lustre/osd-ldiskfs/osd_scrub.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2012, 2015, Intel Corporation.
  24  */
  25 /*
  26  * lustre/osd-ldiskfs/osd_scrub.c
  27  *
  28  * Top-level entry points into osd module
  29  *
  30  * The OI scrub is used for rebuilding Object Index files when restores MDT from
  31  * file-level backup.
  32  *
  33  * The otable based iterator scans ldiskfs inode table to feed up layer LFSCK.
  34  *
  35  * Author: Fan Yong <yong.fan@whamcloud.com>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_LFSCK
  39
  40 #include <linux/kthread.h>
  41 #include <lustre/lustre_idl.h>
  42 #include <lustre_disk.h>
  43 #include <dt_object.h>
  44 #include <linux/xattr.h>
  45
  46 #include "osd_internal.h"
  47 #include "osd_oi.h"
  48 #include "osd_scrub.h"
  49
  50 #define HALF_SEC        msecs_to_jiffies(MSEC_PER_SEC >> 1)
  51
  52 #define OSD_OTABLE_MAX_HASH             0x00000000ffffffffULL
  53
  54 #define SCRUB_NEXT_BREAK        1 /* exit current loop and process next group */
  55 #define SCRUB_NEXT_CONTINUE     2 /* skip current object and process next bit */
  56 #define SCRUB_NEXT_EXIT         3 /* exit all the loops */
  57 #define SCRUB_NEXT_WAIT         4 /* wait for free cache slot */
  58 #define SCRUB_NEXT_CRASH        5 /* simulate system crash during OI scrub */
  59 #define SCRUB_NEXT_FATAL        6 /* simulate failure during OI scrub */
  60 #define SCRUB_NEXT_NOSCRUB      7 /* new created object, no scrub on it */
  61 #define SCRUB_NEXT_NOLMA        8 /* the inode has no FID-in-LMA */
  62 #define SCRUB_NEXT_OSTOBJ       9 /* for OST-object */
  63 #define SCRUB_NEXT_OSTOBJ_OLD   10 /* old OST-object, no LMA or no FID-on-OST
  64                                     * flags in LMA */
  65
  66 /* misc functions */
  67
  68 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
  69 {
  70         return container_of0(scrub, struct osd_device, od_scrub);
  71 }
  72
  73 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
  74 {
  75         return osd_sb(osd_scrub2dev(scrub));
  76 }
  77
  78 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
  79                                        struct osd_otable_cache *ooc)
  80 {
  81         return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
  82 }
  83
  84 static inline const char *osd_scrub2name(struct osd_scrub *scrub)
  85 {
  86         return LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
  87 }
  88
  89 /**
  90  * update/insert/delete the specified OI mapping (@fid @id) according to the ops
  91  *
  92  * \retval   1, changed nothing
  93  * \retval   0, changed successfully
  94  * \retval -ve, on error
  95  */
  96 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
  97                                      struct osd_device *dev,
  98                                      const struct lu_fid *fid,
  99                                      const struct osd_inode_id *id,
 100                                      int ops, bool force,
 101                                      enum oi_check_flags flags)
 102 {
 103         handle_t *th;
 104         int       rc;
 105         ENTRY;
 106
 107         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
 108                 RETURN(0);
 109
 110         /* DTO_INDEX_INSERT is enough for other two ops:
 111          * delete/update, but save stack. */
 112         th = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 113                                 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
 114         if (IS_ERR(th)) {
 115                 rc = PTR_ERR(th);
 116                 CDEBUG(D_LFSCK, "%s: fail to start trans for scrub op %d "
 117                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 118                        PFID(fid), id->oii_ino, id->oii_gen, rc);
 119                 RETURN(rc);
 120         }
 121
 122         switch (ops) {
 123         case DTO_INDEX_UPDATE:
 124                 rc = osd_oi_update(info, dev, fid, id, th, flags);
 125                 if (unlikely(rc == -ENOENT)) {
 126                         /* Some unlink thread may removed the OI mapping. */
 127                         rc = 1;
 128                 }
 129                 break;
 130         case DTO_INDEX_INSERT:
 131                 rc = osd_oi_insert(info, dev, fid, id, th, flags);
 132                 if (unlikely(rc == -EEXIST)) {
 133                         rc = 1;
 134                         /* XXX: There are trouble things when adding OI
 135                          *      mapping for IGIF object, which may cause
 136                          *      multiple objects to be mapped to the same
 137                          *      IGIF formatted FID. Consider the following
 138                          *      situations:
 139                          *
 140                          *      1) The MDT is upgrading from 1.8 device.
 141                          *      The OI scrub generates IGIF FID1 for the
 142                          *      OBJ1 and adds the OI mapping.
 143                          *
 144                          *      2) For some reason, the OI scrub does not
 145                          *      process all the IGIF objects completely.
 146                          *
 147                          *      3) The MDT is backuped and restored against
 148                          *      this device.
 149                          *
 150                          *      4) When the MDT mounts up, the OI scrub will
 151                          *      try to rebuild the OI files. For some IGIF
 152                          *      object, OBJ2, which was not processed by the
 153                          *      OI scrub before the backup/restore, and the
 154                          *      new generated IGIF formatted FID may be just
 155                          *      the FID1, the same as OBJ1.
 156                          *
 157                          *      Under such case, the OI scrub cannot know how
 158                          *      to generate new FID for the OBJ2.
 159                          *
 160                          *      Currently, we do nothing for that. One possible
 161                          *      solution is to generate new normal FID for the
 162                          *      conflict object.
 163                          *
 164                          *      Anyway, it is rare, only exists in theory. */
 165                 }
 166                 break;
 167         case DTO_INDEX_DELETE:
 168                 rc = osd_oi_delete(info, dev, fid, th, flags);
 169                 if (rc == -ENOENT) {
 170                         /* It is normal that the unlink thread has removed the
 171                          * OI mapping already. */
 172                         rc = 1;
 173                 }
 174                 break;
 175         default:
 176                 LASSERTF(0, "Unexpected ops %d\n", ops);
 177                 break;
 178         }
 179
 180         ldiskfs_journal_stop(th);
 181         if (rc < 0)
 182                 CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d "
 183                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 184                        PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1,
 185                        rc);
 186
 187         RETURN(rc);
 188 }
 189
 190 /* OI_scrub file ops */
 191
 192 static void osd_scrub_file_to_cpu(struct scrub_file *des,
 193                                   struct scrub_file *src)
 194 {
 195         memcpy(des->sf_uuid, src->sf_uuid, 16);
 196         des->sf_flags   = le64_to_cpu(src->sf_flags);
 197         des->sf_magic   = le32_to_cpu(src->sf_magic);
 198         des->sf_status  = le16_to_cpu(src->sf_status);
 199         des->sf_param   = le16_to_cpu(src->sf_param);
 200         des->sf_time_last_complete      =
 201                                 le64_to_cpu(src->sf_time_last_complete);
 202         des->sf_time_latest_start       =
 203                                 le64_to_cpu(src->sf_time_latest_start);
 204         des->sf_time_last_checkpoint    =
 205                                 le64_to_cpu(src->sf_time_last_checkpoint);
 206         des->sf_pos_latest_start        =
 207                                 le64_to_cpu(src->sf_pos_latest_start);
 208         des->sf_pos_last_checkpoint     =
 209                                 le64_to_cpu(src->sf_pos_last_checkpoint);
 210         des->sf_pos_first_inconsistent  =
 211                                 le64_to_cpu(src->sf_pos_first_inconsistent);
 212         des->sf_items_checked           =
 213                                 le64_to_cpu(src->sf_items_checked);
 214         des->sf_items_updated           =
 215                                 le64_to_cpu(src->sf_items_updated);
 216         des->sf_items_failed            =
 217                                 le64_to_cpu(src->sf_items_failed);
 218         des->sf_items_updated_prior     =
 219                                 le64_to_cpu(src->sf_items_updated_prior);
 220         des->sf_run_time        = le32_to_cpu(src->sf_run_time);
 221         des->sf_success_count   = le32_to_cpu(src->sf_success_count);
 222         des->sf_oi_count        = le16_to_cpu(src->sf_oi_count);
 223         des->sf_internal_flags  = le16_to_cpu(src->sf_internal_flags);
 224         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 225 }
 226
 227 static void osd_scrub_file_to_le(struct scrub_file *des,
 228                                  struct scrub_file *src)
 229 {
 230         memcpy(des->sf_uuid, src->sf_uuid, 16);
 231         des->sf_flags   = cpu_to_le64(src->sf_flags);
 232         des->sf_magic   = cpu_to_le32(src->sf_magic);
 233         des->sf_status  = cpu_to_le16(src->sf_status);
 234         des->sf_param   = cpu_to_le16(src->sf_param);
 235         des->sf_time_last_complete      =
 236                                 cpu_to_le64(src->sf_time_last_complete);
 237         des->sf_time_latest_start       =
 238                                 cpu_to_le64(src->sf_time_latest_start);
 239         des->sf_time_last_checkpoint    =
 240                                 cpu_to_le64(src->sf_time_last_checkpoint);
 241         des->sf_pos_latest_start        =
 242                                 cpu_to_le64(src->sf_pos_latest_start);
 243         des->sf_pos_last_checkpoint     =
 244                                 cpu_to_le64(src->sf_pos_last_checkpoint);
 245         des->sf_pos_first_inconsistent  =
 246                                 cpu_to_le64(src->sf_pos_first_inconsistent);
 247         des->sf_items_checked           =
 248                                 cpu_to_le64(src->sf_items_checked);
 249         des->sf_items_updated           =
 250                                 cpu_to_le64(src->sf_items_updated);
 251         des->sf_items_failed            =
 252                                 cpu_to_le64(src->sf_items_failed);
 253         des->sf_items_updated_prior     =
 254                                 cpu_to_le64(src->sf_items_updated_prior);
 255         des->sf_run_time        = cpu_to_le32(src->sf_run_time);
 256         des->sf_success_count   = cpu_to_le32(src->sf_success_count);
 257         des->sf_oi_count        = cpu_to_le16(src->sf_oi_count);
 258         des->sf_internal_flags  = cpu_to_le16(src->sf_internal_flags);
 259         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 260 }
 261
 262 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
 263 {
 264         struct scrub_file *sf = &scrub->os_file;
 265
 266         memset(sf, 0, sizeof(*sf));
 267         memcpy(sf->sf_uuid, uuid, 16);
 268         sf->sf_magic = SCRUB_MAGIC_V1;
 269         sf->sf_status = SS_INIT;
 270 }
 271
 272 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
 273 {
 274         struct scrub_file *sf = &scrub->os_file;
 275
 276         CDEBUG(D_LFSCK, "%.16s: reset OI scrub file, old flags = "
 277                LPX64", add flags = "LPX64"\n",
 278                osd_scrub2name(scrub), sf->sf_flags, flags);
 279
 280         memcpy(sf->sf_uuid, uuid, 16);
 281         sf->sf_status = SS_INIT;
 282         sf->sf_flags |= flags;
 283         sf->sf_flags &= ~SF_AUTO;
 284         sf->sf_run_time = 0;
 285         sf->sf_time_latest_start = 0;
 286         sf->sf_time_last_checkpoint = 0;
 287         sf->sf_pos_latest_start = 0;
 288         sf->sf_pos_last_checkpoint = 0;
 289         sf->sf_pos_first_inconsistent = 0;
 290         sf->sf_items_checked = 0;
 291         sf->sf_items_updated = 0;
 292         sf->sf_items_failed = 0;
 293         if (!scrub->os_in_join)
 294                 sf->sf_items_updated_prior = 0;
 295
 296         sf->sf_items_noscrub = 0;
 297         sf->sf_items_igif = 0;
 298 }
 299
 300 static int osd_scrub_file_load(struct osd_scrub *scrub)
 301 {
 302         loff_t  pos  = 0;
 303         int     len  = sizeof(scrub->os_file_disk);
 304         int     rc;
 305
 306         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
 307         if (rc == len) {
 308                 struct scrub_file *sf = &scrub->os_file;
 309
 310                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
 311                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
 312                         CDEBUG(D_LFSCK, "%.16s: invalid scrub magic "
 313                                "0x%x != 0x%x\n", osd_scrub2name(scrub),
 314                                sf->sf_magic, SCRUB_MAGIC_V1);
 315                         /* Process it as new scrub file. */
 316                         rc = -ENOENT;
 317                 } else {
 318                         rc = 0;
 319                 }
 320         } else if (rc != 0) {
 321                 CDEBUG(D_LFSCK, "%.16s: fail to load scrub file, "
 322                        "expected = %d: rc = %d\n",
 323                        osd_scrub2name(scrub), len, rc);
 324                 if (rc > 0)
 325                         rc = -EFAULT;
 326         } else {
 327                 /* return -ENOENT for empty scrub file case. */
 328                 rc = -ENOENT;
 329         }
 330
 331         return rc;
 332 }
 333
 334 int osd_scrub_file_store(struct osd_scrub *scrub)
 335 {
 336         struct osd_device *dev;
 337         handle_t          *jh;
 338         loff_t             pos     = 0;
 339         int                len     = sizeof(scrub->os_file_disk);
 340         int                credits;
 341         int                rc;
 342
 343         dev = container_of0(scrub, struct osd_device, od_scrub);
 344         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
 345                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
 346         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits);
 347         if (IS_ERR(jh)) {
 348                 rc = PTR_ERR(jh);
 349                 CDEBUG(D_LFSCK, "%.16s: fail to start trans for scrub store: "
 350                        "rc = %d\n", osd_scrub2name(scrub), rc);
 351                 return rc;
 352         }
 353
 354         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
 355         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
 356                                       len, 0, &pos, jh);
 357         ldiskfs_journal_stop(jh);
 358         if (rc != 0)
 359                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file, "
 360                        "expected = %d: rc = %d\n",
 361                        osd_scrub2name(scrub), len, rc);
 362
 363         scrub->os_time_last_checkpoint = cfs_time_current();
 364         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
 365                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
 366         return rc;
 367 }
 368
 369 static int
 370 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
 371                      struct inode *inode, const struct lu_fid *fid)
 372 {
 373         struct filter_fid_old   *ff      = &info->oti_ff;
 374         struct dentry           *dentry  = &info->oti_obj_dentry;
 375         struct lu_fid           *tfid    = &info->oti_fid;
 376         handle_t                *jh;
 377         int                      size    = 0;
 378         int                      rc;
 379         bool                     removed = false;
 380         bool                     reset   = true;
 381         ENTRY;
 382
 383         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
 384                 RETURN(0);
 385
 386         if (fid_is_idif(fid) && dev->od_index_in_idif == 0) {
 387                 struct ost_id *oi = &info->oti_ostid;
 388
 389                 fid_to_ostid(fid, oi);
 390                 ostid_to_fid(tfid, oi, 0);
 391         } else {
 392                 *tfid = *fid;
 393         }
 394
 395         /* We want the LMA to fit into the 256-byte OST inode, so operate
 396          * as following:
 397          * 1) read old XATTR_NAME_FID and save the parent FID;
 398          * 2) delete the old XATTR_NAME_FID;
 399          * 3) make new LMA and add it;
 400          * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
 401          *
 402          * Making the LMA to fit into the 256-byte OST inode can save time for
 403          * normal osd_check_lma() and for other OI scrub scanning in future.
