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