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, 2016, Intel Corporation.
  24  */
  25 /*
  26  * lustre/osd-ldiskfs/osd_scrub.c
  27  *
  28  * Top-level entry points into osd module
  29  *
  30  * The OI scrub is used for rebuilding Object Index files when restores MDT from
  31  * file-level backup.
  32  *
  33  * The otable based iterator scans ldiskfs inode table to feed up layer LFSCK.
  34  *
  35  * Author: Fan Yong <yong.fan@whamcloud.com>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_LFSCK
  39
  40 #include <linux/kthread.h>
  41 #include <lustre/lustre_idl.h>
  42 #include <lustre_disk.h>
  43 #include <dt_object.h>
  44 #include <linux/xattr.h>
  45
  46 #include "osd_internal.h"
  47 #include "osd_oi.h"
  48 #include "osd_scrub.h"
  49
  50 #define HALF_SEC        msecs_to_jiffies(MSEC_PER_SEC >> 1)
  51
  52 #define OSD_OTABLE_MAX_HASH             0x00000000ffffffffULL
  53
  54 #define SCRUB_NEXT_BREAK        1 /* exit current loop and process next group */
  55 #define SCRUB_NEXT_CONTINUE     2 /* skip current object and process next bit */
  56 #define SCRUB_NEXT_EXIT         3 /* exit all the loops */
  57 #define SCRUB_NEXT_WAIT         4 /* wait for free cache slot */
  58 #define SCRUB_NEXT_CRASH        5 /* simulate system crash during OI scrub */
  59 #define SCRUB_NEXT_FATAL        6 /* simulate failure during OI scrub */
  60 #define SCRUB_NEXT_NOSCRUB      7 /* new created object, no scrub on it */
  61 #define SCRUB_NEXT_NOLMA        8 /* the inode has no FID-in-LMA */
  62 #define SCRUB_NEXT_OSTOBJ       9 /* for OST-object */
  63 #define SCRUB_NEXT_OSTOBJ_OLD   10 /* old OST-object, no LMA or no FID-on-OST
  64                                     * flags in LMA */
  65
  66 /* misc functions */
  67
  68 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
  69 {
  70         return container_of0(scrub, struct osd_device, od_scrub);
  71 }
  72
  73 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
  74 {
  75         return osd_sb(osd_scrub2dev(scrub));
  76 }
  77
  78 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
  79                                        struct osd_otable_cache *ooc)
  80 {
  81         return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
  82 }
  83
  84 static inline const char *osd_scrub2name(struct osd_scrub *scrub)
  85 {
  86         return LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
  87 }
  88
  89 /**
  90  * update/insert/delete the specified OI mapping (@fid @id) according to the ops
  91  *
  92  * \retval   1, changed nothing
  93  * \retval   0, changed successfully
  94  * \retval -ve, on error
  95  */
  96 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
  97                                      struct osd_device *dev,
  98                                      const struct lu_fid *fid,
  99                                      const struct osd_inode_id *id,
 100                                      int ops, bool force,
 101                                      enum oi_check_flags flags, bool *exist)
 102 {
 103         handle_t *th;
 104         int       rc;
 105         ENTRY;
 106
 107         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
 108                 RETURN(0);
 109
 110         /* DTO_INDEX_INSERT is enough for other two ops:
 111          * delete/update, but save stack. */
 112         th = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 113                                 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
 114         if (IS_ERR(th)) {
 115                 rc = PTR_ERR(th);
 116                 CDEBUG(D_LFSCK, "%s: fail to start trans for scrub op %d "
 117                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 118                        PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1,
 119                        rc);
 120                 RETURN(rc);
 121         }
 122
 123         switch (ops) {
 124         case DTO_INDEX_UPDATE:
 125                 rc = osd_oi_update(info, dev, fid, id, th, flags);
 126                 if (unlikely(rc == -ENOENT)) {
 127                         /* Some unlink thread may removed the OI mapping. */
 128                         rc = 1;
 129                 }
 130                 break;
 131         case DTO_INDEX_INSERT:
 132                 rc = osd_oi_insert(info, dev, fid, id, th, flags, exist);
 133                 if (unlikely(rc == -EEXIST)) {
 134                         rc = 1;
 135                         /* XXX: There are trouble things when adding OI
 136                          *      mapping for IGIF object, which may cause
 137                          *      multiple objects to be mapped to the same
 138                          *      IGIF formatted FID. Consider the following
 139                          *      situations:
 140                          *
 141                          *      1) The MDT is upgrading from 1.8 device.
 142                          *      The OI scrub generates IGIF FID1 for the
 143                          *      OBJ1 and adds the OI mapping.
 144                          *
 145                          *      2) For some reason, the OI scrub does not
 146                          *      process all the IGIF objects completely.
 147                          *
 148                          *      3) The MDT is backuped and restored against
 149                          *      this device.
 150                          *
 151                          *      4) When the MDT mounts up, the OI scrub will
 152                          *      try to rebuild the OI files. For some IGIF
 153                          *      object, OBJ2, which was not processed by the
 154                          *      OI scrub before the backup/restore, and the
 155                          *      new generated IGIF formatted FID may be just
 156                          *      the FID1, the same as OBJ1.
 157                          *
 158                          *      Under such case, the OI scrub cannot know how
 159                          *      to generate new FID for the OBJ2.
 160                          *
 161                          *      Currently, we do nothing for that. One possible
 162                          *      solution is to generate new normal FID for the
 163                          *      conflict object.
 164                          *
 165                          *      Anyway, it is rare, only exists in theory. */
 166                 }
 167                 break;
 168         case DTO_INDEX_DELETE:
 169                 rc = osd_oi_delete(info, dev, fid, th, flags);
 170                 if (rc == -ENOENT) {
 171                         /* It is normal that the unlink thread has removed the
 172                          * OI mapping already. */
 173                         rc = 1;
 174                 }
 175                 break;
 176         default:
 177                 LASSERTF(0, "Unexpected ops %d\n", ops);
 178                 break;
 179         }
 180
 181         ldiskfs_journal_stop(th);
 182         if (rc < 0)
 183                 CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d "
 184                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 185                        PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1,
 186                        rc);
 187
 188         RETURN(rc);
 189 }
 190
 191 /* OI_scrub file ops */
 192
 193 static void osd_scrub_file_to_cpu(struct scrub_file *des,
 194                                   struct scrub_file *src)
 195 {
 196         memcpy(des->sf_uuid, src->sf_uuid, 16);
 197         des->sf_flags   = le64_to_cpu(src->sf_flags);
 198         des->sf_magic   = le32_to_cpu(src->sf_magic);
 199         des->sf_status  = le16_to_cpu(src->sf_status);
 200         des->sf_param   = le16_to_cpu(src->sf_param);
 201         des->sf_time_last_complete      =
 202                                 le64_to_cpu(src->sf_time_last_complete);
 203         des->sf_time_latest_start       =
 204                                 le64_to_cpu(src->sf_time_latest_start);
 205         des->sf_time_last_checkpoint    =
 206                                 le64_to_cpu(src->sf_time_last_checkpoint);
 207         des->sf_pos_latest_start        =
 208                                 le64_to_cpu(src->sf_pos_latest_start);
 209         des->sf_pos_last_checkpoint     =
 210                                 le64_to_cpu(src->sf_pos_last_checkpoint);
 211         des->sf_pos_first_inconsistent  =
 212                                 le64_to_cpu(src->sf_pos_first_inconsistent);
 213         des->sf_items_checked           =
 214                                 le64_to_cpu(src->sf_items_checked);
 215         des->sf_items_updated           =
 216                                 le64_to_cpu(src->sf_items_updated);
 217         des->sf_items_failed            =
 218                                 le64_to_cpu(src->sf_items_failed);
 219         des->sf_items_updated_prior     =
 220                                 le64_to_cpu(src->sf_items_updated_prior);
 221         des->sf_run_time        = le32_to_cpu(src->sf_run_time);
 222         des->sf_success_count   = le32_to_cpu(src->sf_success_count);
 223         des->sf_oi_count        = le16_to_cpu(src->sf_oi_count);
 224         des->sf_internal_flags  = le16_to_cpu(src->sf_internal_flags);
 225         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 226 }
 227
 228 static void osd_scrub_file_to_le(struct scrub_file *des,
 229                                  struct scrub_file *src)
 230 {
 231         memcpy(des->sf_uuid, src->sf_uuid, 16);
 232         des->sf_flags   = cpu_to_le64(src->sf_flags);
 233         des->sf_magic   = cpu_to_le32(src->sf_magic);
 234         des->sf_status  = cpu_to_le16(src->sf_status);
 235         des->sf_param   = cpu_to_le16(src->sf_param);
 236         des->sf_time_last_complete      =
 237                                 cpu_to_le64(src->sf_time_last_complete);
 238         des->sf_time_latest_start       =
 239                                 cpu_to_le64(src->sf_time_latest_start);
 240         des->sf_time_last_checkpoint    =
 241                                 cpu_to_le64(src->sf_time_last_checkpoint);
 242         des->sf_pos_latest_start        =
 243                                 cpu_to_le64(src->sf_pos_latest_start);
 244         des->sf_pos_last_checkpoint     =
 245                                 cpu_to_le64(src->sf_pos_last_checkpoint);
 246         des->sf_pos_first_inconsistent  =
 247                                 cpu_to_le64(src->sf_pos_first_inconsistent);
 248         des->sf_items_checked           =
 249                                 cpu_to_le64(src->sf_items_checked);
 250         des->sf_items_updated           =
 251                                 cpu_to_le64(src->sf_items_updated);
 252         des->sf_items_failed            =
 253                                 cpu_to_le64(src->sf_items_failed);
 254         des->sf_items_updated_prior     =
 255                                 cpu_to_le64(src->sf_items_updated_prior);
 256         des->sf_run_time        = cpu_to_le32(src->sf_run_time);
 257         des->sf_success_count   = cpu_to_le32(src->sf_success_count);
 258         des->sf_oi_count        = cpu_to_le16(src->sf_oi_count);
 259         des->sf_internal_flags  = cpu_to_le16(src->sf_internal_flags);
 260         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 261 }
 262
 263 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
 264 {
 265         struct scrub_file *sf = &scrub->os_file;
 266
 267         memset(sf, 0, sizeof(*sf));
 268         memcpy(sf->sf_uuid, uuid, 16);
 269         sf->sf_magic = SCRUB_MAGIC_V1;
 270         sf->sf_status = SS_INIT;
 271 }
 272
 273 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
 274 {
 275         struct scrub_file *sf = &scrub->os_file;
 276
 277         CDEBUG(D_LFSCK, "%.16s: reset OI scrub file, old flags = "
 278                "%#llx, add flags = %#llx\n",
 279                osd_scrub2name(scrub), sf->sf_flags, flags);
 280
 281         memcpy(sf->sf_uuid, uuid, 16);
 282         sf->sf_status = SS_INIT;
 283         sf->sf_flags |= flags;
 284         sf->sf_flags &= ~SF_AUTO;
 285         sf->sf_run_time = 0;
 286         sf->sf_time_latest_start = 0;
 287         sf->sf_time_last_checkpoint = 0;
 288         sf->sf_pos_latest_start = 0;
 289         sf->sf_pos_last_checkpoint = 0;
 290         sf->sf_pos_first_inconsistent = 0;
 291         sf->sf_items_checked = 0;
 292         sf->sf_items_updated = 0;
 293         sf->sf_items_failed = 0;
 294         if (!scrub->os_in_join)
 295                 sf->sf_items_updated_prior = 0;
 296
 297         sf->sf_items_noscrub = 0;
 298         sf->sf_items_igif = 0;
 299 }
 300
 301 static int osd_scrub_file_load(struct osd_scrub *scrub)
 302 {
 303         loff_t  pos  = 0;
 304         int     len  = sizeof(scrub->os_file_disk);
 305         int     rc;
 306
 307         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
 308         if (rc == len) {
 309                 struct scrub_file *sf = &scrub->os_file;
 310
 311                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
 312                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
 313                         CDEBUG(D_LFSCK, "%.16s: invalid scrub magic "
 314                                "0x%x != 0x%x\n", osd_scrub2name(scrub),
 315                                sf->sf_magic, SCRUB_MAGIC_V1);
 316                         /* Process it as new scrub file. */
 317                         rc = -ENOENT;
 318                 } else {
 319                         rc = 0;
 320                 }
 321         } else if (rc != 0) {
 322                 CDEBUG(D_LFSCK, "%.16s: fail to load scrub file, "
 323                        "expected = %d: rc = %d\n",
 324                        osd_scrub2name(scrub), len, rc);
 325                 if (rc > 0)
 326                         rc = -EFAULT;
 327         } else {
 328                 /* return -ENOENT for empty scrub file case. */
 329                 rc = -ENOENT;
 330         }
 331
 332         return rc;
 333 }
 334
 335 int osd_scrub_file_store(struct osd_scrub *scrub)
 336 {
 337         struct osd_device *dev;
 338         handle_t          *jh;
 339         loff_t             pos     = 0;
 340         int                len     = sizeof(scrub->os_file_disk);
 341         int                credits;
 342         int                rc;
 343
 344         dev = container_of0(scrub, struct osd_device, od_scrub);
 345         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
 346                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
 347         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits);
 348         if (IS_ERR(jh)) {
 349                 rc = PTR_ERR(jh);
 350                 CDEBUG(D_LFSCK, "%.16s: fail to start trans for scrub store: "
 351                        "rc = %d\n", osd_scrub2name(scrub), rc);
 352                 return rc;
 353         }
 354
 355         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
 356         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
 357                                       len, 0, &pos, jh);
 358         ldiskfs_journal_stop(jh);
 359         if (rc != 0)
 360                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file, "
 361                        "expected = %d: rc = %d\n",
 362                        osd_scrub2name(scrub), len, rc);
 363
 364         scrub->os_time_last_checkpoint = cfs_time_current();
 365         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
 366                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
 367         return rc;
 368 }
 369
 370 static int
 371 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
 372                      struct inode *inode, const struct lu_fid *fid)
 373 {
 374         struct filter_fid_old   *ff      = &info->oti_ff;
 375         struct dentry           *dentry  = &info->oti_obj_dentry;
 376         struct lu_fid           *tfid    = &info->oti_fid;
 377         handle_t                *jh;
 378         int                      size    = 0;
 379         int                      rc;
 380         bool                     reset   = false;
 381         ENTRY;
 382
 383         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
 384                 RETURN(0);
 385
 386         if (fid_is_idif(fid) && dev->od_index_in_idif == 0) {
 387                 struct ost_id *oi = &info->oti_ostid;
 388
 389                 fid_to_ostid(fid, oi);
 390                 ostid_to_fid(tfid, oi, 0);
 391         } else {
 392                 *tfid = *fid;
 393         }
 394
 395         /* We want the LMA to fit into the 256-byte OST inode, so operate
 396          * as following:
 397          * 1) read old XATTR_NAME_FID and save the parent FID;
 398          * 2) delete the old XATTR_NAME_FID;
 399          * 3) make new LMA and add it;
 400          * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
 401          *
 402          * Making the LMA to fit into the 256-byte OST inode can save time for
 403          * normal osd_check_lma() and for other OI scrub scanning in future.