 404          * So it is worth to make some slow conversion here. */
 405         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 406                                 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
 407         if (IS_ERR(jh)) {
 408                 rc = PTR_ERR(jh);
 409                 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
 410                        DFID": rc = %d\n", osd_name(dev), PFID(tfid), rc);
 411                 RETURN(rc);
 412         }
 413
 414         /* 1) read old XATTR_NAME_FID and save the parent FID */
 415         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 416         if (rc == sizeof(*ff)) {
 417                 /* 2) delete the old XATTR_NAME_FID */
 418                 ll_vfs_dq_init(inode);
 419                 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
 420                 if (rc != 0)
 421                         GOTO(stop, rc);
 422
 423                 removed = true;
 424         } else if (unlikely(rc == -ENODATA)) {
 425                 reset = false;
 426         } else if (rc != sizeof(struct filter_fid)) {
 427                 GOTO(stop, rc = -EINVAL);
 428         }
 429
 430         /* 3) make new LMA and add it */
 431         rc = osd_ea_fid_set(info, inode, tfid, LMAC_FID_ON_OST, 0);
 432         if (rc == 0 && reset)
 433                 size = sizeof(struct filter_fid);
 434         else if (rc != 0 && removed)
 435                 /* If failed, we should try to add the old back. */
 436                 size = sizeof(struct filter_fid_old);
 437
 438         /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
 439         if (size > 0) {
 440                 int rc1;
 441
 442                 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
 443                                       XATTR_CREATE);
 444                 if (rc1 != 0 && rc == 0)
 445                         rc = rc1;
 446         }
 447
 448         GOTO(stop, rc);
 449
 450 stop:
 451         ldiskfs_journal_stop(jh);
 452         if (rc < 0)
 453                 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
 454                        osd_name(dev), PFID(tfid), rc);
 455         return rc;
 456 }
 457
 458 static int
 459 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
 460                        struct osd_idmap_cache *oic, int val)
 461 {
 462         struct osd_scrub             *scrub  = &dev->od_scrub;
 463         struct scrub_file            *sf     = &scrub->os_file;
 464         struct lu_fid                *fid    = &oic->oic_fid;
 465         struct osd_inode_id          *lid    = &oic->oic_lid;
 466         struct osd_inode_id          *lid2   = &info->oti_id;
 467         struct osd_inconsistent_item *oii    = NULL;
 468         struct inode                 *inode  = NULL;
 469         int                           ops    = DTO_INDEX_UPDATE;
 470         int                           idx;
 471         int                           rc;
 472         bool                          converted = false;
 473         ENTRY;
 474
 475         down_write(&scrub->os_rwsem);
 476         scrub->os_new_checked++;
 477         if (val < 0)
 478                 GOTO(out, rc = val);
 479
 480         if (scrub->os_in_prior)
 481                 oii = list_entry(oic, struct osd_inconsistent_item,
 482                                  oii_cache);
 483
 484         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
 485                 GOTO(out, rc = 0);
 486
 487         if (fid_is_igif(fid))
 488                 sf->sf_items_igif++;
 489
 490         if (val == SCRUB_NEXT_OSTOBJ_OLD) {
 491                 inode = osd_iget(info, dev, lid);
 492                 if (IS_ERR(inode)) {
 493                         rc = PTR_ERR(inode);
 494                         /* Someone removed the inode. */
 495                         if (rc == -ENOENT || rc == -ESTALE)
 496                                 rc = 0;
 497                         GOTO(out, rc);
 498                 }
 499
 500                 /* The inode has been reused as EA inode, ignore it. */
 501                 if (unlikely(osd_is_ea_inode(inode)))
 502                         GOTO(out, rc = 0);
 503
 504                 sf->sf_flags |= SF_UPGRADE;
 505                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
 506                 dev->od_check_ff = 1;
 507                 rc = osd_scrub_convert_ff(info, dev, inode, fid);
 508                 if (rc != 0)
 509                         GOTO(out, rc);
 510
 511                 converted = true;
 512         }
 513
 514         if ((val == SCRUB_NEXT_NOLMA) &&
 515             (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
 516                 GOTO(out, rc = 0);
 517
 518         if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA)) {
 519                 ops = DTO_INDEX_INSERT;
 520
 521                 goto iget;
 522         }
 523
 524         rc = osd_oi_lookup(info, dev, fid, lid2,
 525                 (val == SCRUB_NEXT_OSTOBJ ||
 526                  val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 527         if (rc != 0) {
 528                 if (rc == -ENOENT)
 529                         ops = DTO_INDEX_INSERT;
 530                 else if (rc != -ESTALE)
 531                         GOTO(out, rc);
 532
 533 iget:
 534                 if (inode == NULL) {
 535                         inode = osd_iget(info, dev, lid);
 536                         if (IS_ERR(inode)) {
 537                                 rc = PTR_ERR(inode);
 538                                 /* Someone removed the inode. */
 539                                 if (rc == -ENOENT || rc == -ESTALE)
 540                                         rc = 0;
 541                                 GOTO(out, rc);
 542                         }
 543
 544                         /* The inode has been reused as EA inode, ignore it. */
 545                         if (unlikely(osd_is_ea_inode(inode)))
 546                                 GOTO(out, rc = 0);
 547                 }
 548
 549                 if (!scrub->os_partial_scan)
 550                         scrub->os_full_speed = 1;
 551
 552                 idx = osd_oi_fid2idx(dev, fid);
 553                 switch (val) {
 554                 case SCRUB_NEXT_NOLMA:
 555                         sf->sf_flags |= SF_UPGRADE;
 556                         if (!(sf->sf_param & SP_DRYRUN)) {
 557                                 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
 558                                 if (rc != 0)
 559                                         GOTO(out, rc);
 560                         }
 561
 562                         if (!(sf->sf_flags & SF_INCONSISTENT))
 563                                 dev->od_igif_inoi = 0;
 564                         break;
 565                 case SCRUB_NEXT_OSTOBJ:
 566                         sf->sf_flags |= SF_INCONSISTENT;
 567                 case SCRUB_NEXT_OSTOBJ_OLD:
 568                         break;
 569                 default:
 570                         sf->sf_flags |= SF_RECREATED;
 571                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 572                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 573                         break;
 574                 }
 575         } else if (osd_id_eq(lid, lid2)) {
 576                 if (converted)
 577                         sf->sf_items_updated++;
 578
 579                 GOTO(out, rc = 0);
 580         } else {
 581                 if (!scrub->os_partial_scan)
 582                         scrub->os_full_speed = 1;
 583
 584                 sf->sf_flags |= SF_INCONSISTENT;
 585
 586                 /* XXX: If the device is restored from file-level backup, then
 587                  *      some IGIFs may have been already in OI files, and some
 588                  *      may be not yet. Means upgrading from 1.8 may be partly
 589                  *      processed, but some clients may hold some immobilized
 590                  *      IGIFs, and use them to access related objects. Under
 591                  *      such case, OSD does not know whether an given IGIF has
 592                  *      been processed or to be processed, and it also cannot
 593                  *      generate local ino#/gen# directly from the immobilized
 594                  *      IGIF because of the backup/restore. Then force OSD to
 595                  *      lookup the given IGIF in OI files, and if no entry,
 596                  *      then ask the client to retry after upgrading completed.
 597                  *      No better choice. */
 598                 dev->od_igif_inoi = 1;
 599         }
 600
 601         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
 602                         (val == SCRUB_NEXT_OSTOBJ ||
 603                          val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 604         if (rc == 0) {
 605                 if (scrub->os_in_prior)
 606                         sf->sf_items_updated_prior++;
 607                 else
 608                         sf->sf_items_updated++;
 609         }
 610
 611         GOTO(out, rc);
 612
 613 out:
 614         if (rc < 0) {
 615                 sf->sf_items_failed++;
 616                 if (sf->sf_pos_first_inconsistent == 0 ||
 617                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 618                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 619         } else {
 620                 rc = 0;
 621         }
 622
 623         /* There may be conflict unlink during the OI scrub,
 624          * if happend, then remove the new added OI mapping. */
 625         if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
 626             unlikely(ldiskfs_test_inode_state(inode,
 627                                               LDISKFS_STATE_LUSTRE_DESTROY)))
 628                 osd_scrub_refresh_mapping(info, dev, fid, lid,
 629                                 DTO_INDEX_DELETE, false,
 630                                 (val == SCRUB_NEXT_OSTOBJ ||
 631                                  val == SCRUB_NEXT_OSTOBJ_OLD) ?
