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, bool *exist)
 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, exist);
 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                           rc;
 471         bool                          converted = false;
 472         bool                          exist     = 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                 switch (val) {
 553                 case SCRUB_NEXT_NOLMA:
 554                         sf->sf_flags |= SF_UPGRADE;
 555                         if (!(sf->sf_param & SP_DRYRUN)) {
 556                                 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
 557                                 if (rc != 0)
 558                                         GOTO(out, rc);
 559                         }
 560
 561                         if (!(sf->sf_flags & SF_INCONSISTENT))
 562                                 dev->od_igif_inoi = 0;
 563                         break;
 564                 case SCRUB_NEXT_OSTOBJ:
 565                         sf->sf_flags |= SF_INCONSISTENT;
 566                 case SCRUB_NEXT_OSTOBJ_OLD:
 567                         break;
 568                 default:
 569                         break;
 570                 }
 571         } else if (osd_id_eq(lid, lid2)) {
 572                 if (converted)
 573                         sf->sf_items_updated++;
 574
 575                 GOTO(out, rc = 0);
 576         } else {
 577                 if (!scrub->os_partial_scan)
 578                         scrub->os_full_speed = 1;
 579
 580                 sf->sf_flags |= SF_INCONSISTENT;
 581
 582                 /* XXX: If the device is restored from file-level backup, then
 583                  *      some IGIFs may have been already in OI files, and some
 584                  *      may be not yet. Means upgrading from 1.8 may be partly
 585                  *      processed, but some clients may hold some immobilized
 586                  *      IGIFs, and use them to access related objects. Under
 587                  *      such case, OSD does not know whether an given IGIF has
 588                  *      been processed or to be processed, and it also cannot
 589                  *      generate local ino#/gen# directly from the immobilized
 590                  *      IGIF because of the backup/restore. Then force OSD to
 591                  *      lookup the given IGIF in OI files, and if no entry,
 592                  *      then ask the client to retry after upgrading completed.
 593                  *      No better choice. */
 594                 dev->od_igif_inoi = 1;
 595         }
 596
 597         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
 598                         (val == SCRUB_NEXT_OSTOBJ ||
 599                          val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0,
 600                         &exist);
 601         if (rc == 0) {
 602                 if (scrub->os_in_prior)
 603                         sf->sf_items_updated_prior++;
 604                 else
 605                         sf->sf_items_updated++;
 606
 607                 if (ops == DTO_INDEX_INSERT && val == 0 && !exist) {
 608                         int idx = osd_oi_fid2idx(dev, fid);
 609
 610                         sf->sf_flags |= SF_RECREATED;
 611                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 612                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 613                 }
 614         }
 615
 616         GOTO(out, rc);
 617
 618 out:
 619         if (rc < 0) {
 620                 sf->sf_items_failed++;
 621                 if (sf->sf_pos_first_inconsistent == 0 ||
 622                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 623                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 624         } else {
 625                 rc = 0;
 626         }
 627
 628         /* There may be conflict unlink during the OI scrub,
 629          * if happend, then remove the new added OI mapping. */
 630         if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
 631             unlikely(ldiskfs_test_inode_state(inode,
 632                                               LDISKFS_STATE_LUSTRE_DESTROY)))
 633                 osd_scrub_refresh_mapping(info, dev, fid, lid,
 634                                 DTO_INDEX_DELETE, false,
 635                                 (val == SCRUB_NEXT_OSTOBJ ||
 636                                  val == SCRUB_NEXT_OSTOBJ_OLD) ?
 637                                 OI_KNOWN_ON_OST : 0, NULL);
 638         up_write(&scrub->os_rwsem);
 639
 640         if (inode != NULL && !IS_ERR(inode))
 641                 iput(inode);
 642
 643         if (oii != NULL) {
 644                 LASSERT(list_empty(&oii->oii_list));
 645
 646                 OBD_FREE_PTR(oii);
 647         }
 648
 649         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 650 }
 651
 652 /* OI scrub APIs */
 653
 654 static int osd_scrub_prep(struct osd_device *dev)
 655 {
 656         struct osd_scrub     *scrub  = &dev->od_scrub;
 657         struct ptlrpc_thread *thread = &scrub->os_thread;
 658         struct scrub_file    *sf     = &scrub->os_file;
 659         __u32                 flags  = scrub->os_start_flags;
 660         int                   rc;
 661         bool                  drop_dryrun = false;
 662         ENTRY;
 663
 664         CDEBUG(D_LFSCK, "%.16s: OI scrub prep, flags = 0x%x\n",
 665                osd_scrub2name(scrub), flags);
 666
 667         down_write(&scrub->os_rwsem);
 668         if (flags & SS_SET_FAILOUT)
 669                 sf->sf_param |= SP_FAILOUT;
 670         else if (flags & SS_CLEAR_FAILOUT)
 671                 sf->sf_param &= ~SP_FAILOUT;
 672
 673         if (flags & SS_SET_DRYRUN) {
 674                 sf->sf_param |= SP_DRYRUN;
 675         } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
 676                 sf->sf_param &= ~SP_DRYRUN;
 677                 drop_dryrun = true;
 678         }
 679
 680         if (flags & SS_RESET)
 681                 osd_scrub_file_reset(scrub,
 682                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
 683
 684         if (flags & SS_AUTO_FULL) {
 685                 scrub->os_full_speed = 1;
 686                 scrub->os_partial_scan = 0;
 687                 sf->sf_flags |= SF_AUTO;
 688         } else if (flags & SS_AUTO_PARTIAL) {
 689                 scrub->os_full_speed = 0;
 690                 scrub->os_partial_scan = 1;
 691                 sf->sf_flags |= SF_AUTO;
 692         } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
 693                                    SF_UPGRADE)) {
 694                 scrub->os_full_speed = 1;
 695                 scrub->os_partial_scan = 0;
 696         } else {
 697                 scrub->os_full_speed = 0;
 698                 scrub->os_partial_scan = 0;
 699         }
 700
 701         spin_lock(&scrub->os_lock);
 702         scrub->os_in_prior = 0;
 703         scrub->os_waiting = 0;
 704         scrub->os_paused = 0;
 705         scrub->os_in_join = 0;
 706         scrub->os_full_scrub = 0;
 707         spin_unlock(&scrub->os_lock);
 708         scrub->os_new_checked = 0;
 709         if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
 710                 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
 711         else if (sf->sf_pos_last_checkpoint != 0)
 712                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 713         else
 714                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 715
 716         scrub->os_pos_current = sf->sf_pos_latest_start;
 717         sf->sf_status = SS_SCANNING;
 718         sf->sf_time_latest_start = cfs_time_current_sec();
 719         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 720         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
 721         rc = osd_scrub_file_store(scrub);
 722         if (rc == 0) {
 723                 spin_lock(&scrub->os_lock);
 724                 thread_set_flags(thread, SVC_RUNNING);
 725                 spin_unlock(&scrub->os_lock);
 726                 wake_up_all(&thread->t_ctl_waitq);
 727         }
 728         up_write(&scrub->os_rwsem);
 729
 730         RETURN(rc);
 731 }
 732
 733 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
 734 {
 735         struct scrub_file *sf = &scrub->os_file;
 736         int                rc;
 737
 738         if (likely(cfs_time_before(cfs_time_current(),
 739                                    scrub->os_time_next_checkpoint) ||
 740                    scrub->os_new_checked == 0))
 741                 return 0;
 742
 743         down_write(&scrub->os_rwsem);
 744         sf->sf_items_checked += scrub->os_new_checked;
 745         scrub->os_new_checked = 0;
 746         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 747         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 748         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 749                                             scrub->os_time_last_checkpoint);
 750         rc = osd_scrub_file_store(scrub);
 751         up_write(&scrub->os_rwsem);
 752
 753         return rc;
 754 }
 755
 756 static int osd_scrub_post(struct osd_scrub *scrub, int result)
 757 {
 758         struct scrub_file *sf = &scrub->os_file;
 759         int rc;
 760         ENTRY;
 761
 762         CDEBUG(D_LFSCK, "%.16s: OI scrub post, result = %d\n",
 763                osd_scrub2name(scrub), result);
 764
 765         down_write(&scrub->os_rwsem);
 766         spin_lock(&scrub->os_lock);
 767         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 768         spin_unlock(&scrub->os_lock);
 769         if (scrub->os_new_checked > 0) {
 770                 sf->sf_items_checked += scrub->os_new_checked;
 771                 scrub->os_new_checked = 0;
 772                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 773         }
 774         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 775         if (result > 0) {
 776                 struct osd_device *dev =
 777                         container_of0(scrub, struct osd_device, od_scrub);
 778
 779                 dev->od_igif_inoi = 1;
 780                 dev->od_check_ff = 0;
 781                 sf->sf_status = SS_COMPLETED;
 782                 if (!(sf->sf_param & SP_DRYRUN)) {
 783                         memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 784                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 785                                           SF_UPGRADE | SF_AUTO);
 786                 }
 787                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 788                 sf->sf_success_count++;
 789         } else if (result == 0) {
 790                 if (scrub->os_paused)
 791                         sf->sf_status = SS_PAUSED;
 792                 else
 793                         sf->sf_status = SS_STOPPED;
 794         } else {
 795                 sf->sf_status = SS_FAILED;
 796         }
 797         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 798                                             scrub->os_time_last_checkpoint);
 799         rc = osd_scrub_file_store(scrub);
 800         up_write(&scrub->os_rwsem);
 801
 802         RETURN(rc < 0 ? rc : result);
 803 }
 804
 805 /* iteration engine */
 806
 807 struct osd_iit_param {
 808         struct super_block *sb;
 809         struct buffer_head *bitmap;
 810         ldiskfs_group_t bg;
 811         __u32 gbase;
 812         __u32 offset;
 813 };
 814
 815 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 816                                    struct osd_device *dev,
 817                                    struct osd_iit_param *param,
 818                                    struct osd_idmap_cache **oic,
 819                                    const bool noslot);
 820
 821 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 822                                    struct osd_device *dev,
 823                                    struct osd_iit_param *param,
 824                                    struct osd_idmap_cache *oic,
 825                                    bool *noslot, int rc);
 826
 827 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 828 {
 829         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 830                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 831         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 832                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 833                 return SCRUB_NEXT_BREAK;
 834         } else {
 835                 *pos = param->gbase + param->offset;
 836                 return 0;
 837         }
 838 }
 839
 840 /**
 841  * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
 842  * \retval 0: FID-on-MDT
 843  */
 844 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
 845                                       struct osd_device *dev,
 846                                       struct lu_fid *fid)
 847 {
 848         /* XXX: The initial OI scrub will scan the top level /O to generate
 849          *      a small local FLDB according to the <seq>. If the given FID
 850          *      is in the local FLDB, then it is FID-on-OST; otherwise it's
 851          *      quite possible for FID-on-MDT. */
 852         if (dev->od_is_ost)
 853                 return SCRUB_NEXT_OSTOBJ_OLD;
 854         else
 855                 return 0;
 856 }
 857
 858 static int osd_scrub_get_fid(struct osd_thread_info *info,
 859                              struct osd_device *dev, struct inode *inode,
 860                              struct lu_fid *fid, bool scrub)
 861 {
 862         struct lustre_mdt_attrs *lma     = &info->oti_mdt_attrs;
 863         int                      rc;
 864         bool                     has_lma = false;
 865
 866         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 867         if (rc == 0) {
 868                 has_lma = true;
 869                 if (lma->lma_compat & LMAC_NOT_IN_OI ||
 870                     lma->lma_incompat & LMAI_AGENT)
 871                         return SCRUB_NEXT_CONTINUE;
 872
 873                 *fid = lma->lma_self_fid;
 874                 if (!scrub)
 875                         return 0;
 876
 877                 if (lma->lma_compat & LMAC_FID_ON_OST)
 878                         return SCRUB_NEXT_OSTOBJ;
 879
 880                 if (fid_is_idif(fid))
 881                         return SCRUB_NEXT_OSTOBJ_OLD;
 882
 883                 /* For local object. */
 884                 if (fid_is_internal(fid))
 885                         return 0;
 886
 887                 /* For external visible MDT-object with non-normal FID. */
 888                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
 889                         return 0;
 890
 891                 /* For the object with normal FID, it may be MDT-object,
 892                  * or may be 2.4 OST-object, need further distinguish.