 404          * So it is worth to make some slow conversion here. */
 405         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 406                                 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
 407         if (IS_ERR(jh)) {
 408                 rc = PTR_ERR(jh);
 409                 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
 410                        DFID": rc = %d\n", osd_name(dev), PFID(tfid), rc);
 411                 RETURN(rc);
 412         }
 413
 414         /* 1) read old XATTR_NAME_FID and save the parent FID */
 415         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 416         if (rc == sizeof(*ff)) {
 417                 /* 2) delete the old XATTR_NAME_FID */
 418                 ll_vfs_dq_init(inode);
 419                 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
 420                 if (rc)
 421                         GOTO(stop, rc);
 422
 423                 reset = true;
 424         } else if (rc != -ENODATA && rc != sizeof(struct filter_fid)) {
 425                 GOTO(stop, rc = -EINVAL);
 426         }
 427
 428         /* 3) make new LMA and add it */
 429         rc = osd_ea_fid_set(info, inode, tfid, LMAC_FID_ON_OST, 0);
 430         if (reset) {
 431                 if (rc)
 432                         /* If failed, we should try to add the old back. */
 433                         size = sizeof(*ff);
 434                 else
 435                         /* The new PFID EA will only contains ::ff_parent */
 436                         size = sizeof(ff->ff_parent);
 437         }
 438
 439         /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
 440         if (size > 0) {
 441                 int rc1;
 442
 443                 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
 444                                       XATTR_CREATE);
 445                 if (rc1 != 0 && rc == 0)
 446                         rc = rc1;
 447         }
 448
 449         GOTO(stop, rc);
 450
 451 stop:
 452         ldiskfs_journal_stop(jh);
 453         if (rc < 0)
 454                 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
 455                        osd_name(dev), PFID(tfid), rc);
 456         return rc;
 457 }
 458
 459 static int
 460 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
 461                        struct osd_idmap_cache *oic, int val)
 462 {
 463         struct osd_scrub             *scrub  = &dev->od_scrub;
 464         struct scrub_file            *sf     = &scrub->os_file;
 465         struct lu_fid                *fid    = &oic->oic_fid;
 466         struct osd_inode_id          *lid    = &oic->oic_lid;
 467         struct osd_inode_id          *lid2   = &info->oti_id;
 468         struct osd_inconsistent_item *oii    = NULL;
 469         struct inode                 *inode  = NULL;
 470         int                           ops    = DTO_INDEX_UPDATE;
 471         int                           rc;
 472         bool                          converted = false;
 473         bool                          exist     = false;
 474         ENTRY;
 475
 476         down_write(&scrub->os_rwsem);
 477         scrub->os_new_checked++;
 478         if (val < 0)
 479                 GOTO(out, rc = val);
 480
 481         if (scrub->os_in_prior)
 482                 oii = list_entry(oic, struct osd_inconsistent_item,
 483                                  oii_cache);
 484
 485         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
 486                 GOTO(out, rc = 0);
 487
 488         if (fid_is_igif(fid))
 489                 sf->sf_items_igif++;
 490
 491         if (val == SCRUB_NEXT_OSTOBJ_OLD) {
 492                 inode = osd_iget(info, dev, lid);
 493                 if (IS_ERR(inode)) {
 494                         rc = PTR_ERR(inode);
 495                         /* Someone removed the inode. */
 496                         if (rc == -ENOENT || rc == -ESTALE)
 497                                 rc = 0;
 498                         GOTO(out, rc);
 499                 }
 500
 501                 /* The inode has been reused as EA inode, ignore it. */
 502                 if (unlikely(osd_is_ea_inode(inode)))
 503                         GOTO(out, rc = 0);
 504
 505                 sf->sf_flags |= SF_UPGRADE;
 506                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
 507                 dev->od_check_ff = 1;
 508                 rc = osd_scrub_convert_ff(info, dev, inode, fid);
 509                 if (rc != 0)
 510                         GOTO(out, rc);
 511
 512                 converted = true;
 513         }
 514
 515         if ((val == SCRUB_NEXT_NOLMA) &&
 516             (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
 517                 GOTO(out, rc = 0);
 518
 519         if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA)) {
 520                 ops = DTO_INDEX_INSERT;
 521
 522                 goto iget;
 523         }
 524
 525         rc = osd_oi_lookup(info, dev, fid, lid2,
 526                 (val == SCRUB_NEXT_OSTOBJ ||
 527                  val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 528         if (rc != 0) {
 529                 if (rc == -ENOENT)
 530                         ops = DTO_INDEX_INSERT;
 531                 else if (rc != -ESTALE)
 532                         GOTO(out, rc);
 533
 534 iget:
 535                 if (inode == NULL) {
 536                         inode = osd_iget(info, dev, lid);
 537                         if (IS_ERR(inode)) {
 538                                 rc = PTR_ERR(inode);
 539                                 /* Someone removed the inode. */
 540                                 if (rc == -ENOENT || rc == -ESTALE)
 541                                         rc = 0;
 542                                 GOTO(out, rc);
 543                         }
 544
 545                         /* The inode has been reused as EA inode, ignore it. */
 546                         if (unlikely(osd_is_ea_inode(inode)))
 547                                 GOTO(out, rc = 0);
 548                 }
 549
 550                 if (!scrub->os_partial_scan)
 551                         scrub->os_full_speed = 1;
 552
 553                 switch (val) {
 554                 case SCRUB_NEXT_NOLMA:
 555                         sf->sf_flags |= SF_UPGRADE;
 556                         if (!(sf->sf_param & SP_DRYRUN)) {
 557                                 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
 558                                 if (rc != 0)
 559                                         GOTO(out, rc);
 560                         }
 561
 562                         if (!(sf->sf_flags & SF_INCONSISTENT))
 563                                 dev->od_igif_inoi = 0;
 564                         break;
 565                 case SCRUB_NEXT_OSTOBJ:
 566                         sf->sf_flags |= SF_INCONSISTENT;
 567                 case SCRUB_NEXT_OSTOBJ_OLD:
 568                         break;
 569                 default:
 570                         break;
 571                 }
 572         } else if (osd_id_eq(lid, lid2)) {
 573                 if (converted)
 574                         sf->sf_items_updated++;
 575
 576                 GOTO(out, rc = 0);
 577         } else {
 578                 if (!scrub->os_partial_scan)
 579                         scrub->os_full_speed = 1;
 580
 581                 sf->sf_flags |= SF_INCONSISTENT;
 582
 583                 /* XXX: If the device is restored from file-level backup, then
 584                  *      some IGIFs may have been already in OI files, and some
 585                  *      may be not yet. Means upgrading from 1.8 may be partly
 586                  *      processed, but some clients may hold some immobilized
 587                  *      IGIFs, and use them to access related objects. Under
 588                  *      such case, OSD does not know whether an given IGIF has
 589                  *      been processed or to be processed, and it also cannot
 590                  *      generate local ino#/gen# directly from the immobilized
 591                  *      IGIF because of the backup/restore. Then force OSD to
 592                  *      lookup the given IGIF in OI files, and if no entry,
 593                  *      then ask the client to retry after upgrading completed.
 594                  *      No better choice. */
 595                 dev->od_igif_inoi = 1;
 596         }
 597
 598         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
 599                         (val == SCRUB_NEXT_OSTOBJ ||
 600                          val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0,
 601                         &exist);
 602         if (rc == 0) {
 603                 if (scrub->os_in_prior)
 604                         sf->sf_items_updated_prior++;
 605                 else
 606                         sf->sf_items_updated++;
 607
 608                 if (ops == DTO_INDEX_INSERT && val == 0 && !exist) {
 609                         int idx = osd_oi_fid2idx(dev, fid);
 610
 611                         sf->sf_flags |= SF_RECREATED;
 612                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 613                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 614                 }
 615         }
 616
 617         GOTO(out, rc);
 618
 619 out:
 620         if (rc < 0) {
 621                 sf->sf_items_failed++;
 622                 if (sf->sf_pos_first_inconsistent == 0 ||
 623                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 624                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 625         } else {
 626                 rc = 0;
 627         }
 628
 629         /* There may be conflict unlink during the OI scrub,
 630          * if happend, then remove the new added OI mapping. */
 631         if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
 632             unlikely(ldiskfs_test_inode_state(inode,
 633                                               LDISKFS_STATE_LUSTRE_DESTROY)))
 634                 osd_scrub_refresh_mapping(info, dev, fid, lid,
 635                                 DTO_INDEX_DELETE, false,
 636                                 (val == SCRUB_NEXT_OSTOBJ ||
 637                                  val == SCRUB_NEXT_OSTOBJ_OLD) ?
 638                                 OI_KNOWN_ON_OST : 0, NULL);
 639         up_write(&scrub->os_rwsem);
 640
 641         if (inode != NULL && !IS_ERR(inode))
 642                 iput(inode);
 643
 644         if (oii != NULL) {
 645                 spin_lock(&scrub->os_lock);
 646                 if (likely(!list_empty(&oii->oii_list)))
 647                         list_del(&oii->oii_list);
 648                 spin_unlock(&scrub->os_lock);
 649
 650                 OBD_FREE_PTR(oii);
 651         }
 652
 653         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 654 }
 655
 656 /* OI scrub APIs */
 657
 658 static int osd_scrub_prep(struct osd_device *dev)
 659 {
 660         struct osd_scrub     *scrub  = &dev->od_scrub;
 661         struct ptlrpc_thread *thread = &scrub->os_thread;
 662         struct scrub_file    *sf     = &scrub->os_file;
 663         __u32                 flags  = scrub->os_start_flags;
 664         int                   rc;
 665         bool                  drop_dryrun = false;
 666         ENTRY;
 667
 668         CDEBUG(D_LFSCK, "%.16s: OI scrub prep, flags = 0x%x\n",
 669                osd_scrub2name(scrub), flags);
 670
 671         down_write(&scrub->os_rwsem);
 672         if (flags & SS_SET_FAILOUT)
 673                 sf->sf_param |= SP_FAILOUT;
 674         else if (flags & SS_CLEAR_FAILOUT)
 675                 sf->sf_param &= ~SP_FAILOUT;
 676
 677         if (flags & SS_SET_DRYRUN) {
 678                 sf->sf_param |= SP_DRYRUN;
 679         } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
 680                 sf->sf_param &= ~SP_DRYRUN;
 681                 drop_dryrun = true;
 682         }
 683
 684         if (flags & SS_RESET)
 685                 osd_scrub_file_reset(scrub,
 686                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
 687
 688         if (flags & SS_AUTO_FULL) {
 689                 scrub->os_full_speed = 1;
 690                 scrub->os_partial_scan = 0;
 691                 sf->sf_flags |= SF_AUTO;
 692         } else if (flags & SS_AUTO_PARTIAL) {
 693                 scrub->os_full_speed = 0;
 694                 scrub->os_partial_scan = 1;
 695                 sf->sf_flags |= SF_AUTO;
 696         } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
 697                                    SF_UPGRADE)) {
 698                 scrub->os_full_speed = 1;
 699                 scrub->os_partial_scan = 0;
 700         } else {
 701                 scrub->os_full_speed = 0;
 702                 scrub->os_partial_scan = 0;
 703         }
 704
 705         spin_lock(&scrub->os_lock);
 706         scrub->os_in_prior = 0;
 707         scrub->os_waiting = 0;
 708         scrub->os_paused = 0;
 709         scrub->os_in_join = 0;
 710         scrub->os_full_scrub = 0;
 711         spin_unlock(&scrub->os_lock);
 712         scrub->os_new_checked = 0;
 713         if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
 714                 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
 715         else if (sf->sf_pos_last_checkpoint != 0)
 716                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 717         else
 718                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 719
 720         scrub->os_pos_current = sf->sf_pos_latest_start;
 721         sf->sf_status = SS_SCANNING;
 722         sf->sf_time_latest_start = cfs_time_current_sec();
 723         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 724         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
 725         rc = osd_scrub_file_store(scrub);
 726         if (rc == 0) {
 727                 spin_lock(&scrub->os_lock);
 728                 thread_set_flags(thread, SVC_RUNNING);
 729                 spin_unlock(&scrub->os_lock);
 730                 wake_up_all(&thread->t_ctl_waitq);
 731         }
 732         up_write(&scrub->os_rwsem);
 733
 734         RETURN(rc);
 735 }
 736
 737 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
 738 {
 739         struct scrub_file *sf = &scrub->os_file;
 740         int                rc;
 741
 742         if (likely(cfs_time_before(cfs_time_current(),
 743                                    scrub->os_time_next_checkpoint) ||
 744                    scrub->os_new_checked == 0))
 745                 return 0;
 746
 747         down_write(&scrub->os_rwsem);
 748         sf->sf_items_checked += scrub->os_new_checked;
 749         scrub->os_new_checked = 0;
 750         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 751         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 752         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 753                                             scrub->os_time_last_checkpoint);
 754         rc = osd_scrub_file_store(scrub);
 755         up_write(&scrub->os_rwsem);
 756
 757         return rc;
 758 }
 759
 760 static int osd_scrub_post(struct osd_scrub *scrub, int result)
 761 {
 762         struct scrub_file *sf = &scrub->os_file;
 763         int rc;
 764         ENTRY;
 765
 766         CDEBUG(D_LFSCK, "%.16s: OI scrub post, result = %d\n",
 767                osd_scrub2name(scrub), result);
 768
 769         down_write(&scrub->os_rwsem);
 770         spin_lock(&scrub->os_lock);
 771         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 772         spin_unlock(&scrub->os_lock);
 773         if (scrub->os_new_checked > 0) {
 774                 sf->sf_items_checked += scrub->os_new_checked;
 775                 scrub->os_new_checked = 0;
 776                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 777         }
 778         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 779         if (result > 0) {
 780                 struct osd_device *dev =
 781                         container_of0(scrub, struct osd_device, od_scrub);
 782
 783                 dev->od_igif_inoi = 1;
 784                 dev->od_check_ff = 0;
 785                 sf->sf_status = SS_COMPLETED;
 786                 if (!(sf->sf_param & SP_DRYRUN)) {
 787                         memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 788                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 789                                           SF_UPGRADE | SF_AUTO);
 790                 }
 791                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 792                 sf->sf_success_count++;
 793         } else if (result == 0) {
 794                 if (scrub->os_paused)
 795                         sf->sf_status = SS_PAUSED;
 796                 else
 797                         sf->sf_status = SS_STOPPED;
 798         } else {
 799                 sf->sf_status = SS_FAILED;
 800         }
 801         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 802                                             scrub->os_time_last_checkpoint);
 803         rc = osd_scrub_file_store(scrub);
 804         up_write(&scrub->os_rwsem);
 805
 806         RETURN(rc < 0 ? rc : result);
 807 }
 808
 809 /* iteration engine */
 810
 811 struct osd_iit_param {
 812         struct super_block *sb;
 813         struct buffer_head *bitmap;
 814         ldiskfs_group_t bg;
 815         __u32 gbase;
 816         __u32 offset;
 817 };
 818
 819 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 820                                    struct osd_device *dev,
 821                                    struct osd_iit_param *param,
 822                                    struct osd_idmap_cache **oic,
 823                                    const bool noslot);
 824
 825 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 826                                    struct osd_device *dev,
 827                                    struct osd_iit_param *param,
 828                                    struct osd_idmap_cache *oic,
 829                                    bool *noslot, int rc);
 830
 831 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 832 {
 833         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 834                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 835         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 836                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 837                 return SCRUB_NEXT_BREAK;
 838         } else {
 839                 *pos = param->gbase + param->offset;
 840                 return 0;
 841         }
 842 }
 843
 844 /**
 845  * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
 846  * \retval 0: FID-on-MDT
 847  */
 848 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
 849                                       struct osd_device *dev,
 850                                       struct lu_fid *fid)
 851 {
 852         /* XXX: The initial OI scrub will scan the top level /O to generate
 853          *      a small local FLDB according to the <seq>. If the given FID
 854          *      is in the local FLDB, then it is FID-on-OST; otherwise it's
 855          *      quite possible for FID-on-MDT. */
 856         if (dev->od_is_ost)
 857                 return SCRUB_NEXT_OSTOBJ_OLD;
 858         else
 859                 return 0;
 860 }
 861
 862 static int osd_scrub_get_fid(struct osd_thread_info *info,
 863                              struct osd_device *dev, struct inode *inode,
 864                              struct lu_fid *fid, bool scrub)
 865 {
 866         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
 867         int rc;
 868         bool has_lma = false;
 869
 870         rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
 871                          &info->oti_ost_attrs);
 872         if (rc == 0) {
 873                 has_lma = true;
 874                 if (lma->lma_compat & LMAC_NOT_IN_OI ||
 875                     lma->lma_incompat & LMAI_AGENT)
 876                         return SCRUB_NEXT_CONTINUE;
 877
 878                 *fid = lma->lma_self_fid;
 879                 if (!scrub)
 880                         return 0;
 881
 882                 if (lma->lma_compat & LMAC_FID_ON_OST)
 883                         return SCRUB_NEXT_OSTOBJ;
 884
 885                 if (fid_is_idif(fid))
 886                         return SCRUB_NEXT_OSTOBJ_OLD;
 887
 888                 /* For local object. */
 889                 if (fid_is_internal(fid))
 890                         return 0;
 891
 892                 /* For external visible MDT-object with non-normal FID. */
 893                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
 894                         return 0;
 895
 896                 /* For the object with normal FID, it may be MDT-object,
 897                  * or may be 2.4 OST-object, need further distinguish.
