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 void osd_scrub_post(struct osd_scrub *scrub, int result)
 744 {
 745         struct scrub_file *sf = &scrub->os_file;
 746         ENTRY;
 747
 748         CDEBUG(D_LFSCK, "%.16s: OI scrub post, result = %d\n",
 749                osd_scrub2name(scrub), result);
 750
 751         down_write(&scrub->os_rwsem);
 752         spin_lock(&scrub->os_lock);
 753         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 754         spin_unlock(&scrub->os_lock);
 755         if (scrub->os_new_checked > 0) {
 756                 sf->sf_items_checked += scrub->os_new_checked;
 757                 scrub->os_new_checked = 0;
 758                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 759         }
 760         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 761         if (result > 0) {
 762                 struct osd_device *dev =
 763                         container_of0(scrub, struct osd_device, od_scrub);
 764
 765                 dev->od_igif_inoi = 1;
 766                 dev->od_check_ff = 0;
 767                 sf->sf_status = SS_COMPLETED;
 768                 if (!(sf->sf_param & SP_DRYRUN)) {
 769                         memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 770                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 771                                           SF_UPGRADE | SF_AUTO);
 772                 }
 773                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 774                 sf->sf_success_count++;
 775         } else if (result == 0) {
 776                 if (scrub->os_paused)
 777                         sf->sf_status = SS_PAUSED;
 778                 else
 779                         sf->sf_status = SS_STOPPED;
 780         } else {
 781                 sf->sf_status = SS_FAILED;
 782         }
 783         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 784                                             scrub->os_time_last_checkpoint);
 785         result = osd_scrub_file_store(scrub);
 786         up_write(&scrub->os_rwsem);
 787
 788         EXIT;
 789 }
 790
 791 /* iteration engine */
 792
 793 struct osd_iit_param {
 794         struct super_block *sb;
 795         struct buffer_head *bitmap;
 796         ldiskfs_group_t bg;
 797         __u32 gbase;
 798         __u32 offset;
 799 };
 800
 801 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 802                                    struct osd_device *dev,
 803                                    struct osd_iit_param *param,
 804                                    struct osd_idmap_cache **oic,
 805                                    const bool noslot);
 806
 807 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 808                                    struct osd_device *dev,
 809                                    struct osd_iit_param *param,
 810                                    struct osd_idmap_cache *oic,
 811                                    bool *noslot, int rc);
 812
 813 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 814 {
 815         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 816                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 817         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 818                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 819                 return SCRUB_NEXT_BREAK;
 820         } else {
 821                 *pos = param->gbase + param->offset;
 822                 return 0;
 823         }
 824 }
 825
 826 /**
 827  * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
 828  * \retval 0: FID-on-MDT
 829  */
 830 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
 831                                       struct osd_device *dev,
 832                                       struct lu_fid *fid)
 833 {
 834         /* XXX: The initial OI scrub will scan the top level /O to generate
 835          *      a small local FLDB according to the <seq>. If the given FID
 836          *      is in the local FLDB, then it is FID-on-OST; otherwise it's
 837          *      quite possible for FID-on-MDT. */
 838         if (dev->od_is_ost)
 839                 return SCRUB_NEXT_OSTOBJ_OLD;
 840         else
 841                 return 0;
 842 }
 843
 844 static int osd_scrub_get_fid(struct osd_thread_info *info,
 845                              struct osd_device *dev, struct inode *inode,
 846                              struct lu_fid *fid, bool scrub)
 847 {
 848         struct lustre_mdt_attrs *lma     = &info->oti_mdt_attrs;
 849         int                      rc;
 850         bool                     has_lma = false;
 851
 852         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 853         if (rc == 0) {
 854                 has_lma = true;
 855                 if (lma->lma_compat & LMAC_NOT_IN_OI ||
 856                     lma->lma_incompat & LMAI_AGENT)
 857                         return SCRUB_NEXT_CONTINUE;
 858
 859                 *fid = lma->lma_self_fid;
 860                 if (!scrub)
 861                         return 0;
 862
 863                 if (lma->lma_compat & LMAC_FID_ON_OST)
 864                         return SCRUB_NEXT_OSTOBJ;
 865
 866                 if (fid_is_idif(fid))
 867                         return SCRUB_NEXT_OSTOBJ_OLD;
 868
 869                 /* For local object. */
 870                 if (fid_is_internal(fid))
 871                         return 0;
 872
 873                 /* For external visible MDT-object with non-normal FID. */
 874                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
 875                         return 0;
 876
 877                 /* For the object with normal FID, it may be MDT-object,
 878                  * or may be 2.4 OST-object, need further distinguish.
 879                  * Fall through to next section. */
 880         }
 881
 882         if (rc == -ENODATA || rc == 0) {
 883                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
 884                 if (rc == 0) {
 885                         if (scrub)
 886                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
 887                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 888                         return rc;
 889                 }
 890
 891                 if (rc > 0) {
 892                         if (!has_lma)
 893                                 /* It is FID-on-OST, but we do not know how
 894                                  * to generate its FID, ignore it directly. */
 895                                 rc = SCRUB_NEXT_CONTINUE;
 896                         else
 897                                 /* It is 2.4 OST-object. */
 898                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 899                         return rc;
 900                 }
 901
 902                 if (rc != -ENODATA)
 903                         return rc;
 904
 905                 if (!has_lma) {
 906                         if (dev->od_scrub.os_convert_igif) {
 907                                 lu_igif_build(fid, inode->i_ino,
 908                                               inode->i_generation);
 909                                 if (scrub)
 910                                         rc = SCRUB_NEXT_NOLMA;
 911                                 else
 912                                         rc = 0;
 913                         } else {
 914                                 /* It may be FID-on-OST, or may be FID for
 915                                  * non-MDT0, anyway, we do not know how to
 916                                  * generate its FID, ignore it directly. */
 917                                 rc = SCRUB_NEXT_CONTINUE;
 918                         }
 919                         return rc;
 920                 }
 921
 922                 /* For OI scrub case only: the object has LMA but has no ff
 923                  * (or ff crashed). It may be MDT-object, may be OST-object
 924                  * with crashed ff. The last check is local FLDB. */
 925                 rc = osd_scrub_check_local_fldb(info, dev, fid);
 926         }
 927
 928         return rc;
 929 }
 930
 931 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 932                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 933                         struct super_block *sb, bool scrub)
 934 {
 935         struct inode *inode;
 936         int           rc;
 937         ENTRY;
 938
 939         /* Not handle the backend root object and agent parent object.
 940          * They are neither visible to namespace nor have OI mappings. */
 941         if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
 942                      pos == osd_remote_parent_ino(dev)))
 943                 RETURN(SCRUB_NEXT_CONTINUE);
 944
 945         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 946         inode = osd_iget(info, dev, lid);
 947         if (IS_ERR(inode)) {
 948                 rc = PTR_ERR(inode);
 949                 /* The inode may be removed after bitmap searching, or the
 950                  * file is new created without inode initialized yet. */
 951                 if (rc == -ENOENT || rc == -ESTALE)
 952                         RETURN(SCRUB_NEXT_CONTINUE);
 953
 954                 CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: "
 955                        "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name,
 956                        pos, rc);
 957                 RETURN(rc);
 958         }
 959
 960         if (scrub &&
 961             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 962                 /* Only skip it for the first OI scrub accessing. */
 963                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 964                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
 965         }
 966
 967         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
 968
 969         GOTO(put, rc);
 970
 971 put:
 972         iput(inode);
 973         return rc;
 974 }
 975
 976 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 977                           struct osd_iit_param *param,
 978                           struct osd_idmap_cache **oic, const bool noslot)
 979 {
 980         struct osd_scrub     *scrub  = &dev->od_scrub;
 981         struct ptlrpc_thread *thread = &scrub->os_thread;
 982         struct lu_fid        *fid;
 983         struct osd_inode_id  *lid;
 984         int                   rc;
 985
 986         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
 987                 struct l_wait_info lwi;
 988
 989                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
 990                 if (likely(lwi.lwi_timeout > 0))
 991                         l_wait_event(thread->t_ctl_waitq,
 992                                 !list_empty(&scrub->os_inconsistent_items) ||
 993                                 !thread_is_running(thread),
 994                                 &lwi);
 995         }
 996
 997         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 998                 spin_lock(&scrub->os_lock);
 999                 thread_set_flags(thread, SVC_STOPPING);
1000                 spin_unlock(&scrub->os_lock);
1001                 return SCRUB_NEXT_CRASH;
1002         }
1003
1004         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
1005                 return SCRUB_NEXT_FATAL;
1006
1007         if (unlikely(!thread_is_running(thread)))
1008                 return SCRUB_NEXT_EXIT;
1009
1010         if (!list_empty(&scrub->os_inconsistent_items)) {
1011                 spin_lock(&scrub->os_lock);
1012                 if (likely(!list_empty(&scrub->os_inconsistent_items))) {
