&scrub->os_lock);
 710                 thread_set_flags(thread, SVC_RUNNING);
 711                 spin_unlock(&scrub->os_lock);
 712                 wake_up_all(&thread->t_ctl_waitq);
 713         }
 714         up_write(&scrub->os_rwsem);
 715
 716         RETURN(rc);
 717 }
 718
 719 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
 720 {
 721         struct scrub_file *sf = &scrub->os_file;
 722         int                rc;
 723
 724         if (likely(cfs_time_before(cfs_time_current(),
 725                                    scrub->os_time_next_checkpoint) ||
 726                    scrub->os_new_checked == 0))
 727                 return 0;
 728
 729         down_write(&scrub->os_rwsem);
 730         sf->sf_items_checked += scrub->os_new_checked;
 731         scrub->os_new_checked = 0;
 732         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 733         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 734         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 735                                             scrub->os_time_last_checkpoint);
 736         rc = osd_scrub_file_store(scrub);
 737         up_write(&scrub->os_rwsem);
 738
 739         return rc;
 740 }
 741
 742 static void osd_scrub_post(struct osd_scrub *scrub, int result)
 743 {
 744         struct scrub_file *sf = &scrub->os_file;
 745         ENTRY;
 746
 747         CDEBUG(D_LFSCK, "%.16s: OI scrub post, result = %d\n",
 748                osd_scrub2name(scrub), result);
 749
 750         down_write(&scrub->os_rwsem);
 751         spin_lock(&scrub->os_lock);
 752         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 753         spin_unlock(&scrub->os_lock);
 754         if (scrub->os_new_checked > 0) {
 755                 sf->sf_items_checked += scrub->os_new_checked;
 756                 scrub->os_new_checked = 0;
 757                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 758         }
 759         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 760         if (result > 0) {
 761                 struct osd_device *dev =
 762                         container_of0(scrub, struct osd_device, od_scrub);
 763
 764                 dev->od_igif_inoi = 1;
 765                 dev->od_check_ff = 0;
 766                 sf->sf_status = SS_COMPLETED;
 767                 if (!(sf->sf_param & SP_DRYRUN)) {
 768                         memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 769                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 770                                           SF_UPGRADE | SF_AUTO);
 771                 }
 772                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 773                 sf->sf_success_count++;
 774         } else if (result == 0) {
 775                 if (scrub->os_paused)
 776                         sf->sf_status = SS_PAUSED;
 777                 else
 778                         sf->sf_status = SS_STOPPED;
 779         } else {
 780                 sf->sf_status = SS_FAILED;
 781         }
 782         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 783                                             scrub->os_time_last_checkpoint);
 784         result = osd_scrub_file_store(scrub);
 785         up_write(&scrub->os_rwsem);
 786
 787         EXIT;
 788 }
 789
 790 /* iteration engine */
 791
 792 struct osd_iit_param {
 793         struct super_block *sb;
 794         struct buffer_head *bitmap;
 795         ldiskfs_group_t bg;
 796         __u32 gbase;
 797         __u32 offset;
 798 };
 799
 800 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 801                                    struct osd_device *dev,
 802                                    struct osd_iit_param *param,
 803                                    struct osd_idmap_cache **oic,
 804                                    const bool noslot);
 805
 806 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 807                                    struct osd_device *dev,
 808                                    struct osd_iit_param *param,
 809                                    struct osd_idmap_cache *oic,
 810                                    bool *noslot, int rc);
 811
 812 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 813 {
 814         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 815                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 816         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 817                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 818                 return SCRUB_NEXT_BREAK;
 819         } else {
 820                 *pos = param->gbase + param->offset;
 821                 return 0;
 822         }
 823 }
 824
 825 /**
 826  * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
 827  * \retval 0: FID-on-MDT
 828  */
 829 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
 830                                       struct osd_device *dev,
 831                                       struct lu_fid *fid)
 832 {
 833         /* XXX: The initial OI scrub will scan the top level /O to generate
 834          *      a small local FLDB according to the <seq>. If the given FID
 835          *      is in the local FLDB, then it is FID-on-OST; otherwise it's
 836          *      quite possible for FID-on-MDT. */
 837         if (dev->od_is_ost)
 838                 return SCRUB_NEXT_OSTOBJ_OLD;
 839         else
 840                 return 0;
 841 }
 842
 843 static int osd_scrub_get_fid(struct osd_thread_info *info,
 844                              struct osd_device *dev, struct inode *inode,
 845                              struct lu_fid *fid, bool scrub)
 846 {
 847         struct lustre_mdt_attrs *lma     = &info->oti_mdt_attrs;
 848         int                      rc;
 849         bool                     has_lma = false;
 850
 851         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 852         if (rc == 0) {
 853                 has_lma = true;
 854                 if (lma->lma_compat & LMAC_NOT_IN_OI ||
 855                     lma->lma_incompat & LMAI_AGENT)
 856                         return SCRUB_NEXT_CONTINUE;
 857
 858                 *fid = lma->lma_self_fid;
 859                 if (!scrub)
 860                         return 0;
 861
 862                 if (lma->lma_compat & LMAC_FID_ON_OST)
 863                         return SCRUB_NEXT_OSTOBJ;
 864
 865                 if (fid_is_idif(fid))
 866                         return SCRUB_NEXT_OSTOBJ_OLD;
 867
 868                 /* For local object. */
 869                 if (fid_is_internal(fid))
 870                         return 0;
 871
 872                 /* For external visible MDT-object with non-normal FID. */
 873                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
 874                         return 0;
 875
 876                 /* For the object with normal FID, it may be MDT-object,
 877                  * or may be 2.4 OST-object, need further distinguish.
 878                  * Fall through to next section. */
 879         }
 880
 881         if (rc == -ENODATA || rc == 0) {
 882                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
 883                 if (rc == 0) {
 884                         if (scrub)
 885                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
 886                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 887                         return rc;
 888                 }
 889
 890                 if (rc > 0) {
 891                         if (!has_lma)
 892                                 /* It is FID-on-OST, but we do not know how
 893                                  * to generate its FID, ignore it directly. */
 894                                 rc = SCRUB_NEXT_CONTINUE;
 895                         else
 896                                 /* It is 2.4 OST-object. */
 897                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 898                         return rc;
 899                 }
 900
 901                 if (rc != -ENODATA)
 902                         return rc;
 903
 904                 if (!has_lma) {
 905                         if (dev->od_scrub.os_convert_igif) {
 906                                 lu_igif_build(fid, inode->i_ino,
 907                                               inode->i_generation);
 908                                 if (scrub)
 909                                         rc = SCRUB_NEXT_NOLMA;
 910                                 else
 911                                         rc = 0;
 912                         } else {
 913                                 /* It may be FID-on-OST, or may be FID for
 914                                  * non-MDT0, anyway, we do not know how to
 915                                  * generate its FID, ignore it directly. */
 916                                 rc = SCRUB_NEXT_CONTINUE;
 917                         }
 918                         return rc;
 919                 }
 920
 921                 /* For OI scrub case only: the object has LMA but has no ff
 922                  * (or ff crashed). It may be MDT-object, may be OST-object
 923                  * with crashed ff. The last check is local FLDB. */
 924                 rc = osd_scrub_check_local_fldb(info, dev, fid);
 925         }
 926
 927         return rc;
 928 }
 929
 930 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 931                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 932                         struct super_block *sb, bool scrub)
 933 {
 934         struct inode *inode;
 935         int           rc;
 936         ENTRY;
 937
 938         /* Not handle the backend root object and agent parent object.
