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 Whamcloud, Inc.
  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 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
  54 {
  55         return container_of0(scrub, struct osd_device, od_scrub);
  56 }
  57
  58 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
  59 {
  60         return osd_sb(osd_scrub2dev(scrub));
  61 }
  62
  63 static void osd_scrub_file_to_cpu(struct scrub_file *des,
  64                                   struct scrub_file *src)
  65 {
  66         memcpy(des->sf_uuid, src->sf_uuid, 16);
  67         des->sf_flags   = le64_to_cpu(src->sf_flags);
  68         des->sf_magic   = le32_to_cpu(src->sf_magic);
  69         des->sf_status  = le16_to_cpu(src->sf_status);
  70         des->sf_param   = le16_to_cpu(src->sf_param);
  71         des->sf_time_last_complete      =
  72                                 le64_to_cpu(src->sf_time_last_complete);
  73         des->sf_time_latest_start       =
  74                                 le64_to_cpu(src->sf_time_latest_start);
  75         des->sf_time_last_checkpoint    =
  76                                 le64_to_cpu(src->sf_time_last_checkpoint);
  77         des->sf_pos_latest_start        =
  78                                 le64_to_cpu(src->sf_pos_latest_start);
  79         des->sf_pos_last_checkpoint     =
  80                                 le64_to_cpu(src->sf_pos_last_checkpoint);
  81         des->sf_pos_first_inconsistent  =
  82                                 le64_to_cpu(src->sf_pos_first_inconsistent);
  83         des->sf_items_checked           =
  84                                 le64_to_cpu(src->sf_items_checked);
  85         des->sf_items_updated           =
  86                                 le64_to_cpu(src->sf_items_updated);
  87         des->sf_items_failed            =
  88                                 le64_to_cpu(src->sf_items_failed);
  89         des->sf_items_updated_prior     =
  90                                 le64_to_cpu(src->sf_items_updated_prior);
  91         des->sf_run_time        = le32_to_cpu(src->sf_run_time);
  92         des->sf_success_count   = le32_to_cpu(src->sf_success_count);
  93         des->sf_oi_count        = le16_to_cpu(src->sf_oi_count);
  94         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
  95 }
  96
  97 static void osd_scrub_file_to_le(struct scrub_file *des,
  98                                  struct scrub_file *src)
  99 {
 100         memcpy(des->sf_uuid, src->sf_uuid, 16);
 101         des->sf_flags   = cpu_to_le64(src->sf_flags);
 102         des->sf_magic   = cpu_to_le32(src->sf_magic);
 103         des->sf_status  = cpu_to_le16(src->sf_status);
 104         des->sf_param   = cpu_to_le16(src->sf_param);
 105         des->sf_time_last_complete      =
 106                                 cpu_to_le64(src->sf_time_last_complete);
 107         des->sf_time_latest_start       =
 108                                 cpu_to_le64(src->sf_time_latest_start);
 109         des->sf_time_last_checkpoint    =
 110                                 cpu_to_le64(src->sf_time_last_checkpoint);
 111         des->sf_pos_latest_start        =
 112                                 cpu_to_le64(src->sf_pos_latest_start);
 113         des->sf_pos_last_checkpoint     =
 114                                 cpu_to_le64(src->sf_pos_last_checkpoint);
 115         des->sf_pos_first_inconsistent  =
 116                                 cpu_to_le64(src->sf_pos_first_inconsistent);
 117         des->sf_items_checked           =
 118                                 cpu_to_le64(src->sf_items_checked);
 119         des->sf_items_updated           =
 120                                 cpu_to_le64(src->sf_items_updated);
 121         des->sf_items_failed            =
 122                                 cpu_to_le64(src->sf_items_failed);
 123         des->sf_items_updated_prior     =
 124                                 cpu_to_le64(src->sf_items_updated_prior);
 125         des->sf_run_time        = cpu_to_le32(src->sf_run_time);
 126         des->sf_success_count   = cpu_to_le32(src->sf_success_count);
 127         des->sf_oi_count        = cpu_to_le16(src->sf_oi_count);
 128         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 129 }
 130
 131 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
 132 {
 133         struct scrub_file *sf = &scrub->os_file;
 134
 135         memset(sf, 0, sizeof(*sf));
 136         memcpy(sf->sf_uuid, uuid, 16);
 137         sf->sf_magic = SCRUB_MAGIC_V1;
 138         sf->sf_status = SS_INIT;
 139 }
 140
 141 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
 142 {
 143         struct scrub_file *sf = &scrub->os_file;
 144
 145         CDEBUG(D_LFSCK, "Reset OI scrub file, flags = "LPX64"\n", flags);
 146         memcpy(sf->sf_uuid, uuid, 16);
 147         sf->sf_status = SS_INIT;
 148         sf->sf_flags |= flags;
 149         sf->sf_param = 0;
 150         sf->sf_run_time = 0;
 151         sf->sf_time_latest_start = 0;
 152         sf->sf_time_last_checkpoint = 0;
 153         sf->sf_pos_latest_start = 0;
 154         sf->sf_pos_last_checkpoint = 0;
 155         sf->sf_pos_first_inconsistent = 0;
 156         sf->sf_items_checked = 0;
 157         sf->sf_items_updated = 0;
 158         sf->sf_items_failed = 0;
 159         sf->sf_items_updated_prior = 0;
 160 }
 161
 162 static int osd_scrub_file_load(struct osd_scrub *scrub)
 163 {
 164         loff_t  pos  = 0;
 165         char   *name = LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
 166         int     len  = sizeof(scrub->os_file_disk);
 167         int     rc;
 168
 169         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
 170         if (rc == len) {
 171                 struct scrub_file *sf = &scrub->os_file;
 172
 173                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
 174                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
 175                         CWARN("%.16s: invalid scrub magic 0x%x != 0x%x\n,",
 176                               name, sf->sf_magic, SCRUB_MAGIC_V1);
 177                         /* Process it as new scrub file. */
 178                         rc = -ENOENT;
 179                 } else {
 180                         rc = 0;
 181                 }
 182         } else if (rc != 0) {
 183                 CERROR("%.16s: fail to load scrub file, expected = %d, "
 184                        "rc = %d\n", name, len, rc);
 185                 if (rc > 0)
 186                         rc = -EFAULT;