1013                         struct osd_inconsistent_item *oii;
1014
1015                         oii = list_entry(scrub->os_inconsistent_items.next,
1016                                 struct osd_inconsistent_item, oii_list);
1017                         list_del_init(&oii->oii_list);
1018                         spin_unlock(&scrub->os_lock);
1019
1020                         *oic = &oii->oii_cache;
1021                         scrub->os_in_prior = 1;
1022
1023                         return 0;
1024                 }
1025                 spin_unlock(&scrub->os_lock);
1026         }
1027
1028         if (noslot)
1029                 return SCRUB_NEXT_WAIT;
1030
1031         rc = osd_iit_next(param, &scrub->os_pos_current);
1032         if (rc != 0)
1033                 return rc;
1034
1035         *oic = &scrub->os_oic;
1036         fid = &(*oic)->oic_fid;
1037         lid = &(*oic)->oic_lid;
1038         rc = osd_iit_iget(info, dev, fid, lid,
1039                           scrub->os_pos_current, param->sb, true);
1040         return rc;
1041 }
1042
1043 static int osd_preload_next(struct osd_thread_info *info,
1044                             struct osd_device *dev, struct osd_iit_param *param,
1045                             struct osd_idmap_cache **oic, const bool noslot)
1046 {
1047         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
1048         struct osd_scrub        *scrub;
1049         struct ptlrpc_thread    *thread;
1050         int                      rc;
1051
1052         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1053         if (rc != 0)
1054                 return rc;
1055
1056         scrub = &dev->od_scrub;
1057         thread = &scrub->os_thread;
1058         if (thread_is_running(thread) &&
1059             ooc->ooc_pos_preload >= scrub->os_pos_current)
1060                 return SCRUB_NEXT_EXIT;
1061
1062         rc = osd_iit_iget(info, dev,
1063                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1064                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1065                           ooc->ooc_pos_preload, param->sb, false);
1066         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1067          * ignore the failure, so it still need to skip the inode next time. */
1068         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1069         return rc;
1070 }
1071
1072 static inline int
1073 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1074 {
1075         spin_lock(&scrub->os_lock);
1076         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1077             !list_empty(&scrub->os_inconsistent_items) ||
1078             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1079                 scrub->os_waiting = 0;
1080         else
1081                 scrub->os_waiting = 1;
1082         spin_unlock(&scrub->os_lock);
1083
1084         return !scrub->os_waiting;
1085 }
1086
1087 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1088                           struct osd_iit_param *param,
1089                           struct osd_idmap_cache *oic, bool *noslot, int rc)
1090 {
1091         struct l_wait_info       lwi    = { 0 };
1092         struct osd_scrub        *scrub  = &dev->od_scrub;
1093         struct scrub_file       *sf     = &scrub->os_file;
1094         struct ptlrpc_thread    *thread = &scrub->os_thread;
1095         struct osd_otable_it    *it     = dev->od_otable_it;
1096         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
1097
1098         switch (rc) {
1099         case SCRUB_NEXT_CONTINUE:
1100                 goto next;
1101         case SCRUB_NEXT_WAIT:
1102                 goto wait;
1103         case SCRUB_NEXT_NOSCRUB:
1104                 down_write(&scrub->os_rwsem);
1105                 scrub->os_new_checked++;
1106                 sf->sf_items_noscrub++;
1107                 up_write(&scrub->os_rwsem);
1108                 goto next;
1109         }
1110
1111         rc = osd_scrub_check_update(info, dev, oic, rc);
1112         if (rc != 0) {
1113                 scrub->os_in_prior = 0;
1114                 return rc;
1115         }
1116
1117         rc = osd_scrub_checkpoint(scrub);
1118         if (rc != 0) {
1119                 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1120                        "rc = %d\n", osd_scrub2name(scrub),
1121                        scrub->os_pos_current, rc);
1122                 /* Continue, as long as the scrub itself can go ahead. */
1123         }
1124
1125         if (scrub->os_in_prior) {
1126                 scrub->os_in_prior = 0;
1127                 return 0;
1128         }
1129
1130 next:
1131         scrub->os_pos_current = param->gbase + ++(param->offset);
1132
1133 wait:
1134         if (it != NULL && it->ooi_waiting && ooc != NULL &&
1135             ooc->ooc_pos_preload < scrub->os_pos_current) {
1136                 spin_lock(&scrub->os_lock);
1137                 it->ooi_waiting = 0;
1138                 wake_up_all(&thread->t_ctl_waitq);
1139                 spin_unlock(&scrub->os_lock);
1140         }
1141
1142         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1143                 return 0;
1144
1145         if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1146                 *noslot = false;
1147                 return 0;
1148         }
1149
1150         if (it != NULL)
1151                 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1152                              &lwi);
1153
1154         if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1155                 *noslot = false;
1156         else
1157                 *noslot = true;
1158         return 0;
1159 }
1160
1161 static int osd_preload_exec(struct osd_thread_info *info,
1162                             struct osd_device *dev, struct osd_iit_param *param,
1163                             struct osd_idmap_cache *oic, bool *noslot, int rc)
1164 {
1165         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1166
1167         if (rc == 0) {
1168                 ooc->ooc_cached_items++;
1169                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1170                                         ~OSD_OTABLE_IT_CACHE_MASK;
1171         }
1172         return rc > 0 ? 0 : rc;
1173 }
1174
1175 #define SCRUB_IT_ALL    1
1176 #define SCRUB_IT_CRASH  2
1177
1178 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1179                            bool inconsistent)
1180 {
1181         struct osd_scrub     *scrub  = &dev->od_scrub;
1182         struct ptlrpc_thread *thread = &scrub->os_thread;
1183         struct scrub_file    *sf     = &scrub->os_file;
1184         int                   rc;
1185         ENTRY;
1186
1187         LASSERT(!(flags & SS_AUTO_PARTIAL));
1188
1189         down_write(&scrub->os_rwsem);
1190         scrub->os_in_join = 1;
1191         if (flags & SS_SET_FAILOUT)
1192                 sf->sf_param |= SP_FAILOUT;
1193         else if (flags & SS_CLEAR_FAILOUT)
1194                 sf->sf_param &= ~SP_FAILOUT;
1195
1196         if (flags & SS_SET_DRYRUN)
1197                 sf->sf_param |= SP_DRYRUN;
1198         else if (flags & SS_CLEAR_DRYRUN)
1199                 sf->sf_param &= ~SP_DRYRUN;
1200
1201         if (flags & SS_RESET) {
1202                 osd_scrub_file_reset(scrub,
1203                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1204                         inconsistent ? SF_INCONSISTENT : 0);
1205                 sf->sf_status = SS_SCANNING;
1206         }
1207
1208         if (flags & SS_AUTO_FULL) {
1209                 sf->sf_flags |= SF_AUTO;
1210                 scrub->os_full_speed = 1;
1211         }
1212
1213         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1214                 scrub->os_full_speed = 1;
1215         else
1216                 scrub->os_full_speed = 0;
1217
1218         scrub->os_new_checked = 0;
1219         if (sf->sf_pos_last_checkpoint != 0)
1220                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1221         else
1222                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1223
1224         scrub->os_pos_current = sf->sf_pos_latest_start;
1225         sf->sf_time_latest_start = cfs_time_current_sec();
1226         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1227         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1228         rc = osd_scrub_file_store(scrub);
1229         if (rc != 0)
1230                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
1231                        "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1232
1233         spin_lock(&scrub->os_lock);
1234         scrub->os_waiting = 0;
1235         scrub->os_paused = 0;
1236         scrub->os_partial_scan = 0;
1237         scrub->os_in_join = 0;
1238         scrub->os_full_scrub = 0;
1239         spin_unlock(&scrub->os_lock);
1240         wake_up_all(&thread->t_ctl_waitq);
1241         up_write(&scrub->os_rwsem);
1242
1243         EXIT;
1244 }
1245
1246 static int osd_inode_iteration(struct osd_thread_info *info,
1247                                struct osd_device *dev, __u32 max, bool preload)
1248 {
1249         struct osd_scrub     *scrub  = &dev->od_scrub;
1250         struct ptlrpc_thread *thread = &scrub->os_thread;
1251         struct scrub_file    *sf     = &scrub->os_file;
1252         osd_iit_next_policy   next;
1253         osd_iit_exec_policy   exec;
1254         __u32                *pos;
1255         __u32                *count;
1256         struct osd_iit_param  param  = { NULL };
1257         struct l_wait_info    lwi    = { 0 };
1258         __u32                 limit;
1259         int                   rc;
1260         bool                  noslot = true;
1261         ENTRY;
1262
1263         param.sb = osd_sb(dev);
1264         if (preload)
1265                 goto full;
1266
1267         while (scrub->os_partial_scan && !scrub->os_in_join) {
1268                 struct osd_idmap_cache *oic = NULL;
1269
1270                 rc = osd_scrub_next(info, dev, &param, &oic, noslot);
1271                 switch (rc) {
1272                 case SCRUB_NEXT_EXIT:
1273                         RETURN(0);
1274                 case SCRUB_NEXT_CRASH:
1275                         RETURN(SCRUB_IT_CRASH);
1276                 case SCRUB_NEXT_FATAL:
1277                         RETURN(-EINVAL);
1278                 case SCRUB_NEXT_WAIT: {
1279                         struct kstatfs *ksfs = &info->oti_ksfs;
1280                         __u64 saved_flags;
1281
1282                         if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1283                             unlikely(sf->sf_items_updated_prior == 0))
1284                                 goto wait;
1285
1286                         if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1287                             scrub->os_full_scrub) {
1288                                 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1289                                                true);
1290                                 goto full;
1291                         }
1292
1293                         rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1294                         if (rc == 0) {
1295                                 __u64 used = ksfs->f_files - ksfs->f_ffree;
1296
1297                                 do_div(used, sf->sf_items_updated_prior);
1298                                 /* If we hit too much inconsistent OI
1299                                  * mappings during the partial scan,
1300                                  * then scan the device completely. */
1301                                 if (used < dev->od_full_scrub_ratio) {
