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, 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 #ifndef EXPORT_SYMTAB
  39 # define EXPORT_SYMTAB
  40 #endif
  41 #define DEBUG_SUBSYSTEM S_MDS
  42
  43 #include <lustre/lustre_idl.h>
  44 #include <lustre_disk.h>
  45 #include <dt_object.h>
  46
  47 #include "osd_internal.h"
  48 #include "osd_oi.h"
  49 #include "osd_scrub.h"
  50
  51 #define HALF_SEC        (CFS_HZ >> 1)
  52
  53 #define OSD_OTABLE_MAX_HASH             0x00000000ffffffffULL
  54
  55 #define SCRUB_NEXT_BREAK        1 /* exit current loop and process next group */
  56 #define SCRUB_NEXT_CONTINUE     2 /* skip current object and process next bit */
  57 #define SCRUB_NEXT_EXIT         3 /* exit all the loops */
  58 #define SCRUB_NEXT_WAIT         4 /* wait for free cache slot */
  59 #define SCRUB_NEXT_CRASH        5 /* simulate system crash during OI scrub */
  60 #define SCRUB_NEXT_FATAL        6 /* simulate failure during OI scrub */
  61 #define SCRUB_NEXT_NOSCRUB      7 /* new created object, no scrub on it */
  62 #define SCRUB_NEXT_NOLMA        8 /* the inode has no FID-in-LMA */
  63
  64 /* misc functions */
  65
  66 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
  67 {
  68         return container_of0(scrub, struct osd_device, od_scrub);
  69 }
  70
  71 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
  72 {
  73         return osd_sb(osd_scrub2dev(scrub));
  74 }
  75
  76 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
  77                                        struct osd_otable_cache *ooc)
  78 {
  79         return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
  80 }
  81
  82 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
  83                                      struct osd_device *dev,
  84                                      const struct lu_fid *fid,
  85                                      const struct osd_inode_id *id, int ops)
  86 {
  87         struct lu_fid         *oi_fid = &info->oti_fid2;
  88         struct osd_inode_id   *oi_id  = &info->oti_id2;
  89         struct iam_container  *bag;
  90         struct iam_path_descr *ipd;
  91         handle_t              *jh;
  92         int                    rc;
  93         ENTRY;
  94
  95         fid_cpu_to_be(oi_fid, fid);
  96         osd_id_pack(oi_id, id);
  97         jh = ldiskfs_journal_start_sb(osd_sb(dev),
  98                                       osd_dto_credits_noquota[ops]);
  99         if (IS_ERR(jh)) {
 100                 rc = PTR_ERR(jh);
 101                 CERROR("%.16s: fail to start trans for scrub store: rc = %d\n",
 102                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
 103                 RETURN(rc);
 104         }
 105
 106         bag = &osd_fid2oi(dev, fid)->oi_dir.od_container;
 107         ipd = osd_idx_ipd_get(info->oti_env, bag);
 108         if (unlikely(ipd == NULL)) {
 109                 ldiskfs_journal_stop(jh);
 110                 CERROR("%.16s: fail to get ipd for scrub store\n",
 111                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name);
 112                 RETURN(-ENOMEM);
 113         }
 114
 115         if (ops == DTO_INDEX_UPDATE) {
 116                 rc = iam_update(jh, bag, (const struct iam_key *)oi_fid,
 117                                 (struct iam_rec *)oi_id, ipd);
 118         } else {
 119                 rc = iam_insert(jh, bag, (const struct iam_key *)oi_fid,
 120                                 (struct iam_rec *)oi_id, ipd);
 121                 if (rc == -EEXIST) {
 122                         rc = 1;
 123                         /* XXX: There are trouble things when adding OI
 124                          *      mapping for IGIF object, which may cause
 125                          *      multiple objects to be mapped to the same
 126                          *      IGIF formatted FID. Consider the following
 127                          *      situations:
 128                          *
 129                          *      1) The MDT is upgrading from 1.8 device.
 130                          *      The OI scrub generates IGIF FID1 for the
 131                          *      OBJ1 and adds the OI mapping.
 132                          *
 133                          *      2) For some reason, the OI scrub does not
 134                          *      process all the IGIF objects completely.
 135                          *
 136                          *      3) The MDT is backuped and restored against
 137                          *      this device.
 138                          *
 139                          *      4) When the MDT mounts up, the OI scrub will
 140                          *      try to rebuild the OI files. For some IGIF
 141                          *      object, OBJ2, which was not processed by the
 142                          *      OI scrub before the backup/restore, and the
 143                          *      new generated IGIF formatted FID may be just
 144                          *      the FID1, the same as OBJ1.
 145                          *
 146                          *      Under such case, the OI scrub cannot know how
 147                          *      to generate new FID for the OBJ2.
 148                          *
 149                          *      Currently, we do nothing for that. One possible
 150                          *      solution is to generate new normal FID for the
 151                          *      conflict object.
 152                          *
 153                          *      Anyway, it is rare, only exists in theory. */
 154                 }
 155         }
 156         osd_ipd_put(info->oti_env, bag, ipd);
 157         ldiskfs_journal_stop(jh);
 158         RETURN(rc);
 159 }
 160
 161 /* OI_scrub file ops */
 162
 163 static void osd_scrub_file_to_cpu(struct scrub_file *des,
 164                                   struct scrub_file *src)
 165 {
 166         memcpy(des->sf_uuid, src->sf_uuid, 16);
 167         des->sf_flags   = le64_to_cpu(src->sf_flags);
 168         des->sf_magic   = le32_to_cpu(src->sf_magic);
 169         des->sf_status  = le16_to_cpu(src->sf_status);
 170         des->sf_param   = le16_to_cpu(src->sf_param);
 171         des->sf_time_last_complete      =
 172                                 le64_to_cpu(src->sf_time_last_complete);
 173         des->sf_time_latest_start       =
 174                                 le64_to_cpu(src->sf_time_latest_start);
 175         des->sf_time_last_checkpoint    =
 176                                 le64_to_cpu(src->sf_time_last_checkpoint);
 177         des->sf_pos_latest_start        =
 178                                 le64_to_cpu(src->sf_pos_latest_start);
 179         des->sf_pos_last_checkpoint     =
 180                                 le64_to_cpu(src->sf_pos_last_checkpoint);
 181         des->sf_pos_first_inconsistent  =
 182                                 le64_to_cpu(src->sf_pos_first_inconsistent);
 183         des->sf_items_checked           =
 184                                 le64_to_cpu(src->sf_items_checked);
 185         des->sf_items_updated           =
 186                                 le64_to_cpu(src->sf_items_updated);
 187         des->sf_items_failed            =
 188                                 le64_to_cpu(src->sf_items_failed);
 189         des->sf_items_updated_prior     =
 190                                 le64_to_cpu(src->sf_items_updated_prior);
 191         des->sf_run_time        = le32_to_cpu(src->sf_run_time);
 192         des->sf_success_count   = le32_to_cpu(src->sf_success_count);
 193         des->sf_oi_count        = le16_to_cpu(src->sf_oi_count);
 194         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 195 }
 196
 197 static void osd_scrub_file_to_le(struct scrub_file *des,
 198                                  struct scrub_file *src)
 199 {
 200         memcpy(des->sf_uuid, src->sf_uuid, 16);
 201         des->sf_flags   = cpu_to_le64(src->sf_flags);
 202         des->sf_magic   = cpu_to_le32(src->sf_magic);
 203         des->sf_status  = cpu_to_le16(src->sf_status);
 204         des->sf_param   = cpu_to_le16(src->sf_param);
 205         des->sf_time_last_complete      =
 206                                 cpu_to_le64(src->sf_time_last_complete);
 207         des->sf_time_latest_start       =
 208                                 cpu_to_le64(src->sf_time_latest_start);
 209         des->sf_time_last_checkpoint    =
 210                                 cpu_to_le64(src->sf_time_last_checkpoint);
 211         des->sf_pos_latest_start        =
 212                                 cpu_to_le64(src->sf_pos_latest_start);
 213         des->sf_pos_last_checkpoint     =
 214                                 cpu_to_le64(src->sf_pos_last_checkpoint);
 215         des->sf_pos_first_inconsistent  =
 216                                 cpu_to_le64(src->sf_pos_first_inconsistent);
 217         des->sf_items_checked           =
 218                                 cpu_to_le64(src->sf_items_checked);
 219         des->sf_items_updated           =
 220                                 cpu_to_le64(src->sf_items_updated);
 221         des->sf_items_failed            =
 222                                 cpu_to_le64(src->sf_items_failed);
 223         des->sf_items_updated_prior     =
 224                                 cpu_to_le64(src->sf_items_updated_prior);
 225         des->sf_run_time        = cpu_to_le32(src->sf_run_time);
 226         des->sf_success_count   = cpu_to_le32(src->sf_success_count);
 227         des->sf_oi_count        = cpu_to_le16(src->sf_oi_count);
 228         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 229 }
 230
 231 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
 232 {
 233         struct scrub_file *sf = &scrub->os_file;
 234
 235         memset(sf, 0, sizeof(*sf));
 236         memcpy(sf->sf_uuid, uuid, 16);
 237         sf->sf_magic = SCRUB_MAGIC_V1;
 238         sf->sf_status = SS_INIT;
 239 }
 240
 241 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
 242 {
 243         struct scrub_file *sf = &scrub->os_file;
 244
 245         CDEBUG(D_LFSCK, "Reset OI scrub file, flags = "LPX64"\n", flags);
 246         memcpy(sf->sf_uuid, uuid, 16);
 247         sf->sf_status = SS_INIT;
 248         sf->sf_flags |= flags;
 249         sf->sf_param = 0;
 250         sf->sf_run_time = 0;
 251         sf->sf_time_latest_start = 0;
 252         sf->sf_time_last_checkpoint = 0;
 253         sf->sf_pos_latest_start = 0;
 254         sf->sf_pos_last_checkpoint = 0;
 255         sf->sf_pos_first_inconsistent = 0;
 256         sf->sf_items_checked = 0;