 898                  * Fall through to next section. */
 899         }
 900
 901         if (rc == -ENODATA || rc == 0) {
 902                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
 903                 if (rc == 0) {
 904                         if (scrub)
 905                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
 906                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 907                         return rc;
 908                 }
 909
 910                 if (rc > 0) {
 911                         if (!has_lma)
 912                                 /* It is FID-on-OST, but we do not know how
 913                                  * to generate its FID, ignore it directly. */
 914                                 rc = SCRUB_NEXT_CONTINUE;
 915                         else
 916                                 /* It is 2.4 OST-object. */
 917                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 918                         return rc;
 919                 }
 920
 921                 if (rc != -ENODATA)
 922                         return rc;
 923
 924                 if (!has_lma) {
 925                         if (dev->od_scrub.os_convert_igif) {
 926                                 lu_igif_build(fid, inode->i_ino,
 927                                               inode->i_generation);
 928                                 if (scrub)
 929                                         rc = SCRUB_NEXT_NOLMA;
 930                                 else
 931                                         rc = 0;
 932                         } else {
 933                                 /* It may be FID-on-OST, or may be FID for
 934                                  * non-MDT0, anyway, we do not know how to
 935                                  * generate its FID, ignore it directly. */
 936                                 rc = SCRUB_NEXT_CONTINUE;
 937                         }
 938                         return rc;
 939                 }
 940
 941                 /* For OI scrub case only: the object has LMA but has no ff
 942                  * (or ff crashed). It may be MDT-object, may be OST-object
 943                  * with crashed ff. The last check is local FLDB. */
 944                 rc = osd_scrub_check_local_fldb(info, dev, fid);
 945         }
 946
 947         return rc;
 948 }
 949
 950 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 951                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 952                         struct super_block *sb, bool scrub)
 953 {
 954         struct inode *inode;
 955         int           rc;
 956         ENTRY;
 957
 958         /* Not handle the backend root object and agent parent object.
 959          * They are neither visible to namespace nor have OI mappings. */
 960         if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
 961                      pos == osd_remote_parent_ino(dev)))
 962                 RETURN(SCRUB_NEXT_CONTINUE);
 963
 964         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 965         inode = osd_iget(info, dev, lid);
 966         if (IS_ERR(inode)) {
 967                 rc = PTR_ERR(inode);
 968                 /* The inode may be removed after bitmap searching, or the
 969                  * file is new created without inode initialized yet. */
 970                 if (rc == -ENOENT || rc == -ESTALE)
 971                         RETURN(SCRUB_NEXT_CONTINUE);
 972
 973                 CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: "
 974                        "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name,
 975                        pos, rc);
 976                 RETURN(rc);
 977         }
 978
 979         /* It is an EA inode, no OI mapping for it, skip it. */
 980         if (osd_is_ea_inode(inode))
 981                 GOTO(put, rc = SCRUB_NEXT_CONTINUE);
 982
 983         if (scrub &&
 984             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 985                 /* Only skip it for the first OI scrub accessing. */
 986                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 987                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
 988         }
 989
 990         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
 991
 992         GOTO(put, rc);
 993
 994 put:
 995         iput(inode);
 996         return rc;
 997 }
 998
 999 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
1000                           struct osd_iit_param *param,
1001                           struct osd_idmap_cache **oic, const bool noslot)
1002 {
1003         struct osd_scrub     *scrub  = &dev->od_scrub;
1004         struct ptlrpc_thread *thread = &scrub->os_thread;
1005         struct lu_fid        *fid;
1006         struct osd_inode_id  *lid;
1007         int                   rc;
1008
1009         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
1010                 struct l_wait_info lwi;
1011
1012                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
1013                 if (likely(lwi.lwi_timeout > 0))
1014                         l_wait_event(thread->t_ctl_waitq,
1015                                 !list_empty(&scrub->os_inconsistent_items) ||
1016                                 !thread_is_running(thread),
1017                                 &lwi);
1018         }
1019
1020         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
1021                 spin_lock(&scrub->os_lock);
1022                 thread_set_flags(thread, SVC_STOPPING);
1023                 spin_unlock(&scrub->os_lock);
1024                 return SCRUB_NEXT_CRASH;
1025         }
1026
1027         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
1028                 return SCRUB_NEXT_FATAL;
1029
1030         if (unlikely(!thread_is_running(thread)))
1031                 return SCRUB_NEXT_EXIT;
1032
1033         if (!list_empty(&scrub->os_inconsistent_items)) {
1034                 spin_lock(&scrub->os_lock);
1035                 if (likely(!list_empty(&scrub->os_inconsistent_items))) {
1036                         struct osd_inconsistent_item *oii;
1037
1038                         oii = list_entry(scrub->os_inconsistent_items.next,
1039                                 struct osd_inconsistent_item, oii_list);
1040                         spin_unlock(&scrub->os_lock);
1041
1042                         *oic = &oii->oii_cache;
1043                         scrub->os_in_prior = 1;
1044
1045                         return 0;
1046                 }
1047                 spin_unlock(&scrub->os_lock);
1048         }
1049
1050         if (noslot)
1051                 return SCRUB_NEXT_WAIT;
1052
1053         rc = osd_iit_next(param, &scrub->os_pos_current);
1054         if (rc != 0)
1055                 return rc;
1056
1057         *oic = &scrub->os_oic;
1058         fid = &(*oic)->oic_fid;
1059         lid = &(*oic)->oic_lid;
1060         rc = osd_iit_iget(info, dev, fid, lid,
1061                           scrub->os_pos_current, param->sb, true);
1062         return rc;
1063 }
1064
1065 static int osd_preload_next(struct osd_thread_info *info,
1066                             struct osd_device *dev, struct osd_iit_param *param,
1067                             struct osd_idmap_cache **oic, const bool noslot)
1068 {
1069         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
1070         struct osd_scrub        *scrub;
1071         struct ptlrpc_thread    *thread;
1072         int                      rc;
1073
1074         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1075         if (rc != 0)
1076                 return rc;
1077
1078         scrub = &dev->od_scrub;
1079         thread = &scrub->os_thread;
1080         if (thread_is_running(thread) &&
1081             ooc->ooc_pos_preload >= scrub->os_pos_current)
1082                 return SCRUB_NEXT_EXIT;
1083
1084         rc = osd_iit_iget(info, dev,
1085                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1086                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1087                           ooc->ooc_pos_preload, param->sb, false);
1088         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1089          * ignore the failure, so it still need to skip the inode next time. */
1090         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1091         return rc;
1092 }
1093
1094 static inline int
1095 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1096 {
1097         spin_lock(&scrub->os_lock);
1098         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1099             !list_empty(&scrub->os_inconsistent_items) ||
1100             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1101                 scrub->os_waiting = 0;
1102         else
1103                 scrub->os_waiting = 1;
1104         spin_unlock(&scrub->os_lock);
1105
1106         return !scrub->os_waiting;
1107 }
1108
1109 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1110                           struct osd_iit_param *param,
1111                           struct osd_idmap_cache *oic, bool *noslot, int rc)
1112 {
1113         struct l_wait_info       lwi    = { 0 };
1114         struct osd_scrub        *scrub  = &dev->od_scrub;
1115         struct scrub_file       *sf     = &scrub->os_file;
1116         struct ptlrpc_thread    *thread = &scrub->os_thread;
1117         struct osd_otable_it    *it     = dev->od_otable_it;
1118         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
1119
1120         switch (rc) {
1121         case SCRUB_NEXT_CONTINUE:
1122                 goto next;
1123         case SCRUB_NEXT_WAIT:
1124                 goto wait;
1125         case SCRUB_NEXT_NOSCRUB:
1126                 down_write(&scrub->os_rwsem);
1127                 scrub->os_new_checked++;
1128                 sf->sf_items_noscrub++;
1129                 up_write(&scrub->os_rwsem);
1130                 goto next;
1131         }
1132
1133         rc = osd_scrub_check_update(info, dev, oic, rc);
1134         if (rc != 0) {
1135                 scrub->os_in_prior = 0;
1136                 return rc;
1137         }
1138
1139         rc = osd_scrub_checkpoint(scrub);
1140         if (rc != 0) {
1141                 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1142                        "rc = %d\n", osd_scrub2name(scrub),
1143                        scrub->os_pos_current, rc);
1144                 /* Continue, as long as the scrub itself can go ahead. */
1145         }
1146
1147         if (scrub->os_in_prior) {
1148                 scrub->os_in_prior = 0;
1149                 return 0;
1150         }
1151
1152 next:
1153         scrub->os_pos_current = param->gbase + ++(param->offset);
1154
1155 wait:
1156         if (it != NULL && it->ooi_waiting && ooc != NULL &&
1157             ooc->ooc_pos_preload < scrub->os_pos_current) {
1158                 spin_lock(&scrub->os_lock);
1159                 it->ooi_waiting = 0;
1160                 wake_up_all(&thread->t_ctl_waitq);
1161                 spin_unlock(&scrub->os_lock);
1162         }
1163
1164         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1165                 return 0;
1166
1167         if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1168                 *noslot = false;
1169                 return 0;
1170         }
1171
1172         if (it != NULL)
1173                 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1174                              &lwi);
1175
1176         if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1177                 *noslot = false;
1178         else
1179                 *noslot = true;
1180         return 0;
1181 }
1182
1183 static int osd_preload_exec(struct osd_thread_info *info,
1184                             struct osd_device *dev, struct osd_iit_param *param,
1185                             struct osd_idmap_cache *oic, bool *noslot, int rc)
1186 {
1187         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1188
1189         if (rc == 0) {
1190                 ooc->ooc_cached_items++;
1191                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1192                                         ~OSD_OTABLE_IT_CACHE_MASK;
1193         }
1194         return rc > 0 ? 0 : rc;
1195 }
1196
1197 #define SCRUB_IT_ALL    1
1198 #define SCRUB_IT_CRASH  2
1199
1200 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1201                            bool inconsistent)
1202 {
1203         struct osd_scrub     *scrub  = &dev->od_scrub;
1204         struct ptlrpc_thread *thread = &scrub->os_thread;
1205         struct scrub_file    *sf     = &scrub->os_file;
1206         int                   rc;
1207         ENTRY;
1208
1209         LASSERT(!(flags & SS_AUTO_PARTIAL));
1210
1211         down_write(&scrub->os_rwsem);
1212         scrub->os_in_join = 1;
1213         if (flags & SS_SET_FAILOUT)
1214                 sf->sf_param |= SP_FAILOUT;
1215         else if (flags & SS_CLEAR_FAILOUT)
1216                 sf->sf_param &= ~SP_FAILOUT;
1217
1218         if (flags & SS_SET_DRYRUN)
1219                 sf->sf_param |= SP_DRYRUN;
1220         else if (flags & SS_CLEAR_DRYRUN)
1221                 sf->sf_param &= ~SP_DRYRUN;
1222
1223         if (flags & SS_RESET) {
1224                 osd_scrub_file_reset(scrub,
1225                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1226                         inconsistent ? SF_INCONSISTENT : 0);
1227                 sf->sf_status = SS_SCANNING;
1228         }
1229
1230         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1231                 scrub->os_full_speed = 1;
1232         else
1233                 scrub->os_full_speed = 0;
1234
1235         if (flags & SS_AUTO_FULL) {
1236                 sf->sf_flags |= SF_AUTO;
1237                 scrub->os_full_speed = 1;
1238         }
1239
1240         scrub->os_new_checked = 0;
1241         if (sf->sf_pos_last_checkpoint != 0)
1242                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1243         else
1244                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1245
1246         scrub->os_pos_current = sf->sf_pos_latest_start;
1247         sf->sf_time_latest_start = cfs_time_current_sec();
1248         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1249         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1250         rc = osd_scrub_file_store(scrub);
1251         if (rc != 0)
1252                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
1253                        "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1254
1255         spin_lock(&scrub->os_lock);
1256         scrub->os_waiting = 0;
1257         scrub->os_paused = 0;
1258         scrub->os_partial_scan = 0;