1302                                         osd_scrub_join(dev,
1303                                                 SS_AUTO_FULL | SS_RESET, true);
1304                                         goto full;
1305                                 }
1306                         }
1307
1308 wait:
1309                         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1310                             cfs_fail_val > 0)
1311                                 continue;
1312
1313                         saved_flags = sf->sf_flags;
1314                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1315                                           SF_UPGRADE | SF_AUTO);
1316                         sf->sf_status = SS_COMPLETED;
1317                         l_wait_event(thread->t_ctl_waitq,
1318                                      !thread_is_running(thread) ||
1319                                      !scrub->os_partial_scan ||
1320                                      scrub->os_in_join ||
1321                                      !list_empty(&scrub->os_inconsistent_items),
1322                                      &lwi);
1323                         sf->sf_flags = saved_flags;
1324                         sf->sf_status = SS_SCANNING;
1325
1326                         if (unlikely(!thread_is_running(thread)))
1327                                 RETURN(0);
1328
1329                         if (!scrub->os_partial_scan || scrub->os_in_join)
1330                                 goto full;
1331
1332                         continue;
1333                 }
1334                 default:
1335                         LASSERTF(rc == 0, "rc = %d\n", rc);
1336
1337                         osd_scrub_exec(info, dev, &param, oic, &noslot, rc);
1338                         break;
1339                 }
1340         }
1341
1342 full:
1343         if (!preload) {
1344                 l_wait_event(thread->t_ctl_waitq,
1345                              !thread_is_running(thread) || !scrub->os_in_join,
1346                              &lwi);
1347
1348                 if (unlikely(!thread_is_running(thread)))
1349                         RETURN(0);
1350         }
1351
1352         noslot = false;
1353         if (!preload) {
1354                 next = osd_scrub_next;
1355                 exec = osd_scrub_exec;
1356                 pos = &scrub->os_pos_current;
1357                 count = &scrub->os_new_checked;
1358         } else {
1359                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1360
1361                 next = osd_preload_next;
1362                 exec = osd_preload_exec;
1363                 pos = &ooc->ooc_pos_preload;
1364                 count = &ooc->ooc_cached_items;
1365         }
1366         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1367
1368         while (*pos <= limit && *count < max) {
1369                 struct osd_idmap_cache *oic = NULL;
1370                 struct ldiskfs_group_desc *desc;
1371
1372                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1373                 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1374                 if (desc == NULL)
1375                         RETURN(-EIO);
1376
1377                 ldiskfs_lock_group(param.sb, param.bg);
1378                 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1379                         ldiskfs_unlock_group(param.sb, param.bg);
1380                         *pos = 1 + (param.bg + 1) *
1381                                 LDISKFS_INODES_PER_GROUP(param.sb);
1382                         continue;
1383                 }
1384                 ldiskfs_unlock_group(param.sb, param.bg);
1385
1386                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1387                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1388                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1389                 if (param.bitmap == NULL) {
1390                         CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1391                                "scrub will stop, urgent mode\n",
1392                                osd_scrub2name(scrub), (__u32)param.bg);
1393                         RETURN(-EIO);
1394                 }
1395
1396                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1397                        *count < max) {
1398                         if (param.offset +
1399                                 ldiskfs_itable_unused_count(param.sb, desc) >
1400                             LDISKFS_INODES_PER_GROUP(param.sb))
1401                                 goto next_group;
1402
1403                         rc = next(info, dev, &param, &oic, noslot);
1404                         switch (rc) {
1405                         case SCRUB_NEXT_BREAK:
1406                                 goto next_group;
1407                         case SCRUB_NEXT_EXIT:
1408                                 brelse(param.bitmap);
1409                                 RETURN(0);
1410                         case SCRUB_NEXT_CRASH:
1411                                 brelse(param.bitmap);
1412                                 RETURN(SCRUB_IT_CRASH);
1413                         case SCRUB_NEXT_FATAL:
1414                                 brelse(param.bitmap);
1415                                 RETURN(-EINVAL);
1416                         }
1417
1418                         rc = exec(info, dev, &param, oic, &noslot, rc);
1419                         if (rc != 0) {
1420                                 brelse(param.bitmap);
1421                                 RETURN(rc);
1422                         }
1423                 }
1424
1425 next_group:
1426                 brelse(param.bitmap);
1427         }
1428
1429         if (*pos > limit)
1430                 RETURN(SCRUB_IT_ALL);
1431         RETURN(0);
1432 }
1433
1434 static int osd_otable_it_preload(const struct lu_env *env,
1435                                  struct osd_otable_it *it)
1436 {
1437         struct osd_device       *dev   = it->ooi_dev;
1438         struct osd_scrub        *scrub = &dev->od_scrub;
1439         struct osd_otable_cache *ooc   = &it->ooi_cache;
1440         int                      rc;
1441         ENTRY;
1442
1443         rc = osd_inode_iteration(osd_oti_get(env), dev,
1444                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1445         if (rc == SCRUB_IT_ALL)
1446                 it->ooi_all_cached = 1;
1447
1448         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1449                 spin_lock(&scrub->os_lock);
1450                 scrub->os_waiting = 0;
1451                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1452                 spin_unlock(&scrub->os_lock);
1453         }
1454
1455         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1456 }
1457
1458 static int osd_scrub_main(void *args)
1459 {
1460         struct lu_env         env;
1461         struct osd_device    *dev    = (struct osd_device *)args;
1462         struct osd_scrub     *scrub  = &dev->od_scrub;
1463         struct ptlrpc_thread *thread = &scrub->os_thread;
1464         int                   rc;
1465         ENTRY;
1466
1467         rc = lu_env_init(&env, LCT_LOCAL);
1468         if (rc != 0) {
1469                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1470                        osd_scrub2name(scrub), rc);
1471                 GOTO(noenv, rc);
1472         }
1473
1474         rc = osd_scrub_prep(dev);
1475         if (rc != 0) {
1476                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1477                        osd_scrub2name(scrub), rc);
1478                 GOTO(out, rc);
1479         }
1480
1481         if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1482                 struct l_wait_info lwi = { 0 };
1483                 struct osd_otable_it *it = dev->od_otable_it;
1484                 struct osd_otable_cache *ooc = &it->ooi_cache;
1485
1486                 l_wait_event(thread->t_ctl_waitq,
1487                              it->ooi_user_ready || !thread_is_running(thread),
1488                              &lwi);
1489                 if (unlikely(!thread_is_running(thread)))
1490                         GOTO(post, rc = 0);
1491
1492                 scrub->os_pos_current = ooc->ooc_pos_preload;
1493         }
1494
1495         CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1496                osd_scrub2name(scrub), scrub->os_start_flags,
1497                scrub->os_pos_current);
1498
1499         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1500         if (unlikely(rc == SCRUB_IT_CRASH))
1501                 GOTO(out, rc = -EINVAL);
1502         GOTO(post, rc);
1503
1504 post:
1505         osd_scrub_post(scrub, rc);
1506         CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1507                osd_scrub2name(scrub), scrub->os_pos_current, rc);
1508
1509 out:
1510         while (!list_empty(&scrub->os_inconsistent_items)) {
1511                 struct osd_inconsistent_item *oii;
1512
1513                 oii = list_entry(scrub->os_inconsistent_items.next,
1514                                      struct osd_inconsistent_item, oii_list);
1515                 list_del_init(&oii->oii_list);
1516                 OBD_FREE_PTR(oii);
1517         }
1518         lu_env_fini(&env);
1519
1520 noenv:
1521         spin_lock(&scrub->os_lock);
1522         thread_set_flags(thread, SVC_STOPPED);
1523         wake_up_all(&thread->t_ctl_waitq);
1524         spin_unlock(&scrub->os_lock);
1525         return rc;
1526 }
1527
1528 /* initial OI scrub */
1529
1530 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1531                          struct dentry *, filldir_t filldir);
1532
1533 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1534                                loff_t offset, __u64 ino, unsigned d_type);
1535 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1536                            loff_t offset, __u64 ino, unsigned d_type);
1537 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1538                            loff_t offset, __u64 ino, unsigned d_type);
1539 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1540                             loff_t offset, __u64 ino, unsigned d_type);
1541
1542 static int
1543 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1544                      struct dentry *dentry, filldir_t filldir);
1545 static int
1546 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1547                   struct dentry *dentry, filldir_t filldir);
1548
1549 static int
1550 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1551                      struct dentry *dentry, filldir_t filldir);
1552
1553 enum osd_lf_flags {
1554         OLF_SCAN_SUBITEMS       = 0x0001,
1555         OLF_HIDE_FID            = 0x0002,
1556         OLF_SHOW_NAME           = 0x0004,
1557         OLF_NO_OI               = 0x0008,
1558         OLF_IDX_IN_FID          = 0x0010,
1559 };
1560
1561 struct osd_lf_map {
1562         char            *olm_name;
1563         struct lu_fid    olm_fid;
1564         __u16            olm_flags;
1565         __u16            olm_namelen;
1566         scandir_t        olm_scandir;
1567         filldir_t        olm_filldir;
1568 };
1569
1570 /* Add the new introduced local files in the list in the future. */
1571 static const struct osd_lf_map osd_lf_maps[] = {
1572         /* CATALOGS */
1573         { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1574                 sizeof(CATLIST) - 1, NULL, NULL },
1575
1576         /* CONFIGS */
1577         { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1578                 OLF_SCAN_SUBITEMS, sizeof(MOUNT_CONFIGS_DIR) - 1,
1579                 osd_ios_general_scan, osd_ios_varfid_fill },
