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 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         sf->sf_items_noscrub = 0;
 161         sf->sf_items_igif = 0;
 162 }
 163
 164 static int osd_scrub_file_load(struct osd_scrub *scrub)
 165 {
 166         loff_t  pos  = 0;
 167         char   *name = LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
 168         int     len  = sizeof(scrub->os_file_disk);
 169         int     rc;
 170
 171         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
 172         if (rc == len) {
 173                 struct scrub_file *sf = &scrub->os_file;
 174
 175                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
 176                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
 177                         CWARN("%.16s: invalid scrub magic 0x%x != 0x%x\n,",
 178                               name, sf->sf_magic, SCRUB_MAGIC_V1);
 179                         /* Process it as new scrub file. */
 180                         rc = -ENOENT;
 181                 } else {
 182                         rc = 0;
 183                 }
 184         } else if (rc != 0) {
 185                 CERROR("%.16s: fail to load scrub file, expected = %d, "
 186                        "rc = %d\n", name, len, rc);
 187                 if (rc > 0)
 188                         rc = -EFAULT;
 189         } else {
 190                 /* return -ENOENT for empty scrub file case. */
 191                 rc = -ENOENT;
 192         }
 193
 194         return rc;
 195 }
 196
 197 int osd_scrub_file_store(struct osd_scrub *scrub)
 198 {
 199         struct osd_device *dev;
 200         handle_t          *jh;
 201         loff_t             pos     = 0;
 202         int                len     = sizeof(scrub->os_file_disk);
 203         int                credits;
 204         int                rc;
 205
 206         dev = container_of0(scrub, struct osd_device, od_scrub);
 207         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
 208                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
 209         jh = ldiskfs_journal_start_sb(osd_sb(dev), credits);
 210         if (IS_ERR(jh)) {
 211                 rc = PTR_ERR(jh);
 212                 CERROR("%.16s: fail to start trans for scrub store, rc = %d\n",
 213                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,rc);
 214                 return rc;
 215         }
 216
 217         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
 218         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
 219                                       len, 0, &pos, jh);
 220         ldiskfs_journal_stop(jh);
 221         if (rc != 0)
 222                 CERROR("%.16s: fail to store scrub file, expected = %d, "
 223                        "rc = %d\n",
 224                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
 225                        len, rc);
 226         scrub->os_time_last_checkpoint = cfs_time_current();
 227         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
 228                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
 229         return rc;
 230 }
 231
 232 static int osd_scrub_prep(struct osd_device *dev)
 233 {
 234         struct osd_scrub     *scrub  = &dev->od_scrub;
 235         struct ptlrpc_thread *thread = &scrub->os_thread;
 236         struct scrub_file    *sf     = &scrub->os_file;
 237         __u32                 flags  = scrub->os_start_flags;
 238         int                   rc;
 239         ENTRY;
 240
 241         cfs_down_write(&scrub->os_rwsem);
 242         if (flags & SS_SET_FAILOUT)
 243                 sf->sf_param |= SP_FAILOUT;
 244
 245         if (flags & SS_CLEAR_FAILOUT)
 246                 sf->sf_param &= ~SP_FAILOUT;
 247
 248         if (flags & SS_RESET)
 249                 osd_scrub_file_reset(scrub,
 250                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, sf->sf_flags);
 251
 252         if (flags & SS_AUTO) {
 253                 scrub->os_full_speed = 1;
 254                 sf->sf_flags |= SF_AUTO;
 255         } else {
 256                 scrub->os_full_speed = 0;
 257         }
 258
 259         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT))
 260                 scrub->os_full_speed = 1;
 261
 262         scrub->os_in_prior = 0;
 263         scrub->os_waiting = 0;
 264         scrub->os_new_checked = 0;
 265         if (sf->sf_pos_last_checkpoint != 0)
 266                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 267         else
 268                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev));
 269
 270         scrub->os_pos_current = sf->sf_pos_latest_start;
 271         sf->sf_status = SS_SCANNING;