1259         scrub->os_in_join = 0;
1260         scrub->os_full_scrub = 0;
1261         spin_unlock(&scrub->os_lock);
1262         wake_up_all(&thread->t_ctl_waitq);
1263         up_write(&scrub->os_rwsem);
1264
1265         EXIT;
1266 }
1267
1268 static int osd_inode_iteration(struct osd_thread_info *info,
1269                                struct osd_device *dev, __u32 max, bool preload)
1270 {
1271         struct osd_scrub     *scrub  = &dev->od_scrub;
1272         struct ptlrpc_thread *thread = &scrub->os_thread;
1273         struct scrub_file    *sf     = &scrub->os_file;
1274         osd_iit_next_policy   next;
1275         osd_iit_exec_policy   exec;
1276         __u32                *pos;
1277         __u32                *count;
1278         struct osd_iit_param  param  = { NULL };
1279         struct l_wait_info    lwi    = { 0 };
1280         __u32                 limit;
1281         int                   rc;
1282         bool                  noslot = true;
1283         ENTRY;
1284
1285         param.sb = osd_sb(dev);
1286         if (preload)
1287                 goto full;
1288
1289         while (scrub->os_partial_scan && !scrub->os_in_join) {
1290                 struct osd_idmap_cache *oic = NULL;
1291
1292                 rc = osd_scrub_next(info, dev, &param, &oic, noslot);
1293                 switch (rc) {
1294                 case SCRUB_NEXT_EXIT:
1295                         RETURN(0);
1296                 case SCRUB_NEXT_CRASH:
1297                         RETURN(SCRUB_IT_CRASH);
1298                 case SCRUB_NEXT_FATAL:
1299                         RETURN(-EINVAL);
1300                 case SCRUB_NEXT_WAIT: {
1301                         struct kstatfs *ksfs = &info->oti_ksfs;
1302                         __u64 saved_flags;
1303
1304                         if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1305                             unlikely(sf->sf_items_updated_prior == 0))
1306                                 goto wait;
1307
1308                         if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1309                             scrub->os_full_scrub) {
1310                                 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1311                                                true);
1312                                 goto full;
1313                         }
1314
1315                         rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1316                         if (rc == 0) {
1317                                 __u64 used = ksfs->f_files - ksfs->f_ffree;
1318
1319                                 do_div(used, sf->sf_items_updated_prior);
1320                                 /* If we hit too much inconsistent OI
1321                                  * mappings during the partial scan,
1322                                  * then scan the device completely. */
1323                                 if (used < dev->od_full_scrub_ratio) {
1324                                         osd_scrub_join(dev,
1325                                                 SS_AUTO_FULL | SS_RESET, true);
1326                                         goto full;
1327                                 }
1328                         }
1329
1330 wait:
1331                         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1332                             cfs_fail_val > 0)
1333                                 continue;
1334
1335                         saved_flags = sf->sf_flags;
1336                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1337                                           SF_UPGRADE | SF_AUTO);
1338                         sf->sf_status = SS_COMPLETED;
1339                         l_wait_event(thread->t_ctl_waitq,
1340                                      !thread_is_running(thread) ||
1341                                      !scrub->os_partial_scan ||
1342                                      scrub->os_in_join ||
1343                                      !list_empty(&scrub->os_inconsistent_items),
1344                                      &lwi);
1345                         sf->sf_flags = saved_flags;
1346                         sf->sf_status = SS_SCANNING;
1347
1348                         if (unlikely(!thread_is_running(thread)))
1349                                 RETURN(0);
1350
1351                         if (!scrub->os_partial_scan || scrub->os_in_join)
1352                                 goto full;
1353
1354                         continue;
1355                 }
1356                 default:
1357                         LASSERTF(rc == 0, "rc = %d\n", rc);
1358
1359                         osd_scrub_exec(info, dev, &param, oic, &noslot, rc);
1360                         break;
1361                 }
1362         }
1363
1364 full:
1365         if (!preload) {
1366                 l_wait_event(thread->t_ctl_waitq,
1367                              !thread_is_running(thread) || !scrub->os_in_join,
1368                              &lwi);
1369
1370                 if (unlikely(!thread_is_running(thread)))
1371                         RETURN(0);
1372         }
1373
1374         noslot = false;
1375         if (!preload) {
1376                 next = osd_scrub_next;
1377                 exec = osd_scrub_exec;
1378                 pos = &scrub->os_pos_current;
1379                 count = &scrub->os_new_checked;
1380         } else {
1381                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1382
1383                 next = osd_preload_next;
1384                 exec = osd_preload_exec;
1385                 pos = &ooc->ooc_pos_preload;
1386                 count = &ooc->ooc_cached_items;
1387         }
1388
1389         param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1390         param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1391         param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1392         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1393
1394         while (*pos <= limit && *count < max) {
1395                 struct ldiskfs_group_desc *desc;
1396                 bool next_group = false;
1397
1398                 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1399                 if (!desc)
1400                         RETURN(-EIO);
1401
1402                 ldiskfs_lock_group(param.sb, param.bg);
1403                 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1404                         ldiskfs_unlock_group(param.sb, param.bg);
1405                         next_group = true;
1406                         goto next_group;
1407                 }
1408                 ldiskfs_unlock_group(param.sb, param.bg);
1409
1410                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1411                 if (!param.bitmap) {
1412                         CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1413                                "scrub will stop, urgent mode\n",
1414                                osd_scrub2name(scrub), (__u32)param.bg);
1415                         RETURN(-EIO);
1416                 }
1417
1418                 while (*count < max) {
1419                         struct osd_idmap_cache *oic = NULL;
1420
1421                         if (param.offset +
1422                                 ldiskfs_itable_unused_count(param.sb, desc) >=
1423                             LDISKFS_INODES_PER_GROUP(param.sb)) {
1424                                 next_group = true;
1425                                 goto next_group;
1426                         }
1427
1428                         rc = next(info, dev, &param, &oic, noslot);
1429                         switch (rc) {
1430                         case SCRUB_NEXT_BREAK:
1431                                 next_group = true;
1432                                 goto next_group;
1433                         case SCRUB_NEXT_EXIT:
1434                                 brelse(param.bitmap);
1435                                 RETURN(0);
1436                         case SCRUB_NEXT_CRASH:
1437                                 brelse(param.bitmap);
1438                                 RETURN(SCRUB_IT_CRASH);
1439                         case SCRUB_NEXT_FATAL:
1440                                 brelse(param.bitmap);
1441                                 RETURN(-EINVAL);
1442                         }
1443
1444                         rc = exec(info, dev, &param, oic, &noslot, rc);
1445                         if (rc != 0) {
1446                                 brelse(param.bitmap);
1447                                 RETURN(rc);
1448                         }
1449                 }
1450
1451 next_group:
1452                 if (param.bitmap) {
1453                         brelse(param.bitmap);
1454                         param.bitmap = NULL;
1455                 }
1456
1457                 if (next_group) {
1458                         param.bg++;
1459                         param.offset = 0;
1460                         param.gbase = 1 +
1461                                 param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1462                         *pos = param.gbase;
1463                 }
1464         }
1465
1466         if (*pos > limit)
1467                 RETURN(SCRUB_IT_ALL);
1468         RETURN(0);
1469 }
1470
1471 static int osd_otable_it_preload(const struct lu_env *env,
1472                                  struct osd_otable_it *it)
1473 {
1474         struct osd_device       *dev   = it->ooi_dev;
1475         struct osd_scrub        *scrub = &dev->od_scrub;
1476         struct osd_otable_cache *ooc   = &it->ooi_cache;
1477         int                      rc;
1478         ENTRY;
1479
1480         rc = osd_inode_iteration(osd_oti_get(env), dev,
1481                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1482         if (rc == SCRUB_IT_ALL)
1483                 it->ooi_all_cached = 1;
1484
1485         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1486                 spin_lock(&scrub->os_lock);
1487                 scrub->os_waiting = 0;
1488                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1489                 spin_unlock(&scrub->os_lock);
1490         }
1491
1492         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1493 }
1494
1495 static int osd_scrub_main(void *args)
1496 {
1497         struct lu_env         env;
1498         struct osd_device    *dev    = (struct osd_device *)args;
1499         struct osd_scrub     *scrub  = &dev->od_scrub;
1500         struct ptlrpc_thread *thread = &scrub->os_thread;
1501         int                   rc;
1502         ENTRY;
1503
1504         rc = lu_env_init(&env, LCT_LOCAL);
1505         if (rc != 0) {
1506                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1507                        osd_scrub2name(scrub), rc);
1508                 GOTO(noenv, rc);
1509         }
1510
1511         rc = osd_scrub_prep(dev);
1512         if (rc != 0) {
1513                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1514                        osd_scrub2name(scrub), rc);
1515                 GOTO(out, rc);
1516         }
1517
1518         if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1519                 struct l_wait_info lwi = { 0 };
1520                 struct osd_otable_it *it = dev->od_otable_it;
1521                 struct osd_otable_cache *ooc = &it->ooi_cache;
1522
1523                 l_wait_event(thread->t_ctl_waitq,
1524                              it->ooi_user_ready || !thread_is_running(thread),
1525                              &lwi);
1526                 if (unlikely(!thread_is_running(thread)))
1527                         GOTO(post, rc = 0);
1528
1529                 scrub->os_pos_current = ooc->ooc_pos_preload;
1530         }
1531
1532         CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1533                osd_scrub2name(scrub), scrub->os_start_flags,
1534                scrub->os_pos_current);
1535
1536         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1537         if (unlikely(rc == SCRUB_IT_CRASH))
1538                 GOTO(out, rc = -EINVAL);
1539         GOTO(post, rc);
1540
1541 post:
1542         rc = osd_scrub_post(scrub, rc);
1543         CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1544                osd_scrub2name(scrub), scrub->os_pos_current, rc);
1545
1546 out:
1547         while (!list_empty(&scrub->os_inconsistent_items)) {
1548                 struct osd_inconsistent_item *oii;
1549
1550                 oii = list_entry(scrub->os_inconsistent_items.next,
1551                                      struct osd_inconsistent_item, oii_list);
1552                 list_del_init(&oii->oii_list);
1553                 OBD_FREE_PTR(oii);
1554         }
1555         lu_env_fini(&env);
1556
1557 noenv:
1558         spin_lock(&scrub->os_lock);
1559         thread_set_flags(thread, SVC_STOPPED);
1560         wake_up_all(&thread->t_ctl_waitq);
1561         spin_unlock(&scrub->os_lock);
1562         return rc;
1563 }
1564
1565 /* initial OI scrub */
1566
1567 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1568                          struct dentry *, filldir_t filldir);
1569
1570 #ifdef HAVE_FILLDIR_USE_CTX
1571 static int osd_ios_varfid_fill(struct dir_context *buf, const char *name,
1572                                int namelen, loff_t offset, __u64 ino,
1573                                unsigned d_type);
1574 static int osd_ios_lf_fill(struct dir_context *buf, const char *name,
1575                            int namelen, loff_t offset, __u64 ino,
1576                            unsigned d_type);
1577 static int osd_ios_dl_fill(struct dir_context *buf, const char *name,
1578                            int namelen, loff_t offset, __u64 ino,
1579                            unsigned d_type);
1580 static int osd_ios_uld_fill(struct dir_context *buf, const char *name,
1581                             int namelen, loff_t offset, __u64 ino,
1582                             unsigned d_type);
1583 #else
1584 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1585                                loff_t offset, __u64 ino, unsigned d_type);
1586 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1587                            loff_t offset, __u64 ino, unsigned d_type);
1588 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1589                            loff_t offset, __u64 ino, unsigned d_type);
1590 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1591                             loff_t offset, __u64 ino, unsigned d_type);
1592 #endif
1593
1594 static int
1595 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1596                      struct dentry *dentry, filldir_t filldir);
1597 static int