 893                  * Fall through to next section. */
 894         }
 895
 896         if (rc == -ENODATA || rc == 0) {
 897                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
 898                 if (rc == 0) {
 899                         if (scrub)
 900                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
 901                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 902                         return rc;
 903                 }
 904
 905                 if (rc > 0) {
 906                         if (!has_lma)
 907                                 /* It is FID-on-OST, but we do not know how
 908                                  * to generate its FID, ignore it directly. */
 909                                 rc = SCRUB_NEXT_CONTINUE;
 910                         else
 911                                 /* It is 2.4 OST-object. */
 912                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 913                         return rc;
 914                 }
 915
 916                 if (rc != -ENODATA)
 917                         return rc;
 918
 919                 if (!has_lma) {
 920                         if (dev->od_scrub.os_convert_igif) {
 921                                 lu_igif_build(fid, inode->i_ino,
 922                                               inode->i_generation);
 923                                 if (scrub)
 924                                         rc = SCRUB_NEXT_NOLMA;
 925                                 else
 926                                         rc = 0;
 927                         } else {
 928                                 /* It may be FID-on-OST, or may be FID for
 929                                  * non-MDT0, anyway, we do not know how to
 930                                  * generate its FID, ignore it directly. */
 931                                 rc = SCRUB_NEXT_CONTINUE;
 932                         }
 933                         return rc;
 934                 }
 935
 936                 /* For OI scrub case only: the object has LMA but has no ff
 937                  * (or ff crashed). It may be MDT-object, may be OST-object
 938                  * with crashed ff. The last check is local FLDB. */
 939                 rc = osd_scrub_check_local_fldb(info, dev, fid);
 940         }
 941
 942         return rc;
 943 }
 944
 945 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 946                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 947                         struct super_block *sb, bool scrub)
 948 {
 949         struct inode *inode;
 950         int           rc;
 951         ENTRY;
 952
 953         /* Not handle the backend root object and agent parent object.
 954          * They are neither visible to namespace nor have OI mappings. */
 955         if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
 956                      pos == osd_remote_parent_ino(dev)))
 957                 RETURN(SCRUB_NEXT_CONTINUE);
 958
 959         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 960         inode = osd_iget(info, dev, lid);
 961         if (IS_ERR(inode)) {
 962                 rc = PTR_ERR(inode);
 963                 /* The inode may be removed after bitmap searching, or the
 964                  * file is new created without inode initialized yet. */
 965                 if (rc == -ENOENT || rc == -ESTALE)
 966                         RETURN(SCRUB_NEXT_CONTINUE);
 967
 968                 CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: "
 969                        "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name,
 970                        pos, rc);
 971                 RETURN(rc);
 972         }
 973
 974         /* It is an EA inode, no OI mapping for it, skip it. */
 975         if (osd_is_ea_inode(inode))
 976                 GOTO(put, rc = SCRUB_NEXT_CONTINUE);
 977
 978         if (scrub &&
 979             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 980                 /* Only skip it for the first OI scrub accessing. */
 981                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 982                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
 983         }
 984
 985         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
 986
 987         GOTO(put, rc);
 988
 989 put:
 990         iput(inode);
 991         return rc;
 992 }
 993
 994 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 995                           struct osd_iit_param *param,
 996                           struct osd_idmap_cache **oic, const bool noslot)
 997 {
 998         struct osd_scrub     *scrub  = &dev->od_scrub;
 999         struct ptlrpc_thread *thread = &scrub->os_thread;
1000         struct lu_fid        *fid;
1001         struct osd_inode_id  *lid;
1002         int                   rc;
1003
1004         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
1005                 struct l_wait_info lwi;
1006
1007                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
1008                 if (likely(lwi.lwi_timeout > 0))
1009                         l_wait_event(thread->t_ctl_waitq,
1010                                 !list_empty(&scrub->os_inconsistent_items) ||
1011                                 !thread_is_running(thread),
1012                                 &lwi);
1013         }
1014
1015         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
1016                 spin_lock(&scrub->os_lock);
1017                 thread_set_flags(thread, SVC_STOPPING);
1018                 spin_unlock(&scrub->os_lock);
1019                 return SCRUB_NEXT_CRASH;
1020         }
1021
1022         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
1023                 return SCRUB_NEXT_FATAL;
1024
1025         if (unlikely(!thread_is_running(thread)))
1026                 return SCRUB_NEXT_EXIT;
1027
1028         if (!list_empty(&scrub->os_inconsistent_items)) {
1029                 spin_lock(&scrub->os_lock);
1030                 if (likely(!list_empty(&scrub->os_inconsistent_items))) {
1031                         struct osd_inconsistent_item *oii;
1032
1033                         oii = list_entry(scrub->os_inconsistent_items.next,
1034                                 struct osd_inconsistent_item, oii_list);
1035                         list_del_init(&oii->oii_list);
1036                         spin_unlock(&scrub->os_lock);
1037
1038                         *oic = &oii->oii_cache;
1039                         scrub->os_in_prior = 1;
1040
1041                         return 0;
1042                 }
1043                 spin_unlock(&scrub->os_lock);
1044         }
1045
1046         if (noslot)
1047                 return SCRUB_NEXT_WAIT;
1048
1049         rc = osd_iit_next(param, &scrub->os_pos_current);
1050         if (rc != 0)
1051                 return rc;
1052
1053         *oic = &scrub->os_oic;
1054         fid = &(*oic)->oic_fid;
1055         lid = &(*oic)->oic_lid;
1056         rc = osd_iit_iget(info, dev, fid, lid,
1057                           scrub->os_pos_current, param->sb, true);
1058         return rc;
1059 }
1060
1061 static int osd_preload_next(struct osd_thread_info *info,
1062                             struct osd_device *dev, struct osd_iit_param *param,
1063                             struct osd_idmap_cache **oic, const bool noslot)
1064 {
1065         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
1066         struct osd_scrub        *scrub;
1067         struct ptlrpc_thread    *thread;
1068         int                      rc;
1069
1070         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1071         if (rc != 0)
1072                 return rc;
1073
1074         scrub = &dev->od_scrub;
1075         thread = &scrub->os_thread;
1076         if (thread_is_running(thread) &&
1077             ooc->ooc_pos_preload >= scrub->os_pos_current)
1078                 return SCRUB_NEXT_EXIT;
1079
1080         rc = osd_iit_iget(info, dev,
1081                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1082                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1083                           ooc->ooc_pos_preload, param->sb, false);
1084         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1085          * ignore the failure, so it still need to skip the inode next time. */
1086         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1087         return rc;
1088 }
1089
1090 static inline int
1091 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1092 {
1093         spin_lock(&scrub->os_lock);
1094         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1095             !list_empty(&scrub->os_inconsistent_items) ||
1096             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1097                 scrub->os_waiting = 0;
1098         else
1099                 scrub->os_waiting = 1;
1100         spin_unlock(&scrub->os_lock);
1101
1102         return !scrub->os_waiting;
1103 }
1104
1105 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1106                           struct osd_iit_param *param,
1107                           struct osd_idmap_cache *oic, bool *noslot, int rc)
1108 {
1109         struct l_wait_info       lwi    = { 0 };
1110         struct osd_scrub        *scrub  = &dev->od_scrub;
1111         struct scrub_file       *sf     = &scrub->os_file;
1112         struct ptlrpc_thread    *thread = &scrub->os_thread;
1113         struct osd_otable_it    *it     = dev->od_otable_it;
1114         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
1115
1116         switch (rc) {
1117         case SCRUB_NEXT_CONTINUE:
1118                 goto next;
1119         case SCRUB_NEXT_WAIT:
1120                 goto wait;
1121         case SCRUB_NEXT_NOSCRUB:
1122                 down_write(&scrub->os_rwsem);
1123                 scrub->os_new_checked++;
1124                 sf->sf_items_noscrub++;
1125                 up_write(&scrub->os_rwsem);
1126                 goto next;
1127         }
1128
1129         rc = osd_scrub_check_update(info, dev, oic, rc);
1130         if (rc != 0) {
1131                 scrub->os_in_prior = 0;
1132                 return rc;
1133         }
1134
1135         rc = osd_scrub_checkpoint(scrub);
1136         if (rc != 0) {
1137                 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1138                        "rc = %d\n", osd_scrub2name(scrub),
1139                        scrub->os_pos_current, rc);
1140                 /* Continue, as long as the scrub itself can go ahead. */
1141         }
1142
1143         if (scrub->os_in_prior) {
1144                 scrub->os_in_prior = 0;
1145                 return 0;
1146         }
1147
1148 next:
1149         scrub->os_pos_current = param->gbase + ++(param->offset);
1150
1151 wait:
1152         if (it != NULL && it->ooi_waiting && ooc != NULL &&
1153             ooc->ooc_pos_preload < scrub->os_pos_current) {
1154                 spin_lock(&scrub->os_lock);
1155                 it->ooi_waiting = 0;
1156                 wake_up_all(&thread->t_ctl_waitq);
1157                 spin_unlock(&scrub->os_lock);
1158         }
1159
1160         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1161                 return 0;
1162
1163         if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1164                 *noslot = false;
1165                 return 0;
1166         }
1167
1168         if (it != NULL)
1169                 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1170                              &lwi);
1171
1172         if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1173                 *noslot = false;
1174         else
1175                 *noslot = true;
1176         return 0;
1177 }
1178
1179 static int osd_preload_exec(struct osd_thread_info *info,
1180                             struct osd_device *dev, struct osd_iit_param *param,
1181                             struct osd_idmap_cache *oic, bool *noslot, int rc)
1182 {
1183         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1184
1185         if (rc == 0) {
1186                 ooc->ooc_cached_items++;
1187                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1188                                         ~OSD_OTABLE_IT_CACHE_MASK;
1189         }
1190         return rc > 0 ? 0 : rc;
1191 }
1192
1193 #define SCRUB_IT_ALL    1
1194 #define SCRUB_IT_CRASH  2
1195
1196 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1197                            bool inconsistent)
1198 {
1199         struct osd_scrub     *scrub  = &dev->od_scrub;
1200         struct ptlrpc_thread *thread = &scrub->os_thread;
1201         struct scrub_file    *sf     = &scrub->os_file;
1202         int                   rc;
1203         ENTRY;
1204
1205         LASSERT(!(flags & SS_AUTO_PARTIAL));
1206
1207         down_write(&scrub->os_rwsem);