 272         sf->sf_time_latest_start = cfs_time_current_sec();
 273         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 274         rc = osd_scrub_file_store(scrub);
 275         if (rc == 0) {
 276                 cfs_spin_lock(&scrub->os_lock);
 277                 thread_set_flags(thread, SVC_RUNNING);
 278                 cfs_spin_unlock(&scrub->os_lock);
 279                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
 280         }
 281         cfs_up_write(&scrub->os_rwsem);
 282
 283         RETURN(rc);
 284 }
 285
 286 static int
 287 osd_scrub_error(struct osd_device *dev, struct osd_inode_id *lid, int rc)
 288 {
 289         struct osd_scrub  *scrub = &dev->od_scrub;
 290         struct scrub_file *sf    = &scrub->os_file;
 291
 292         cfs_down_write(&scrub->os_rwsem);
 293         scrub->os_new_checked++;
 294         sf->sf_items_failed++;
 295         if (sf->sf_pos_first_inconsistent == 0 ||
 296             sf->sf_pos_first_inconsistent > lid->oii_ino)
 297                 sf->sf_pos_first_inconsistent = lid->oii_ino;
 298         cfs_up_write(&scrub->os_rwsem);
 299         return sf->sf_param & SP_FAILOUT ? rc : 0;
 300 }
 301
 302 static int
 303 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
 304                        struct osd_idmap_cache *oic)
 305 {
 306         struct osd_scrub             *scrub  = &dev->od_scrub;
 307         struct scrub_file            *sf     = &scrub->os_file;
 308         struct osd_inode_id          *lid2   = &info->oti_id;
 309         struct lu_fid                *oi_fid = &info->oti_fid;
 310         struct osd_inode_id          *oi_id  = &info->oti_id2;
 311         handle_t                     *jh     = NULL;
 312         struct osd_inconsistent_item *oii    = NULL;
 313         struct inode                 *inode  = NULL;
 314         struct lu_fid                *fid    = &oic->oic_fid;
 315         struct osd_inode_id          *lid    = &oic->oic_lid;
 316         struct iam_container         *bag;
 317         struct iam_path_descr        *ipd;
 318         int                           ops    = DTO_INDEX_UPDATE;
 319         int                           idx;
 320         int                           rc;
 321         ENTRY;
 322
 323         if (scrub->os_in_prior)
 324                 oii = cfs_list_entry(oic, struct osd_inconsistent_item,
 325                                      oii_cache);
 326
 327         cfs_down_write(&scrub->os_rwsem);
 328         scrub->os_new_checked++;
 329         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
 330                 GOTO(out, rc = 0);
 331
 332         if (oii != NULL && oii->oii_insert)
 333                 goto iget;
 334
 335         rc = osd_oi_lookup(info, dev, fid, lid2);
 336         if (rc != 0) {
 337                 if (rc != -ENOENT)
 338                         GOTO(out, rc);
 339
 340 iget:
 341                 inode = osd_iget(info, dev, lid);
 342                 if (IS_ERR(inode)) {
 343                         rc = PTR_ERR(inode);
 344                         /* Someone removed the inode. */
 345                         if (rc == -ENOENT || rc == -ESTALE)
 346                                 rc = 0;
 347                         GOTO(out, rc);
 348                 }
 349
 350                 /* Prevent the inode to be unlinked during OI scrub. */
 351                 cfs_mutex_lock(&inode->i_mutex);
 352                 if (unlikely(inode->i_nlink == 0)) {
 353                         cfs_mutex_unlock(&inode->i_mutex);
 354                         iput(inode);
 355                         GOTO(out, rc = 0);
 356                 }
 357
 358                 ops = DTO_INDEX_INSERT;
 359                 idx = osd_oi_fid2idx(dev, fid);
 360                 if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 361                         ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 362                 sf->sf_flags |= SF_RECREATED;
 363         } else if (osd_id_eq(lid, lid2)) {
 364                         GOTO(out, rc = 0);
 365         }
 366
 367         sf->sf_flags |= SF_INCONSISTENT;
 368         fid_cpu_to_be(oi_fid, fid);
 369         osd_id_pack(oi_id, &oic->oic_lid);
 370         jh = ldiskfs_journal_start_sb(osd_sb(dev),
 371                                 osd_dto_credits_noquota[ops]);
 372         if (IS_ERR(jh)) {
 373                 rc = PTR_ERR(jh);
 374                 CERROR("%.16s: fail to start trans for scrub store, rc = %d\n",
 375                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
 376                 GOTO(out, rc);
 377         }
 378
 379         bag = &osd_fid2oi(dev, fid)->oi_dir.od_container;
 380         ipd = osd_idx_ipd_get(info->oti_env, bag);
 381         if (unlikely(ipd == NULL)) {