 187         } else {
 188                 /* return -ENOENT for empty scrub file case. */
 189                 rc = -ENOENT;
 190         }
 191
 192         return rc;
 193 }
 194
 195 int osd_scrub_file_store(struct osd_scrub *scrub)
 196 {
 197         struct osd_device *dev;
 198         handle_t          *jh;
 199         loff_t             pos     = 0;
 200         int                len     = sizeof(scrub->os_file_disk);
 201         int                credits;
 202         int                rc;
 203
 204         dev = container_of0(scrub, struct osd_device, od_scrub);
 205         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
 206                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
 207         jh = ldiskfs_journal_start_sb(osd_sb(dev), credits);
 208         if (IS_ERR(jh)) {
 209                 rc = PTR_ERR(jh);
 210                 CERROR("%.16s: fail to start trans for scrub store, rc = %d\n",
 211                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,rc);
 212                 return rc;
 213         }
 214
 215         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
 216         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
 217                                       len, &pos, jh);
 218         ldiskfs_journal_stop(jh);
 219         if (rc != 0)
 220                 CERROR("%.16s: fail to store scrub file, expected = %d, "
 221                        "rc = %d\n",
 222                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
 223                        len, rc);
 224         scrub->os_time_last_checkpoint = cfs_time_current();
 225         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
 226                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
 227         return rc;
 228 }
 229
 230 static int osd_scrub_prep(struct osd_device *dev)
 231 {
 232         struct osd_scrub     *scrub  = &dev->od_scrub;
 233         struct ptlrpc_thread *thread = &scrub->os_thread;
 234         struct scrub_file    *sf     = &scrub->os_file;
 235         __u32                 flags  = scrub->os_start_flags;
 236         int                   rc;
 237         ENTRY;
 238
 239         cfs_down_write(&scrub->os_rwsem);
 240         if (flags & SS_SET_FAILOUT)
 241                 sf->sf_param |= SP_FAILOUT;
 242
 243         if (flags & SS_CLEAR_FAILOUT)
 244                 sf->sf_param &= ~SP_FAILOUT;
 245
 246         if (flags & SS_RESET)
 247                 osd_scrub_file_reset(scrub,
 248                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, sf->sf_flags);
 249
 250         if (flags & SS_AUTO) {
 251                 scrub->os_full_speed = 1;
 252                 sf->sf_flags |= SF_AUTO;
 253         } else {
 254                 scrub->os_full_speed = 0;
 255         }
 256
 257         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT))
 258                 scrub->os_full_speed = 1;
 259
 260         scrub->os_in_prior = 0;
 261         scrub->os_waiting = 0;
 262         scrub->os_new_checked = 0;
 263         if (sf->sf_pos_last_checkpoint != 0)
 264                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 265         else
 266                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev));
 267
 268         scrub->os_pos_current = sf->sf_pos_latest_start;
 269         sf->sf_status = SS_SCANNING;
 270         sf->sf_time_latest_start = cfs_time_current_sec();
 271         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 272         rc = osd_scrub_file_store(scrub);
 273         if (rc == 0) {
 274                 cfs_spin_lock(&scrub->os_lock);
 275                 thread_set_flags(thread, SVC_RUNNING);
 276                 cfs_spin_unlock(&scrub->os_lock);
 277                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
 278         }
 279         cfs_up_write(&scrub->os_rwsem);
 280
 281         RETURN(rc);
 282 }
 283
 284 static int
 285 osd_scrub_error(struct osd_device *dev, struct osd_inode_id *lid, int rc)
 286 {
 287         struct osd_scrub  *scrub = &dev->od_scrub;
 288         struct scrub_file *sf    = &scrub->os_file;
 289
 290         cfs_down_write(&scrub->os_rwsem);
 291         scrub->os_new_checked++;
 292         sf->sf_items_failed++;
 293         if (sf->sf_pos_first_inconsistent == 0 ||
 294             sf->sf_pos_first_inconsistent > lid->oii_ino)
 295                 sf->sf_pos_first_inconsistent = lid->oii_ino;
 296         cfs_up_write(&scrub->os_rwsem);
 297         return sf->sf_param & SP_FAILOUT ? rc : 0;
 298 }
 299
 300 static int
 301 osd_scrub_check_update(struct osd_thread_info *info,  struct osd_device *dev,
 302                        struct osd_idmap_cache *oic)
 303 {
 304         struct osd_scrub             *scrub  = &dev->od_scrub;
 305         struct scrub_file            *sf     = &scrub->os_file;
 306         struct osd_inode_id          *lid2   = &info->oti_id;
 307         struct lu_fid                *oi_fid = &info->oti_fid;
 308         struct osd_inode_id          *oi_id  = &info->oti_id;
 309         handle_t                     *jh     = NULL;
 310         struct osd_inconsistent_item *oii    = NULL;
 311         struct inode                 *inode  = NULL;
 312         struct lu_fid                *fid    = &oic->oic_fid;
 313         struct osd_inode_id          *lid    = &oic->oic_lid;
 314         struct iam_container         *bag;
 315         struct iam_path_descr        *ipd;
 316         int                           ops    = DTO_INDEX_UPDATE;
 317         int                           idx;
 318         int                           rc;
 319         ENTRY;
 320
 321         if (scrub->os_in_prior)
 322                 oii = cfs_list_entry(oic, struct osd_inconsistent_item,
 323                                      oii_cache);
 324
 325         cfs_down_write(&scrub->os_rwsem);
 326         scrub->os_new_checked++;
 327         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
 328                 GOTO(out, rc = 0);
 329
 330         if (oii != NULL && oii->oii_insert)
 331                 goto iget;
 332
 333         rc = osd_oi_lookup(info, dev, fid, lid2);
 334         if (rc != 0) {
 335                 if (rc != -ENOENT)
 336                         GOTO(out, rc);
 337
 338 iget:
 339                 inode = osd_iget(info, dev, lid);
 340                 if (IS_ERR(inode)) {
 341                         rc = PTR_ERR(inode);
 342                         /* Someone removed the inode. */
 343                         if (rc == -ENOENT || rc == -ESTALE)