1580
1581         /* NIDTBL_VERSIONS */
1582         { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1583                 sizeof(MGS_NIDTBL_DIR) - 1, osd_ios_general_scan,
1584                 osd_ios_varfid_fill },
1585
1586         /* PENDING */
1587         { "PENDING", { 0, 0, 0 }, 0, sizeof("PENDING") - 1, NULL, NULL },
1588
1589         /* ROOT */
1590         { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 },
1591                 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, sizeof("ROOT") - 1,
1592                 osd_ios_ROOT_scan, NULL },
1593
1594         /* changelog_catalog */
1595         { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, sizeof(CHANGELOG_CATALOG) - 1,
1596                 NULL, NULL },
1597
1598         /* changelog_users */
1599         { CHANGELOG_USERS, { 0, 0, 0 }, 0, sizeof(CHANGELOG_USERS) - 1,
1600                 NULL, NULL },
1601
1602         /* fld */
1603         { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1604                 sizeof("fld") - 1, NULL, NULL },
1605
1606         /* last_rcvd */
1607         { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1608                 sizeof(LAST_RCVD) - 1, NULL, NULL },
1609
1610         /* reply_data */
1611         { REPLY_DATA, { FID_SEQ_LOCAL_FILE, REPLY_DATA_OID, 0 }, OLF_SHOW_NAME,
1612                 sizeof(REPLY_DATA) - 1, NULL, NULL },
1613
1614         /* lov_objid */
1615         { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1616                 sizeof(LOV_OBJID) - 1, NULL, NULL },
1617
1618         /* lov_objseq */
1619         { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1620                 OLF_SHOW_NAME, sizeof(LOV_OBJSEQ) - 1, NULL, NULL },
1621
1622         /* quota_master */
1623         { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QMT_DIR) - 1,
1624                 osd_ios_general_scan, osd_ios_varfid_fill },
1625
1626         /* quota_slave */
1627         { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QSD_DIR) - 1,
1628                 osd_ios_general_scan, osd_ios_varfid_fill },
1629
1630         /* seq_ctl */
1631         { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1632                 OLF_SHOW_NAME, sizeof("seq_ctl") - 1, NULL, NULL },
1633
1634         /* seq_srv */
1635         { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1636                 OLF_SHOW_NAME, sizeof("seq_srv") - 1, NULL, NULL },
1637
1638         /* health_check */
1639         { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1640                 OLF_SHOW_NAME, sizeof(HEALTH_CHECK) - 1, NULL, NULL },
1641
1642         /* LFSCK */
1643         { LFSCK_DIR, { 0, 0, 0 }, 0, sizeof(LFSCK_DIR) - 1,
1644                 osd_ios_general_scan, osd_ios_varfid_fill },
1645
1646         /* lfsck_bookmark */
1647         { LFSCK_BOOKMARK, { 0, 0, 0 }, 0, sizeof(LFSCK_BOOKMARK) - 1,
1648                 NULL, NULL },
1649
1650         /* lfsck_layout */
1651         { LFSCK_LAYOUT, { 0, 0, 0 }, 0, sizeof(LFSCK_LAYOUT) - 1,
1652                 NULL, NULL },
1653
1654         /* lfsck_namespace */
1655         { LFSCK_NAMESPACE, { 0, 0, 0 }, 0, sizeof(LFSCK_NAMESPACE) - 1,
1656                 NULL, NULL },
1657
1658         /* OBJECTS, upgrade from old device */
1659         { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(OBJECTS) - 1,
1660                 osd_ios_OBJECTS_scan, NULL },
1661
1662         /* lquota_v2.user, upgrade from old device */
1663         { "lquota_v2.user", { 0, 0, 0 }, 0, sizeof("lquota_v2.user") - 1,
1664                 NULL, NULL },
1665
1666         /* lquota_v2.group, upgrade from old device */
1667         { "lquota_v2.group", { 0, 0, 0 }, 0, sizeof("lquota_v2.group") - 1,
1668                 NULL, NULL },
1669
1670         /* LAST_GROUP, upgrade from old device */
1671         { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1672                 OLF_SHOW_NAME, sizeof("LAST_GROUP") - 1, NULL, NULL },
1673
1674         /* committed batchid for cross-MDT operation */
1675         { "BATCHID", { FID_SEQ_LOCAL_FILE, BATCHID_COMMITTED_OID, 0 },
1676                 OLF_SHOW_NAME, sizeof("BATCHID") - 1, NULL, NULL },
1677
1678         /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1679          * and f_oid = index for their log files.  See lu_update_log{_dir}_fid()
1680          * for more details. */
1681
1682         /* update_log */
1683         { "update_log", { FID_SEQ_UPDATE_LOG, 0, 0 },
1684                 OLF_SHOW_NAME | OLF_IDX_IN_FID, sizeof("update_log") - 1,
1685                 NULL, NULL },
1686
1687         /* update_log_dir */
1688         { "update_log_dir", { FID_SEQ_UPDATE_LOG_DIR, 0, 0 },
1689                 OLF_SHOW_NAME | OLF_SCAN_SUBITEMS | OLF_IDX_IN_FID,
1690                 sizeof("update_log_dir") - 1,
1691                 osd_ios_general_scan, osd_ios_uld_fill },
1692
1693         /* lost+found */
1694         { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 },
1695                 OLF_SCAN_SUBITEMS, sizeof("lost+found") - 1,
1696                 osd_ios_general_scan, osd_ios_lf_fill },
1697
1698         { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1699 };
1700
1701 /* Add the new introduced files under .lustre/ in the list in the future. */
1702 static const struct osd_lf_map osd_dl_maps[] = {
1703         /* .lustre/fid */
1704         { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1705                 sizeof("fid") - 1, NULL, NULL },
1706
1707         /* .lustre/lost+found */
1708         { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1709                 sizeof("lost+found") - 1, NULL, NULL },
1710
1711         { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1712 };
1713
1714 struct osd_ios_item {
1715         struct list_head oii_list;
1716         struct dentry   *oii_dentry;
1717         scandir_t        oii_scandir;
1718         filldir_t        oii_filldir;
1719 };
1720
1721 struct osd_ios_filldir_buf {
1722 #ifdef HAVE_DIR_CONTEXT
1723         /* please keep it as first member */
1724         struct dir_context       ctx;
1725 #endif
1726         struct osd_thread_info  *oifb_info;
1727         struct osd_device       *oifb_dev;
1728         struct dentry           *oifb_dentry;
1729 };
1730
1731 static inline struct dentry *
1732 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1733 {
1734         struct dentry *dentry;
1735
1736         dentry = ll_lookup_one_len(name, parent, namelen);
1737         if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1738                 dput(dentry);
1739                 return ERR_PTR(-ENOENT);
1740         }
1741
1742         return dentry;
1743 }
1744
1745 static int
1746 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1747                  scandir_t scandir, filldir_t filldir)
1748 {
1749         struct osd_ios_item *item;
1750         ENTRY;
1751
1752         OBD_ALLOC_PTR(item);
1753         if (item == NULL)
1754                 RETURN(-ENOMEM);
1755
1756         INIT_LIST_HEAD(&item->oii_list);
1757         item->oii_dentry = dget(dentry);
1758         item->oii_scandir = scandir;
1759         item->oii_filldir = filldir;
1760         list_add_tail(&item->oii_list, &dev->od_ios_list);
1761
1762         RETURN(0);
1763 }
1764
1765 /**
1766  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1767  *
1768  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1769  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1770  * reference the inode, or fixed if it is missing or references another inode.
1771  */
1772 static int
1773 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1774                  struct inode *inode, const struct lu_fid *fid, int flags)
1775 {
1776         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
1777         struct osd_inode_id     *id     = &info->oti_id;
1778         struct osd_inode_id     *id2    = &info->oti_id2;
1779         struct osd_scrub        *scrub  = &dev->od_scrub;
1780         struct scrub_file       *sf     = &scrub->os_file;
1781         struct lu_fid            tfid;
1782         int                      rc;
1783         ENTRY;
1784
1785         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1786         if (rc != 0 && rc != -ENODATA) {
1787                 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1788                        "rc = %d\n", osd_name(dev), rc);
1789
1790                 RETURN(rc);
1791         }
1792
1793         osd_id_gen(id, inode->i_ino, inode->i_generation);
1794         if (rc == -ENODATA) {
1795                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
1796                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1797                 } else {
1798                         tfid = *fid;
1799                         if (flags & OLF_IDX_IN_FID) {
1800                                 LASSERT(dev->od_index >= 0);
1801
1802                                 tfid.f_oid = dev->od_index;
1803                         }
1804                 }
1805                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1806                 if (rc != 0) {
1807                         CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1808                               "scrub: rc = %d\n", osd_name(dev), rc);
1809
1810                         RETURN(rc);
1811                 }
1812         } else {
1813                 if (lma->lma_compat & LMAC_NOT_IN_OI)
1814                         RETURN(0);
1815
1816                 tfid = lma->lma_self_fid;
1817         }
1818
1819         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1820         if (rc != 0) {
1821                 if (rc != -ENOENT)
1822                         RETURN(rc);
1823
1824                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1825                                                DTO_INDEX_INSERT, true, 0);
1826                 if (rc > 0)
1827                         rc = 0;
1828
1829                 RETURN(rc);
1830         }
1831
1832         if (osd_id_eq_strict(id, id2))
1833                 RETURN(0);
1834
1835         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1836                 osd_scrub_file_reset(scrub,
1837                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1838                                      SF_INCONSISTENT);
1839                 rc = osd_scrub_file_store(scrub);
1840                 if (rc != 0)
1841                         RETURN(rc);
1842         }
1843
1844         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1845                                        DTO_INDEX_UPDATE, true, 0);
1846         if (rc > 0)
1847                 rc = 0;
1848
1849         RETURN(rc);
1850 }
1851
1852 /**
1853  * It scans the /lost+found, and for the OST-object (with filter_fid
1854  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1855  */
1856 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1857                            loff_t offset, __u64 ino, unsigned d_type)
1858 {
1859         struct osd_ios_filldir_buf *fill_buf = buf;
1860         struct osd_thread_info     *info     = fill_buf->oifb_info;
1861         struct osd_device          *dev      = fill_buf->oifb_dev;
1862         struct lu_fid              *fid      = &info->oti_fid;
1863         struct osd_scrub           *scrub    = &dev->od_scrub;
1864         struct dentry              *parent   = fill_buf->oifb_dentry;
1865         struct dentry              *child;
1866         struct inode               *dir      = parent->d_inode;
1867         struct inode               *inode;
1868         int                         rc;
1869         ENTRY;
1870
1871         /* skip any '.' started names */
1872         if (name[0] == '.')