 382                 ldiskfs_journal_stop(jh);
 383                 CERROR("%.16s: fail to get ipd for scrub store\n",
 384                         LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name);
 385                 GOTO(out, rc = -ENOMEM);
 386         }
 387
 388         if (ops == DTO_INDEX_UPDATE) {
 389                 rc = iam_update(jh, bag, (const struct iam_key *)oi_fid,
 390                                 (struct iam_rec *)oi_id, ipd);
 391         } else {
 392                 rc = iam_insert(jh, bag, (const struct iam_key *)oi_fid,
 393                                 (struct iam_rec *)oi_id, ipd);
 394                 if (rc == -EEXIST)
 395                         rc = 1;
 396         }
 397         osd_ipd_put(info->oti_env, bag, ipd);
 398         ldiskfs_journal_stop(jh);
 399         if (rc == 0) {
 400                 if (scrub->os_in_prior)
 401                         sf->sf_items_updated_prior++;
 402                 else
 403                         sf->sf_items_updated++;
 404         }
 405
 406         GOTO(out, rc);
 407
 408 out:
 409         if (rc < 0) {
 410                 sf->sf_items_failed++;
 411                 if (sf->sf_pos_first_inconsistent == 0 ||
 412                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 413                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 414         } else {
 415                 rc = 0;
 416         }
 417
 418         if (ops == DTO_INDEX_INSERT) {
 419                 cfs_mutex_unlock(&inode->i_mutex);
 420                 iput(inode);
 421         }
 422         cfs_up_write(&scrub->os_rwsem);
 423
 424         if (oii != NULL) {
 425                 LASSERT(!cfs_list_empty(&oii->oii_list));
 426
 427                 cfs_spin_lock(&scrub->os_lock);
 428                 cfs_list_del_init(&oii->oii_list);
 429                 cfs_spin_unlock(&scrub->os_lock);
 430                 OBD_FREE_PTR(oii);
 431         }
 432         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 433 }
 434
 435 static int do_osd_scrub_checkpoint(struct osd_scrub *scrub)
 436 {
 437         struct scrub_file *sf = &scrub->os_file;
 438         int                rc;
 439         ENTRY;
 440
 441         cfs_down_write(&scrub->os_rwsem);
 442         sf->sf_items_checked += scrub->os_new_checked;
 443         scrub->os_new_checked = 0;
 444         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 445         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 446         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 447                                             scrub->os_time_last_checkpoint);
 448         rc = osd_scrub_file_store(scrub);
 449         cfs_up_write(&scrub->os_rwsem);
 450
 451         RETURN(rc);
 452 }
 453
 454 static inline int osd_scrub_checkpoint(struct osd_scrub *scrub)
 455 {
 456         if (unlikely(cfs_time_beforeq(scrub->os_time_next_checkpoint,
 457                                       cfs_time_current()) &&
 458                      scrub->os_new_checked > 0))
 459                 return do_osd_scrub_checkpoint(scrub);
 460         return 0;
 461 }
 462
 463 static void osd_scrub_post(struct osd_scrub *scrub, int result)
 464 {
 465         struct scrub_file *sf = &scrub->os_file;
 466         ENTRY;
 467
 468         cfs_down_write(&scrub->os_rwsem);
 469         cfs_spin_lock(&scrub->os_lock);
 470         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 471         cfs_spin_unlock(&scrub->os_lock);
 472         if (scrub->os_new_checked > 0) {
 473                 sf->sf_items_checked += scrub->os_new_checked;
 474                 scrub->os_new_checked = 0;
 475                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 476         }
 477         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 478         if (result > 0) {
 479                 sf->sf_status = SS_COMPLETED;
 480                 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 481                 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT | SF_AUTO);
 482                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 483                 sf->sf_success_count++;
 484         } else if (result == 0) {
 485                 if (scrub->os_paused)
 486                         sf->sf_status = SS_PAUSED;
 487                 else
 488                         sf->sf_status = SS_STOPPED;
 489         } else {
 490                 sf->sf_status = SS_FAILED;
 491         }
 492         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 493                                             scrub->os_time_last_checkpoint);
 494         result = osd_scrub_file_store(scrub);
 495         if (result < 0)
 496                 CERROR("%.16s: fail to osd_scrub_post, rc = %d\n",