 939          * They are neither visible to namespace nor have OI mappings. */
 940         if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
 941                      pos == osd_remote_parent_ino(dev)))
 942                 RETURN(SCRUB_NEXT_CONTINUE);
 943
 944         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 945         inode = osd_iget(info, dev, lid);
 946         if (IS_ERR(inode)) {
 947                 rc = PTR_ERR(inode);
 948                 /* The inode may be removed after bitmap searching, or the
 949                  * file is new created without inode initialized yet. */
 950                 if (rc == -ENOENT || rc == -ESTALE)
 951                         RETURN(SCRUB_NEXT_CONTINUE);
 952
 953                 CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: "
 954                        "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name,
 955                        pos, rc);
 956                 RETURN(rc);
 957         }
 958
 959         if (scrub &&
 960             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 961                 /* Only skip it for the first OI scrub accessing. */
 962                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 963                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
 964         }
 965
 966         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
 967
 968         GOTO(put, rc);
 969
 970 put:
 971         iput(inode);
 972         return rc;
 973 }
 974
 975 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 976                           struct osd_iit_param *param,
 977                           struct osd_idmap_cache **oic, const bool noslot)
 978 {
 979         struct osd_scrub     *scrub  = &dev->od_scrub;
 980         struct ptlrpc_thread *thread = &scrub->os_thread;
 981         struct lu_fid        *fid;
 982         struct osd_inode_id  *lid;
 983         int                   rc;
 984
 985         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
 986                 struct l_wait_info lwi;
 987
 988                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
 989                 if (likely(lwi.lwi_timeout > 0))
 990                         l_wait_event(thread->t_ctl_waitq,
 991                                 !list_empty(&scrub->os_inconsistent_items) ||
 992                                 !thread_is_running(thread),
 993                                 &lwi);
 994         }
 995
 996         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 997                 spin_lock(&scrub->os_lock);
 998                 thread_set_flags(thread, SVC_STOPPING);
 999                 spin_unlock(&scrub->os_lock);
1000                 return SCRUB_NEXT_CRASH;
1001         }
1002
1003         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
1004                 return SCRUB_NEXT_FATAL;
1005
1006         if (unlikely(!thread_is_running(thread)))
1007                 return SCRUB_NEXT_EXIT;
1008
1009         if (!list_empty(&scrub->os_inconsistent_items)) {
1010                 struct osd_inconsistent_item *oii;
1011
1012                 oii = list_entry(scrub->os_inconsistent_items.next,
1013                                  struct osd_inconsistent_item, oii_list);
1014                 *oic = &oii->oii_cache;
1015                 scrub->os_in_prior = 1;
1016                 return 0;
1017         }
1018
1019         if (noslot)
1020                 return SCRUB_NEXT_WAIT;
1021
1022         rc = osd_iit_next(param, &scrub->os_pos_current);
1023         if (rc != 0)
1024                 return rc;
1025
1026         *oic = &scrub->os_oic;
1027         fid = &(*oic)->oic_fid;
1028         lid = &(*oic)->oic_lid;
1029         rc = osd_iit_iget(info, dev, fid, lid,
1030                           scrub->os_pos_current, param->sb, true);
1031         return rc;
1032 }
1033
1034 static int osd_preload_next(struct osd_thread_info *info,
1035                             struct osd_device *dev, struct osd_iit_param *param,
1036                             struct osd_idmap_cache **oic, const bool noslot)
1037 {
1038         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
1039         struct osd_scrub        *scrub;
1040         struct ptlrpc_thread    *thread;
1041         int                      rc;
1042
1043         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1044         if (rc != 0)
1045                 return rc;
1046
1047         scrub = &dev->od_scrub;
1048         thread = &scrub->os_thread;
1049         if (thread_is_running(thread) &&
1050             ooc->ooc_pos_preload >= scrub->os_pos_current)
1051                 return SCRUB_NEXT_EXIT;
1052
1053         rc = osd_iit_iget(info, dev,
1054                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1055                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1056                           ooc->ooc_pos_preload, param->sb, false);
1057         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1058          * ignore the failure, so it still need to skip the inode next time. */
1059         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1060         return rc;
1061 }
1062
1063 static inline int
1064 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1065 {
1066         spin_lock(&scrub->os_lock);
1067         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1068             !list_empty(&scrub->os_inconsistent_items) ||
1069             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1070                 scrub->os_waiting = 0;
1071         else
1072                 scrub->os_waiting = 1;
1073         spin_unlock(&scrub->os_lock);
1074
1075         return !scrub->os_waiting;
1076 }
1077
1078 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1079                           struct osd_iit_param *param,
1080                           struct osd_idmap_cache *oic, bool *noslot, int rc)
1081 {
1082         struct l_wait_info       lwi    = { 0 };
1083         struct osd_scrub        *scrub  = &dev->od_scrub;
1084         struct scrub_file       *sf     = &scrub->os_file;
1085         struct ptlrpc_thread    *thread = &scrub->os_thread;
1086         struct osd_otable_it    *it     = dev->od_otable_it;
1087         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
1088
1089         switch (rc) {
1090         case SCRUB_NEXT_CONTINUE:
1091                 goto next;
1092         case SCRUB_NEXT_WAIT:
1093                 goto wait;
1094         case SCRUB_NEXT_NOSCRUB:
1095                 down_write(&scrub->os_rwsem);
1096                 scrub->os_new_checked++;
1097                 sf->sf_items_noscrub++;
1098                 up_write(&scrub->os_rwsem);
1099                 goto next;
1100         }
1101
1102         rc = osd_scrub_check_update(info, dev, oic, rc);
1103         if (rc != 0)
1104                 return rc;
1105
1106         rc = osd_scrub_checkpoint(scrub);
1107         if (rc != 0) {
1108                 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1109                        "rc = %d\n", osd_scrub2name(scrub),
1110                        scrub->os_pos_current, rc);
1111                 /* Continue, as long as the scrub itself can go ahead. */
1112         }
1113
1114         if (scrub->os_in_prior) {
1115                 scrub->os_in_prior = 0;
1116                 return 0;
1117         }
1118
1119 next:
1120         scrub->os_pos_current = param->gbase + ++(param->offset);
1121
1122 wait:
1123         if (it != NULL && it->ooi_waiting && ooc != NULL &&
1124             ooc->ooc_pos_preload < scrub->os_pos_current) {
1125                 spin_lock(&scrub->os_lock);
1126                 it->ooi_waiting = 0;
1127                 wake_up_all(&thread->t_ctl_waitq);
1128                 spin_unlock(&scrub->os_lock);
1129         }
1130
1131         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1132                 return 0;
1133
1134         if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1135                 *noslot = false;
1136                 return 0;
1137         }
1138
1139         if (it != NULL)
1140                 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1141                              &lwi);
1142
1143         if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1144                 *noslot = false;
1145         else
1146                 *noslot = true;
1147         return 0;
1148 }
1149
1150 static int osd_preload_exec(struct osd_thread_info *info,
1151                             struct osd_device *dev, struct osd_iit_param *param,
1152                             struct osd_idmap_cache *oic, bool *noslot, int rc)
1153 {
1154         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1155
1156         if (rc == 0) {
1157                 ooc->ooc_cached_items++;
1158                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1159                                         ~OSD_OTABLE_IT_CACHE_MASK;
1160         }
1161         return rc > 0 ? 0 : rc;
1162 }
1163
1164 #define SCRUB_IT_ALL    1
1165 #define SCRUB_IT_CRASH  2
1166
1167 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1168                            bool inconsistent)
1169 {
1170         struct osd_scrub     *scrub  = &dev->od_scrub;
1171         struct ptlrpc_thread *thread = &scrub->os_thread;
1172         struct scrub_file    *sf     = &scrub->os_file;
1173         int                   rc;
1174         ENTRY;
1175
1176         LASSERT(!(flags & SS_AUTO_PARTIAL));
1177
1178         down_write(&scrub->os_rwsem);
1179         scrub->os_in_join = 1;
1180         if (flags & SS_SET_FAILOUT)
1181                 sf->sf_param |= SP_FAILOUT;
1182         else if (flags & SS_CLEAR_FAILOUT)
1183                 sf->sf_param &= ~SP_FAILOUT;
1184
1185         if (flags & SS_SET_DRYRUN)
1186                 sf->sf_param |= SP_DRYRUN;
1187         else if (flags & SS_CLEAR_DRYRUN)
1188                 sf->sf_param &= ~SP_DRYRUN;
1189
1190         if (flags & SS_RESET) {
1191                 osd_scrub_file_reset(scrub,
1192                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1193                         inconsistent ? SF_INCONSISTENT : 0);
1194                 sf->sf_status = SS_SCANNING;
1195         }
1196
1197         if (flags & SS_AUTO_FULL) {
1198                 sf->sf_flags |= SF_AUTO;
1199                 scrub->os_full_speed = 1;
1200         }
1201
1202         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1203                 scrub->os_full_speed = 1;
1204         else
1205                 scrub->os_full_speed = 0;
1206
1207         scrub->os_new_checked = 0;
1208         if (sf->sf_pos_last_checkpoint != 0)
1209                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1210         else
1211                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1212
1213         scrub->os_pos_current = sf->sf_pos_latest_start;
1214         sf->sf_time_latest_start = cfs_time_current_sec();
1215         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1216         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1217         rc = osd_scrub_file_store(scrub);
1218         if (rc != 0)
1219                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
1220                        "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1221
1222         spin_lock(&scrub->os_lock);
1223         scrub->os_waiting = 0;
1224         scrub->os_paused = 0;
1225         scrub->os_partial_scan = 0;
1226         scrub->os_in_join = 0;
1227         scrub->os_full_scrub = 0;
1228         spin_unlock(&scrub->os_lock);
1229         wake_up_all(&thread->t_ctl_waitq);
1230         up_write(&scrub->os_rwsem);
1231
1232         EXIT;
1233 }
1234
1235 static int osd_inode_iteration(struct osd_thread_info *info,
1236                                struct osd_device *dev, __u32 max, bool preload)
1237 {
1238         struct osd_scrub     *scrub  = &dev->od_scrub;
1239         struct ptlrpc_thread *thread = &scrub->os_thread;
1240         struct scrub_file    *sf     = &scrub->os_file;
1241         osd_iit_next_policy   next;
1242         osd_iit_exec_policy   exec;
1243         __u32                *pos;
1244         __u32                *count;