 632                                 OI_KNOWN_ON_OST : 0);
 633         up_write(&scrub->os_rwsem);
 634
 635         if (inode != NULL && !IS_ERR(inode))
 636                 iput(inode);
 637
 638         if (oii != NULL) {
 639                 LASSERT(list_empty(&oii->oii_list));
 640
 641                 OBD_FREE_PTR(oii);
 642         }
 643
 644         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 645 }
 646
 647 /* OI scrub APIs */
 648
 649 static int osd_scrub_prep(struct osd_device *dev)
 650 {
 651         struct osd_scrub     *scrub  = &dev->od_scrub;
 652         struct ptlrpc_thread *thread = &scrub->os_thread;
 653         struct scrub_file    *sf     = &scrub->os_file;
 654         __u32                 flags  = scrub->os_start_flags;
 655         int                   rc;
 656         bool                  drop_dryrun = false;
 657         ENTRY;
 658
 659         CDEBUG(D_LFSCK, "%.16s: OI scrub prep, flags = 0x%x\n",
 660                osd_scrub2name(scrub), flags);
 661
 662         down_write(&scrub->os_rwsem);
 663         if (flags & SS_SET_FAILOUT)
 664                 sf->sf_param |= SP_FAILOUT;
 665         else if (flags & SS_CLEAR_FAILOUT)
 666                 sf->sf_param &= ~SP_FAILOUT;
 667
 668         if (flags & SS_SET_DRYRUN) {
 669                 sf->sf_param |= SP_DRYRUN;
 670         } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
 671                 sf->sf_param &= ~SP_DRYRUN;
 672                 drop_dryrun = true;
 673         }
 674
 675         if (flags & SS_RESET)
 676                 osd_scrub_file_reset(scrub,
 677                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
 678
 679         if (flags & SS_AUTO_FULL) {
 680                 scrub->os_full_speed = 1;
 681                 scrub->os_partial_scan = 0;
 682                 sf->sf_flags |= SF_AUTO;
 683         } else if (flags & SS_AUTO_PARTIAL) {
 684                 scrub->os_full_speed = 0;
 685                 scrub->os_partial_scan = 1;
 686                 sf->sf_flags |= SF_AUTO;
 687         } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
 688                                    SF_UPGRADE)) {
 689                 scrub->os_full_speed = 1;
 690                 scrub->os_partial_scan = 0;
 691         } else {
 692                 scrub->os_full_speed = 0;
 693                 scrub->os_partial_scan = 0;
 694         }
 695
 696         spin_lock(&scrub->os_lock);
 697         scrub->os_in_prior = 0;
 698         scrub->os_waiting = 0;
 699         scrub->os_paused = 0;
 700         scrub->os_in_join = 0;
 701         scrub->os_full_scrub = 0;
 702         spin_unlock(&scrub->os_lock);
 703         scrub->os_new_checked = 0;
 704         if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
 705                 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
 706         else if (sf->sf_pos_last_checkpoint != 0)
 707                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 708         else
 709                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 710
 711         scrub->os_pos_current = sf->sf_pos_latest_start;
 712         sf->sf_status = SS_SCANNING;
 713         sf->sf_time_latest_start = cfs_time_current_sec();
 714         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 715         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
 716         rc = osd_scrub_file_store(scrub);
 717         if (rc == 0) {
 718                 spin_lock(&scrub->os_lock);
 719                 thread_set_flags(thread, SVC_RUNNING);
 720                 spin_unlock(&scrub->os_lock);
 721                 wake_up_all(&thread->t_ctl_waitq);
 722         }
 723         up_write(&scrub->os_rwsem);
 724
 725         RETURN(rc);
 726 }
 727
 728 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
 729 {
 730         struct scrub_file *sf = &scrub->os_file;
 731         int                rc;
 732
 733         if (likely(cfs_time_before(cfs_time_current(),
 734                                    scrub->os_time_next_checkpoint) ||
 735                    scrub->os_new_checked == 0))
 736                 return 0;
 737
 738         down_write(&scrub->os_rwsem);
 739         sf->sf_items_checked += scrub->os_new_checked;
 740         scrub->os_new_checked = 0;
 741         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 742         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 743         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 744                                             scrub->os_time_last_checkpoint);
 745         rc = osd_scrub_file_store(scrub);
 746         up_write(&scrub->os_rwsem);
 747
 748         return rc;
 749 }
 750
 751 static int osd_scrub_post(struct osd_scrub *scrub, int result)
 752 {
 753         struct scrub_file *sf = &scrub->os_file;
 754         int rc;
 755         ENTRY;
 756
 757         CDEBUG(D_LFSCK, "%.16s: OI scrub post, result = %d\n",
 758                osd_scrub2name(scrub), result);
 759
 760         down_write(&scrub->os_rwsem);
 761         spin_lock(&scrub->os_lock);
 762         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 763         spin_unlock(&scrub->os_lock);
 764         if (scrub->os_new_checked > 0) {
 765                 sf->sf_items_checked += scrub->os_new_checked;
 766                 scrub->os_new_checked = 0;
 767                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 768         }
 769         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 770         if (result > 0) {
 771                 struct osd_device *dev =
 772                         container_of0(scrub, struct osd_device, od_scrub);
 773
 774                 dev->od_igif_inoi = 1;
 775                 dev->od_check_ff = 0;
 776                 sf->sf_status = SS_COMPLETED;
 777                 if (!(sf->sf_param & SP_DRYRUN)) {
 778                         memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 779                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 780                                           SF_UPGRADE | SF_AUTO);
 781                 }
 782                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 783                 sf->sf_success_count++;
 784         } else if (result == 0) {
 785                 if (scrub->os_paused)
 786                         sf->sf_status = SS_PAUSED;
 787                 else
 788                         sf->sf_status = SS_STOPPED;
 789         } else {
 790                 sf->sf_status = SS_FAILED;
 791         }
 792         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 793                                             scrub->os_time_last_checkpoint);
 794         rc = osd_scrub_file_store(scrub);
 795         up_write(&scrub->os_rwsem);
 796
 797         RETURN(rc < 0 ? rc : result);
 798 }
 799
 800 /* iteration engine */
 801
 802 struct osd_iit_param {
 803         struct super_block *sb;
 804         struct buffer_head *bitmap;
 805         ldiskfs_group_t bg;
 806         __u32 gbase;
 807         __u32 offset;
 808 };
 809
 810 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 811                                    struct osd_device *dev,
 812                                    struct osd_iit_param *param,
 813                                    struct osd_idmap_cache **oic,
 814                                    const bool noslot);
 815
 816 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 817                                    struct osd_device *dev,
 818                                    struct osd_iit_param *param,
 819                                    struct osd_idmap_cache *oic,
 820                                    bool *noslot, int rc);
 821
 822 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 823 {
 824         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 825                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 826         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 827                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 828                 return SCRUB_NEXT_BREAK;
 829         } else {
 830                 *pos = param->gbase + param->offset;
 831                 return 0;
 832         }
 833 }
 834
 835 /**
 836  * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
 837  * \retval 0: FID-on-MDT
 838  */
 839 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
 840                                       struct osd_device *dev,
 841                                       struct lu_fid *fid)
 842 {
 843         /* XXX: The initial OI scrub will scan the top level /O to generate
 844          *      a small local FLDB according to the <seq>. If the given FID
 845          *      is in the local FLDB, then it is FID-on-OST; otherwise it's
 846          *      quite possible for FID-on-MDT. */
 847         if (dev->od_is_ost)
 848                 return SCRUB_NEXT_OSTOBJ_OLD;
 849         else
 850                 return 0;
 851 }
 852
 853 static int osd_scrub_get_fid(struct osd_thread_info *info,
 854                              struct osd_device *dev, struct inode *inode,
 855                              struct lu_fid *fid, bool scrub)
 856 {
 857         struct lustre_mdt_attrs *lma     = &info->oti_mdt_attrs;
 858         int                      rc;
 859         bool                     has_lma = false;
 860
 861         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 862         if (rc == 0) {
 863                 has_lma = true;
 864                 if (lma->lma_compat & LMAC_NOT_IN_OI ||
 865                     lma->lma_incompat & LMAI_AGENT)
 866                         return SCRUB_NEXT_CONTINUE;
 867
 868                 *fid = lma->lma_self_fid;
 869                 if (!scrub)
 870                         return 0;
 871
 872                 if (lma->lma_compat & LMAC_FID_ON_OST)
 873                         return SCRUB_NEXT_OSTOBJ;
 874
 875                 if (fid_is_idif(fid))
 876                         return SCRUB_NEXT_OSTOBJ_OLD;
 877
 878                 /* For local object. */
 879                 if (fid_is_internal(fid))
 880                         return 0;
 881
 882                 /* For external visible MDT-object with non-normal FID. */
 883                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
 884                         return 0;
 885
 886                 /* For the object with normal FID, it may be MDT-object,
 887                  * or may be 2.4 OST-object, need further distinguish.
 888                  * Fall through to next section. */
 889         }
 890
 891         if (rc == -ENODATA || rc == 0) {
 892                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
 893                 if (rc == 0) {
 894                         if (scrub)
 895                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
 896                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 897                         return rc;
 898                 }
 899
 900                 if (rc > 0) {
 901                         if (!has_lma)
 902                                 /* It is FID-on-OST, but we do not know how
 903                                  * to generate its FID, ignore it directly. */
 904                                 rc = SCRUB_NEXT_CONTINUE;
 905                         else
 906                                 /* It is 2.4 OST-object. */
 907                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 908                         return rc;
 909                 }
 910
 911                 if (rc != -ENODATA)
 912                         return rc;
 913
 914                 if (!has_lma) {
 915                         if (dev->od_scrub.os_convert_igif) {
 916                                 lu_igif_build(fid, inode->i_ino,
 917                                               inode->i_generation);
 918                                 if (scrub)
 919                                         rc = SCRUB_NEXT_NOLMA;
 920                                 else
 921                                         rc = 0;
 922                         } else {
 923                                 /* It may be FID-on-OST, or may be FID for
 924                                  * non-MDT0, anyway, we do not know how to
 925                                  * generate its FID, ignore it directly. */
 926                                 rc = SCRUB_NEXT_CONTINUE;
 927                         }
 928                         return rc;
 929                 }
 930
 931                 /* For OI scrub case only: the object has LMA but has no ff
 932                  * (or ff crashed). It may be MDT-object, may be OST-object
 933                  * with crashed ff. The last check is local FLDB. */
 934                 rc = osd_scrub_check_local_fldb(info, dev, fid);
 935         }
 936
 937         return rc;
 938 }
 939
 940 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 941                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 942                         struct super_block *sb, bool scrub)
 943 {
 944         struct inode *inode;
 945         int           rc;
 946         ENTRY;
 947
 948         /* Not handle the backend root object and agent parent object.
 949          * They are neither visible to namespace nor have OI mappings. */
 950         if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
 951                      pos == osd_remote_parent_ino(dev)))
 952                 RETURN(SCRUB_NEXT_CONTINUE);
 953
 954         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 955         inode = osd_iget(info, dev, lid);
 956         if (IS_ERR(inode)) {
 957                 rc = PTR_ERR(inode);
 958                 /* The inode may be removed after bitmap searching, or the
 959                  * file is new created without inode initialized yet. */
 960                 if (rc == -ENOENT || rc == -ESTALE)
 961                         RETURN(SCRUB_NEXT_CONTINUE);
 962
 963                 CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: "
 964                        "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name,
 965                        pos, rc);
 966                 RETURN(rc);
 967         }
 968
 969         /* It is an EA inode, no OI mapping for it, skip it. */
 970         if (osd_is_ea_inode(inode))
 971                 GOTO(put, rc = SCRUB_NEXT_CONTINUE);
 972
 973         if (scrub &&
 974             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 975                 /* Only skip it for the first OI scrub accessing. */
 976                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 977                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
 978         }
 979
 980         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
 981
 982         GOTO(put, rc);
 983
 984 put:
 985         iput(inode);
 986         return rc;
 987 }
 988
 989 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 990                           struct osd_iit_param *param,
 991                           struct osd_idmap_cache **oic, const bool noslot)
 992 {
 993         struct osd_scrub     *scrub  = &dev->od_scrub;
 994         struct ptlrpc_thread *thread = &scrub->os_thread;
 995         struct lu_fid        *fid;
 996         struct osd_inode_id  *lid;
 997         int                   rc;
 998
 999         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
1000                 struct l_wait_info lwi;
1001
1002                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
1003                 if (likely(lwi.lwi_timeout > 0))
1004                         l_wait_event(thread->t_ctl_waitq,
1005                                 !list_empty(&scrub->os_inconsistent_items) ||
1006                                 !thread_is_running(thread),
1007                                 &lwi);
1008         }
1009
1010         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
1011                 spin_lock(&scrub->os_lock);
1012                 thread_set_flags(thread, SVC_STOPPING);
1013                 spin_unlock(&scrub->os_lock);
1014                 return SCRUB_NEXT_CRASH;
1015         }
1016
1017         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
1018                 return SCRUB_NEXT_FATAL;
1019
1020         if (unlikely(!thread_is_running(thread)))
1021                 return SCRUB_NEXT_EXIT;
1022
1023         if (!list_empty(&scrub->os_inconsistent_items)) {
1024                 spin_lock(&scrub->os_lock);
1025                 if (likely(!list_empty(&scrub->os_inconsistent_items))) {
1026                         struct osd_inconsistent_item *oii;
1027
1028                         oii = list_entry(scrub->os_inconsistent_items.next,
1029                                 struct osd_inconsistent_item, oii_list);
1030                         list_del_init(&oii->oii_list);
1031                         spin_unlock(&scrub->os_lock);
1032
1033                         *oic = &oii->oii_cache;
1034                         scrub->os_in_prior = 1;
1035
1036                         return 0;
1037                 }
1038                 spin_unlock(&scrub->os_lock);
1039         }
1040
1041         if (noslot)
1042                 return SCRUB_NEXT_WAIT;
1043
1044         rc = osd_iit_next(param, &scrub->os_pos_current);
1045         if (rc != 0)
1046                 return rc;
1047
1048         *oic = &scrub->os_oic;
