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, 2015, Intel Corporation.
  24  */
  25 /*
  26  * lustre/osd-ldiskfs/osd_scrub.c
  27  *
  28  * Top-level entry points into osd module
  29  *
  30  * The OI scrub is used for rebuilding Object Index files when restores MDT from
  31  * file-level backup.
  32  *
  33  * The otable based iterator scans ldiskfs inode table to feed up layer LFSCK.
  34  *
  35  * Author: Fan Yong <yong.fan@whamcloud.com>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_LFSCK
  39
  40 #include <linux/kthread.h>
  41 #include <lustre/lustre_idl.h>
  42 #include <lustre_disk.h>
  43 #include <dt_object.h>
  44 #include <linux/xattr.h>
  45
  46 #include "osd_internal.h"
  47 #include "osd_oi.h"
  48 #include "osd_scrub.h"
  49
  50 #define HALF_SEC        msecs_to_jiffies(MSEC_PER_SEC >> 1)
  51
  52 #define OSD_OTABLE_MAX_HASH             0x00000000ffffffffULL
  53
  54 #define SCRUB_NEXT_BREAK        1 /* exit current loop and process next group */
  55 #define SCRUB_NEXT_CONTINUE     2 /* skip current object and process next bit */
  56 #define SCRUB_NEXT_EXIT         3 /* exit all the loops */
  57 #define SCRUB_NEXT_WAIT         4 /* wait for free cache slot */
  58 #define SCRUB_NEXT_CRASH        5 /* simulate system crash during OI scrub */
  59 #define SCRUB_NEXT_FATAL        6 /* simulate failure during OI scrub */
  60 #define SCRUB_NEXT_NOSCRUB      7 /* new created object, no scrub on it */
  61 #define SCRUB_NEXT_NOLMA        8 /* the inode has no FID-in-LMA */
  62 #define SCRUB_NEXT_OSTOBJ       9 /* for OST-object */
  63 #define SCRUB_NEXT_OSTOBJ_OLD   10 /* old OST-object, no LMA or no FID-on-OST
  64                                     * flags in LMA */
  65
  66 /* misc functions */
  67
  68 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
  69 {
  70         return container_of0(scrub, struct osd_device, od_scrub);
  71 }
  72
  73 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
  74 {
  75         return osd_sb(osd_scrub2dev(scrub));
  76 }
  77
  78 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
  79                                        struct osd_otable_cache *ooc)
  80 {
  81         return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
  82 }
  83
  84 static inline const char *osd_scrub2name(struct osd_scrub *scrub)
  85 {
  86         return LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
  87 }
  88
  89 /**
  90  * update/insert/delete the specified OI mapping (@fid @id) according to the ops
  91  *
  92  * \retval   1, changed nothing
  93  * \retval   0, changed successfully
  94  * \retval -ve, on error
  95  */
  96 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
  97                                      struct osd_device *dev,
  98                                      const struct lu_fid *fid,
  99                                      const struct osd_inode_id *id,
 100                                      int ops, bool force,
 101                                      enum oi_check_flags flags)
 102 {
 103         handle_t *th;
 104         int       rc;
 105         ENTRY;
 106
 107         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
 108                 RETURN(0);
 109
 110         /* DTO_INDEX_INSERT is enough for other two ops:
 111          * delete/update, but save stack. */
 112         th = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 113                                 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
 114         if (IS_ERR(th)) {
 115                 rc = PTR_ERR(th);
 116                 CDEBUG(D_LFSCK, "%s: fail to start trans for scrub op %d "
 117                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 118                        PFID(fid), id->oii_ino, id->oii_gen, rc);
 119                 RETURN(rc);
 120         }
 121
 122         switch (ops) {
 123         case DTO_INDEX_UPDATE:
 124                 rc = osd_oi_update(info, dev, fid, id, th, flags);
 125                 if (unlikely(rc == -ENOENT)) {
 126                         /* Some unlink thread may removed the OI mapping. */
 127                         rc = 1;
 128                 }
 129                 break;
 130         case DTO_INDEX_INSERT:
 131                 rc = osd_oi_insert(info, dev, fid, id, th, flags);
 132                 if (unlikely(rc == -EEXIST)) {
 133                         rc = 1;
 134                         /* XXX: There are trouble things when adding OI
 135                          *      mapping for IGIF object, which may cause
 136                          *      multiple objects to be mapped to the same
 137                          *      IGIF formatted FID. Consider the following
 138                          *      situations:
 139                          *
 140                          *      1) The MDT is upgrading from 1.8 device.
 141                          *      The OI scrub generates IGIF FID1 for the
 142                          *      OBJ1 and adds the OI mapping.
 143                          *
 144                          *      2) For some reason, the OI scrub does not
 145                          *      process all the IGIF objects completely.
 146                          *
 147                          *      3) The MDT is backuped and restored against
 148                          *      this device.
 149                          *
 150                          *      4) When the MDT mounts up, the OI scrub will
 151                          *      try to rebuild the OI files. For some IGIF
 152                          *      object, OBJ2, which was not processed by the
 153                          *      OI scrub before the backup/restore, and the
 154                          *      new generated IGIF formatted FID may be just
 155                          *      the FID1, the same as OBJ1.
 156                          *
 157                          *      Under such case, the OI scrub cannot know how
 158                          *      to generate new FID for the OBJ2.
 159                          *
 160                          *      Currently, we do nothing for that. One possible
 161                          *      solution is to generate new normal FID for the
 162                          *      conflict object.
 163                          *
 164                          *      Anyway, it is rare, only exists in theory. */
 165                 }
 166                 break;
 167         case DTO_INDEX_DELETE:
 168                 rc = osd_oi_delete(info, dev, fid, th, flags);
 169                 if (rc == -ENOENT) {
 170                         /* It is normal that the unlink thread has removed the
 171                          * OI mapping already. */
 172                         rc = 1;
 173                 }
 174                 break;
 175         default:
 176                 LASSERTF(0, "Unexpected ops %d\n", ops);
 177                 break;
 178         }
 179
 180         ldiskfs_journal_stop(th);
 181         if (rc < 0)
 182                 CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d "
 183                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 184                        PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1,
 185                        rc);
 186
 187         RETURN(rc);
 188 }
 189
 190 /* OI_scrub file ops */
 191
 192 static void osd_scrub_file_to_cpu(struct scrub_file *des,
 193                                   struct scrub_file *src)
 194 {
 195         memcpy(des->sf_uuid, src->sf_uuid, 16);
 196         des->sf_flags   = le64_to_cpu(src->sf_flags);
 197         des->sf_magic   = le32_to_cpu(src->sf_magic);
 198         des->sf_status  = le16_to_cpu(src->sf_status);
 199         des->sf_param   = le16_to_cpu(src->sf_param);
 200         des->sf_time_last_complete      =
 201                                 le64_to_cpu(src->sf_time_last_complete);
 202         des->sf_time_latest_start       =
 203                                 le64_to_cpu(src->sf_time_latest_start);
 204         des->sf_time_last_checkpoint    =
 205                                 le64_to_cpu(src->sf_time_last_checkpoint);
 206         des->sf_pos_latest_start        =
 207                                 le64_to_cpu(src->sf_pos_latest_start);
 208         des->sf_pos_last_checkpoint     =
 209                                 le64_to_cpu(src->sf_pos_last_checkpoint);
 210         des->sf_pos_first_inconsistent  =
 211                                 le64_to_cpu(src->sf_pos_first_inconsistent);
 212         des->sf_items_checked           =
 213                                 le64_to_cpu(src->sf_items_checked);
 214         des->sf_items_updated           =
 215                                 le64_to_cpu(src->sf_items_updated);
 216         des->sf_items_failed            =
 217                                 le64_to_cpu(src->sf_items_failed);
 218         des->sf_items_updated_prior     =
 219                                 le64_to_cpu(src->sf_items_updated_prior);
 220         des->sf_run_time        = le32_to_cpu(src->sf_run_time);
 221         des->sf_success_count   = le32_to_cpu(src->sf_success_count);
 222         des->sf_oi_count        = le16_to_cpu(src->sf_oi_count);
 223         des->sf_internal_flags  = le16_to_cpu(src->sf_internal_flags);
 224         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 225 }
 226
 227 static void osd_scrub_file_to_le(struct scrub_file *des,
 228                                  struct scrub_file *src)
 229 {
 230         memcpy(des->sf_uuid, src->sf_uuid, 16);
 231         des->sf_flags   = cpu_to_le64(src->sf_flags);
 232         des->sf_magic   = cpu_to_le32(src->sf_magic);
 233         des->sf_status  = cpu_to_le16(src->sf_status);
 234         des->sf_param   = cpu_to_le16(src->sf_param);
 235         des->sf_time_last_complete      =
 236                                 cpu_to_le64(src->sf_time_last_complete);
 237         des->sf_time_latest_start       =
 238                                 cpu_to_le64(src->sf_time_latest_start);
 239         des->sf_time_last_checkpoint    =
 240                                 cpu_to_le64(src->sf_time_last_checkpoint);
 241         des->sf_pos_latest_start        =
 242                                 cpu_to_le64(src->sf_pos_latest_start);
 243         des->sf_pos_last_checkpoint     =
 244                                 cpu_to_le64(src->sf_pos_last_checkpoint);
 245         des->sf_pos_first_inconsistent  =
 246                                 cpu_to_le64(src->sf_pos_first_inconsistent);
 247         des->sf_items_checked           =
 248                                 cpu_to_le64(src->sf_items_checked);
 249         des->sf_items_updated           =
 250                                 cpu_to_le64(src->sf_items_updated);
 251         des->sf_items_failed            =
 252                                 cpu_to_le64(src->sf_items_failed);
 253         des->sf_items_updated_prior     =
 254                                 cpu_to_le64(src->sf_items_updated_prior);
 255         des->sf_run_time        = cpu_to_le32(src->sf_run_time);
 256         des->sf_success_count   = cpu_to_le32(src->sf_success_count);
 257         des->sf_oi_count        = cpu_to_le16(src->sf_oi_count);
 258         des->sf_internal_flags  = cpu_to_le16(src->sf_internal_flags);
 259         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 260 }
 261
 262 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
 263 {
 264         struct scrub_file *sf = &scrub->os_file;
 265
 266         memset(sf, 0, sizeof(*sf));
 267         memcpy(sf->sf_uuid, uuid, 16);
 268         sf->sf_magic = SCRUB_MAGIC_V1;
 269         sf->sf_status = SS_INIT;
 270 }
 271
 272 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
 273 {
 274         struct scrub_file *sf = &scrub->os_file;
 275
 276         CDEBUG(D_LFSCK, "%.16s: reset OI scrub file, old flags = "
 277                LPX64", add flags = "LPX64"\n",
 278                osd_scrub2name(scrub), sf->sf_flags, flags);
 279
 280         memcpy(sf->sf_uuid, uuid, 16);
 281         sf->sf_status = SS_INIT;
 282         sf->sf_flags |= flags;
 283         sf->sf_flags &= ~SF_AUTO;
 284         sf->sf_run_time = 0;
 285         sf->sf_time_latest_start = 0;
 286         sf->sf_time_last_checkpoint = 0;
 287         sf->sf_pos_latest_start = 0;
 288         sf->sf_pos_last_checkpoint = 0;
 289         sf->sf_pos_first_inconsistent = 0;
 290         sf->sf_items_checked = 0;
 291         sf->sf_items_updated = 0;
 292         sf->sf_items_failed = 0;
 293         if (!scrub->os_in_join)
 294                 sf->sf_items_updated_prior = 0;
 295
 296         sf->sf_items_noscrub = 0;
 297         sf->sf_items_igif = 0;
 298 }
 299
 300 static int osd_scrub_file_load(struct osd_scrub *scrub)
 301 {
 302         loff_t  pos  = 0;
 303         int     len  = sizeof(scrub->os_file_disk);
 304         int     rc;
 305
 306         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
 307         if (rc == len) {
 308                 struct scrub_file *sf = &scrub->os_file;
 309
 310                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
 311                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
 312                         CDEBUG(D_LFSCK, "%.16s: invalid scrub magic "
 313                                "0x%x != 0x%x\n", osd_scrub2name(scrub),
 314                                sf->sf_magic, SCRUB_MAGIC_V1);
 315                         /* Process it as new scrub file. */
 316                         rc = -ENOENT;
 317                 } else {
 318                         rc = 0;
 319                 }
 320         } else if (rc != 0) {
 321                 CDEBUG(D_LFSCK, "%.16s: fail to load scrub file, "
 322                        "expected = %d: rc = %d\n",
 323                        osd_scrub2name(scrub), len, rc);
 324                 if (rc > 0)
 325                         rc = -EFAULT;
 326         } else {
 327                 /* return -ENOENT for empty scrub file case. */
 328                 rc = -ENOENT;
 329         }
 330
 331         return rc;
 332 }
 333
 334 int osd_scrub_file_store(struct osd_scrub *scrub)
 335 {
 336         struct osd_device *dev;
 337         handle_t          *jh;
 338         loff_t             pos     = 0;
 339         int                len     = sizeof(scrub->os_file_disk);
 340         int                credits;
 341         int                rc;
 342
 343         dev = container_of0(scrub, struct osd_device, od_scrub);
 344         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
 345                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
 346         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits);
 347         if (IS_ERR(jh)) {
 348                 rc = PTR_ERR(jh);
 349                 CDEBUG(D_LFSCK, "%.16s: fail to start trans for scrub store: "
 350                        "rc = %d\n", osd_scrub2name(scrub), rc);
 351                 return rc;
 352         }
 353
 354         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
 355         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
 356                                       len, 0, &pos, jh);
 357         ldiskfs_journal_stop(jh);
 358         if (rc != 0)
 359                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file, "
 360                        "expected = %d: rc = %d\n",
 361                        osd_scrub2name(scrub), len, rc);
 362
 363         scrub->os_time_last_checkpoint = cfs_time_current();
 364         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
 365                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
 366         return rc;
 367 }
 368
 369 static int
 370 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
 371                      struct inode *inode, const struct lu_fid *fid)
 372 {
 373         struct filter_fid_old   *ff      = &info->oti_ff;
 374         struct dentry           *dentry  = &info->oti_obj_dentry;
 375         struct lu_fid           *tfid    = &info->oti_fid;
 376         handle_t                *jh;
 377         int                      size    = 0;
 378         int                      rc;
 379         bool                     removed = false;
 380         bool                     reset   = true;
 381         ENTRY;
 382
 383         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
 384                 RETURN(0);
 385
 386         if (fid_is_idif(fid) && dev->od_index_in_idif == 0) {
 387                 struct ost_id *oi = &info->oti_ostid;
 388
 389                 fid_to_ostid(fid, oi);
 390                 ostid_to_fid(tfid, oi, 0);
 391         } else {
 392                 *tfid = *fid;
 393         }
 394
 395         /* We want the LMA to fit into the 256-byte OST inode, so operate
 396          * as following:
 397          * 1) read old XATTR_NAME_FID and save the parent FID;
 398          * 2) delete the old XATTR_NAME_FID;
 399          * 3) make new LMA and add it;
 400          * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
 401          *
 402          * Making the LMA to fit into the 256-byte OST inode can save time for
 403          * normal osd_check_lma() and for other OI scrub scanning in future.