1245         struct osd_iit_param  param  = { NULL };
1246         struct l_wait_info    lwi    = { 0 };
1247         __u32                 limit;
1248         int                   rc;
1249         bool                  noslot = true;
1250         ENTRY;
1251
1252         param.sb = osd_sb(dev);
1253         if (preload)
1254                 goto full;
1255
1256         while (scrub->os_partial_scan && !scrub->os_in_join) {
1257                 struct osd_idmap_cache *oic = NULL;
1258
1259                 rc = osd_scrub_next(info, dev, &param, &oic, noslot);
1260                 switch (rc) {
1261                 case SCRUB_NEXT_EXIT:
1262                         RETURN(0);
1263                 case SCRUB_NEXT_CRASH:
1264                         RETURN(SCRUB_IT_CRASH);
1265                 case SCRUB_NEXT_FATAL:
1266                         RETURN(-EINVAL);
1267                 case SCRUB_NEXT_WAIT: {
1268                         struct kstatfs *ksfs = &info->oti_ksfs;
1269                         __u64 saved_flags;
1270
1271                         if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1272                             unlikely(sf->sf_items_updated_prior == 0))
1273                                 goto wait;
1274
1275                         if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1276                             scrub->os_full_scrub) {
1277                                 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1278                                                true);
1279                                 goto full;
1280                         }
1281
1282                         rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1283                         if (rc == 0) {
1284                                 __u64 used = ksfs->f_files - ksfs->f_ffree;
1285
1286                                 do_div(used, sf->sf_items_updated_prior);
1287                                 /* If we hit too much inconsistent OI
1288                                  * mappings during the partial scan,
1289                                  * then scan the device completely. */
1290                                 if (used < dev->od_full_scrub_ratio) {
1291                                         osd_scrub_join(dev,
1292                                                 SS_AUTO_FULL | SS_RESET, true);
1293                                         goto full;
1294                                 }
1295                         }
1296
1297 wait:
1298                         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1299                             cfs_fail_val > 0)
1300                                 continue;
1301
1302                         saved_flags = sf->sf_flags;
1303                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1304                                           SF_UPGRADE | SF_AUTO);
1305                         sf->sf_status = SS_COMPLETED;
1306                         l_wait_event(thread->t_ctl_waitq,
1307                                      !thread_is_running(thread) ||
1308                                      !scrub->os_partial_scan ||
1309                                      scrub->os_in_join ||
1310                                      !list_empty(&scrub->os_inconsistent_items),
1311                                      &lwi);
1312                         sf->sf_flags = saved_flags;
1313                         sf->sf_status = SS_SCANNING;
1314
1315                         if (unlikely(!thread_is_running(thread)))
1316                                 RETURN(0);
1317
1318                         if (!scrub->os_partial_scan || scrub->os_in_join)
1319                                 goto full;
1320
1321                         continue;
1322                 }
1323                 default:
1324                         LASSERTF(rc == 0, "rc = %d\n", rc);
1325
1326                         osd_scrub_exec(info, dev, &param, oic, &noslot, rc);
1327                         break;
1328                 }
1329         }
1330
1331 full:
1332         if (!preload) {
1333                 l_wait_event(thread->t_ctl_waitq,
1334                              !thread_is_running(thread) || !scrub->os_in_join,
1335                              &lwi);
1336
1337                 if (unlikely(!thread_is_running(thread)))
1338                         RETURN(0);
1339         }
1340
1341         noslot = false;
1342         if (!preload) {
1343                 next = osd_scrub_next;
1344                 exec = osd_scrub_exec;
1345                 pos = &scrub->os_pos_current;
1346                 count = &scrub->os_new_checked;
1347         } else {
1348                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1349
1350                 next = osd_preload_next;
1351                 exec = osd_preload_exec;
1352                 pos = &ooc->ooc_pos_preload;
1353                 count = &ooc->ooc_cached_items;
1354         }
1355         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1356
1357         while (*pos <= limit && *count < max) {
1358                 struct osd_idmap_cache *oic = NULL;
1359                 struct ldiskfs_group_desc *desc;
1360
1361                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1362                 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1363                 if (desc == NULL)
1364                         RETURN(-EIO);
1365
1366                 ldiskfs_lock_group(param.sb, param.bg);
1367                 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1368                         ldiskfs_unlock_group(param.sb, param.bg);
1369                         *pos = 1 + (param.bg + 1) *
1370                                 LDISKFS_INODES_PER_GROUP(param.sb);
1371                         continue;
1372                 }
1373                 ldiskfs_unlock_group(param.sb, param.bg);
1374
1375                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1376                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1377                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1378                 if (param.bitmap == NULL) {
1379                         CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1380                                "scrub will stop, urgent mode\n",
1381                                osd_scrub2name(scrub), (__u32)param.bg);
1382                         RETURN(-EIO);
1383                 }
1384
1385                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1386                        *count < max) {
1387                         if (param.offset +
1388                                 ldiskfs_itable_unused_count(param.sb, desc) >
1389                             LDISKFS_INODES_PER_GROUP(param.sb))
1390                                 goto next_group;
1391
1392                         rc = next(info, dev, &param, &oic, noslot);
1393                         switch (rc) {
1394                         case SCRUB_NEXT_BREAK:
1395                                 goto next_group;
1396                         case SCRUB_NEXT_EXIT:
1397                                 brelse(param.bitmap);
1398                                 RETURN(0);
1399                         case SCRUB_NEXT_CRASH:
1400                                 brelse(param.bitmap);
1401                                 RETURN(SCRUB_IT_CRASH);
1402                         case SCRUB_NEXT_FATAL:
1403                                 brelse(param.bitmap);
1404                                 RETURN(-EINVAL);
1405                         }
1406
1407                         rc = exec(info, dev, &param, oic, &noslot, rc);
1408                         if (rc != 0) {
1409                                 brelse(param.bitmap);
1410                                 RETURN(rc);
1411                         }
1412                 }
1413
1414 next_group:
1415                 brelse(param.bitmap);
1416         }
1417
1418         if (*pos > limit)
1419                 RETURN(SCRUB_IT_ALL);
1420         RETURN(0);
1421 }
1422
1423 static int osd_otable_it_preload(const struct lu_env *env,
1424                                  struct osd_otable_it *it)
1425 {
1426         struct osd_device       *dev   = it->ooi_dev;
1427         struct osd_scrub        *scrub = &dev->od_scrub;
1428         struct osd_otable_cache *ooc   = &it->ooi_cache;
1429         int                      rc;
1430         ENTRY;
1431
1432         rc = osd_inode_iteration(osd_oti_get(env), dev,
1433                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1434         if (rc == SCRUB_IT_ALL)
1435                 it->ooi_all_cached = 1;
1436
1437         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1438                 spin_lock(&scrub->os_lock);
1439                 scrub->os_waiting = 0;
1440                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1441                 spin_unlock(&scrub->os_lock);
1442         }
1443
1444         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1445 }
1446
1447 static int osd_scrub_main(void *args)
1448 {
1449         struct lu_env         env;
1450         struct osd_device    *dev    = (struct osd_device *)args;
1451         struct osd_scrub     *scrub  = &dev->od_scrub;
1452         struct ptlrpc_thread *thread = &scrub->os_thread;
1453         int                   rc;
1454         ENTRY;
1455
1456         rc = lu_env_init(&env, LCT_LOCAL);
1457         if (rc != 0) {
1458                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1459                        osd_scrub2name(scrub), rc);
1460                 GOTO(noenv, rc);
1461         }
1462
1463         rc = osd_scrub_prep(dev);
1464         if (rc != 0) {
1465                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1466                        osd_scrub2name(scrub), rc);
1467                 GOTO(out, rc);
1468         }
1469
1470         if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1471                 struct l_wait_info lwi = { 0 };
1472                 struct osd_otable_it *it = dev->od_otable_it;
1473                 struct osd_otable_cache *ooc = &it->ooi_cache;
1474
1475                 l_wait_event(thread->t_ctl_waitq,
1476                              it->ooi_user_ready || !thread_is_running(thread),
1477                              &lwi);
1478                 if (unlikely(!thread_is_running(thread)))
1479                         GOTO(post, rc = 0);
1480
1481                 scrub->os_pos_current = ooc->ooc_pos_preload;
1482         }
1483
1484         CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1485                osd_scrub2name(scrub), scrub->os_start_flags,
1486                scrub->os_pos_current);
1487
1488         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1489         if (unlikely(rc == SCRUB_IT_CRASH))
1490                 GOTO(out, rc = -EINVAL);
1491         GOTO(post, rc);
1492
1493 post:
1494         osd_scrub_post(scrub, rc);
1495         CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1496                osd_scrub2name(scrub), scrub->os_pos_current, rc);
1497
1498 out:
1499         while (!list_empty(&scrub->os_inconsistent_items)) {
1500                 struct osd_inconsistent_item *oii;
1501
1502                 oii = list_entry(scrub->os_inconsistent_items.next,
1503                                      struct osd_inconsistent_item, oii_list);
1504                 list_del_init(&oii->oii_list);
1505                 OBD_FREE_PTR(oii);
1506         }
1507         lu_env_fini(&env);
1508
1509 noenv:
1510         spin_lock(&scrub->os_lock);
1511         thread_set_flags(thread, SVC_STOPPED);
1512         wake_up_all(&thread->t_ctl_waitq);
1513         spin_unlock(&scrub->os_lock);
1514         return rc;
1515 }
1516
1517 /* initial OI scrub */
1518
1519 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1520                          struct dentry *, filldir_t filldir);
1521
1522 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1523                                loff_t offset, __u64 ino, unsigned d_type);
1524 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1525                            loff_t offset, __u64 ino, unsigned d_type);