1049         fid = &(*oic)->oic_fid;
1050         lid = &(*oic)->oic_lid;
1051         rc = osd_iit_iget(info, dev, fid, lid,
1052                           scrub->os_pos_current, param->sb, true);
1053         return rc;
1054 }
1055
1056 static int osd_preload_next(struct osd_thread_info *info,
1057                             struct osd_device *dev, struct osd_iit_param *param,
1058                             struct osd_idmap_cache **oic, const bool noslot)
1059 {
1060         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
1061         struct osd_scrub        *scrub;
1062         struct ptlrpc_thread    *thread;
1063         int                      rc;
1064
1065         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1066         if (rc != 0)
1067                 return rc;
1068
1069         scrub = &dev->od_scrub;
1070         thread = &scrub->os_thread;
1071         if (thread_is_running(thread) &&
1072             ooc->ooc_pos_preload >= scrub->os_pos_current)
1073                 return SCRUB_NEXT_EXIT;
1074
1075         rc = osd_iit_iget(info, dev,
1076                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1077                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1078                           ooc->ooc_pos_preload, param->sb, false);
1079         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1080          * ignore the failure, so it still need to skip the inode next time. */
1081         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1082         return rc;
1083 }
1084
1085 static inline int
1086 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1087 {
1088         spin_lock(&scrub->os_lock);
1089         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1090             !list_empty(&scrub->os_inconsistent_items) ||
1091             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1092                 scrub->os_waiting = 0;
1093         else
1094                 scrub->os_waiting = 1;
1095         spin_unlock(&scrub->os_lock);
1096
1097         return !scrub->os_waiting;
1098 }
1099
1100 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1101                           struct osd_iit_param *param,
1102                           struct osd_idmap_cache *oic, bool *noslot, int rc)
1103 {
1104         struct l_wait_info       lwi    = { 0 };
1105         struct osd_scrub        *scrub  = &dev->od_scrub;
1106         struct scrub_file       *sf     = &scrub->os_file;
1107         struct ptlrpc_thread    *thread = &scrub->os_thread;
1108         struct osd_otable_it    *it     = dev->od_otable_it;
1109         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
1110
1111         switch (rc) {
1112         case SCRUB_NEXT_CONTINUE:
1113                 goto next;
1114         case SCRUB_NEXT_WAIT:
1115                 goto wait;
1116         case SCRUB_NEXT_NOSCRUB:
1117                 down_write(&scrub->os_rwsem);
1118                 scrub->os_new_checked++;
1119                 sf->sf_items_noscrub++;
1120                 up_write(&scrub->os_rwsem);
1121                 goto next;
1122         }
1123
1124         rc = osd_scrub_check_update(info, dev, oic, rc);
1125         if (rc != 0) {
1126                 scrub->os_in_prior = 0;
1127                 return rc;
1128         }
1129
1130         rc = osd_scrub_checkpoint(scrub);
1131         if (rc != 0) {
1132                 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1133                        "rc = %d\n", osd_scrub2name(scrub),
1134                        scrub->os_pos_current, rc);
1135                 /* Continue, as long as the scrub itself can go ahead. */
1136         }
1137
1138         if (scrub->os_in_prior) {
1139                 scrub->os_in_prior = 0;
1140                 return 0;
1141         }
1142
1143 next:
1144         scrub->os_pos_current = param->gbase + ++(param->offset);
1145
1146 wait:
1147         if (it != NULL && it->ooi_waiting && ooc != NULL &&
1148             ooc->ooc_pos_preload < scrub->os_pos_current) {
1149                 spin_lock(&scrub->os_lock);
1150                 it->ooi_waiting = 0;
1151                 wake_up_all(&thread->t_ctl_waitq);
1152                 spin_unlock(&scrub->os_lock);
1153         }
1154
1155         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1156                 return 0;
1157
1158         if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1159                 *noslot = false;
1160                 return 0;
1161         }
1162
1163         if (it != NULL)
1164                 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1165                              &lwi);
1166
1167         if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1168                 *noslot = false;
1169         else
1170                 *noslot = true;
1171         return 0;
1172 }
1173
1174 static int osd_preload_exec(struct osd_thread_info *info,
1175                             struct osd_device *dev, struct osd_iit_param *param,
1176                             struct osd_idmap_cache *oic, bool *noslot, int rc)
1177 {
1178         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1179
1180         if (rc == 0) {
1181                 ooc->ooc_cached_items++;
1182                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1183                                         ~OSD_OTABLE_IT_CACHE_MASK;
1184         }
1185         return rc > 0 ? 0 : rc;
1186 }
1187
1188 #define SCRUB_IT_ALL    1
1189 #define SCRUB_IT_CRASH  2
1190
1191 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1192                            bool inconsistent)
1193 {
1194         struct osd_scrub     *scrub  = &dev->od_scrub;
1195         struct ptlrpc_thread *thread = &scrub->os_thread;
1196         struct scrub_file    *sf     = &scrub->os_file;
1197         int                   rc;
1198         ENTRY;
1199
1200         LASSERT(!(flags & SS_AUTO_PARTIAL));
1201
1202         down_write(&scrub->os_rwsem);
1203         scrub->os_in_join = 1;
1204         if (flags & SS_SET_FAILOUT)
1205                 sf->sf_param |= SP_FAILOUT;
1206         else if (flags & SS_CLEAR_FAILOUT)
1207                 sf->sf_param &= ~SP_FAILOUT;
1208
1209         if (flags & SS_SET_DRYRUN)
1210                 sf->sf_param |= SP_DRYRUN;
1211         else if (flags & SS_CLEAR_DRYRUN)
1212                 sf->sf_param &= ~SP_DRYRUN;
1213
1214         if (flags & SS_RESET) {
1215                 osd_scrub_file_reset(scrub,
1216                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1217                         inconsistent ? SF_INCONSISTENT : 0);
1218                 sf->sf_status = SS_SCANNING;
1219         }
1220
1221         if (flags & SS_AUTO_FULL) {
1222                 sf->sf_flags |= SF_AUTO;
1223                 scrub->os_full_speed = 1;
1224         }
1225
1226         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1227                 scrub->os_full_speed = 1;
1228         else
1229                 scrub->os_full_speed = 0;
1230
1231         scrub->os_new_checked = 0;
1232         if (sf->sf_pos_last_checkpoint != 0)
1233                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1234         else
1235                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1236
1237         scrub->os_pos_current = sf->sf_pos_latest_start;
1238         sf->sf_time_latest_start = cfs_time_current_sec();
1239         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1240         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1241         rc = osd_scrub_file_store(scrub);
1242         if (rc != 0)
1243                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
1244                        "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1245
1246         spin_lock(&scrub->os_lock);
1247         scrub->os_waiting = 0;
1248         scrub->os_paused = 0;
1249         scrub->os_partial_scan = 0;
1250         scrub->os_in_join = 0;
1251         scrub->os_full_scrub = 0;
1252         spin_unlock(&scrub->os_lock);
1253         wake_up_all(&thread->t_ctl_waitq);
1254         up_write(&scrub->os_rwsem);
1255
1256         EXIT;
1257 }
1258
1259 static int osd_inode_iteration(struct osd_thread_info *info,
1260                                struct osd_device *dev, __u32 max, bool preload)
1261 {
1262         struct osd_scrub     *scrub  = &dev->od_scrub;
1263         struct ptlrpc_thread *thread = &scrub->os_thread;
1264         struct scrub_file    *sf     = &scrub->os_file;
1265         osd_iit_next_policy   next;
1266         osd_iit_exec_policy   exec;
1267         __u32                *pos;
1268         __u32                *count;
1269         struct osd_iit_param  param  = { NULL };
1270         struct l_wait_info    lwi    = { 0 };
1271         __u32                 limit;
1272         int                   rc;
1273         bool                  noslot = true;
1274         ENTRY;
1275
1276         param.sb = osd_sb(dev);
1277         if (preload)
1278                 goto full;
1279
1280         while (scrub->os_partial_scan && !scrub->os_in_join) {
1281                 struct osd_idmap_cache *oic = NULL;
1282
1283                 rc = osd_scrub_next(info, dev, &param, &oic, noslot);
1284                 switch (rc) {
1285                 case SCRUB_NEXT_EXIT:
1286                         RETURN(0);
1287                 case SCRUB_NEXT_CRASH:
1288                         RETURN(SCRUB_IT_CRASH);
1289                 case SCRUB_NEXT_FATAL:
1290                         RETURN(-EINVAL);
1291                 case SCRUB_NEXT_WAIT: {
1292                         struct kstatfs *ksfs = &info->oti_ksfs;
1293                         __u64 saved_flags;
1294
1295                         if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1296                             unlikely(sf->sf_items_updated_prior == 0))
1297                                 goto wait;
1298
1299                         if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1300                             scrub->os_full_scrub) {
1301                                 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1302                                                true);
1303                                 goto full;
1304                         }
1305
1306                         rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1307                         if (rc == 0) {
1308                                 __u64 used = ksfs->f_files - ksfs->f_ffree;
1309
1310                                 do_div(used, sf->sf_items_updated_prior);
1311                                 /* If we hit too much inconsistent OI
1312                                  * mappings during the partial scan,
1313                                  * then scan the device completely. */
1314                                 if (used < dev->od_full_scrub_ratio) {
1315                                         osd_scrub_join(dev,
1316                                                 SS_AUTO_FULL | SS_RESET, true);
1317                                         goto full;
1318                                 }
1319                         }
1320
1321 wait:
1322                         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1323                             cfs_fail_val > 0)
1324                                 continue;
1325
1326                         saved_flags = sf->sf_flags;
1327                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1328                                           SF_UPGRADE | SF_AUTO);
1329                         sf->sf_status = SS_COMPLETED;
1330                         l_wait_event(thread->t_ctl_waitq,
1331                                      !thread_is_running(thread) ||
1332                                      !scrub->os_partial_scan ||
1333                                      scrub->os_in_join ||
1334                                      !list_empty(&scrub->os_inconsistent_items),
1335                                      &lwi);
1336                         sf->sf_flags = saved_flags;
1337                         sf->sf_status = SS_SCANNING;
1338
1339                         if (unlikely(!thread_is_running(thread)))
1340                                 RETURN(0);
1341
1342                         if (!scrub->os_partial_scan || scrub->os_in_join)
1343                                 goto full;
1344
1345                         continue;
1346                 }
1347                 default:
1348                         LASSERTF(rc == 0, "rc = %d\n", rc);
1349
1350                         osd_scrub_exec(info, dev, &param, oic, &noslot, rc);
1351                         break;
1352                 }
1353         }
1354
1355 full:
1356         if (!preload) {
1357                 l_wait_event(thread->t_ctl_waitq,
1358                              !thread_is_running(thread) || !scrub->os_in_join,
1359                              &lwi);
1360
1361                 if (unlikely(!thread_is_running(thread)))
1362                         RETURN(0);
1363         }
1364
1365         noslot = false;
1366         if (!preload) {
1367                 next = osd_scrub_next;
1368                 exec = osd_scrub_exec;
1369                 pos = &scrub->os_pos_current;
1370                 count = &scrub->os_new_checked;
1371         } else {
1372                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1373
1374                 next = osd_preload_next;
1375                 exec = osd_preload_exec;
1376                 pos = &ooc->ooc_pos_preload;
1377                 count = &ooc->ooc_cached_items;
1378         }
1379         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1380
1381         while (*pos <= limit && *count < max) {
1382                 struct osd_idmap_cache *oic = NULL;
1383                 struct ldiskfs_group_desc *desc;
1384
1385                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1386                 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1387                 if (desc == NULL)
1388                         RETURN(-EIO);
1389
1390                 ldiskfs_lock_group(param.sb, param.bg);
1391                 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1392                         ldiskfs_unlock_group(param.sb, param.bg);
1393                         *pos = 1 + (param.bg + 1) *
1394                                 LDISKFS_INODES_PER_GROUP(param.sb);
1395                         continue;
1396                 }
1397                 ldiskfs_unlock_group(param.sb, param.bg);
1398
1399                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1400                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1401                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1402                 if (param.bitmap == NULL) {
1403                         CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1404                                "scrub will stop, urgent mode\n",
1405                                osd_scrub2name(scrub), (__u32)param.bg);
1406                         RETURN(-EIO);
1407                 }
1408
1409                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1410                        *count < max) {
1411                         if (param.offset +
1412                                 ldiskfs_itable_unused_count(param.sb, desc) >
1413                             LDISKFS_INODES_PER_GROUP(param.sb))
1414                                 goto next_group;
1415
1416                         rc = next(info, dev, &param, &oic, noslot);
1417                         switch (rc) {
1418                         case SCRUB_NEXT_BREAK:
1419                                 goto next_group;
1420                         case SCRUB_NEXT_EXIT:
1421                                 brelse(param.bitmap);
1422                                 RETURN(0);
1423                         case SCRUB_NEXT_CRASH:
1424                                 brelse(param.bitmap);
1425                                 RETURN(SCRUB_IT_CRASH);
1426                         case SCRUB_NEXT_FATAL:
1427                                 brelse(param.bitmap);
1428                                 RETURN(-EINVAL);
1429                         }
1430
1431                         rc = exec(info, dev, &param, oic, &noslot, rc);
1432                         if (rc != 0) {
1433                                 brelse(param.bitmap);
1434                                 RETURN(rc);
1435                         }
1436                 }
1437
1438 next_group:
1439                 brelse(param.bitmap);
1440         }
1441
1442         if (*pos > limit)
1443                 RETURN(SCRUB_IT_ALL);
1444         RETURN(0);
1445 }
1446
1447 static int osd_otable_it_preload(const struct lu_env *env,
1448                                  struct osd_otable_it *it)
1449 {
1450         struct osd_device       *dev   = it->ooi_dev;
1451         struct osd_scrub        *scrub = &dev->od_scrub;
1452         struct osd_otable_cache *ooc   = &it->ooi_cache;
1453         int                      rc;
1454         ENTRY;
1455
1456         rc = osd_inode_iteration(osd_oti_get(env), dev,
1457                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1458         if (rc == SCRUB_IT_ALL)
1459                 it->ooi_all_cached = 1;
1460
1461         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1462                 spin_lock(&scrub->os_lock);