1873                 RETURN(0);
1874
1875         scrub->os_lf_scanned++;
1876         child = osd_ios_lookup_one_len(name, parent, namelen);
1877         if (IS_ERR(child)) {
1878                 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1879                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
1880                 RETURN(0);
1881         }
1882
1883         inode = child->d_inode;
1884         if (S_ISDIR(inode->i_mode)) {
1885                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1886                                       osd_ios_lf_fill);
1887                 if (rc != 0)
1888                         CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1889                               "rc = %d\n", osd_name(dev), namelen, name, rc);
1890                 GOTO(put, rc);
1891         }
1892
1893         if (!S_ISREG(inode->i_mode))
1894                 GOTO(put, rc = 0);
1895
1896         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1897         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1898                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1899                 if (rc == 0) {
1900                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1901                                "/lost+found.\n", namelen, name, PFID(fid));
1902                         scrub->os_lf_repaired++;
1903                 } else {
1904                         CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1905                                DFID": rc = %d\n",
1906                                osd_name(dev), namelen, name, PFID(fid), rc);
1907                 }
1908         }
1909
1910         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1911          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1912          *      can process them in furtuer. */
1913
1914         GOTO(put, rc);
1915
1916 put:
1917         if (rc < 0)
1918                 scrub->os_lf_failed++;
1919         dput(child);
1920         /* skip the failure to make the scanning to continue. */
1921         return 0;
1922 }
1923
1924 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1925                                loff_t offset, __u64 ino, unsigned d_type)
1926 {
1927         struct osd_ios_filldir_buf *fill_buf = buf;
1928         struct osd_device          *dev      = fill_buf->oifb_dev;
1929         struct dentry              *child;
1930         int                         rc;
1931         ENTRY;
1932
1933         /* skip any '.' started names */
1934         if (name[0] == '.')
1935                 RETURN(0);
1936
1937         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1938         if (IS_ERR(child))
1939                 RETURN(PTR_ERR(child));
1940
1941         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1942                               NULL, 0);
1943         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1944                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1945                                       osd_ios_varfid_fill);
1946         dput(child);
1947
1948         RETURN(rc);
1949 }
1950
1951 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1952                            loff_t offset, __u64 ino, unsigned d_type)
1953 {
1954         struct osd_ios_filldir_buf *fill_buf = buf;
1955         struct osd_device          *dev      = fill_buf->oifb_dev;
1956         const struct osd_lf_map    *map;
1957         struct dentry              *child;
1958         int                         rc       = 0;
1959         ENTRY;
1960
1961         /* skip any '.' started names */
1962         if (name[0] == '.')
1963                 RETURN(0);
1964
1965         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
1966                 if (map->olm_namelen != namelen)
1967                         continue;
1968
1969                 if (strncmp(map->olm_name, name, namelen) == 0)
1970                         break;
1971         }
1972
1973         if (map->olm_name == NULL)
1974                 RETURN(0);
1975
1976         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1977         if (IS_ERR(child))
1978                 RETURN(PTR_ERR(child));
1979
1980         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1981                               &map->olm_fid, map->olm_flags);
1982         dput(child);
1983
1984         RETURN(rc);
1985 }
1986
1987 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1988                             loff_t offset, __u64 ino, unsigned d_type)
1989 {
1990         struct osd_ios_filldir_buf *fill_buf = buf;
1991         struct dentry              *child;
1992         struct lu_fid               tfid;
1993         int                         rc       = 0;
1994         ENTRY;
1995
1996         /* skip any non-DFID format name */
1997         if (name[0] != '[')
1998                 RETURN(0);
1999
2000         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2001         if (IS_ERR(child))
2002                 RETURN(PTR_ERR(child));
2003
2004         /* skip the start '[' */
2005         sscanf(&name[1], SFID, RFID(&tfid));
2006         if (fid_is_sane(&tfid))
2007                 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
2008                                       child->d_inode, &tfid, 0);
2009         else
2010                 rc = -EIO;
2011         dput(child);
2012
2013         RETURN(rc);
2014 }
2015
2016 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
2017                              loff_t offset, __u64 ino, unsigned d_type)
2018 {
2019         struct osd_ios_filldir_buf *fill_buf = buf;
2020         struct osd_device          *dev      = fill_buf->oifb_dev;
2021         const struct osd_lf_map    *map;
2022         struct dentry              *child;
2023         int                         rc       = 0;
2024         ENTRY;
2025
2026         /* skip any '.' started names */
2027         if (name[0] == '.')
2028                 RETURN(0);
2029
2030         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2031                 if (map->olm_namelen != namelen)
2032                         continue;
2033
2034                 if (strncmp(map->olm_name, name, namelen) == 0)
2035                         break;
2036         }
2037
2038         if (map->olm_name == NULL)
2039                 RETURN(0);
2040
2041         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2042         if (IS_ERR(child))
2043                 RETURN(PTR_ERR(child));
2044
2045         if (!(map->olm_flags & OLF_NO_OI))
2046                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2047                                       &map->olm_fid, map->olm_flags);
2048         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2049                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2050                                       map->olm_filldir);
2051         dput(child);
2052
2053         RETURN(rc);
2054 }
2055
2056 static int
2057 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2058                      struct dentry *dentry, filldir_t filldir)
2059 {
2060         struct osd_ios_filldir_buf    buf   = {
2061 #ifdef HAVE_DIR_CONTEXT
2062                                                 .ctx.actor = filldir,
2063 #endif
2064                                                 .oifb_info = info,
2065                                                 .oifb_dev = dev,
2066                                                 .oifb_dentry = dentry };
2067         struct file                  *filp  = &info->oti_file;
2068         struct inode                 *inode = dentry->d_inode;
2069         const struct file_operations *fops  = inode->i_fop;
2070         int                           rc;
2071         ENTRY;
2072
2073         LASSERT(filldir != NULL);
2074
2075         filp->f_pos = 0;
2076         filp->f_path.dentry = dentry;
2077         filp->f_mode = FMODE_64BITHASH;
2078         filp->f_mapping = inode->i_mapping;
2079         filp->f_op = fops;
2080         filp->private_data = NULL;
2081         set_file_inode(filp, inode);
2082
2083 #ifdef HAVE_DIR_CONTEXT
2084         buf.ctx.pos = filp->f_pos;
2085         rc = fops->iterate(filp, &buf.ctx);
2086         filp->f_pos = buf.ctx.pos;
2087 #else
2088         rc = fops->readdir(filp, &buf, filldir);
2089 #endif
2090         fops->release(inode, filp);
2091
2092         RETURN(rc);
2093 }
2094
2095 static int
2096 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2097                   struct dentry *dentry, filldir_t filldir)
2098 {
2099         struct osd_scrub  *scrub  = &dev->od_scrub;
2100         struct scrub_file *sf     = &scrub->os_file;
2101         struct dentry     *child;
2102         int                rc;
2103         ENTRY;
2104
2105         /* It is existing MDT0 device. We only allow the case of object without
2106          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2107          * can generate IGIF mode FID for the object and related OI mapping. If
2108          * it is on other MDTs, then becuase file-level backup/restore, related
2109          * OI mapping may be invalid already, we do not know which is the right
2110          * FID for the object. We only allow IGIF objects to reside on the MDT0.
2111          *
2112          * XXX: For the case of object on non-MDT0 device with neither LMA nor
2113          *      "fid" xattr, then something crashed. We cannot re-generate the
2114          *      FID directly, instead, the OI scrub will scan the OI structure
2115          *      and try to re-generate the LMA from the OI mapping. But if the
2116          *      OI mapping crashed or lost also, then we have to give up under
2117          *      double failure cases. */
2118         scrub->os_convert_igif = 1;
2119         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2120                                        strlen(dot_lustre_name));
2121         if (IS_ERR(child)) {
2122                 rc = PTR_ERR(child);
2123                 if (rc == -ENOENT) {
2124                         /* It is 1.8 MDT device. */
2125                         if (!(sf->sf_flags & SF_UPGRADE)) {
2126                                 osd_scrub_file_reset(scrub,
2127                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2128                                         SF_UPGRADE);
2129                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2130                                 rc = osd_scrub_file_store(scrub);
2131                         } else {
2132                                 rc = 0;
2133                         }
2134                 }
2135         } else {
2136                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2137                  * so the client will get IGIF for the ".lustre" object when
2138                  * the MDT restart.
2139                  *
2140                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2141                  * it does not know whether there are some old clients cached
2142                  * the ".lustre" IGIF during the upgrading. Two choices:
2143                  *
2144                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2145                  *    It will allow the old connected clients to access the
2146                  *    ".lustre" with cached IGIF. But it will cause others
2147                  *    on the MDT failed to check "fid_is_dot_lustre()".
2148                  *
2149                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2150                  *    for ".lustre" in spite of whether there are some clients
2151                  *    cached the ".lustre" IGIF or not. It enables the check
2152                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
2153                  *    that the old connected clients cannot access the ".lustre"
2154                  *    with the cached IGIF.