 257         sf->sf_items_updated = 0;
 258         sf->sf_items_failed = 0;
 259         sf->sf_items_updated_prior = 0;
 260         sf->sf_items_noscrub = 0;
 261         sf->sf_items_igif = 0;
 262 }
 263
 264 static int osd_scrub_file_load(struct osd_scrub *scrub)
 265 {
 266         loff_t  pos  = 0;
 267         char   *name = LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
 268         int     len  = sizeof(scrub->os_file_disk);
 269         int     rc;
 270
 271         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
 272         if (rc == len) {
 273                 struct scrub_file *sf = &scrub->os_file;
 274
 275                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
 276                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
 277                         CWARN("%.16s: invalid scrub magic 0x%x != 0x%x\n,",
 278                               name, sf->sf_magic, SCRUB_MAGIC_V1);
 279                         /* Process it as new scrub file. */
 280                         rc = -ENOENT;
 281                 } else {
 282                         rc = 0;
 283                 }
 284         } else if (rc != 0) {
 285                 CERROR("%.16s: fail to load scrub file, expected = %d, "
 286                        "rc = %d\n", name, len, rc);
 287                 if (rc > 0)
 288                         rc = -EFAULT;
 289         } else {
 290                 /* return -ENOENT for empty scrub file case. */
 291                 rc = -ENOENT;
 292         }
 293
 294         return rc;
 295 }
 296
 297 int osd_scrub_file_store(struct osd_scrub *scrub)
 298 {
 299         struct osd_device *dev;
 300         handle_t          *jh;
 301         loff_t             pos     = 0;
 302         int                len     = sizeof(scrub->os_file_disk);
 303         int                credits;
 304         int                rc;
 305
 306         dev = container_of0(scrub, struct osd_device, od_scrub);
 307         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
 308                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
 309         jh = ldiskfs_journal_start_sb(osd_sb(dev), credits);
 310         if (IS_ERR(jh)) {
 311                 rc = PTR_ERR(jh);
 312                 CERROR("%.16s: fail to start trans for scrub store, rc = %d\n",
 313                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,rc);
 314                 return rc;
 315         }
 316
 317         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
 318         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
 319                                       len, 0, &pos, jh);
 320         ldiskfs_journal_stop(jh);
 321         if (rc != 0)
 322                 CERROR("%.16s: fail to store scrub file, expected = %d, "
 323                        "rc = %d\n",
 324                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
 325                        len, rc);
 326         scrub->os_time_last_checkpoint = cfs_time_current();
 327         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
 328                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
 329         return rc;
 330 }
 331
 332 /* OI scrub APIs */
 333
 334 static int osd_scrub_prep(struct osd_device *dev)
 335 {
 336         struct osd_scrub     *scrub  = &dev->od_scrub;
 337         struct ptlrpc_thread *thread = &scrub->os_thread;
 338         struct scrub_file    *sf     = &scrub->os_file;
 339         __u32                 flags  = scrub->os_start_flags;
 340         int                   rc;
 341         ENTRY;
 342
 343         down_write(&scrub->os_rwsem);
 344         if (flags & SS_SET_FAILOUT)
 345                 sf->sf_param |= SP_FAILOUT;
 346
 347         if (flags & SS_CLEAR_FAILOUT)
 348                 sf->sf_param &= ~SP_FAILOUT;
 349
 350         if (flags & SS_RESET)
 351                 osd_scrub_file_reset(scrub,
 352                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
 353
 354         if (flags & SS_AUTO) {
 355                 scrub->os_full_speed = 1;
 356                 sf->sf_flags |= SF_AUTO;
 357         } else {
 358                 scrub->os_full_speed = 0;
 359         }
 360
 361         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
 362                 scrub->os_full_speed = 1;
 363
 364         scrub->os_in_prior = 0;
 365         scrub->os_waiting = 0;
 366         scrub->os_paused = 0;
 367         scrub->os_new_checked = 0;
 368         if (sf->sf_pos_last_checkpoint != 0)
 369                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 370         else
 371                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 372
 373         scrub->os_pos_current = sf->sf_pos_latest_start;
 374         sf->sf_status = SS_SCANNING;
 375         sf->sf_time_latest_start = cfs_time_current_sec();
 376         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 377         rc = osd_scrub_file_store(scrub);
 378         if (rc == 0) {
 379                 spin_lock(&scrub->os_lock);
 380                 thread_set_flags(thread, SVC_RUNNING);
 381                 spin_unlock(&scrub->os_lock);
 382                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
 383         }
 384         up_write(&scrub->os_rwsem);
 385
 386         RETURN(rc);
 387 }
 388
 389 static int
 390 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
 391                        struct osd_idmap_cache *oic, int val)
 392 {
 393         struct osd_scrub             *scrub  = &dev->od_scrub;
 394         struct scrub_file            *sf     = &scrub->os_file;
 395         struct lu_fid                *fid    = &oic->oic_fid;
 396         struct osd_inode_id          *lid    = &oic->oic_lid;
 397         struct osd_inode_id          *lid2   = &info->oti_id;
 398         struct osd_inconsistent_item *oii    = NULL;
 399         struct inode                 *inode  = NULL;
 400         int                           ops    = DTO_INDEX_UPDATE;
 401         int                           idx;
 402         int                           rc;
 403         ENTRY;
 404
 405         down_write(&scrub->os_rwsem);
 406         scrub->os_new_checked++;
 407         if (val < 0)
 408                 GOTO(out, rc = val);
 409
 410         if (scrub->os_in_prior)
 411                 oii = cfs_list_entry(oic, struct osd_inconsistent_item,
 412                                      oii_cache);
 413
 414         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
 415                 GOTO(out, rc = 0);
 416
 417         if (fid_is_igif(fid))
 418                 sf->sf_items_igif++;
 419
 420         if ((val == SCRUB_NEXT_NOLMA) &&
 421             (!dev->od_handle_nolma || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
 422                 GOTO(out, rc = 0);
 423
 424         if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA))
 425                 goto iget;
 426
 427         /* XXX: Currently, no FID-in-LMA for OST object, so osd_oi_lookup()
 428          *      without checking FLD is enough.
 429          *
 430          *      It should be updated if FID-in-LMA for OSD object introduced
 431          *      in the future. */
 432         rc = osd_oi_lookup(info, dev, fid, lid2, false);
 433         if (rc != 0) {
 434                 if (rc != -ENOENT)
 435                         GOTO(out, rc);
 436
 437 iget:
 438                 inode = osd_iget(info, dev, lid);
 439                 if (IS_ERR(inode)) {
 440                         rc = PTR_ERR(inode);
 441                         /* Someone removed the inode. */
 442                         if (rc == -ENOENT || rc == -ESTALE)
 443                                 rc = 0;
 444                         GOTO(out, rc);
 445                 }
 446
 447                 /* Prevent the inode to be unlinked during OI scrub. */
 448                 mutex_lock(&inode->i_mutex);
 449                 if (unlikely(inode->i_nlink == 0)) {
 450                         mutex_unlock(&inode->i_mutex);
 451                         iput(inode);
 452                         GOTO(out, rc = 0);
 453                 }
 454
 455                 ops = DTO_INDEX_INSERT;
 456                 idx = osd_oi_fid2idx(dev, fid);
 457                 if (val == SCRUB_NEXT_NOLMA) {
 458                         rc = osd_ea_fid_set(info, inode, fid, 0);
 459                         if (rc != 0)
 460                                 GOTO(out, rc);
 461                 } else {
 462                         sf->sf_flags |= SF_RECREATED | SF_INCONSISTENT;
 463                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 464                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 465                 }
 466         } else if (osd_id_eq(lid, lid2)) {
 467                 GOTO(out, rc = 0);
 468         } else {
 469                 sf->sf_flags |= SF_INCONSISTENT;
 470         }
 471
 472         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops);
 473         if (rc == 0) {
 474                 if (scrub->os_in_prior)
 475                         sf->sf_items_updated_prior++;
 476                 else
 477                         sf->sf_items_updated++;
 478         }
 479
 480         GOTO(out, rc);
 481
 482 out:
 483         if (rc < 0) {
 484                 sf->sf_items_failed++;
 485                 if (sf->sf_pos_first_inconsistent == 0 ||
 486                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 487                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 488         } else {
 489                 rc = 0;
 490         }
 491
 492         if (ops == DTO_INDEX_INSERT) {
 493                 mutex_unlock(&inode->i_mutex);
 494                 iput(inode);
 495         }
 496         up_write(&scrub->os_rwsem);
 497
 498         if (oii != NULL) {
 499                 LASSERT(!cfs_list_empty(&oii->oii_list));
 500
 501                 spin_lock(&scrub->os_lock);
 502                 cfs_list_del_init(&oii->oii_list);
 503                 spin_unlock(&scrub->os_lock);
 504                 OBD_FREE_PTR(oii);
 505         }
 506         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 507 }
 508
 509 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
 510 {
 511         struct scrub_file *sf = &scrub->os_file;
 512         int                rc;
 513
 514         if (likely(cfs_time_before(cfs_time_current(),
 515                                    scrub->os_time_next_checkpoint) ||
 516                    scrub->os_new_checked == 0))
 517                 return 0;
 518
 519         down_write(&scrub->os_rwsem);
 520         sf->sf_items_checked += scrub->os_new_checked;