1208         scrub->os_in_join = 1;
1209         if (flags & SS_SET_FAILOUT)
1210                 sf->sf_param |= SP_FAILOUT;
1211         else if (flags & SS_CLEAR_FAILOUT)
1212                 sf->sf_param &= ~SP_FAILOUT;
1213
1214         if (flags & SS_SET_DRYRUN)
1215                 sf->sf_param |= SP_DRYRUN;
1216         else if (flags & SS_CLEAR_DRYRUN)
1217                 sf->sf_param &= ~SP_DRYRUN;
1218
1219         if (flags & SS_RESET) {
1220                 osd_scrub_file_reset(scrub,
1221                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1222                         inconsistent ? SF_INCONSISTENT : 0);
1223                 sf->sf_status = SS_SCANNING;
1224         }
1225
1226         if (flags & SS_AUTO_FULL) {
1227                 sf->sf_flags |= SF_AUTO;
1228                 scrub->os_full_speed = 1;
1229         }
1230
1231         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1232                 scrub->os_full_speed = 1;
1233         else
1234                 scrub->os_full_speed = 0;
1235
1236         scrub->os_new_checked = 0;
1237         if (sf->sf_pos_last_checkpoint != 0)
1238                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1239         else
1240                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1241
1242         scrub->os_pos_current = sf->sf_pos_latest_start;
1243         sf->sf_time_latest_start = cfs_time_current_sec();
1244         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1245         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1246         rc = osd_scrub_file_store(scrub);
1247         if (rc != 0)
1248                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
1249                        "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1250
1251         spin_lock(&scrub->os_lock);
1252         scrub->os_waiting = 0;
1253         scrub->os_paused = 0;
1254         scrub->os_partial_scan = 0;
1255         scrub->os_in_join = 0;
1256         scrub->os_full_scrub = 0;
1257         spin_unlock(&scrub->os_lock);
1258         wake_up_all(&thread->t_ctl_waitq);
1259         up_write(&scrub->os_rwsem);
1260
1261         EXIT;
1262 }
1263
1264 static int osd_inode_iteration(struct osd_thread_info *info,
1265                                struct osd_device *dev, __u32 max, bool preload)
1266 {
1267         struct osd_scrub     *scrub  = &dev->od_scrub;
1268         struct ptlrpc_thread *thread = &scrub->os_thread;
1269         struct scrub_file    *sf     = &scrub->os_file;
1270         osd_iit_next_policy   next;
1271         osd_iit_exec_policy   exec;
1272         __u32                *pos;
1273         __u32                *count;
1274         struct osd_iit_param  param  = { NULL };
1275         struct l_wait_info    lwi    = { 0 };
1276         __u32                 limit;
1277         int                   rc;
1278         bool                  noslot = true;
1279         ENTRY;
1280
1281         param.sb = osd_sb(dev);
1282         if (preload)
1283                 goto full;
1284
1285         while (scrub->os_partial_scan && !scrub->os_in_join) {
1286                 struct osd_idmap_cache *oic = NULL;
1287
1288                 rc = osd_scrub_next(info, dev, &param, &oic, noslot);
1289                 switch (rc) {
1290                 case SCRUB_NEXT_EXIT:
1291                         RETURN(0);
1292                 case SCRUB_NEXT_CRASH:
1293                         RETURN(SCRUB_IT_CRASH);
1294                 case SCRUB_NEXT_FATAL:
1295                         RETURN(-EINVAL);
1296                 case SCRUB_NEXT_WAIT: {
1297                         struct kstatfs *ksfs = &info->oti_ksfs;
1298                         __u64 saved_flags;
1299
1300                         if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1301                             unlikely(sf->sf_items_updated_prior == 0))
1302                                 goto wait;
1303
1304                         if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1305                             scrub->os_full_scrub) {
1306                                 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1307                                                true);
1308                                 goto full;
1309                         }
1310
1311                         rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1312                         if (rc == 0) {
1313                                 __u64 used = ksfs->f_files - ksfs->f_ffree;
1314
1315                                 do_div(used, sf->sf_items_updated_prior);
1316                                 /* If we hit too much inconsistent OI
1317                                  * mappings during the partial scan,
1318                                  * then scan the device completely. */
1319                                 if (used < dev->od_full_scrub_ratio) {
1320                                         osd_scrub_join(dev,
1321                                                 SS_AUTO_FULL | SS_RESET, true);
1322                                         goto full;
1323                                 }
1324                         }
1325
1326 wait:
1327                         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1328                             cfs_fail_val > 0)
1329                                 continue;
1330
1331                         saved_flags = sf->sf_flags;
1332                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1333                                           SF_UPGRADE | SF_AUTO);
1334                         sf->sf_status = SS_COMPLETED;
1335                         l_wait_event(thread->t_ctl_waitq,
1336                                      !thread_is_running(thread) ||
1337                                      !scrub->os_partial_scan ||
1338                                      scrub->os_in_join ||
1339                                      !list_empty(&scrub->os_inconsistent_items),
1340                                      &lwi);
1341                         sf->sf_flags = saved_flags;
1342                         sf->sf_status = SS_SCANNING;
1343
1344                         if (unlikely(!thread_is_running(thread)))
1345                                 RETURN(0);
1346
1347                         if (!scrub->os_partial_scan || scrub->os_in_join)
1348                                 goto full;
1349
1350                         continue;
1351                 }
1352                 default:
1353                         LASSERTF(rc == 0, "rc = %d\n", rc);
1354
1355                         osd_scrub_exec(info, dev, &param, oic, &noslot, rc);
1356                         break;
1357                 }
1358         }
1359
1360 full:
1361         if (!preload) {
1362                 l_wait_event(thread->t_ctl_waitq,
1363                              !thread_is_running(thread) || !scrub->os_in_join,
1364                              &lwi);
1365
1366                 if (unlikely(!thread_is_running(thread)))
1367                         RETURN(0);
1368         }
1369
1370         noslot = false;
1371         if (!preload) {
1372                 next = osd_scrub_next;
1373                 exec = osd_scrub_exec;
1374                 pos = &scrub->os_pos_current;
1375                 count = &scrub->os_new_checked;
1376         } else {
1377                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1378
1379                 next = osd_preload_next;
1380                 exec = osd_preload_exec;
1381                 pos = &ooc->ooc_pos_preload;
1382                 count = &ooc->ooc_cached_items;
1383         }
1384         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1385
1386         while (*pos <= limit && *count < max) {
1387                 struct osd_idmap_cache *oic = NULL;
1388                 struct ldiskfs_group_desc *desc;
1389
1390                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1391                 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1392                 if (desc == NULL)
1393                         RETURN(-EIO);
1394
1395                 ldiskfs_lock_group(param.sb, param.bg);
1396                 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1397                         ldiskfs_unlock_group(param.sb, param.bg);
1398                         *pos = 1 + (param.bg + 1) *
1399                                 LDISKFS_INODES_PER_GROUP(param.sb);
1400                         continue;
1401                 }
1402                 ldiskfs_unlock_group(param.sb, param.bg);
1403
1404                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1405                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1406                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1407                 if (param.bitmap == NULL) {
1408                         CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1409                                "scrub will stop, urgent mode\n",
1410                                osd_scrub2name(scrub), (__u32)param.bg);
1411                         RETURN(-EIO);
1412                 }
1413
1414                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1415                        *count < max) {
1416                         if (param.offset +
1417                                 ldiskfs_itable_unused_count(param.sb, desc) >
1418                             LDISKFS_INODES_PER_GROUP(param.sb)) {
1419                                 *pos = 1 + (param.bg + 1) *
1420                                         LDISKFS_INODES_PER_GROUP(param.sb);
1421                                 goto next_group;
1422                         }
1423
1424                         rc = next(info, dev, &param, &oic, noslot);
1425                         switch (rc) {
1426                         case SCRUB_NEXT_BREAK:
1427                                 goto next_group;
1428                         case SCRUB_NEXT_EXIT:
1429                                 brelse(param.bitmap);
1430                                 RETURN(0);
1431                         case SCRUB_NEXT_CRASH:
1432                                 brelse(param.bitmap);
1433                                 RETURN(SCRUB_IT_CRASH);
1434                         case SCRUB_NEXT_FATAL:
1435                                 brelse(param.bitmap);
1436                                 RETURN(-EINVAL);
1437                         }
1438
1439                         rc = exec(info, dev, &param, oic, &noslot, rc);
1440                         if (rc != 0) {
1441                                 brelse(param.bitmap);
1442                                 RETURN(rc);
1443                         }
1444                 }
1445
1446 next_group:
1447                 brelse(param.bitmap);
1448         }
1449
1450         if (*pos > limit)
1451                 RETURN(SCRUB_IT_ALL);
1452         RETURN(0);
1453 }
1454
1455 static int osd_otable_it_preload(const struct lu_env *env,
1456                                  struct osd_otable_it *it)
1457 {
1458         struct osd_device       *dev   = it->ooi_dev;
1459         struct osd_scrub        *scrub = &dev->od_scrub;
1460         struct osd_otable_cache *ooc   = &it->ooi_cache;
1461         int                      rc;
1462         ENTRY;
1463
1464         rc = osd_inode_iteration(osd_oti_get(env), dev,
1465                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1466         if (rc == SCRUB_IT_ALL)
1467                 it->ooi_all_cached = 1;
1468
1469         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1470                 spin_lock(&scrub->os_lock);
1471                 scrub->os_waiting = 0;
1472                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1473                 spin_unlock(&scrub->os_lock);
1474         }
1475
1476         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1477 }
1478
1479 static int osd_scrub_main(void *args)
1480 {
1481         struct lu_env         env;
1482         struct osd_device    *dev    = (struct osd_device *)args;
1483         struct osd_scrub     *scrub  = &dev->od_scrub;
1484         struct ptlrpc_thread *thread = &scrub->os_thread;
1485         int                   rc;
1486         ENTRY;
1487
1488         rc = lu_env_init(&env, LCT_LOCAL);
1489         if (rc != 0) {
1490                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1491                        osd_scrub2name(scrub), rc);
1492                 GOTO(noenv, rc);
1493         }
1494
1495         rc = osd_scrub_prep(dev);
1496         if (rc != 0) {
1497                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1498                        osd_scrub2name(scrub), rc);
1499                 GOTO(out, rc);
1500         }
1501
1502         if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1503                 struct l_wait_info lwi = { 0 };