1463                 scrub->os_waiting = 0;
1464                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1465                 spin_unlock(&scrub->os_lock);
1466         }
1467
1468         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1469 }
1470
1471 static int osd_scrub_main(void *args)
1472 {
1473         struct lu_env         env;
1474         struct osd_device    *dev    = (struct osd_device *)args;
1475         struct osd_scrub     *scrub  = &dev->od_scrub;
1476         struct ptlrpc_thread *thread = &scrub->os_thread;
1477         int                   rc;
1478         ENTRY;
1479
1480         rc = lu_env_init(&env, LCT_LOCAL);
1481         if (rc != 0) {
1482                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1483                        osd_scrub2name(scrub), rc);
1484                 GOTO(noenv, rc);
1485         }
1486
1487         rc = osd_scrub_prep(dev);
1488         if (rc != 0) {
1489                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1490                        osd_scrub2name(scrub), rc);
1491                 GOTO(out, rc);
1492         }
1493
1494         if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1495                 struct l_wait_info lwi = { 0 };
1496                 struct osd_otable_it *it = dev->od_otable_it;
1497                 struct osd_otable_cache *ooc = &it->ooi_cache;
1498
1499                 l_wait_event(thread->t_ctl_waitq,
1500                              it->ooi_user_ready || !thread_is_running(thread),
1501                              &lwi);
1502                 if (unlikely(!thread_is_running(thread)))
1503                         GOTO(post, rc = 0);
1504
1505                 scrub->os_pos_current = ooc->ooc_pos_preload;
1506         }
1507
1508         CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1509                osd_scrub2name(scrub), scrub->os_start_flags,
1510                scrub->os_pos_current);
1511
1512         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1513         if (unlikely(rc == SCRUB_IT_CRASH))
1514                 GOTO(out, rc = -EINVAL);
1515         GOTO(post, rc);
1516
1517 post:
1518         rc = osd_scrub_post(scrub, rc);
1519         CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1520                osd_scrub2name(scrub), scrub->os_pos_current, rc);
1521
1522 out:
1523         while (!list_empty(&scrub->os_inconsistent_items)) {
1524                 struct osd_inconsistent_item *oii;
1525
1526                 oii = list_entry(scrub->os_inconsistent_items.next,
1527                                      struct osd_inconsistent_item, oii_list);
1528                 list_del_init(&oii->oii_list);
1529                 OBD_FREE_PTR(oii);
1530         }
1531         lu_env_fini(&env);
1532
1533 noenv:
1534         spin_lock(&scrub->os_lock);
1535         thread_set_flags(thread, SVC_STOPPED);
1536         wake_up_all(&thread->t_ctl_waitq);
1537         spin_unlock(&scrub->os_lock);
1538         return rc;
1539 }
1540
1541 /* initial OI scrub */
1542
1543 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1544                          struct dentry *, filldir_t filldir);
1545
1546 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1547                                loff_t offset, __u64 ino, unsigned d_type);
1548 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1549                            loff_t offset, __u64 ino, unsigned d_type);
1550 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1551                            loff_t offset, __u64 ino, unsigned d_type);
1552 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1553                             loff_t offset, __u64 ino, unsigned d_type);
1554
1555 static int
1556 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1557                      struct dentry *dentry, filldir_t filldir);
1558 static int
1559 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1560                   struct dentry *dentry, filldir_t filldir);
1561
1562 static int
1563 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1564                      struct dentry *dentry, filldir_t filldir);
1565
1566 enum osd_lf_flags {
1567         OLF_SCAN_SUBITEMS       = 0x0001,
1568         OLF_HIDE_FID            = 0x0002,
1569         OLF_SHOW_NAME           = 0x0004,
1570         OLF_NO_OI               = 0x0008,
1571         OLF_IDX_IN_FID          = 0x0010,
1572 };
1573
1574 struct osd_lf_map {
1575         char            *olm_name;
1576         struct lu_fid    olm_fid;
1577         __u16            olm_flags;
1578         __u16            olm_namelen;
1579         scandir_t        olm_scandir;
1580         filldir_t        olm_filldir;
1581 };
1582
1583 /* Add the new introduced local files in the list in the future. */
1584 static const struct osd_lf_map osd_lf_maps[] = {
1585         /* CATALOGS */
1586         { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1587                 sizeof(CATLIST) - 1, NULL, NULL },
1588
1589         /* CONFIGS */
1590         { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1591                 OLF_SCAN_SUBITEMS, sizeof(MOUNT_CONFIGS_DIR) - 1,
1592                 osd_ios_general_scan, osd_ios_varfid_fill },
1593
1594         /* NIDTBL_VERSIONS */
1595         { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1596                 sizeof(MGS_NIDTBL_DIR) - 1, osd_ios_general_scan,
1597                 osd_ios_varfid_fill },
1598
1599         /* PENDING */
1600         { "PENDING", { 0, 0, 0 }, 0, sizeof("PENDING") - 1, NULL, NULL },
1601
1602         /* ROOT */
1603         { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 },
1604                 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, sizeof("ROOT") - 1,
1605                 osd_ios_ROOT_scan, NULL },
1606
1607         /* changelog_catalog */
1608         { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, sizeof(CHANGELOG_CATALOG) - 1,
1609                 NULL, NULL },
1610
1611         /* changelog_users */
1612         { CHANGELOG_USERS, { 0, 0, 0 }, 0, sizeof(CHANGELOG_USERS) - 1,
1613                 NULL, NULL },
1614
1615         /* fld */
1616         { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1617                 sizeof("fld") - 1, NULL, NULL },
1618
1619         /* last_rcvd */
1620         { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1621                 sizeof(LAST_RCVD) - 1, NULL, NULL },
1622
1623         /* reply_data */
1624         { REPLY_DATA, { FID_SEQ_LOCAL_FILE, REPLY_DATA_OID, 0 }, OLF_SHOW_NAME,
1625                 sizeof(REPLY_DATA) - 1, NULL, NULL },
1626
1627         /* lov_objid */
1628         { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1629                 sizeof(LOV_OBJID) - 1, NULL, NULL },
1630
1631         /* lov_objseq */
1632         { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1633                 OLF_SHOW_NAME, sizeof(LOV_OBJSEQ) - 1, NULL, NULL },
1634
1635         /* quota_master */
1636         { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QMT_DIR) - 1,
1637                 osd_ios_general_scan, osd_ios_varfid_fill },
1638
1639         /* quota_slave */
1640         { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QSD_DIR) - 1,
1641                 osd_ios_general_scan, osd_ios_varfid_fill },
1642
1643         /* seq_ctl */
1644         { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1645                 OLF_SHOW_NAME, sizeof("seq_ctl") - 1, NULL, NULL },
1646
1647         /* seq_srv */
1648         { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1649                 OLF_SHOW_NAME, sizeof("seq_srv") - 1, NULL, NULL },
1650
1651         /* health_check */
1652         { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1653                 OLF_SHOW_NAME, sizeof(HEALTH_CHECK) - 1, NULL, NULL },
1654
1655         /* LFSCK */
1656         { LFSCK_DIR, { 0, 0, 0 }, 0, sizeof(LFSCK_DIR) - 1,
1657                 osd_ios_general_scan, osd_ios_varfid_fill },
1658
1659         /* lfsck_bookmark */
1660         { LFSCK_BOOKMARK, { 0, 0, 0 }, 0, sizeof(LFSCK_BOOKMARK) - 1,
1661                 NULL, NULL },
1662
1663         /* lfsck_layout */
1664         { LFSCK_LAYOUT, { 0, 0, 0 }, 0, sizeof(LFSCK_LAYOUT) - 1,
1665                 NULL, NULL },
1666
1667         /* lfsck_namespace */
1668         { LFSCK_NAMESPACE, { 0, 0, 0 }, 0, sizeof(LFSCK_NAMESPACE) - 1,
1669                 NULL, NULL },
1670
1671         /* OBJECTS, upgrade from old device */
1672         { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(OBJECTS) - 1,
1673                 osd_ios_OBJECTS_scan, NULL },
1674
1675         /* lquota_v2.user, upgrade from old device */
1676         { "lquota_v2.user", { 0, 0, 0 }, 0, sizeof("lquota_v2.user") - 1,
1677                 NULL, NULL },
1678
1679         /* lquota_v2.group, upgrade from old device */
1680         { "lquota_v2.group", { 0, 0, 0 }, 0, sizeof("lquota_v2.group") - 1,
1681                 NULL, NULL },
1682
1683         /* LAST_GROUP, upgrade from old device */
1684         { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1685                 OLF_SHOW_NAME, sizeof("LAST_GROUP") - 1, NULL, NULL },
1686
1687         /* committed batchid for cross-MDT operation */
1688         { "BATCHID", { FID_SEQ_LOCAL_FILE, BATCHID_COMMITTED_OID, 0 },
1689                 OLF_SHOW_NAME, sizeof("BATCHID") - 1, NULL, NULL },
1690
1691         /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1692          * and f_oid = index for their log files.  See lu_update_log{_dir}_fid()
1693          * for more details. */
1694
1695         /* update_log */
1696         { "update_log", { FID_SEQ_UPDATE_LOG, 0, 0 },
1697                 OLF_SHOW_NAME | OLF_IDX_IN_FID, sizeof("update_log") - 1,
1698                 NULL, NULL },
1699
1700         /* update_log_dir */
1701         { "update_log_dir", { FID_SEQ_UPDATE_LOG_DIR, 0, 0 },
1702                 OLF_SHOW_NAME | OLF_SCAN_SUBITEMS | OLF_IDX_IN_FID,
1703                 sizeof("update_log_dir") - 1,
1704                 osd_ios_general_scan, osd_ios_uld_fill },
1705
1706         /* lost+found */
1707         { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 },
1708                 OLF_SCAN_SUBITEMS, sizeof("lost+found") - 1,
1709                 osd_ios_general_scan, osd_ios_lf_fill },
1710
1711         { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1712 };
1713
1714 /* Add the new introduced files under .lustre/ in the list in the future. */
1715 static const struct osd_lf_map osd_dl_maps[] = {
1716         /* .lustre/fid */
1717         { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1718                 sizeof("fid") - 1, NULL, NULL },
1719
1720         /* .lustre/lost+found */
1721         { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1722                 sizeof("lost+found") - 1, NULL, NULL },
1723
1724         { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1725 };
1726
1727 struct osd_ios_item {
1728         struct list_head oii_list;
1729         struct dentry   *oii_dentry;
1730         scandir_t        oii_scandir;
1731         filldir_t        oii_filldir;
1732 };
1733
1734 struct osd_ios_filldir_buf {
1735 #ifdef HAVE_DIR_CONTEXT
1736         /* please keep it as first member */
1737         struct dir_context       ctx;
1738 #endif
1739         struct osd_thread_info  *oifb_info;
1740         struct osd_device       *oifb_dev;
1741         struct dentry           *oifb_dentry;
1742 };
1743
1744 static inline struct dentry *
1745 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1746 {
1747         struct dentry *dentry;
1748
1749         dentry = ll_lookup_one_len(name, parent, namelen);
1750         if (IS_ERR(dentry)) {
1751                 int rc = PTR_ERR(dentry);
1752
1753                 if (rc != -ENOENT)
1754                         CERROR("Fail to find %.*s in %.*s (%lu/%u): rc = %d\n",
1755                                namelen, name, parent->d_name.len,
1756                                parent->d_name.name, parent->d_inode->i_ino,
1757                                parent->d_inode->i_generation, rc);
1758
1759                 return dentry;
1760         }
1761
1762         if (dentry->d_inode == NULL) {
1763                 dput(dentry);
1764                 return ERR_PTR(-ENOENT);
1765         }
1766
1767         return dentry;
1768 }
1769
1770 static int
1771 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1772                  scandir_t scandir, filldir_t filldir)
1773 {
1774         struct osd_ios_item *item;
1775         ENTRY;
1776
1777         OBD_ALLOC_PTR(item);
1778         if (item == NULL)
1779                 RETURN(-ENOMEM);
1780
1781         INIT_LIST_HEAD(&item->oii_list);
1782         item->oii_dentry = dget(dentry);
1783         item->oii_scandir = scandir;
1784         item->oii_filldir = filldir;
1785         list_add_tail(&item->oii_list, &dev->od_ios_list);
1786
1787         RETURN(0);
1788 }
1789
1790 /**
1791  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1792  *
1793  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1794  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1795  * reference the inode, or fixed if it is missing or references another inode.
1796  */
1797 static int
1798 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1799                  struct inode *inode, const struct lu_fid *fid, int flags)
1800 {
1801         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
1802         struct osd_inode_id     *id     = &info->oti_id;
1803         struct osd_inode_id     *id2    = &info->oti_id2;
1804         struct osd_scrub        *scrub  = &dev->od_scrub;
1805         struct scrub_file       *sf     = &scrub->os_file;
1806         struct lu_fid            tfid;
1807         int                      rc;
1808         ENTRY;
1809
1810         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1811         if (rc != 0 && rc != -ENODATA) {
1812                 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1813                        "rc = %d\n", osd_name(dev), rc);
1814
1815                 RETURN(rc);
1816         }
1817
1818         osd_id_gen(id, inode->i_ino, inode->i_generation);
1819         if (rc == -ENODATA) {
1820                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
1821                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1822                 } else {
1823                         tfid = *fid;
1824                         if (flags & OLF_IDX_IN_FID) {
1825                                 LASSERT(dev->od_index >= 0);
1826
1827                                 tfid.f_oid = dev->od_index;
1828                         }
1829                 }
1830                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1831                 if (rc != 0) {
1832                         CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1833                               "scrub: rc = %d\n", osd_name(dev), rc);
1834
1835                         RETURN(rc);
1836                 }
1837         } else {
1838                 if (lma->lma_compat & LMAC_NOT_IN_OI)
1839                         RETURN(0);
1840
1841                 tfid = lma->lma_self_fid;
1842         }
1843
1844         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1845         if (rc != 0) {
1846                 if (rc != -ENOENT)
1847                         RETURN(rc);
1848
1849                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1850                                                DTO_INDEX_INSERT, true, 0);
1851                 if (rc > 0)
1852                         rc = 0;
1853
1854                 RETURN(rc);
1855         }
1856
1857         if (osd_id_eq_strict(id, id2))
1858                 RETURN(0);
1859
1860         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1861                 osd_scrub_file_reset(scrub,
1862                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1863                                      SF_INCONSISTENT);
1864                 rc = osd_scrub_file_store(scrub);
1865                 if (rc != 0)
1866                         RETURN(rc);
1867         }
1868
1869         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1870                                        DTO_INDEX_UPDATE, true, 0);
1871         if (rc > 0)
1872                 rc = 0;
1873
1874         RETURN(rc);
1875 }
1876
1877 /**
1878  * It scans the /lost+found, and for the OST-object (with filter_fid
1879  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1880  */
1881 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1882                            loff_t offset, __u64 ino, unsigned d_type)
1883 {
1884         struct osd_ios_filldir_buf *fill_buf = buf;
1885         struct osd_thread_info     *info     = fill_buf->oifb_info;
1886         struct osd_device          *dev      = fill_buf->oifb_dev;
1887         struct lu_fid              *fid      = &info->oti_fid;
1888         struct osd_scrub           *scrub    = &dev->od_scrub;
1889         struct dentry              *parent   = fill_buf->oifb_dentry;
1890         struct dentry              *child;
1891         struct inode               *dir      = parent->d_inode;
1892         struct inode               *inode;
1893         int                         rc;
1894         ENTRY;
1895
1896         /* skip any '.' started names */
1897         if (name[0] == '.')