 521         scrub->os_new_checked = 0;
 522         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 523         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 524         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 525                                             scrub->os_time_last_checkpoint);
 526         rc = osd_scrub_file_store(scrub);
 527         up_write(&scrub->os_rwsem);
 528
 529         return rc;
 530 }
 531
 532 static void osd_scrub_post(struct osd_scrub *scrub, int result)
 533 {
 534         struct scrub_file *sf = &scrub->os_file;
 535         ENTRY;
 536
 537         down_write(&scrub->os_rwsem);
 538         spin_lock(&scrub->os_lock);
 539         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 540         spin_unlock(&scrub->os_lock);
 541         if (scrub->os_new_checked > 0) {
 542                 sf->sf_items_checked += scrub->os_new_checked;
 543                 scrub->os_new_checked = 0;
 544                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 545         }
 546         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 547         if (result > 0) {
 548                 sf->sf_status = SS_COMPLETED;
 549                 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 550                 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 551                                   SF_UPGRADE | SF_AUTO);
 552                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 553                 sf->sf_success_count++;
 554         } else if (result == 0) {
 555                 if (scrub->os_paused)
 556                         sf->sf_status = SS_PAUSED;
 557                 else
 558                         sf->sf_status = SS_STOPPED;
 559         } else {
 560                 sf->sf_status = SS_FAILED;
 561         }
 562         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 563                                             scrub->os_time_last_checkpoint);
 564         result = osd_scrub_file_store(scrub);
 565         if (result < 0)
 566                 CERROR("%.16s: fail to osd_scrub_post, rc = %d\n",
 567                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
 568                        result);
 569         up_write(&scrub->os_rwsem);
 570
 571         EXIT;
 572 }
 573
 574 /* iteration engine */
 575
 576 struct osd_iit_param {
 577         struct super_block *sb;
 578         struct buffer_head *bitmap;
 579         ldiskfs_group_t bg;
 580         __u32 gbase;
 581         __u32 offset;
 582 };
 583
 584 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 585                                    struct osd_device *dev,
 586                                    struct osd_iit_param *param,
 587                                    struct osd_idmap_cache **oic,
 588                                    int noslot);
 589
 590 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 591                                    struct osd_device *dev,
 592                                    struct osd_iit_param *param,
 593                                    struct osd_idmap_cache *oic,
 594                                    int *noslot, int rc);
 595
 596 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 597 {
 598         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 599                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 600         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 601                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 602                 return SCRUB_NEXT_BREAK;
 603         } else {
 604                 *pos = param->gbase + param->offset;
 605                 return 0;
 606         }
 607 }
 608
 609 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 610                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 611                         struct super_block *sb, bool scrub)
 612 {
 613         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 614         struct inode            *inode;
 615         int                      rc;
 616
 617         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 618         inode = osd_iget(info, dev, lid);
 619         if (IS_ERR(inode)) {
 620                 rc = PTR_ERR(inode);
 621                 /* The inode may be removed after bitmap searching, or the
 622                  * file is new created without inode initialized yet. */
 623                 if (rc == -ENOENT || rc == -ESTALE)
 624                         return SCRUB_NEXT_CONTINUE;
 625
 626                 CERROR("%.16s: fail to read inode, ino# = %u, rc = %d\n",
 627                        LDISKFS_SB(sb)->s_es->s_volume_name, pos, rc);
 628                 return rc;
 629         }
 630
 631         /* If the inode has no OI mapping, then it is special locally used,
 632          * should be invisible to OI scrub or up layer LFSCK. */
 633         if (ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI)) {
 634                 iput(inode);
 635                 return SCRUB_NEXT_CONTINUE;
 636         }
 637
 638         if (scrub &&
 639             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 640                 /* Only skip it for the first OI scrub accessing. */
 641                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 642                 iput(inode);
 643                 return SCRUB_NEXT_NOSCRUB;
 644         }
 645
 646         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 647         if (rc == 0) {
 648                 if (!scrub) {
 649                         if (!fid_is_client_visible(&lma->lma_self_fid))
 650                                 rc = SCRUB_NEXT_CONTINUE;
 651                         else
 652                                 *fid = lma->lma_self_fid;
 653                 }
 654         } else if (rc == -ENODATA) {
 655                 lu_igif_build(fid, inode->i_ino, inode->i_generation);
 656                 if (scrub)
 657                         rc = SCRUB_NEXT_NOLMA;
 658                 else
 659                         rc = 0;
 660         }
 661         iput(inode);
 662         return rc;
 663 }
 664
 665 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 666                           struct osd_iit_param *param,
 667                           struct osd_idmap_cache **oic, int noslot)
 668 {
 669         struct osd_scrub     *scrub  = &dev->od_scrub;
 670         struct ptlrpc_thread *thread = &scrub->os_thread;
 671         struct lu_fid        *fid;
 672         struct osd_inode_id  *lid;
 673         int                   rc;
 674
 675         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
 676                 struct l_wait_info lwi;
 677
 678                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
 679                 l_wait_event(thread->t_ctl_waitq,
 680                              !cfs_list_empty(&scrub->os_inconsistent_items) ||
 681                              !thread_is_running(thread),
 682                              &lwi);
 683         }
 684
 685         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 686                 spin_lock(&scrub->os_lock);
 687                 thread_set_flags(thread, SVC_STOPPING);
 688                 spin_unlock(&scrub->os_lock);
 689                 return SCRUB_NEXT_CRASH;
 690         }
 691
 692         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
 693                 return SCRUB_NEXT_FATAL;
 694
 695         if (unlikely(!thread_is_running(thread)))
 696                 return SCRUB_NEXT_EXIT;
 697
 698         if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
 699                 struct osd_inconsistent_item *oii;
 700
 701                 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
 702                                      struct osd_inconsistent_item, oii_list);
 703                 *oic = &oii->oii_cache;
 704                 scrub->os_in_prior = 1;
 705                 return 0;
 706         }
 707
 708         if (noslot != 0)
 709                 return SCRUB_NEXT_WAIT;
 710
 711         rc = osd_iit_next(param, &scrub->os_pos_current);
 712         if (rc != 0)
 713                 return rc;
 714
 715         *oic = &scrub->os_oic;
 716         fid = &(*oic)->oic_fid;
 717         lid = &(*oic)->oic_lid;
 718         rc = osd_iit_iget(info, dev, fid, lid,
 719                           scrub->os_pos_current, param->sb, true);
 720         return rc;
 721 }
 722
 723 static int osd_preload_next(struct osd_thread_info *info,
 724                             struct osd_device *dev, struct osd_iit_param *param,
 725                             struct osd_idmap_cache **oic, int noslot)
 726 {
 727         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
 728         struct osd_scrub        *scrub;
 729         struct ptlrpc_thread    *thread;
 730         int                      rc;
 731
 732         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
 733         if (rc != 0)
 734                 return rc;
 735
 736         scrub = &dev->od_scrub;
 737         thread = &scrub->os_thread;
 738         if (thread_is_running(thread) &&
 739             ooc->ooc_pos_preload >= scrub->os_pos_current)
 740                 return SCRUB_NEXT_EXIT;
 741
 742         rc = osd_iit_iget(info, dev,
 743                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
 744                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
 745                           ooc->ooc_pos_preload, param->sb, false);
 746         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
 747          * ignore the failure, so it still need to skip the inode next time. */
 748         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
 749         return rc;
 750 }
 751
 752 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
 753                           struct osd_iit_param *param,
 754                           struct osd_idmap_cache *oic, int *noslot, int rc)
 755 {
 756         struct l_wait_info       lwi    = { 0 };
 757         struct osd_scrub        *scrub  = &dev->od_scrub;
 758         struct scrub_file       *sf     = &scrub->os_file;
 759         struct ptlrpc_thread    *thread = &scrub->os_thread;
 760         struct osd_otable_it    *it     = dev->od_otable_it;
 761         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
 762
 763         switch (rc) {
 764         case SCRUB_NEXT_CONTINUE:
 765                 goto next;
 766         case SCRUB_NEXT_WAIT:
 767                 goto wait;
 768         case SCRUB_NEXT_NOSCRUB:
 769                 down_write(&scrub->os_rwsem);
 770                 scrub->os_new_checked++;