1526 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1527                            loff_t offset, __u64 ino, unsigned d_type);
1528
1529 static int
1530 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1531                      struct dentry *dentry, filldir_t filldir);
1532 static int
1533 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1534                   struct dentry *dentry, filldir_t filldir);
1535
1536 static int
1537 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1538                      struct dentry *dentry, filldir_t filldir);
1539
1540 enum osd_lf_flags {
1541         OLF_SCAN_SUBITEMS       = 0x0001,
1542         OLF_HIDE_FID            = 0x0002,
1543         OLF_SHOW_NAME           = 0x0004,
1544         OLF_NO_OI               = 0x0008,
1545 };
1546
1547 struct osd_lf_map {
1548         char            *olm_name;
1549         struct lu_fid    olm_fid;
1550         __u16            olm_flags;
1551         __u16            olm_namelen;
1552         scandir_t        olm_scandir;
1553         filldir_t        olm_filldir;
1554 };
1555
1556 /* Add the new introduced local files in the list in the future. */
1557 static const struct osd_lf_map osd_lf_maps[] = {
1558         /* CATALOGS */
1559         { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1560                 sizeof(CATLIST) - 1, NULL, NULL },
1561
1562         /* CONFIGS */
1563         { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1564                 OLF_SCAN_SUBITEMS, sizeof(MOUNT_CONFIGS_DIR) - 1,
1565                 osd_ios_general_scan, osd_ios_varfid_fill },
1566
1567         /* NIDTBL_VERSIONS */
1568         { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1569                 sizeof(MGS_NIDTBL_DIR) - 1, osd_ios_general_scan,
1570                 osd_ios_varfid_fill },
1571
1572         /* PENDING */
1573         { "PENDING", { 0, 0, 0 }, 0, sizeof("PENDING") - 1, NULL, NULL },
1574
1575         /* ROOT */
1576         { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 },
1577                 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, sizeof("ROOT") - 1,
1578                 osd_ios_ROOT_scan, NULL },
1579
1580         /* changelog_catalog */
1581         { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, sizeof(CHANGELOG_CATALOG) - 1,
1582                 NULL, NULL },
1583
1584         /* changelog_users */
1585         { CHANGELOG_USERS, { 0, 0, 0 }, 0, sizeof(CHANGELOG_USERS) - 1,
1586                 NULL, NULL },
1587
1588         /* fld */
1589         { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1590                 sizeof("fld") - 1, NULL, NULL },
1591
1592         /* last_rcvd */
1593         { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1594                 sizeof(LAST_RCVD) - 1, NULL, NULL },
1595
1596         /* lov_objid */
1597         { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1598                 sizeof(LOV_OBJID) - 1, NULL, NULL },
1599
1600         /* lov_objseq */
1601         { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1602                 OLF_SHOW_NAME, sizeof(LOV_OBJSEQ) - 1, NULL, NULL },
1603
1604         /* quota_master */
1605         { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QMT_DIR) - 1,
1606                 osd_ios_general_scan, osd_ios_varfid_fill },
1607
1608         /* quota_slave */
1609         { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QSD_DIR) - 1,
1610                 osd_ios_general_scan, osd_ios_varfid_fill },
1611
1612         /* seq_ctl */
1613         { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1614                 OLF_SHOW_NAME, sizeof("seq_ctl") - 1, NULL, NULL },
1615
1616         /* seq_srv */
1617         { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1618                 OLF_SHOW_NAME, sizeof("seq_srv") - 1, NULL, NULL },
1619
1620         /* health_check */
1621         { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1622                 OLF_SHOW_NAME, sizeof(HEALTH_CHECK) - 1, NULL, NULL },
1623
1624         /* LFSCK */
1625         { LFSCK_DIR, { 0, 0, 0 }, 0, sizeof(LFSCK_DIR) - 1,
1626                 osd_ios_general_scan, osd_ios_varfid_fill },
1627
1628         /* lfsck_bookmark */
1629         { LFSCK_BOOKMARK, { 0, 0, 0 }, 0, sizeof(LFSCK_BOOKMARK) - 1,
1630                 NULL, NULL },
1631
1632         /* lfsck_layout */
1633         { LFSCK_LAYOUT, { 0, 0, 0 }, 0, sizeof(LFSCK_LAYOUT) - 1,
1634                 NULL, NULL },
1635
1636         /* lfsck_namespace */
1637         { LFSCK_NAMESPACE, { 0, 0, 0 }, 0, sizeof(LFSCK_NAMESPACE) - 1,
1638                 NULL, NULL },
1639
1640         /* OBJECTS, upgrade from old device */
1641         { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(OBJECTS) - 1,
1642                 osd_ios_OBJECTS_scan, NULL },
1643
1644         /* lquota_v2.user, upgrade from old device */
1645         { "lquota_v2.user", { 0, 0, 0 }, 0, sizeof("lquota_v2.user") - 1,
1646                 NULL, NULL },
1647
1648         /* lquota_v2.group, upgrade from old device */
1649         { "lquota_v2.group", { 0, 0, 0 }, 0, sizeof("lquota_v2.group") - 1,
1650                 NULL, NULL },
1651
1652         /* LAST_GROUP, upgrade from old device */
1653         { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1654                 OLF_SHOW_NAME, sizeof("LAST_GROUP") - 1, NULL, NULL },
1655
1656         /* SLAVE_LOG, llog for destroy slave stripes of striped dir */
1657         { "SLAVE_LOG", { FID_SEQ_LOCAL_FILE, SLAVE_LLOG_CATALOGS_OID, 0 },
1658                OLF_SHOW_NAME, sizeof("SLAVE_LOG") - 1, NULL, NULL },
1659
1660         /* lost+found */
1661         { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 },
1662                 OLF_SCAN_SUBITEMS, sizeof("lost+found") - 1,
1663                 osd_ios_general_scan, osd_ios_lf_fill },
1664
1665         { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1666 };
1667
1668 /* Add the new introduced files under .lustre/ in the list in the future. */
1669 static const struct osd_lf_map osd_dl_maps[] = {
1670         /* .lustre/fid */
1671         { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1672                 sizeof("fid") - 1, NULL, NULL },
1673
1674         /* .lustre/lost+found */
1675         { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1676                 sizeof("lost+found") - 1, NULL, NULL },
1677
1678         { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1679 };
1680
1681 struct osd_ios_item {
1682         struct list_head oii_list;
1683         struct dentry   *oii_dentry;
1684         scandir_t        oii_scandir;
1685         filldir_t        oii_filldir;
1686 };
1687
1688 struct osd_ios_filldir_buf {
1689 #ifdef HAVE_DIR_CONTEXT
1690         /* please keep it as first member */
1691         struct dir_context       ctx;
1692 #endif
1693         struct osd_thread_info  *oifb_info;
1694         struct osd_device       *oifb_dev;
1695         struct dentry           *oifb_dentry;
1696 };
1697
1698 static inline struct dentry *
1699 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1700 {
1701         struct dentry *dentry;
1702
1703         dentry = ll_lookup_one_len(name, parent, namelen);
1704         if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1705                 dput(dentry);
1706                 return ERR_PTR(-ENOENT);
1707         }
1708
1709         return dentry;
1710 }
1711
1712 static int
1713 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1714                  scandir_t scandir, filldir_t filldir)
1715 {
1716         struct osd_ios_item *item;
1717         ENTRY;
1718
1719         OBD_ALLOC_PTR(item);
1720         if (item == NULL)
1721                 RETURN(-ENOMEM);
1722
1723         INIT_LIST_HEAD(&item->oii_list);
1724         item->oii_dentry = dget(dentry);
1725         item->oii_scandir = scandir;
1726         item->oii_filldir = filldir;
1727         list_add_tail(&item->oii_list, &dev->od_ios_list);
1728
1729         RETURN(0);
1730 }
1731
1732 /**
1733  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1734  *
1735  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1736  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1737  * reference the inode, or fixed if it is missing or references another inode.
1738  */
1739 static int
1740 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1741                  struct inode *inode, const struct lu_fid *fid, int flags)
1742 {
1743         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
1744         struct osd_inode_id     *id     = &info->oti_id;
1745         struct osd_inode_id     *id2    = &info->oti_id2;
1746         struct osd_scrub        *scrub  = &dev->od_scrub;
1747         struct scrub_file       *sf     = &scrub->os_file;
1748         struct lu_fid            tfid;
1749         int                      rc;
1750         ENTRY;
1751
1752         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1753         if (rc != 0 && rc != -ENODATA) {
1754                 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1755                        "rc = %d\n", osd_name(dev), rc);
1756
1757                 RETURN(rc);
1758         }
1759
1760         osd_id_gen(id, inode->i_ino, inode->i_generation);
1761         if (rc == -ENODATA) {
1762                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID)
1763                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1764                 else
1765                         tfid = *fid;
1766                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1767                 if (rc != 0) {
1768                         CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1769                               "scrub: rc = %d\n", osd_name(dev), rc);
1770
1771                         RETURN(rc);
1772                 }
1773         } else {
1774                 if (lma->lma_compat & LMAC_NOT_IN_OI)
1775                         RETURN(0);
1776
1777                 tfid = lma->lma_self_fid;
1778         }
1779
1780         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1781         if (rc != 0) {
1782                 if (rc != -ENOENT)
1783                         RETURN(rc);
1784
1785                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1786                                                DTO_INDEX_INSERT, true, 0);
1787                 if (rc > 0)
1788                         rc = 0;
1789
1790                 RETURN(rc);
1791         }
1792
1793         if (osd_id_eq_strict(id, id2))
1794                 RETURN(0);
1795
1796         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1797                 osd_scrub_file_reset(scrub,
1798                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1799                                      SF_INCONSISTENT);
1800                 rc = osd_scrub_file_store(scrub);
1801                 if (rc != 0)
1802                         RETURN(rc);
1803         }
1804
1805         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1806                                        DTO_INDEX_UPDATE, true, 0);
1807         if (rc > 0)
1808                 rc = 0;
1809
1810         RETURN(rc);
1811 }
1812
1813 /**
1814  * It scans the /lost+found, and for the OST-object (with filter_fid
1815  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1816  */
1817 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1818                            loff_t offset, __u64 ino, unsigned d_type)
1819 {
1820         struct osd_ios_filldir_buf *fill_buf = buf;
1821         struct osd_thread_info     *info     = fill_buf->oifb_info;
1822         struct osd_device          *dev      = fill_buf->oifb_dev;
1823         struct lu_fid              *fid      = &info->oti_fid;
1824         struct osd_scrub           *scrub    = &dev->od_scrub;
1825         struct dentry              *parent   = fill_buf->oifb_dentry;
1826         struct dentry              *child;
1827         struct inode               *dir      = parent->d_inode;
1828         struct inode               *inode;
1829         int                         rc;
1830         ENTRY;
1831
1832         /* skip any '.' started names */
1833         if (name[0] == '.')