 344                                 rc = 0;
 345                         GOTO(out, rc);
 346                 }
 347
 348                 /* Prevent the inode to be unlinked during OI scrub. */
 349                 cfs_mutex_lock(&inode->i_mutex);
 350                 if (unlikely(inode->i_nlink == 0)) {
 351                         cfs_mutex_unlock(&inode->i_mutex);
 352                         iput(inode);
 353                         GOTO(out, rc = 0);
 354                 }
 355
 356                 ops = DTO_INDEX_INSERT;
 357                 idx = osd_oi_fid2idx(dev, fid);
 358                 if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 359                         ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 360                 sf->sf_flags |= SF_RECREATED;
 361         } else if (osd_id_eq(lid, lid2)) {
 362                         GOTO(out, rc = 0);
 363         }
 364
 365         sf->sf_flags |= SF_INCONSISTENT;
 366         fid_cpu_to_be(oi_fid, fid);
 367         osd_id_pack(oi_id, &oic->oic_lid);
 368         jh = ldiskfs_journal_start_sb(osd_sb(dev),
 369                                 osd_dto_credits_noquota[ops]);
 370         if (IS_ERR(jh)) {
 371                 rc = PTR_ERR(jh);
 372                 CERROR("%.16s: fail to start trans for scrub store, rc = %d\n",
 373                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
 374                 GOTO(out, rc);
 375         }
 376
 377         bag = &osd_fid2oi(dev, fid)->oi_dir.od_container;
 378         ipd = osd_idx_ipd_get(info->oti_env, bag);
 379         if (unlikely(ipd == NULL)) {
 380                 ldiskfs_journal_stop(jh);
 381                 CERROR("%.16s: fail to get ipd for scrub store\n",
 382                         LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name);
 383                 GOTO(out, rc = -ENOMEM);
 384         }
 385
 386         if (ops == DTO_INDEX_UPDATE)
 387                 rc = iam_update(jh, bag, (const struct iam_key *)oi_fid,
 388                                 (struct iam_rec *)oi_id, ipd);
 389         else
 390                 rc = iam_insert(jh, bag, (const struct iam_key *)oi_fid,
 391                                 (struct iam_rec *)oi_id, ipd);
 392         osd_ipd_put(info->oti_env, bag, ipd);
 393         ldiskfs_journal_stop(jh);
 394         if (rc == 0) {
 395                 if (scrub->os_in_prior)
 396                         sf->sf_items_updated_prior++;
 397                 else
 398                         sf->sf_items_updated++;
 399         }
 400
 401         GOTO(out, rc);
 402
 403 out:
 404         if (rc != 0) {
 405                 sf->sf_items_failed++;
 406                 if (sf->sf_pos_first_inconsistent == 0 ||
 407                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 408                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 409         }
 410
 411         if (ops == DTO_INDEX_INSERT) {
 412                 cfs_mutex_unlock(&inode->i_mutex);
 413                 iput(inode);
 414         }
 415         cfs_up_write(&scrub->os_rwsem);
 416
 417         if (oii != NULL) {
 418                 LASSERT(!cfs_list_empty(&oii->oii_list));
 419
 420                 cfs_spin_lock(&scrub->os_lock);
 421                 cfs_list_del_init(&oii->oii_list);
 422                 cfs_spin_unlock(&scrub->os_lock);
 423                 OBD_FREE_PTR(oii);
 424         }
 425         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 426 }
 427
 428 static int do_osd_scrub_checkpoint(struct osd_scrub *scrub)
 429 {
 430         struct scrub_file *sf = &scrub->os_file;
 431         int                rc;
 432         ENTRY;
 433
 434         cfs_down_write(&scrub->os_rwsem);
 435         sf->sf_items_checked += scrub->os_new_checked;
 436         scrub->os_new_checked = 0;
 437         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 438         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 439         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 440                                             scrub->os_time_last_checkpoint);
 441         rc = osd_scrub_file_store(scrub);
 442         cfs_up_write(&scrub->os_rwsem);
 443
 444         RETURN(rc);
 445 }
 446
 447 static inline int osd_scrub_checkpoint(struct osd_scrub *scrub)
 448 {
 449         if (unlikely(cfs_time_beforeq(scrub->os_time_next_checkpoint,
 450                                       cfs_time_current()) &&
 451                      scrub->os_new_checked > 0))
 452                 return do_osd_scrub_checkpoint(scrub);
 453         return 0;
 454 }
 455
 456 static void osd_scrub_post(struct osd_scrub *scrub, int result)
 457 {
 458         struct scrub_file *sf = &scrub->os_file;
 459         ENTRY;
 460
 461         cfs_down_write(&scrub->os_rwsem);
 462         cfs_spin_lock(&scrub->os_lock);
 463         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 464         cfs_spin_unlock(&scrub->os_lock);
 465         if (scrub->os_new_checked > 0) {
 466                 sf->sf_items_checked += scrub->os_new_checked;
 467                 scrub->os_new_checked = 0;
 468                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 469         }
 470         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 471         if (result > 0) {
 472                 sf->sf_status = SS_COMPLETED;
 473                 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 474                 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT | SF_AUTO);
 475                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 476                 sf->sf_success_count++;
 477         } else if (result == 0) {
 478                 sf->sf_status = SS_PAUSED;
 479         } else {
 480                 sf->sf_status = SS_FAILED;
 481         }
 482         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 483                                             scrub->os_time_last_checkpoint);
 484         result = osd_scrub_file_store(scrub);
 485         if (result < 0)
 486                 CERROR("%.16s: fail to osd_scrub_post, rc = %d\n",
 487                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
 488                        result);
 489         cfs_up_write(&scrub->os_rwsem);
 490
 491         EXIT;
 492 }
 493
 494 #define SCRUB_NEXT_BREAK        1 /* exit current loop and process next group */
 495 #define SCRUB_NEXT_CONTINUE     2 /* skip current object and process next bit */
 496 #define SCRUB_NEXT_EXIT         3 /* exit all the loops */