 404          * So it is worth to make some slow conversion here. */
 405         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 406                                 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
 407         if (IS_ERR(jh)) {
 408                 rc = PTR_ERR(jh);
 409                 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
 410                        DFID": rc = %d\n", osd_name(dev), PFID(tfid), rc);
 411                 RETURN(rc);
 412         }
 413
 414         /* 1) read old XATTR_NAME_FID and save the parent FID */
 415         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 416         if (rc == sizeof(*ff)) {
 417                 /* 2) delete the old XATTR_NAME_FID */
 418                 ll_vfs_dq_init(inode);
 419                 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
 420                 if (rc != 0)
 421                         GOTO(stop, rc);
 422
 423                 removed = true;
 424         } else if (unlikely(rc == -ENODATA)) {
 425                 reset = false;
 426         } else if (rc != sizeof(struct filter_fid)) {
 427                 GOTO(stop, rc = -EINVAL);
 428         }
 429
 430         /* 3) make new LMA and add it */
 431         rc = osd_ea_fid_set(info, inode, tfid, LMAC_FID_ON_OST, 0);
 432         if (rc == 0 && reset)
 433                 size = sizeof(struct filter_fid);
 434         else if (rc != 0 && removed)
 435                 /* If failed, we should try to add the old back. */
 436                 size = sizeof(struct filter_fid_old);
 437
 438         /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
 439         if (size > 0) {
 440                 int rc1;
 441
 442                 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
 443                                       XATTR_CREATE);
 444                 if (rc1 != 0 && rc == 0)
 445                         rc = rc1;
 446         }
 447
 448         GOTO(stop, rc);
 449
 450 stop:
 451         ldiskfs_journal_stop(jh);
 452         if (rc < 0)
 453                 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
 454                        osd_name(dev), PFID(tfid), rc);
 455         return rc;
 456 }
 457
 458 static int
 459 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
 460                        struct osd_idmap_cache *oic, int val)
 461 {
 462         struct osd_scrub             *scrub  = &dev->od_scrub;
 463         struct scrub_file            *sf     = &scrub->os_file;
 464         struct lu_fid                *fid    = &oic->oic_fid;
 465         struct osd_inode_id          *lid    = &oic->oic_lid;
 466         struct osd_inode_id          *lid2   = &info->oti_id;
 467         struct osd_inconsistent_item *oii    = NULL;
 468         struct inode                 *inode  = NULL;
 469         int                           ops    = DTO_INDEX_UPDATE;
 470         int                           idx;
 471         int                           rc;
 472         bool                          converted = false;
 473         ENTRY;
 474
 475         down_write(&scrub->os_rwsem);
 476         scrub->os_new_checked++;
 477         if (val < 0)
 478                 GOTO(out, rc = val);
 479
 480         if (scrub->os_in_prior)
 481                 oii = list_entry(oic, struct osd_inconsistent_item,
 482                                  oii_cache);
 483
 484         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
 485                 GOTO(out, rc = 0);
 486
 487         if (fid_is_igif(fid))
 488                 sf->sf_items_igif++;
 489
 490         if (val == SCRUB_NEXT_OSTOBJ_OLD) {
 491                 inode = osd_iget(info, dev, lid);
 492                 if (IS_ERR(inode)) {
 493                         rc = PTR_ERR(inode);
 494                         /* Someone removed the inode. */
 495                         if (rc == -ENOENT || rc == -ESTALE)
 496                                 rc = 0;
 497                         GOTO(out, rc);
 498                 }
 499
 500                 sf->sf_flags |= SF_UPGRADE;
 501                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
 502                 dev->od_check_ff = 1;
 503                 rc = osd_scrub_convert_ff(info, dev, inode, fid);
 504                 if (rc != 0)
 505                         GOTO(out, rc);
 506
 507                 converted = true;
 508         }
 509
 510         if ((val == SCRUB_NEXT_NOLMA) &&
 511             (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
 512                 GOTO(out, rc = 0);
 513
 514         if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA)) {
 515                 ops = DTO_INDEX_INSERT;
 516
 517                 goto iget;
 518         }
 519
 520         rc = osd_oi_lookup(info, dev, fid, lid2,
 521                 (val == SCRUB_NEXT_OSTOBJ ||
 522                  val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 523         if (rc != 0) {
 524                 if (rc == -ENOENT)
 525                         ops = DTO_INDEX_INSERT;
 526                 else if (rc != -ESTALE)
 527                         GOTO(out, rc);
 528
 529 iget:
 530                 if (inode == NULL) {
 531                         inode = osd_iget(info, dev, lid);
 532                         if (IS_ERR(inode)) {
 533                                 rc = PTR_ERR(inode);
 534                                 /* Someone removed the inode. */
 535                                 if (rc == -ENOENT || rc == -ESTALE)
 536                                         rc = 0;
 537                                 GOTO(out, rc);
 538                         }
 539                 }
 540
 541                 if (!scrub->os_partial_scan)
 542                         scrub->os_full_speed = 1;
 543
 544                 idx = osd_oi_fid2idx(dev, fid);
 545                 switch (val) {
 546                 case SCRUB_NEXT_NOLMA:
 547                         sf->sf_flags |= SF_UPGRADE;
 548                         if (!(sf->sf_param & SP_DRYRUN)) {
 549                                 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
 550                                 if (rc != 0)
 551                                         GOTO(out, rc);
 552                         }
 553
 554                         if (!(sf->sf_flags & SF_INCONSISTENT))
 555                                 dev->od_igif_inoi = 0;
 556                         break;
 557                 case SCRUB_NEXT_OSTOBJ:
 558                         sf->sf_flags |= SF_INCONSISTENT;
 559                 case SCRUB_NEXT_OSTOBJ_OLD:
 560                         break;
 561                 default:
 562                         sf->sf_flags |= SF_RECREATED;
 563                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 564                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 565                         break;
 566                 }
 567         } else if (osd_id_eq(lid, lid2)) {
 568                 if (converted)
 569                         sf->sf_items_updated++;
 570
 571                 GOTO(out, rc = 0);
 572         } else {
 573                 if (!scrub->os_partial_scan)
 574                         scrub->os_full_speed = 1;
 575
 576                 sf->sf_flags |= SF_INCONSISTENT;
 577
 578                 /* XXX: If the device is restored from file-level backup, then
 579                  *      some IGIFs may have been already in OI files, and some
 580                  *      may be not yet. Means upgrading from 1.8 may be partly
 581                  *      processed, but some clients may hold some immobilized
 582                  *      IGIFs, and use them to access related objects. Under
 583                  *      such case, OSD does not know whether an given IGIF has
 584                  *      been processed or to be processed, and it also cannot
 585                  *      generate local ino#/gen# directly from the immobilized
 586                  *      IGIF because of the backup/restore. Then force OSD to
 587                  *      lookup the given IGIF in OI files, and if no entry,
 588                  *      then ask the client to retry after upgrading completed.
 589                  *      No better choice. */
 590                 dev->od_igif_inoi = 1;
 591         }
 592
 593         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
 594                         (val == SCRUB_NEXT_OSTOBJ ||
 595                          val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 596         if (rc == 0) {
 597                 if (scrub->os_in_prior)
 598                         sf->sf_items_updated_prior++;
 599                 else
 600                         sf->sf_items_updated++;
 601         }
 602
 603         GOTO(out, rc);
 604
 605 out:
 606         if (rc < 0) {
 607                 sf->sf_items_failed++;
 608                 if (sf->sf_pos_first_inconsistent == 0 ||
 609                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 610                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 611         } else {
 612                 rc = 0;
 613         }
 614
 615         /* There may be conflict unlink during the OI scrub,
 616          * if happend, then remove the new added OI mapping. */
 617         if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
 618             unlikely(ldiskfs_test_inode_state(inode,
 619                                               LDISKFS_STATE_LUSTRE_DESTROY)))
 620                 osd_scrub_refresh_mapping(info, dev, fid, lid,
 621                                 DTO_INDEX_DELETE, false,
 622                                 (val == SCRUB_NEXT_OSTOBJ ||
 623                                  val == SCRUB_NEXT_OSTOBJ_OLD) ?