 771                 sf->sf_items_noscrub++;
 772                 up_write(&scrub->os_rwsem);
 773                 goto next;
 774         }
 775
 776         rc = osd_scrub_check_update(info, dev, oic, rc);
 777         if (rc != 0)
 778                 return rc;
 779
 780         rc = osd_scrub_checkpoint(scrub);
 781         if (rc != 0) {
 782                 CERROR("%.16s: fail to checkpoint, pos = %u, rc = %d\n",
 783                        LDISKFS_SB(param->sb)->s_es->s_volume_name,
 784                        scrub->os_pos_current, rc);
 785                 /* Continue, as long as the scrub itself can go ahead. */
 786         }
 787
 788         if (scrub->os_in_prior) {
 789                 scrub->os_in_prior = 0;
 790                 return 0;
 791         }
 792
 793 next:
 794         scrub->os_pos_current = param->gbase + ++(param->offset);
 795         if (it != NULL && it->ooi_waiting &&
 796             ooc->ooc_pos_preload < scrub->os_pos_current) {
 797                 it->ooi_waiting = 0;
 798                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
 799         }
 800
 801         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
 802                 return 0;
 803
 804 wait:
 805         if (osd_scrub_has_window(scrub, ooc)) {
 806                 *noslot = 0;
 807                 return 0;
 808         }
 809
 810         scrub->os_waiting = 1;
 811         l_wait_event(thread->t_ctl_waitq,
 812                      osd_scrub_has_window(scrub, ooc) ||
 813                      !cfs_list_empty(&scrub->os_inconsistent_items) ||
 814                      !thread_is_running(thread),
 815                      &lwi);
 816         scrub->os_waiting = 0;
 817
 818         if (osd_scrub_has_window(scrub, ooc))
 819                 *noslot = 0;
 820         else
 821                 *noslot = 1;
 822         return 0;
 823 }
 824
 825 static int osd_preload_exec(struct osd_thread_info *info,
 826                             struct osd_device *dev, struct osd_iit_param *param,
 827                             struct osd_idmap_cache *oic, int *noslot, int rc)
 828 {
 829         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
 830
 831         if (rc == 0) {
 832                 ooc->ooc_cached_items++;
 833                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
 834                                         ~OSD_OTABLE_IT_CACHE_MASK;
 835         }
 836         return rc > 0 ? 0 : rc;
 837 }
 838
 839 #define SCRUB_IT_ALL    1
 840 #define SCRUB_IT_CRASH  2
 841
 842 static int osd_inode_iteration(struct osd_thread_info *info,
 843                                struct osd_device *dev, __u32 max, bool preload)
 844 {
 845         osd_iit_next_policy   next;
 846         osd_iit_exec_policy   exec;
 847         __u32                *pos;
 848         __u32                *count;
 849         struct osd_iit_param  param;
 850         __u32                 limit;
 851         int                   noslot = 0;
 852         int                   rc;
 853         ENTRY;
 854
 855         if (!preload) {
 856                 struct osd_scrub *scrub = &dev->od_scrub;
 857
 858                 next = osd_scrub_next;
 859                 exec = osd_scrub_exec;
 860                 pos = &scrub->os_pos_current;
 861                 count = &scrub->os_new_checked;
 862         } else {
 863                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
 864
 865                 next = osd_preload_next;
 866                 exec = osd_preload_exec;
 867                 pos = &ooc->ooc_pos_preload;
 868                 count = &ooc->ooc_cached_items;
 869         }
 870         param.sb = osd_sb(dev);
 871         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
 872
 873         while (*pos <= limit && *count < max) {
 874                 struct osd_idmap_cache *oic = NULL;
 875
 876                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
 877                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
 878                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
 879                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
 880                 if (param.bitmap == NULL) {
 881                         CERROR("%.16s: fail to read bitmap for %u, "
 882                                "scrub will stop, urgent mode\n",
 883                                LDISKFS_SB(param.sb)->s_es->s_volume_name,
 884                                (__u32)param.bg);
 885                         RETURN(-EIO);
 886                 }
 887
 888                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
 889                        *count < max) {
 890                         rc = next(info, dev, &param, &oic, noslot);
 891                         switch (rc) {
 892                         case SCRUB_NEXT_BREAK:
 893                                 goto next_group;
 894                         case SCRUB_NEXT_EXIT:
 895                                 brelse(param.bitmap);
 896                                 RETURN(0);
 897                         case SCRUB_NEXT_CRASH:
 898                                 brelse(param.bitmap);
 899                                 RETURN(SCRUB_IT_CRASH);
 900                         case SCRUB_NEXT_FATAL:
 901                                 brelse(param.bitmap);
 902                                 RETURN(-EINVAL);
 903                         }
 904
 905                         rc = exec(info, dev, &param, oic, &noslot, rc);
 906                         if (rc != 0) {
 907                                 brelse(param.bitmap);
 908                                 RETURN(rc);
 909                         }
 910
 911                         if (preload && dev->od_otable_it->ooi_stopping) {
 912                                 brelse(param.bitmap);
 913                                 RETURN(0);
 914                         }
 915                 }
 916
 917 next_group:
 918                 brelse(param.bitmap);
 919
 920                 if (preload && dev->od_otable_it->ooi_stopping)
 921                         RETURN(0);
 922         }
 923
 924         if (*pos > limit)
 925                 RETURN(SCRUB_IT_ALL);
 926         RETURN(0);
 927 }
 928
 929 static int osd_otable_it_preload(const struct lu_env *env,
 930                                  struct osd_otable_it *it)
 931 {
 932         struct osd_device       *dev   = it->ooi_dev;
 933         struct osd_scrub        *scrub = &dev->od_scrub;
 934         struct osd_otable_cache *ooc   = &it->ooi_cache;
 935         int                      rc;
 936         ENTRY;
 937
 938         rc = osd_inode_iteration(osd_oti_get(env), dev,
 939                                  OSD_OTABLE_IT_CACHE_SIZE, true);
 940         if (rc == SCRUB_IT_ALL)
 941                 it->ooi_all_cached = 1;
 942
 943         CDEBUG(D_LFSCK, "OSD pre-loaded: max = %u, preload = %u, rc = %d\n",
 944                le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count),
 945                ooc->ooc_pos_preload, rc);
 946
 947         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
 948                 scrub->os_waiting = 0;
 949                 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
 950         }
 951
 952         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
 953 }
 954
 955 static int osd_scrub_main(void *args)
 956 {
 957         struct lu_env         env;
 958         struct osd_device    *dev    = (struct osd_device *)args;
 959         struct osd_scrub     *scrub  = &dev->od_scrub;
 960         struct ptlrpc_thread *thread = &scrub->os_thread;
 961         struct super_block   *sb     = osd_sb(dev);
 962         int                   rc;
 963         ENTRY;
 964
 965         cfs_daemonize("OI_scrub");
 966         rc = lu_env_init(&env, LCT_DT_THREAD);
 967         if (rc != 0) {
 968                 CERROR("%.16s: OI scrub, fail to init env, rc = %d\n",
 969                        LDISKFS_SB(sb)->s_es->s_volume_name, rc);
 970                 GOTO(noenv, rc);
 971         }
 972
 973         rc = osd_scrub_prep(dev);
 974         if (rc != 0) {
 975                 CERROR("%.16s: OI scrub, fail to scrub prep, rc = %d\n",
 976                        LDISKFS_SB(sb)->s_es->s_volume_name, rc);
 977                 GOTO(out, rc);
 978         }
 979
 980         if (!scrub->os_full_speed) {
 981                 struct l_wait_info lwi = { 0 };
 982                 struct osd_otable_it *it = dev->od_otable_it;
 983                 struct osd_otable_cache *ooc = &it->ooi_cache;
 984
 985                 l_wait_event(thread->t_ctl_waitq,
 986                              it->ooi_user_ready || !thread_is_running(thread),
 987                              &lwi);
 988                 if (unlikely(!thread_is_running(thread)))
 989                         GOTO(post, rc = 0);
 990
 991                 LASSERT(scrub->os_pos_current >= ooc->ooc_pos_preload);
 992                 scrub->os_pos_current = ooc->ooc_pos_preload;
 993         }
 994
 995         CDEBUG(D_LFSCK, "OI scrub: flags = 0x%x, pos = %u\n",
 996                scrub->os_start_flags, scrub->os_pos_current);
 997
 998         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
 999         if (unlikely(rc == SCRUB_IT_CRASH))
1000                 GOTO(out, rc = -EINVAL);
1001         GOTO(post, rc);
1002
1003 post:
1004         osd_scrub_post(scrub, rc);
1005         CDEBUG(D_LFSCK, "OI scrub: stop, rc = %d, pos = %u\n",
1006                rc, scrub->os_pos_current);
1007
1008 out:
1009         while (!cfs_list_empty(&scrub->os_inconsistent_items)) {
1010                 struct osd_inconsistent_item *oii;
1011
1012                 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
1013                                      struct osd_inconsistent_item, oii_list);
1014                 cfs_list_del_init(&oii->oii_list);
1015                 OBD_FREE_PTR(oii);
1016         }
1017         lu_env_fini(&env);
1018
1019 noenv:
1020         spin_lock(&scrub->os_lock);
1021         thread_set_flags(thread, SVC_STOPPED);
1022         cfs_waitq_broadcast(&thread->t_ctl_waitq);
1023         spin_unlock(&scrub->os_lock);
1024         return rc;
1025 }
1026
1027 /* initial OI scrub */
1028
1029 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1030                          struct dentry *, filldir_t filldir);
1031
1032 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1033                                loff_t offset, __u64 ino, unsigned d_type);
1034
1035 static int
1036 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1037                      struct dentry *dentry, filldir_t filldir);