 497 #define SCRUB_NEXT_WAIT         4 /* wait for free cache slot */
 498
 499 struct osd_iit_param {
 500         struct super_block *sb;
 501         struct buffer_head *bitmap;
 502         ldiskfs_group_t bg;
 503         __u32 gbase;
 504         __u32 offset;
 505 };
 506
 507 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 508                                    struct osd_device *dev,
 509                                    struct osd_iit_param *param,
 510                                    struct osd_idmap_cache **oic,
 511                                    int noslot);
 512
 513 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 514                                    struct osd_device *dev,
 515                                    struct osd_iit_param *param,
 516                                    struct osd_idmap_cache *oic,
 517                                    int *noslot, int rc);
 518
 519 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
 520                                        struct osd_otable_cache *ooc)
 521 {
 522         return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
 523 }
 524
 525 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 526 {
 527         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 528                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 529         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 530                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 531                 return SCRUB_NEXT_BREAK;
 532         } else {
 533                 *pos = param->gbase + param->offset;
 534                 return 0;
 535         }
 536 }
 537
 538 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 539                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 540                         struct super_block *sb, struct inode **pinode)
 541 {
 542         struct inode *inode;
 543         int           rc;
 544
 545         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 546         inode = osd_iget_fid(info, dev, lid, fid);
 547         if (IS_ERR(inode)) {
 548                 rc = PTR_ERR(inode);
 549                 /* The inode may be removed after bitmap searching, or the
 550                  * file is new created without inode initialized yet. */
 551                 if (rc == -ENOENT || rc == -ESTALE)
 552                         return SCRUB_NEXT_CONTINUE;
 553
 554                 CERROR("%.16s: fail to read inode, ino# = %u, rc = %d\n",
 555                        LDISKFS_SB(sb)->s_es->s_volume_name, pos, rc);
 556                 return rc;
 557         }
 558
 559         *pinode = inode;
 560         return 0;
 561 }
 562
 563 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 564                           struct osd_iit_param *param,
 565                           struct osd_idmap_cache **oic, int noslot)
 566 {
 567         struct osd_scrub     *scrub  = &dev->od_scrub;
 568         struct ptlrpc_thread *thread = &scrub->os_thread;
 569         struct lu_fid        *fid;
 570         struct osd_inode_id  *lid;
 571         struct inode         *inode;
 572         int                   rc;
 573
 574         if (unlikely(!thread_is_running(thread)))
 575                 return SCRUB_NEXT_EXIT;
 576
 577         if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
 578                 struct osd_inconsistent_item *oii;
 579
 580                 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
 581                                      struct osd_inconsistent_item, oii_list);
 582                 *oic = &oii->oii_cache;
 583                 scrub->os_in_prior = 1;
 584                 return 0;
 585         }
 586
 587         if (noslot != 0)
 588                 return SCRUB_NEXT_WAIT;
 589
 590         rc = osd_iit_next(param, &scrub->os_pos_current);
 591         if (rc != 0)
 592                 return rc;
 593
 594         *oic = &scrub->os_oic;
 595         fid = &(*oic)->oic_fid;
 596         lid = &(*oic)->oic_lid;
 597         rc = osd_iit_iget(info, dev, fid, lid,
 598                           scrub->os_pos_current, param->sb, &inode);
 599         if (rc != 0)
 600                 return rc;
 601
 602         if (!fid_is_norm(fid) || inode->i_state & I_LUSTRE_NOSCRUB)
 603                 rc = SCRUB_NEXT_CONTINUE;
 604         iput(inode);
 605         return rc;
 606 }
 607
 608 static int osd_preload_next(struct osd_thread_info *info,
 609                             struct osd_device *dev, struct osd_iit_param *param,
 610                             struct osd_idmap_cache **oic, int noslot)
 611 {
 612         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
 613         struct osd_scrub        *scrub;
 614         struct ptlrpc_thread    *thread;
 615         struct inode            *inode;
 616         int                      rc;
 617
 618         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
 619         if (rc != 0)
 620                 return rc;
 621
 622         scrub = &dev->od_scrub;
 623         thread = &scrub->os_thread;
 624         if (thread_is_running(thread) &&
 625             ooc->ooc_pos_preload >= scrub->os_pos_current)
 626                 return SCRUB_NEXT_EXIT;
 627
 628         rc = osd_iit_iget(info, dev,
 629                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
 630                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
 631                           ooc->ooc_pos_preload, param->sb, &inode);
 632         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
 633          * ignore the failure, so it still need to skip the inode next time. */
 634         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
 635         if (rc == 0)
 636                 iput(inode);
 637         return rc;
 638 }
 639
 640 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
 641                           struct osd_iit_param *param,
 642                           struct osd_idmap_cache *oic, int *noslot, int rc)
 643 {
 644         struct l_wait_info       lwi    = { 0 };
 645         struct osd_scrub        *scrub  = &dev->od_scrub;
 646         struct ptlrpc_thread    *thread = &scrub->os_thread;
 647         struct osd_otable_it    *it     = dev->od_otable_it;
 648         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
 649
 650         switch (rc) {
 651         case SCRUB_NEXT_CONTINUE:
 652                 goto next;
 653         case SCRUB_NEXT_WAIT:
 654                 goto wait;
 655         }