1598 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1599                   struct dentry *dentry, filldir_t filldir);
1600
1601 static int
1602 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1603                      struct dentry *dentry, filldir_t filldir);
1604
1605 enum osd_lf_flags {
1606         OLF_SCAN_SUBITEMS       = 0x0001,
1607         OLF_HIDE_FID            = 0x0002,
1608         OLF_SHOW_NAME           = 0x0004,
1609         OLF_NO_OI               = 0x0008,
1610         OLF_IDX_IN_FID          = 0x0010,
1611 };
1612
1613 struct osd_lf_map {
1614         char            *olm_name;
1615         struct lu_fid    olm_fid;
1616         __u16            olm_flags;
1617         __u16            olm_namelen;
1618         scandir_t        olm_scandir;
1619         filldir_t        olm_filldir;
1620 };
1621
1622 /* Add the new introduced local files in the list in the future. */
1623 static const struct osd_lf_map osd_lf_maps[] = {
1624         /* CATALOGS */
1625         {
1626                 .olm_name       = CATLIST,
1627                 .olm_fid        = {
1628                         .f_seq  = FID_SEQ_LOCAL_FILE,
1629                         .f_oid  = LLOG_CATALOGS_OID,
1630                 },
1631                 .olm_flags      = OLF_SHOW_NAME,
1632                 .olm_namelen    = sizeof(CATLIST) - 1,
1633         },
1634
1635         /* CONFIGS */
1636         {
1637                 .olm_name       = MOUNT_CONFIGS_DIR,
1638                 .olm_fid        = {
1639                         .f_seq  = FID_SEQ_LOCAL_FILE,
1640                         .f_oid  = MGS_CONFIGS_OID,
1641                 },
1642                 .olm_flags      = OLF_SCAN_SUBITEMS,
1643                 .olm_namelen    = sizeof(MOUNT_CONFIGS_DIR) - 1,
1644                 .olm_scandir    = osd_ios_general_scan,
1645                 .olm_filldir    = osd_ios_varfid_fill,
1646         },
1647
1648         /* NIDTBL_VERSIONS */
1649         {
1650                 .olm_name       = MGS_NIDTBL_DIR,
1651                 .olm_flags      = OLF_SCAN_SUBITEMS,
1652                 .olm_namelen    = sizeof(MGS_NIDTBL_DIR) - 1,
1653                 .olm_scandir    = osd_ios_general_scan,
1654                 .olm_filldir    = osd_ios_varfid_fill,
1655         },
1656
1657         /* PENDING */
1658         {
1659                 .olm_name       = "PENDING",
1660                 .olm_namelen    = sizeof("PENDING") - 1,
1661         },
1662
1663         /* ROOT */
1664         {
1665                 .olm_name       = "ROOT",
1666                 .olm_fid        = {
1667                         .f_seq  = FID_SEQ_ROOT,
1668                         .f_oid  = FID_OID_ROOT,
1669                 },
1670                 .olm_flags      = OLF_SCAN_SUBITEMS | OLF_HIDE_FID,
1671                 .olm_namelen    = sizeof("ROOT") - 1,
1672                 .olm_scandir    = osd_ios_ROOT_scan,
1673         },
1674
1675         /* changelog_catalog */
1676         {
1677                 .olm_name       = CHANGELOG_CATALOG,
1678                 .olm_namelen    = sizeof(CHANGELOG_CATALOG) - 1,
1679         },
1680
1681         /* changelog_users */
1682         {
1683                 .olm_name       = CHANGELOG_USERS,
1684                 .olm_namelen    = sizeof(CHANGELOG_USERS) - 1,
1685         },
1686
1687         /* fld */
1688         {
1689                 .olm_name       = "fld",
1690                 .olm_fid        = {
1691                         .f_seq  = FID_SEQ_LOCAL_FILE,
1692                         .f_oid  = FLD_INDEX_OID,
1693                 },
1694                 .olm_flags      = OLF_SHOW_NAME,
1695                 .olm_namelen    = sizeof("fld") - 1,
1696         },
1697
1698         /* last_rcvd */
1699         {
1700                 .olm_name       = LAST_RCVD,
1701                 .olm_fid        = {
1702                         .f_seq  = FID_SEQ_LOCAL_FILE,
1703                         .f_oid  = LAST_RECV_OID,
1704                 },
1705                 .olm_flags      = OLF_SHOW_NAME,
1706                 .olm_namelen    = sizeof(LAST_RCVD) - 1,
1707         },
1708
1709         /* reply_data */
1710         {
1711                 .olm_name       = REPLY_DATA,
1712                 .olm_fid        = {
1713                         .f_seq  = FID_SEQ_LOCAL_FILE,
1714                         .f_oid  = REPLY_DATA_OID,
1715                 },
1716                 .olm_flags      = OLF_SHOW_NAME,
1717                 .olm_namelen    = sizeof(REPLY_DATA) - 1,
1718         },
1719
1720         /* lov_objid */
1721         {
1722                 .olm_name       = LOV_OBJID,
1723                 .olm_fid        = {
1724                         .f_seq  = FID_SEQ_LOCAL_FILE,
1725                         .f_oid  = MDD_LOV_OBJ_OID,
1726                 },
1727                 .olm_flags      = OLF_SHOW_NAME,
1728                 .olm_namelen    = sizeof(LOV_OBJID) - 1,
1729         },
1730
1731         /* lov_objseq */
1732         {
1733                 .olm_name       = LOV_OBJSEQ,
1734                 .olm_fid        = {
1735                         .f_seq  = FID_SEQ_LOCAL_FILE,
1736                         .f_oid  = MDD_LOV_OBJ_OSEQ,
1737                 },
1738                 .olm_flags      = OLF_SHOW_NAME,
1739                 .olm_namelen    = sizeof(LOV_OBJSEQ) - 1,
1740         },
1741
1742         /* quota_master */
1743         {
1744                 .olm_name       = QMT_DIR,
1745                 .olm_flags      = OLF_SCAN_SUBITEMS,
1746                 .olm_namelen    = sizeof(QMT_DIR) - 1,
1747                 .olm_scandir    = osd_ios_general_scan,
1748                 .olm_filldir    = osd_ios_varfid_fill,
1749         },
1750
1751         /* quota_slave */
1752         {
1753                 .olm_name       = QSD_DIR,
1754                 .olm_flags      = OLF_SCAN_SUBITEMS,
1755                 .olm_namelen    = sizeof(QSD_DIR) - 1,
1756                 .olm_scandir    = osd_ios_general_scan,
1757                 .olm_filldir    = osd_ios_varfid_fill,
1758         },
1759
1760         /* seq_ctl */
1761         {
1762                 .olm_name       = "seq_ctl",
1763                 .olm_fid        = {
1764                         .f_seq  = FID_SEQ_LOCAL_FILE,
1765                         .f_oid  = FID_SEQ_CTL_OID,
1766                 },
1767                 .olm_flags      = OLF_SHOW_NAME,
1768                 .olm_namelen    = sizeof("seq_ctl") - 1,
1769         },
1770
1771         /* seq_srv */
1772         {
1773                 .olm_name       = "seq_srv",
1774                 .olm_fid        = {
1775                         .f_seq  = FID_SEQ_LOCAL_FILE,
1776                         .f_oid  = FID_SEQ_SRV_OID,
1777                 },
1778                 .olm_flags      = OLF_SHOW_NAME,
1779                 .olm_namelen    = sizeof("seq_srv") - 1,
1780         },
1781
1782         /* health_check */
1783         {
1784                 .olm_name       = HEALTH_CHECK,
1785                 .olm_fid        = {
1786                         .f_seq  = FID_SEQ_LOCAL_FILE,
1787                         .f_oid  = OFD_HEALTH_CHECK_OID,
1788                 },
1789                 .olm_flags      = OLF_SHOW_NAME,
1790                 .olm_namelen    = sizeof(HEALTH_CHECK) - 1,
1791         },
1792
1793         /* LFSCK */
1794         {
1795                 .olm_name       = LFSCK_DIR,
1796                 .olm_namelen    = sizeof(LFSCK_DIR) - 1,
1797                 .olm_scandir    = osd_ios_general_scan,
1798                 .olm_filldir    = osd_ios_varfid_fill,
1799         },
1800
1801         /* lfsck_bookmark */
1802         {
1803                 .olm_name       = LFSCK_BOOKMARK,
1804                 .olm_namelen    = sizeof(LFSCK_BOOKMARK) - 1,
1805         },
1806
1807         /* lfsck_layout */
1808         {
1809                 .olm_name       = LFSCK_LAYOUT,
1810                 .olm_namelen    = sizeof(LFSCK_LAYOUT) - 1,
1811         },
1812
1813         /* lfsck_namespace */
1814         {
1815                 .olm_name       = LFSCK_NAMESPACE,
1816                 .olm_namelen    = sizeof(LFSCK_NAMESPACE) - 1,
1817         },
1818
1819         /* OBJECTS, upgrade from old device */
1820         {
1821                 .olm_name       = OBJECTS,
1822                 .olm_flags      = OLF_SCAN_SUBITEMS,
1823                 .olm_namelen    = sizeof(OBJECTS) - 1,
1824                 .olm_scandir    = osd_ios_OBJECTS_scan,
1825         },
1826
1827         /* lquota_v2.user, upgrade from old device */
1828         {
1829                 .olm_name       = "lquota_v2.user",
1830                 .olm_namelen    = sizeof("lquota_v2.user") - 1,
1831         },
1832
1833         /* lquota_v2.group, upgrade from old device */
1834         {
1835                 .olm_name       = "lquota_v2.group",
1836                 .olm_namelen    = sizeof("lquota_v2.group") - 1,
1837         },
1838
1839         /* LAST_GROUP, upgrade from old device */
1840         {
1841                 .olm_name       = "LAST_GROUP",
1842                 .olm_fid        = {
1843                         .f_seq  = FID_SEQ_LOCAL_FILE,
1844                         .f_oid  = OFD_LAST_GROUP_OID,
1845                 },
1846                 .olm_flags      = OLF_SHOW_NAME,
1847                 .olm_namelen    = sizeof("LAST_GROUP") - 1,
1848         },
1849
1850         /* committed batchid for cross-MDT operation */
1851         {
1852                 .olm_name       = "BATCHID",
1853                 .olm_fid        = {
1854                         .f_seq  = FID_SEQ_LOCAL_FILE,
1855                         .f_oid  = BATCHID_COMMITTED_OID,
1856                 },
1857                 .olm_flags      = OLF_SHOW_NAME,
1858                 .olm_namelen    = sizeof("BATCHID") - 1,
1859         },
1860
1861         /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1862          * and f_oid = index for their log files.  See lu_update_log{_dir}_fid()
1863          * for more details. */
1864
1865         /* update_log */
1866         {
1867                 .olm_name       = "update_log",
1868                 .olm_fid        = {
1869                         .f_seq  = FID_SEQ_UPDATE_LOG,
1870                 },
1871                 .olm_flags      = OLF_SHOW_NAME | OLF_IDX_IN_FID,
1872                 .olm_namelen    = sizeof("update_log") - 1,
1873         },
1874
1875         /* update_log_dir */
1876         {
1877                 .olm_name       = "update_log_dir",
1878                 .olm_fid        = {
1879                         .f_seq  = FID_SEQ_UPDATE_LOG_DIR,
1880                 },
1881                 .olm_flags      = OLF_SHOW_NAME | OLF_SCAN_SUBITEMS |
1882                                   OLF_IDX_IN_FID,
1883                 .olm_namelen    = sizeof("update_log_dir") - 1,
1884                 .olm_scandir    = osd_ios_general_scan,
1885                 .olm_filldir    = osd_ios_uld_fill,
1886         },
1887
1888         /* lost+found */
1889         {
1890                 .olm_name       = "lost+found",
1891                 .olm_fid        = {
1892                         .f_seq  = FID_SEQ_LOCAL_FILE,
1893                         .f_oid  = OSD_LPF_OID,
1894                 },
1895                 .olm_flags      = OLF_SCAN_SUBITEMS,
1896                 .olm_namelen    = sizeof("lost+found") - 1,
1897                 .olm_scandir    = osd_ios_general_scan,
1898                 .olm_filldir    = osd_ios_lf_fill,
1899         },
1900
1901         {
1902                 .olm_name       = NULL
1903         }
1904 };
1905
1906 /* Add the new introduced files under .lustre/ in the list in the future. */
1907 static const struct osd_lf_map osd_dl_maps[] = {
1908         /* .lustre/fid */
1909         {
1910                 .olm_name       = "fid",
1911                 .olm_fid        = {
1912                         .f_seq  = FID_SEQ_DOT_LUSTRE,
1913                         .f_oid  = FID_OID_DOT_LUSTRE_OBF,
1914                 },
1915                 .olm_namelen    = sizeof("fid") - 1,
1916         },
1917         /* .lustre/lost+found */
1918         {
1919                 .olm_name       = "lost+found",
1920                 .olm_fid        = {
1921                         .f_seq  = FID_SEQ_DOT_LUSTRE,
1922                         .f_oid  = FID_OID_DOT_LUSTRE_LPF,
1923                 },
1924                 .olm_namelen    = sizeof("lost+found") - 1,
1925         },
1926         {
1927                 .olm_name       = NULL
1928         }
1929 };
1930
1931 struct osd_ios_item {
1932         struct list_head oii_list;
1933         struct dentry   *oii_dentry;
1934         scandir_t        oii_scandir;
1935         filldir_t        oii_filldir;
1936 };
1937
1938 struct osd_ios_filldir_buf {
1939 #ifdef HAVE_DIR_CONTEXT
1940         /* please keep it as first member */
1941         struct dir_context       ctx;
1942 #endif
1943         struct osd_thread_info  *oifb_info;
1944         struct osd_device       *oifb_dev;
1945         struct dentry           *oifb_dentry;
1946 };
1947
1948 static inline struct dentry *
1949 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1950 {
1951         struct dentry *dentry;
1952
1953         dentry = ll_lookup_one_len(name, parent, namelen);
1954         if (IS_ERR(dentry)) {
1955                 int rc = PTR_ERR(dentry);
1956
1957                 if (rc != -ENOENT)
1958                         CERROR("Fail to find %.*s in %.*s (%lu/%u): rc = %d\n",
1959                                namelen, name, parent->d_name.len,
1960                                parent->d_name.name, parent->d_inode->i_ino,
1961                                parent->d_inode->i_generation, rc);
1962
1963                 return dentry;
1964         }
1965
1966         if (dentry->d_inode == NULL) {
1967                 dput(dentry);
1968                 return ERR_PTR(-ENOENT);
1969         }
1970
1971         return dentry;
1972 }
1973
1974 static int
1975 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1976                  scandir_t scandir, filldir_t filldir)
1977 {
1978         struct osd_ios_item *item;
1979         ENTRY;
1980
1981         OBD_ALLOC_PTR(item);
1982         if (item == NULL)
1983                 RETURN(-ENOMEM);
1984
1985         INIT_LIST_HEAD(&item->oii_list);
1986         item->oii_dentry = dget(dentry);
1987         item->oii_scandir = scandir;
1988         item->oii_filldir = filldir;
1989         list_add_tail(&item->oii_list, &dev->od_ios_list);
1990
1991         RETURN(0);
1992 }
1993
1994 /**
1995  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1996  *
1997  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1998  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1999  * reference the inode, or fixed if it is missing or references another inode.