1038 static int
1039 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1040                   struct dentry *dentry, filldir_t filldir);
1041
1042 static int
1043 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1044                      struct dentry *dentry, filldir_t filldir);
1045
1046 enum osd_lf_flags {
1047         OLF_SCAN_SUBITEMS       = 0x0001,
1048         OLF_HIDE_FID            = 0x0002,
1049         OLF_SHOW_NAME           = 0x0004,
1050 };
1051
1052 struct osd_lf_map {
1053         char            *olm_name;
1054         struct lu_fid    olm_fid;
1055         __u16            olm_flags;
1056         scandir_t        olm_scandir;
1057         filldir_t        olm_filldir;
1058 };
1059
1060 /* Add the new introduced local files in the list in the future. */
1061 static const struct osd_lf_map osd_lf_maps[] = {
1062         /* CATALOGS */
1063         { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1064                 NULL, NULL },
1065
1066         /* CONFIGS */
1067         { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1068                 OLF_SCAN_SUBITEMS, osd_ios_general_scan,
1069                 osd_ios_varfid_fill },
1070
1071         /* NIDTBL_VERSIONS */
1072         { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1073                 osd_ios_general_scan, osd_ios_varfid_fill },
1074
1075         /* PENDING */
1076         { "PENDING", { FID_SEQ_LOCAL_FILE, MDD_ORPHAN_OID, 0 }, 0, NULL, NULL },
1077
1078         /* ROOT */
1079         { "ROOT", { FID_SEQ_LOCAL_FILE, MDD_ROOT_INDEX_OID, 0 },
1080                 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, osd_ios_ROOT_scan, NULL },
1081
1082         /* capa_keys */
1083         { CAPA_KEYS, { FID_SEQ_LOCAL_FILE, MDD_CAPA_KEYS_OID, 0 }, 0,
1084                 NULL, NULL },
1085
1086         /* changelog_catalog */
1087         { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, NULL, NULL },
1088
1089         /* changelog_users */
1090         { CHANGELOG_USERS, { 0, 0, 0 }, 0, NULL, NULL },
1091
1092         /* fld */
1093         { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1094                 NULL, NULL },
1095
1096         /* last_rcvd */
1097         { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1098                 NULL, NULL },
1099
1100         /* lfsck_bookmark */
1101         { "lfsck_bookmark", { FID_SEQ_LOCAL_FILE, LFSCK_BOOKMARK_OID, 0 }, 0,
1102                 NULL, NULL },
1103
1104         /* lov_objid */
1105         { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1106                 NULL, NULL },
1107
1108         /* lov_objseq */
1109         { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1110                 OLF_SHOW_NAME, NULL, NULL },
1111
1112         /* quota_master */
1113         { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1114                 osd_ios_general_scan, osd_ios_varfid_fill },
1115
1116         /* quota_slave */
1117         { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1118                 osd_ios_general_scan, osd_ios_varfid_fill },
1119
1120         /* seq-200000003-lastid */
1121         { "seq-200000003-lastid", { FID_SEQ_LOCAL_NAME, 1, 0 }, 0,
1122                 NULL, NULL },
1123
1124         /* seq_ctl */
1125         { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1126                 OLF_SHOW_NAME, NULL, NULL },
1127
1128         /* seq_srv */
1129         { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1130                 OLF_SHOW_NAME, NULL, NULL },
1131
1132         /* LAST_GROUP */
1133         { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1134                 OLF_SHOW_NAME, NULL, NULL },
1135
1136         /* health_check */
1137         { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1138                 OLF_SHOW_NAME, NULL, NULL },
1139
1140         /* lfsck_namespace */
1141         { "lfsck_namespace", { FID_SEQ_LOCAL_FILE, LFSCK_NAMESPACE_OID, 0 }, 0,
1142                 NULL, NULL },
1143
1144         /* OBJECTS, upgrade from old device */
1145         { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, osd_ios_OBJECTS_scan, NULL },
1146
1147         /* lquota_v2.user, upgrade from old device */
1148         { "lquota_v2.user", { 0, 0, 0 }, 0, NULL, NULL },
1149
1150         /* lquota_v2.group, upgrade from old device */
1151         { "lquota_v2.group", { 0, 0, 0 }, 0, NULL, NULL },
1152
1153         { NULL, { 0, 0, 0 }, 0, NULL, NULL }
1154 };
1155
1156 struct osd_ios_item {
1157         cfs_list_t       oii_list;
1158         struct dentry   *oii_dentry;
1159         scandir_t        oii_scandir;
1160         filldir_t        oii_filldir;
1161 };
1162
1163 struct osd_ios_filldir_buf {
1164         struct osd_thread_info  *oifb_info;
1165         struct osd_device       *oifb_dev;
1166         struct dentry           *oifb_dentry;
1167 };
1168
1169 static inline struct dentry *
1170 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1171 {
1172         struct dentry *dentry;
1173
1174         dentry = ll_lookup_one_len(name, parent, namelen);
1175         if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1176                 dput(dentry);
1177                 return ERR_PTR(-ENOENT);
1178         }
1179
1180         return dentry;
1181 }
1182
1183 static inline void
1184 osd_ios_llogname2fid(struct lu_fid *fid, const char *name, int namelen)
1185 {
1186         obd_id id = 0;
1187         int    i  = 0;
1188
1189         fid->f_seq = FID_SEQ_LLOG;
1190         while (i < namelen)
1191                 id = id * 10 + name[i++] - '0';
1192
1193         fid->f_oid = id & 0x00000000ffffffffULL;
1194         fid->f_ver = id >> 32;
1195 }
1196
1197 static inline void
1198 osd_ios_Oname2fid(struct lu_fid *fid, const char *name, int namelen)
1199 {
1200         __u64 seq = 0;
1201         int   i   = 0;
1202
1203         while (i < namelen)
1204                 seq = seq * 10 + name[i++] - '0';
1205
1206         lu_last_id_fid(fid, seq);
1207 }
1208
1209 static int
1210 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1211                  scandir_t scandir, filldir_t filldir)
1212 {
1213         struct osd_ios_item *item;
1214
1215         OBD_ALLOC_PTR(item);
1216         if (item == NULL)
1217                 return -ENOMEM;
1218
1219         CFS_INIT_LIST_HEAD(&item->oii_list);
1220         item->oii_dentry = dget(dentry);
1221         item->oii_scandir = scandir;
1222         item->oii_filldir = filldir;
1223         cfs_list_add_tail(&item->oii_list, &dev->od_ios_list);
1224         return 0;
1225 }
1226
1227 /**
1228  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1229  *
1230  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1231  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1232  * reference the inode, or fixed if it is missing or references another inode.
1233  */
1234 static int
1235 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1236                  struct inode *inode, const struct lu_fid *fid, int flags)
1237 {
1238         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
1239         struct osd_inode_id     *id     = &info->oti_id;
1240         struct osd_inode_id     *id2    = &info->oti_id2;
1241         struct osd_scrub        *scrub  = &dev->od_scrub;
1242         struct scrub_file       *sf     = &scrub->os_file;
1243         struct lu_fid            tfid;
1244         int                      rc;
1245         ENTRY;
1246
1247         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1248         if (rc != 0 && rc != -ENODATA)
1249                 RETURN(rc);
1250
1251         osd_id_gen(id, inode->i_ino, inode->i_generation);
1252         if (rc == -ENODATA) {
1253                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID)
1254                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1255                 else
1256                         tfid = *fid;
1257                 rc = osd_ea_fid_set(info, inode, &tfid, 0);
1258                 if (rc != 0)
1259                         RETURN(rc);
1260         } else {
1261                 tfid = lma->lma_self_fid;
1262         }
1263
1264         rc = __osd_oi_lookup(info, dev, &tfid, id2);
1265         if (rc != 0) {
1266                 if (rc != -ENOENT)
1267                         RETURN(rc);
1268
1269                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1270                                                DTO_INDEX_INSERT);
1271                 RETURN(rc);
1272         }
1273
1274         if (osd_id_eq_strict(id, id2))
1275                 RETURN(0);
1276
1277         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1278                 osd_scrub_file_reset(scrub,
1279                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1280                                      SF_INCONSISTENT);
1281                 rc = osd_scrub_file_store(scrub);
1282                 if (rc != 0)
1283                         RETURN(rc);
1284         }
1285
1286         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id, DTO_INDEX_UPDATE);
1287
1288         RETURN(rc);
1289 }
1290
1291 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1292                                loff_t offset, __u64 ino, unsigned d_type)
1293 {
1294         struct osd_ios_filldir_buf *fill_buf = buf;
1295         struct osd_device          *dev      = fill_buf->oifb_dev;
1296         struct dentry              *child;
1297         int                         rc;
1298         ENTRY;
1299
1300         /* skip any '.' started names */
1301         if (name[0] == '.')
1302                 RETURN(0);
1303
1304         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1305         if (IS_ERR(child))
1306                 RETURN(PTR_ERR(child));
1307
1308         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1309                               NULL, 0);
1310         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1311                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1312                                       osd_ios_varfid_fill);
1313         dput(child);
1314
1315         RETURN(rc);
1316 }
1317
1318 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
1319                              loff_t offset, __u64 ino, unsigned d_type)
1320 {
1321         struct osd_ios_filldir_buf *fill_buf = buf;
1322         struct osd_device          *dev      = fill_buf->oifb_dev;
1323         const struct osd_lf_map    *map;
1324         struct dentry              *child;
1325         int                         rc       = 0;
1326         ENTRY;
1327
1328         /* skip any '.' started names */
1329         if (name[0] == '.')