1504                 struct osd_otable_it *it = dev->od_otable_it;
1505                 struct osd_otable_cache *ooc = &it->ooi_cache;
1506
1507                 l_wait_event(thread->t_ctl_waitq,
1508                              it->ooi_user_ready || !thread_is_running(thread),
1509                              &lwi);
1510                 if (unlikely(!thread_is_running(thread)))
1511                         GOTO(post, rc = 0);
1512
1513                 scrub->os_pos_current = ooc->ooc_pos_preload;
1514         }
1515
1516         CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1517                osd_scrub2name(scrub), scrub->os_start_flags,
1518                scrub->os_pos_current);
1519
1520         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1521         if (unlikely(rc == SCRUB_IT_CRASH))
1522                 GOTO(out, rc = -EINVAL);
1523         GOTO(post, rc);
1524
1525 post:
1526         rc = osd_scrub_post(scrub, rc);
1527         CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1528                osd_scrub2name(scrub), scrub->os_pos_current, rc);
1529
1530 out:
1531         while (!list_empty(&scrub->os_inconsistent_items)) {
1532                 struct osd_inconsistent_item *oii;
1533
1534                 oii = list_entry(scrub->os_inconsistent_items.next,
1535                                      struct osd_inconsistent_item, oii_list);
1536                 list_del_init(&oii->oii_list);
1537                 OBD_FREE_PTR(oii);
1538         }
1539         lu_env_fini(&env);
1540
1541 noenv:
1542         spin_lock(&scrub->os_lock);
1543         thread_set_flags(thread, SVC_STOPPED);
1544         wake_up_all(&thread->t_ctl_waitq);
1545         spin_unlock(&scrub->os_lock);
1546         return rc;
1547 }
1548
1549 /* initial OI scrub */
1550
1551 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1552                          struct dentry *, filldir_t filldir);
1553
1554 #ifdef HAVE_FILLDIR_USE_CTX
1555 static int osd_ios_varfid_fill(struct dir_context *buf, const char *name,
1556                                int namelen, loff_t offset, __u64 ino,
1557                                unsigned d_type);
1558 static int osd_ios_lf_fill(struct dir_context *buf, const char *name,
1559                            int namelen, loff_t offset, __u64 ino,
1560                            unsigned d_type);
1561 static int osd_ios_dl_fill(struct dir_context *buf, const char *name,
1562                            int namelen, loff_t offset, __u64 ino,
1563                            unsigned d_type);
1564 static int osd_ios_uld_fill(struct dir_context *buf, const char *name,
1565                             int namelen, loff_t offset, __u64 ino,
1566                             unsigned d_type);
1567 #else
1568 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1569                                loff_t offset, __u64 ino, unsigned d_type);
1570 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1571                            loff_t offset, __u64 ino, unsigned d_type);
1572 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1573                            loff_t offset, __u64 ino, unsigned d_type);
1574 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1575                             loff_t offset, __u64 ino, unsigned d_type);
1576 #endif
1577
1578 static int
1579 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1580                      struct dentry *dentry, filldir_t filldir);
1581 static int
1582 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1583                   struct dentry *dentry, filldir_t filldir);
1584
1585 static int
1586 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1587                      struct dentry *dentry, filldir_t filldir);
1588
1589 enum osd_lf_flags {
1590         OLF_SCAN_SUBITEMS       = 0x0001,
1591         OLF_HIDE_FID            = 0x0002,
1592         OLF_SHOW_NAME           = 0x0004,
1593         OLF_NO_OI               = 0x0008,
1594         OLF_IDX_IN_FID          = 0x0010,
1595 };
1596
1597 struct osd_lf_map {
1598         char            *olm_name;
1599         struct lu_fid    olm_fid;
1600         __u16            olm_flags;
1601         __u16            olm_namelen;
1602         scandir_t        olm_scandir;
1603         filldir_t        olm_filldir;
1604 };
1605
1606 /* Add the new introduced local files in the list in the future. */
1607 static const struct osd_lf_map osd_lf_maps[] = {
1608         /* CATALOGS */
1609         { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1610                 sizeof(CATLIST) - 1, NULL, NULL },
1611
1612         /* CONFIGS */
1613         { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1614                 OLF_SCAN_SUBITEMS, sizeof(MOUNT_CONFIGS_DIR) - 1,
1615                 osd_ios_general_scan, osd_ios_varfid_fill },
1616
1617         /* NIDTBL_VERSIONS */
1618         { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1619                 sizeof(MGS_NIDTBL_DIR) - 1, osd_ios_general_scan,
1620                 osd_ios_varfid_fill },
1621
1622         /* PENDING */
1623         { "PENDING", { 0, 0, 0 }, 0, sizeof("PENDING") - 1, NULL, NULL },
1624
1625         /* ROOT */
1626         { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 },
1627                 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, sizeof("ROOT") - 1,
1628                 osd_ios_ROOT_scan, NULL },
1629
1630         /* changelog_catalog */
1631         { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, sizeof(CHANGELOG_CATALOG) - 1,
1632                 NULL, NULL },
1633
1634         /* changelog_users */
1635         { CHANGELOG_USERS, { 0, 0, 0 }, 0, sizeof(CHANGELOG_USERS) - 1,
1636                 NULL, NULL },
1637
1638         /* fld */
1639         { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1640                 sizeof("fld") - 1, NULL, NULL },
1641
1642         /* last_rcvd */
1643         { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1644                 sizeof(LAST_RCVD) - 1, NULL, NULL },
1645
1646         /* reply_data */
1647         { REPLY_DATA, { FID_SEQ_LOCAL_FILE, REPLY_DATA_OID, 0 }, OLF_SHOW_NAME,
1648                 sizeof(REPLY_DATA) - 1, NULL, NULL },
1649
1650         /* lov_objid */
1651         { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1652                 sizeof(LOV_OBJID) - 1, NULL, NULL },
1653
1654         /* lov_objseq */
1655         { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1656                 OLF_SHOW_NAME, sizeof(LOV_OBJSEQ) - 1, NULL, NULL },
1657
1658         /* quota_master */
1659         { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QMT_DIR) - 1,
1660                 osd_ios_general_scan, osd_ios_varfid_fill },
1661
1662         /* quota_slave */
1663         { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QSD_DIR) - 1,
1664                 osd_ios_general_scan, osd_ios_varfid_fill },
1665
1666         /* seq_ctl */
1667         { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1668                 OLF_SHOW_NAME, sizeof("seq_ctl") - 1, NULL, NULL },
1669
1670         /* seq_srv */
1671         { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1672                 OLF_SHOW_NAME, sizeof("seq_srv") - 1, NULL, NULL },
1673
1674         /* health_check */
1675         { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1676                 OLF_SHOW_NAME, sizeof(HEALTH_CHECK) - 1, NULL, NULL },
1677
1678         /* LFSCK */
1679         { LFSCK_DIR, { 0, 0, 0 }, 0, sizeof(LFSCK_DIR) - 1,
1680                 osd_ios_general_scan, osd_ios_varfid_fill },
1681
1682         /* lfsck_bookmark */
1683         { LFSCK_BOOKMARK, { 0, 0, 0 }, 0, sizeof(LFSCK_BOOKMARK) - 1,
1684                 NULL, NULL },
1685
1686         /* lfsck_layout */
1687         { LFSCK_LAYOUT, { 0, 0, 0 }, 0, sizeof(LFSCK_LAYOUT) - 1,
1688                 NULL, NULL },
1689
1690         /* lfsck_namespace */
1691         { LFSCK_NAMESPACE, { 0, 0, 0 }, 0, sizeof(LFSCK_NAMESPACE) - 1,
1692                 NULL, NULL },
1693
1694         /* OBJECTS, upgrade from old device */
1695         { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(OBJECTS) - 1,
1696                 osd_ios_OBJECTS_scan, NULL },
1697
1698         /* lquota_v2.user, upgrade from old device */
1699         { "lquota_v2.user", { 0, 0, 0 }, 0, sizeof("lquota_v2.user") - 1,
1700                 NULL, NULL },
1701
1702         /* lquota_v2.group, upgrade from old device */
1703         { "lquota_v2.group", { 0, 0, 0 }, 0, sizeof("lquota_v2.group") - 1,
1704                 NULL, NULL },
1705
1706         /* LAST_GROUP, upgrade from old device */
1707         { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1708                 OLF_SHOW_NAME, sizeof("LAST_GROUP") - 1, NULL, NULL },
1709
1710         /* committed batchid for cross-MDT operation */
1711         { "BATCHID", { FID_SEQ_LOCAL_FILE, BATCHID_COMMITTED_OID, 0 },
1712                 OLF_SHOW_NAME, sizeof("BATCHID") - 1, NULL, NULL },
1713
1714         /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1715          * and f_oid = index for their log files.  See lu_update_log{_dir}_fid()
1716          * for more details. */
1717
1718         /* update_log */
1719         { "update_log", { FID_SEQ_UPDATE_LOG, 0, 0 },
1720                 OLF_SHOW_NAME | OLF_IDX_IN_FID, sizeof("update_log") - 1,
1721                 NULL, NULL },
1722
1723         /* update_log_dir */
1724         { "update_log_dir", { FID_SEQ_UPDATE_LOG_DIR, 0, 0 },
1725                 OLF_SHOW_NAME | OLF_SCAN_SUBITEMS | OLF_IDX_IN_FID,
1726                 sizeof("update_log_dir") - 1,
1727                 osd_ios_general_scan, osd_ios_uld_fill },
1728
1729         /* lost+found */
1730         { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 },
1731                 OLF_SCAN_SUBITEMS, sizeof("lost+found") - 1,
1732                 osd_ios_general_scan, osd_ios_lf_fill },
1733
1734         { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1735 };
1736
1737 /* Add the new introduced files under .lustre/ in the list in the future. */
1738 static const struct osd_lf_map osd_dl_maps[] = {
1739         /* .lustre/fid */
1740         { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1741                 sizeof("fid") - 1, NULL, NULL },
1742
1743         /* .lustre/lost+found */
1744         { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1745                 sizeof("lost+found") - 1, NULL, NULL },
1746
1747         { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1748 };
1749
1750 struct osd_ios_item {
1751         struct list_head oii_list;
1752         struct dentry   *oii_dentry;
1753         scandir_t        oii_scandir;
1754         filldir_t        oii_filldir;
1755 };
1756
1757 struct osd_ios_filldir_buf {
1758 #ifdef HAVE_DIR_CONTEXT
1759         /* please keep it as first member */
1760         struct dir_context       ctx;
1761 #endif
1762         struct osd_thread_info  *oifb_info;
1763         struct osd_device       *oifb_dev;
1764         struct dentry           *oifb_dentry;
1765 };
1766
1767 static inline struct dentry *
1768 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1769 {
1770         struct dentry *dentry;
1771
1772         dentry = ll_lookup_one_len(name, parent, namelen);
1773         if (IS_ERR(dentry)) {
1774                 int rc = PTR_ERR(dentry);
1775
1776                 if (rc != -ENOENT)
1777                         CERROR("Fail to find %.*s in %.*s (%lu/%u): rc = %d\n",
1778                                namelen, name, parent->d_name.len,
1779                                parent->d_name.name, parent->d_inode->i_ino,
1780                                parent->d_inode->i_generation, rc);
1781
1782                 return dentry;
1783         }
1784
1785         if (dentry->d_inode == NULL) {
1786                 dput(dentry);
1787                 return ERR_PTR(-ENOENT);
1788         }
1789
1790         return dentry;
1791 }
1792
1793 static int
1794 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1795                  scandir_t scandir, filldir_t filldir)
1796 {
1797         struct osd_ios_item *item;
1798         ENTRY;
1799
1800         OBD_ALLOC_PTR(item);
1801         if (item == NULL)
1802                 RETURN(-ENOMEM);
1803
1804         INIT_LIST_HEAD(&item->oii_list);
1805         item->oii_dentry = dget(dentry);
1806         item->oii_scandir = scandir;
1807         item->oii_filldir = filldir;
1808         list_add_tail(&item->oii_list, &dev->od_ios_list);
1809
1810         RETURN(0);
1811 }
1812
1813 /**
1814  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1815  *
1816  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1817  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1818  * reference the inode, or fixed if it is missing or references another inode.