2000  */
2001 static int
2002 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
2003                  struct inode *inode, const struct lu_fid *fid, int flags)
2004 {
2005         struct lustre_mdt_attrs *lma    = &info->oti_ost_attrs.loa_lma;
2006         struct osd_inode_id     *id     = &info->oti_id;
2007         struct osd_inode_id     *id2    = &info->oti_id2;
2008         struct osd_scrub        *scrub  = &dev->od_scrub;
2009         struct scrub_file       *sf     = &scrub->os_file;
2010         struct lu_fid            tfid;
2011         int                      rc;
2012         ENTRY;
2013
2014         rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
2015                          &info->oti_ost_attrs);
2016         if (rc != 0 && rc != -ENODATA) {
2017                 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
2018                        "rc = %d\n", osd_name(dev), rc);
2019
2020                 RETURN(rc);
2021         }
2022
2023         osd_id_gen(id, inode->i_ino, inode->i_generation);
2024         if (rc == -ENODATA) {
2025                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
2026                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
2027                 } else {
2028                         tfid = *fid;
2029                         if (flags & OLF_IDX_IN_FID) {
2030                                 LASSERT(dev->od_index >= 0);
2031
2032                                 tfid.f_oid = dev->od_index;
2033                         }
2034                 }
2035                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
2036                 if (rc != 0) {
2037                         CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
2038                               "scrub: rc = %d\n", osd_name(dev), rc);
2039
2040                         RETURN(rc);
2041                 }
2042         } else {
2043                 if (lma->lma_compat & LMAC_NOT_IN_OI)
2044                         RETURN(0);
2045
2046                 tfid = lma->lma_self_fid;
2047         }
2048
2049         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
2050         if (rc != 0) {
2051                 if (rc != -ENOENT)
2052                         RETURN(rc);
2053
2054                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
2055                                                DTO_INDEX_INSERT, true, 0, NULL);
2056                 if (rc > 0)
2057                         rc = 0;
2058
2059                 RETURN(rc);
2060         }
2061
2062         if (osd_id_eq_strict(id, id2))
2063                 RETURN(0);
2064
2065         if (!(sf->sf_flags & SF_INCONSISTENT)) {
2066                 osd_scrub_file_reset(scrub,
2067                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2068                                      SF_INCONSISTENT);
2069                 rc = osd_scrub_file_store(scrub);
2070                 if (rc != 0)
2071                         RETURN(rc);
2072         }
2073
2074         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
2075                                        DTO_INDEX_UPDATE, true, 0, NULL);
2076         if (rc > 0)
2077                 rc = 0;
2078
2079         RETURN(rc);
2080 }
2081
2082 /**
2083  * It scans the /lost+found, and for the OST-object (with filter_fid
2084  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
2085  */
2086 #ifdef HAVE_FILLDIR_USE_CTX
2087 static int osd_ios_lf_fill(struct dir_context *buf,
2088 #else
2089 static int osd_ios_lf_fill(void *buf,
2090 #endif
2091                            const char *name, int namelen,
2092                            loff_t offset, __u64 ino, unsigned d_type)
2093 {
2094         struct osd_ios_filldir_buf *fill_buf =
2095                 (struct osd_ios_filldir_buf *)buf;
2096         struct osd_thread_info     *info     = fill_buf->oifb_info;
2097         struct osd_device          *dev      = fill_buf->oifb_dev;
2098         struct lu_fid              *fid      = &info->oti_fid;
2099         struct osd_scrub           *scrub    = &dev->od_scrub;
2100         struct dentry              *parent   = fill_buf->oifb_dentry;
2101         struct dentry              *child;
2102         struct inode               *dir      = parent->d_inode;
2103         struct inode               *inode;
2104         int                         rc;
2105         ENTRY;
2106
2107         /* skip any '.' started names */
2108         if (name[0] == '.')
2109                 RETURN(0);
2110
2111         scrub->os_lf_scanned++;
2112         child = osd_ios_lookup_one_len(name, parent, namelen);
2113         if (IS_ERR(child)) {
2114                 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
2115                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
2116                 RETURN(0);
2117         }
2118
2119         inode = child->d_inode;
2120         if (S_ISDIR(inode->i_mode)) {
2121                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2122                                       osd_ios_lf_fill);
2123                 if (rc != 0)
2124                         CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
2125                               "rc = %d\n", osd_name(dev), namelen, name, rc);
2126                 GOTO(put, rc);
2127         }
2128
2129         if (!S_ISREG(inode->i_mode))
2130                 GOTO(put, rc = 0);
2131
2132         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
2133         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
2134                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
2135                 if (rc == 0) {
2136                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
2137                                "/lost+found.\n", namelen, name, PFID(fid));
2138                         scrub->os_lf_repaired++;
2139                 } else {
2140                         CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
2141                                DFID": rc = %d\n",
2142                                osd_name(dev), namelen, name, PFID(fid), rc);
2143                 }
2144         }
2145
2146         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
2147          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
2148          *      can process them in furtuer. */
2149
2150         GOTO(put, rc);
2151
2152 put:
2153         if (rc < 0)
2154                 scrub->os_lf_failed++;
2155         dput(child);
2156         /* skip the failure to make the scanning to continue. */
2157         return 0;
2158 }
2159
2160 #ifdef HAVE_FILLDIR_USE_CTX
2161 static int osd_ios_varfid_fill(struct dir_context *buf,
2162 #else
2163 static int osd_ios_varfid_fill(void *buf,
2164 #endif
2165                                const char *name, int namelen,
2166                                loff_t offset, __u64 ino, unsigned d_type)
2167 {
2168         struct osd_ios_filldir_buf *fill_buf =
2169                 (struct osd_ios_filldir_buf *)buf;
2170         struct osd_device          *dev      = fill_buf->oifb_dev;
2171         struct dentry              *child;
2172         int                         rc;
2173         ENTRY;
2174
2175         /* skip any '.' started names */
2176         if (name[0] == '.')
2177                 RETURN(0);
2178
2179         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2180         if (IS_ERR(child))
2181                 RETURN(PTR_ERR(child));
2182
2183         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2184                               NULL, 0);
2185         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
2186                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2187                                       osd_ios_varfid_fill);
2188         dput(child);
2189
2190         RETURN(rc);
2191 }
2192
2193 #ifdef HAVE_FILLDIR_USE_CTX
2194 static int osd_ios_dl_fill(struct dir_context *buf,
2195 #else
2196 static int osd_ios_dl_fill(void *buf,
2197 #endif
2198                            const char *name, int namelen,
2199                            loff_t offset, __u64 ino, unsigned d_type)
2200 {
2201         struct osd_ios_filldir_buf *fill_buf =
2202                 (struct osd_ios_filldir_buf *)buf;
2203         struct osd_device          *dev      = fill_buf->oifb_dev;
2204         const struct osd_lf_map    *map;
2205         struct dentry              *child;
2206         int                         rc       = 0;
2207         ENTRY;
2208
2209         /* skip any '.' started names */
2210         if (name[0] == '.')
2211                 RETURN(0);
2212
2213         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
2214                 if (map->olm_namelen != namelen)
2215                         continue;
2216
2217                 if (strncmp(map->olm_name, name, namelen) == 0)
2218                         break;
2219         }
2220
2221         if (map->olm_name == NULL)
2222                 RETURN(0);
2223
2224         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2225         if (IS_ERR(child))
2226                 RETURN(PTR_ERR(child));
2227
2228         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2229                               &map->olm_fid, map->olm_flags);
2230         dput(child);
2231
2232         RETURN(rc);
2233 }
2234
2235 #ifdef HAVE_FILLDIR_USE_CTX
2236 static int osd_ios_uld_fill(struct dir_context *buf,
2237 #else
2238 static int osd_ios_uld_fill(void *buf,
2239 #endif
2240                             const char *name, int namelen,
2241                             loff_t offset, __u64 ino, unsigned d_type)
2242 {
2243         struct osd_ios_filldir_buf *fill_buf =
2244                 (struct osd_ios_filldir_buf *)buf;
2245         struct dentry              *child;
2246         struct lu_fid               tfid;
2247         int                         rc       = 0;
2248         ENTRY;
2249
2250         /* skip any non-DFID format name */
2251         if (name[0] != '[')
2252                 RETURN(0);
2253
2254         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2255         if (IS_ERR(child))
2256                 RETURN(PTR_ERR(child));
2257
2258         /* skip the start '[' */
2259         sscanf(&name[1], SFID, RFID(&tfid));
2260         if (fid_is_sane(&tfid))
2261                 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
2262                                       child->d_inode, &tfid, 0);
2263         else
2264                 rc = -EIO;
2265         dput(child);
2266
2267         RETURN(rc);
2268 }
2269
2270 #ifdef HAVE_FILLDIR_USE_CTX
2271 static int osd_ios_root_fill(struct dir_context *buf,
2272 #else
2273 static int osd_ios_root_fill(void *buf,
2274 #endif
2275                              const char *name, int namelen,
2276                              loff_t offset, __u64 ino, unsigned d_type)
2277 {
2278         struct osd_ios_filldir_buf *fill_buf =
2279                 (struct osd_ios_filldir_buf *)buf;
2280         struct osd_device          *dev      = fill_buf->oifb_dev;
2281         const struct osd_lf_map    *map;
2282         struct dentry              *child;
2283         int                         rc       = 0;
2284         ENTRY;
2285
2286         /* skip any '.' started names */
2287         if (name[0] == '.')
2288                 RETURN(0);
2289
2290         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2291                 if (map->olm_namelen != namelen)
2292                         continue;
2293
2294                 if (strncmp(map->olm_name, name, namelen) == 0)
2295                         break;
2296         }
2297
2298         if (map->olm_name == NULL)
2299                 RETURN(0);
2300
2301         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2302         if (IS_ERR(child))
2303                 RETURN(PTR_ERR(child));
2304
2305         if (!(map->olm_flags & OLF_NO_OI))
2306                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2307                                       &map->olm_fid, map->olm_flags);
2308         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2309                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2310                                       map->olm_filldir);
2311         dput(child);
2312
2313         RETURN(rc);
2314 }
2315
2316 static int
2317 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2318                      struct dentry *dentry, filldir_t filldir)
2319 {
2320         struct osd_ios_filldir_buf    buf   = {
2321 #ifdef HAVE_DIR_CONTEXT
2322                                                 .ctx.actor = filldir,
2323 #endif
2324                                                 .oifb_info = info,
2325                                                 .oifb_dev = dev,
2326                                                 .oifb_dentry = dentry };
2327         struct file                  *filp  = &info->oti_file;
2328         struct inode                 *inode = dentry->d_inode;
2329         const struct file_operations *fops  = inode->i_fop;
2330         int                           rc;
2331         ENTRY;
2332
2333         LASSERT(filldir != NULL);
2334
2335         filp->f_pos = 0;
2336         filp->f_path.dentry = dentry;
2337         filp->f_mode = FMODE_64BITHASH;
2338         filp->f_mapping = inode->i_mapping;
2339         filp->f_op = fops;
2340         filp->private_data = NULL;
2341         set_file_inode(filp, inode);
2342
2343 #ifdef HAVE_DIR_CONTEXT
2344         buf.ctx.pos = filp->f_pos;
2345         rc = fops->iterate(filp, &buf.ctx);
2346         filp->f_pos = buf.ctx.pos;
2347 #else
2348         rc = fops->readdir(filp, &buf, filldir);
2349 #endif
2350         fops->release(inode, filp);
2351
2352         RETURN(rc);
2353 }
2354
2355 static int
2356 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2357                   struct dentry *dentry, filldir_t filldir)
2358 {
2359         struct osd_scrub  *scrub  = &dev->od_scrub;
2360         struct scrub_file *sf     = &scrub->os_file;
2361         struct dentry     *child;
2362         int                rc;
2363         ENTRY;
2364
2365         /* It is existing MDT0 device. We only allow the case of object without
2366          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2367          * can generate IGIF mode FID for the object and related OI mapping. If
2368          * it is on other MDTs, then becuase file-level backup/restore, related
2369          * OI mapping may be invalid already, we do not know which is the right
2370          * FID for the object. We only allow IGIF objects to reside on the MDT0.
2371          *
2372          * XXX: For the case of object on non-MDT0 device with neither LMA nor
2373          *      "fid" xattr, then something crashed. We cannot re-generate the
2374          *      FID directly, instead, the OI scrub will scan the OI structure
2375          *      and try to re-generate the LMA from the OI mapping. But if the
2376          *      OI mapping crashed or lost also, then we have to give up under
2377          *      double failure cases. */
2378         scrub->os_convert_igif = 1;
2379         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2380                                        strlen(dot_lustre_name));
2381         if (IS_ERR(child)) {
2382                 rc = PTR_ERR(child);
2383                 if (rc == -ENOENT) {
2384                         /* It is 1.8 MDT device. */
2385                         if (!(sf->sf_flags & SF_UPGRADE)) {
2386                                 osd_scrub_file_reset(scrub,
2387                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2388                                         SF_UPGRADE);
2389                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2390                                 rc = osd_scrub_file_store(scrub);
2391                         } else {
2392                                 rc = 0;
2393                         }
2394                 }
2395         } else {
2396                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2397                  * so the client will get IGIF for the ".lustre" object when
2398                  * the MDT restart.
2399                  *
2400                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2401                  * it does not know whether there are some old clients cached
2402                  * the ".lustre" IGIF during the upgrading. Two choices:
2403                  *
2404                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2405                  *    It will allow the old connected clients to access the
2406                  *    ".lustre" with cached IGIF. But it will cause others
2407                  *    on the MDT failed to check "fid_is_dot_lustre()".
2408                  *
2409                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2410                  *    for ".lustre" in spite of whether there are some clients
2411                  *    cached the ".lustre" IGIF or not. It enables the check
2412                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
2413                  *    that the old connected clients cannot access the ".lustre"
2414                  *    with the cached IGIF.