 624                                 OI_KNOWN_ON_OST : 0);
 625         up_write(&scrub->os_rwsem);
 626
 627         if (inode != NULL && !IS_ERR(inode))
 628                 iput(inode);
 629
 630         if (oii != NULL) {
 631                 LASSERT(list_empty(&oii->oii_list));
 632
 633                 OBD_FREE_PTR(oii);
 634         }
 635
 636         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 637 }
 638
 639 /* OI scrub APIs */
 640
 641 static int osd_scrub_prep(struct osd_device *dev)
 642 {
 643         struct osd_scrub     *scrub  = &dev->od_scrub;
 644         struct ptlrpc_thread *thread = &scrub->os_thread;
 645         struct scrub_file    *sf     = &scrub->os_file;
 646         __u32                 flags  = scrub->os_start_flags;
 647         int                   rc;
 648         bool                  drop_dryrun = false;
 649         ENTRY;
 650
 651         CDEBUG(D_LFSCK, "%.16s: OI scrub prep, flags = 0x%x\n",
 652                osd_scrub2name(scrub), flags);
 653
 654         down_write(&scrub->os_rwsem);
 655         if (flags & SS_SET_FAILOUT)
 656                 sf->sf_param |= SP_FAILOUT;
 657         else if (flags & SS_CLEAR_FAILOUT)
 658                 sf->sf_param &= ~SP_FAILOUT;
 659
 660         if (flags & SS_SET_DRYRUN) {
 661                 sf->sf_param |= SP_DRYRUN;
 662         } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
 663                 sf->sf_param &= ~SP_DRYRUN;
 664                 drop_dryrun = true;
 665         }
 666
 667         if (flags & SS_RESET)
 668                 osd_scrub_file_reset(scrub,
 669                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
 670
 671         if (flags & SS_AUTO_FULL) {
 672                 scrub->os_full_speed = 1;
 673                 scrub->os_partial_scan = 0;
 674                 sf->sf_flags |= SF_AUTO;
 675         } else if (flags & SS_AUTO_PARTIAL) {
 676                 scrub->os_full_speed = 0;
 677                 scrub->os_partial_scan = 1;
 678                 sf->sf_flags |= SF_AUTO;
 679         } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
 680                                    SF_UPGRADE)) {
 681                 scrub->os_full_speed = 1;
 682                 scrub->os_partial_scan = 0;
 683         } else {
 684                 scrub->os_full_speed = 0;
 685                 scrub->os_partial_scan = 0;
 686         }
 687
 688         spin_lock(&scrub->os_lock);
 689         scrub->os_in_prior = 0;
 690         scrub->os_waiting = 0;
 691         scrub->os_paused = 0;
 692         scrub->os_in_join = 0;
 693         scrub->os_full_scrub = 0;
 694         spin_unlock(&scrub->os_lock);
 695         scrub->os_new_checked = 0;
 696         if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
 697                 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
 698         else if (sf->sf_pos_last_checkpoint != 0)
 699                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 700         else
 701                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 702
 703         scrub->os_pos_current = sf->sf_pos_latest_start;
 704         sf->sf_status = SS_SCANNING;
 705         sf->sf_time_latest_start = cfs_time_current_sec();
 706         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 707         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
 708         rc = osd_scrub_file_store(scrub);
 709         if (rc == 0) {
 710                 spin_lock(&scrub->os_lock);
 711                 thread_set_flags(thread, SVC_RUNNING);
 712                 spin_unlock(&scrub->os_lock);
 713                 wake_up_all(&thread->t_ctl_waitq);
 714         }
 715         up_write(&scrub->os_rwsem);
 716
 717         RETURN(rc);
 718 }
 719
 720 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
 721 {
 722         struct scrub_file *sf = &scrub->os_file;
 723         int                rc;
 724
 725         if (likely(cfs_time_before(cfs_time_current(),
 726                                    scrub->os_time_next_checkpoint) ||
 727                    scrub->os_new_checked == 0))
 728                 return 0;
 729
 730         down_write(&scrub->os_rwsem);
 731         sf->sf_items_checked += scrub->os_new_checked;
 732         scrub->os_new_checked = 0;
 733         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 734         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 735         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 736                                             scrub->os_time_last_checkpoint);
 737         rc = osd_scrub_file_store(scrub);
 738         up_write(&scrub->os_rwsem);
 739
 740         return rc;
 741 }
 742
 743 static int osd_scrub_post(struct osd_scrub *scrub, int result)
 744 {
 745         struct scrub_file *sf = &scrub->os_file;
 746         int rc;
 747         ENTRY;
 748
 749         CDEBUG(D_LFSCK, "%.16s: OI scrub post, result = %d\n",
 750                osd_scrub2name(scrub), result);
 751
 752         down_write(&scrub->os_rwsem);
 753         spin_lock(&scrub->os_lock);
 754         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 755         spin_unlock(&scrub->os_lock);
 756         if (scrub->os_new_checked > 0) {
 757                 sf->sf_items_checked += scrub->os_new_checked;
 758                 scrub->os_new_checked = 0;
 759                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 760         }
 761         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 762         if (result > 0) {
 763                 struct osd_device *dev =
 764                         container_of0(scrub, struct osd_device, od_scrub);
 765
 766                 dev->od_igif_inoi = 1;
 767                 dev->od_check_ff = 0;
 768                 sf->sf_status = SS_COMPLETED;
 769                 if (!(sf->sf_param & SP_DRYRUN)) {
 770                         memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 771                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 772                                           SF_UPGRADE | SF_AUTO);
 773                 }
 774                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 775                 sf->sf_success_count++;
 776         } else if (result == 0) {
 777                 if (scrub->os_paused)
 778                         sf->sf_status = SS_PAUSED;
 779                 else
 780                         sf->sf_status = SS_STOPPED;
 781         } else {
 782                 sf->sf_status = SS_FAILED;
 783         }
 784         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 785                                             scrub->os_time_last_checkpoint);
 786         rc = osd_scrub_file_store(scrub);
 787         up_write(&scrub->os_rwsem);
 788
 789         RETURN(rc < 0 ? rc : result);
 790 }
 791
 792 /* iteration engine */
 793
 794 struct osd_iit_param {
 795         struct super_block *sb;
 796         struct buffer_head *bitmap;
 797         ldiskfs_group_t bg;
 798         __u32 gbase;
 799         __u32 offset;
 800 };
 801
 802 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 803                                    struct osd_device *dev,
 804                                    struct osd_iit_param *param,
 805                                    struct osd_idmap_cache **oic,
 806                                    const bool noslot);
 807
 808 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 809                                    struct osd_device *dev,
 810                                    struct osd_iit_param *param,
 811                                    struct osd_idmap_cache *oic,
 812                                    bool *noslot, int rc);
 813
 814 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 815 {
 816         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 817                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 818         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 819                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 820                 return SCRUB_NEXT_BREAK;
 821         } else {
 822                 *pos = param->gbase + param->offset;
 823                 return 0;
 824         }
 825 }
 826
 827 /**
 828  * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
 829  * \retval 0: FID-on-MDT
 830  */
 831 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
 832                                       struct osd_device *dev,
 833                                       struct lu_fid *fid)
 834 {
 835         /* XXX: The initial OI scrub will scan the top level /O to generate
 836          *      a small local FLDB according to the <seq>. If the given FID
 837          *      is in the local FLDB, then it is FID-on-OST; otherwise it's
 838          *      quite possible for FID-on-MDT. */
 839         if (dev->od_is_ost)
 840                 return SCRUB_NEXT_OSTOBJ_OLD;
 841         else
 842                 return 0;
 843 }
 844
 845 static int osd_scrub_get_fid(struct osd_thread_info *info,
 846                              struct osd_device *dev, struct inode *inode,
 847                              struct lu_fid *fid, bool scrub)
 848 {
 849         struct lustre_mdt_attrs *lma     = &info->oti_mdt_attrs;
 850         int                      rc;
 851         bool                     has_lma = false;
 852
 853         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 854         if (rc == 0) {
 855                 has_lma = true;
 856                 if (lma->lma_compat & LMAC_NOT_IN_OI ||
 857                     lma->lma_incompat & LMAI_AGENT)
 858                         return SCRUB_NEXT_CONTINUE;
 859
 860                 *fid = lma->lma_self_fid;
 861                 if (!scrub)
 862                         return 0;
 863
 864                 if (lma->lma_compat & LMAC_FID_ON_OST)
 865                         return SCRUB_NEXT_OSTOBJ;
 866
 867                 if (fid_is_idif(fid))
 868                         return SCRUB_NEXT_OSTOBJ_OLD;
 869
 870                 /* For local object. */
 871                 if (fid_is_internal(fid))
 872                         return 0;
 873
 874                 /* For external visible MDT-object with non-normal FID. */
 875                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
 876                         return 0;
 877
 878                 /* For the object with normal FID, it may be MDT-object,
 879                  * or may be 2.4 OST-object, need further distinguish.
 880                  * Fall through to next section. */
 881         }
 882
 883         if (rc == -ENODATA || rc == 0) {
 884                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
 885                 if (rc == 0) {
 886                         if (scrub)
 887                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
 888                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 889                         return rc;
 890                 }
 891
 892                 if (rc > 0) {
 893                         if (!has_lma)
 894                                 /* It is FID-on-OST, but we do not know how
 895                                  * to generate its FID, ignore it directly. */
 896                                 rc = SCRUB_NEXT_CONTINUE;
 897                         else
 898                                 /* It is 2.4 OST-object. */
 899                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 900                         return rc;
 901                 }
 902
 903                 if (rc != -ENODATA)
 904                         return rc;
 905
 906                 if (!has_lma) {
 907                         if (dev->od_scrub.os_convert_igif) {
 908                                 lu_igif_build(fid, inode->i_ino,
 909                                               inode->i_generation);
 910                                 if (scrub)
 911                                         rc = SCRUB_NEXT_NOLMA;
 912                                 else
 913                                         rc = 0;
 914                         } else {
 915                                 /* It may be FID-on-OST, or may be FID for
 916                                  * non-MDT0, anyway, we do not know how to
 917                                  * generate its FID, ignore it directly. */
 918                                 rc = SCRUB_NEXT_CONTINUE;
 919                         }
 920                         return rc;
 921                 }
 922
 923                 /* For OI scrub case only: the object has LMA but has no ff
 924                  * (or ff crashed). It may be MDT-object, may be OST-object
 925                  * with crashed ff. The last check is local FLDB. */
 926                 rc = osd_scrub_check_local_fldb(info, dev, fid);
 927         }
 928
 929         return rc;
 930 }
 931
 932 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 933                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 934                         struct super_block *sb, bool scrub)
 935 {
 936         struct inode *inode;
 937         int           rc;
 938         ENTRY;
 939
 940         /* Not handle the backend root object and agent parent object.
 941          * They are neither visible to namespace nor have OI mappings. */
 942         if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
 943                      pos == osd_remote_parent_ino(dev)))
 944                 RETURN(SCRUB_NEXT_CONTINUE);
 945
 946         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 947         inode = osd_iget(info, dev, lid);
 948         if (IS_ERR(inode)) {
 949                 rc = PTR_ERR(inode);
 950                 /* The inode may be removed after bitmap searching, or the
 951                  * file is new created without inode initialized yet. */
 952                 if (rc == -ENOENT || rc == -ESTALE)
 953                         RETURN(SCRUB_NEXT_CONTINUE);
 954
 955                 CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: "
 956                        "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name,
 957                        pos, rc);
 958                 RETURN(rc);
 959         }
 960
 961         if (scrub &&
 962             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 963                 /* Only skip it for the first OI scrub accessing. */
 964                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 965                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
 966         }
 967
 968         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
 969
 970         GOTO(put, rc);
 971
 972 put:
 973         iput(inode);
 974         return rc;
 975 }
 976
 977 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 978                           struct osd_iit_param *param,
 979                           struct osd_idmap_cache **oic, const bool noslot)
 980 {
 981         struct osd_scrub     *scrub  = &dev->od_scrub;
 982         struct ptlrpc_thread *thread = &scrub->os_thread;
 983         struct lu_fid        *fid;
 984         struct osd_inode_id  *lid;
 985         int                   rc;
 986
 987         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
 988                 struct l_wait_info lwi;
 989
 990                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
 991                 if (likely(lwi.lwi_timeout > 0))
 992                         l_wait_event(thread->t_ctl_waitq,
 993                                 !list_empty(&scrub->os_inconsistent_items) ||
 994                                 !thread_is_running(thread),
 995                                 &lwi);
 996         }
 997
 998         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 999                 spin_lock(&scrub->os_lock);
1000                 thread_set_flags(thread, SVC_STOPPING);
1001                 spin_unlock(&scrub->os_lock);
1002                 return SCRUB_NEXT_CRASH;
1003         }
1004
1005         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
1006                 return SCRUB_NEXT_FATAL;
1007
1008         if (unlikely(!thread_is_running(thread)))
1009                 return SCRUB_NEXT_EXIT;
1010
1011         if (!list_empty(&scrub->os_inconsistent_items)) {
1012                 spin_lock(&scrub->os_lock);
1013                 if (likely(!list_empty(&scrub->os_inconsistent_items))) {
1014                         struct osd_inconsistent_item *oii;
1015
1016                         oii = list_entry(scrub->os_inconsistent_items.next,
1017                                 struct osd_inconsistent_item, oii_list);
1018                         list_del_init(&oii->oii_list);
1019                         spin_unlock(&scrub->os_lock);
1020
1021                         *oic = &oii->oii_cache;
1022                         scrub->os_in_prior = 1;
1023
1024                         return 0;
1025                 }
1026                 spin_unlock(&scrub->os_lock);
1027         }
1028
1029         if (noslot)
1030                 return SCRUB_NEXT_WAIT;
1031
1032         rc = osd_iit_next(param, &scrub->os_pos_current);
1033         if (rc != 0)
1034                 return rc;