1819  */
1820 static int
1821 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1822                  struct inode *inode, const struct lu_fid *fid, int flags)
1823 {
1824         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
1825         struct osd_inode_id     *id     = &info->oti_id;
1826         struct osd_inode_id     *id2    = &info->oti_id2;
1827         struct osd_scrub        *scrub  = &dev->od_scrub;
1828         struct scrub_file       *sf     = &scrub->os_file;
1829         struct lu_fid            tfid;
1830         int                      rc;
1831         ENTRY;
1832
1833         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1834         if (rc != 0 && rc != -ENODATA) {
1835                 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1836                        "rc = %d\n", osd_name(dev), rc);
1837
1838                 RETURN(rc);
1839         }
1840
1841         osd_id_gen(id, inode->i_ino, inode->i_generation);
1842         if (rc == -ENODATA) {
1843                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
1844                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1845                 } else {
1846                         tfid = *fid;
1847                         if (flags & OLF_IDX_IN_FID) {
1848                                 LASSERT(dev->od_index >= 0);
1849
1850                                 tfid.f_oid = dev->od_index;
1851                         }
1852                 }
1853                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1854                 if (rc != 0) {
1855                         CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1856                               "scrub: rc = %d\n", osd_name(dev), rc);
1857
1858                         RETURN(rc);
1859                 }
1860         } else {
1861                 if (lma->lma_compat & LMAC_NOT_IN_OI)
1862                         RETURN(0);
1863
1864                 tfid = lma->lma_self_fid;
1865         }
1866
1867         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1868         if (rc != 0) {
1869                 if (rc != -ENOENT)
1870                         RETURN(rc);
1871
1872                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1873                                                DTO_INDEX_INSERT, true, 0, NULL);
1874                 if (rc > 0)
1875                         rc = 0;
1876
1877                 RETURN(rc);
1878         }
1879
1880         if (osd_id_eq_strict(id, id2))
1881                 RETURN(0);
1882
1883         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1884                 osd_scrub_file_reset(scrub,
1885                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1886                                      SF_INCONSISTENT);
1887                 rc = osd_scrub_file_store(scrub);
1888                 if (rc != 0)
1889                         RETURN(rc);
1890         }
1891
1892         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1893                                        DTO_INDEX_UPDATE, true, 0, NULL);
1894         if (rc > 0)
1895                 rc = 0;
1896
1897         RETURN(rc);
1898 }
1899
1900 /**
1901  * It scans the /lost+found, and for the OST-object (with filter_fid
1902  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1903  */
1904 #ifdef HAVE_FILLDIR_USE_CTX
1905 static int osd_ios_lf_fill(struct dir_context *buf, const char *name,
1906                            int namelen,
1907 #else
1908 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1909 #endif
1910                            loff_t offset, __u64 ino, unsigned d_type)
1911 {
1912         struct osd_ios_filldir_buf *fill_buf =
1913                 (struct osd_ios_filldir_buf *)buf;
1914         struct osd_thread_info     *info     = fill_buf->oifb_info;
1915         struct osd_device          *dev      = fill_buf->oifb_dev;
1916         struct lu_fid              *fid      = &info->oti_fid;
1917         struct osd_scrub           *scrub    = &dev->od_scrub;
1918         struct dentry              *parent   = fill_buf->oifb_dentry;
1919         struct dentry              *child;
1920         struct inode               *dir      = parent->d_inode;
1921         struct inode               *inode;
1922         int                         rc;
1923         ENTRY;
1924
1925         /* skip any '.' started names */
1926         if (name[0] == '.')
1927                 RETURN(0);
1928
1929         scrub->os_lf_scanned++;
1930         child = osd_ios_lookup_one_len(name, parent, namelen);
1931         if (IS_ERR(child)) {
1932                 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1933                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
1934                 RETURN(0);
1935         }
1936
1937         inode = child->d_inode;
1938         if (S_ISDIR(inode->i_mode)) {
1939                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1940                                       osd_ios_lf_fill);
1941                 if (rc != 0)
1942                         CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1943                               "rc = %d\n", osd_name(dev), namelen, name, rc);
1944                 GOTO(put, rc);
1945         }
1946
1947         if (!S_ISREG(inode->i_mode))
1948                 GOTO(put, rc = 0);
1949
1950         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1951         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1952                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1953                 if (rc == 0) {
1954                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1955                                "/lost+found.\n", namelen, name, PFID(fid));
1956                         scrub->os_lf_repaired++;
1957                 } else {
1958                         CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1959                                DFID": rc = %d\n",
1960                                osd_name(dev), namelen, name, PFID(fid), rc);
1961                 }
1962         }
1963
1964         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1965          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1966          *      can process them in furtuer. */
1967
1968         GOTO(put, rc);
1969
1970 put:
1971         if (rc < 0)
1972                 scrub->os_lf_failed++;
1973         dput(child);
1974         /* skip the failure to make the scanning to continue. */
1975         return 0;
1976 }
1977
1978 #ifdef HAVE_FILLDIR_USE_CTX
1979 static int osd_ios_varfid_fill(struct dir_context  *buf, const char *name,
1980                                int namelen,
1981 #else
1982 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1983 #endif
1984                                loff_t offset, __u64 ino, unsigned d_type)
1985 {
1986         struct osd_ios_filldir_buf *fill_buf =
1987                 (struct osd_ios_filldir_buf *)buf;
1988         struct osd_device          *dev      = fill_buf->oifb_dev;
1989         struct dentry              *child;
1990         int                         rc;
1991         ENTRY;
1992
1993         /* skip any '.' started names */
1994         if (name[0] == '.')
1995                 RETURN(0);
1996
1997         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1998         if (IS_ERR(child))
1999                 RETURN(PTR_ERR(child));
2000
2001         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2002                               NULL, 0);
2003         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
2004                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2005                                       osd_ios_varfid_fill);
2006         dput(child);
2007
2008         RETURN(rc);
2009 }
2010
2011 #ifdef HAVE_FILLDIR_USE_CTX
2012 static int osd_ios_dl_fill(struct dir_context  *buf, const char *name,
2013                            int namelen,
2014 #else
2015 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
2016 #endif
2017                            loff_t offset, __u64 ino, unsigned d_type)
2018 {
2019         struct osd_ios_filldir_buf *fill_buf =
2020                 (struct osd_ios_filldir_buf *)buf;
2021         struct osd_device          *dev      = fill_buf->oifb_dev;
2022         const struct osd_lf_map    *map;
2023         struct dentry              *child;
2024         int                         rc       = 0;
2025         ENTRY;
2026
2027         /* skip any '.' started names */
2028         if (name[0] == '.')
2029                 RETURN(0);
2030
2031         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
2032                 if (map->olm_namelen != namelen)
2033                         continue;
2034
2035                 if (strncmp(map->olm_name, name, namelen) == 0)
2036                         break;
2037         }
2038
2039         if (map->olm_name == NULL)
2040                 RETURN(0);
2041
2042         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2043         if (IS_ERR(child))
2044                 RETURN(PTR_ERR(child));
2045
2046         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2047                               &map->olm_fid, map->olm_flags);
2048         dput(child);
2049
2050         RETURN(rc);
2051 }
2052
2053 #ifdef HAVE_FILLDIR_USE_CTX
2054 static int osd_ios_uld_fill(struct dir_context *buf, const char *name,
2055                             int namelen,
2056 #else
2057 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
2058 #endif
2059                             loff_t offset, __u64 ino, unsigned d_type)
2060 {
2061         struct osd_ios_filldir_buf *fill_buf =
2062                 (struct osd_ios_filldir_buf *)buf;
2063         struct dentry              *child;
2064         struct lu_fid               tfid;
2065         int                         rc       = 0;
2066         ENTRY;
2067
2068         /* skip any non-DFID format name */
2069         if (name[0] != '[')
2070                 RETURN(0);
2071
2072         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2073         if (IS_ERR(child))
2074                 RETURN(PTR_ERR(child));
2075
2076         /* skip the start '[' */
2077         sscanf(&name[1], SFID, RFID(&tfid));
2078         if (fid_is_sane(&tfid))
2079                 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
2080                                       child->d_inode, &tfid, 0);
2081         else
2082                 rc = -EIO;
2083         dput(child);
2084
2085         RETURN(rc);
2086 }
2087
2088 #ifdef HAVE_FILLDIR_USE_CTX
2089 static int osd_ios_root_fill(struct dir_context *buf, const char *name,
2090                              int namelen,
2091 #else
2092 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
2093 #endif
2094                              loff_t offset, __u64 ino, unsigned d_type)
2095 {
2096         struct osd_ios_filldir_buf *fill_buf =
2097                 (struct osd_ios_filldir_buf *)buf;
2098         struct osd_device          *dev      = fill_buf->oifb_dev;
2099         const struct osd_lf_map    *map;
2100         struct dentry              *child;
2101         int                         rc       = 0;
2102         ENTRY;
2103
2104         /* skip any '.' started names */
2105         if (name[0] == '.')
2106                 RETURN(0);
2107
2108         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2109                 if (map->olm_namelen != namelen)
2110                         continue;
2111
2112                 if (strncmp(map->olm_name, name, namelen) == 0)
2113                         break;
2114         }
2115
2116         if (map->olm_name == NULL)
2117                 RETURN(0);
2118
2119         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2120         if (IS_ERR(child))
2121                 RETURN(PTR_ERR(child));
2122
2123         if (!(map->olm_flags & OLF_NO_OI))
2124                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2125                                       &map->olm_fid, map->olm_flags);
2126         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2127                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2128                                       map->olm_filldir);
2129         dput(child);
2130
2131         RETURN(rc);
2132 }
2133
2134 static int
2135 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2136                      struct dentry *dentry, filldir_t filldir)
2137 {
2138         struct osd_ios_filldir_buf    buf   = {
2139 #ifdef HAVE_DIR_CONTEXT
2140                                                 .ctx.actor = filldir,
2141 #endif
2142                                                 .oifb_info = info,
2143                                                 .oifb_dev = dev,
2144                                                 .oifb_dentry = dentry };
2145         struct file                  *filp  = &info->oti_file;
2146         struct inode                 *inode = dentry->d_inode;
2147         const struct file_operations *fops  = inode->i_fop;
2148         int                           rc;
2149         ENTRY;
2150
2151         LASSERT(filldir != NULL);
2152
2153         filp->f_pos = 0;
2154         filp->f_path.dentry = dentry;
2155         filp->f_mode = FMODE_64BITHASH;
2156         filp->f_mapping = inode->i_mapping;
2157         filp->f_op = fops;
2158         filp->private_data = NULL;
2159         set_file_inode(filp, inode);
2160
2161 #ifdef HAVE_DIR_CONTEXT
2162         buf.ctx.pos = filp->f_pos;
2163         rc = fops->iterate(filp, &buf.ctx);
2164         filp->f_pos = buf.ctx.pos;
2165 #else
2166         rc = fops->readdir(filp, &buf, filldir);
2167 #endif
2168         fops->release(inode, filp);
2169
2170         RETURN(rc);
2171 }
2172
2173 static int
2174 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2175                   struct dentry *dentry, filldir_t filldir)
2176 {
2177         struct osd_scrub  *scrub  = &dev->od_scrub;
2178         struct scrub_file *sf     = &scrub->os_file;
2179         struct dentry     *child;
2180         int                rc;
2181         ENTRY;
2182
2183         /* It is existing MDT0 device. We only allow the case of object without
2184          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2185          * can generate IGIF mode FID for the object and related OI mapping. If
2186          * it is on other MDTs, then becuase file-level backup/restore, related
2187          * OI mapping may be invalid already, we do not know which is the right
2188          * FID for the object. We only allow IGIF objects to reside on the MDT0.