2155                  *
2156                  * Usually, it is rare case for the old connected clients
2157                  * to access the ".lustre" with cached IGIF. So we prefer
2158                  * to the solution 2). */
2159                 rc = osd_ios_scan_one(info, dev, child->d_inode,
2160                                       &LU_DOT_LUSTRE_FID, 0);
2161                 if (rc == 0)
2162                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2163                                               osd_ios_dl_fill);
2164                 dput(child);
2165         }
2166
2167         RETURN(rc);
2168 }
2169
2170 static int
2171 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2172                      struct dentry *dentry, filldir_t filldir)
2173 {
2174         struct osd_scrub  *scrub  = &dev->od_scrub;
2175         struct scrub_file *sf     = &scrub->os_file;
2176         struct dentry     *child;
2177         int                rc;
2178         ENTRY;
2179
2180         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2181                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2182                 rc = osd_scrub_file_store(scrub);
2183                 if (rc != 0)
2184                         RETURN(rc);
2185         }
2186
2187         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2188         if (!IS_ERR(child)) {
2189                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2190                 dput(child);
2191         } else {
2192                 rc = PTR_ERR(child);
2193         }
2194
2195         if (rc != 0 && rc != -ENOENT)
2196                 RETURN(rc);
2197
2198         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2199         if (!IS_ERR(child)) {
2200                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2201                 dput(child);
2202         } else {
2203                 rc = PTR_ERR(child);
2204         }
2205
2206         if (rc == -ENOENT)
2207                 rc = 0;
2208
2209         RETURN(rc);
2210 }
2211
2212 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2213                                 struct osd_device *dev)
2214 {
2215         struct osd_ios_item     *item    = NULL;
2216         scandir_t                scandir = osd_ios_general_scan;
2217         filldir_t                filldir = osd_ios_root_fill;
2218         struct dentry           *dentry  = osd_sb(dev)->s_root;
2219         const struct osd_lf_map *map     = osd_lf_maps;
2220         int                      rc;
2221         ENTRY;
2222
2223         /* Lookup IGIF in OI by force for initial OI scrub. */
2224         dev->od_igif_inoi = 1;
2225
2226         while (1) {
2227                 rc = scandir(info, dev, dentry, filldir);
2228                 if (item != NULL) {
2229                         dput(item->oii_dentry);
2230                         OBD_FREE_PTR(item);
2231                 }
2232
2233                 if (rc != 0)
2234                         break;
2235
2236                 if (list_empty(&dev->od_ios_list))
2237                         break;
2238
2239                 item = list_entry(dev->od_ios_list.next,
2240                                   struct osd_ios_item, oii_list);
2241                 list_del_init(&item->oii_list);
2242
2243                 LASSERT(item->oii_scandir != NULL);
2244                 scandir = item->oii_scandir;
2245                 filldir = item->oii_filldir;
2246                 dentry = item->oii_dentry;
2247         }
2248
2249         while (!list_empty(&dev->od_ios_list)) {
2250                 item = list_entry(dev->od_ios_list.next,
2251                                   struct osd_ios_item, oii_list);
2252                 list_del_init(&item->oii_list);
2253                 dput(item->oii_dentry);
2254                 OBD_FREE_PTR(item);
2255         }
2256
2257         if (rc != 0)
2258                 RETURN(rc);
2259
2260         /* There maybe the case that the object has been removed, but its OI
2261          * mapping is still in the OI file, such as the "CATALOGS" after MDT
2262          * file-level backup/restore. So here cleanup the stale OI mappings. */
2263         while (map->olm_name != NULL) {
2264                 struct dentry *child;
2265
2266                 if (fid_is_zero(&map->olm_fid)) {
2267                         map++;
2268                         continue;
2269                 }
2270
2271                 child = osd_ios_lookup_one_len(map->olm_name,
2272                                                osd_sb(dev)->s_root,
2273                                                map->olm_namelen);
2274                 if (!IS_ERR(child))
2275                         dput(child);
2276                 else if (PTR_ERR(child) == -ENOENT)
2277                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2278                                                   NULL, DTO_INDEX_DELETE,
2279                                                   true, 0);
2280                 map++;
2281         }
2282
2283         RETURN(0);
2284 }
2285
2286 char *osd_lf_fid2name(const struct lu_fid *fid)
2287 {
2288         const struct osd_lf_map *map = osd_lf_maps;
2289
2290         while (map->olm_name != NULL) {
2291                 if (!lu_fid_eq(fid, &map->olm_fid)) {
2292                         map++;
2293                         continue;
2294                 }
2295
2296                 if (map->olm_flags & OLF_SHOW_NAME)
2297                         return map->olm_name;
2298                 else
2299                         return "";
2300         }
2301
2302         return NULL;
2303 }
2304
2305 /* OI scrub start/stop */
2306
2307 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2308 {
2309         struct osd_scrub     *scrub  = &dev->od_scrub;
2310         struct ptlrpc_thread *thread = &scrub->os_thread;
2311         struct l_wait_info    lwi    = { 0 };
2312         struct task_struct   *task;
2313         int                   rc;
2314         ENTRY;
2315
2316         /* os_lock: sync status between stop and scrub thread */
2317         spin_lock(&scrub->os_lock);
2318
2319 again:
2320         if (thread_is_running(thread)) {
2321                 spin_unlock(&scrub->os_lock);
2322                 if (!(scrub->os_file.sf_flags & SF_AUTO) ||
2323                      (flags & (SS_AUTO_FULL | SS_AUTO_PARTIAL)))
2324                         RETURN(-EALREADY);
2325
2326                 osd_scrub_join(dev, flags, false);
2327                 spin_lock(&scrub->os_lock);
2328                 if (!thread_is_running(thread))
2329                         goto again;
2330
2331                 spin_unlock(&scrub->os_lock);
2332                 RETURN(0);
2333         }
2334
2335         if (unlikely(thread_is_stopping(thread))) {
2336                 spin_unlock(&scrub->os_lock);
2337                 l_wait_event(thread->t_ctl_waitq,
2338                              thread_is_stopped(thread),
2339                              &lwi);
2340                 spin_lock(&scrub->os_lock);
2341                 goto again;
2342         }
2343         spin_unlock(&scrub->os_lock);
2344
2345         if (scrub->os_file.sf_status == SS_COMPLETED) {
2346                 if (!(flags & SS_SET_FAILOUT))
2347                         flags |= SS_CLEAR_FAILOUT;
2348
2349                 if (!(flags & SS_SET_DRYRUN))
2350                         flags |= SS_CLEAR_DRYRUN;
2351
2352                 flags |= SS_RESET;
2353         }
2354
2355         scrub->os_start_flags = flags;
2356         thread_set_flags(thread, 0);
2357         task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2358         if (IS_ERR(task)) {
2359                 rc = PTR_ERR(task);
2360                 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2361                        osd_scrub2name(scrub), rc);
2362                 RETURN(rc);
2363         }
2364
2365         l_wait_event(thread->t_ctl_waitq,
2366                      thread_is_running(thread) || thread_is_stopped(thread),
2367                      &lwi);
2368
2369         RETURN(0);
2370 }
2371
2372 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2373 {
2374         int rc;
2375         ENTRY;
2376
2377         /* od_otable_mutex: prevent curcurrent start/stop */
2378         mutex_lock(&dev->od_otable_mutex);
2379         rc = do_osd_scrub_start(dev, flags);
2380         mutex_unlock(&dev->od_otable_mutex);
2381
2382         RETURN(rc == -EALREADY ? 0 : rc);
2383 }
2384
2385 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2386 {
2387         struct ptlrpc_thread *thread = &scrub->os_thread;
2388         struct l_wait_info    lwi    = { 0 };
2389
2390         /* os_lock: sync status between stop and scrub thread */
2391         spin_lock(&scrub->os_lock);
2392         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2393                 thread_set_flags(thread, SVC_STOPPING);
2394                 spin_unlock(&scrub->os_lock);
2395                 wake_up_all(&thread->t_ctl_waitq);
2396                 l_wait_event(thread->t_ctl_waitq,
2397                              thread_is_stopped(thread),
2398                              &lwi);
2399                 /* Do not skip the last lock/unlock, which can guarantee that
2400                  * the caller cannot return until the OI scrub thread exit. */
2401                 spin_lock(&scrub->os_lock);
2402         }
2403         spin_unlock(&scrub->os_lock);
2404 }
2405
2406 static void osd_scrub_stop(struct osd_device *dev)
2407 {
2408         /* od_otable_mutex: prevent curcurrent start/stop */
2409         mutex_lock(&dev->od_otable_mutex);
2410         dev->od_scrub.os_paused = 1;
2411         do_osd_scrub_stop(&dev->od_scrub);
2412         mutex_unlock(&dev->od_otable_mutex);
2413 }
2414
2415 /* OI scrub setup/cleanup */
2416
2417 static const char osd_scrub_name[] = "OI_scrub";
2418
2419 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2420 {
2421         struct osd_thread_info     *info   = osd_oti_get(env);
2422         struct osd_scrub           *scrub  = &dev->od_scrub;
2423         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
2424         struct scrub_file          *sf     = &scrub->os_file;
2425         struct super_block         *sb     = osd_sb(dev);
2426         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
2427         struct lvfs_run_ctxt        saved;
2428         struct file                *filp;
2429         struct inode               *inode;
2430         struct lu_fid              *fid    = &info->oti_fid;
2431         int                         dirty  = 0;
2432         int                         rc     = 0;
2433         ENTRY;
2434
2435         memset(scrub, 0, sizeof(*scrub));
2436         OBD_SET_CTXT_MAGIC(ctxt);
2437         ctxt->pwdmnt = dev->od_mnt;
2438         ctxt->pwd = dev->od_mnt->mnt_root;
2439         ctxt->fs = get_ds();
2440
2441         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2442         init_rwsem(&scrub->os_rwsem);
2443         spin_lock_init(&scrub->os_lock);
2444         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2445
2446         push_ctxt(&saved, ctxt);
2447         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2448         if (IS_ERR(filp)) {
2449                 pop_ctxt(&saved, ctxt);
2450                 RETURN(PTR_ERR(filp));
2451         }
2452
2453         inode = filp->f_path.dentry->d_inode;
2454         /* 'What the @fid is' is not imporatant, because the object
2455          * has no OI mapping, and only is visible inside the OSD.*/
2456         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2457         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2458         if (rc != 0) {
2459                 filp_close(filp, NULL);
2460                 pop_ctxt(&saved, ctxt);
2461                 RETURN(rc);
2462         }
2463
2464         scrub->os_inode = igrab(inode);
2465         filp_close(filp, NULL);
2466         pop_ctxt(&saved, ctxt);
2467
2468         rc = osd_scrub_file_load(scrub);
2469         if (rc == -ENOENT) {
2470                 osd_scrub_file_init(scrub, es->s_uuid);
2471                 /* If the "/O" dir does not exist when mount (indicated by
2472                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2473                  * then it is quite probably that the device is a new one,
2474                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2475                  *
2476                  * For the rare case that "/O" and "OI_scrub" both lost on
2477                  * an old device, it can be found and cleared later.