 656
 657         LASSERTF(rc <= 0, "unexpected rc = %d\n", rc);
 658
 659         if (rc != 0)
 660                 rc = osd_scrub_error(dev, &oic->oic_lid, rc);
 661         else
 662                 rc = osd_scrub_check_update(info, dev, oic);
 663         if (rc != 0)
 664                 return rc;
 665
 666         rc = osd_scrub_checkpoint(scrub);
 667         if (rc != 0) {
 668                 CERROR("%.16s: fail to checkpoint, pos = %u, rc = %d\n",
 669                        LDISKFS_SB(param->sb)->s_es->s_volume_name,
 670                        scrub->os_pos_current, rc);
 671                 /* Continue, as long as the scrub itself can go ahead. */
 672         }
 673
 674         if (scrub->os_in_prior) {
 675                 scrub->os_in_prior = 0;
 676                 return 0;
 677         }
 678
 679 next:
 680         scrub->os_pos_current = param->gbase + ++(param->offset);
 681         if (it != NULL && it->ooi_waiting &&
 682             ooc->ooc_pos_preload < scrub->os_pos_current) {
 683                 it->ooi_waiting = 0;
 684                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
 685         }
 686
 687         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
 688                 return 0;
 689
 690 wait:
 691         if (osd_scrub_has_window(scrub, ooc)) {
 692                 *noslot = 0;
 693                 return 0;
 694         }
 695
 696         scrub->os_waiting = 1;
 697         l_wait_event(thread->t_ctl_waitq,
 698                      osd_scrub_has_window(scrub, ooc) ||
 699                      !cfs_list_empty(&scrub->os_inconsistent_items) ||
 700                      !thread_is_running(thread),
 701                      &lwi);
 702         scrub->os_waiting = 0;
 703
 704         if (osd_scrub_has_window(scrub, ooc))
 705                 *noslot = 0;
 706         else
 707                 *noslot = 1;
 708         return 0;
 709 }
 710
 711 static int osd_preload_exec(struct osd_thread_info *info,
 712                             struct osd_device *dev, struct osd_iit_param *param,
 713                             struct osd_idmap_cache *oic, int *noslot, int rc)
 714 {
 715         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
 716
 717         if (rc == 0) {
 718                 ooc->ooc_cached_items++;
 719                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
 720                                         ~OSD_OTABLE_IT_CACHE_MASK;
 721         }
 722         return rc > 0 ? 0 : rc;
 723 }
 724
 725 #define SCRUB_IT_ALL    1
 726
 727 static int osd_inode_iteration(struct osd_thread_info *info,
 728                                struct osd_device *dev, __u32 max, int preload)
 729 {
 730         osd_iit_next_policy   next;
 731         osd_iit_exec_policy   exec;
 732         __u32                *pos;
 733         __u32                *count;
 734         struct osd_iit_param  param;
 735         __u32                 limit;
 736         int                   noslot = 0;
 737         int                   rc;
 738         ENTRY;
 739
 740         if (preload == 0) {
 741                 struct osd_scrub *scrub = &dev->od_scrub;
 742
 743                 next = osd_scrub_next;
 744                 exec = osd_scrub_exec;
 745                 pos = &scrub->os_pos_current;
 746                 count = &scrub->os_new_checked;
 747         } else {
 748                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
 749
 750                 next = osd_preload_next;
 751                 exec = osd_preload_exec;
 752                 pos = &ooc->ooc_pos_preload;
 753                 count = &ooc->ooc_cached_items;
 754         }
 755         param.sb = osd_sb(dev);
 756         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
 757
 758         while (*pos <= limit && *count < max) {
 759                 struct osd_idmap_cache *oic = NULL;
 760
 761                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
 762                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
 763                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
 764                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
 765                 if (param.bitmap == NULL) {
 766                         CERROR("%.16s: fail to read bitmap for %u, "
 767                                "scrub will stop, urgent mode\n",
 768                                LDISKFS_SB(param.sb)->s_es->s_volume_name,
 769                                (__u32)param.bg);
 770                         RETURN(-EIO);
 771                 }
 772
 773                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
 774                        *count < max) {
 775                         rc = next(info, dev, &param, &oic, noslot);
 776                         switch (rc) {
 777                         case SCRUB_NEXT_BREAK:
 778                                 goto next_group;
 779                         case SCRUB_NEXT_EXIT:
 780                                 brelse(param.bitmap);
 781                                 RETURN(0);
 782                         }
 783
 784                         rc = exec(info, dev, &param, oic, &noslot, rc);
 785                         if (rc != 0) {
 786                                 brelse(param.bitmap);
 787                                 RETURN(rc);
 788                         }
 789                 }
 790
 791 next_group:
 792                 brelse(param.bitmap);
 793         }
 794
 795         if (*pos > limit)
 796                 RETURN(SCRUB_IT_ALL);
 797         RETURN(0);
 798 }
 799
 800 static int osd_scrub_main(void *args)
 801 {
 802         struct lu_env         env;
 803         struct osd_device    *dev    = (struct osd_device *)args;
 804         struct osd_scrub     *scrub  = &dev->od_scrub;
 805         struct ptlrpc_thread *thread = &scrub->os_thread;
 806         struct super_block   *sb     = osd_sb(dev);
 807         int                   rc;
 808         ENTRY;
 809
 810         cfs_daemonize("OI_scrub");
 811         rc = lu_env_init(&env, LCT_DT_THREAD);
 812         if (rc != 0) {
 813                 CERROR("%.16s: OI scrub, fail to init env, rc = %d\n",
 814                        LDISKFS_SB(sb)->s_es->s_volume_name, rc);
 815                 GOTO(noenv, rc);
 816         }
 817
 818         rc = osd_scrub_prep(dev);
 819         if (rc != 0) {
 820                 CERROR("%.16s: OI scrub, fail to scrub prep, rc = %d\n",
 821                        LDISKFS_SB(sb)->s_es->s_volume_name, rc);