1035
1036         *oic = &scrub->os_oic;
1037         fid = &(*oic)->oic_fid;
1038         lid = &(*oic)->oic_lid;
1039         rc = osd_iit_iget(info, dev, fid, lid,
1040                           scrub->os_pos_current, param->sb, true);
1041         return rc;
1042 }
1043
1044 static int osd_preload_next(struct osd_thread_info *info,
1045                             struct osd_device *dev, struct osd_iit_param *param,
1046                             struct osd_idmap_cache **oic, const bool noslot)
1047 {
1048         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
1049         struct osd_scrub        *scrub;
1050         struct ptlrpc_thread    *thread;
1051         int                      rc;
1052
1053         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1054         if (rc != 0)
1055                 return rc;
1056
1057         scrub = &dev->od_scrub;
1058         thread = &scrub->os_thread;
1059         if (thread_is_running(thread) &&
1060             ooc->ooc_pos_preload >= scrub->os_pos_current)
1061                 return SCRUB_NEXT_EXIT;
1062
1063         rc = osd_iit_iget(info, dev,
1064                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1065                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1066                           ooc->ooc_pos_preload, param->sb, false);
1067         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1068          * ignore the failure, so it still need to skip the inode next time. */
1069         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1070         return rc;
1071 }
1072
1073 static inline int
1074 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1075 {
1076         spin_lock(&scrub->os_lock);
1077         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1078             !list_empty(&scrub->os_inconsistent_items) ||
1079             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1080                 scrub->os_waiting = 0;
1081         else
1082                 scrub->os_waiting = 1;
1083         spin_unlock(&scrub->os_lock);
1084
1085         return !scrub->os_waiting;
1086 }
1087
1088 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1089                           struct osd_iit_param *param,
1090                           struct osd_idmap_cache *oic, bool *noslot, int rc)
1091 {
1092         struct l_wait_info       lwi    = { 0 };
1093         struct osd_scrub        *scrub  = &dev->od_scrub;
1094         struct scrub_file       *sf     = &scrub->os_file;
1095         struct ptlrpc_thread    *thread = &scrub->os_thread;
1096         struct osd_otable_it    *it     = dev->od_otable_it;
1097         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
1098
1099         switch (rc) {
1100         case SCRUB_NEXT_CONTINUE:
1101                 goto next;
1102         case SCRUB_NEXT_WAIT:
1103                 goto wait;
1104         case SCRUB_NEXT_NOSCRUB:
1105                 down_write(&scrub->os_rwsem);
1106                 scrub->os_new_checked++;
1107                 sf->sf_items_noscrub++;
1108                 up_write(&scrub->os_rwsem);
1109                 goto next;
1110         }
1111
1112         rc = osd_scrub_check_update(info, dev, oic, rc);
1113         if (rc != 0) {
1114                 scrub->os_in_prior = 0;
1115                 return rc;
1116         }
1117
1118         rc = osd_scrub_checkpoint(scrub);
1119         if (rc != 0) {
1120                 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1121                        "rc = %d\n", osd_scrub2name(scrub),
1122                        scrub->os_pos_current, rc);
1123                 /* Continue, as long as the scrub itself can go ahead. */
1124         }
1125
1126         if (scrub->os_in_prior) {
1127                 scrub->os_in_prior = 0;
1128                 return 0;
1129         }
1130
1131 next:
1132         scrub->os_pos_current = param->gbase + ++(param->offset);
1133
1134 wait:
1135         if (it != NULL && it->ooi_waiting && ooc != NULL &&
1136             ooc->ooc_pos_preload < scrub->os_pos_current) {
1137                 spin_lock(&scrub->os_lock);
1138                 it->ooi_waiting = 0;
1139                 wake_up_all(&thread->t_ctl_waitq);
1140                 spin_unlock(&scrub->os_lock);
1141         }
1142
1143         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1144                 return 0;
1145
1146         if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1147                 *noslot = false;
1148                 return 0;
1149         }
1150
1151         if (it != NULL)
1152                 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1153                              &lwi);
1154
1155         if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1156                 *noslot = false;
1157         else
1158                 *noslot = true;
1159         return 0;
1160 }
1161
1162 static int osd_preload_exec(struct osd_thread_info *info,
1163                             struct osd_device *dev, struct osd_iit_param *param,
1164                             struct osd_idmap_cache *oic, bool *noslot, int rc)
1165 {
1166         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1167
1168         if (rc == 0) {
1169                 ooc->ooc_cached_items++;
1170                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1171                                         ~OSD_OTABLE_IT_CACHE_MASK;
1172         }
1173         return rc > 0 ? 0 : rc;
1174 }
1175
1176 #define SCRUB_IT_ALL    1
1177 #define SCRUB_IT_CRASH  2
1178
1179 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1180                            bool inconsistent)
1181 {
1182         struct osd_scrub     *scrub  = &dev->od_scrub;
1183         struct ptlrpc_thread *thread = &scrub->os_thread;
1184         struct scrub_file    *sf     = &scrub->os_file;
1185         int                   rc;
1186         ENTRY;
1187
1188         LASSERT(!(flags & SS_AUTO_PARTIAL));
1189
1190         down_write(&scrub->os_rwsem);
1191         scrub->os_in_join = 1;
1192         if (flags & SS_SET_FAILOUT)
1193                 sf->sf_param |= SP_FAILOUT;
1194         else if (flags & SS_CLEAR_FAILOUT)
1195                 sf->sf_param &= ~SP_FAILOUT;
1196
1197         if (flags & SS_SET_DRYRUN)
1198                 sf->sf_param |= SP_DRYRUN;
1199         else if (flags & SS_CLEAR_DRYRUN)
1200                 sf->sf_param &= ~SP_DRYRUN;
1201
1202         if (flags & SS_RESET) {
1203                 osd_scrub_file_reset(scrub,
1204                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1205                         inconsistent ? SF_INCONSISTENT : 0);
1206                 sf->sf_status = SS_SCANNING;
1207         }
1208
1209         if (flags & SS_AUTO_FULL) {
1210                 sf->sf_flags |= SF_AUTO;
1211                 scrub->os_full_speed = 1;
1212         }
1213
1214         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1215                 scrub->os_full_speed = 1;
1216         else
1217                 scrub->os_full_speed = 0;
1218
1219         scrub->os_new_checked = 0;
1220         if (sf->sf_pos_last_checkpoint != 0)
1221                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1222         else
1223                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1224
1225         scrub->os_pos_current = sf->sf_pos_latest_start;
1226         sf->sf_time_latest_start = cfs_time_current_sec();
1227         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1228         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1229         rc = osd_scrub_file_store(scrub);
1230         if (rc != 0)
1231                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
1232                        "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1233
1234         spin_lock(&scrub->os_lock);
1235         scrub->os_waiting = 0;
1236         scrub->os_paused = 0;
1237         scrub->os_partial_scan = 0;
1238         scrub->os_in_join = 0;
1239         scrub->os_full_scrub = 0;
1240         spin_unlock(&scrub->os_lock);
1241         wake_up_all(&thread->t_ctl_waitq);
1242         up_write(&scrub->os_rwsem);
1243
1244         EXIT;
1245 }
1246
1247 static int osd_inode_iteration(struct osd_thread_info *info,
1248                                struct osd_device *dev, __u32 max, bool preload)
1249 {
1250         struct osd_scrub     *scrub  = &dev->od_scrub;
1251         struct ptlrpc_thread *thread = &scrub->os_thread;
1252         struct scrub_file    *sf     = &scrub->os_file;
1253         osd_iit_next_policy   next;
1254         osd_iit_exec_policy   exec;
1255         __u32                *pos;
1256         __u32                *count;
1257         struct osd_iit_param  param  = { NULL };
1258         struct l_wait_info    lwi    = { 0 };
1259         __u32                 limit;
1260         int                   rc;
1261         bool                  noslot = true;
1262         ENTRY;
1263
1264         param.sb = osd_sb(dev);
1265         if (preload)
1266                 goto full;
1267
1268         while (scrub->os_partial_scan && !scrub->os_in_join) {
1269                 struct osd_idmap_cache *oic = NULL;
1270
1271                 rc = osd_scrub_next(info, dev, &param, &oic, noslot);
1272                 switch (rc) {
1273                 case SCRUB_NEXT_EXIT:
1274                         RETURN(0);
1275                 case SCRUB_NEXT_CRASH:
1276                         RETURN(SCRUB_IT_CRASH);
1277                 case SCRUB_NEXT_FATAL:
1278                         RETURN(-EINVAL);
1279                 case SCRUB_NEXT_WAIT: {
1280                         struct kstatfs *ksfs = &info->oti_ksfs;
1281                         __u64 saved_flags;
1282
1283                         if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1284                             unlikely(sf->sf_items_updated_prior == 0))
1285                                 goto wait;
1286
1287                         if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1288                             scrub->os_full_scrub) {
1289                                 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1290                                                true);
1291                                 goto full;
1292                         }
1293
1294                         rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1295                         if (rc == 0) {
1296                                 __u64 used = ksfs->f_files - ksfs->f_ffree;
1297
1298                                 do_div(used, sf->sf_items_updated_prior);
1299                                 /* If we hit too much inconsistent OI
1300                                  * mappings during the partial scan,
1301                                  * then scan the device completely. */
1302                                 if (used < dev->od_full_scrub_ratio) {
1303                                         osd_scrub_join(dev,
1304                                                 SS_AUTO_FULL | SS_RESET, true);
1305                                         goto full;
1306                                 }
1307                         }
1308
1309 wait:
1310                         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1311                             cfs_fail_val > 0)
1312                                 continue;
1313
1314                         saved_flags = sf->sf_flags;
1315                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1316                                           SF_UPGRADE | SF_AUTO);
1317                         sf->sf_status = SS_COMPLETED;
1318                         l_wait_event(thread->t_ctl_waitq,
1319                                      !thread_is_running(thread) ||
1320                                      !scrub->os_partial_scan ||
1321                                      scrub->os_in_join ||
1322                                      !list_empty(&scrub->os_inconsistent_items),
1323                                      &lwi);
1324                         sf->sf_flags = saved_flags;
1325                         sf->sf_status = SS_SCANNING;
1326
1327                         if (unlikely(!thread_is_running(thread)))
1328                                 RETURN(0);
1329
1330                         if (!scrub->os_partial_scan || scrub->os_in_join)
1331                                 goto full;
1332
1333                         continue;
1334                 }
1335                 default:
1336                         LASSERTF(rc == 0, "rc = %d\n", rc);
1337
1338                         osd_scrub_exec(info, dev, &param, oic, &noslot, rc);
1339                         break;
1340                 }
1341         }
1342
1343 full:
1344         if (!preload) {
1345                 l_wait_event(thread->t_ctl_waitq,
1346                              !thread_is_running(thread) || !scrub->os_in_join,
1347                              &lwi);
1348
1349                 if (unlikely(!thread_is_running(thread)))
1350                         RETURN(0);
1351         }
1352
1353         noslot = false;
1354         if (!preload) {
1355                 next = osd_scrub_next;
1356                 exec = osd_scrub_exec;
1357                 pos = &scrub->os_pos_current;
1358                 count = &scrub->os_new_checked;
1359         } else {
1360                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1361
1362                 next = osd_preload_next;
1363                 exec = osd_preload_exec;
1364                 pos = &ooc->ooc_pos_preload;
1365                 count = &ooc->ooc_cached_items;
1366         }
1367         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1368
1369         while (*pos <= limit && *count < max) {
1370                 struct osd_idmap_cache *oic = NULL;
1371                 struct ldiskfs_group_desc *desc;
1372
1373                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1374                 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1375                 if (desc == NULL)
1376                         RETURN(-EIO);
1377
1378                 ldiskfs_lock_group(param.sb, param.bg);
1379                 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1380                         ldiskfs_unlock_group(param.sb, param.bg);
1381                         *pos = 1 + (param.bg + 1) *
1382                                 LDISKFS_INODES_PER_GROUP(param.sb);
1383                         continue;
1384                 }
1385                 ldiskfs_unlock_group(param.sb, param.bg);
1386
1387                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1388                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1389                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1390                 if (param.bitmap == NULL) {
1391                         CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1392                                "scrub will stop, urgent mode\n",
1393                                osd_scrub2name(scrub), (__u32)param.bg);
1394                         RETURN(-EIO);
1395                 }
1396
1397                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1398                        *count < max) {
1399                         if (param.offset +
1400                                 ldiskfs_itable_unused_count(param.sb, desc) >
1401                             LDISKFS_INODES_PER_GROUP(param.sb))
1402                                 goto next_group;
1403
1404                         rc = next(info, dev, &param, &oic, noslot);
1405                         switch (rc) {
1406                         case SCRUB_NEXT_BREAK:
1407                                 goto next_group;
1408                         case SCRUB_NEXT_EXIT:
1409                                 brelse(param.bitmap);
1410                                 RETURN(0);
1411                         case SCRUB_NEXT_CRASH:
1412                                 brelse(param.bitmap);
1413                                 RETURN(SCRUB_IT_CRASH);
1414                         case SCRUB_NEXT_FATAL:
1415                                 brelse(param.bitmap);
1416                                 RETURN(-EINVAL);
1417                         }
1418
1419                         rc = exec(info, dev, &param, oic, &noslot, rc);
1420                         if (rc != 0) {
1421                                 brelse(param.bitmap);
1422                                 RETURN(rc);
1423                         }
1424                 }
1425
1426 next_group:
1427                 brelse(param.bitmap);
1428         }
1429
1430         if (*pos > limit)
1431                 RETURN(SCRUB_IT_ALL);
1432         RETURN(0);
1433 }
1434
1435 static int osd_otable_it_preload(const struct lu_env *env,
1436                                  struct osd_otable_it *it)
1437 {
1438         struct osd_device       *dev   = it->ooi_dev;
1439         struct osd_scrub        *scrub = &dev->od_scrub;
1440         struct osd_otable_cache *ooc   = &it->ooi_cache;
1441         int                      rc;
1442         ENTRY;
1443
1444         rc = osd_inode_iteration(osd_oti_get(env), dev,