2478                  *
2479                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2480                  * need to check "filter_fid_old" and to convert it to
2481                  * "filter_fid" for each object, and all the IGIF should
2482                  * have their FID mapping in OI files already. */
2483                 if (dev->od_maybe_new)
2484                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2485                 dirty = 1;
2486         } else if (rc != 0) {
2487                 GOTO(cleanup_inode, rc);
2488         } else {
2489                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2490                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2491                         dirty = 1;
2492                 } else if (sf->sf_status == SS_SCANNING) {
2493                         sf->sf_status = SS_CRASHED;
2494                         dirty = 1;
2495                 }
2496         }
2497
2498         if (sf->sf_pos_last_checkpoint != 0)
2499                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2500         else
2501                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2502
2503         if (dirty != 0) {
2504                 rc = osd_scrub_file_store(scrub);
2505                 if (rc != 0)
2506                         GOTO(cleanup_inode, rc);
2507         }
2508
2509         /* Initialize OI files. */
2510         rc = osd_oi_init(info, dev);
2511         if (rc < 0)
2512                 GOTO(cleanup_inode, rc);
2513
2514         rc = osd_initial_OI_scrub(info, dev);
2515         if (rc != 0)
2516                 GOTO(cleanup_oi, rc);
2517
2518         if (sf->sf_flags & SF_UPGRADE ||
2519             !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2520               sf->sf_success_count > 0)) {
2521                 dev->od_igif_inoi = 0;
2522                 dev->od_check_ff = dev->od_is_ost;
2523         } else {
2524                 dev->od_igif_inoi = 1;
2525                 dev->od_check_ff = 0;
2526         }
2527
2528         if (sf->sf_flags & SF_INCONSISTENT)
2529                 /* The 'od_igif_inoi' will be set under the
2530                  * following cases:
2531                  * 1) new created system, or
2532                  * 2) restored from file-level backup, or
2533                  * 3) the upgrading completed.
2534                  *
2535                  * The 'od_igif_inoi' may be cleared by OI scrub
2536                  * later if found that the system is upgrading. */
2537                 dev->od_igif_inoi = 1;
2538
2539         if (!dev->od_noscrub &&
2540             ((sf->sf_status == SS_PAUSED) ||
2541              (sf->sf_status == SS_CRASHED &&
2542               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2543                               SF_UPGRADE | SF_AUTO)) ||
2544              (sf->sf_status == SS_INIT &&
2545               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2546                               SF_UPGRADE))))
2547                 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2548
2549         if (rc != 0)
2550                 GOTO(cleanup_oi, rc);
2551
2552         /* it is possible that dcache entries may keep objects after they are
2553          * deleted by OSD. While it looks safe this can cause object data to
2554          * stay until umount causing failures in tests calculating free space,
2555          * e.g. replay-ost-single. Since those dcache entries are not used
2556          * anymore let's just free them after use here */
2557         shrink_dcache_sb(sb);
2558
2559         RETURN(0);
2560 cleanup_oi:
2561         osd_oi_fini(info, dev);
2562 cleanup_inode:
2563         iput(scrub->os_inode);
2564         scrub->os_inode = NULL;
2565
2566         return rc;
2567 }
2568
2569 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2570 {
2571         struct osd_scrub *scrub = &dev->od_scrub;
2572
2573         LASSERT(dev->od_otable_it == NULL);
2574
2575         if (scrub->os_inode != NULL) {
2576                 osd_scrub_stop(dev);
2577                 iput(scrub->os_inode);
2578                 scrub->os_inode = NULL;
2579         }
2580         if (dev->od_oi_table != NULL)
2581                 osd_oi_fini(osd_oti_get(env), dev);
2582 }
2583
2584 /* object table based iteration APIs */
2585
2586 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2587                                        struct dt_object *dt, __u32 attr)
2588 {
2589         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2590         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2591         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2592         struct osd_scrub       *scrub = &dev->od_scrub;
2593         struct osd_otable_it   *it;
2594         __u32                   start = 0;
2595         int                     rc;
2596         ENTRY;
2597
2598         /* od_otable_mutex: prevent curcurrent init/fini */
2599         mutex_lock(&dev->od_otable_mutex);
2600         if (dev->od_otable_it != NULL)
2601                 GOTO(out, it = ERR_PTR(-EALREADY));
2602
2603         OBD_ALLOC_PTR(it);
2604         if (it == NULL)
2605                 GOTO(out, it = ERR_PTR(-ENOMEM));
2606
2607         dev->od_otable_it = it;
2608         it->ooi_dev = dev;
2609         it->ooi_cache.ooc_consumer_idx = -1;
2610         if (flags & DOIF_OUTUSED)
2611                 it->ooi_used_outside = 1;
2612
2613         if (flags & DOIF_RESET)
2614                 start |= SS_RESET;
2615
2616         if (valid & DOIV_ERROR_HANDLE) {
2617                 if (flags & DOIF_FAILOUT)
2618                         start |= SS_SET_FAILOUT;
2619                 else
2620                         start |= SS_CLEAR_FAILOUT;
2621         }
2622
2623         if (valid & DOIV_DRYRUN) {
2624                 if (flags & DOIF_DRYRUN)
2625                         start |= SS_SET_DRYRUN;
2626                 else
2627                         start |= SS_CLEAR_DRYRUN;
2628         }
2629
2630         rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2631         if (rc < 0 && rc != -EALREADY) {
2632                 dev->od_otable_it = NULL;
2633                 OBD_FREE_PTR(it);
2634                 GOTO(out, it = ERR_PTR(rc));
2635         }
2636
2637         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2638
2639         GOTO(out, it);
2640
2641 out:
2642         mutex_unlock(&dev->od_otable_mutex);
2643         return (struct dt_it *)it;
2644 }
2645
2646 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2647 {
2648         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2649         struct osd_device    *dev = it->ooi_dev;
2650
2651         /* od_otable_mutex: prevent curcurrent init/fini */
2652         mutex_lock(&dev->od_otable_mutex);
2653         do_osd_scrub_stop(&dev->od_scrub);
2654         LASSERT(dev->od_otable_it == it);
2655
2656         dev->od_otable_it = NULL;
2657         mutex_unlock(&dev->od_otable_mutex);
2658         OBD_FREE_PTR(it);
2659 }
2660
2661 static int osd_otable_it_get(const struct lu_env *env,
2662                              struct dt_it *di, const struct dt_key *key)
2663 {
2664         return 0;
2665 }
2666
2667 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2668 {
2669 }
2670
2671 static inline int
2672 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2673 {
2674         spin_lock(&scrub->os_lock);
2675         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2676             scrub->os_waiting ||
2677             !thread_is_running(&scrub->os_thread))
2678                 it->ooi_waiting = 0;
2679         else
2680                 it->ooi_waiting = 1;
2681         spin_unlock(&scrub->os_lock);
2682
2683         return !it->ooi_waiting;
2684 }
2685
2686 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2687 {
2688         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2689         struct osd_device       *dev    = it->ooi_dev;
2690         struct osd_scrub        *scrub  = &dev->od_scrub;
2691         struct osd_otable_cache *ooc    = &it->ooi_cache;
2692         struct ptlrpc_thread    *thread = &scrub->os_thread;
2693         struct l_wait_info       lwi    = { 0 };
2694         int                      rc;
2695         ENTRY;
2696
2697         LASSERT(it->ooi_user_ready);
2698
2699 again:
2700         if (!thread_is_running(thread) && !it->ooi_used_outside)
2701                 RETURN(1);
2702
2703         if (ooc->ooc_cached_items > 0) {
2704                 ooc->ooc_cached_items--;
2705                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2706                                         ~OSD_OTABLE_IT_CACHE_MASK;
2707                 RETURN(0);
2708         }
2709
2710         if (it->ooi_all_cached) {
2711                 l_wait_event(thread->t_ctl_waitq,
2712                              !thread_is_running(thread),
2713                              &lwi);
2714                 RETURN(1);
2715         }
2716
2717         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2718                 spin_lock(&scrub->os_lock);
2719                 scrub->os_waiting = 0;
2720                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2721                 spin_unlock(&scrub->os_lock);
2722         }
2723
2724         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2725                 l_wait_event(thread->t_ctl_waitq,
2726                              osd_otable_it_wakeup(scrub, it),
2727                              &lwi);
2728
2729         if (!thread_is_running(thread) && !it->ooi_used_outside)
2730                 RETURN(1);
2731
2732         rc = osd_otable_it_preload(env, it);
2733         if (rc >= 0)
2734                 goto again;
2735
2736         RETURN(rc);
2737 }
2738
2739 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2740                                         const struct dt_it *di)
2741 {
2742         return NULL;
2743 }
2744
2745 static int osd_otable_it_key_size(const struct lu_env *env,
2746                                   const struct dt_it *di)
2747 {
2748         return sizeof(__u64);
2749 }
2750
2751 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2752                              struct dt_rec *rec, __u32 attr)
2753 {
2754         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2755         struct osd_otable_cache *ooc = &it->ooi_cache;
2756
2757         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2758
2759         /* Filter out Invald FID already. */
2760         LASSERTF(fid_is_sane((struct lu_fid *)rec),
2761                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2762                  PFID((struct lu_fid *)rec),
2763                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2764
2765         return 0;
2766 }
2767
2768 static __u64 osd_otable_it_store(const struct lu_env *env,
2769                                  const struct dt_it *di)
2770 {
2771         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2772         struct osd_otable_cache *ooc = &it->ooi_cache;
2773         __u64                    hash;
2774
2775         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2776                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2777         else
2778                 hash = ooc->ooc_pos_preload;
2779         return hash;
2780 }
2781
2782 /**
2783  * Set the OSD layer iteration start position as the specified hash.