1834                 RETURN(0);
1835
1836         scrub->os_lf_scanned++;
1837         child = osd_ios_lookup_one_len(name, parent, namelen);
1838         if (IS_ERR(child)) {
1839                 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1840                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
1841                 RETURN(0);
1842         }
1843
1844         inode = child->d_inode;
1845         if (S_ISDIR(inode->i_mode)) {
1846                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1847                                       osd_ios_lf_fill);
1848                 if (rc != 0)
1849                         CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1850                               "rc = %d\n", osd_name(dev), namelen, name, rc);
1851                 GOTO(put, rc);
1852         }
1853
1854         if (!S_ISREG(inode->i_mode))
1855                 GOTO(put, rc = 0);
1856
1857         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1858         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1859                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1860                 if (rc == 0) {
1861                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1862                                "/lost+found.\n", namelen, name, PFID(fid));
1863                         scrub->os_lf_repaired++;
1864                 } else {
1865                         CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1866                                DFID": rc = %d\n",
1867                                osd_name(dev), namelen, name, PFID(fid), rc);
1868                 }
1869         }
1870
1871         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1872          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1873          *      can process them in furtuer. */
1874
1875         GOTO(put, rc);
1876
1877 put:
1878         if (rc < 0)
1879                 scrub->os_lf_failed++;
1880         dput(child);
1881         /* skip the failure to make the scanning to continue. */
1882         return 0;
1883 }
1884
1885 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1886                                loff_t offset, __u64 ino, unsigned d_type)
1887 {
1888         struct osd_ios_filldir_buf *fill_buf = buf;
1889         struct osd_device          *dev      = fill_buf->oifb_dev;
1890         struct dentry              *child;
1891         int                         rc;
1892         ENTRY;
1893
1894         /* skip any '.' started names */
1895         if (name[0] == '.')
1896                 RETURN(0);
1897
1898         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1899         if (IS_ERR(child))
1900                 RETURN(PTR_ERR(child));
1901
1902         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1903                               NULL, 0);
1904         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1905                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1906                                       osd_ios_varfid_fill);
1907         dput(child);
1908
1909         RETURN(rc);
1910 }
1911
1912 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1913                            loff_t offset, __u64 ino, unsigned d_type)
1914 {
1915         struct osd_ios_filldir_buf *fill_buf = buf;
1916         struct osd_device          *dev      = fill_buf->oifb_dev;
1917         const struct osd_lf_map    *map;
1918         struct dentry              *child;
1919         int                         rc       = 0;
1920         ENTRY;
1921
1922         /* skip any '.' started names */
1923         if (name[0] == '.')
1924                 RETURN(0);
1925
1926         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
1927                 if (map->olm_namelen != namelen)
1928                         continue;
1929
1930                 if (strncmp(map->olm_name, name, namelen) == 0)
1931                         break;
1932         }
1933
1934         if (map->olm_name == NULL)
1935                 RETURN(0);
1936
1937         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1938         if (IS_ERR(child))
1939                 RETURN(PTR_ERR(child));
1940
1941         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1942                               &map->olm_fid, map->olm_flags);
1943         dput(child);
1944
1945         RETURN(rc);
1946 }
1947
1948 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
1949                              loff_t offset, __u64 ino, unsigned d_type)
1950 {
1951         struct osd_ios_filldir_buf *fill_buf = buf;
1952         struct osd_device          *dev      = fill_buf->oifb_dev;
1953         const struct osd_lf_map    *map;
1954         struct dentry              *child;
1955         int                         rc       = 0;
1956         ENTRY;
1957
1958         /* skip any '.' started names */
1959         if (name[0] == '.')
1960                 RETURN(0);
1961
1962         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
1963                 if (map->olm_namelen != namelen)
1964                         continue;
1965
1966                 if (strncmp(map->olm_name, name, namelen) == 0)
1967                         break;
1968         }
1969
1970         if (map->olm_name == NULL)
1971                 RETURN(0);
1972
1973         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1974         if (IS_ERR(child))
1975                 RETURN(PTR_ERR(child));
1976
1977         if (!(map->olm_flags & OLF_NO_OI))
1978                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1979                                       &map->olm_fid, map->olm_flags);
1980         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
1981                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
1982                                       map->olm_filldir);
1983         dput(child);
1984
1985         RETURN(rc);
1986 }
1987
1988 static int
1989 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1990                      struct dentry *dentry, filldir_t filldir)
1991 {
1992         struct osd_ios_filldir_buf    buf   = {
1993 #ifdef HAVE_DIR_CONTEXT
1994                                                 .ctx.actor = filldir,
1995 #endif
1996                                                 .oifb_info = info,
1997                                                 .oifb_dev = dev,
1998                                                 .oifb_dentry = dentry };
1999         struct file                  *filp  = &info->oti_file;
2000         struct inode                 *inode = dentry->d_inode;
2001         const struct file_operations *fops  = inode->i_fop;
2002         int                           rc;
2003         ENTRY;
2004
2005         LASSERT(filldir != NULL);
2006
2007         filp->f_pos = 0;
2008         filp->f_dentry = dentry;
2009         filp->f_mode = FMODE_64BITHASH;
2010         filp->f_mapping = inode->i_mapping;
2011         filp->f_op = fops;
2012         filp->private_data = NULL;
2013         set_file_inode(filp, inode);
2014
2015 #ifdef HAVE_DIR_CONTEXT
2016         buf.ctx.pos = filp->f_pos;
2017         rc = fops->iterate(filp, &buf.ctx);
2018         filp->f_pos = buf.ctx.pos;
2019 #else
2020         rc = fops->readdir(filp, &buf, filldir);
2021 #endif
2022         fops->release(inode, filp);
2023
2024         RETURN(rc);
2025 }
2026
2027 static int
2028 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2029                   struct dentry *dentry, filldir_t filldir)
2030 {
2031         struct osd_scrub  *scrub  = &dev->od_scrub;
2032         struct scrub_file *sf     = &scrub->os_file;
2033         struct dentry     *child;
2034         int                rc;
2035         ENTRY;
2036
2037         /* It is existing MDT0 device. We only allow the case of object without
2038          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2039          * can generate IGIF mode FID for the object and related OI mapping. If
2040          * it is on other MDTs, then becuase file-level backup/restore, related
2041          * OI mapping may be invalid already, we do not know which is the right
2042          * FID for the object. We only allow IGIF objects to reside on the MDT0.
2043          *
2044          * XXX: For the case of object on non-MDT0 device with neither LMA nor
2045          *      "fid" xattr, then something crashed. We cannot re-generate the
2046          *      FID directly, instead, the OI scrub will scan the OI structure
2047          *      and try to re-generate the LMA from the OI mapping. But if the
2048          *      OI mapping crashed or lost also, then we have to give up under
2049          *      double failure cases. */
2050         scrub->os_convert_igif = 1;
2051         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2052                                        strlen(dot_lustre_name));
2053         if (IS_ERR(child)) {
2054                 rc = PTR_ERR(child);
2055                 if (rc == -ENOENT) {
2056                         /* It is 1.8 MDT device. */
2057                         if (!(sf->sf_flags & SF_UPGRADE)) {
2058                                 osd_scrub_file_reset(scrub,
2059                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2060                                         SF_UPGRADE);
2061                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2062                                 rc = osd_scrub_file_store(scrub);
2063                         } else {
2064                                 rc = 0;
2065                         }
2066                 }
2067         } else {
2068                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2069                  * so the client will get IGIF for the ".lustre" object when
2070                  * the MDT restart.
2071                  *
2072                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2073                  * it does not know whether there are some old clients cached
2074                  * the ".lustre" IGIF during the upgrading. Two choices:
2075                  *
2076                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2077                  *    It will allow the old connected clients to access the
2078                  *    ".lustre" with cached IGIF. But it will cause others
2079                  *    on the MDT failed to check "fid_is_dot_lustre()".
2080                  *
2081                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2082                  *    for ".lustre" in spite of whether there are some clients
2083                  *    cached the ".lustre" IGIF or not. It enables the check
2084                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
2085                  *    that the old connected clients cannot access the ".lustre"
2086                  *    with the cached IGIF.