2189          *
2190          * XXX: For the case of object on non-MDT0 device with neither LMA nor
2191          *      "fid" xattr, then something crashed. We cannot re-generate the
2192          *      FID directly, instead, the OI scrub will scan the OI structure
2193          *      and try to re-generate the LMA from the OI mapping. But if the
2194          *      OI mapping crashed or lost also, then we have to give up under
2195          *      double failure cases. */
2196         scrub->os_convert_igif = 1;
2197         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2198                                        strlen(dot_lustre_name));
2199         if (IS_ERR(child)) {
2200                 rc = PTR_ERR(child);
2201                 if (rc == -ENOENT) {
2202                         /* It is 1.8 MDT device. */
2203                         if (!(sf->sf_flags & SF_UPGRADE)) {
2204                                 osd_scrub_file_reset(scrub,
2205                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2206                                         SF_UPGRADE);
2207                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2208                                 rc = osd_scrub_file_store(scrub);
2209                         } else {
2210                                 rc = 0;
2211                         }
2212                 }
2213         } else {
2214                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2215                  * so the client will get IGIF for the ".lustre" object when
2216                  * the MDT restart.
2217                  *
2218                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2219                  * it does not know whether there are some old clients cached
2220                  * the ".lustre" IGIF during the upgrading. Two choices:
2221                  *
2222                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2223                  *    It will allow the old connected clients to access the
2224                  *    ".lustre" with cached IGIF. But it will cause others
2225                  *    on the MDT failed to check "fid_is_dot_lustre()".
2226                  *
2227                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2228                  *    for ".lustre" in spite of whether there are some clients
2229                  *    cached the ".lustre" IGIF or not. It enables the check
2230                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
2231                  *    that the old connected clients cannot access the ".lustre"
2232                  *    with the cached IGIF.
2233                  *
2234                  * Usually, it is rare case for the old connected clients
2235                  * to access the ".lustre" with cached IGIF. So we prefer
2236                  * to the solution 2). */
2237                 rc = osd_ios_scan_one(info, dev, child->d_inode,
2238                                       &LU_DOT_LUSTRE_FID, 0);
2239                 if (rc == 0)
2240                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2241                                               osd_ios_dl_fill);
2242                 dput(child);
2243         }
2244
2245         RETURN(rc);
2246 }
2247
2248 static int
2249 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2250                      struct dentry *dentry, filldir_t filldir)
2251 {
2252         struct osd_scrub  *scrub  = &dev->od_scrub;
2253         struct scrub_file *sf     = &scrub->os_file;
2254         struct dentry     *child;
2255         int                rc;
2256         ENTRY;
2257
2258         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2259                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2260                 rc = osd_scrub_file_store(scrub);
2261                 if (rc != 0)
2262                         RETURN(rc);
2263         }
2264
2265         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2266         if (!IS_ERR(child)) {
2267                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2268                 dput(child);
2269         } else {
2270                 rc = PTR_ERR(child);
2271         }
2272
2273         if (rc != 0 && rc != -ENOENT)
2274                 RETURN(rc);
2275
2276         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2277         if (!IS_ERR(child)) {
2278                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2279                 dput(child);
2280         } else {
2281                 rc = PTR_ERR(child);
2282         }
2283
2284         if (rc == -ENOENT)
2285                 rc = 0;
2286
2287         RETURN(rc);
2288 }
2289
2290 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2291                                 struct osd_device *dev)
2292 {
2293         struct osd_ios_item     *item    = NULL;
2294         scandir_t                scandir = osd_ios_general_scan;
2295         filldir_t                filldir = osd_ios_root_fill;
2296         struct dentry           *dentry  = osd_sb(dev)->s_root;
2297         const struct osd_lf_map *map     = osd_lf_maps;
2298         int                      rc;
2299         ENTRY;
2300
2301         /* Lookup IGIF in OI by force for initial OI scrub. */
2302         dev->od_igif_inoi = 1;
2303
2304         while (1) {
2305                 rc = scandir(info, dev, dentry, filldir);
2306                 if (item != NULL) {
2307                         dput(item->oii_dentry);
2308                         OBD_FREE_PTR(item);
2309                 }
2310
2311                 if (rc != 0)
2312                         break;
2313
2314                 if (list_empty(&dev->od_ios_list))
2315                         break;
2316
2317                 item = list_entry(dev->od_ios_list.next,
2318                                   struct osd_ios_item, oii_list);
2319                 list_del_init(&item->oii_list);
2320
2321                 LASSERT(item->oii_scandir != NULL);
2322                 scandir = item->oii_scandir;
2323                 filldir = item->oii_filldir;
2324                 dentry = item->oii_dentry;
2325         }
2326
2327         while (!list_empty(&dev->od_ios_list)) {
2328                 item = list_entry(dev->od_ios_list.next,
2329                                   struct osd_ios_item, oii_list);
2330                 list_del_init(&item->oii_list);
2331                 dput(item->oii_dentry);
2332                 OBD_FREE_PTR(item);
2333         }
2334
2335         if (rc != 0)
2336                 RETURN(rc);
2337
2338         /* There maybe the case that the object has been removed, but its OI
2339          * mapping is still in the OI file, such as the "CATALOGS" after MDT
2340          * file-level backup/restore. So here cleanup the stale OI mappings. */
2341         while (map->olm_name != NULL) {
2342                 struct dentry *child;
2343
2344                 if (fid_is_zero(&map->olm_fid)) {
2345                         map++;
2346                         continue;
2347                 }
2348
2349                 child = osd_ios_lookup_one_len(map->olm_name,
2350                                                osd_sb(dev)->s_root,
2351                                                map->olm_namelen);
2352                 if (!IS_ERR(child))
2353                         dput(child);
2354                 else if (PTR_ERR(child) == -ENOENT)
2355                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2356                                                   NULL, DTO_INDEX_DELETE,
2357                                                   true, 0, NULL);
2358                 map++;
2359         }
2360
2361         RETURN(0);
2362 }
2363
2364 char *osd_lf_fid2name(const struct lu_fid *fid)
2365 {
2366         const struct osd_lf_map *map = osd_lf_maps;
2367
2368         while (map->olm_name != NULL) {
2369                 if (!lu_fid_eq(fid, &map->olm_fid)) {
2370                         map++;
2371                         continue;
2372                 }
2373
2374                 if (map->olm_flags & OLF_SHOW_NAME)
2375                         return map->olm_name;
2376                 else
2377                         return "";
2378         }
2379
2380         return NULL;
2381 }
2382
2383 /* OI scrub start/stop */
2384
2385 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2386 {
2387         struct osd_scrub     *scrub  = &dev->od_scrub;
2388         struct ptlrpc_thread *thread = &scrub->os_thread;
2389         struct l_wait_info    lwi    = { 0 };
2390         struct task_struct   *task;
2391         int                   rc;
2392         ENTRY;
2393
2394         /* os_lock: sync status between stop and scrub thread */
2395         spin_lock(&scrub->os_lock);
2396
2397 again:
2398         if (thread_is_running(thread)) {
2399                 spin_unlock(&scrub->os_lock);
2400                 if (!(scrub->os_file.sf_flags & SF_AUTO ||
2401                       scrub->os_partial_scan) ||
2402                      (flags & SS_AUTO_PARTIAL))
2403                         RETURN(-EALREADY);
2404
2405                 osd_scrub_join(dev, flags, false);
2406                 spin_lock(&scrub->os_lock);
2407                 if (!thread_is_running(thread))
2408                         goto again;
2409
2410                 spin_unlock(&scrub->os_lock);
2411                 RETURN(0);
2412         }
2413
2414         if (unlikely(thread_is_stopping(thread))) {
2415                 spin_unlock(&scrub->os_lock);
2416                 l_wait_event(thread->t_ctl_waitq,
2417                              thread_is_stopped(thread),
2418                              &lwi);
2419                 spin_lock(&scrub->os_lock);
2420                 goto again;
2421         }
2422         spin_unlock(&scrub->os_lock);
2423
2424         if (scrub->os_file.sf_status == SS_COMPLETED) {
2425                 if (!(flags & SS_SET_FAILOUT))
2426                         flags |= SS_CLEAR_FAILOUT;
2427
2428                 if (!(flags & SS_SET_DRYRUN))
2429                         flags |= SS_CLEAR_DRYRUN;
2430
2431                 flags |= SS_RESET;
2432         }
2433
2434         scrub->os_start_flags = flags;
2435         thread_set_flags(thread, 0);
2436         task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2437         if (IS_ERR(task)) {
2438                 rc = PTR_ERR(task);
2439                 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2440                        osd_scrub2name(scrub), rc);
2441                 RETURN(rc);
2442         }
2443
2444         l_wait_event(thread->t_ctl_waitq,
2445                      thread_is_running(thread) || thread_is_stopped(thread),
2446                      &lwi);
2447
2448         RETURN(0);
2449 }
2450
2451 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2452 {
2453         int rc;
2454         ENTRY;
2455
2456         /* od_otable_mutex: prevent curcurrent start/stop */
2457         mutex_lock(&dev->od_otable_mutex);
2458         rc = do_osd_scrub_start(dev, flags);
2459         mutex_unlock(&dev->od_otable_mutex);
2460
2461         RETURN(rc == -EALREADY ? 0 : rc);
2462 }
2463
2464 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2465 {
2466         struct ptlrpc_thread *thread = &scrub->os_thread;
2467         struct l_wait_info    lwi    = { 0 };
2468
2469         /* os_lock: sync status between stop and scrub thread */
2470         spin_lock(&scrub->os_lock);
2471         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2472                 thread_set_flags(thread, SVC_STOPPING);
2473                 spin_unlock(&scrub->os_lock);
2474                 wake_up_all(&thread->t_ctl_waitq);
2475                 l_wait_event(thread->t_ctl_waitq,
2476                              thread_is_stopped(thread),
2477                              &lwi);
2478                 /* Do not skip the last lock/unlock, which can guarantee that
2479                  * the caller cannot return until the OI scrub thread exit. */
2480                 spin_lock(&scrub->os_lock);
2481         }
2482         spin_unlock(&scrub->os_lock);
2483 }
2484
2485 static void osd_scrub_stop(struct osd_device *dev)
2486 {
2487         /* od_otable_mutex: prevent curcurrent start/stop */
2488         mutex_lock(&dev->od_otable_mutex);
2489         dev->od_scrub.os_paused = 1;
2490         do_osd_scrub_stop(&dev->od_scrub);
2491         mutex_unlock(&dev->od_otable_mutex);
2492 }
2493
2494 /* OI scrub setup/cleanup */
2495
2496 static const char osd_scrub_name[] = "OI_scrub";
2497
2498 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2499 {
2500         struct osd_thread_info     *info   = osd_oti_get(env);
2501         struct osd_scrub           *scrub  = &dev->od_scrub;
2502         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
2503         struct scrub_file          *sf     = &scrub->os_file;
2504         struct super_block         *sb     = osd_sb(dev);
2505         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
2506         struct lvfs_run_ctxt        saved;
2507         struct file                *filp;
2508         struct inode               *inode;
2509         struct lu_fid              *fid    = &info->oti_fid;
2510         bool                        dirty  = false;
2511         bool                        restored = false;
2512         int                         rc     = 0;
2513         ENTRY;
2514
2515         memset(scrub, 0, sizeof(*scrub));
2516         OBD_SET_CTXT_MAGIC(ctxt);
2517         ctxt->pwdmnt = dev->od_mnt;
2518         ctxt->pwd = dev->od_mnt->mnt_root;
2519         ctxt->fs = get_ds();
2520
2521         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2522         init_rwsem(&scrub->os_rwsem);
2523         spin_lock_init(&scrub->os_lock);
2524         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2525
2526         push_ctxt(&saved, ctxt);
2527         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2528         if (IS_ERR(filp)) {
2529                 pop_ctxt(&saved, ctxt);
2530                 RETURN(PTR_ERR(filp));
2531         }
2532
2533         inode = file_inode(filp);
2534         /* 'What the @fid is' is not imporatant, because the object
2535          * has no OI mapping, and only is visible inside the OSD.*/
2536         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2537         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2538         if (rc != 0) {
2539                 filp_close(filp, NULL);
2540                 pop_ctxt(&saved, ctxt);
2541                 RETURN(rc);
2542         }
2543
2544         scrub->os_inode = igrab(inode);
2545         filp_close(filp, NULL);
2546         pop_ctxt(&saved, ctxt);
2547
2548         rc = osd_scrub_file_load(scrub);
2549         if (rc == -ENOENT) {
2550                 osd_scrub_file_init(scrub, es->s_uuid);
2551                 /* If the "/O" dir does not exist when mount (indicated by
2552                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2553                  * then it is quite probably that the device is a new one,
2554                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2555                  *
2556                  * For the rare case that "/O" and "OI_scrub" both lost on
2557                  * an old device, it can be found and cleared later.