 497                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
 498                        result);
 499         cfs_up_write(&scrub->os_rwsem);
 500
 501         EXIT;
 502 }
 503
 504 #define SCRUB_NEXT_BREAK        1 /* exit current loop and process next group */
 505 #define SCRUB_NEXT_CONTINUE     2 /* skip current object and process next bit */
 506 #define SCRUB_NEXT_EXIT         3 /* exit all the loops */
 507 #define SCRUB_NEXT_WAIT         4 /* wait for free cache slot */
 508 #define SCRUB_NEXT_CRASH        5 /* simulate system crash during OI scrub */
 509 #define SCRUB_NEXT_FATAL        6 /* simulate failure during OI scrub */
 510 #define SCRUB_NEXT_NOSCRUB      7 /* new created object, no scrub on it */
 511 #define SCRUB_NEXT_IGIF         8 /* IGIF object */
 512
 513 struct osd_iit_param {
 514         struct super_block *sb;
 515         struct buffer_head *bitmap;
 516         ldiskfs_group_t bg;
 517         __u32 gbase;
 518         __u32 offset;
 519 };
 520
 521 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 522                                    struct osd_device *dev,
 523                                    struct osd_iit_param *param,
 524                                    struct osd_idmap_cache **oic,
 525                                    int noslot);
 526
 527 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 528                                    struct osd_device *dev,
 529                                    struct osd_iit_param *param,
 530                                    struct osd_idmap_cache *oic,
 531                                    int *noslot, int rc);
 532
 533 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
 534                                        struct osd_otable_cache *ooc)
 535 {
 536         return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
 537 }
 538
 539 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 540 {
 541         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 542                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 543         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 544                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 545                 return SCRUB_NEXT_BREAK;
 546         } else {
 547                 *pos = param->gbase + param->offset;
 548                 return 0;
 549         }
 550 }
 551
 552 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 553                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 554                         struct super_block *sb, struct inode **pinode)
 555 {
 556         struct inode *inode;
 557         int           rc;
 558
 559         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 560         inode = osd_iget_fid(info, dev, lid, fid);
 561         if (IS_ERR(inode)) {
 562                 rc = PTR_ERR(inode);
 563                 /* The inode may be removed after bitmap searching, or the
 564                  * file is new created without inode initialized yet. */
 565                 if (rc == -ENOENT || rc == -ESTALE)
 566                         return SCRUB_NEXT_CONTINUE;
 567
 568                 CERROR("%.16s: fail to read inode, ino# = %u, rc = %d\n",
 569                        LDISKFS_SB(sb)->s_es->s_volume_name, pos, rc);
 570                 return rc;
 571         }
 572
 573         *pinode = inode;
 574         return 0;
 575 }
 576
 577 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 578                           struct osd_iit_param *param,
 579                           struct osd_idmap_cache **oic, int noslot)
 580 {
 581         struct osd_scrub     *scrub  = &dev->od_scrub;
 582         struct ptlrpc_thread *thread = &scrub->os_thread;
 583         struct lu_fid        *fid;
 584         struct osd_inode_id  *lid;
 585         struct inode         *inode;
 586         int                   rc;
 587
 588         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
 589                 struct l_wait_info lwi;
 590
 591                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
 592                 l_wait_event(thread->t_ctl_waitq,
 593                              !cfs_list_empty(&scrub->os_inconsistent_items) ||
 594                              !thread_is_running(thread),
 595                              &lwi);
 596         }
 597
 598         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 599                 cfs_spin_lock(&scrub->os_lock);
 600                 thread_set_flags(thread, SVC_STOPPING);
 601                 cfs_spin_unlock(&scrub->os_lock);
 602                 return SCRUB_NEXT_CRASH;
 603         }
 604
 605         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
 606                 return SCRUB_NEXT_FATAL;