2087                  *
2088                  * Usually, it is rare case for the old connected clients
2089                  * to access the ".lustre" with cached IGIF. So we prefer
2090                  * to the solution 2). */
2091                 rc = osd_ios_scan_one(info, dev, child->d_inode,
2092                                       &LU_DOT_LUSTRE_FID, 0);
2093                 if (rc == 0)
2094                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2095                                               osd_ios_dl_fill);
2096                 dput(child);
2097         }
2098
2099         RETURN(rc);
2100 }
2101
2102 static int
2103 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2104                      struct dentry *dentry, filldir_t filldir)
2105 {
2106         struct osd_scrub  *scrub  = &dev->od_scrub;
2107         struct scrub_file *sf     = &scrub->os_file;
2108         struct dentry     *child;
2109         int                rc;
2110         ENTRY;
2111
2112         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2113                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2114                 rc = osd_scrub_file_store(scrub);
2115                 if (rc != 0)
2116                         RETURN(rc);
2117         }
2118
2119         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2120         if (!IS_ERR(child)) {
2121                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2122                 dput(child);
2123         } else {
2124                 rc = PTR_ERR(child);
2125         }
2126
2127         if (rc != 0 && rc != -ENOENT)
2128                 RETURN(rc);
2129
2130         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2131         if (!IS_ERR(child)) {
2132                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2133                 dput(child);
2134         } else {
2135                 rc = PTR_ERR(child);
2136         }
2137
2138         if (rc == -ENOENT)
2139                 rc = 0;
2140
2141         RETURN(rc);
2142 }
2143
2144 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2145                                 struct osd_device *dev)
2146 {
2147         struct osd_ios_item     *item    = NULL;
2148         scandir_t                scandir = osd_ios_general_scan;
2149         filldir_t                filldir = osd_ios_root_fill;
2150         struct dentry           *dentry  = osd_sb(dev)->s_root;
2151         const struct osd_lf_map *map     = osd_lf_maps;
2152         int                      rc;
2153         ENTRY;
2154
2155         /* Lookup IGIF in OI by force for initial OI scrub. */
2156         dev->od_igif_inoi = 1;
2157
2158         while (1) {
2159                 rc = scandir(info, dev, dentry, filldir);
2160                 if (item != NULL) {
2161                         dput(item->oii_dentry);
2162                         OBD_FREE_PTR(item);
2163                 }
2164
2165                 if (rc != 0)
2166                         break;
2167
2168                 if (list_empty(&dev->od_ios_list))
2169                         break;
2170
2171                 item = list_entry(dev->od_ios_list.next,
2172                                   struct osd_ios_item, oii_list);
2173                 list_del_init(&item->oii_list);
2174
2175                 LASSERT(item->oii_scandir != NULL);
2176                 scandir = item->oii_scandir;
2177                 filldir = item->oii_filldir;
2178                 dentry = item->oii_dentry;
2179         }
2180
2181         while (!list_empty(&dev->od_ios_list)) {
2182                 item = list_entry(dev->od_ios_list.next,
2183                                   struct osd_ios_item, oii_list);
2184                 list_del_init(&item->oii_list);
2185                 dput(item->oii_dentry);
2186                 OBD_FREE_PTR(item);
2187         }
2188
2189         if (rc != 0)
2190                 RETURN(rc);
2191
2192         /* There maybe the case that the object has been removed, but its OI
2193          * mapping is still in the OI file, such as the "CATALOGS" after MDT
2194          * file-level backup/restore. So here cleanup the stale OI mappings. */
2195         while (map->olm_name != NULL) {
2196                 struct dentry *child;
2197
2198                 if (fid_is_zero(&map->olm_fid)) {
2199                         map++;
2200                         continue;
2201                 }
2202
2203                 child = osd_ios_lookup_one_len(map->olm_name,
2204                                                osd_sb(dev)->s_root,
2205                                                map->olm_namelen);
2206                 if (!IS_ERR(child))
2207                         dput(child);
2208                 else if (PTR_ERR(child) == -ENOENT)
2209                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2210                                                   NULL, DTO_INDEX_DELETE,
2211                                                   true, 0);
2212                 map++;
2213         }
2214
2215         RETURN(0);
2216 }
2217
2218 char *osd_lf_fid2name(const struct lu_fid *fid)
2219 {
2220         const struct osd_lf_map *map = osd_lf_maps;
2221
2222         while (map->olm_name != NULL) {
2223                 if (!lu_fid_eq(fid, &map->olm_fid)) {
2224                         map++;
2225                         continue;
2226                 }
2227
2228                 if (map->olm_flags & OLF_SHOW_NAME)
2229                         return map->olm_name;
2230                 else
2231                         return "";
2232         }
2233
2234         return NULL;
2235 }
2236
2237 /* OI scrub start/stop */
2238
2239 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2240 {
2241         struct osd_scrub     *scrub  = &dev->od_scrub;
2242         struct ptlrpc_thread *thread = &scrub->os_thread;
2243         struct l_wait_info    lwi    = { 0 };
2244         struct task_struct   *task;
2245         int                   rc;
2246         ENTRY;
2247
2248         /* os_lock: sync status between stop and scrub thread */
2249         spin_lock(&scrub->os_lock);
2250
2251 again:
2252         if (thread_is_running(thread)) {
2253                 spin_unlock(&scrub->os_lock);
2254                 if (!(scrub->os_file.sf_flags & SF_AUTO) ||
2255                      (flags & (SS_AUTO_FULL | SS_AUTO_PARTIAL)))
2256                         RETURN(-EALREADY);
2257
2258                 osd_scrub_join(dev, flags, false);
2259                 spin_lock(&scrub->os_lock);
2260                 if (!thread_is_running(thread))
2261                         goto again;
2262
2263                 spin_unlock(&scrub->os_lock);
2264                 RETURN(0);
2265         }
2266
2267         if (unlikely(thread_is_stopping(thread))) {
2268                 spin_unlock(&scrub->os_lock);
2269                 l_wait_event(thread->t_ctl_waitq,
2270                              thread_is_stopped(thread),
2271                              &lwi);
2272                 spin_lock(&scrub->os_lock);
2273                 goto again;
2274         }
2275         spin_unlock(&scrub->os_lock);
2276
2277         if (scrub->os_file.sf_status == SS_COMPLETED) {
2278                 if (!(flags & SS_SET_FAILOUT))
2279                         flags |= SS_CLEAR_FAILOUT;
2280
2281                 if (!(flags & SS_SET_DRYRUN))
2282                         flags |= SS_CLEAR_DRYRUN;
2283
2284                 flags |= SS_RESET;
2285         }
2286
2287         scrub->os_start_flags = flags;
2288         thread_set_flags(thread, 0);
2289         task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2290         if (IS_ERR(task)) {
2291                 rc = PTR_ERR(task);
2292                 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2293                        osd_scrub2name(scrub), rc);
2294                 RETURN(rc);
2295         }
2296
2297         l_wait_event(thread->t_ctl_waitq,
2298                      thread_is_running(thread) || thread_is_stopped(thread),
2299                      &lwi);
2300
2301         RETURN(0);
2302 }
2303
2304 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2305 {
2306         int rc;
2307         ENTRY;
2308
2309         /* od_otable_mutex: prevent curcurrent start/stop */
2310         mutex_lock(&dev->od_otable_mutex);
2311         rc = do_osd_scrub_start(dev, flags);
2312         mutex_unlock(&dev->od_otable_mutex);
2313
2314         RETURN(rc == -EALREADY ? 0 : rc);
2315 }
2316
2317 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2318 {
2319         struct ptlrpc_thread *thread = &scrub->os_thread;
2320         struct l_wait_info    lwi    = { 0 };
2321
2322         /* os_lock: sync status between stop and scrub thread */
2323         spin_lock(&scrub->os_lock);
2324         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2325                 thread_set_flags(thread, SVC_STOPPING);
2326                 spin_unlock(&scrub->os_lock);
2327                 wake_up_all(&thread->t_ctl_waitq);
2328                 l_wait_event(thread->t_ctl_waitq,
2329                              thread_is_stopped(thread),
2330                              &lwi);
2331                 /* Do not skip the last lock/unlock, which can guarantee that
2332                  * the caller cannot return until the OI scrub thread exit. */
2333                 spin_lock(&scrub->os_lock);
2334         }
2335         spin_unlock(&scrub->os_lock);
2336 }
2337
2338 static void osd_scrub_stop(struct osd_device *dev)
2339 {
2340         /* od_otable_mutex: prevent curcurrent start/stop */
2341         mutex_lock(&dev->od_otable_mutex);
2342         dev->od_scrub.os_paused = 1;
2343         do_osd_scrub_stop(&dev->od_scrub);
2344         mutex_unlock(&dev->od_otable_mutex);
2345 }
2346
2347 /* OI scrub setup/cleanup */
2348
2349 static const char osd_scrub_name[] = "OI_scrub";
2350
2351 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2352 {
2353         struct osd_thread_info     *info   = osd_oti_get(env);
2354         struct osd_scrub           *scrub  = &dev->od_scrub;
2355         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
2356         struct scrub_file          *sf     = &scrub->os_file;
2357         struct super_block         *sb     = osd_sb(dev);
2358         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
2359         struct lvfs_run_ctxt        saved;
2360         struct file                *filp;
2361         struct inode               *inode;
2362         struct lu_fid              *fid    = &info->oti_fid;
2363         int                         dirty  = 0;
2364         int                         rc     = 0;
2365         ENTRY;
2366
2367         memset(scrub, 0, sizeof(*scrub));
2368         OBD_SET_CTXT_MAGIC(ctxt);
2369         ctxt->pwdmnt = dev->od_mnt;
2370         ctxt->pwd = dev->od_mnt->mnt_root;
2371         ctxt->fs = get_ds();
2372
2373         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2374         init_rwsem(&scrub->os_rwsem);
2375         spin_lock_init(&scrub->os_lock);
2376         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2377
2378         push_ctxt(&saved, ctxt);
2379         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2380         if (IS_ERR(filp)) {
2381                 pop_ctxt(&saved, ctxt);
2382                 RETURN(PTR_ERR(filp));
2383         }
2384
2385         inode = filp->f_dentry->d_inode;
2386         /* 'What the @fid is' is not imporatant, because the object
2387          * has no OI mapping, and only is visible inside the OSD.*/
2388         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2389         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2390         if (rc != 0) {
2391                 filp_close(filp, NULL);
2392                 pop_ctxt(&saved, ctxt);
2393                 RETURN(rc);
2394         }
2395
2396         scrub->os_inode = igrab(inode);
2397         filp_close(filp, NULL);
2398         pop_ctxt(&saved, ctxt);
2399
2400         rc = osd_scrub_file_load(scrub);
2401         if (rc == -ENOENT) {
2402                 osd_scrub_file_init(scrub, es->s_uuid);
2403                 /* If the "/O" dir does not exist when mount (indicated by
2404                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2405                  * then it is quite probably that the device is a new one,
2406                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2407                  *
2408                  * For the rare case that "/O" and "OI_scrub" both lost on
2409                  * an old device, it can be found and cleared later.
2410                  *
2411                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2412                  * need to check "filter_fid_old" and to convert it to
2413                  * "filter_fid" for each object, and all the IGIF should
2414                  * have their FID mapping in OI files already. */
2415                 if (dev->od_maybe_new)
2416                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2417                 dirty = 1;
2418         } else if (rc != 0) {
2419                 GOTO(cleanup_inode, rc);
2420         } else {
2421                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2422                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2423                         dirty = 1;
2424                 } else if (sf->sf_status == SS_SCANNING) {
2425                         sf->sf_status = SS_CRASHED;
2426                         dirty = 1;
2427                 }
2428         }
2429
2430         if (sf->sf_pos_last_checkpoint != 0)
2431                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2432         else
2433                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2434
2435         if (dirty != 0) {
2436                 rc = osd_scrub_file_store(scrub);
2437                 if (rc != 0)
2438                         GOTO(cleanup_inode, rc);
2439         }
2440
2441         /* Initialize OI files. */
2442         rc = osd_oi_init(info, dev);
2443         if (rc < 0)
2444                 GOTO(cleanup_inode, rc);
2445
2446         rc = osd_initial_OI_scrub(info, dev);
2447         if (rc != 0)
2448                 GOTO(cleanup_oi, rc);
2449
2450         if (sf->sf_flags & SF_UPGRADE ||
2451             !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2452               sf->sf_success_count > 0)) {
2453                 dev->od_igif_inoi = 0;
2454                 dev->od_check_ff = dev->od_is_ost;
2455         } else {
2456                 dev->od_igif_inoi = 1;
2457                 dev->od_check_ff = 0;
2458         }
2459
2460         if (sf->sf_flags & SF_INCONSISTENT)
2461                 /* The 'od_igif_inoi' will be set under the
2462                  * following cases:
2463                  * 1) new created system, or
2464                  * 2) restored from file-level backup, or
2465                  * 3) the upgrading completed.