1445                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1446         if (rc == SCRUB_IT_ALL)
1447                 it->ooi_all_cached = 1;
1448
1449         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1450                 spin_lock(&scrub->os_lock);
1451                 scrub->os_waiting = 0;
1452                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1453                 spin_unlock(&scrub->os_lock);
1454         }
1455
1456         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1457 }
1458
1459 static int osd_scrub_main(void *args)
1460 {
1461         struct lu_env         env;
1462         struct osd_device    *dev    = (struct osd_device *)args;
1463         struct osd_scrub     *scrub  = &dev->od_scrub;
1464         struct ptlrpc_thread *thread = &scrub->os_thread;
1465         int                   rc;
1466         ENTRY;
1467
1468         rc = lu_env_init(&env, LCT_LOCAL);
1469         if (rc != 0) {
1470                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1471                        osd_scrub2name(scrub), rc);
1472                 GOTO(noenv, rc);
1473         }
1474
1475         rc = osd_scrub_prep(dev);
1476         if (rc != 0) {
1477                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1478                        osd_scrub2name(scrub), rc);
1479                 GOTO(out, rc);
1480         }
1481
1482         if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1483                 struct l_wait_info lwi = { 0 };
1484                 struct osd_otable_it *it = dev->od_otable_it;
1485                 struct osd_otable_cache *ooc = &it->ooi_cache;
1486
1487                 l_wait_event(thread->t_ctl_waitq,
1488                              it->ooi_user_ready || !thread_is_running(thread),
1489                              &lwi);
1490                 if (unlikely(!thread_is_running(thread)))
1491                         GOTO(post, rc = 0);
1492
1493                 scrub->os_pos_current = ooc->ooc_pos_preload;
1494         }
1495
1496         CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1497                osd_scrub2name(scrub), scrub->os_start_flags,
1498                scrub->os_pos_current);
1499
1500         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1501         if (unlikely(rc == SCRUB_IT_CRASH))
1502                 GOTO(out, rc = -EINVAL);
1503         GOTO(post, rc);
1504
1505 post:
1506         rc = osd_scrub_post(scrub, rc);
1507         CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1508                osd_scrub2name(scrub), scrub->os_pos_current, rc);
1509
1510 out:
1511         while (!list_empty(&scrub->os_inconsistent_items)) {
1512                 struct osd_inconsistent_item *oii;
1513
1514                 oii = list_entry(scrub->os_inconsistent_items.next,
1515                                      struct osd_inconsistent_item, oii_list);
1516                 list_del_init(&oii->oii_list);
1517                 OBD_FREE_PTR(oii);
1518         }
1519         lu_env_fini(&env);
1520
1521 noenv:
1522         spin_lock(&scrub->os_lock);
1523         thread_set_flags(thread, SVC_STOPPED);
1524         wake_up_all(&thread->t_ctl_waitq);
1525         spin_unlock(&scrub->os_lock);
1526         return rc;
1527 }
1528
1529 /* initial OI scrub */
1530
1531 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1532                          struct dentry *, filldir_t filldir);
1533
1534 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1535                                loff_t offset, __u64 ino, unsigned d_type);
1536 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1537                            loff_t offset, __u64 ino, unsigned d_type);
1538 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1539                            loff_t offset, __u64 ino, unsigned d_type);
1540 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1541                             loff_t offset, __u64 ino, unsigned d_type);
1542
1543 static int
1544 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1545                      struct dentry *dentry, filldir_t filldir);
1546 static int
1547 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1548                   struct dentry *dentry, filldir_t filldir);
1549
1550 static int
1551 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1552                      struct dentry *dentry, filldir_t filldir);
1553
1554 enum osd_lf_flags {
1555         OLF_SCAN_SUBITEMS       = 0x0001,
1556         OLF_HIDE_FID            = 0x0002,
1557         OLF_SHOW_NAME           = 0x0004,
1558         OLF_NO_OI               = 0x0008,
1559         OLF_IDX_IN_FID          = 0x0010,
1560 };
1561
1562 struct osd_lf_map {
1563         char            *olm_name;
1564         struct lu_fid    olm_fid;
1565         __u16            olm_flags;
1566         __u16            olm_namelen;
1567         scandir_t        olm_scandir;
1568         filldir_t        olm_filldir;
1569 };
1570
1571 /* Add the new introduced local files in the list in the future. */
1572 static const struct osd_lf_map osd_lf_maps[] = {
1573         /* CATALOGS */
1574         { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1575                 sizeof(CATLIST) - 1, NULL, NULL },
1576
1577         /* CONFIGS */
1578         { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1579                 OLF_SCAN_SUBITEMS, sizeof(MOUNT_CONFIGS_DIR) - 1,
1580                 osd_ios_general_scan, osd_ios_varfid_fill },
1581
1582         /* NIDTBL_VERSIONS */
1583         { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1584                 sizeof(MGS_NIDTBL_DIR) - 1, osd_ios_general_scan,
1585                 osd_ios_varfid_fill },
1586
1587         /* PENDING */
1588         { "PENDING", { 0, 0, 0 }, 0, sizeof("PENDING") - 1, NULL, NULL },
1589
1590         /* ROOT */
1591         { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 },
1592                 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, sizeof("ROOT") - 1,
1593                 osd_ios_ROOT_scan, NULL },
1594
1595         /* changelog_catalog */
1596         { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, sizeof(CHANGELOG_CATALOG) - 1,
1597                 NULL, NULL },
1598
1599         /* changelog_users */
1600         { CHANGELOG_USERS, { 0, 0, 0 }, 0, sizeof(CHANGELOG_USERS) - 1,
1601                 NULL, NULL },
1602
1603         /* fld */
1604         { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1605                 sizeof("fld") - 1, NULL, NULL },
1606
1607         /* last_rcvd */
1608         { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1609                 sizeof(LAST_RCVD) - 1, NULL, NULL },
1610
1611         /* reply_data */
1612         { REPLY_DATA, { FID_SEQ_LOCAL_FILE, REPLY_DATA_OID, 0 }, OLF_SHOW_NAME,
1613                 sizeof(REPLY_DATA) - 1, NULL, NULL },
1614
1615         /* lov_objid */
1616         { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1617                 sizeof(LOV_OBJID) - 1, NULL, NULL },
1618
1619         /* lov_objseq */
1620         { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1621                 OLF_SHOW_NAME, sizeof(LOV_OBJSEQ) - 1, NULL, NULL },
1622
1623         /* quota_master */
1624         { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QMT_DIR) - 1,
1625                 osd_ios_general_scan, osd_ios_varfid_fill },
1626
1627         /* quota_slave */
1628         { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QSD_DIR) - 1,
1629                 osd_ios_general_scan, osd_ios_varfid_fill },
1630
1631         /* seq_ctl */
1632         { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1633                 OLF_SHOW_NAME, sizeof("seq_ctl") - 1, NULL, NULL },
1634
1635         /* seq_srv */
1636         { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1637                 OLF_SHOW_NAME, sizeof("seq_srv") - 1, NULL, NULL },
1638
1639         /* health_check */
1640         { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1641                 OLF_SHOW_NAME, sizeof(HEALTH_CHECK) - 1, NULL, NULL },
1642
1643         /* LFSCK */
1644         { LFSCK_DIR, { 0, 0, 0 }, 0, sizeof(LFSCK_DIR) - 1,
1645                 osd_ios_general_scan, osd_ios_varfid_fill },
1646
1647         /* lfsck_bookmark */
1648         { LFSCK_BOOKMARK, { 0, 0, 0 }, 0, sizeof(LFSCK_BOOKMARK) - 1,
1649                 NULL, NULL },
1650
1651         /* lfsck_layout */
1652         { LFSCK_LAYOUT, { 0, 0, 0 }, 0, sizeof(LFSCK_LAYOUT) - 1,
1653                 NULL, NULL },
1654
1655         /* lfsck_namespace */
1656         { LFSCK_NAMESPACE, { 0, 0, 0 }, 0, sizeof(LFSCK_NAMESPACE) - 1,
1657                 NULL, NULL },
1658
1659         /* OBJECTS, upgrade from old device */
1660         { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(OBJECTS) - 1,
1661                 osd_ios_OBJECTS_scan, NULL },
1662
1663         /* lquota_v2.user, upgrade from old device */
1664         { "lquota_v2.user", { 0, 0, 0 }, 0, sizeof("lquota_v2.user") - 1,
1665                 NULL, NULL },
1666
1667         /* lquota_v2.group, upgrade from old device */
1668         { "lquota_v2.group", { 0, 0, 0 }, 0, sizeof("lquota_v2.group") - 1,
1669                 NULL, NULL },
1670
1671         /* LAST_GROUP, upgrade from old device */
1672         { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1673                 OLF_SHOW_NAME, sizeof("LAST_GROUP") - 1, NULL, NULL },
1674
1675         /* committed batchid for cross-MDT operation */
1676         { "BATCHID", { FID_SEQ_LOCAL_FILE, BATCHID_COMMITTED_OID, 0 },
1677                 OLF_SHOW_NAME, sizeof("BATCHID") - 1, NULL, NULL },
1678
1679         /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1680          * and f_oid = index for their log files.  See lu_update_log{_dir}_fid()
1681          * for more details. */
1682
1683         /* update_log */
1684         { "update_log", { FID_SEQ_UPDATE_LOG, 0, 0 },
1685                 OLF_SHOW_NAME | OLF_IDX_IN_FID, sizeof("update_log") - 1,
1686                 NULL, NULL },
1687
1688         /* update_log_dir */
1689         { "update_log_dir", { FID_SEQ_UPDATE_LOG_DIR, 0, 0 },
1690                 OLF_SHOW_NAME | OLF_SCAN_SUBITEMS | OLF_IDX_IN_FID,
1691                 sizeof("update_log_dir") - 1,
1692                 osd_ios_general_scan, osd_ios_uld_fill },
1693
1694         /* lost+found */
1695         { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 },
1696                 OLF_SCAN_SUBITEMS, sizeof("lost+found") - 1,
1697                 osd_ios_general_scan, osd_ios_lf_fill },
1698
1699         { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1700 };
1701
1702 /* Add the new introduced files under .lustre/ in the list in the future. */
1703 static const struct osd_lf_map osd_dl_maps[] = {
1704         /* .lustre/fid */
1705         { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1706                 sizeof("fid") - 1, NULL, NULL },
1707
1708         /* .lustre/lost+found */
1709         { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1710                 sizeof("lost+found") - 1, NULL, NULL },
1711
1712         { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1713 };
1714
1715 struct osd_ios_item {
1716         struct list_head oii_list;
1717         struct dentry   *oii_dentry;
1718         scandir_t        oii_scandir;
1719         filldir_t        oii_filldir;
1720 };
1721
1722 struct osd_ios_filldir_buf {
1723 #ifdef HAVE_DIR_CONTEXT
1724         /* please keep it as first member */
1725         struct dir_context       ctx;
1726 #endif
1727         struct osd_thread_info  *oifb_info;
1728         struct osd_device       *oifb_dev;
1729         struct dentry           *oifb_dentry;
1730 };
1731
1732 static inline struct dentry *
1733 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1734 {
1735         struct dentry *dentry;
1736
1737         dentry = ll_lookup_one_len(name, parent, namelen);
1738         if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1739                 dput(dentry);
1740                 return ERR_PTR(-ENOENT);
1741         }
1742
1743         return dentry;
1744 }
1745
1746 static int
1747 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1748                  scandir_t scandir, filldir_t filldir)
1749 {
1750         struct osd_ios_item *item;
1751         ENTRY;
1752
1753         OBD_ALLOC_PTR(item);
1754         if (item == NULL)
1755                 RETURN(-ENOMEM);
1756
1757         INIT_LIST_HEAD(&item->oii_list);
1758         item->oii_dentry = dget(dentry);
1759         item->oii_scandir = scandir;
1760         item->oii_filldir = filldir;
1761         list_add_tail(&item->oii_list, &dev->od_ios_list);
1762
1763         RETURN(0);
1764 }
1765
1766 /**
1767  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1768  *
1769  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1770  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1771  * reference the inode, or fixed if it is missing or references another inode.
1772  */
1773 static int
1774 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1775                  struct inode *inode, const struct lu_fid *fid, int flags)
1776 {
1777         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
1778         struct osd_inode_id     *id     = &info->oti_id;
1779         struct osd_inode_id     *id2    = &info->oti_id2;
1780         struct osd_scrub        *scrub  = &dev->od_scrub;
1781         struct scrub_file       *sf     = &scrub->os_file;
1782         struct lu_fid            tfid;
1783         int                      rc;
1784         ENTRY;
1785
1786         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1787         if (rc != 0 && rc != -ENODATA) {
1788                 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1789                        "rc = %d\n", osd_name(dev), rc);
1790
1791                 RETURN(rc);
1792         }
1793
1794         osd_id_gen(id, inode->i_ino, inode->i_generation);
1795         if (rc == -ENODATA) {
1796                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
1797                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1798                 } else {
1799                         tfid = *fid;
1800                         if (flags & OLF_IDX_IN_FID) {
1801                                 LASSERT(dev->od_index >= 0);
1802
1803                                 tfid.f_oid = dev->od_index;
1804                         }
1805                 }
1806                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1807                 if (rc != 0) {
1808                         CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1809                               "scrub: rc = %d\n", osd_name(dev), rc);
1810
1811                         RETURN(rc);
1812                 }
1813         } else {
1814                 if (lma->lma_compat & LMAC_NOT_IN_OI)
1815                         RETURN(0);
1816
1817                 tfid = lma->lma_self_fid;
1818         }
1819
1820         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1821         if (rc != 0) {
1822                 if (rc != -ENOENT)
1823                         RETURN(rc);
1824
1825                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1826                                                DTO_INDEX_INSERT, true, 0);
1827                 if (rc > 0)
1828                         rc = 0;
1829
1830                 RETURN(rc);
1831         }
1832
1833         if (osd_id_eq_strict(id, id2))
1834                 RETURN(0);
1835
1836         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1837                 osd_scrub_file_reset(scrub,
1838                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1839                                      SF_INCONSISTENT);
1840                 rc = osd_scrub_file_store(scrub);
1841                 if (rc != 0)
1842                         RETURN(rc);
1843         }
1844
1845         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1846                                        DTO_INDEX_UPDATE, true, 0);
1847         if (rc > 0)
1848                 rc = 0;
1849
1850         RETURN(rc);
1851 }
1852
1853 /**
1854  * It scans the /lost+found, and for the OST-object (with filter_fid
1855  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1856  */
1857 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1858                            loff_t offset, __u64 ino, unsigned d_type)
1859 {
1860         struct osd_ios_filldir_buf *fill_buf = buf;
1861         struct osd_thread_info     *info     = fill_buf->oifb_info;
1862         struct osd_device          *dev      = fill_buf->oifb_dev;
1863         struct lu_fid              *fid      = &info->oti_fid;
1864         struct osd_scrub           *scrub    = &dev->od_scrub;
1865         struct dentry              *parent   = fill_buf->oifb_dentry;
1866         struct dentry              *child;
1867         struct inode               *dir      = parent->d_inode;
1868         struct inode               *inode;
1869         int                         rc;
1870         ENTRY;
1871
1872         /* skip any '.' started names */
1873         if (name[0] == '.')