 822                 GOTO(out, rc);
 823         }
 824
 825         if (!scrub->os_full_speed) {
 826                 struct l_wait_info lwi = { 0 };
 827                 struct osd_otable_it *it = dev->od_otable_it;
 828                 struct osd_otable_cache *ooc = &it->ooi_cache;
 829
 830                 l_wait_event(thread->t_ctl_waitq,
 831                              it->ooi_user_ready || !thread_is_running(thread),
 832                              &lwi);
 833                 if (unlikely(!thread_is_running(thread)))
 834                         GOTO(post, rc = 0);
 835
 836                 LASSERT(scrub->os_pos_current >= ooc->ooc_pos_preload);
 837                 scrub->os_pos_current = ooc->ooc_pos_preload;
 838         }
 839
 840         CDEBUG(D_LFSCK, "OI scrub: flags = 0x%x, pos = %u\n",
 841                scrub->os_start_flags, scrub->os_pos_current);
 842
 843         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, 0);
 844         GOTO(post, rc);
 845
 846 post:
 847         osd_scrub_post(scrub, rc);
 848         CDEBUG(D_LFSCK, "OI scrub: stop, rc = %d, pos = %u\n",
 849                rc, scrub->os_pos_current);
 850
 851 out:
 852         while (!cfs_list_empty(&scrub->os_inconsistent_items)) {
 853                 struct osd_inconsistent_item *oii;
 854
 855                 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
 856                                      struct osd_inconsistent_item, oii_list);
 857                 cfs_list_del_init(&oii->oii_list);
 858                 OBD_FREE_PTR(oii);
 859         }
 860         lu_env_fini(&env);
 861
 862 noenv:
 863         cfs_spin_lock(&scrub->os_lock);
 864         thread_set_flags(thread, SVC_STOPPED);
 865         cfs_waitq_broadcast(&thread->t_ctl_waitq);
 866         cfs_spin_unlock(&scrub->os_lock);
 867         return rc;
 868 }
 869
 870 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
 871 {
 872         struct osd_scrub     *scrub  = &dev->od_scrub;
 873         struct ptlrpc_thread *thread = &scrub->os_thread;
 874         struct l_wait_info    lwi    = { 0 };
 875         int                   rc;
 876         ENTRY;
 877
 878 again:
 879         /* os_lock: sync status between stop and scrub thread */
 880         cfs_spin_lock(&scrub->os_lock);
 881         if (thread_is_running(thread)) {
 882                 cfs_spin_unlock(&scrub->os_lock);
 883                 RETURN(-EALREADY);
 884         } else if (unlikely(thread_is_stopping(thread))) {
 885                 cfs_spin_unlock(&scrub->os_lock);
 886                 l_wait_event(thread->t_ctl_waitq,
 887                              thread_is_stopped(thread),
 888                              &lwi);
 889                 goto again;
 890         }
 891         cfs_spin_unlock(&scrub->os_lock);
 892
 893         scrub->os_start_flags = flags;
 894         thread_set_flags(thread, 0);
 895         rc = cfs_create_thread(osd_scrub_main, dev, 0);
 896         if (rc < 0) {
 897                 CERROR("%.16s: cannot start iteration thread, rc = %d\n",
 898                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
 899                 RETURN(rc);
 900         }
 901
 902         l_wait_event(thread->t_ctl_waitq,
 903                      thread_is_running(thread) || thread_is_stopped(thread),
 904                      &lwi);
 905
 906         RETURN(0);
 907 }
 908
 909 int osd_scrub_start(struct osd_device *dev)
 910 {
 911         __u32 flags = SS_AUTO;
 912         int   rc;
 913         ENTRY;
 914
 915         if (dev->od_scrub.os_file.sf_status == SS_COMPLETED)
 916                 flags |= SS_RESET;
 917
 918         /* od_otable_mutex: prevent curcurrent start/stop */
 919         cfs_mutex_lock(&dev->od_otable_mutex);
 920         rc = do_osd_scrub_start(dev, flags);
 921         cfs_mutex_unlock(&dev->od_otable_mutex);
 922
 923         RETURN(rc == -EALREADY ? 0 : rc);
 924 }
 925
 926 static void do_osd_scrub_stop(struct osd_scrub *scrub)
 927 {
 928         struct ptlrpc_thread *thread = &scrub->os_thread;
 929         struct l_wait_info    lwi    = { 0 };
 930
 931         /* os_lock: sync status between stop and scrub thread */
 932         cfs_spin_lock(&scrub->os_lock);
 933         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
 934                 thread_set_flags(thread, SVC_STOPPING);
 935                 cfs_spin_unlock(&scrub->os_lock);
 936                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
 937                 l_wait_event(thread->t_ctl_waitq,
 938                              thread_is_stopped(thread),
 939                              &lwi);
 940                 /* Do not skip the last lock/unlock, which can guarantee that
 941                  * the caller cannot return until the OI scrub thread exit. */
 942                 cfs_spin_lock(&scrub->os_lock);
 943         }
 944         cfs_spin_unlock(&scrub->os_lock);
 945 }
 946
 947 static void osd_scrub_stop(struct osd_device *dev)
 948 {
 949         /* od_otable_mutex: prevent curcurrent start/stop */
 950         cfs_mutex_lock(&dev->od_otable_mutex);
 951         do_osd_scrub_stop(&dev->od_scrub);
 952         cfs_mutex_unlock(&dev->od_otable_mutex);
 953 }
 954
 955 static const char osd_scrub_name[] = "OI_scrub";
 956
 957 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
 958 {
 959         struct osd_thread_info     *info   = osd_oti_get(env);
 960         struct osd_scrub           *scrub  = &dev->od_scrub;
 961         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
 962         struct scrub_file          *sf     = &scrub->os_file;
 963         struct osd_inode_id        *id     = &scrub->os_oic.oic_lid;
 964         struct super_block         *sb     = osd_sb(dev);
 965         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
 966         struct inode               *inode;
 967         struct lvfs_run_ctxt        saved;
 968         struct file                *filp;
 969         int                         dirty  = 0;
 970         int                         init   = 0;
 971         int                         rc     = 0;
 972         ENTRY;
 973
 974         OBD_SET_CTXT_MAGIC(ctxt);
 975         ctxt->pwdmnt = dev->od_mnt;
 976         ctxt->pwd = dev->od_mnt->mnt_root;
 977         ctxt->fs = get_ds();
 978
 979         cfs_waitq_init(&scrub->os_thread.t_ctl_waitq);
 980         cfs_init_rwsem(&scrub->os_rwsem);
 981         cfs_spin_lock_init(&scrub->os_lock);
 982         CFS_INIT_LIST_HEAD(&scrub->os_inconsistent_items);