2784  */
2785 static int osd_otable_it_load(const struct lu_env *env,
2786                               const struct dt_it *di, __u64 hash)
2787 {
2788         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
2789         struct osd_device       *dev   = it->ooi_dev;
2790         struct osd_otable_cache *ooc   = &it->ooi_cache;
2791         struct osd_scrub        *scrub = &dev->od_scrub;
2792         int                      rc;
2793         ENTRY;
2794
2795         /* Forbid to set iteration position after iteration started. */
2796         if (it->ooi_user_ready)
2797                 RETURN(-EPERM);
2798
2799         if (hash > OSD_OTABLE_MAX_HASH)
2800                 hash = OSD_OTABLE_MAX_HASH;
2801
2802         ooc->ooc_pos_preload = hash;
2803         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2804                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2805
2806         it->ooi_user_ready = 1;
2807         if (!scrub->os_full_speed)
2808                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2809
2810         /* Unplug OSD layer iteration by the first next() call. */
2811         rc = osd_otable_it_next(env, (struct dt_it *)it);
2812
2813         RETURN(rc);
2814 }
2815
2816 static int osd_otable_it_key_rec(const struct lu_env *env,
2817                                  const struct dt_it *di, void *key_rec)
2818 {
2819         return 0;
2820 }
2821
2822 const struct dt_index_operations osd_otable_ops = {
2823         .dio_it = {
2824                 .init     = osd_otable_it_init,
2825                 .fini     = osd_otable_it_fini,
2826                 .get      = osd_otable_it_get,
2827                 .put      = osd_otable_it_put,
2828                 .next     = osd_otable_it_next,
2829                 .key      = osd_otable_it_key,
2830                 .key_size = osd_otable_it_key_size,
2831                 .rec      = osd_otable_it_rec,
2832                 .store    = osd_otable_it_store,
2833                 .load     = osd_otable_it_load,
2834                 .key_rec  = osd_otable_it_key_rec,
2835         }
2836 };
2837
2838 /* high priority inconsistent items list APIs */
2839
2840 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL  60
2841
2842 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2843                    int insert)
2844 {
2845         struct osd_inconsistent_item *oii;
2846         struct osd_scrub             *scrub  = &dev->od_scrub;
2847         struct ptlrpc_thread         *thread = &scrub->os_thread;
2848         int                           wakeup = 0;
2849         ENTRY;
2850
2851         OBD_ALLOC_PTR(oii);
2852         if (unlikely(oii == NULL))
2853                 RETURN(-ENOMEM);
2854
2855         INIT_LIST_HEAD(&oii->oii_list);
2856         oii->oii_cache = *oic;
2857         oii->oii_insert = insert;
2858
2859         if (scrub->os_partial_scan) {
2860                 __u64 now = cfs_time_current_sec();
2861
2862                 /* If there haven't been errors in a long time,
2863                  * decay old count until either the errors are
2864                  * gone or we reach the current interval. */
2865                 while (unlikely(scrub->os_bad_oimap_count > 0 &&
2866                                 scrub->os_bad_oimap_time +
2867                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
2868                         scrub->os_bad_oimap_count >>= 1;
2869                         scrub->os_bad_oimap_time +=
2870                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
2871                 }
2872
2873                 scrub->os_bad_oimap_time = now;
2874                 if (++scrub->os_bad_oimap_count >
2875                     dev->od_full_scrub_threshold_rate)
2876                         scrub->os_full_scrub = 1;
2877         }
2878
2879         spin_lock(&scrub->os_lock);
2880         if (unlikely(!thread_is_running(thread))) {
2881                 spin_unlock(&scrub->os_lock);
2882                 OBD_FREE_PTR(oii);
2883                 RETURN(-EAGAIN);
2884         }
2885
2886         if (list_empty(&scrub->os_inconsistent_items))
2887                 wakeup = 1;
2888         list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2889         spin_unlock(&scrub->os_lock);
2890
2891         if (wakeup != 0)
2892                 wake_up_all(&thread->t_ctl_waitq);
2893
2894         RETURN(0);
2895 }
2896
2897 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2898                    struct osd_inode_id *id)
2899 {
2900         struct osd_scrub             *scrub = &dev->od_scrub;
2901         struct osd_inconsistent_item *oii;
2902         ENTRY;
2903
2904         spin_lock(&scrub->os_lock);
2905         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2906                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2907                         *id = oii->oii_cache.oic_lid;
2908                         spin_unlock(&scrub->os_lock);
2909                         RETURN(0);
2910                 }
2911         }
2912         spin_unlock(&scrub->os_lock);
2913
2914         RETURN(-ENOENT);
2915 }
2916
2917 /* OI scrub dump */
2918
2919 static const char *scrub_status_names[] = {
2920         "init",
2921         "scanning",
2922         "completed",
2923         "failed",
2924         "stopped",
2925         "paused",
2926         "crashed",
2927         NULL
2928 };
2929
2930 static const char *scrub_flags_names[] = {
2931         "recreated",
2932         "inconsistent",
2933         "auto",
2934         "upgrade",
2935         NULL
2936 };
2937
2938 static const char *scrub_param_names[] = {
2939         "failout",
2940         "dryrun",
2941         NULL
2942 };
2943
2944 static int scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
2945                            const char *prefix)
2946 {
2947         int flag;
2948         int rc;
2949         int i;
2950
2951         rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2952         if (rc < 0)
2953                 return rc;
2954
2955         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2956                 if (flag & bits) {
2957                         bits &= ~flag;
2958                         rc = seq_printf(m, "%s%c", names[i],
2959                                         bits != 0 ? ',' : '\n');
2960                         if (rc < 0)
2961                                 return rc;
2962                 }
2963         }
2964         return 0;
2965 }
2966
2967 static int scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
2968 {
2969         int rc;
2970
2971         if (time != 0)
2972                 rc = seq_printf(m, "%s: "LPU64" seconds\n", prefix,
2973                               cfs_time_current_sec() - time);
2974         else
2975                 rc = seq_printf(m, "%s: N/A\n", prefix);
2976         return rc;
2977 }
2978
2979 static int scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
2980 {
2981         int rc;
2982
2983         if (pos != 0)
2984                 rc = seq_printf(m, "%s: "LPU64"\n", prefix, pos);
2985         else
2986                 rc = seq_printf(m, "%s: N/A\n", prefix);
2987         return rc;
2988 }
2989
2990 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
2991 {
2992         struct osd_scrub  *scrub   = &dev->od_scrub;
2993         struct scrub_file *sf      = &scrub->os_file;
2994         __u64              checked;
2995         __u64              speed;
2996         int                rc;
2997
2998         down_read(&scrub->os_rwsem);
2999         rc = seq_printf(m, "name: OI_scrub\n"
3000                         "magic: 0x%x\n"
3001                         "oi_files: %d\n"
3002                         "status: %s\n",
3003                         sf->sf_magic, (int)sf->sf_oi_count,
3004                         scrub_status_names[sf->sf_status]);
3005         if (rc < 0)
3006                 goto out;
3007
3008         rc = scrub_bits_dump(m, sf->sf_flags, scrub_flags_names,
3009                              "flags");
3010         if (rc < 0)
3011                 goto out;
3012
3013         rc = scrub_bits_dump(m, sf->sf_param, scrub_param_names,
3014                              "param");
3015         if (rc < 0)
3016                 goto out;
3017
3018         rc = scrub_time_dump(m, sf->sf_time_last_complete,
3019                              "time_since_last_completed");
3020         if (rc < 0)
3021                 goto out;
3022
3023         rc = scrub_time_dump(m, sf->sf_time_latest_start,
3024                              "time_since_latest_start");
3025         if (rc < 0)
3026                 goto out;
3027
3028         rc = scrub_time_dump(m, sf->sf_time_last_checkpoint,
3029                              "time_since_last_checkpoint");
3030         if (rc < 0)
3031                 goto out;
3032
3033         rc = scrub_pos_dump(m, sf->sf_pos_latest_start,
3034                             "latest_start_position");
3035         if (rc < 0)
3036                 goto out;
3037
3038         rc = scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
3039                             "last_checkpoint_position");
3040         if (rc < 0)
3041                 goto out;
3042
3043         rc = scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
3044                             "first_failure_position");
3045         if (rc < 0)
3046                 goto out;
3047
3048         checked = sf->sf_items_checked + scrub->os_new_checked;
3049         rc = seq_printf(m, "checked: "LPU64"\n"
3050                       "updated: "LPU64"\n"
3051                       "failed: "LPU64"\n"
3052                       "prior_updated: "LPU64"\n"
3053                       "noscrub: "LPU64"\n"
3054                       "igif: "LPU64"\n"
3055                       "success_count: %u\n",
3056                       checked, sf->sf_items_updated, sf->sf_items_failed,
3057                       sf->sf_items_updated_prior, sf->sf_items_noscrub,
3058                       sf->sf_items_igif, sf->sf_success_count);
3059         if (rc < 0)
3060                 goto out;
3061
3062         speed = checked;
3063         if (thread_is_running(&scrub->os_thread)) {
3064                 cfs_duration_t duration = cfs_time_current() -
3065                                           scrub->os_time_last_checkpoint;
3066                 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
3067                                                      MSEC_PER_SEC);
3068                 __u32 rtime = sf->sf_run_time +
3069                               cfs_duration_sec(duration + HALF_SEC);
3070
3071                 if (duration != 0)
3072                         do_div(new_checked, duration);
3073                 if (rtime != 0)
3074                         do_div(speed, rtime);
3075                 rc = seq_printf(m, "run_time: %u seconds\n"
3076                               "average_speed: "LPU64" objects/sec\n"
3077                               "real-time_speed: "LPU64" objects/sec\n"
3078                               "current_position: %u\n"
3079                               "lf_scanned: "LPU64"\n"
3080                               "lf_repaired: "LPU64"\n"
3081                               "lf_failed: "LPU64"\n",
3082                               rtime, speed, new_checked, scrub->os_pos_current,
3083                               scrub->os_lf_scanned, scrub->os_lf_repaired,
3084                               scrub->os_lf_failed);
3085         } else {
3086                 if (sf->sf_run_time != 0)
3087                         do_div(speed, sf->sf_run_time);
3088                 rc = seq_printf(m, "run_time: %u seconds\n"
3089                               "average_speed: "LPU64" objects/sec\n"
3090                               "real-time_speed: N/A\n"
3091                               "current_position: N/A\n"
3092                               "lf_scanned: "LPU64"\n"
3093                               "lf_repaired: "LPU64"\n"
3094                               "lf_failed: "LPU64"\n",
3095                               sf->sf_run_time, speed, scrub->os_lf_scanned,
3096                               scrub->os_lf_repaired, scrub->os_lf_failed);
3097         }
3098
3099 out:
3100         up_read(&scrub->os_rwsem);
3101         return (rc < 0 ? -ENOSPC : 0);
3102 }