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