2558                  *
2559                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2560                  * need to check "filter_fid_old" and to convert it to
2561                  * "filter_fid" for each object, and all the IGIF should
2562                  * have their FID mapping in OI files already. */
2563                 if (dev->od_maybe_new)
2564                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2565                 dirty = true;
2566         } else if (rc != 0) {
2567                 GOTO(cleanup_inode, rc);
2568         } else {
2569                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2570                         struct obd_uuid *old_uuid;
2571                         struct obd_uuid *new_uuid;
2572
2573                         OBD_ALLOC_PTR(old_uuid);
2574                         OBD_ALLOC_PTR(new_uuid);
2575                         if (old_uuid == NULL || new_uuid == NULL) {
2576                                 CERROR("%.16s: UUID has been changed, but"
2577                                        "failed to allocate RAM for report\n",
2578                                        LDISKFS_SB(sb)->s_es->s_volume_name);
2579                         } else {
2580                                 class_uuid_unparse(sf->sf_uuid, old_uuid);
2581                                 class_uuid_unparse(es->s_uuid, new_uuid);
2582                                 CERROR("%.16s: UUID has been changed from "
2583                                        "%s to %s\n",
2584                                        LDISKFS_SB(sb)->s_es->s_volume_name,
2585                                        old_uuid->uuid, new_uuid->uuid);
2586                         }
2587                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2588                         dirty = true;
2589                         restored = true;
2590                         if (old_uuid != NULL)
2591                                 OBD_FREE_PTR(old_uuid);
2592                         if (new_uuid != NULL)
2593                                 OBD_FREE_PTR(new_uuid);
2594                 } else if (sf->sf_status == SS_SCANNING) {
2595                         sf->sf_status = SS_CRASHED;
2596                         dirty = true;
2597                 }
2598         }
2599
2600         if (sf->sf_pos_last_checkpoint != 0)
2601                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2602         else
2603                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2604
2605         if (dirty) {
2606                 rc = osd_scrub_file_store(scrub);
2607                 if (rc != 0)
2608                         GOTO(cleanup_inode, rc);
2609         }
2610
2611         /* Initialize OI files. */
2612         rc = osd_oi_init(info, dev, restored);
2613         if (rc < 0)
2614                 GOTO(cleanup_inode, rc);
2615
2616         rc = osd_initial_OI_scrub(info, dev);
2617         if (rc != 0)
2618                 GOTO(cleanup_oi, rc);
2619
2620         if (sf->sf_flags & SF_UPGRADE ||
2621             !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2622               sf->sf_success_count > 0)) {
2623                 dev->od_igif_inoi = 0;
2624                 dev->od_check_ff = dev->od_is_ost;
2625         } else {
2626                 dev->od_igif_inoi = 1;
2627                 dev->od_check_ff = 0;
2628         }
2629
2630         if (sf->sf_flags & SF_INCONSISTENT)
2631                 /* The 'od_igif_inoi' will be set under the
2632                  * following cases:
2633                  * 1) new created system, or
2634                  * 2) restored from file-level backup, or
2635                  * 3) the upgrading completed.
2636                  *
2637                  * The 'od_igif_inoi' may be cleared by OI scrub
2638                  * later if found that the system is upgrading. */
2639                 dev->od_igif_inoi = 1;
2640
2641         if (!dev->od_noscrub &&
2642             ((sf->sf_status == SS_PAUSED) ||
2643              (sf->sf_status == SS_CRASHED &&
2644               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2645                               SF_UPGRADE | SF_AUTO)) ||
2646              (sf->sf_status == SS_INIT &&
2647               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2648                               SF_UPGRADE))))
2649                 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2650
2651         if (rc != 0)
2652                 GOTO(cleanup_oi, rc);
2653
2654         /* it is possible that dcache entries may keep objects after they are
2655          * deleted by OSD. While it looks safe this can cause object data to
2656          * stay until umount causing failures in tests calculating free space,
2657          * e.g. replay-ost-single. Since those dcache entries are not used
2658          * anymore let's just free them after use here */
2659         shrink_dcache_sb(sb);
2660
2661         RETURN(0);
2662 cleanup_oi:
2663         osd_oi_fini(info, dev);
2664 cleanup_inode:
2665         iput(scrub->os_inode);
2666         scrub->os_inode = NULL;
2667
2668         return rc;
2669 }
2670
2671 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2672 {
2673         struct osd_scrub *scrub = &dev->od_scrub;
2674
2675         LASSERT(dev->od_otable_it == NULL);
2676
2677         if (scrub->os_inode != NULL) {
2678                 osd_scrub_stop(dev);
2679                 iput(scrub->os_inode);
2680                 scrub->os_inode = NULL;
2681         }
2682         if (dev->od_oi_table != NULL)
2683                 osd_oi_fini(osd_oti_get(env), dev);
2684 }
2685
2686 /* object table based iteration APIs */
2687
2688 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2689                                        struct dt_object *dt, __u32 attr)
2690 {
2691         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2692         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2693         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2694         struct osd_scrub       *scrub = &dev->od_scrub;
2695         struct osd_otable_it   *it;
2696         __u32                   start = 0;
2697         int                     rc;
2698         ENTRY;
2699
2700         /* od_otable_mutex: prevent curcurrent init/fini */
2701         mutex_lock(&dev->od_otable_mutex);
2702         if (dev->od_otable_it != NULL)
2703                 GOTO(out, it = ERR_PTR(-EALREADY));
2704
2705         OBD_ALLOC_PTR(it);
2706         if (it == NULL)
2707                 GOTO(out, it = ERR_PTR(-ENOMEM));
2708
2709         dev->od_otable_it = it;
2710         it->ooi_dev = dev;
2711         it->ooi_cache.ooc_consumer_idx = -1;
2712         if (flags & DOIF_OUTUSED)
2713                 it->ooi_used_outside = 1;
2714
2715         if (flags & DOIF_RESET)
2716                 start |= SS_RESET;
2717
2718         if (valid & DOIV_ERROR_HANDLE) {
2719                 if (flags & DOIF_FAILOUT)
2720                         start |= SS_SET_FAILOUT;
2721                 else
2722                         start |= SS_CLEAR_FAILOUT;
2723         }
2724
2725         if (valid & DOIV_DRYRUN) {
2726                 if (flags & DOIF_DRYRUN)
2727                         start |= SS_SET_DRYRUN;
2728                 else
2729                         start |= SS_CLEAR_DRYRUN;
2730         }
2731
2732         rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2733         if (rc < 0 && rc != -EALREADY) {
2734                 dev->od_otable_it = NULL;
2735                 OBD_FREE_PTR(it);
2736                 GOTO(out, it = ERR_PTR(rc));
2737         }
2738
2739         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2740
2741         GOTO(out, it);
2742
2743 out:
2744         mutex_unlock(&dev->od_otable_mutex);
2745         return (struct dt_it *)it;
2746 }
2747
2748 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2749 {
2750         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2751         struct osd_device    *dev = it->ooi_dev;
2752
2753         /* od_otable_mutex: prevent curcurrent init/fini */
2754         mutex_lock(&dev->od_otable_mutex);
2755         do_osd_scrub_stop(&dev->od_scrub);
2756         LASSERT(dev->od_otable_it == it);
2757
2758         dev->od_otable_it = NULL;
2759         mutex_unlock(&dev->od_otable_mutex);
2760         OBD_FREE_PTR(it);
2761 }
2762
2763 static int osd_otable_it_get(const struct lu_env *env,
2764                              struct dt_it *di, const struct dt_key *key)
2765 {
2766         return 0;
2767 }
2768
2769 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2770 {
2771 }
2772
2773 static inline int
2774 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2775 {
2776         spin_lock(&scrub->os_lock);
2777         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2778             scrub->os_waiting ||
2779             !thread_is_running(&scrub->os_thread))
2780                 it->ooi_waiting = 0;
2781         else
2782                 it->ooi_waiting = 1;
2783         spin_unlock(&scrub->os_lock);
2784
2785         return !it->ooi_waiting;
2786 }
2787
2788 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2789 {
2790         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2791         struct osd_device       *dev    = it->ooi_dev;
2792         struct osd_scrub        *scrub  = &dev->od_scrub;
2793         struct osd_otable_cache *ooc    = &it->ooi_cache;
2794         struct ptlrpc_thread    *thread = &scrub->os_thread;
2795         struct l_wait_info       lwi    = { 0 };
2796         int                      rc;
2797         ENTRY;
2798
2799         LASSERT(it->ooi_user_ready);
2800
2801 again:
2802         if (!thread_is_running(thread) && !it->ooi_used_outside)
2803                 RETURN(1);
2804
2805         if (ooc->ooc_cached_items > 0) {
2806                 ooc->ooc_cached_items--;
2807                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2808                                         ~OSD_OTABLE_IT_CACHE_MASK;
2809                 RETURN(0);
2810         }
2811
2812         if (it->ooi_all_cached) {
2813                 l_wait_event(thread->t_ctl_waitq,
2814                              !thread_is_running(thread),
2815                              &lwi);
2816                 RETURN(1);
2817         }
2818
2819         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2820                 spin_lock(&scrub->os_lock);
2821                 scrub->os_waiting = 0;
2822                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2823                 spin_unlock(&scrub->os_lock);
2824         }
2825
2826         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2827                 l_wait_event(thread->t_ctl_waitq,
2828                              osd_otable_it_wakeup(scrub, it),
2829                              &lwi);
2830
2831         if (!thread_is_running(thread) && !it->ooi_used_outside)
2832                 RETURN(1);
2833
2834         rc = osd_otable_it_preload(env, it);
2835         if (rc >= 0)
2836                 goto again;
2837
2838         RETURN(rc);
2839 }
2840
2841 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2842                                         const struct dt_it *di)
2843 {
2844         return NULL;
2845 }
2846
2847 static int osd_otable_it_key_size(const struct lu_env *env,
2848                                   const struct dt_it *di)
2849 {
2850         return sizeof(__u64);
2851 }
2852
2853 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2854                              struct dt_rec *rec, __u32 attr)
2855 {
2856         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2857         struct osd_otable_cache *ooc = &it->ooi_cache;
2858
2859         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2860
2861         /* Filter out Invald FID already. */
2862         LASSERTF(fid_is_sane((struct lu_fid *)rec),
2863                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2864                  PFID((struct lu_fid *)rec),
2865                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2866
2867         return 0;
2868 }
2869
2870 static __u64 osd_otable_it_store(const struct lu_env *env,
2871                                  const struct dt_it *di)
2872 {
2873         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2874         struct osd_otable_cache *ooc = &it->ooi_cache;
2875         __u64                    hash;
2876
2877         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2878                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2879         else
2880                 hash = ooc->ooc_pos_preload;
2881         return hash;
2882 }
2883
2884 /**
2885  * Set the OSD layer iteration start position as the specified hash.