2466                  *
2467                  * The 'od_igif_inoi' may be cleared by OI scrub
2468                  * later if found that the system is upgrading. */
2469                 dev->od_igif_inoi = 1;
2470
2471         if (!dev->od_noscrub &&
2472             ((sf->sf_status == SS_PAUSED) ||
2473              (sf->sf_status == SS_CRASHED &&
2474               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2475                               SF_UPGRADE | SF_AUTO)) ||
2476              (sf->sf_status == SS_INIT &&
2477               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2478                               SF_UPGRADE))))
2479                 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2480
2481         if (rc != 0)
2482                 GOTO(cleanup_oi, rc);
2483
2484         /* it is possible that dcache entries may keep objects after they are
2485          * deleted by OSD. While it looks safe this can cause object data to
2486          * stay until umount causing failures in tests calculating free space,
2487          * e.g. replay-ost-single. Since those dcache entries are not used
2488          * anymore let's just free them after use here */
2489         shrink_dcache_sb(sb);
2490
2491         RETURN(0);
2492 cleanup_oi:
2493         osd_oi_fini(info, dev);
2494 cleanup_inode:
2495         iput(scrub->os_inode);
2496         scrub->os_inode = NULL;
2497
2498         return rc;
2499 }
2500
2501 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2502 {
2503         struct osd_scrub *scrub = &dev->od_scrub;
2504
2505         LASSERT(dev->od_otable_it == NULL);
2506
2507         if (scrub->os_inode != NULL) {
2508                 osd_scrub_stop(dev);
2509                 iput(scrub->os_inode);
2510                 scrub->os_inode = NULL;
2511         }
2512         if (dev->od_oi_table != NULL)
2513                 osd_oi_fini(osd_oti_get(env), dev);
2514 }
2515
2516 /* object table based iteration APIs */
2517
2518 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2519                                        struct dt_object *dt, __u32 attr)
2520 {
2521         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2522         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2523         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2524         struct osd_scrub       *scrub = &dev->od_scrub;
2525         struct osd_otable_it   *it;
2526         __u32                   start = 0;
2527         int                     rc;
2528         ENTRY;
2529
2530         /* od_otable_mutex: prevent curcurrent init/fini */
2531         mutex_lock(&dev->od_otable_mutex);
2532         if (dev->od_otable_it != NULL)
2533                 GOTO(out, it = ERR_PTR(-EALREADY));
2534
2535         OBD_ALLOC_PTR(it);
2536         if (it == NULL)
2537                 GOTO(out, it = ERR_PTR(-ENOMEM));
2538
2539         dev->od_otable_it = it;
2540         it->ooi_dev = dev;
2541         it->ooi_cache.ooc_consumer_idx = -1;
2542         if (flags & DOIF_OUTUSED)
2543                 it->ooi_used_outside = 1;
2544
2545         if (flags & DOIF_RESET)
2546                 start |= SS_RESET;
2547
2548         if (valid & DOIV_ERROR_HANDLE) {
2549                 if (flags & DOIF_FAILOUT)
2550                         start |= SS_SET_FAILOUT;
2551                 else
2552                         start |= SS_CLEAR_FAILOUT;
2553         }
2554
2555         if (valid & DOIV_DRYRUN) {
2556                 if (flags & DOIF_DRYRUN)
2557                         start |= SS_SET_DRYRUN;
2558                 else
2559                         start |= SS_CLEAR_DRYRUN;
2560         }
2561
2562         rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2563         if (rc < 0 && rc != -EALREADY) {
2564                 dev->od_otable_it = NULL;
2565                 OBD_FREE_PTR(it);
2566                 GOTO(out, it = ERR_PTR(rc));
2567         }
2568
2569         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2570
2571         GOTO(out, it);
2572
2573 out:
2574         mutex_unlock(&dev->od_otable_mutex);
2575         return (struct dt_it *)it;
2576 }
2577
2578 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2579 {
2580         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2581         struct osd_device    *dev = it->ooi_dev;
2582
2583         /* od_otable_mutex: prevent curcurrent init/fini */
2584         mutex_lock(&dev->od_otable_mutex);
2585         do_osd_scrub_stop(&dev->od_scrub);
2586         LASSERT(dev->od_otable_it == it);
2587
2588         dev->od_otable_it = NULL;
2589         mutex_unlock(&dev->od_otable_mutex);
2590         OBD_FREE_PTR(it);
2591 }
2592
2593 static int osd_otable_it_get(const struct lu_env *env,
2594                              struct dt_it *di, const struct dt_key *key)
2595 {
2596         return 0;
2597 }
2598
2599 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2600 {
2601 }
2602
2603 static inline int
2604 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2605 {
2606         spin_lock(&scrub->os_lock);
2607         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2608             scrub->os_waiting ||
2609             !thread_is_running(&scrub->os_thread))
2610                 it->ooi_waiting = 0;
2611         else
2612                 it->ooi_waiting = 1;
2613         spin_unlock(&scrub->os_lock);
2614
2615         return !it->ooi_waiting;
2616 }
2617
2618 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2619 {
2620         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2621         struct osd_device       *dev    = it->ooi_dev;
2622         struct osd_scrub        *scrub  = &dev->od_scrub;
2623         struct osd_otable_cache *ooc    = &it->ooi_cache;
2624         struct ptlrpc_thread    *thread = &scrub->os_thread;
2625         struct l_wait_info       lwi    = { 0 };
2626         int                      rc;
2627         ENTRY;
2628
2629         LASSERT(it->ooi_user_ready);
2630
2631 again:
2632         if (!thread_is_running(thread) && !it->ooi_used_outside)
2633                 RETURN(1);
2634
2635         if (ooc->ooc_cached_items > 0) {
2636                 ooc->ooc_cached_items--;
2637                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2638                                         ~OSD_OTABLE_IT_CACHE_MASK;
2639                 RETURN(0);
2640         }
2641
2642         if (it->ooi_all_cached) {
2643                 l_wait_event(thread->t_ctl_waitq,
2644                              !thread_is_running(thread),
2645                              &lwi);
2646                 RETURN(1);
2647         }
2648
2649         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2650                 spin_lock(&scrub->os_lock);
2651                 scrub->os_waiting = 0;
2652                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2653                 spin_unlock(&scrub->os_lock);
2654         }
2655
2656         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2657                 l_wait_event(thread->t_ctl_waitq,
2658                              osd_otable_it_wakeup(scrub, it),
2659                              &lwi);
2660
2661         if (!thread_is_running(thread) && !it->ooi_used_outside)
2662                 RETURN(1);
2663
2664         rc = osd_otable_it_preload(env, it);
2665         if (rc >= 0)
2666                 goto again;
2667
2668         RETURN(rc);
2669 }
2670
2671 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2672                                         const struct dt_it *di)
2673 {
2674         return NULL;
2675 }
2676
2677 static int osd_otable_it_key_size(const struct lu_env *env,
2678                                   const struct dt_it *di)
2679 {
2680         return sizeof(__u64);
2681 }
2682
2683 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2684                              struct dt_rec *rec, __u32 attr)
2685 {
2686         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2687         struct osd_otable_cache *ooc = &it->ooi_cache;
2688
2689         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2690
2691         /* Filter out Invald FID already. */
2692         LASSERTF(fid_is_sane((struct lu_fid *)rec),
2693                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2694                  PFID((struct lu_fid *)rec),
2695                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2696
2697         return 0;
2698 }
2699
2700 static __u64 osd_otable_it_store(const struct lu_env *env,
2701                                  const struct dt_it *di)
2702 {
2703         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2704         struct osd_otable_cache *ooc = &it->ooi_cache;
2705         __u64                    hash;
2706
2707         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2708                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2709         else
2710                 hash = ooc->ooc_pos_preload;
2711         return hash;
2712 }
2713
2714 /**
2715  * Set the OSD layer iteration start position as the specified hash.