1898                 RETURN(0);
1899
1900         scrub->os_lf_scanned++;
1901         child = osd_ios_lookup_one_len(name, parent, namelen);
1902         if (IS_ERR(child)) {
1903                 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1904                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
1905                 RETURN(0);
1906         }
1907
1908         inode = child->d_inode;
1909         if (S_ISDIR(inode->i_mode)) {
1910                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1911                                       osd_ios_lf_fill);
1912                 if (rc != 0)
1913                         CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1914                               "rc = %d\n", osd_name(dev), namelen, name, rc);
1915                 GOTO(put, rc);
1916         }
1917
1918         if (!S_ISREG(inode->i_mode))
1919                 GOTO(put, rc = 0);
1920
1921         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1922         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1923                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1924                 if (rc == 0) {
1925                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1926                                "/lost+found.\n", namelen, name, PFID(fid));
1927                         scrub->os_lf_repaired++;
1928                 } else {
1929                         CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1930                                DFID": rc = %d\n",
1931                                osd_name(dev), namelen, name, PFID(fid), rc);
1932                 }
1933         }
1934
1935         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1936          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1937          *      can process them in furtuer. */
1938
1939         GOTO(put, rc);
1940
1941 put:
1942         if (rc < 0)
1943                 scrub->os_lf_failed++;
1944         dput(child);
1945         /* skip the failure to make the scanning to continue. */
1946         return 0;
1947 }
1948
1949 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1950                                loff_t offset, __u64 ino, unsigned d_type)
1951 {
1952         struct osd_ios_filldir_buf *fill_buf = buf;
1953         struct osd_device          *dev      = fill_buf->oifb_dev;
1954         struct dentry              *child;
1955         int                         rc;
1956         ENTRY;
1957
1958         /* skip any '.' started names */
1959         if (name[0] == '.')
1960                 RETURN(0);
1961
1962         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1963         if (IS_ERR(child))
1964                 RETURN(PTR_ERR(child));
1965
1966         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1967                               NULL, 0);
1968         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1969                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1970                                       osd_ios_varfid_fill);
1971         dput(child);
1972
1973         RETURN(rc);
1974 }
1975
1976 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1977                            loff_t offset, __u64 ino, unsigned d_type)
1978 {
1979         struct osd_ios_filldir_buf *fill_buf = buf;
1980         struct osd_device          *dev      = fill_buf->oifb_dev;
1981         const struct osd_lf_map    *map;
1982         struct dentry              *child;
1983         int                         rc       = 0;
1984         ENTRY;
1985
1986         /* skip any '.' started names */
1987         if (name[0] == '.')
1988                 RETURN(0);
1989
1990         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
1991                 if (map->olm_namelen != namelen)
1992                         continue;
1993
1994                 if (strncmp(map->olm_name, name, namelen) == 0)
1995                         break;
1996         }
1997
1998         if (map->olm_name == NULL)
1999                 RETURN(0);
2000
2001         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2002         if (IS_ERR(child))
2003                 RETURN(PTR_ERR(child));
2004
2005         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2006                               &map->olm_fid, map->olm_flags);
2007         dput(child);
2008
2009         RETURN(rc);
2010 }
2011
2012 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
2013                             loff_t offset, __u64 ino, unsigned d_type)
2014 {
2015         struct osd_ios_filldir_buf *fill_buf = buf;
2016         struct dentry              *child;
2017         struct lu_fid               tfid;
2018         int                         rc       = 0;
2019         ENTRY;
2020
2021         /* skip any non-DFID format name */
2022         if (name[0] != '[')
2023                 RETURN(0);
2024
2025         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2026         if (IS_ERR(child))
2027                 RETURN(PTR_ERR(child));
2028
2029         /* skip the start '[' */
2030         sscanf(&name[1], SFID, RFID(&tfid));
2031         if (fid_is_sane(&tfid))
2032                 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
2033                                       child->d_inode, &tfid, 0);
2034         else
2035                 rc = -EIO;
2036         dput(child);
2037
2038         RETURN(rc);
2039 }
2040
2041 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
2042                              loff_t offset, __u64 ino, unsigned d_type)
2043 {
2044         struct osd_ios_filldir_buf *fill_buf = buf;
2045         struct osd_device          *dev      = fill_buf->oifb_dev;
2046         const struct osd_lf_map    *map;
2047         struct dentry              *child;
2048         int                         rc       = 0;
2049         ENTRY;
2050
2051         /* skip any '.' started names */
2052         if (name[0] == '.')
2053                 RETURN(0);
2054
2055         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2056                 if (map->olm_namelen != namelen)
2057                         continue;
2058
2059                 if (strncmp(map->olm_name, name, namelen) == 0)
2060                         break;
2061         }
2062
2063         if (map->olm_name == NULL)
2064                 RETURN(0);
2065
2066         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2067         if (IS_ERR(child))
2068                 RETURN(PTR_ERR(child));
2069
2070         if (!(map->olm_flags & OLF_NO_OI))
2071                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2072                                       &map->olm_fid, map->olm_flags);
2073         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2074                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2075                                       map->olm_filldir);
2076         dput(child);
2077
2078         RETURN(rc);
2079 }
2080
2081 static int
2082 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2083                      struct dentry *dentry, filldir_t filldir)
2084 {
2085         struct osd_ios_filldir_buf    buf   = {
2086 #ifdef HAVE_DIR_CONTEXT
2087                                                 .ctx.actor = filldir,
2088 #endif
2089                                                 .oifb_info = info,
2090                                                 .oifb_dev = dev,
2091                                                 .oifb_dentry = dentry };
2092         struct file                  *filp  = &info->oti_file;
2093         struct inode                 *inode = dentry->d_inode;
2094         const struct file_operations *fops  = inode->i_fop;
2095         int                           rc;
2096         ENTRY;
2097
2098         LASSERT(filldir != NULL);
2099
2100         filp->f_pos = 0;
2101         filp->f_path.dentry = dentry;
2102         filp->f_mode = FMODE_64BITHASH;
2103         filp->f_mapping = inode->i_mapping;
2104         filp->f_op = fops;
2105         filp->private_data = NULL;
2106         set_file_inode(filp, inode);
2107
2108 #ifdef HAVE_DIR_CONTEXT
2109         buf.ctx.pos = filp->f_pos;
2110         rc = fops->iterate(filp, &buf.ctx);
2111         filp->f_pos = buf.ctx.pos;
2112 #else
2113         rc = fops->readdir(filp, &buf, filldir);
2114 #endif
2115         fops->release(inode, filp);
2116
2117         RETURN(rc);
2118 }
2119
2120 static int
2121 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2122                   struct dentry *dentry, filldir_t filldir)
2123 {
2124         struct osd_scrub  *scrub  = &dev->od_scrub;
2125         struct scrub_file *sf     = &scrub->os_file;
2126         struct dentry     *child;
2127         int                rc;
2128         ENTRY;
2129
2130         /* It is existing MDT0 device. We only allow the case of object without
2131          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2132          * can generate IGIF mode FID for the object and related OI mapping. If
2133          * it is on other MDTs, then becuase file-level backup/restore, related
2134          * OI mapping may be invalid already, we do not know which is the right
2135          * FID for the object. We only allow IGIF objects to reside on the MDT0.
2136          *
2137          * XXX: For the case of object on non-MDT0 device with neither LMA nor
2138          *      "fid" xattr, then something crashed. We cannot re-generate the
2139          *      FID directly, instead, the OI scrub will scan the OI structure
2140          *      and try to re-generate the LMA from the OI mapping. But if the
2141          *      OI mapping crashed or lost also, then we have to give up under
2142          *      double failure cases. */
2143         scrub->os_convert_igif = 1;
2144         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2145                                        strlen(dot_lustre_name));
2146         if (IS_ERR(child)) {
2147                 rc = PTR_ERR(child);
2148                 if (rc == -ENOENT) {
2149                         /* It is 1.8 MDT device. */
2150                         if (!(sf->sf_flags & SF_UPGRADE)) {
2151                                 osd_scrub_file_reset(scrub,
2152                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2153                                         SF_UPGRADE);
2154                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2155                                 rc = osd_scrub_file_store(scrub);
2156                         } else {
2157                                 rc = 0;
2158                         }
2159                 }
2160         } else {
2161                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2162                  * so the client will get IGIF for the ".lustre" object when
2163                  * the MDT restart.
2164                  *
2165                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2166                  * it does not know whether there are some old clients cached
2167                  * the ".lustre" IGIF during the upgrading. Two choices:
2168                  *
2169                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2170                  *    It will allow the old connected clients to access the
2171                  *    ".lustre" with cached IGIF. But it will cause others
2172                  *    on the MDT failed to check "fid_is_dot_lustre()".
2173                  *
2174                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2175                  *    for ".lustre" in spite of whether there are some clients
2176                  *    cached the ".lustre" IGIF or not. It enables the check
2177                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
2178                  *    that the old connected clients cannot access the ".lustre"
2179                  *    with the cached IGIF.
2180                  *
2181                  * Usually, it is rare case for the old connected clients
2182                  * to access the ".lustre" with cached IGIF. So we prefer
2183                  * to the solution 2). */
2184                 rc = osd_ios_scan_one(info, dev, child->d_inode,
2185                                       &LU_DOT_LUSTRE_FID, 0);
2186                 if (rc == 0)
2187                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2188                                               osd_ios_dl_fill);
2189                 dput(child);
2190         }
2191
2192         RETURN(rc);
2193 }
2194
2195 static int
2196 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2197                      struct dentry *dentry, filldir_t filldir)
2198 {
2199         struct osd_scrub  *scrub  = &dev->od_scrub;
2200         struct scrub_file *sf     = &scrub->os_file;
2201         struct dentry     *child;
2202         int                rc;
2203         ENTRY;
2204
2205         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2206                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2207                 rc = osd_scrub_file_store(scrub);
2208                 if (rc != 0)
2209                         RETURN(rc);
2210         }
2211
2212         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2213         if (!IS_ERR(child)) {
2214                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2215                 dput(child);
2216         } else {
2217                 rc = PTR_ERR(child);
2218         }
2219
2220         if (rc != 0 && rc != -ENOENT)
2221                 RETURN(rc);
2222
2223         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2224         if (!IS_ERR(child)) {
2225                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2226                 dput(child);
2227         } else {
2228                 rc = PTR_ERR(child);
2229         }
2230
2231         if (rc == -ENOENT)
2232                 rc = 0;
2233
2234         RETURN(rc);
2235 }
2236
2237 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2238                                 struct osd_device *dev)
2239 {
2240         struct osd_ios_item     *item    = NULL;
2241         scandir_t                scandir = osd_ios_general_scan;
2242         filldir_t                filldir = osd_ios_root_fill;
2243         struct dentry           *dentry  = osd_sb(dev)->s_root;
2244         const struct osd_lf_map *map     = osd_lf_maps;
2245         int                      rc;
2246         ENTRY;
2247
2248         /* Lookup IGIF in OI by force for initial OI scrub. */
2249         dev->od_igif_inoi = 1;
2250
2251         while (1) {
2252                 rc = scandir(info, dev, dentry, filldir);
2253                 if (item != NULL) {
2254                         dput(item->oii_dentry);
2255                         OBD_FREE_PTR(item);
2256                 }
2257
2258                 if (rc != 0)
2259                         break;
2260
2261                 if (list_empty(&dev->od_ios_list))
2262                         break;
2263
2264                 item = list_entry(dev->od_ios_list.next,
2265                                   struct osd_ios_item, oii_list);
2266                 list_del_init(&item->oii_list);
2267
2268                 LASSERT(item->oii_scandir != NULL);
2269                 scandir = item->oii_scandir;
2270                 filldir = item->oii_filldir;
2271                 dentry = item->oii_dentry;
2272         }
2273
2274         while (!list_empty(&dev->od_ios_list)) {
2275                 item = list_entry(dev->od_ios_list.next,
2276                                   struct osd_ios_item, oii_list);
2277                 list_del_init(&item->oii_list);
2278                 dput(item->oii_dentry);
2279                 OBD_FREE_PTR(item);
2280         }
2281
2282         if (rc != 0)
2283                 RETURN(rc);
2284
2285         /* There maybe the case that the object has been removed, but its OI
2286          * mapping is still in the OI file, such as the "CATALOGS" after MDT
2287          * file-level backup/restore. So here cleanup the stale OI mappings. */
2288         while (map->olm_name != NULL) {
2289                 struct dentry *child;
2290
2291                 if (fid_is_zero(&map->olm_fid)) {
2292                         map++;
2293                         continue;
2294                 }
2295
2296                 child = osd_ios_lookup_one_len(map->olm_name,
2297                                                osd_sb(dev)->s_root,
2298                                                map->olm_namelen);
2299                 if (!IS_ERR(child))
2300                         dput(child);
2301                 else if (PTR_ERR(child) == -ENOENT)
2302                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2303                                                   NULL, DTO_INDEX_DELETE,
2304                                                   true, 0);
2305                 map++;
2306         }
2307
2308         RETURN(0);
2309 }
2310
2311 char *osd_lf_fid2name(const struct lu_fid *fid)
2312 {
2313         const struct osd_lf_map *map = osd_lf_maps;
2314
2315         while (map->olm_name != NULL) {
2316                 if (!lu_fid_eq(fid, &map->olm_fid)) {
2317                         map++;
2318                         continue;
2319                 }
2320
2321                 if (map->olm_flags & OLF_SHOW_NAME)
2322                         return map->olm_name;
2323                 else
2324                         return "";
2325         }
2326
2327         return NULL;
2328 }
2329
2330 /* OI scrub start/stop */
2331
2332 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2333 {
2334         struct osd_scrub     *scrub  = &dev->od_scrub;
2335         struct ptlrpc_thread *thread = &scrub->os_thread;
2336         struct l_wait_info    lwi    = { 0 };
2337         struct task_struct   *task;
2338         int                   rc;
2339         ENTRY;
2340
2341         /* os_lock: sync status between stop and scrub thread */
2342         spin_lock(&scrub->os_lock);
2343
2344 again:
2345         if (thread_is_running(thread)) {
2346                 spin_unlock(&scrub->os_lock);
2347                 if (!(scrub->os_file.sf_flags & SF_AUTO) ||
2348                      (flags & (SS_AUTO_FULL | SS_AUTO_PARTIAL)))
2349                         RETURN(-EALREADY);
2350
2351                 osd_scrub_join(dev, flags, false);
2352                 spin_lock(&scrub->os_lock);
2353                 if (!thread_is_running(thread))
2354                         goto again;
2355
2356                 spin_unlock(&scrub->os_lock);
2357                 RETURN(0);