1874                 RETURN(0);
1875
1876         scrub->os_lf_scanned++;
1877         child = osd_ios_lookup_one_len(name, parent, namelen);
1878         if (IS_ERR(child)) {
1879                 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1880                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
1881                 RETURN(0);
1882         }
1883
1884         inode = child->d_inode;
1885         if (S_ISDIR(inode->i_mode)) {
1886                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1887                                       osd_ios_lf_fill);
1888                 if (rc != 0)
1889                         CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1890                               "rc = %d\n", osd_name(dev), namelen, name, rc);
1891                 GOTO(put, rc);
1892         }
1893
1894         if (!S_ISREG(inode->i_mode))
1895                 GOTO(put, rc = 0);
1896
1897         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1898         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1899                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1900                 if (rc == 0) {
1901                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1902                                "/lost+found.\n", namelen, name, PFID(fid));
1903                         scrub->os_lf_repaired++;
1904                 } else {
1905                         CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1906                                DFID": rc = %d\n",
1907                                osd_name(dev), namelen, name, PFID(fid), rc);
1908                 }
1909         }
1910
1911         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1912          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1913          *      can process them in furtuer. */
1914
1915         GOTO(put, rc);
1916
1917 put:
1918         if (rc < 0)
1919                 scrub->os_lf_failed++;
1920         dput(child);
1921         /* skip the failure to make the scanning to continue. */
1922         return 0;
1923 }
1924
1925 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1926                                loff_t offset, __u64 ino, unsigned d_type)
1927 {
1928         struct osd_ios_filldir_buf *fill_buf = buf;
1929         struct osd_device          *dev      = fill_buf->oifb_dev;
1930         struct dentry              *child;
1931         int                         rc;
1932         ENTRY;
1933
1934         /* skip any '.' started names */
1935         if (name[0] == '.')
1936                 RETURN(0);
1937
1938         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1939         if (IS_ERR(child))
1940                 RETURN(PTR_ERR(child));
1941
1942         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1943                               NULL, 0);
1944         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1945                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1946                                       osd_ios_varfid_fill);
1947         dput(child);
1948
1949         RETURN(rc);
1950 }
1951
1952 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1953                            loff_t offset, __u64 ino, unsigned d_type)
1954 {
1955         struct osd_ios_filldir_buf *fill_buf = buf;
1956         struct osd_device          *dev      = fill_buf->oifb_dev;
1957         const struct osd_lf_map    *map;
1958         struct dentry              *child;
1959         int                         rc       = 0;
1960         ENTRY;
1961
1962         /* skip any '.' started names */
1963         if (name[0] == '.')
1964                 RETURN(0);
1965
1966         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
1967                 if (map->olm_namelen != namelen)
1968                         continue;
1969
1970                 if (strncmp(map->olm_name, name, namelen) == 0)
1971                         break;
1972         }
1973
1974         if (map->olm_name == NULL)
1975                 RETURN(0);
1976
1977         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1978         if (IS_ERR(child))
1979                 RETURN(PTR_ERR(child));
1980
1981         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1982                               &map->olm_fid, map->olm_flags);
1983         dput(child);
1984
1985         RETURN(rc);
1986 }
1987
1988 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1989                             loff_t offset, __u64 ino, unsigned d_type)
1990 {
1991         struct osd_ios_filldir_buf *fill_buf = buf;
1992         struct dentry              *child;
1993         struct lu_fid               tfid;
1994         int                         rc       = 0;
1995         ENTRY;
1996
1997         /* skip any non-DFID format name */
1998         if (name[0] != '[')
1999                 RETURN(0);
2000
2001         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2002         if (IS_ERR(child))
2003                 RETURN(PTR_ERR(child));
2004
2005         /* skip the start '[' */
2006         sscanf(&name[1], SFID, RFID(&tfid));
2007         if (fid_is_sane(&tfid))
2008                 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
2009                                       child->d_inode, &tfid, 0);
2010         else
2011                 rc = -EIO;
2012         dput(child);
2013
2014         RETURN(rc);
2015 }
2016
2017 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
2018                              loff_t offset, __u64 ino, unsigned d_type)
2019 {
2020         struct osd_ios_filldir_buf *fill_buf = buf;
2021         struct osd_device          *dev      = fill_buf->oifb_dev;
2022         const struct osd_lf_map    *map;
2023         struct dentry              *child;
2024         int                         rc       = 0;
2025         ENTRY;
2026
2027         /* skip any '.' started names */
2028         if (name[0] == '.')
2029                 RETURN(0);
2030
2031         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2032                 if (map->olm_namelen != namelen)
2033                         continue;
2034
2035                 if (strncmp(map->olm_name, name, namelen) == 0)
2036                         break;
2037         }
2038
2039         if (map->olm_name == NULL)
2040                 RETURN(0);
2041
2042         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2043         if (IS_ERR(child))
2044                 RETURN(PTR_ERR(child));
2045
2046         if (!(map->olm_flags & OLF_NO_OI))
2047                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2048                                       &map->olm_fid, map->olm_flags);
2049         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2050                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2051                                       map->olm_filldir);
2052         dput(child);
2053
2054         RETURN(rc);
2055 }
2056
2057 static int
2058 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2059                      struct dentry *dentry, filldir_t filldir)
2060 {
2061         struct osd_ios_filldir_buf    buf   = {
2062 #ifdef HAVE_DIR_CONTEXT
2063                                                 .ctx.actor = filldir,
2064 #endif
2065                                                 .oifb_info = info,
2066                                                 .oifb_dev = dev,
2067                                                 .oifb_dentry = dentry };
2068         struct file                  *filp  = &info->oti_file;
2069         struct inode                 *inode = dentry->d_inode;
2070         const struct file_operations *fops  = inode->i_fop;
2071         int                           rc;
2072         ENTRY;
2073
2074         LASSERT(filldir != NULL);
2075
2076         filp->f_pos = 0;
2077         filp->f_path.dentry = dentry;
2078         filp->f_mode = FMODE_64BITHASH;
2079         filp->f_mapping = inode->i_mapping;
2080         filp->f_op = fops;
2081         filp->private_data = NULL;
2082         set_file_inode(filp, inode);
2083
2084 #ifdef HAVE_DIR_CONTEXT
2085         buf.ctx.pos = filp->f_pos;
2086         rc = fops->iterate(filp, &buf.ctx);
2087         filp->f_pos = buf.ctx.pos;
2088 #else
2089         rc = fops->readdir(filp, &buf, filldir);
2090 #endif
2091         fops->release(inode, filp);
2092
2093         RETURN(rc);
2094 }
2095
2096 static int
2097 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2098                   struct dentry *dentry, filldir_t filldir)
2099 {
2100         struct osd_scrub  *scrub  = &dev->od_scrub;
2101         struct scrub_file *sf     = &scrub->os_file;
2102         struct dentry     *child;
2103         int                rc;
2104         ENTRY;
2105
2106         /* It is existing MDT0 device. We only allow the case of object without
2107          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2108          * can generate IGIF mode FID for the object and related OI mapping. If
2109          * it is on other MDTs, then becuase file-level backup/restore, related
2110          * OI mapping may be invalid already, we do not know which is the right
2111          * FID for the object. We only allow IGIF objects to reside on the MDT0.
2112          *
2113          * XXX: For the case of object on non-MDT0 device with neither LMA nor
2114          *      "fid" xattr, then something crashed. We cannot re-generate the
2115          *      FID directly, instead, the OI scrub will scan the OI structure
2116          *      and try to re-generate the LMA from the OI mapping. But if the
2117          *      OI mapping crashed or lost also, then we have to give up under
2118          *      double failure cases. */
2119         scrub->os_convert_igif = 1;
2120         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2121                                        strlen(dot_lustre_name));
2122         if (IS_ERR(child)) {
2123                 rc = PTR_ERR(child);
2124                 if (rc == -ENOENT) {
2125                         /* It is 1.8 MDT device. */
2126                         if (!(sf->sf_flags & SF_UPGRADE)) {
2127                                 osd_scrub_file_reset(scrub,
2128                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2129                                         SF_UPGRADE);
2130                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2131                                 rc = osd_scrub_file_store(scrub);
2132                         } else {
2133                                 rc = 0;
2134                         }
2135                 }
2136         } else {
2137                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2138                  * so the client will get IGIF for the ".lustre" object when
2139                  * the MDT restart.
2140                  *
2141                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2142                  * it does not know whether there are some old clients cached
2143                  * the ".lustre" IGIF during the upgrading. Two choices:
2144                  *
2145                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2146                  *    It will allow the old connected clients to access the
2147                  *    ".lustre" with cached IGIF. But it will cause others
2148                  *    on the MDT failed to check "fid_is_dot_lustre()".
2149                  *
2150                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2151                  *    for ".lustre" in spite of whether there are some clients
2152                  *    cached the ".lustre" IGIF or not. It enables the check
2153                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
2154                  *    that the old connected clients cannot access the ".lustre"
2155                  *    with the cached IGIF.
2156                  *
2157                  * Usually, it is rare case for the old connected clients
2158                  * to access the ".lustre" with cached IGIF. So we prefer
2159                  * to the solution 2). */
2160                 rc = osd_ios_scan_one(info, dev, child->d_inode,
2161                                       &LU_DOT_LUSTRE_FID, 0);
2162                 if (rc == 0)
2163                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2164                                               osd_ios_dl_fill);
2165                 dput(child);
2166         }
2167
2168         RETURN(rc);
2169 }
2170
2171 static int
2172 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2173                      struct dentry *dentry, filldir_t filldir)
2174 {
2175         struct osd_scrub  *scrub  = &dev->od_scrub;
2176         struct scrub_file *sf     = &scrub->os_file;
2177         struct dentry     *child;
2178         int                rc;
2179         ENTRY;
2180
2181         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2182                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2183                 rc = osd_scrub_file_store(scrub);
2184                 if (rc != 0)
2185                         RETURN(rc);
2186         }
2187
2188         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2189         if (!IS_ERR(child)) {
2190                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2191                 dput(child);
2192         } else {
2193                 rc = PTR_ERR(child);
2194         }
2195
2196         if (rc != 0 && rc != -ENOENT)
2197                 RETURN(rc);
2198
2199         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2200         if (!IS_ERR(child)) {
2201                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2202                 dput(child);
2203         } else {
2204                 rc = PTR_ERR(child);
2205         }
2206
2207         if (rc == -ENOENT)
2208                 rc = 0;
2209
2210         RETURN(rc);
2211 }
2212
2213 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2214                                 struct osd_device *dev)
2215 {
2216         struct osd_ios_item     *item    = NULL;
2217         scandir_t                scandir = osd_ios_general_scan;
2218         filldir_t                filldir = osd_ios_root_fill;
2219         struct dentry           *dentry  = osd_sb(dev)->s_root;
2220         const struct osd_lf_map *map     = osd_lf_maps;
2221         int                      rc;
2222         ENTRY;
2223
2224         /* Lookup IGIF in OI by force for initial OI scrub. */
2225         dev->od_igif_inoi = 1;
2226
2227         while (1) {
2228                 rc = scandir(info, dev, dentry, filldir);
2229                 if (item != NULL) {
2230                         dput(item->oii_dentry);
2231                         OBD_FREE_PTR(item);
2232                 }
2233
2234                 if (rc != 0)
2235                         break;
2236
2237                 if (list_empty(&dev->od_ios_list))
2238                         break;
2239
2240                 item = list_entry(dev->od_ios_list.next,
2241                                   struct osd_ios_item, oii_list);
2242                 list_del_init(&item->oii_list);
2243
2244                 LASSERT(item->oii_scandir != NULL);
2245                 scandir = item->oii_scandir;
2246                 filldir = item->oii_filldir;
2247                 dentry = item->oii_dentry;
2248         }
2249
2250         while (!list_empty(&dev->od_ios_list)) {
2251                 item = list_entry(dev->od_ios_list.next,
2252                                   struct osd_ios_item, oii_list);
2253                 list_del_init(&item->oii_list);
2254                 dput(item->oii_dentry);
2255                 OBD_FREE_PTR(item);
2256         }
2257
2258         if (rc != 0)
2259                 RETURN(rc);
2260
2261         /* There maybe the case that the object has been removed, but its OI
2262          * mapping is still in the OI file, such as the "CATALOGS" after MDT
2263          * file-level backup/restore. So here cleanup the stale OI mappings. */
2264         while (map->olm_name != NULL) {
2265                 struct dentry *child;
2266
2267                 if (fid_is_zero(&map->olm_fid)) {
2268                         map++;
2269                         continue;
2270                 }
2271
2272                 child = osd_ios_lookup_one_len(map->olm_name,
2273                                                osd_sb(dev)->s_root,
2274                                                map->olm_namelen);
2275                 if (!IS_ERR(child))
2276                         dput(child);
2277                 else if (PTR_ERR(child) == -ENOENT)
2278                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2279                                                   NULL, DTO_INDEX_DELETE,
2280                                                   true, 0);
2281                 map++;
2282         }
2283
2284         RETURN(0);
2285 }
2286
2287 char *osd_lf_fid2name(const struct lu_fid *fid)
2288 {
2289         const struct osd_lf_map *map = osd_lf_maps;
2290
2291         while (map->olm_name != NULL) {
2292                 if (!lu_fid_eq(fid, &map->olm_fid)) {
2293                         map++;
2294                         continue;
2295                 }
2296
2297                 if (map->olm_flags & OLF_SHOW_NAME)
2298                         return map->olm_name;
2299                 else
2300                         return "";
2301         }
2302
2303         return NULL;
2304 }
2305
2306 /* OI scrub start/stop */
2307
2308 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2309 {
2310         struct osd_scrub     *scrub  = &dev->od_scrub;
2311         struct ptlrpc_thread *thread = &scrub->os_thread;
2312         struct l_wait_info    lwi    = { 0 };
2313         struct task_struct   *task;
2314         int                   rc;
2315         ENTRY;
2316
2317         /* os_lock: sync status between stop and scrub thread */