1330                 RETURN(0);
1331
1332         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
1333                 if (strlen(map->olm_name) != namelen)
1334                         continue;
1335
1336                 if (strncmp(map->olm_name, name, namelen) == 0)
1337                         break;
1338         }
1339
1340         if (map->olm_name == NULL)
1341                 RETURN(0);
1342
1343         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1344         if (IS_ERR(child))
1345                 RETURN(PTR_ERR(child));
1346
1347         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1348                               &map->olm_fid, map->olm_flags);
1349         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
1350                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
1351                                       map->olm_filldir);
1352         dput(child);
1353
1354         RETURN(rc);
1355 }
1356
1357 static int
1358 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1359                      struct dentry *dentry, filldir_t filldir)
1360 {
1361         struct osd_ios_filldir_buf    buf   = { info, dev, dentry };
1362         struct file                  *filp  = &info->oti_it_ea.oie_file;
1363         struct inode                 *inode = dentry->d_inode;
1364         const struct file_operations *fops  = inode->i_fop;
1365         int                           rc;
1366         ENTRY;
1367
1368         LASSERT(filldir != NULL);
1369
1370         filp->f_pos = 0;
1371         filp->f_dentry = dentry;
1372         filp->f_mode = FMODE_64BITHASH;
1373         filp->f_mapping = inode->i_mapping;
1374         filp->f_op = fops;
1375         filp->private_data = NULL;
1376
1377         rc = fops->readdir(filp, &buf, filldir);
1378         fops->release(inode, filp);
1379
1380         RETURN(rc);
1381 }
1382
1383 static int
1384 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1385                   struct dentry *dentry, filldir_t filldir)
1386 {
1387         struct osd_scrub  *scrub  = &dev->od_scrub;
1388         struct scrub_file *sf     = &scrub->os_file;
1389         struct dentry     *child;
1390         int                rc;
1391         ENTRY;
1392
1393         /* It is existing MDT device. */
1394         dev->od_handle_nolma = 1;
1395         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
1396                                        strlen(dot_lustre_name));
1397         if (IS_ERR(child)) {
1398                 rc = PTR_ERR(child);
1399                 if (rc == -ENOENT) {
1400                         /* It is 1.8 MDT device. */
1401                         if (!(sf->sf_flags & SF_UPGRADE)) {
1402                                 osd_scrub_file_reset(scrub,
1403                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1404                                         SF_UPGRADE);
1405                                 rc = osd_scrub_file_store(scrub);
1406                         } else {
1407                                 rc = 0;
1408                         }
1409                 }
1410         } else {
1411                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1412                 dput(child);
1413         }
1414
1415         RETURN(rc);
1416 }
1417
1418 static int
1419 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1420                      struct dentry *dentry, filldir_t filldir)
1421 {
1422         struct dentry *child;
1423         int            rc;
1424         ENTRY;
1425
1426         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
1427         if (!IS_ERR(child)) {
1428                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1429                 dput(child);
1430         } else {
1431                 rc = PTR_ERR(child);
1432         }
1433
1434         if (rc != 0 && rc != -ENOENT)
1435                 RETURN(rc);
1436
1437         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
1438         if (!IS_ERR(child)) {
1439                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1440                 dput(child);
1441         } else {
1442                 rc = PTR_ERR(child);
1443         }
1444
1445         if (rc == -ENOENT)
1446                 rc = 0;
1447
1448         RETURN(rc);
1449 }
1450
1451 static int osd_initial_OI_scrub(struct osd_thread_info *info,
1452                                 struct osd_device *dev)
1453 {
1454         struct osd_ios_item *item    = NULL;
1455         scandir_t            scandir = osd_ios_general_scan;
1456         filldir_t            filldir = osd_ios_root_fill;
1457         struct dentry       *dentry  = osd_sb(dev)->s_root;
1458         int                  rc;
1459         ENTRY;
1460
1461         while (1) {
1462                 rc = scandir(info, dev, dentry, filldir);
1463                 if (item != NULL) {
1464                         dput(item->oii_dentry);
1465                         OBD_FREE_PTR(item);
1466                 }
1467
1468                 if (rc != 0)
1469                         break;
1470
1471                 if (cfs_list_empty(&dev->od_ios_list))
1472                         break;
1473
1474                 item = cfs_list_entry(dev->od_ios_list.next,
1475                                       struct osd_ios_item, oii_list);
1476                 cfs_list_del_init(&item->oii_list);
1477
1478                 LASSERT(item->oii_scandir != NULL);
1479                 scandir = item->oii_scandir;
1480                 filldir = item->oii_filldir;
1481                 dentry = item->oii_dentry;
1482         }
1483
1484         while (!cfs_list_empty(&dev->od_ios_list)) {
1485                 item = cfs_list_entry(dev->od_ios_list.next,
1486                                       struct osd_ios_item, oii_list);
1487                 cfs_list_del_init(&item->oii_list);
1488                 dput(item->oii_dentry);
1489                 OBD_FREE_PTR(item);
1490         }
1491
1492         RETURN(rc);
1493 }
1494
1495 char *osd_lf_fid2name(const struct lu_fid *fid)
1496 {
1497         const struct osd_lf_map *map = osd_lf_maps;
1498
1499         while (map->olm_name != NULL) {
1500                 if (!lu_fid_eq(fid, &map->olm_fid)) {
1501                         map++;
1502                         continue;
1503                 }
1504
1505                 if (map->olm_flags & OLF_SHOW_NAME)
1506                         return map->olm_name;
1507                 else
1508                         return "";
1509         }
1510
1511         return NULL;
1512 }
1513
1514 /* OI scrub start/stop */
1515
1516 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
1517 {
1518         struct osd_scrub     *scrub  = &dev->od_scrub;
1519         struct ptlrpc_thread *thread = &scrub->os_thread;
1520         struct l_wait_info    lwi    = { 0 };
1521         int                   rc;
1522         ENTRY;
1523
1524 again:
1525         /* os_lock: sync status between stop and scrub thread */
1526         spin_lock(&scrub->os_lock);
1527         if (thread_is_running(thread)) {
1528                 spin_unlock(&scrub->os_lock);
1529                 RETURN(-EALREADY);
1530         } else if (unlikely(thread_is_stopping(thread))) {
1531                 spin_unlock(&scrub->os_lock);
1532                 l_wait_event(thread->t_ctl_waitq,
1533                              thread_is_stopped(thread),
1534                              &lwi);
1535                 goto again;
1536         }
1537         spin_unlock(&scrub->os_lock);
1538
1539         if (scrub->os_file.sf_status == SS_COMPLETED)
1540                 flags |= SS_RESET;
1541
1542         scrub->os_start_flags = flags;
1543         thread_set_flags(thread, 0);
1544         rc = cfs_create_thread(osd_scrub_main, dev, 0);
1545         if (rc < 0) {
1546                 CERROR("%.16s: cannot start iteration thread, rc = %d\n",
1547                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
1548                 RETURN(rc);
1549         }
1550
1551         l_wait_event(thread->t_ctl_waitq,
1552                      thread_is_running(thread) || thread_is_stopped(thread),
1553                      &lwi);
1554
1555         RETURN(0);
1556 }
1557
1558 int osd_scrub_start(struct osd_device *dev)
1559 {
1560         int rc;
1561         ENTRY;
1562
1563         /* od_otable_mutex: prevent curcurrent start/stop */
1564         mutex_lock(&dev->od_otable_mutex);
1565         rc = do_osd_scrub_start(dev, SS_AUTO);
1566         mutex_unlock(&dev->od_otable_mutex);
1567
1568         RETURN(rc == -EALREADY ? 0 : rc);
1569 }
1570
1571 static void do_osd_scrub_stop(struct osd_scrub *scrub)
1572 {
1573         struct ptlrpc_thread *thread = &scrub->os_thread;
1574         struct l_wait_info    lwi    = { 0 };
1575
1576         /* os_lock: sync status between stop and scrub thread */
1577         spin_lock(&scrub->os_lock);
1578         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
1579                 thread_set_flags(thread, SVC_STOPPING);
1580                 spin_unlock(&scrub->os_lock);
1581                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
1582                 l_wait_event(thread->t_ctl_waitq,
1583                              thread_is_stopped(thread),
1584                              &lwi);
1585                 /* Do not skip the last lock/unlock, which can guarantee that
1586                  * the caller cannot return until the OI scrub thread exit. */
1587                 spin_lock(&scrub->os_lock);
1588         }
1589         spin_unlock(&scrub->os_lock);
1590 }
1591
1592 static void osd_scrub_stop(struct osd_device *dev)
1593 {
1594         /* od_otable_mutex: prevent curcurrent start/stop */
1595         mutex_lock(&dev->od_otable_mutex);
1596         dev->od_scrub.os_paused = 1;
1597         do_osd_scrub_stop(&dev->od_scrub);
1598         mutex_unlock(&dev->od_otable_mutex);
1599 }
1600
1601 /* OI scrub setup/cleanup */
1602
1603 static const char osd_scrub_name[] = "OI_scrub";
1604
1605 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
1606 {
1607         struct osd_thread_info     *info   = osd_oti_get(env);
1608         struct osd_scrub           *scrub  = &dev->od_scrub;
1609         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
1610         struct scrub_file          *sf     = &scrub->os_file;
1611         struct super_block         *sb     = osd_sb(dev);
1612         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
1613         struct lvfs_run_ctxt        saved;
1614         struct file                *filp;