2358         }
2359
2360         if (unlikely(thread_is_stopping(thread))) {
2361                 spin_unlock(&scrub->os_lock);
2362                 l_wait_event(thread->t_ctl_waitq,
2363                              thread_is_stopped(thread),
2364                              &lwi);
2365                 spin_lock(&scrub->os_lock);
2366                 goto again;
2367         }
2368         spin_unlock(&scrub->os_lock);
2369
2370         if (scrub->os_file.sf_status == SS_COMPLETED) {
2371                 if (!(flags & SS_SET_FAILOUT))
2372                         flags |= SS_CLEAR_FAILOUT;
2373
2374                 if (!(flags & SS_SET_DRYRUN))
2375                         flags |= SS_CLEAR_DRYRUN;
2376
2377                 flags |= SS_RESET;
2378         }
2379
2380         scrub->os_start_flags = flags;
2381         thread_set_flags(thread, 0);
2382         task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2383         if (IS_ERR(task)) {
2384                 rc = PTR_ERR(task);
2385                 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2386                        osd_scrub2name(scrub), rc);
2387                 RETURN(rc);
2388         }
2389
2390         l_wait_event(thread->t_ctl_waitq,
2391                      thread_is_running(thread) || thread_is_stopped(thread),
2392                      &lwi);
2393
2394         RETURN(0);
2395 }
2396
2397 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2398 {
2399         int rc;
2400         ENTRY;
2401
2402         /* od_otable_mutex: prevent curcurrent start/stop */
2403         mutex_lock(&dev->od_otable_mutex);
2404         rc = do_osd_scrub_start(dev, flags);
2405         mutex_unlock(&dev->od_otable_mutex);
2406
2407         RETURN(rc == -EALREADY ? 0 : rc);
2408 }
2409
2410 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2411 {
2412         struct ptlrpc_thread *thread = &scrub->os_thread;
2413         struct l_wait_info    lwi    = { 0 };
2414
2415         /* os_lock: sync status between stop and scrub thread */
2416         spin_lock(&scrub->os_lock);
2417         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2418                 thread_set_flags(thread, SVC_STOPPING);
2419                 spin_unlock(&scrub->os_lock);
2420                 wake_up_all(&thread->t_ctl_waitq);
2421                 l_wait_event(thread->t_ctl_waitq,
2422                              thread_is_stopped(thread),
2423                              &lwi);
2424                 /* Do not skip the last lock/unlock, which can guarantee that
2425                  * the caller cannot return until the OI scrub thread exit. */
2426                 spin_lock(&scrub->os_lock);
2427         }
2428         spin_unlock(&scrub->os_lock);
2429 }
2430
2431 static void osd_scrub_stop(struct osd_device *dev)
2432 {
2433         /* od_otable_mutex: prevent curcurrent start/stop */
2434         mutex_lock(&dev->od_otable_mutex);
2435         dev->od_scrub.os_paused = 1;
2436         do_osd_scrub_stop(&dev->od_scrub);
2437         mutex_unlock(&dev->od_otable_mutex);
2438 }
2439
2440 /* OI scrub setup/cleanup */
2441
2442 static const char osd_scrub_name[] = "OI_scrub";
2443
2444 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2445 {
2446         struct osd_thread_info     *info   = osd_oti_get(env);
2447         struct osd_scrub           *scrub  = &dev->od_scrub;
2448         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
2449         struct scrub_file          *sf     = &scrub->os_file;
2450         struct super_block         *sb     = osd_sb(dev);
2451         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
2452         struct lvfs_run_ctxt        saved;
2453         struct file                *filp;
2454         struct inode               *inode;
2455         struct lu_fid              *fid    = &info->oti_fid;
2456         int                         dirty  = 0;
2457         int                         rc     = 0;
2458         ENTRY;
2459
2460         memset(scrub, 0, sizeof(*scrub));
2461         OBD_SET_CTXT_MAGIC(ctxt);
2462         ctxt->pwdmnt = dev->od_mnt;
2463         ctxt->pwd = dev->od_mnt->mnt_root;
2464         ctxt->fs = get_ds();
2465
2466         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2467         init_rwsem(&scrub->os_rwsem);
2468         spin_lock_init(&scrub->os_lock);
2469         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2470
2471         push_ctxt(&saved, ctxt);
2472         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2473         if (IS_ERR(filp)) {
2474                 pop_ctxt(&saved, ctxt);
2475                 RETURN(PTR_ERR(filp));
2476         }
2477
2478         inode = filp->f_path.dentry->d_inode;
2479         /* 'What the @fid is' is not imporatant, because the object
2480          * has no OI mapping, and only is visible inside the OSD.*/
2481         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2482         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2483         if (rc != 0) {
2484                 filp_close(filp, NULL);
2485                 pop_ctxt(&saved, ctxt);
2486                 RETURN(rc);
2487         }
2488
2489         scrub->os_inode = igrab(inode);
2490         filp_close(filp, NULL);
2491         pop_ctxt(&saved, ctxt);
2492
2493         rc = osd_scrub_file_load(scrub);
2494         if (rc == -ENOENT) {
2495                 osd_scrub_file_init(scrub, es->s_uuid);
2496                 /* If the "/O" dir does not exist when mount (indicated by
2497                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2498                  * then it is quite probably that the device is a new one,
2499                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2500                  *
2501                  * For the rare case that "/O" and "OI_scrub" both lost on
2502                  * an old device, it can be found and cleared later.
2503                  *
2504                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2505                  * need to check "filter_fid_old" and to convert it to
2506                  * "filter_fid" for each object, and all the IGIF should
2507                  * have their FID mapping in OI files already. */
2508                 if (dev->od_maybe_new)
2509                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2510                 dirty = 1;
2511         } else if (rc != 0) {
2512                 GOTO(cleanup_inode, rc);
2513         } else {
2514                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2515                         struct obd_uuid *old_uuid;
2516                         struct obd_uuid *new_uuid;
2517
2518                         OBD_ALLOC_PTR(old_uuid);
2519                         OBD_ALLOC_PTR(new_uuid);
2520                         if (old_uuid == NULL || new_uuid == NULL) {
2521                                 CERROR("%.16s: UUID has been changed, but"
2522                                        "failed to allocate RAM for report\n",
2523                                        LDISKFS_SB(sb)->s_es->s_volume_name);
2524                         } else {
2525                                 class_uuid_unparse(sf->sf_uuid, old_uuid);
2526                                 class_uuid_unparse(es->s_uuid, new_uuid);
2527                                 CERROR("%.16s: UUID has been changed from "
2528                                        "%s to %s\n",
2529                                        LDISKFS_SB(sb)->s_es->s_volume_name,
2530                                        old_uuid->uuid, new_uuid->uuid);
2531                         }
2532                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2533                         dirty = 1;
2534                         if (old_uuid != NULL)
2535                                 OBD_FREE_PTR(old_uuid);
2536                         if (new_uuid != NULL)
2537                                 OBD_FREE_PTR(new_uuid);
2538                 } else if (sf->sf_status == SS_SCANNING) {
2539                         sf->sf_status = SS_CRASHED;
2540                         dirty = 1;
2541                 }
2542         }
2543
2544         if (sf->sf_pos_last_checkpoint != 0)
2545                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2546         else
2547                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2548
2549         if (dirty != 0) {
2550                 rc = osd_scrub_file_store(scrub);
2551                 if (rc != 0)
2552                         GOTO(cleanup_inode, rc);
2553         }
2554
2555         /* Initialize OI files. */
2556         rc = osd_oi_init(info, dev);
2557         if (rc < 0)
2558                 GOTO(cleanup_inode, rc);
2559
2560         rc = osd_initial_OI_scrub(info, dev);
2561         if (rc != 0)
2562                 GOTO(cleanup_oi, rc);
2563
2564         if (sf->sf_flags & SF_UPGRADE ||
2565             !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2566               sf->sf_success_count > 0)) {
2567                 dev->od_igif_inoi = 0;
2568                 dev->od_check_ff = dev->od_is_ost;
2569         } else {
2570                 dev->od_igif_inoi = 1;
2571                 dev->od_check_ff = 0;
2572         }
2573
2574         if (sf->sf_flags & SF_INCONSISTENT)
2575                 /* The 'od_igif_inoi' will be set under the
2576                  * following cases:
2577                  * 1) new created system, or
2578                  * 2) restored from file-level backup, or
2579                  * 3) the upgrading completed.
2580                  *
2581                  * The 'od_igif_inoi' may be cleared by OI scrub
2582                  * later if found that the system is upgrading. */
2583                 dev->od_igif_inoi = 1;
2584
2585         if (!dev->od_noscrub &&
2586             ((sf->sf_status == SS_PAUSED) ||
2587              (sf->sf_status == SS_CRASHED &&
2588               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2589                               SF_UPGRADE | SF_AUTO)) ||
2590              (sf->sf_status == SS_INIT &&
2591               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2592                               SF_UPGRADE))))
2593                 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2594
2595         if (rc != 0)
2596                 GOTO(cleanup_oi, rc);
2597
2598         /* it is possible that dcache entries may keep objects after they are
2599          * deleted by OSD. While it looks safe this can cause object data to
2600          * stay until umount causing failures in tests calculating free space,
2601          * e.g. replay-ost-single. Since those dcache entries are not used
2602          * anymore let's just free them after use here */
2603         shrink_dcache_sb(sb);
2604
2605         RETURN(0);
2606 cleanup_oi:
2607         osd_oi_fini(info, dev);
2608 cleanup_inode:
2609         iput(scrub->os_inode);
2610         scrub->os_inode = NULL;
2611
2612         return rc;
2613 }
2614
2615 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2616 {
2617         struct osd_scrub *scrub = &dev->od_scrub;
2618
2619         LASSERT(dev->od_otable_it == NULL);
2620
2621         if (scrub->os_inode != NULL) {
2622                 osd_scrub_stop(dev);
2623                 iput(scrub->os_inode);
2624                 scrub->os_inode = NULL;
2625         }
2626         if (dev->od_oi_table != NULL)
2627                 osd_oi_fini(osd_oti_get(env), dev);
2628 }
2629
2630 /* object table based iteration APIs */
2631
2632 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2633                                        struct dt_object *dt, __u32 attr)
2634 {
2635         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2636         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2637         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2638         struct osd_scrub       *scrub = &dev->od_scrub;
2639         struct osd_otable_it   *it;
2640         __u32                   start = 0;
2641         int                     rc;
2642         ENTRY;
2643
2644         /* od_otable_mutex: prevent curcurrent init/fini */
2645         mutex_lock(&dev->od_otable_mutex);
2646         if (dev->od_otable_it != NULL)
2647                 GOTO(out, it = ERR_PTR(-EALREADY));
2648
2649         OBD_ALLOC_PTR(it);
2650         if (it == NULL)
2651                 GOTO(out, it = ERR_PTR(-ENOMEM));
2652
2653         dev->od_otable_it = it;
2654         it->ooi_dev = dev;
2655         it->ooi_cache.ooc_consumer_idx = -1;
2656         if (flags & DOIF_OUTUSED)
2657                 it->ooi_used_outside = 1;
2658
2659         if (flags & DOIF_RESET)
2660                 start |= SS_RESET;
2661
2662         if (valid & DOIV_ERROR_HANDLE) {
2663                 if (flags & DOIF_FAILOUT)
2664                         start |= SS_SET_FAILOUT;
2665                 else
2666                         start |= SS_CLEAR_FAILOUT;
2667         }
2668
2669         if (valid & DOIV_DRYRUN) {
2670                 if (flags & DOIF_DRYRUN)
2671                         start |= SS_SET_DRYRUN;
2672                 else
2673                         start |= SS_CLEAR_DRYRUN;
2674         }
2675
2676         rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2677         if (rc < 0 && rc != -EALREADY) {
2678                 dev->od_otable_it = NULL;
2679                 OBD_FREE_PTR(it);
2680                 GOTO(out, it = ERR_PTR(rc));
2681         }
2682
2683         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2684
2685         GOTO(out, it);
2686
2687 out:
2688         mutex_unlock(&dev->od_otable_mutex);
2689         return (struct dt_it *)it;
2690 }
2691
2692 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2693 {
2694         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2695         struct osd_device    *dev = it->ooi_dev;
2696
2697         /* od_otable_mutex: prevent curcurrent init/fini */
2698         mutex_lock(&dev->od_otable_mutex);
2699         do_osd_scrub_stop(&dev->od_scrub);
2700         LASSERT(dev->od_otable_it == it);
2701
2702         dev->od_otable_it = NULL;
2703         mutex_unlock(&dev->od_otable_mutex);
2704         OBD_FREE_PTR(it);
2705 }
2706
2707 static int osd_otable_it_get(const struct lu_env *env,
2708                              struct dt_it *di, const struct dt_key *key)
2709 {
2710         return 0;
2711 }
2712
2713 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2714 {
2715 }
2716
2717 static inline int
2718 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2719 {
2720         spin_lock(&scrub->os_lock);
2721         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2722             scrub->os_waiting ||
2723             !thread_is_running(&scrub->os_thread))
2724                 it->ooi_waiting = 0;
2725         else
2726                 it->ooi_waiting = 1;
2727         spin_unlock(&scrub->os_lock);
2728
2729         return !it->ooi_waiting;
2730 }
2731
2732 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2733 {
2734         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2735         struct osd_device       *dev    = it->ooi_dev;
2736         struct osd_scrub        *scrub  = &dev->od_scrub;
2737         struct osd_otable_cache *ooc    = &it->ooi_cache;
2738         struct ptlrpc_thread    *thread = &scrub->os_thread;
2739         struct l_wait_info       lwi    = { 0 };
2740         int                      rc;
2741         ENTRY;
2742
2743         LASSERT(it->ooi_user_ready);
2744
2745 again:
2746         if (!thread_is_running(thread) && !it->ooi_used_outside)
2747                 RETURN(1);
2748
2749         if (ooc->ooc_cached_items > 0) {
2750                 ooc->ooc_cached_items--;
2751                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2752                                         ~OSD_OTABLE_IT_CACHE_MASK;
2753                 RETURN(0);
2754         }
2755
2756         if (it->ooi_all_cached) {
2757                 l_wait_event(thread->t_ctl_waitq,
2758                              !thread_is_running(thread),
2759                              &lwi);
2760                 RETURN(1);
2761         }
2762
2763         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2764                 spin_lock(&scrub->os_lock);
2765                 scrub->os_waiting = 0;
2766                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2767                 spin_unlock(&scrub->os_lock);
2768         }
2769
2770         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2771                 l_wait_event(thread->t_ctl_waitq,
2772                              osd_otable_it_wakeup(scrub, it),
2773                              &lwi);
2774
2775         if (!thread_is_running(thread) && !it->ooi_used_outside)
2776                 RETURN(1);
2777
2778         rc = osd_otable_it_preload(env, it);
2779         if (rc >= 0)
2780                 goto again;
2781
2782         RETURN(rc);
2783 }
2784
2785 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2786                                         const struct dt_it *di)
2787 {
2788         return NULL;
2789 }
2790
2791 static int osd_otable_it_key_size(const struct lu_env *env,
2792                                   const struct dt_it *di)
2793 {
2794         return sizeof(__u64);
2795 }
2796
2797 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2798                              struct dt_rec *rec, __u32 attr)
2799 {
2800         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2801         struct osd_otable_cache *ooc = &it->ooi_cache;
2802
2803         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2804
2805         /* Filter out Invald FID already. */
2806         LASSERTF(fid_is_sane((struct lu_fid *)rec),
2807                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2808                  PFID((struct lu_fid *)rec),
2809                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2810
2811         return 0;
2812 }
2813
2814 static __u64 osd_otable_it_store(const struct lu_env *env,
2815                                  const struct dt_it *di)
2816 {
2817         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2818         struct osd_otable_cache *ooc = &it->ooi_cache;
2819         __u64                    hash;
2820
2821         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2822                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2823         else
2824                 hash = ooc->ooc_pos_preload;
2825         return hash;
2826 }
2827
2828 /**
2829  * Set the OSD layer iteration start position as the specified hash.