2318         spin_lock(&scrub->os_lock);
2319
2320 again:
2321         if (thread_is_running(thread)) {
2322                 spin_unlock(&scrub->os_lock);
2323                 if (!(scrub->os_file.sf_flags & SF_AUTO) ||
2324                      (flags & (SS_AUTO_FULL | SS_AUTO_PARTIAL)))
2325                         RETURN(-EALREADY);
2326
2327                 osd_scrub_join(dev, flags, false);
2328                 spin_lock(&scrub->os_lock);
2329                 if (!thread_is_running(thread))
2330                         goto again;
2331
2332                 spin_unlock(&scrub->os_lock);
2333                 RETURN(0);
2334         }
2335
2336         if (unlikely(thread_is_stopping(thread))) {
2337                 spin_unlock(&scrub->os_lock);
2338                 l_wait_event(thread->t_ctl_waitq,
2339                              thread_is_stopped(thread),
2340                              &lwi);
2341                 spin_lock(&scrub->os_lock);
2342                 goto again;
2343         }
2344         spin_unlock(&scrub->os_lock);
2345
2346         if (scrub->os_file.sf_status == SS_COMPLETED) {
2347                 if (!(flags & SS_SET_FAILOUT))
2348                         flags |= SS_CLEAR_FAILOUT;
2349
2350                 if (!(flags & SS_SET_DRYRUN))
2351                         flags |= SS_CLEAR_DRYRUN;
2352
2353                 flags |= SS_RESET;
2354         }
2355
2356         scrub->os_start_flags = flags;
2357         thread_set_flags(thread, 0);
2358         task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2359         if (IS_ERR(task)) {
2360                 rc = PTR_ERR(task);
2361                 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2362                        osd_scrub2name(scrub), rc);
2363                 RETURN(rc);
2364         }
2365
2366         l_wait_event(thread->t_ctl_waitq,
2367                      thread_is_running(thread) || thread_is_stopped(thread),
2368                      &lwi);
2369
2370         RETURN(0);
2371 }
2372
2373 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2374 {
2375         int rc;
2376         ENTRY;
2377
2378         /* od_otable_mutex: prevent curcurrent start/stop */
2379         mutex_lock(&dev->od_otable_mutex);
2380         rc = do_osd_scrub_start(dev, flags);
2381         mutex_unlock(&dev->od_otable_mutex);
2382
2383         RETURN(rc == -EALREADY ? 0 : rc);
2384 }
2385
2386 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2387 {
2388         struct ptlrpc_thread *thread = &scrub->os_thread;
2389         struct l_wait_info    lwi    = { 0 };
2390
2391         /* os_lock: sync status between stop and scrub thread */
2392         spin_lock(&scrub->os_lock);
2393         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2394                 thread_set_flags(thread, SVC_STOPPING);
2395                 spin_unlock(&scrub->os_lock);
2396                 wake_up_all(&thread->t_ctl_waitq);
2397                 l_wait_event(thread->t_ctl_waitq,
2398                              thread_is_stopped(thread),
2399                              &lwi);
2400                 /* Do not skip the last lock/unlock, which can guarantee that
2401                  * the caller cannot return until the OI scrub thread exit. */
2402                 spin_lock(&scrub->os_lock);
2403         }
2404         spin_unlock(&scrub->os_lock);
2405 }
2406
2407 static void osd_scrub_stop(struct osd_device *dev)
2408 {
2409         /* od_otable_mutex: prevent curcurrent start/stop */
2410         mutex_lock(&dev->od_otable_mutex);
2411         dev->od_scrub.os_paused = 1;
2412         do_osd_scrub_stop(&dev->od_scrub);
2413         mutex_unlock(&dev->od_otable_mutex);
2414 }
2415
2416 /* OI scrub setup/cleanup */
2417
2418 static const char osd_scrub_name[] = "OI_scrub";
2419
2420 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2421 {
2422         struct osd_thread_info     *info   = osd_oti_get(env);
2423         struct osd_scrub           *scrub  = &dev->od_scrub;
2424         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
2425         struct scrub_file          *sf     = &scrub->os_file;
2426         struct super_block         *sb     = osd_sb(dev);
2427         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
2428         struct lvfs_run_ctxt        saved;
2429         struct file                *filp;
2430         struct inode               *inode;
2431         struct lu_fid              *fid    = &info->oti_fid;
2432         int                         dirty  = 0;
2433         int                         rc     = 0;
2434         ENTRY;
2435
2436         memset(scrub, 0, sizeof(*scrub));
2437         OBD_SET_CTXT_MAGIC(ctxt);
2438         ctxt->pwdmnt = dev->od_mnt;
2439         ctxt->pwd = dev->od_mnt->mnt_root;
2440         ctxt->fs = get_ds();
2441
2442         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2443         init_rwsem(&scrub->os_rwsem);
2444         spin_lock_init(&scrub->os_lock);
2445         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2446
2447         push_ctxt(&saved, ctxt);
2448         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2449         if (IS_ERR(filp)) {
2450                 pop_ctxt(&saved, ctxt);
2451                 RETURN(PTR_ERR(filp));
2452         }
2453
2454         inode = filp->f_path.dentry->d_inode;
2455         /* 'What the @fid is' is not imporatant, because the object
2456          * has no OI mapping, and only is visible inside the OSD.*/
2457         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2458         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2459         if (rc != 0) {
2460                 filp_close(filp, NULL);
2461                 pop_ctxt(&saved, ctxt);
2462                 RETURN(rc);
2463         }
2464
2465         scrub->os_inode = igrab(inode);
2466         filp_close(filp, NULL);
2467         pop_ctxt(&saved, ctxt);
2468
2469         rc = osd_scrub_file_load(scrub);
2470         if (rc == -ENOENT) {
2471                 osd_scrub_file_init(scrub, es->s_uuid);
2472                 /* If the "/O" dir does not exist when mount (indicated by
2473                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2474                  * then it is quite probably that the device is a new one,
2475                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2476                  *
2477                  * For the rare case that "/O" and "OI_scrub" both lost on
2478                  * an old device, it can be found and cleared later.
2479                  *
2480                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2481                  * need to check "filter_fid_old" and to convert it to
2482                  * "filter_fid" for each object, and all the IGIF should
2483                  * have their FID mapping in OI files already. */
2484                 if (dev->od_maybe_new)
2485                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2486                 dirty = 1;
2487         } else if (rc != 0) {
2488                 GOTO(cleanup_inode, rc);
2489         } else {
2490                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2491                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2492                         dirty = 1;
2493                 } else if (sf->sf_status == SS_SCANNING) {
2494                         sf->sf_status = SS_CRASHED;
2495                         dirty = 1;
2496                 }
2497         }
2498
2499         if (sf->sf_pos_last_checkpoint != 0)
2500                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2501         else
2502                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2503
2504         if (dirty != 0) {
2505                 rc = osd_scrub_file_store(scrub);
2506                 if (rc != 0)
2507                         GOTO(cleanup_inode, rc);
2508         }
2509
2510         /* Initialize OI files. */
2511         rc = osd_oi_init(info, dev);
2512         if (rc < 0)
2513                 GOTO(cleanup_inode, rc);
2514
2515         rc = osd_initial_OI_scrub(info, dev);
2516         if (rc != 0)
2517                 GOTO(cleanup_oi, rc);
2518
2519         if (sf->sf_flags & SF_UPGRADE ||
2520             !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2521               sf->sf_success_count > 0)) {
2522                 dev->od_igif_inoi = 0;
2523                 dev->od_check_ff = dev->od_is_ost;
2524         } else {
2525                 dev->od_igif_inoi = 1;
2526                 dev->od_check_ff = 0;
2527         }
2528
2529         if (sf->sf_flags & SF_INCONSISTENT)
2530                 /* The 'od_igif_inoi' will be set under the
2531                  * following cases:
2532                  * 1) new created system, or
2533                  * 2) restored from file-level backup, or
2534                  * 3) the upgrading completed.
2535                  *
2536                  * The 'od_igif_inoi' may be cleared by OI scrub
2537                  * later if found that the system is upgrading. */
2538                 dev->od_igif_inoi = 1;
2539
2540         if (!dev->od_noscrub &&
2541             ((sf->sf_status == SS_PAUSED) ||
2542              (sf->sf_status == SS_CRASHED &&
2543               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2544                               SF_UPGRADE | SF_AUTO)) ||
2545              (sf->sf_status == SS_INIT &&
2546               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2547                               SF_UPGRADE))))
2548                 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2549
2550         if (rc != 0)
2551                 GOTO(cleanup_oi, rc);
2552
2553         /* it is possible that dcache entries may keep objects after they are
2554          * deleted by OSD. While it looks safe this can cause object data to
2555          * stay until umount causing failures in tests calculating free space,
2556          * e.g. replay-ost-single. Since those dcache entries are not used
2557          * anymore let's just free them after use here */
2558         shrink_dcache_sb(sb);
2559
2560         RETURN(0);
2561 cleanup_oi:
2562         osd_oi_fini(info, dev);
2563 cleanup_inode:
2564         iput(scrub->os_inode);
2565         scrub->os_inode = NULL;
2566
2567         return rc;
2568 }
2569
2570 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2571 {
2572         struct osd_scrub *scrub = &dev->od_scrub;
2573
2574         LASSERT(dev->od_otable_it == NULL);
2575
2576         if (scrub->os_inode != NULL) {
2577                 osd_scrub_stop(dev);
2578                 iput(scrub->os_inode);
2579                 scrub->os_inode = NULL;
2580         }
2581         if (dev->od_oi_table != NULL)
2582                 osd_oi_fini(osd_oti_get(env), dev);
2583 }
2584
2585 /* object table based iteration APIs */
2586
2587 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2588                                        struct dt_object *dt, __u32 attr)
2589 {
2590         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2591         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2592         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2593         struct osd_scrub       *scrub = &dev->od_scrub;
2594         struct osd_otable_it   *it;
2595         __u32                   start = 0;
2596         int                     rc;
2597         ENTRY;
2598
2599         /* od_otable_mutex: prevent curcurrent init/fini */
2600         mutex_lock(&dev->od_otable_mutex);
2601         if (dev->od_otable_it != NULL)
2602                 GOTO(out, it = ERR_PTR(-EALREADY));
2603
2604         OBD_ALLOC_PTR(it);
2605         if (it == NULL)
2606                 GOTO(out, it = ERR_PTR(-ENOMEM));
2607
2608         dev->od_otable_it = it;
2609         it->ooi_dev = dev;
2610         it->ooi_cache.ooc_consumer_idx = -1;
2611         if (flags & DOIF_OUTUSED)
2612                 it->ooi_used_outside = 1;
2613
2614         if (flags & DOIF_RESET)
2615                 start |= SS_RESET;
2616
2617         if (valid & DOIV_ERROR_HANDLE) {
2618                 if (flags & DOIF_FAILOUT)
2619                         start |= SS_SET_FAILOUT;
2620                 else
2621                         start |= SS_CLEAR_FAILOUT;
2622         }
2623
2624         if (valid & DOIV_DRYRUN) {
2625                 if (flags & DOIF_DRYRUN)
2626                         start |= SS_SET_DRYRUN;
2627                 else
2628                         start |= SS_CLEAR_DRYRUN;
2629         }
2630
2631         rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2632         if (rc < 0 && rc != -EALREADY) {
2633                 dev->od_otable_it = NULL;
2634                 OBD_FREE_PTR(it);
2635                 GOTO(out, it = ERR_PTR(rc));
2636         }
2637
2638         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2639
2640         GOTO(out, it);
2641
2642 out:
2643         mutex_unlock(&dev->od_otable_mutex);
2644         return (struct dt_it *)it;
2645 }
2646
2647 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2648 {
2649         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2650         struct osd_device    *dev = it->ooi_dev;
2651
2652         /* od_otable_mutex: prevent curcurrent init/fini */
2653         mutex_lock(&dev->od_otable_mutex);
2654         do_osd_scrub_stop(&dev->od_scrub);
2655         LASSERT(dev->od_otable_it == it);
2656
2657         dev->od_otable_it = NULL;
2658         mutex_unlock(&dev->od_otable_mutex);
2659         OBD_FREE_PTR(it);
2660 }
2661
2662 static int osd_otable_it_get(const struct lu_env *env,
2663                              struct dt_it *di, const struct dt_key *key)
2664 {
2665         return 0;
2666 }
2667
2668 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2669 {
2670 }
2671
2672 static inline int
2673 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2674 {
2675         spin_lock(&scrub->os_lock);
2676         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2677             scrub->os_waiting ||
2678             !thread_is_running(&scrub->os_thread))
2679                 it->ooi_waiting = 0;
2680         else
2681                 it->ooi_waiting = 1;
2682         spin_unlock(&scrub->os_lock);
2683
2684         return !it->ooi_waiting;
2685 }
2686
2687 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2688 {
2689         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2690         struct osd_device       *dev    = it->ooi_dev;
2691         struct osd_scrub        *scrub  = &dev->od_scrub;
2692         struct osd_otable_cache *ooc    = &it->ooi_cache;
2693         struct ptlrpc_thread    *thread = &scrub->os_thread;
2694         struct l_wait_info       lwi    = { 0 };
2695         int                      rc;
2696         ENTRY;
2697
2698         LASSERT(it->ooi_user_ready);
2699
2700 again:
2701         if (!thread_is_running(thread) && !it->ooi_used_outside)
2702                 RETURN(1);
2703
2704         if (ooc->ooc_cached_items > 0) {
2705                 ooc->ooc_cached_items--;
2706                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2707                                         ~OSD_OTABLE_IT_CACHE_MASK;
2708                 RETURN(0);
2709         }
2710
2711         if (it->ooi_all_cached) {
2712                 l_wait_event(thread->t_ctl_waitq,
2713                              !thread_is_running(thread),
2714                              &lwi);
2715                 RETURN(1);
2716         }
2717
2718         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2719                 spin_lock(&scrub->os_lock);
2720                 scrub->os_waiting = 0;
2721                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2722                 spin_unlock(&scrub->os_lock);
2723         }
2724
2725         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2726                 l_wait_event(thread->t_ctl_waitq,
2727                              osd_otable_it_wakeup(scrub, it),
2728                              &lwi);
2729
2730         if (!thread_is_running(thread) && !it->ooi_used_outside)
2731                 RETURN(1);
2732
2733         rc = osd_otable_it_preload(env, it);
2734         if (rc >= 0)
2735                 goto again;
2736
2737         RETURN(rc);
2738 }
2739
2740 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2741                                         const struct dt_it *di)
2742 {
2743         return NULL;
2744 }
2745
2746 static int osd_otable_it_key_size(const struct lu_env *env,
2747                                   const struct dt_it *di)
2748 {
2749         return sizeof(__u64);
2750 }
2751
2752 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2753                              struct dt_rec *rec, __u32 attr)
2754 {
2755         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2756         struct osd_otable_cache *ooc = &it->ooi_cache;
2757
2758         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2759
2760         /* Filter out Invald FID already. */
2761         LASSERTF(fid_is_sane((struct lu_fid *)rec),
2762                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2763                  PFID((struct lu_fid *)rec),
2764                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2765
2766         return 0;
2767 }
2768
2769 static __u64 osd_otable_it_store(const struct lu_env *env,
2770                                  const struct dt_it *di)
2771 {
2772         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2773         struct osd_otable_cache *ooc = &it->ooi_cache;
2774         __u64                    hash;
2775
2776         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2777                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2778         else
2779                 hash = ooc->ooc_pos_preload;
2780         return hash;
2781 }
2782
2783 /**
2784  * Set the OSD layer iteration start position as the specified hash.
