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