 983         if (get_mount_flags(dev->od_mount->lmi_sb) & LMD_FLG_NOSCRUB)
 984                 scrub->os_no_scrub = 1;
 985
 986         push_ctxt(&saved, ctxt, NULL);
 987         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
 988         if (IS_ERR(filp))
 989                 RETURN(PTR_ERR(filp));
 990
 991         scrub->os_inode = igrab(filp->f_dentry->d_inode);
 992         filp_close(filp, 0);
 993         pop_ctxt(&saved, ctxt, NULL);
 994
 995         rc = osd_scrub_file_load(scrub);
 996         if (rc == -ENOENT) {
 997                 osd_scrub_file_init(scrub, es->s_uuid);
 998                 dirty = 1;
 999                 init = 1;
1000         } else if (rc != 0) {
1001                 RETURN(rc);
1002         } else {
1003                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
1004                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
1005                         dirty = 1;
1006                 } else if (sf->sf_status == SS_SCANNING) {
1007                         sf->sf_status = SS_CRASHED;
1008                         dirty = 1;
1009                 }
1010         }
1011
1012         if (sf->sf_pos_last_checkpoint != 0)
1013                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
1014         else
1015                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb);
1016
1017         if (dirty != 0) {
1018                 rc = osd_scrub_file_store(scrub);
1019                 if (rc != 0)
1020                         RETURN(rc);
1021         }
1022
1023         /* Initialize OI files. */
1024         rc = osd_oi_init(info, dev);
1025         if (rc < 0)
1026                 RETURN(rc);
1027
1028         if (init != 0) {
1029                 rc = __osd_oi_lookup(info, dev, &LU_DOT_LUSTRE_FID, id);
1030                 if (rc == 0) {
1031                         inode = osd_iget(info, dev, id);
1032                         if (IS_ERR(inode)) {
1033                                 rc = PTR_ERR(inode);
1034                                 /* It is restored from old 2.x backup. */
1035                                 if (rc == -ENOENT || rc == -ESTALE) {
1036                                         osd_scrub_file_reset(scrub, es->s_uuid,
1037                                                              SF_INCONSISTENT);
1038                                         rc = osd_scrub_file_store(scrub);
1039                                 }
1040                         } else {
1041                                 iput(inode);
1042                         }
1043                 } else if (rc == -ENOENT) {
1044                         rc = 0;
1045                 }
1046         }
1047
1048         if (rc == 0 && !scrub->os_no_scrub &&
1049             ((sf->sf_status == SS_CRASHED &&
1050               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_AUTO)) ||
1051              (sf->sf_status == SS_INIT &&
1052               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT))))
1053                 rc = osd_scrub_start(dev);
1054
1055         RETURN(rc);
1056 }
1057
1058 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
1059 {
1060         struct osd_scrub *scrub = &dev->od_scrub;
1061
1062         LASSERT(dev->od_otable_it == NULL);
1063
1064         if (scrub->os_inode != NULL) {
1065                 osd_scrub_stop(dev);
1066                 iput(scrub->os_inode);
1067                 scrub->os_inode = NULL;
1068         }
1069         if (dev->od_oi_table != NULL)
1070                 osd_oi_fini(osd_oti_get(env), dev);
1071 }
1072
1073 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
1074                                        struct dt_object *dt, __u32 attr,
1075                                        struct lustre_capa *capa)
1076 {
1077         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
1078         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
1079         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
1080         struct osd_scrub       *scrub = &dev->od_scrub;
1081         struct osd_otable_it   *it;
1082         __u32                   start = 0;
1083         int                     rc;
1084         ENTRY;
1085
1086         /* od_otable_mutex: prevent curcurrent init/fini */
1087         cfs_mutex_lock(&dev->od_otable_mutex);
1088         if (dev->od_otable_it != NULL)
1089                 GOTO(out, it = ERR_PTR(-EALREADY));
1090
1091         OBD_ALLOC_PTR(it);
1092         if (it == NULL)
1093                 GOTO(out, it = ERR_PTR(-ENOMEM));
1094
1095         dev->od_otable_it = it;
1096         it->ooi_dev = dev;
1097         it->ooi_cache.ooc_consumer_idx = -1;
1098         if (flags & DOIF_OUTUSED)
1099                 it->ooi_used_outside = 1;
1100
1101         if (flags & DOIF_RESET)
1102                 start |= SS_RESET;
1103
1104         if (valid & DOIV_ERROR_HANDLE) {
1105                 if (flags & DOIF_FAILOUT)
1106                         start |= SS_SET_FAILOUT;
1107                 else
1108                         start |= SS_CLEAR_FAILOUT;
1109         }
1110
1111         rc = do_osd_scrub_start(dev, start);
1112         if (rc == -EALREADY) {
1113                 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current - 1;
1114         } else if (rc < 0) {
1115                 dev->od_otable_it = NULL;
1116                 OBD_FREE_PTR(it);
1117                 GOTO(out, it = ERR_PTR(-EALREADY));
1118         } else {
1119                 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
1120         }
1121
1122         GOTO(out, it);
1123
1124 out:
1125         cfs_mutex_unlock(&dev->od_otable_mutex);
1126         return (struct dt_it *)it;
1127 }
1128
1129 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
1130 {
1131         struct osd_otable_it *it  = (struct osd_otable_it *)di;
1132         struct osd_device    *dev = it->ooi_dev;
1133
1134         /* od_otable_mutex: prevent curcurrent init/fini */
1135         cfs_mutex_lock(&dev->od_otable_mutex);
1136         do_osd_scrub_stop(&dev->od_scrub);
1137         LASSERT(dev->od_otable_it == it);
1138
1139         dev->od_otable_it = NULL;
1140         cfs_mutex_unlock(&dev->od_otable_mutex);
1141         OBD_FREE_PTR(it);
1142 }
1143
1144 /**
1145  * Set the OSD layer iteration start position as the specified key.
1146  *
1147  * The LFSCK out of OSD layer does not know the detail of the key, so if there
1148  * are several keys, they cannot be compared out of OSD, so call "::get()" for
1149  * each key, and OSD will select the smallest one by itself.