2830  */
2831 static int osd_otable_it_load(const struct lu_env *env,
2832                               const struct dt_it *di, __u64 hash)
2833 {
2834         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
2835         struct osd_device       *dev   = it->ooi_dev;
2836         struct osd_otable_cache *ooc   = &it->ooi_cache;
2837         struct osd_scrub        *scrub = &dev->od_scrub;
2838         int                      rc;
2839         ENTRY;
2840
2841         /* Forbid to set iteration position after iteration started. */
2842         if (it->ooi_user_ready)
2843                 RETURN(-EPERM);
2844
2845         if (hash > OSD_OTABLE_MAX_HASH)
2846                 hash = OSD_OTABLE_MAX_HASH;
2847
2848         ooc->ooc_pos_preload = hash;
2849         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2850                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2851
2852         it->ooi_user_ready = 1;
2853         if (!scrub->os_full_speed)
2854                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2855
2856         /* Unplug OSD layer iteration by the first next() call. */
2857         rc = osd_otable_it_next(env, (struct dt_it *)it);
2858
2859         RETURN(rc);
2860 }
2861
2862 static int osd_otable_it_key_rec(const struct lu_env *env,
2863                                  const struct dt_it *di, void *key_rec)
2864 {
2865         return 0;
2866 }
2867
2868 const struct dt_index_operations osd_otable_ops = {
2869         .dio_it = {
2870                 .init     = osd_otable_it_init,
2871                 .fini     = osd_otable_it_fini,
2872                 .get      = osd_otable_it_get,
2873                 .put      = osd_otable_it_put,
2874                 .next     = osd_otable_it_next,
2875                 .key      = osd_otable_it_key,
2876                 .key_size = osd_otable_it_key_size,
2877                 .rec      = osd_otable_it_rec,
2878                 .store    = osd_otable_it_store,
2879                 .load     = osd_otable_it_load,
2880                 .key_rec  = osd_otable_it_key_rec,
2881         }
2882 };
2883
2884 /* high priority inconsistent items list APIs */
2885
2886 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL  60
2887
2888 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2889                    int insert)
2890 {
2891         struct osd_inconsistent_item *oii;
2892         struct osd_scrub             *scrub  = &dev->od_scrub;
2893         struct ptlrpc_thread         *thread = &scrub->os_thread;
2894         int                           wakeup = 0;
2895         ENTRY;
2896
2897         OBD_ALLOC_PTR(oii);
2898         if (unlikely(oii == NULL))
2899                 RETURN(-ENOMEM);
2900
2901         INIT_LIST_HEAD(&oii->oii_list);
2902         oii->oii_cache = *oic;
2903         oii->oii_insert = insert;
2904
2905         if (scrub->os_partial_scan) {
2906                 __u64 now = cfs_time_current_sec();
2907
2908                 /* If there haven't been errors in a long time,
2909                  * decay old count until either the errors are
2910                  * gone or we reach the current interval. */
2911                 while (unlikely(scrub->os_bad_oimap_count > 0 &&
2912                                 scrub->os_bad_oimap_time +
2913                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
2914                         scrub->os_bad_oimap_count >>= 1;
2915                         scrub->os_bad_oimap_time +=
2916                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
2917                 }
2918
2919                 scrub->os_bad_oimap_time = now;
2920                 if (++scrub->os_bad_oimap_count >
2921                     dev->od_full_scrub_threshold_rate)
2922                         scrub->os_full_scrub = 1;
2923         }
2924
2925         spin_lock(&scrub->os_lock);
2926         if (unlikely(!thread_is_running(thread))) {
2927                 spin_unlock(&scrub->os_lock);
2928                 OBD_FREE_PTR(oii);
2929                 RETURN(-EAGAIN);
2930         }
2931
2932         if (list_empty(&scrub->os_inconsistent_items))
2933                 wakeup = 1;
2934         list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2935         spin_unlock(&scrub->os_lock);
2936
2937         if (wakeup != 0)
2938                 wake_up_all(&thread->t_ctl_waitq);
2939
2940         RETURN(0);
2941 }
2942
2943 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2944                    struct osd_inode_id *id)
2945 {
2946         struct osd_scrub             *scrub = &dev->od_scrub;
2947         struct osd_inconsistent_item *oii;
2948         ENTRY;
2949
2950         spin_lock(&scrub->os_lock);
2951         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2952                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2953                         *id = oii->oii_cache.oic_lid;
2954                         spin_unlock(&scrub->os_lock);
2955                         RETURN(0);
2956                 }
2957         }
2958         spin_unlock(&scrub->os_lock);
2959
2960         RETURN(-ENOENT);
2961 }
2962
2963 /* OI scrub dump */
2964
2965 static const char *scrub_status_names[] = {
2966         "init",
2967         "scanning",
2968         "completed",
2969         "failed",
2970         "stopped",
2971         "paused",
2972         "crashed",
2973         NULL
2974 };
2975
2976 static const char *scrub_flags_names[] = {
2977         "recreated",
2978         "inconsistent",
2979         "auto",
2980         "upgrade",
2981         NULL
2982 };
2983
2984 static const char *scrub_param_names[] = {
2985         "failout",
2986         "dryrun",
2987         NULL
2988 };
2989
2990 static int scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
2991                            const char *prefix)
2992 {
2993         int flag;
2994         int rc;
2995         int i;
2996
2997         rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2998         if (rc < 0)
2999                 return rc;
3000
3001         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
3002                 if (flag & bits) {
3003                         bits &= ~flag;
3004                         rc = seq_printf(m, "%s%c", names[i],
3005                                         bits != 0 ? ',' : '\n');
3006                         if (rc < 0)
3007                                 return rc;
3008                 }
3009         }
3010         return 0;
3011 }
3012
3013 static int scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
3014 {
3015         int rc;
3016
3017         if (time != 0)
3018                 rc = seq_printf(m, "%s: "LPU64" seconds\n", prefix,
3019                               cfs_time_current_sec() - time);
3020         else
3021                 rc = seq_printf(m, "%s: N/A\n", prefix);
3022         return rc;
3023 }
3024
3025 static int scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
3026 {
3027         int rc;
3028
3029         if (pos != 0)
3030                 rc = seq_printf(m, "%s: "LPU64"\n", prefix, pos);
3031         else
3032                 rc = seq_printf(m, "%s: N/A\n", prefix);
3033         return rc;
3034 }
3035
3036 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
3037 {
3038         struct osd_scrub  *scrub   = &dev->od_scrub;
3039         struct scrub_file *sf      = &scrub->os_file;
3040         __u64              checked;
3041         __u64              speed;
3042         int                rc;
3043
3044         down_read(&scrub->os_rwsem);
3045         rc = seq_printf(m, "name: OI_scrub\n"
3046                         "magic: 0x%x\n"
3047                         "oi_files: %d\n"
3048                         "status: %s\n",
3049                         sf->sf_magic, (int)sf->sf_oi_count,
3050                         scrub_status_names[sf->sf_status]);
3051         if (rc < 0)
3052                 goto out;
3053
3054         rc = scrub_bits_dump(m, sf->sf_flags, scrub_flags_names,
3055                              "flags");
3056         if (rc < 0)
3057                 goto out;
3058
3059         rc = scrub_bits_dump(m, sf->sf_param, scrub_param_names,
3060                              "param");
3061         if (rc < 0)
3062                 goto out;
3063
3064         rc = scrub_time_dump(m, sf->sf_time_last_complete,
3065                              "time_since_last_completed");
3066         if (rc < 0)
3067                 goto out;
3068
3069         rc = scrub_time_dump(m, sf->sf_time_latest_start,
3070                              "time_since_latest_start");
3071         if (rc < 0)
3072                 goto out;
3073
3074         rc = scrub_time_dump(m, sf->sf_time_last_checkpoint,
3075                              "time_since_last_checkpoint");
3076         if (rc < 0)
3077                 goto out;
3078
3079         rc = scrub_pos_dump(m, sf->sf_pos_latest_start,
3080                             "latest_start_position");
3081         if (rc < 0)
3082                 goto out;
3083
3084         rc = scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
3085                             "last_checkpoint_position");
3086         if (rc < 0)
3087                 goto out;
3088
3089         rc = scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
3090                             "first_failure_position");
3091         if (rc < 0)
3092                 goto out;
3093
3094         checked = sf->sf_items_checked + scrub->os_new_checked;
3095         rc = seq_printf(m, "checked: "LPU64"\n"
3096                       "updated: "LPU64"\n"
3097                       "failed: "LPU64"\n"
3098                       "prior_updated: "LPU64"\n"
3099                       "noscrub: "LPU64"\n"
3100                       "igif: "LPU64"\n"
3101                       "success_count: %u\n",
3102                       checked, sf->sf_items_updated, sf->sf_items_failed,
3103                       sf->sf_items_updated_prior, sf->sf_items_noscrub,
3104                       sf->sf_items_igif, sf->sf_success_count);
3105         if (rc < 0)
3106                 goto out;
3107
3108         speed = checked;
3109         if (thread_is_running(&scrub->os_thread)) {
3110                 cfs_duration_t duration = cfs_time_current() -
3111                                           scrub->os_time_last_checkpoint;
3112                 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
3113                                                      MSEC_PER_SEC);
3114                 __u32 rtime = sf->sf_run_time +
3115                               cfs_duration_sec(duration + HALF_SEC);
3116
3117                 if (duration != 0)
3118                         do_div(new_checked, duration);
3119                 if (rtime != 0)
3120                         do_div(speed, rtime);
3121                 rc = seq_printf(m, "run_time: %u seconds\n"
3122                               "average_speed: "LPU64" objects/sec\n"
3123                               "real-time_speed: "LPU64" objects/sec\n"
3124                               "current_position: %u\n"
3125                               "lf_scanned: "LPU64"\n"
3126                               "lf_repaired: "LPU64"\n"
3127                               "lf_failed: "LPU64"\n",
3128                               rtime, speed, new_checked, scrub->os_pos_current,
3129                               scrub->os_lf_scanned, scrub->os_lf_repaired,
3130                               scrub->os_lf_failed);
3131         } else {
3132                 if (sf->sf_run_time != 0)
3133                         do_div(speed, sf->sf_run_time);
3134                 rc = seq_printf(m, "run_time: %u seconds\n"
3135                               "average_speed: "LPU64" objects/sec\n"
3136                               "real-time_speed: N/A\n"
3137                               "current_position: N/A\n"
3138                               "lf_scanned: "LPU64"\n"
3139                               "lf_repaired: "LPU64"\n"
3140                               "lf_failed: "LPU64"\n",
3141                               sf->sf_run_time, speed, scrub->os_lf_scanned,
3142                               scrub->os_lf_repaired, scrub->os_lf_failed);
3143         }
3144
3145 out:
3146         up_read(&scrub->os_rwsem);
3147         return (rc < 0 ? -ENOSPC : 0);
3148 }