2785  */
2786 static int osd_otable_it_load(const struct lu_env *env,
2787                               const struct dt_it *di, __u64 hash)
2788 {
2789         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
2790         struct osd_device       *dev   = it->ooi_dev;
2791         struct osd_otable_cache *ooc   = &it->ooi_cache;
2792         struct osd_scrub        *scrub = &dev->od_scrub;
2793         int                      rc;
2794         ENTRY;
2795
2796         /* Forbid to set iteration position after iteration started. */
2797         if (it->ooi_user_ready)
2798                 RETURN(-EPERM);
2799
2800         if (hash > OSD_OTABLE_MAX_HASH)
2801                 hash = OSD_OTABLE_MAX_HASH;
2802
2803         ooc->ooc_pos_preload = hash;
2804         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2805                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2806
2807         it->ooi_user_ready = 1;
2808         if (!scrub->os_full_speed)
2809                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2810
2811         /* Unplug OSD layer iteration by the first next() call. */
2812         rc = osd_otable_it_next(env, (struct dt_it *)it);
2813
2814         RETURN(rc);
2815 }
2816
2817 static int osd_otable_it_key_rec(const struct lu_env *env,
2818                                  const struct dt_it *di, void *key_rec)
2819 {
2820         return 0;
2821 }
2822
2823 const struct dt_index_operations osd_otable_ops = {
2824         .dio_it = {
2825                 .init     = osd_otable_it_init,
2826                 .fini     = osd_otable_it_fini,
2827                 .get      = osd_otable_it_get,
2828                 .put      = osd_otable_it_put,
2829                 .next     = osd_otable_it_next,
2830                 .key      = osd_otable_it_key,
2831                 .key_size = osd_otable_it_key_size,
2832                 .rec      = osd_otable_it_rec,
2833                 .store    = osd_otable_it_store,
2834                 .load     = osd_otable_it_load,
2835                 .key_rec  = osd_otable_it_key_rec,
2836         }
2837 };
2838
2839 /* high priority inconsistent items list APIs */
2840
2841 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL  60
2842
2843 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2844                    int insert)
2845 {
2846         struct osd_inconsistent_item *oii;
2847         struct osd_scrub             *scrub  = &dev->od_scrub;
2848         struct ptlrpc_thread         *thread = &scrub->os_thread;
2849         int                           wakeup = 0;
2850         ENTRY;
2851
2852         OBD_ALLOC_PTR(oii);
2853         if (unlikely(oii == NULL))
2854                 RETURN(-ENOMEM);
2855
2856         INIT_LIST_HEAD(&oii->oii_list);
2857         oii->oii_cache = *oic;
2858         oii->oii_insert = insert;
2859
2860         if (scrub->os_partial_scan) {
2861                 __u64 now = cfs_time_current_sec();
2862
2863                 /* If there haven't been errors in a long time,
2864                  * decay old count until either the errors are
2865                  * gone or we reach the current interval. */
2866                 while (unlikely(scrub->os_bad_oimap_count > 0 &&
2867                                 scrub->os_bad_oimap_time +
2868                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
2869                         scrub->os_bad_oimap_count >>= 1;
2870                         scrub->os_bad_oimap_time +=
2871                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
2872                 }
2873
2874                 scrub->os_bad_oimap_time = now;
2875                 if (++scrub->os_bad_oimap_count >
2876                     dev->od_full_scrub_threshold_rate)
2877                         scrub->os_full_scrub = 1;
2878         }
2879
2880         spin_lock(&scrub->os_lock);
2881         if (unlikely(!thread_is_running(thread))) {
2882                 spin_unlock(&scrub->os_lock);
2883                 OBD_FREE_PTR(oii);
2884                 RETURN(-EAGAIN);
2885         }
2886
2887         if (list_empty(&scrub->os_inconsistent_items))
2888                 wakeup = 1;
2889         list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2890         spin_unlock(&scrub->os_lock);
2891
2892         if (wakeup != 0)
2893                 wake_up_all(&thread->t_ctl_waitq);
2894
2895         RETURN(0);
2896 }
2897
2898 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2899                    struct osd_inode_id *id)
2900 {
2901         struct osd_scrub             *scrub = &dev->od_scrub;
2902         struct osd_inconsistent_item *oii;
2903         ENTRY;
2904
2905         spin_lock(&scrub->os_lock);
2906         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2907                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2908                         *id = oii->oii_cache.oic_lid;
2909                         spin_unlock(&scrub->os_lock);
2910                         RETURN(0);
2911                 }
2912         }
2913         spin_unlock(&scrub->os_lock);
2914
2915         RETURN(-ENOENT);
2916 }
2917
2918 /* OI scrub dump */
2919
2920 static const char *scrub_status_names[] = {
2921         "init",
2922         "scanning",
2923         "completed",
2924         "failed",
2925         "stopped",
2926         "paused",
2927         "crashed",
2928         NULL
2929 };
2930
2931 static const char *scrub_flags_names[] = {
2932         "recreated",
2933         "inconsistent",
2934         "auto",
2935         "upgrade",
2936         NULL
2937 };
2938
2939 static const char *scrub_param_names[] = {
2940         "failout",
2941         "dryrun",
2942         NULL
2943 };
2944
2945 static int scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
2946                            const char *prefix)
2947 {
2948         int flag;
2949         int rc;
2950         int i;
2951
2952         rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2953         if (rc < 0)
2954                 return rc;
2955
2956         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2957                 if (flag & bits) {
2958                         bits &= ~flag;
2959                         rc = seq_printf(m, "%s%c", names[i],
2960                                         bits != 0 ? ',' : '\n');
2961                         if (rc < 0)
2962                                 return rc;
2963                 }
2964         }
2965         return 0;
2966 }
2967
2968 static int scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
2969 {
2970         int rc;
2971
2972         if (time != 0)
2973                 rc = seq_printf(m, "%s: "LPU64" seconds\n", prefix,
2974                               cfs_time_current_sec() - time);
2975         else
2976                 rc = seq_printf(m, "%s: N/A\n", prefix);
2977         return rc;
2978 }
2979
2980 static int scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
2981 {
2982         int rc;
2983
2984         if (pos != 0)
2985                 rc = seq_printf(m, "%s: "LPU64"\n", prefix, pos);
2986         else
2987                 rc = seq_printf(m, "%s: N/A\n", prefix);
2988         return rc;
2989 }
2990
2991 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
2992 {
2993         struct osd_scrub  *scrub   = &dev->od_scrub;
2994         struct scrub_file *sf      = &scrub->os_file;
2995         __u64              checked;
2996         __u64              speed;
2997         int                rc;
2998
2999         down_read(&scrub->os_rwsem);
3000         rc = seq_printf(m, "name: OI_scrub\n"
3001                         "magic: 0x%x\n"
3002                         "oi_files: %d\n"
3003                         "status: %s\n",
3004                         sf->sf_magic, (int)sf->sf_oi_count,
3005                         scrub_status_names[sf->sf_status]);
3006         if (rc < 0)
3007                 goto out;
3008
3009         rc = scrub_bits_dump(m, sf->sf_flags, scrub_flags_names,
3010                              "flags");
3011         if (rc < 0)
3012                 goto out;
3013
3014         rc = scrub_bits_dump(m, sf->sf_param, scrub_param_names,
3015                              "param");
3016         if (rc < 0)
3017                 goto out;
3018
3019         rc = scrub_time_dump(m, sf->sf_time_last_complete,
3020                              "time_since_last_completed");
3021         if (rc < 0)
3022                 goto out;
3023
3024         rc = scrub_time_dump(m, sf->sf_time_latest_start,
3025                              "time_since_latest_start");
3026         if (rc < 0)
3027                 goto out;
3028
3029         rc = scrub_time_dump(m, sf->sf_time_last_checkpoint,
3030                              "time_since_last_checkpoint");
3031         if (rc < 0)
3032                 goto out;
3033
3034         rc = scrub_pos_dump(m, sf->sf_pos_latest_start,
3035                             "latest_start_position");
3036         if (rc < 0)
3037                 goto out;
3038
3039         rc = scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
3040                             "last_checkpoint_position");
3041         if (rc < 0)
3042                 goto out;
3043
3044         rc = scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
3045                             "first_failure_position");
3046         if (rc < 0)
3047                 goto out;
3048
3049         checked = sf->sf_items_checked + scrub->os_new_checked;
3050         rc = seq_printf(m, "checked: "LPU64"\n"
3051                       "updated: "LPU64"\n"
3052                       "failed: "LPU64"\n"
3053                       "prior_updated: "LPU64"\n"
3054                       "noscrub: "LPU64"\n"
3055                       "igif: "LPU64"\n"
3056                       "success_count: %u\n",
3057                       checked, sf->sf_items_updated, sf->sf_items_failed,
3058                       sf->sf_items_updated_prior, sf->sf_items_noscrub,
3059                       sf->sf_items_igif, sf->sf_success_count);
3060         if (rc < 0)
3061                 goto out;
3062
3063         speed = checked;
3064         if (thread_is_running(&scrub->os_thread)) {
3065                 cfs_duration_t duration = cfs_time_current() -
3066                                           scrub->os_time_last_checkpoint;
3067                 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
3068                                                      MSEC_PER_SEC);
3069                 __u32 rtime = sf->sf_run_time +
3070                               cfs_duration_sec(duration + HALF_SEC);
3071
3072                 if (duration != 0)
3073                         do_div(new_checked, duration);
3074                 if (rtime != 0)
3075                         do_div(speed, rtime);
3076                 rc = seq_printf(m, "run_time: %u seconds\n"
3077                               "average_speed: "LPU64" objects/sec\n"
3078                               "real-time_speed: "LPU64" objects/sec\n"
3079                               "current_position: %u\n"
3080                               "lf_scanned: "LPU64"\n"
3081                               "lf_repaired: "LPU64"\n"
3082                               "lf_failed: "LPU64"\n",
3083                               rtime, speed, new_checked, scrub->os_pos_current,
3084                               scrub->os_lf_scanned, scrub->os_lf_repaired,
3085                               scrub->os_lf_failed);
3086         } else {
3087                 if (sf->sf_run_time != 0)
3088                         do_div(speed, sf->sf_run_time);
3089                 rc = seq_printf(m, "run_time: %u seconds\n"
3090                               "average_speed: "LPU64" objects/sec\n"
3091                               "real-time_speed: N/A\n"
3092                               "current_position: N/A\n"
3093                               "lf_scanned: "LPU64"\n"
3094                               "lf_repaired: "LPU64"\n"
3095                               "lf_failed: "LPU64"\n",
3096                               sf->sf_run_time, speed, scrub->os_lf_scanned,
3097                               scrub->os_lf_repaired, scrub->os_lf_failed);
3098         }
3099
3100 out:
3101         up_read(&scrub->os_rwsem);
3102         return (rc < 0 ? -ENOSPC : 0);
3103 }