1615         int                         dirty  = 0;
1616         int                         rc     = 0;
1617         ENTRY;
1618
1619         memset(scrub, 0, sizeof(*scrub));
1620         OBD_SET_CTXT_MAGIC(ctxt);
1621         ctxt->pwdmnt = dev->od_mnt;
1622         ctxt->pwd = dev->od_mnt->mnt_root;
1623         ctxt->fs = get_ds();
1624
1625         cfs_waitq_init(&scrub->os_thread.t_ctl_waitq);
1626         init_rwsem(&scrub->os_rwsem);
1627         spin_lock_init(&scrub->os_lock);
1628         CFS_INIT_LIST_HEAD(&scrub->os_inconsistent_items);
1629
1630         push_ctxt(&saved, ctxt, NULL);
1631         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
1632         if (IS_ERR(filp))
1633                 RETURN(PTR_ERR(filp));
1634
1635         scrub->os_inode = igrab(filp->f_dentry->d_inode);
1636         filp_close(filp, 0);
1637         pop_ctxt(&saved, ctxt, NULL);
1638         ldiskfs_set_inode_state(scrub->os_inode,
1639                                 LDISKFS_STATE_LUSTRE_NO_OI);
1640
1641         rc = osd_scrub_file_load(scrub);
1642         if (rc == -ENOENT) {
1643                 osd_scrub_file_init(scrub, es->s_uuid);
1644                 dirty = 1;
1645         } else if (rc != 0) {
1646                 RETURN(rc);
1647         } else {
1648                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
1649                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
1650                         dirty = 1;
1651                 } else if (sf->sf_status == SS_SCANNING) {
1652                         sf->sf_status = SS_CRASHED;
1653                         dirty = 1;
1654                 }
1655         }
1656
1657         if (sf->sf_pos_last_checkpoint != 0)
1658                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
1659         else
1660                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
1661
1662         if (dirty != 0) {
1663                 rc = osd_scrub_file_store(scrub);
1664                 if (rc != 0)
1665                         RETURN(rc);
1666         }
1667
1668         /* Initialize OI files. */
1669         rc = osd_oi_init(info, dev);
1670         if (rc < 0)
1671                 RETURN(rc);
1672
1673         rc = osd_initial_OI_scrub(info, dev);
1674         if (rc == 0 && !dev->od_noscrub &&
1675             ((sf->sf_status == SS_PAUSED) ||
1676              (sf->sf_status == SS_CRASHED &&
1677               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE |
1678                               SF_AUTO)) ||
1679              (sf->sf_status == SS_INIT &&
1680               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))))
1681                 rc = osd_scrub_start(dev);
1682
1683         RETURN(rc);
1684 }
1685
1686 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
1687 {
1688         struct osd_scrub *scrub = &dev->od_scrub;
1689
1690         LASSERT(dev->od_otable_it == NULL);
1691
1692         if (scrub->os_inode != NULL) {
1693                 osd_scrub_stop(dev);
1694                 iput(scrub->os_inode);
1695                 scrub->os_inode = NULL;
1696         }
1697         if (dev->od_oi_table != NULL)
1698                 osd_oi_fini(osd_oti_get(env), dev);
1699 }
1700
1701 /* object table based iteration APIs */
1702
1703 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
1704                                        struct dt_object *dt, __u32 attr,
1705                                        struct lustre_capa *capa)
1706 {
1707         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
1708         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
1709         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
1710         struct osd_scrub       *scrub = &dev->od_scrub;
1711         struct osd_otable_it   *it;
1712         __u32                   start = 0;
1713         int                     rc;
1714         ENTRY;
1715
1716         /* od_otable_mutex: prevent curcurrent init/fini */
1717         mutex_lock(&dev->od_otable_mutex);
1718         if (dev->od_otable_it != NULL)
1719                 GOTO(out, it = ERR_PTR(-EALREADY));
1720
1721         OBD_ALLOC_PTR(it);
1722         if (it == NULL)
1723                 GOTO(out, it = ERR_PTR(-ENOMEM));
1724
1725         dev->od_otable_it = it;
1726         it->ooi_dev = dev;
1727         it->ooi_pid = cfs_curproc_pid();
1728         it->ooi_cache.ooc_consumer_idx = -1;
1729         if (flags & DOIF_OUTUSED)
1730                 it->ooi_used_outside = 1;
1731
1732         if (flags & DOIF_RESET)
1733                 start |= SS_RESET;
1734
1735         if (valid & DOIV_ERROR_HANDLE) {
1736                 if (flags & DOIF_FAILOUT)
1737                         start |= SS_SET_FAILOUT;
1738                 else
1739                         start |= SS_CLEAR_FAILOUT;
1740         }
1741
1742         rc = do_osd_scrub_start(dev, start);
1743         if (rc < 0 && rc != -EALREADY) {
1744                 dev->od_otable_it = NULL;
1745                 OBD_FREE_PTR(it);
1746                 GOTO(out, it = ERR_PTR(rc));
1747         }
1748
1749         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
1750
1751         GOTO(out, it);
1752
1753 out:
1754         mutex_unlock(&dev->od_otable_mutex);
1755         return (struct dt_it *)it;
1756 }
1757
1758 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
1759 {
1760         struct osd_otable_it *it  = (struct osd_otable_it *)di;
1761         struct osd_device    *dev = it->ooi_dev;
1762
1763         /* od_otable_mutex: prevent curcurrent init/fini */
1764         mutex_lock(&dev->od_otable_mutex);
1765         do_osd_scrub_stop(&dev->od_scrub);
1766         LASSERT(dev->od_otable_it == it);
1767
1768         dev->od_otable_it = NULL;
1769         mutex_unlock(&dev->od_otable_mutex);
1770         OBD_FREE_PTR(it);
1771 }
1772
1773 static int osd_otable_it_get(const struct lu_env *env,
1774                              struct dt_it *di, const struct dt_key *key)
1775 {
1776         return 0;
1777 }
1778
1779 /**
1780  * It is hack here:
1781  *
1782  * Sometimes the otable-based iteration driver (LFSCK) may be blocked in OSD
1783  * layer when someone wants to stop/pause the iteration. Under such case, we
1784  * need some mechanism to notify the event and wakeup the blocker.
1785  */
1786 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
1787 {
1788         struct osd_otable_it *it  = (struct osd_otable_it *)di;
1789         struct osd_device    *dev = it->ooi_dev;
1790
1791         /* od_otable_mutex: prevent curcurrent init/fini */
1792         mutex_lock(&dev->od_otable_mutex);
1793         if (it->ooi_pid == cfs_curproc_pid()) {
1794                 dev->od_scrub.os_paused = 1;
1795         } else {
1796                 struct ptlrpc_thread *thread = &dev->od_scrub.os_thread;
1797
1798                 it->ooi_stopping = 1;
1799                 if (it->ooi_waiting)
1800                         cfs_waitq_broadcast(&thread->t_ctl_waitq);
1801         }
1802         mutex_unlock(&dev->od_otable_mutex);
1803 }
1804
1805 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
1806 {
1807         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
1808         struct osd_device       *dev    = it->ooi_dev;
1809         struct osd_scrub        *scrub  = &dev->od_scrub;
1810         struct osd_otable_cache *ooc    = &it->ooi_cache;
1811         struct ptlrpc_thread    *thread = &scrub->os_thread;
1812         struct l_wait_info       lwi    = { 0 };
1813         int                      rc;
1814         ENTRY;
1815
1816         LASSERT(it->ooi_user_ready);
1817
1818 again:
1819         if (!thread_is_running(thread) && !it->ooi_used_outside)
1820                 RETURN(1);
1821
1822         if (ooc->ooc_cached_items > 0) {
1823                 ooc->ooc_cached_items--;
1824                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
1825                                         ~OSD_OTABLE_IT_CACHE_MASK;
1826                 RETURN(0);
1827         }
1828
1829         if (it->ooi_all_cached) {
1830                 l_wait_event(thread->t_ctl_waitq,
1831                              !thread_is_running(thread),
1832                              &lwi);
1833                 RETURN(1);
1834         }
1835
1836         it->ooi_waiting = 1;
1837         l_wait_event(thread->t_ctl_waitq,
1838                      ooc->ooc_pos_preload < scrub->os_pos_current ||
1839                      !thread_is_running(thread) ||
1840                      it->ooi_stopping,
1841                      &lwi);
1842         it->ooi_waiting = 0;
1843
1844         if (!thread_is_running(thread) && !it->ooi_used_outside)
1845                 RETURN(1);
1846
1847         if (it->ooi_stopping)
1848                 RETURN(0);
1849
1850         rc = osd_otable_it_preload(env, it);
1851         if (rc >= 0)
1852                 goto again;
1853
1854         RETURN(rc);
1855 }
1856
1857 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
1858                                         const struct dt_it *di)
1859 {
1860         return NULL;
1861 }
1862
1863 static int osd_otable_it_key_size(const struct lu_env *env,
1864                                   const struct dt_it *di)
1865 {
1866         return sizeof(__u64);
1867 }
1868
1869 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
1870                              struct dt_rec *rec, __u32 attr)
1871 {
1872         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
1873         struct osd_otable_cache *ooc = &it->ooi_cache;
1874
1875         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
1876
1877         /* Filter out Invald FID already. */
1878         LASSERTF(fid_is_sane((struct lu_fid *)rec),
1879                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
1880                  PFID((struct lu_fid *)rec),
1881                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
1882
1883         return 0;
1884 }
1885
1886 static __u64 osd_otable_it_store(const struct lu_env *env,
1887                                  const struct dt_it *di)
1888 {
1889         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
1890         struct osd_otable_cache *ooc = &it->ooi_cache;
1891         __u64                    hash;
1892
1893         if (it->ooi_user_ready)
1894                 hash = ooc->ooc_pos_preload;
1895         else
1896                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
1897         return hash;
1898 }
1899
1900 /**
1901  * Set the OSD layer iteration start position as the specified hash.