2415                  *
2416                  * Usually, it is rare case for the old connected clients
2417                  * to access the ".lustre" with cached IGIF. So we prefer
2418                  * to the solution 2). */
2419                 rc = osd_ios_scan_one(info, dev, child->d_inode,
2420                                       &LU_DOT_LUSTRE_FID, 0);
2421                 if (rc == 0)
2422                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2423                                               osd_ios_dl_fill);
2424                 dput(child);
2425         }
2426
2427         RETURN(rc);
2428 }
2429
2430 static int
2431 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2432                      struct dentry *dentry, filldir_t filldir)
2433 {
2434         struct osd_scrub  *scrub  = &dev->od_scrub;
2435         struct scrub_file *sf     = &scrub->os_file;
2436         struct dentry     *child;
2437         int                rc;
2438         ENTRY;
2439
2440         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2441                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2442                 rc = osd_scrub_file_store(scrub);
2443                 if (rc != 0)
2444                         RETURN(rc);
2445         }
2446
2447         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2448         if (!IS_ERR(child)) {
2449                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2450                 dput(child);
2451         } else {
2452                 rc = PTR_ERR(child);
2453         }
2454
2455         if (rc != 0 && rc != -ENOENT)
2456                 RETURN(rc);
2457
2458         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2459         if (!IS_ERR(child)) {
2460                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2461                 dput(child);
2462         } else {
2463                 rc = PTR_ERR(child);
2464         }
2465
2466         if (rc == -ENOENT)
2467                 rc = 0;
2468
2469         RETURN(rc);
2470 }
2471
2472 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2473                                 struct osd_device *dev)
2474 {
2475         struct osd_ios_item     *item    = NULL;
2476         scandir_t                scandir = osd_ios_general_scan;
2477         filldir_t                filldir = osd_ios_root_fill;
2478         struct dentry           *dentry  = osd_sb(dev)->s_root;
2479         const struct osd_lf_map *map     = osd_lf_maps;
2480         int                      rc;
2481         ENTRY;
2482
2483         /* Lookup IGIF in OI by force for initial OI scrub. */
2484         dev->od_igif_inoi = 1;
2485
2486         while (1) {
2487                 rc = scandir(info, dev, dentry, filldir);
2488                 if (item != NULL) {
2489                         dput(item->oii_dentry);
2490                         OBD_FREE_PTR(item);
2491                 }
2492
2493                 if (rc != 0)
2494                         break;
2495
2496                 if (list_empty(&dev->od_ios_list))
2497                         break;
2498
2499                 item = list_entry(dev->od_ios_list.next,
2500                                   struct osd_ios_item, oii_list);
2501                 list_del_init(&item->oii_list);
2502
2503                 LASSERT(item->oii_scandir != NULL);
2504                 scandir = item->oii_scandir;
2505                 filldir = item->oii_filldir;
2506                 dentry = item->oii_dentry;
2507         }
2508
2509         while (!list_empty(&dev->od_ios_list)) {
2510                 item = list_entry(dev->od_ios_list.next,
2511                                   struct osd_ios_item, oii_list);
2512                 list_del_init(&item->oii_list);
2513                 dput(item->oii_dentry);
2514                 OBD_FREE_PTR(item);
2515         }
2516
2517         if (rc != 0)
2518                 RETURN(rc);
2519
2520         /* There maybe the case that the object has been removed, but its OI
2521          * mapping is still in the OI file, such as the "CATALOGS" after MDT
2522          * file-level backup/restore. So here cleanup the stale OI mappings. */
2523         while (map->olm_name != NULL) {
2524                 struct dentry *child;
2525
2526                 if (fid_is_zero(&map->olm_fid)) {
2527                         map++;
2528                         continue;
2529                 }
2530
2531                 child = osd_ios_lookup_one_len(map->olm_name,
2532                                                osd_sb(dev)->s_root,
2533                                                map->olm_namelen);
2534                 if (!IS_ERR(child))
2535                         dput(child);
2536                 else if (PTR_ERR(child) == -ENOENT)
2537                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2538                                                   NULL, DTO_INDEX_DELETE,
2539                                                   true, 0, NULL);
2540                 map++;
2541         }
2542
2543         RETURN(0);
2544 }
2545
2546 char *osd_lf_fid2name(const struct lu_fid *fid)
2547 {
2548         const struct osd_lf_map *map = osd_lf_maps;
2549
2550         while (map->olm_name != NULL) {
2551                 if (!lu_fid_eq(fid, &map->olm_fid)) {
2552                         map++;
2553                         continue;
2554                 }
2555
2556                 if (map->olm_flags & OLF_SHOW_NAME)
2557                         return map->olm_name;
2558                 else
2559                         return "";
2560         }
2561
2562         return NULL;
2563 }
2564
2565 /* OI scrub start/stop */
2566
2567 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2568 {
2569         struct osd_scrub     *scrub  = &dev->od_scrub;
2570         struct ptlrpc_thread *thread = &scrub->os_thread;
2571         struct l_wait_info    lwi    = { 0 };
2572         struct task_struct   *task;
2573         int                   rc;
2574         ENTRY;
2575
2576         if (dev->od_dt_dev.dd_rdonly)
2577                 RETURN(-EROFS);
2578
2579         /* os_lock: sync status between stop and scrub thread */
2580         spin_lock(&scrub->os_lock);
2581
2582 again:
2583         if (thread_is_running(thread)) {
2584                 spin_unlock(&scrub->os_lock);
2585                 if (!(scrub->os_file.sf_flags & SF_AUTO ||
2586                       scrub->os_partial_scan) ||
2587                      (flags & SS_AUTO_PARTIAL))
2588                         RETURN(-EALREADY);
2589
2590                 osd_scrub_join(dev, flags, false);
2591                 spin_lock(&scrub->os_lock);
2592                 if (!thread_is_running(thread))
2593                         goto again;
2594
2595                 spin_unlock(&scrub->os_lock);
2596                 RETURN(0);
2597         }
2598
2599         if (unlikely(thread_is_stopping(thread))) {
2600                 spin_unlock(&scrub->os_lock);
2601                 l_wait_event(thread->t_ctl_waitq,
2602                              thread_is_stopped(thread),
2603                              &lwi);
2604                 spin_lock(&scrub->os_lock);
2605                 goto again;
2606         }
2607         spin_unlock(&scrub->os_lock);
2608
2609         if (scrub->os_file.sf_status == SS_COMPLETED) {
2610                 if (!(flags & SS_SET_FAILOUT))
2611                         flags |= SS_CLEAR_FAILOUT;
2612
2613                 if (!(flags & SS_SET_DRYRUN))
2614                         flags |= SS_CLEAR_DRYRUN;
2615
2616                 flags |= SS_RESET;
2617         }
2618
2619         scrub->os_start_flags = flags;
2620         thread_set_flags(thread, 0);
2621         task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2622         if (IS_ERR(task)) {
2623                 rc = PTR_ERR(task);
2624                 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2625                        osd_scrub2name(scrub), rc);
2626                 RETURN(rc);
2627         }
2628
2629         l_wait_event(thread->t_ctl_waitq,
2630                      thread_is_running(thread) || thread_is_stopped(thread),
2631                      &lwi);
2632
2633         RETURN(0);
2634 }
2635
2636 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2637 {
2638         int rc;
2639         ENTRY;
2640
2641         /* od_otable_mutex: prevent curcurrent start/stop */
2642         mutex_lock(&dev->od_otable_mutex);
2643         rc = do_osd_scrub_start(dev, flags);
2644         mutex_unlock(&dev->od_otable_mutex);
2645
2646         RETURN(rc == -EALREADY ? 0 : rc);
2647 }
2648
2649 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2650 {
2651         struct ptlrpc_thread *thread = &scrub->os_thread;
2652         struct l_wait_info    lwi    = { 0 };
2653
2654         /* os_lock: sync status between stop and scrub thread */
2655         spin_lock(&scrub->os_lock);
2656         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2657                 thread_set_flags(thread, SVC_STOPPING);
2658                 spin_unlock(&scrub->os_lock);
2659                 wake_up_all(&thread->t_ctl_waitq);
2660                 l_wait_event(thread->t_ctl_waitq,
2661                              thread_is_stopped(thread),
2662                              &lwi);
2663                 /* Do not skip the last lock/unlock, which can guarantee that
2664                  * the caller cannot return until the OI scrub thread exit. */
2665                 spin_lock(&scrub->os_lock);
2666         }
2667         spin_unlock(&scrub->os_lock);
2668 }
2669
2670 static void osd_scrub_stop(struct osd_device *dev)
2671 {
2672         /* od_otable_mutex: prevent curcurrent start/stop */
2673         mutex_lock(&dev->od_otable_mutex);
2674         dev->od_scrub.os_paused = 1;
2675         do_osd_scrub_stop(&dev->od_scrub);
2676         mutex_unlock(&dev->od_otable_mutex);
2677 }
2678
2679 /* OI scrub setup/cleanup */
2680
2681 static const char osd_scrub_name[] = "OI_scrub";
2682
2683 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2684 {
2685         struct osd_thread_info     *info   = osd_oti_get(env);
2686         struct osd_scrub           *scrub  = &dev->od_scrub;
2687         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
2688         struct scrub_file          *sf     = &scrub->os_file;
2689         struct super_block         *sb     = osd_sb(dev);
2690         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
2691         struct lvfs_run_ctxt        saved;
2692         struct file                *filp;
2693         struct inode               *inode;
2694         struct lu_fid              *fid    = &info->oti_fid;
2695         bool                        dirty  = false;
2696         bool                        restored = false;
2697         int                         rc     = 0;
2698         ENTRY;
2699
2700         memset(scrub, 0, sizeof(*scrub));
2701         OBD_SET_CTXT_MAGIC(ctxt);
2702         ctxt->pwdmnt = dev->od_mnt;
2703         ctxt->pwd = dev->od_mnt->mnt_root;
2704         ctxt->fs = get_ds();
2705
2706         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2707         init_rwsem(&scrub->os_rwsem);
2708         spin_lock_init(&scrub->os_lock);
2709         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2710
2711         push_ctxt(&saved, ctxt);
2712         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2713         if (IS_ERR(filp)) {
2714                 pop_ctxt(&saved, ctxt);
2715                 RETURN(PTR_ERR(filp));
2716         }
2717
2718         inode = file_inode(filp);
2719         /* 'What the @fid is' is not imporatant, because the object
2720          * has no OI mapping, and only is visible inside the OSD.*/
2721         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2722         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2723         if (rc != 0) {
2724                 filp_close(filp, NULL);
2725                 pop_ctxt(&saved, ctxt);
2726                 RETURN(rc);
2727         }
2728
2729         scrub->os_inode = igrab(inode);
2730         filp_close(filp, NULL);
2731         pop_ctxt(&saved, ctxt);
2732
2733         rc = osd_scrub_file_load(scrub);
2734         if (rc == -ENOENT) {
2735                 osd_scrub_file_init(scrub, es->s_uuid);
2736                 /* If the "/O" dir does not exist when mount (indicated by
2737                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2738                  * then it is quite probably that the device is a new one,
2739                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2740                  *
2741                  * For the rare case that "/O" and "OI_scrub" both lost on
2742                  * an old device, it can be found and cleared later.
2743                  *
2744                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2745                  * need to check "filter_fid_old" and to convert it to
2746                  * "filter_fid" for each object, and all the IGIF should
2747                  * have their FID mapping in OI files already. */
2748                 if (dev->od_maybe_new)
2749                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2750                 dirty = true;
2751         } else if (rc != 0) {
2752                 GOTO(cleanup_inode, rc);
2753         } else {
2754                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2755                         struct obd_uuid *old_uuid;
2756                         struct obd_uuid *new_uuid;
2757
2758                         OBD_ALLOC_PTR(old_uuid);
2759                         OBD_ALLOC_PTR(new_uuid);
2760                         if (old_uuid == NULL || new_uuid == NULL) {
2761                                 CERROR("%.16s: UUID has been changed, but"
2762                                        "failed to allocate RAM for report\n",
2763                                        LDISKFS_SB(sb)->s_es->s_volume_name);
2764                         } else {
2765                                 class_uuid_unparse(sf->sf_uuid, old_uuid);
2766                                 class_uuid_unparse(es->s_uuid, new_uuid);
2767                                 CERROR("%.16s: UUID has been changed from "
2768                                        "%s to %s\n",
2769                                        LDISKFS_SB(sb)->s_es->s_volume_name,
2770                                        old_uuid->uuid, new_uuid->uuid);
2771                         }
2772                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2773                         dirty = true;
2774                         restored = true;
2775                         if (old_uuid != NULL)
2776                                 OBD_FREE_PTR(old_uuid);
2777                         if (new_uuid != NULL)
2778                                 OBD_FREE_PTR(new_uuid);
2779                 } else if (sf->sf_status == SS_SCANNING) {
2780                         sf->sf_status = SS_CRASHED;
2781                         dirty = true;
2782                 }
2783         }
2784
2785         if (sf->sf_pos_last_checkpoint != 0)
2786                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2787         else
2788                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2789
2790         if (dirty) {
2791                 rc = osd_scrub_file_store(scrub);
2792                 if (rc != 0)
2793                         GOTO(cleanup_inode, rc);
2794         }
2795
2796         /* Initialize OI files. */
2797         rc = osd_oi_init(info, dev, restored);
2798         if (rc < 0)
2799                 GOTO(cleanup_inode, rc);
2800
2801         rc = osd_initial_OI_scrub(info, dev);
2802         if (rc != 0)
2803                 GOTO(cleanup_oi, rc);
2804
2805         if (sf->sf_flags & SF_UPGRADE ||
2806             !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2807               sf->sf_success_count > 0)) {
2808                 dev->od_igif_inoi = 0;
2809                 dev->od_check_ff = dev->od_is_ost;
2810         } else {
2811                 dev->od_igif_inoi = 1;
2812                 dev->od_check_ff = 0;
2813         }
2814
2815         if (sf->sf_flags & SF_INCONSISTENT)
2816                 /* The 'od_igif_inoi' will be set under the
2817                  * following cases:
2818                  * 1) new created system, or
2819                  * 2) restored from file-level backup, or
2820                  * 3) the upgrading completed.