1150  */
1151 static int osd_otable_it_get(const struct lu_env *env,
1152                              struct dt_it *di, const struct dt_key *key)
1153 {
1154         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
1155         struct osd_otable_cache *ooc = &it->ooi_cache;
1156         const char              *str = (const char *)key;
1157         __u32                    ino;
1158         ENTRY;
1159
1160         /* Forbid to set iteration position after iteration started. */
1161         if (it->ooi_user_ready)
1162                 RETURN(-EPERM);
1163
1164         if (str[0] == '\0')
1165                 RETURN(-EINVAL);
1166
1167         if (sscanf(str, "%u", &ino) <= 0)
1168                 RETURN(-EINVAL);
1169
1170         /* Skip the one that has been processed last time. */
1171         if (ooc->ooc_pos_preload > ++ino)
1172                 ooc->ooc_pos_preload = ino;
1173
1174         RETURN(0);
1175 }
1176
1177 static int osd_otable_it_preload(const struct lu_env *env,
1178                                  struct osd_otable_it *it)
1179 {
1180         struct osd_device       *dev   = it->ooi_dev;
1181         struct osd_scrub        *scrub = &dev->od_scrub;
1182         struct osd_otable_cache *ooc   = &it->ooi_cache;
1183         int                      rc;
1184         ENTRY;
1185
1186         rc = osd_inode_iteration(osd_oti_get(env), dev,
1187                                  OSD_OTABLE_IT_CACHE_SIZE, 1);
1188         if (rc == SCRUB_IT_ALL)
1189                 it->ooi_all_cached = 1;
1190
1191         CDEBUG(D_LFSCK, "OSD pre-loaded: max = %u, preload = %u, rc = %d\n",
1192                le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count),
1193                ooc->ooc_pos_preload, rc);
1194
1195         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1196                 scrub->os_waiting = 0;
1197                 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
1198         }
1199
1200         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1201 }
1202
1203 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
1204 {
1205         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
1206         struct osd_device       *dev    = it->ooi_dev;
1207         struct osd_scrub        *scrub  = &dev->od_scrub;
1208         struct osd_otable_cache *ooc    = &it->ooi_cache;
1209         struct ptlrpc_thread    *thread = &scrub->os_thread;
1210         struct l_wait_info       lwi    = { 0 };
1211         int                      rc;
1212         ENTRY;
1213
1214         LASSERT(it->ooi_user_ready);
1215
1216 again:
1217         if (!thread_is_running(thread) && !it->ooi_used_outside)
1218                 RETURN(1);
1219
1220         if (ooc->ooc_cached_items > 0) {
1221                 ooc->ooc_cached_items--;
1222                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
1223                                         ~OSD_OTABLE_IT_CACHE_MASK;
1224                 RETURN(0);
1225         }
1226
1227         if (it->ooi_all_cached) {
1228                 l_wait_event(thread->t_ctl_waitq,
1229                              !thread_is_running(thread),
1230                              &lwi);
1231                 RETURN(1);
1232         }
1233
1234         it->ooi_waiting = 1;
1235         l_wait_event(thread->t_ctl_waitq,
1236                      ooc->ooc_pos_preload < scrub->os_pos_current ||
1237                      !thread_is_running(thread),
1238                      &lwi);
1239         it->ooi_waiting = 0;
1240
1241         if (!thread_is_running(thread) && !it->ooi_used_outside)
1242                 RETURN(1);
1243
1244         rc = osd_otable_it_preload(env, it);
1245         if (rc >= 0)
1246                 goto again;
1247
1248         RETURN(rc);
1249 }
1250
1251 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
1252                                         const struct dt_it *di)
1253 {
1254         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
1255         struct osd_otable_cache *ooc = &it->ooi_cache;
1256
1257         sprintf(it->ooi_key, "%u",
1258                 ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino);
1259         return (struct dt_key *)it->ooi_key;
1260 }
1261
1262 static int osd_otable_it_key_size(const struct lu_env *env,
1263                                   const struct dt_it *di)
1264 {
1265         return sizeof(((struct osd_otable_it *)di)->ooi_key);
1266 }
1267
1268 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
1269                              struct dt_rec *rec, __u32 attr)
1270 {
1271         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
1272         struct osd_otable_cache *ooc = &it->ooi_cache;
1273
1274         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
1275         return 0;
1276 }
1277
1278 static int osd_otable_it_load(const struct lu_env *env,
1279                               const struct dt_it *di, __u64 hash)
1280 {
1281         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
1282         struct osd_device       *dev   = it->ooi_dev;
1283         struct osd_otable_cache *ooc   = &it->ooi_cache;
1284         struct osd_scrub        *scrub = &dev->od_scrub;
1285
1286         if (it->ooi_user_ready)
1287                 return 0;
1288
1289         if (ooc->ooc_pos_preload < LDISKFS_FIRST_INO(osd_sb(dev)))
1290                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev));
1291         it->ooi_user_ready = 1;
1292         if (!scrub->os_full_speed)
1293                 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
1294
1295         /* Unplug OSD layer iteration by the first next() call. */
1296         return osd_otable_it_next(env, (struct dt_it *)it);
1297 }
1298
1299 const struct dt_index_operations osd_otable_ops = {
1300         .dio_it = {
1301                 .init     = osd_otable_it_init,
1302                 .fini     = osd_otable_it_fini,
1303                 .get      = osd_otable_it_get,
1304                 .next     = osd_otable_it_next,
1305                 .key      = osd_otable_it_key,
1306                 .key_size = osd_otable_it_key_size,
1307                 .rec      = osd_otable_it_rec,
1308                 .load     = osd_otable_it_load,
1309         }
1310 };
1311
1312 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
1313                    int insert)
1314 {
1315         struct osd_inconsistent_item *oii;
1316         struct osd_scrub             *scrub  = &dev->od_scrub;
1317         struct ptlrpc_thread         *thread = &scrub->os_thread;
1318         int                           wakeup = 0;
1319         ENTRY;
1320
1321         OBD_ALLOC_PTR(oii);
1322         if (unlikely(oii == NULL))
1323                 RETURN(-ENOMEM);
1324
1325         CFS_INIT_LIST_HEAD(&oii->oii_list);
1326         oii->oii_cache = *oic;
1327         oii->oii_insert = insert;
1328
1329         cfs_spin_lock(&scrub->os_lock);
1330         if (unlikely(!thread_is_running(thread))) {
1331                 cfs_spin_unlock(&scrub->os_lock);
1332                 OBD_FREE_PTR(oii);
1333                 RETURN(-EAGAIN);
1334         }
1335
1336         if (cfs_list_empty(&scrub->os_inconsistent_items))
1337                 wakeup = 1;
1338         cfs_list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
1339         cfs_spin_unlock(&scrub->os_lock);
1340
1341         if (wakeup != 0)
1342                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
1343
1344         RETURN(0);
1345 }
1346
1347 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
1348                    struct osd_inode_id *id)
1349 {
1350         struct osd_scrub             *scrub = &dev->od_scrub;
1351         struct osd_inconsistent_item *oii;
1352         ENTRY;
1353
1354         cfs_spin_lock(&scrub->os_lock);
1355         cfs_list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
1356                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
1357                         *id = oii->oii_cache.oic_lid;
1358                         cfs_spin_unlock(&scrub->os_lock);
1359                         RETURN(0);
1360                 }
1361         }
1362         cfs_spin_unlock(&scrub->os_lock);
1363
1364         RETURN(-ENOENT);
1365 }