1902  */
1903 static int osd_otable_it_load(const struct lu_env *env,
1904                               const struct dt_it *di, __u64 hash)
1905 {
1906         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
1907         struct osd_device       *dev   = it->ooi_dev;
1908         struct osd_otable_cache *ooc   = &it->ooi_cache;
1909         struct osd_scrub        *scrub = &dev->od_scrub;
1910         int                      rc;
1911         ENTRY;
1912
1913         /* Forbid to set iteration position after iteration started. */
1914         if (it->ooi_user_ready)
1915                 RETURN(-EPERM);
1916
1917         if (hash > OSD_OTABLE_MAX_HASH)
1918                 hash = OSD_OTABLE_MAX_HASH;
1919
1920         /* Skip the one that has been processed last time. */
1921         if (ooc->ooc_pos_preload > hash)
1922                 ooc->ooc_pos_preload = hash;
1923
1924         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
1925                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1926
1927         it->ooi_user_ready = 1;
1928         if (!scrub->os_full_speed)
1929                 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
1930
1931         /* Unplug OSD layer iteration by the first next() call. */
1932         rc = osd_otable_it_next(env, (struct dt_it *)it);
1933
1934         RETURN(rc);
1935 }
1936
1937 static int osd_otable_it_key_rec(const struct lu_env *env,
1938                                  const struct dt_it *di, void *key_rec)
1939 {
1940         return 0;
1941 }
1942
1943 const struct dt_index_operations osd_otable_ops = {
1944         .dio_it = {
1945                 .init     = osd_otable_it_init,
1946                 .fini     = osd_otable_it_fini,
1947                 .get      = osd_otable_it_get,
1948                 .put      = osd_otable_it_put,
1949                 .next     = osd_otable_it_next,
1950                 .key      = osd_otable_it_key,
1951                 .key_size = osd_otable_it_key_size,
1952                 .rec      = osd_otable_it_rec,
1953                 .store    = osd_otable_it_store,
1954                 .load     = osd_otable_it_load,
1955                 .key_rec  = osd_otable_it_key_rec,
1956         }
1957 };
1958
1959 /* high priority inconsistent items list APIs */
1960
1961 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
1962                    int insert)
1963 {
1964         struct osd_inconsistent_item *oii;
1965         struct osd_scrub             *scrub  = &dev->od_scrub;
1966         struct ptlrpc_thread         *thread = &scrub->os_thread;
1967         int                           wakeup = 0;
1968         ENTRY;
1969
1970         OBD_ALLOC_PTR(oii);
1971         if (unlikely(oii == NULL))
1972                 RETURN(-ENOMEM);
1973
1974         CFS_INIT_LIST_HEAD(&oii->oii_list);
1975         oii->oii_cache = *oic;
1976         oii->oii_insert = insert;
1977
1978         spin_lock(&scrub->os_lock);
1979         if (unlikely(!thread_is_running(thread))) {
1980                 spin_unlock(&scrub->os_lock);
1981                 OBD_FREE_PTR(oii);
1982                 RETURN(-EAGAIN);
1983         }
1984
1985         if (cfs_list_empty(&scrub->os_inconsistent_items))
1986                 wakeup = 1;
1987         cfs_list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
1988         spin_unlock(&scrub->os_lock);
1989
1990         if (wakeup != 0)
1991                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
1992
1993         RETURN(0);
1994 }
1995
1996 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
1997                    struct osd_inode_id *id)
1998 {
1999         struct osd_scrub             *scrub = &dev->od_scrub;
2000         struct osd_inconsistent_item *oii;
2001         ENTRY;
2002
2003         spin_lock(&scrub->os_lock);
2004         cfs_list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2005                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2006                         *id = oii->oii_cache.oic_lid;
2007                         spin_unlock(&scrub->os_lock);
2008                         RETURN(0);
2009                 }
2010         }
2011         spin_unlock(&scrub->os_lock);
2012
2013         RETURN(-ENOENT);
2014 }
2015
2016 /* OI scrub dump */
2017
2018 static const char *scrub_status_names[] = {
2019         "init",
2020         "scanning",
2021         "completed",
2022         "failed",
2023         "stopped",
2024         "paused",
2025         "crashed",
2026         NULL
2027 };
2028
2029 static const char *scrub_flags_names[] = {
2030         "recreated",
2031         "inconsistent",
2032         "auto",
2033         "upgrade",
2034         NULL
2035 };
2036
2037 static const char *scrub_param_names[] = {
2038         "failout",
2039         NULL
2040 };
2041
2042 static int scrub_bits_dump(char **buf, int *len, int bits, const char *names[],
2043                            const char *prefix)
2044 {
2045         int save = *len;
2046         int flag;
2047         int rc;
2048         int i;
2049
2050         rc = snprintf(*buf, *len, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2051         if (rc <= 0)
2052                 return -ENOSPC;
2053
2054         *buf += rc;
2055         *len -= rc;
2056         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2057                 if (flag & bits) {
2058                         bits &= ~flag;
2059                         rc = snprintf(*buf, *len, "%s%c", names[i],
2060                                       bits != 0 ? ',' : '\n');
2061                         if (rc <= 0)
2062                                 return -ENOSPC;
2063
2064                         *buf += rc;
2065                         *len -= rc;
2066                 }
2067         }
2068         return save - *len;
2069 }
2070
2071 static int scrub_time_dump(char **buf, int *len, __u64 time, const char *prefix)
2072 {
2073         int rc;
2074
2075         if (time != 0)
2076                 rc = snprintf(*buf, *len, "%s: "LPU64" seconds\n", prefix,
2077                               cfs_time_current_sec() - time);
2078         else
2079                 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
2080         if (rc <= 0)
2081                 return -ENOSPC;
2082
2083         *buf += rc;
2084         *len -= rc;
2085         return rc;
2086 }
2087
2088 static int scrub_pos_dump(char **buf, int *len, __u64 pos, const char *prefix)
2089 {
2090         int rc;
2091
2092         if (pos != 0)
2093                 rc = snprintf(*buf, *len, "%s: "LPU64"\n", prefix, pos);
2094         else
2095                 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
2096         if (rc <= 0)
2097                 return -ENOSPC;
2098
2099         *buf += rc;
2100         *len -= rc;
2101         return rc;
2102 }
2103
2104 int osd_scrub_dump(struct osd_device *dev, char *buf, int len)
2105 {
2106         struct osd_scrub  *scrub   = &dev->od_scrub;
2107         struct scrub_file *sf      = &scrub->os_file;
2108         __u64              checked;
2109         __u64              speed;
2110         int                save    = len;
2111         int                ret     = -ENOSPC;
2112         int                rc;
2113
2114         down_read(&scrub->os_rwsem);
2115         rc = snprintf(buf, len,
2116                       "name: OI_scrub\n"
2117                       "magic: 0x%x\n"
2118                       "oi_files: %d\n"
2119                       "status: %s\n",
2120                       sf->sf_magic, (int)sf->sf_oi_count,
2121                       scrub_status_names[sf->sf_status]);
2122         if (rc <= 0)
2123                 goto out;
2124
2125         buf += rc;
2126         len -= rc;
2127         rc = scrub_bits_dump(&buf, &len, sf->sf_flags, scrub_flags_names,
2128                              "flags");
2129         if (rc < 0)
2130                 goto out;
2131
2132         rc = scrub_bits_dump(&buf, &len, sf->sf_param, scrub_param_names,
2133                              "param");
2134         if (rc < 0)
2135                 goto out;
2136
2137         rc = scrub_time_dump(&buf, &len, sf->sf_time_last_complete,
2138                              "time_since_last_completed");
2139         if (rc < 0)
2140                 goto out;
2141
2142         rc = scrub_time_dump(&buf, &len, sf->sf_time_latest_start,
2143                              "time_since_latest_start");
2144         if (rc < 0)
2145                 goto out;
2146
2147         rc = scrub_time_dump(&buf, &len, sf->sf_time_last_checkpoint,
2148                              "time_since_last_checkpoint");
2149         if (rc < 0)
2150                 goto out;
2151
2152         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_latest_start,
2153                             "latest_start_position");
2154         if (rc < 0)
2155                 goto out;
2156
2157         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_last_checkpoint,
2158                             "last_checkpoint_position");
2159         if (rc < 0)
2160                 goto out;
2161
2162         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_first_inconsistent,
2163                             "first_failure_position");
2164         if (rc < 0)
2165                 goto out;
2166
2167         checked = sf->sf_items_checked + scrub->os_new_checked;
2168         rc = snprintf(buf, len,
2169                       "checked: "LPU64"\n"
2170                       "updated: "LPU64"\n"
2171                       "failed: "LPU64"\n"
2172                       "prior_updated: "LPU64"\n"
2173                       "noscrub: "LPU64"\n"
2174                       "igif: "LPU64"\n"
2175                       "success_count: %u\n",
2176                       checked, sf->sf_items_updated, sf->sf_items_failed,
2177                       sf->sf_items_updated_prior, sf->sf_items_noscrub,
2178                       sf->sf_items_igif, sf->sf_success_count);
2179         if (rc <= 0)
2180                 goto out;
2181
2182         buf += rc;
2183         len -= rc;
2184         speed = checked;
2185         if (thread_is_running(&scrub->os_thread)) {
2186                 cfs_duration_t duration = cfs_time_current() -
2187                                           scrub->os_time_last_checkpoint;
2188                 __u64 new_checked = scrub->os_new_checked * CFS_HZ;
2189                 __u32 rtime = sf->sf_run_time +
2190                               cfs_duration_sec(duration + HALF_SEC);
2191
2192                 if (duration != 0)
2193                         do_div(new_checked, duration);
2194                 if (rtime != 0)
2195                         do_div(speed, rtime);
2196                 rc = snprintf(buf, len,
2197                               "run_time: %u seconds\n"
2198                               "average_speed: "LPU64" objects/sec\n"
2199                               "real-time_speed: "LPU64" objects/sec\n"
2200                               "current_position: %u\n",
2201                               rtime, speed, new_checked, scrub->os_pos_current);
2202         } else {
2203                 if (sf->sf_run_time != 0)
2204                         do_div(speed, sf->sf_run_time);
2205                 rc = snprintf(buf, len,
2206                               "run_time: %u seconds\n"
2207                               "average_speed: "LPU64" objects/sec\n"
2208                               "real-time_speed: N/A\n"
2209                               "current_position: N/A\n",
2210                               sf->sf_run_time, speed);
2211         }
2212         if (rc <= 0)
2213                 goto out;
2214
2215         buf += rc;
2216         len -= rc;
2217         ret = save - len;
2218
2219 out:
2220         up_read(&scrub->os_rwsem);
2221         return ret;
2222 }