2886  */
2887 static int osd_otable_it_load(const struct lu_env *env,
2888                               const struct dt_it *di, __u64 hash)
2889 {
2890         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
2891         struct osd_device       *dev   = it->ooi_dev;
2892         struct osd_otable_cache *ooc   = &it->ooi_cache;
2893         struct osd_scrub        *scrub = &dev->od_scrub;
2894         int                      rc;
2895         ENTRY;
2896
2897         /* Forbid to set iteration position after iteration started. */
2898         if (it->ooi_user_ready)
2899                 RETURN(-EPERM);
2900
2901         LASSERT(!scrub->os_partial_scan);
2902
2903         if (hash > OSD_OTABLE_MAX_HASH)
2904                 hash = OSD_OTABLE_MAX_HASH;
2905
2906         ooc->ooc_pos_preload = hash;
2907         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2908                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2909
2910         it->ooi_user_ready = 1;
2911         if (!scrub->os_full_speed)
2912                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2913
2914         /* Unplug OSD layer iteration by the first next() call. */
2915         rc = osd_otable_it_next(env, (struct dt_it *)it);
2916
2917         RETURN(rc);
2918 }
2919
2920 static int osd_otable_it_key_rec(const struct lu_env *env,
2921                                  const struct dt_it *di, void *key_rec)
2922 {
2923         return 0;
2924 }
2925
2926 const struct dt_index_operations osd_otable_ops = {
2927         .dio_it = {
2928                 .init     = osd_otable_it_init,
2929                 .fini     = osd_otable_it_fini,
2930                 .get      = osd_otable_it_get,
2931                 .put      = osd_otable_it_put,
2932                 .next     = osd_otable_it_next,
2933                 .key      = osd_otable_it_key,
2934                 .key_size = osd_otable_it_key_size,
2935                 .rec      = osd_otable_it_rec,
2936                 .store    = osd_otable_it_store,
2937                 .load     = osd_otable_it_load,
2938                 .key_rec  = osd_otable_it_key_rec,
2939         }
2940 };
2941
2942 /* high priority inconsistent items list APIs */
2943
2944 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL  60
2945
2946 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2947                    int insert)
2948 {
2949         struct osd_inconsistent_item *oii;
2950         struct osd_scrub             *scrub  = &dev->od_scrub;
2951         struct ptlrpc_thread         *thread = &scrub->os_thread;
2952         int                           wakeup = 0;
2953         ENTRY;
2954
2955         OBD_ALLOC_PTR(oii);
2956         if (unlikely(oii == NULL))
2957                 RETURN(-ENOMEM);
2958
2959         INIT_LIST_HEAD(&oii->oii_list);
2960         oii->oii_cache = *oic;
2961         oii->oii_insert = insert;
2962
2963         if (scrub->os_partial_scan) {
2964                 __u64 now = cfs_time_current_sec();
2965
2966                 /* If there haven't been errors in a long time,
2967                  * decay old count until either the errors are
2968                  * gone or we reach the current interval. */
2969                 while (unlikely(scrub->os_bad_oimap_count > 0 &&
2970                                 scrub->os_bad_oimap_time +
2971                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
2972                         scrub->os_bad_oimap_count >>= 1;
2973                         scrub->os_bad_oimap_time +=
2974                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
2975                 }
2976
2977                 scrub->os_bad_oimap_time = now;
2978                 if (++scrub->os_bad_oimap_count >
2979                     dev->od_full_scrub_threshold_rate)
2980                         scrub->os_full_scrub = 1;
2981         }
2982
2983         spin_lock(&scrub->os_lock);
2984         if (unlikely(!thread_is_running(thread))) {
2985                 spin_unlock(&scrub->os_lock);
2986                 OBD_FREE_PTR(oii);
2987                 RETURN(-EAGAIN);
2988         }
2989
2990         if (list_empty(&scrub->os_inconsistent_items))
2991                 wakeup = 1;
2992         list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2993         spin_unlock(&scrub->os_lock);
2994
2995         if (wakeup != 0)
2996                 wake_up_all(&thread->t_ctl_waitq);
2997
2998         RETURN(0);
2999 }
3000
3001 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
3002                    struct osd_inode_id *id)
3003 {
3004         struct osd_scrub             *scrub = &dev->od_scrub;
3005         struct osd_inconsistent_item *oii;
3006         ENTRY;
3007
3008         spin_lock(&scrub->os_lock);
3009         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
3010                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
3011                         *id = oii->oii_cache.oic_lid;
3012                         spin_unlock(&scrub->os_lock);
3013                         RETURN(0);
3014                 }
3015         }
3016         spin_unlock(&scrub->os_lock);
3017
3018         RETURN(-ENOENT);
3019 }
3020
3021 /* OI scrub dump */
3022
3023 static const char *scrub_status_names[] = {
3024         "init",
3025         "scanning",
3026         "completed",
3027         "failed",
3028         "stopped",
3029         "paused",
3030         "crashed",
3031         NULL
3032 };
3033
3034 static const char *scrub_flags_names[] = {
3035         "recreated",
3036         "inconsistent",
3037         "auto",
3038         "upgrade",
3039         NULL
3040 };
3041
3042 static const char *scrub_param_names[] = {
3043         "failout",
3044         "dryrun",
3045         NULL
3046 };
3047
3048 static void scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
3049                             const char *prefix)
3050 {
3051         int flag;
3052         int i;
3053
3054         seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
3055
3056         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
3057                 if (flag & bits) {
3058                         bits &= ~flag;
3059                         seq_printf(m, "%s%c", names[i],
3060                                    bits != 0 ? ',' : '\n');
3061                 }
3062         }
3063 }
3064
3065 static void scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
3066 {
3067         if (time != 0)
3068                 seq_printf(m, "%s: "LPU64" seconds\n", prefix,
3069                            cfs_time_current_sec() - time);
3070         else
3071                 seq_printf(m, "%s: N/A\n", prefix);
3072 }
3073
3074 static void scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
3075 {
3076         if (pos != 0)
3077                 seq_printf(m, "%s: "LPU64"\n", prefix, pos);
3078         else
3079                 seq_printf(m, "%s: N/A\n", prefix);
3080 }
3081
3082 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
3083 {
3084         struct osd_scrub  *scrub   = &dev->od_scrub;
3085         struct scrub_file *sf      = &scrub->os_file;
3086         __u64              checked;
3087         __u64              speed;
3088
3089         down_read(&scrub->os_rwsem);
3090         seq_printf(m, "name: OI_scrub\n"
3091                    "magic: 0x%x\n"
3092                    "oi_files: %d\n"
3093                    "status: %s\n",
3094                    sf->sf_magic, (int)sf->sf_oi_count,
3095                    scrub_status_names[sf->sf_status]);
3096
3097         scrub_bits_dump(m, sf->sf_flags, scrub_flags_names, "flags");
3098
3099         scrub_bits_dump(m, sf->sf_param, scrub_param_names, "param");
3100
3101         scrub_time_dump(m, sf->sf_time_last_complete,
3102                         "time_since_last_completed");
3103
3104         scrub_time_dump(m, sf->sf_time_latest_start,
3105                         "time_since_latest_start");
3106
3107         scrub_time_dump(m, sf->sf_time_last_checkpoint,
3108                         "time_since_last_checkpoint");
3109
3110         scrub_pos_dump(m, sf->sf_pos_latest_start,
3111                         "latest_start_position");
3112
3113         scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
3114                         "last_checkpoint_position");
3115
3116         scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
3117                         "first_failure_position");
3118
3119         checked = sf->sf_items_checked + scrub->os_new_checked;
3120         seq_printf(m, "checked: "LPU64"\n"
3121                    "updated: "LPU64"\n"
3122                    "failed: "LPU64"\n"
3123                    "prior_updated: "LPU64"\n"
3124                    "noscrub: "LPU64"\n"
3125                    "igif: "LPU64"\n"
3126                    "success_count: %u\n",
3127                    checked, sf->sf_items_updated, sf->sf_items_failed,
3128                    sf->sf_items_updated_prior, sf->sf_items_noscrub,
3129                    sf->sf_items_igif, sf->sf_success_count);
3130
3131         speed = checked;
3132         if (thread_is_running(&scrub->os_thread)) {
3133                 cfs_duration_t duration = cfs_time_current() -
3134                                           scrub->os_time_last_checkpoint;
3135                 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
3136                                                      MSEC_PER_SEC);
3137                 __u32 rtime = sf->sf_run_time +
3138                               cfs_duration_sec(duration + HALF_SEC);
3139
3140                 if (duration != 0)
3141                         do_div(new_checked, duration);
3142                 if (rtime != 0)
3143                         do_div(speed, rtime);
3144                 seq_printf(m, "run_time: %u seconds\n"
3145                            "average_speed: "LPU64" objects/sec\n"
3146                            "real-time_speed: "LPU64" objects/sec\n"
3147                            "current_position: %u\n"
3148                            "lf_scanned: "LPU64"\n"
3149                            "lf_repaired: "LPU64"\n"
3150                            "lf_failed: "LPU64"\n",
3151                            rtime, speed, new_checked, scrub->os_pos_current,
3152                            scrub->os_lf_scanned, scrub->os_lf_repaired,
3153                            scrub->os_lf_failed);
3154         } else {
3155                 if (sf->sf_run_time != 0)
3156                         do_div(speed, sf->sf_run_time);
3157                 seq_printf(m, "run_time: %u seconds\n"
3158                            "average_speed: "LPU64" objects/sec\n"
3159                            "real-time_speed: N/A\n"
3160                            "current_position: N/A\n"
3161                            "lf_scanned: "LPU64"\n"
3162                            "lf_repaired: "LPU64"\n"
3163                            "lf_failed: "LPU64"\n",
3164                            sf->sf_run_time, speed, scrub->os_lf_scanned,
3165                            scrub->os_lf_repaired, scrub->os_lf_failed);
3166         }
3167
3168         up_read(&scrub->os_rwsem);
3169         return 0;
3170 }