2716  */
2717 static int osd_otable_it_load(const struct lu_env *env,
2718                               const struct dt_it *di, __u64 hash)
2719 {
2720         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
2721         struct osd_device       *dev   = it->ooi_dev;
2722         struct osd_otable_cache *ooc   = &it->ooi_cache;
2723         struct osd_scrub        *scrub = &dev->od_scrub;
2724         int                      rc;
2725         ENTRY;
2726
2727         /* Forbid to set iteration position after iteration started. */
2728         if (it->ooi_user_ready)
2729                 RETURN(-EPERM);
2730
2731         if (hash > OSD_OTABLE_MAX_HASH)
2732                 hash = OSD_OTABLE_MAX_HASH;
2733
2734         ooc->ooc_pos_preload = hash;
2735         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2736                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2737
2738         it->ooi_user_ready = 1;
2739         if (!scrub->os_full_speed)
2740                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2741
2742         /* Unplug OSD layer iteration by the first next() call. */
2743         rc = osd_otable_it_next(env, (struct dt_it *)it);
2744
2745         RETURN(rc);
2746 }
2747
2748 static int osd_otable_it_key_rec(const struct lu_env *env,
2749                                  const struct dt_it *di, void *key_rec)
2750 {
2751         return 0;
2752 }
2753
2754 const struct dt_index_operations osd_otable_ops = {
2755         .dio_it = {
2756                 .init     = osd_otable_it_init,
2757                 .fini     = osd_otable_it_fini,
2758                 .get      = osd_otable_it_get,
2759                 .put      = osd_otable_it_put,
2760                 .next     = osd_otable_it_next,
2761                 .key      = osd_otable_it_key,
2762                 .key_size = osd_otable_it_key_size,
2763                 .rec      = osd_otable_it_rec,
2764                 .store    = osd_otable_it_store,
2765                 .load     = osd_otable_it_load,
2766                 .key_rec  = osd_otable_it_key_rec,
2767         }
2768 };
2769
2770 /* high priority inconsistent items list APIs */
2771
2772 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL  60
2773
2774 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2775                    int insert)
2776 {
2777         struct osd_inconsistent_item *oii;
2778         struct osd_scrub             *scrub  = &dev->od_scrub;
2779         struct ptlrpc_thread         *thread = &scrub->os_thread;
2780         int                           wakeup = 0;
2781         ENTRY;
2782
2783         OBD_ALLOC_PTR(oii);
2784         if (unlikely(oii == NULL))
2785                 RETURN(-ENOMEM);
2786
2787         INIT_LIST_HEAD(&oii->oii_list);
2788         oii->oii_cache = *oic;
2789         oii->oii_insert = insert;
2790
2791         if (scrub->os_partial_scan) {
2792                 __u64 now = cfs_time_current_sec();
2793
2794                 /* If there haven't been errors in a long time,
2795                  * decay old count until either the errors are
2796                  * gone or we reach the current interval. */
2797                 while (unlikely(scrub->os_bad_oimap_count > 0 &&
2798                                 scrub->os_bad_oimap_time +
2799                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
2800                         scrub->os_bad_oimap_count >>= 1;
2801                         scrub->os_bad_oimap_time +=
2802                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
2803                 }
2804
2805                 scrub->os_bad_oimap_time = now;
2806                 if (++scrub->os_bad_oimap_count >
2807                     dev->od_full_scrub_threshold_rate)
2808                         scrub->os_full_scrub = 1;
2809         }
2810
2811         spin_lock(&scrub->os_lock);
2812         if (unlikely(!thread_is_running(thread))) {
2813                 spin_unlock(&scrub->os_lock);
2814                 OBD_FREE_PTR(oii);
2815                 RETURN(-EAGAIN);
2816         }
2817
2818         if (list_empty(&scrub->os_inconsistent_items))
2819                 wakeup = 1;
2820         list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2821         spin_unlock(&scrub->os_lock);
2822
2823         if (wakeup != 0)
2824                 wake_up_all(&thread->t_ctl_waitq);
2825
2826         RETURN(0);
2827 }
2828
2829 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2830                    struct osd_inode_id *id)
2831 {
2832         struct osd_scrub             *scrub = &dev->od_scrub;
2833         struct osd_inconsistent_item *oii;
2834         ENTRY;
2835
2836         spin_lock(&scrub->os_lock);
2837         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2838                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2839                         *id = oii->oii_cache.oic_lid;
2840                         spin_unlock(&scrub->os_lock);
2841                         RETURN(0);
2842                 }
2843         }
2844         spin_unlock(&scrub->os_lock);
2845
2846         RETURN(-ENOENT);
2847 }
2848
2849 /* OI scrub dump */
2850
2851 static const char *scrub_status_names[] = {
2852         "init",
2853         "scanning",
2854         "completed",
2855         "failed",
2856         "stopped",
2857         "paused",
2858         "crashed",
2859         NULL
2860 };
2861
2862 static const char *scrub_flags_names[] = {
2863         "recreated",
2864         "inconsistent",
2865         "auto",
2866         "upgrade",
2867         NULL
2868 };
2869
2870 static const char *scrub_param_names[] = {
2871         "failout",
2872         "dryrun",
2873         NULL
2874 };
2875
2876 static int scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
2877                            const char *prefix)
2878 {
2879         int flag;
2880         int rc;
2881         int i;
2882
2883         rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2884         if (rc < 0)
2885                 return rc;
2886
2887         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2888                 if (flag & bits) {
2889                         bits &= ~flag;
2890                         rc = seq_printf(m, "%s%c", names[i],
2891                                         bits != 0 ? ',' : '\n');
2892                         if (rc < 0)
2893                                 return rc;
2894                 }
2895         }
2896         return 0;
2897 }
2898
2899 static int scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
2900 {
2901         int rc;
2902
2903         if (time != 0)
2904                 rc = seq_printf(m, "%s: "LPU64" seconds\n", prefix,
2905                               cfs_time_current_sec() - time);
2906         else
2907                 rc = seq_printf(m, "%s: N/A\n", prefix);
2908         return rc;
2909 }
2910
2911 static int scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
2912 {
2913         int rc;
2914
2915         if (pos != 0)
2916                 rc = seq_printf(m, "%s: "LPU64"\n", prefix, pos);
2917         else
2918                 rc = seq_printf(m, "%s: N/A\n", prefix);
2919         return rc;
2920 }
2921
2922 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
2923 {
2924         struct osd_scrub  *scrub   = &dev->od_scrub;
2925         struct scrub_file *sf      = &scrub->os_file;
2926         __u64              checked;
2927         __u64              speed;
2928         int                rc;
2929
2930         down_read(&scrub->os_rwsem);
2931         rc = seq_printf(m, "name: OI_scrub\n"
2932                         "magic: 0x%x\n"
2933                         "oi_files: %d\n"
2934                         "status: %s\n",
2935                         sf->sf_magic, (int)sf->sf_oi_count,
2936                         scrub_status_names[sf->sf_status]);
2937         if (rc < 0)
2938                 goto out;
2939
2940         rc = scrub_bits_dump(m, sf->sf_flags, scrub_flags_names,
2941                              "flags");
2942         if (rc < 0)
2943                 goto out;
2944
2945         rc = scrub_bits_dump(m, sf->sf_param, scrub_param_names,
2946                              "param");
2947         if (rc < 0)
2948                 goto out;
2949
2950         rc = scrub_time_dump(m, sf->sf_time_last_complete,
2951                              "time_since_last_completed");
2952         if (rc < 0)
2953                 goto out;
2954
2955         rc = scrub_time_dump(m, sf->sf_time_latest_start,
2956                              "time_since_latest_start");
2957         if (rc < 0)
2958                 goto out;
2959
2960         rc = scrub_time_dump(m, sf->sf_time_last_checkpoint,
2961                              "time_since_last_checkpoint");
2962         if (rc < 0)
2963                 goto out;
2964
2965         rc = scrub_pos_dump(m, sf->sf_pos_latest_start,
2966                             "latest_start_position");
2967         if (rc < 0)
2968                 goto out;
2969
2970         rc = scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
2971                             "last_checkpoint_position");
2972         if (rc < 0)
2973                 goto out;
2974
2975         rc = scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
2976                             "first_failure_position");
2977         if (rc < 0)
2978                 goto out;
2979
2980         checked = sf->sf_items_checked + scrub->os_new_checked;
2981         rc = seq_printf(m, "checked: "LPU64"\n"
2982                       "updated: "LPU64"\n"
2983                       "failed: "LPU64"\n"
2984                       "prior_updated: "LPU64"\n"
2985                       "noscrub: "LPU64"\n"
2986                       "igif: "LPU64"\n"
2987                       "success_count: %u\n",
2988                       checked, sf->sf_items_updated, sf->sf_items_failed,
2989                       sf->sf_items_updated_prior, sf->sf_items_noscrub,
2990                       sf->sf_items_igif, sf->sf_success_count);
2991         if (rc < 0)
2992                 goto out;
2993
2994         speed = checked;
2995         if (thread_is_running(&scrub->os_thread)) {
2996                 cfs_duration_t duration = cfs_time_current() -
2997                                           scrub->os_time_last_checkpoint;
2998                 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
2999                                                      MSEC_PER_SEC);
3000                 __u32 rtime = sf->sf_run_time +
3001                               cfs_duration_sec(duration + HALF_SEC);
3002
3003                 if (duration != 0)
3004                         do_div(new_checked, duration);
3005                 if (rtime != 0)
3006                         do_div(speed, rtime);
3007                 rc = seq_printf(m, "run_time: %u seconds\n"
3008                               "average_speed: "LPU64" objects/sec\n"
3009                               "real-time_speed: "LPU64" objects/sec\n"
3010                               "current_position: %u\n"
3011                               "lf_scanned: "LPU64"\n"
3012                               "lf_reparied: "LPU64"\n"
3013                               "lf_failed: "LPU64"\n",
3014                               rtime, speed, new_checked, scrub->os_pos_current,
3015                               scrub->os_lf_scanned, scrub->os_lf_repaired,
3016                               scrub->os_lf_failed);
3017         } else {
3018                 if (sf->sf_run_time != 0)
3019                         do_div(speed, sf->sf_run_time);
3020                 rc = seq_printf(m, "run_time: %u seconds\n"
3021                               "average_speed: "LPU64" objects/sec\n"
3022                               "real-time_speed: N/A\n"
3023                               "current_position: N/A\n"
3024                               "lf_scanned: "LPU64"\n"
3025                               "lf_reparied: "LPU64"\n"
3026                               "lf_failed: "LPU64"\n",
3027                               sf->sf_run_time, speed, scrub->os_lf_scanned,
3028                               scrub->os_lf_repaired, scrub->os_lf_failed);
3029         }
3030
3031 out:
3032         up_read(&scrub->os_rwsem);
3033         return (rc < 0 ? -ENOSPC : 0);
3034 }