2821                  *
2822                  * The 'od_igif_inoi' may be cleared by OI scrub
2823                  * later if found that the system is upgrading. */
2824                 dev->od_igif_inoi = 1;
2825
2826         if (!dev->od_dt_dev.dd_rdonly && !dev->od_noscrub &&
2827             ((sf->sf_status == SS_PAUSED) ||
2828              (sf->sf_status == SS_CRASHED &&
2829               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2830                               SF_UPGRADE | SF_AUTO)) ||
2831              (sf->sf_status == SS_INIT &&
2832               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2833                               SF_UPGRADE))))
2834                 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2835
2836         if (rc != 0)
2837                 GOTO(cleanup_oi, rc);
2838
2839         /* it is possible that dcache entries may keep objects after they are
2840          * deleted by OSD. While it looks safe this can cause object data to
2841          * stay until umount causing failures in tests calculating free space,
2842          * e.g. replay-ost-single. Since those dcache entries are not used
2843          * anymore let's just free them after use here */
2844         shrink_dcache_sb(sb);
2845
2846         RETURN(0);
2847 cleanup_oi:
2848         osd_oi_fini(info, dev);
2849 cleanup_inode:
2850         iput(scrub->os_inode);
2851         scrub->os_inode = NULL;
2852
2853         return rc;
2854 }
2855
2856 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2857 {
2858         struct osd_scrub *scrub = &dev->od_scrub;
2859
2860         LASSERT(dev->od_otable_it == NULL);
2861
2862         if (scrub->os_inode != NULL) {
2863                 osd_scrub_stop(dev);
2864                 iput(scrub->os_inode);
2865                 scrub->os_inode = NULL;
2866         }
2867         if (dev->od_oi_table != NULL)
2868                 osd_oi_fini(osd_oti_get(env), dev);
2869 }
2870
2871 /* object table based iteration APIs */
2872
2873 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2874                                        struct dt_object *dt, __u32 attr)
2875 {
2876         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2877         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2878         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2879         struct osd_scrub       *scrub = &dev->od_scrub;
2880         struct osd_otable_it   *it;
2881         __u32                   start = 0;
2882         int                     rc;
2883         ENTRY;
2884
2885         /* od_otable_mutex: prevent curcurrent init/fini */
2886         mutex_lock(&dev->od_otable_mutex);
2887         if (dev->od_otable_it != NULL)
2888                 GOTO(out, it = ERR_PTR(-EALREADY));
2889
2890         OBD_ALLOC_PTR(it);
2891         if (it == NULL)
2892                 GOTO(out, it = ERR_PTR(-ENOMEM));
2893
2894         dev->od_otable_it = it;
2895         it->ooi_dev = dev;
2896         it->ooi_cache.ooc_consumer_idx = -1;
2897         if (flags & DOIF_OUTUSED)
2898                 it->ooi_used_outside = 1;
2899
2900         if (flags & DOIF_RESET)
2901                 start |= SS_RESET;
2902
2903         if (valid & DOIV_ERROR_HANDLE) {
2904                 if (flags & DOIF_FAILOUT)
2905                         start |= SS_SET_FAILOUT;
2906                 else
2907                         start |= SS_CLEAR_FAILOUT;
2908         }
2909
2910         if (valid & DOIV_DRYRUN) {
2911                 if (flags & DOIF_DRYRUN)
2912                         start |= SS_SET_DRYRUN;
2913                 else
2914                         start |= SS_CLEAR_DRYRUN;
2915         }
2916
2917         rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2918         if (rc < 0 && rc != -EALREADY) {
2919                 dev->od_otable_it = NULL;
2920                 OBD_FREE_PTR(it);
2921                 GOTO(out, it = ERR_PTR(rc));
2922         }
2923
2924         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2925
2926         GOTO(out, it);
2927
2928 out:
2929         mutex_unlock(&dev->od_otable_mutex);
2930         return (struct dt_it *)it;
2931 }
2932
2933 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2934 {
2935         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2936         struct osd_device    *dev = it->ooi_dev;
2937
2938         /* od_otable_mutex: prevent curcurrent init/fini */
2939         mutex_lock(&dev->od_otable_mutex);
2940         do_osd_scrub_stop(&dev->od_scrub);
2941         LASSERT(dev->od_otable_it == it);
2942
2943         dev->od_otable_it = NULL;
2944         mutex_unlock(&dev->od_otable_mutex);
2945         OBD_FREE_PTR(it);
2946 }
2947
2948 static int osd_otable_it_get(const struct lu_env *env,
2949                              struct dt_it *di, const struct dt_key *key)
2950 {
2951         return 0;
2952 }
2953
2954 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2955 {
2956 }
2957
2958 static inline int
2959 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2960 {
2961         spin_lock(&scrub->os_lock);
2962         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2963             scrub->os_waiting ||
2964             !thread_is_running(&scrub->os_thread))
2965                 it->ooi_waiting = 0;
2966         else
2967                 it->ooi_waiting = 1;
2968         spin_unlock(&scrub->os_lock);
2969
2970         return !it->ooi_waiting;
2971 }
2972
2973 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2974 {
2975         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2976         struct osd_device       *dev    = it->ooi_dev;
2977         struct osd_scrub        *scrub  = &dev->od_scrub;
2978         struct osd_otable_cache *ooc    = &it->ooi_cache;
2979         struct ptlrpc_thread    *thread = &scrub->os_thread;
2980         struct l_wait_info       lwi    = { 0 };
2981         int                      rc;
2982         ENTRY;
2983
2984         LASSERT(it->ooi_user_ready);
2985
2986 again:
2987         if (!thread_is_running(thread) && !it->ooi_used_outside)
2988                 RETURN(1);
2989
2990         if (ooc->ooc_cached_items > 0) {
2991                 ooc->ooc_cached_items--;
2992                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2993                                         ~OSD_OTABLE_IT_CACHE_MASK;
2994                 RETURN(0);
2995         }
2996
2997         if (it->ooi_all_cached) {
2998                 l_wait_event(thread->t_ctl_waitq,
2999                              !thread_is_running(thread),
3000                              &lwi);
3001                 RETURN(1);
3002         }
3003
3004         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
3005                 spin_lock(&scrub->os_lock);
3006                 scrub->os_waiting = 0;
3007                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
3008                 spin_unlock(&scrub->os_lock);
3009         }
3010
3011         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
3012                 l_wait_event(thread->t_ctl_waitq,
3013                              osd_otable_it_wakeup(scrub, it),
3014                              &lwi);
3015
3016         if (!thread_is_running(thread) && !it->ooi_used_outside)
3017                 RETURN(1);
3018
3019         rc = osd_otable_it_preload(env, it);
3020         if (rc >= 0)
3021                 goto again;
3022
3023         RETURN(rc);
3024 }
3025
3026 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
3027                                         const struct dt_it *di)
3028 {
3029         return NULL;
3030 }
3031
3032 static int osd_otable_it_key_size(const struct lu_env *env,
3033                                   const struct dt_it *di)
3034 {
3035         return sizeof(__u64);
3036 }
3037
3038 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
3039                              struct dt_rec *rec, __u32 attr)
3040 {
3041         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
3042         struct osd_otable_cache *ooc = &it->ooi_cache;
3043
3044         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
3045
3046         /* Filter out Invald FID already. */
3047         LASSERTF(fid_is_sane((struct lu_fid *)rec),
3048                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
3049                  PFID((struct lu_fid *)rec),
3050                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
3051
3052         return 0;
3053 }
3054
3055 static __u64 osd_otable_it_store(const struct lu_env *env,
3056                                  const struct dt_it *di)
3057 {
3058         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
3059         struct osd_otable_cache *ooc = &it->ooi_cache;
3060         __u64                    hash;
3061
3062         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
3063                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
3064         else
3065                 hash = ooc->ooc_pos_preload;
3066         return hash;
3067 }
3068
3069 /**
3070  * Set the OSD layer iteration start position as the specified hash.
3071  */
3072 static int osd_otable_it_load(const struct lu_env *env,
3073                               const struct dt_it *di, __u64 hash)
3074 {
3075         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
3076         struct osd_device       *dev   = it->ooi_dev;
3077         struct osd_otable_cache *ooc   = &it->ooi_cache;
3078         struct osd_scrub        *scrub = &dev->od_scrub;
3079         int                      rc;
3080         ENTRY;
3081
3082         /* Forbid to set iteration position after iteration started. */
3083         if (it->ooi_user_ready)
3084                 RETURN(-EPERM);
3085
3086         LASSERT(!scrub->os_partial_scan);
3087
3088         if (hash > OSD_OTABLE_MAX_HASH)
3089                 hash = OSD_OTABLE_MAX_HASH;
3090
3091         ooc->ooc_pos_preload = hash;
3092         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
3093                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
3094
3095         it->ooi_user_ready = 1;
3096         if (!scrub->os_full_speed)
3097                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
3098
3099         /* Unplug OSD layer iteration by the first next() call. */
3100         rc = osd_otable_it_next(env, (struct dt_it *)it);
3101
3102         RETURN(rc);
3103 }
3104
3105 static int osd_otable_it_key_rec(const struct lu_env *env,
3106                                  const struct dt_it *di, void *key_rec)
3107 {
3108         return 0;
3109 }
3110
3111 const struct dt_index_operations osd_otable_ops = {
3112         .dio_it = {
3113                 .init     = osd_otable_it_init,
3114                 .fini     = osd_otable_it_fini,
3115                 .get      = osd_otable_it_get,
3116                 .put      = osd_otable_it_put,
3117                 .next     = osd_otable_it_next,
3118                 .key      = osd_otable_it_key,
3119                 .key_size = osd_otable_it_key_size,
3120                 .rec      = osd_otable_it_rec,
3121                 .store    = osd_otable_it_store,
3122                 .load     = osd_otable_it_load,
3123                 .key_rec  = osd_otable_it_key_rec,
3124         }
3125 };
3126
3127 /* high priority inconsistent items list APIs */
3128
3129 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL  60
3130
3131 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
3132                    int insert)
3133 {
3134         struct osd_inconsistent_item *oii;
3135         struct osd_scrub             *scrub  = &dev->od_scrub;
3136         struct ptlrpc_thread         *thread = &scrub->os_thread;
3137         int                           wakeup = 0;
3138         ENTRY;
3139
3140         OBD_ALLOC_PTR(oii);
3141         if (unlikely(oii == NULL))
3142                 RETURN(-ENOMEM);
3143
3144         INIT_LIST_HEAD(&oii->oii_list);
3145         oii->oii_cache = *oic;
3146         oii->oii_insert = insert;
3147
3148         if (scrub->os_partial_scan) {
3149                 __u64 now = cfs_time_current_sec();
3150
3151                 /* If there haven't been errors in a long time,
3152                  * decay old count until either the errors are
3153                  * gone or we reach the current interval. */
3154                 while (unlikely(scrub->os_bad_oimap_count > 0 &&
3155                                 scrub->os_bad_oimap_time +
3156                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
3157                         scrub->os_bad_oimap_count >>= 1;
3158                         scrub->os_bad_oimap_time +=
3159                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
3160                 }
3161
3162                 scrub->os_bad_oimap_time = now;
3163                 if (++scrub->os_bad_oimap_count >
3164                     dev->od_full_scrub_threshold_rate)
3165                         scrub->os_full_scrub = 1;
3166         }
3167
3168         spin_lock(&scrub->os_lock);
3169         if (unlikely(!thread_is_running(thread))) {
3170                 spin_unlock(&scrub->os_lock);
3171                 OBD_FREE_PTR(oii);
3172                 RETURN(-EAGAIN);
3173         }
3174
3175         if (list_empty(&scrub->os_inconsistent_items))
3176                 wakeup = 1;
3177         list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
3178         spin_unlock(&scrub->os_lock);
3179
3180         if (wakeup != 0)
3181                 wake_up_all(&thread->t_ctl_waitq);
3182
3183         RETURN(0);
3184 }
3185
3186 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
3187                    struct osd_inode_id *id)
3188 {
3189         struct osd_scrub             *scrub = &dev->od_scrub;
3190         struct osd_inconsistent_item *oii;
3191         ENTRY;
3192
3193         spin_lock(&scrub->os_lock);
3194         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
3195                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
3196                         *id = oii->oii_cache.oic_lid;
3197                         spin_unlock(&scrub->os_lock);
3198                         RETURN(0);
3199                 }
3200         }
3201         spin_unlock(&scrub->os_lock);
3202
3203         RETURN(-ENOENT);
3204 }
3205
3206 /* OI scrub dump */
3207
3208 static const char *scrub_status_names[] = {
3209         "init",
3210         "scanning",
3211         "completed",
3212         "failed",
3213         "stopped",
3214         "paused",
3215         "crashed",
3216         NULL
3217 };
3218
3219 static const char *scrub_flags_names[] = {
3220         "recreated",
3221         "inconsistent",
3222         "auto",
3223         "upgrade",
3224         NULL
3225 };
3226
3227 static const char *scrub_param_names[] = {
3228         "failout",
3229         "dryrun",
3230         NULL
3231 };
3232
3233 static void scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
3234                             const char *prefix)
3235 {
3236         int flag;
3237         int i;
3238
3239         seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
3240
3241         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
3242                 if (flag & bits) {
3243                         bits &= ~flag;
3244                         seq_printf(m, "%s%c", names[i],
3245                                    bits != 0 ? ',' : '\n');
3246                 }
3247         }
3248 }
3249
3250 static void scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
3251 {
3252         if (time != 0)
3253                 seq_printf(m, "%s: %llu seconds\n", prefix,
3254                            cfs_time_current_sec() - time);
3255         else
3256                 seq_printf(m, "%s: N/A\n", prefix);
3257 }
3258
3259 static void scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
3260 {
3261         if (pos != 0)
3262                 seq_printf(m, "%s: %llu\n", prefix, pos);
3263         else
3264                 seq_printf(m, "%s: N/A\n", prefix);
3265 }
3266
3267 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
3268 {
3269         struct osd_scrub  *scrub   = &dev->od_scrub;
3270         struct scrub_file *sf      = &scrub->os_file;
3271         __u64              checked;
3272         __u64              speed;
3273
3274         down_read(&scrub->os_rwsem);
3275         seq_printf(m, "name: OI_scrub\n"
3276                    "magic: 0x%x\n"
3277                    "oi_files: %d\n"
3278                    "status: %s\n",
3279                    sf->sf_magic, (int)sf->sf_oi_count,
3280                    scrub_status_names[sf->sf_status]);
3281
3282         scrub_bits_dump(m, sf->sf_flags, scrub_flags_names, "flags");
3283
3284         scrub_bits_dump(m, sf->sf_param, scrub_param_names, "param");
3285
3286         scrub_time_dump(m, sf->sf_time_last_complete,
3287                         "time_since_last_completed");
3288
3289         scrub_time_dump(m, sf->sf_time_latest_start,
3290                         "time_since_latest_start");
3291
3292         scrub_time_dump(m, sf->sf_time_last_checkpoint,
3293                         "time_since_last_checkpoint");
3294
3295         scrub_pos_dump(m, sf->sf_pos_latest_start,
3296                         "latest_start_position");
3297
3298         scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
3299                         "last_checkpoint_position");
3300
3301         scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
3302                         "first_failure_position");
3303
3304         checked = sf->sf_items_checked + scrub->os_new_checked;
3305         seq_printf(m, "checked: %llu\n"
3306                    "updated: %llu\n"
3307                    "failed: %llu\n"
3308                    "prior_updated: %llu\n"
3309                    "noscrub: %llu\n"
3310                    "igif: %llu\n"
3311                    "success_count: %u\n",
3312                    checked, sf->sf_items_updated, sf->sf_items_failed,
3313                    sf->sf_items_updated_prior, sf->sf_items_noscrub,
3314                    sf->sf_items_igif, sf->sf_success_count);
3315
3316         speed = checked;
3317         if (thread_is_running(&scrub->os_thread)) {
3318                 cfs_duration_t duration = cfs_time_current() -
3319                                           scrub->os_time_last_checkpoint;
3320                 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
3321                                                      MSEC_PER_SEC);
3322                 __u32 rtime = sf->sf_run_time +
3323                               cfs_duration_sec(duration + HALF_SEC);
3324
3325                 if (duration != 0)
3326                         do_div(new_checked, duration);
3327                 if (rtime != 0)
3328                         do_div(speed, rtime);
3329                 seq_printf(m, "run_time: %u seconds\n"
3330                            "average_speed: %llu objects/sec\n"
3331                            "real-time_speed: %llu objects/sec\n"
3332                            "current_position: %u\n"
3333                            "lf_scanned: %llu\n"
3334                            "lf_repaired: %llu\n"
3335                            "lf_failed: %llu\n",
3336                            rtime, speed, new_checked, scrub->os_pos_current,
3337                            scrub->os_lf_scanned, scrub->os_lf_repaired,
3338                            scrub->os_lf_failed);
3339         } else {
3340                 if (sf->sf_run_time != 0)
3341                         do_div(speed, sf->sf_run_time);
3342                 seq_printf(m, "run_time: %u seconds\n"
3343                            "average_speed: %llu objects/sec\n"
3344                            "real-time_speed: N/A\n"
3345                            "current_position: N/A\n"
3346                            "lf_scanned: %llu\n"
3347                            "lf_repaired: %llu\n"
3348                            "lf_failed: %llu\n",
3349                            sf->sf_run_time, speed, scrub->os_lf_scanned,
3350                            scrub->os_lf_repaired, scrub->os_lf_failed);
3351         }
3352
3353         up_read(&scrub->os_rwsem);
3354         return 0;
3355 }