 607
 608         if (unlikely(!thread_is_running(thread)))
 609                 return SCRUB_NEXT_EXIT;
 610
 611         if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
 612                 struct osd_inconsistent_item *oii;
 613
 614                 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
 615                                      struct osd_inconsistent_item, oii_list);
 616                 *oic = &oii->oii_cache;
 617                 scrub->os_in_prior = 1;
 618                 return 0;
 619         }
 620
 621         if (noslot != 0)
 622                 return SCRUB_NEXT_WAIT;
 623
 624         rc = osd_iit_next(param, &scrub->os_pos_current);
 625         if (rc != 0)
 626                 return rc;
 627
 628         *oic = &scrub->os_oic;
 629         fid = &(*oic)->oic_fid;
 630         lid = &(*oic)->oic_lid;
 631         rc = osd_iit_iget(info, dev, fid, lid,
 632                           scrub->os_pos_current, param->sb, &inode);
 633         if (rc != 0)
 634                 return rc;
 635
 636         if (inode->i_state & I_LUSTRE_NOSCRUB) {
 637                 /* Only skip it for the first OI scrub accessing. */
 638                 inode->i_state &= ~I_LUSTRE_NOSCRUB;
 639                 rc = SCRUB_NEXT_NOSCRUB;
 640         } else if (!fid_is_norm(fid)) {
 641                 rc = SCRUB_NEXT_IGIF;
 642         }
 643
 644         iput(inode);
 645         return rc;
 646 }
 647
 648 static int osd_preload_next(struct osd_thread_info *info,
 649                             struct osd_device *dev, struct osd_iit_param *param,
 650                             struct osd_idmap_cache **oic, int noslot)
 651 {
 652         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
 653         struct osd_scrub        *scrub;
 654         struct ptlrpc_thread    *thread;
 655         struct inode            *inode;
 656         int                      rc;
 657
 658         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
 659         if (rc != 0)
 660                 return rc;
 661
 662         scrub = &dev->od_scrub;
 663         thread = &scrub->os_thread;
 664         if (thread_is_running(thread) &&
 665             ooc->ooc_pos_preload >= scrub->os_pos_current)
 666                 return SCRUB_NEXT_EXIT;
 667
 668         rc = osd_iit_iget(info, dev,
 669                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
 670                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
 671                           ooc->ooc_pos_preload, param->sb, &inode);
 672         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
 673          * ignore the failure, so it still need to skip the inode next time. */
 674         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
 675         if (rc == 0)
 676                 iput(inode);
 677         return rc;
 678 }
 679
 680 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
 681                           struct osd_iit_param *param,
 682                           struct osd_idmap_cache *oic, int *noslot, int rc)
 683 {
 684         struct l_wait_info       lwi    = { 0 };
 685         struct osd_scrub        *scrub  = &dev->od_scrub;
 686         struct scrub_file       *sf     = &scrub->os_file;
 687         __u64                   *items  = NULL;
 688         struct ptlrpc_thread    *thread = &scrub->os_thread;
 689         struct osd_otable_it    *it     = dev->od_otable_it;
 690         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
 691
 692         switch (rc) {
 693         case SCRUB_NEXT_CONTINUE:
 694                 goto next;
 695         case SCRUB_NEXT_WAIT:
 696                 goto wait;
 697         case SCRUB_NEXT_NOSCRUB:
 698                 items = &sf->sf_items_noscrub;
 699                 break;
 700         case SCRUB_NEXT_IGIF:
 701                 items = &sf->sf_items_igif;
 702                 break;
 703         }
 704
 705         if (items != NULL) {
 706                 cfs_down_write(&scrub->os_rwsem);
 707                 scrub->os_new_checked++;
 708                 (*items)++;
 709                 cfs_up_write(&scrub->os_rwsem);
 710                 goto next;
 711         }
 712
 713         LASSERTF(rc <= 0, "unexpected rc = %d\n", rc);
 714
 715         if (rc != 0)
 716                 rc = osd_scrub_error(dev, &oic->oic_lid, rc);
 717         else
 718                 rc = osd_scrub_check_update(info, dev, oic);
 719         if (rc != 0)
 720                 return rc;
 721
 722         rc = osd_scrub_checkpoint(scrub);
 723         if (rc != 0) {
 724                 CERROR("%.16s: fail to checkpoint, pos = %u, rc = %d\n",
 725                        LDISKFS_SB(param->sb)->s_es->s_volume_name,
 726                        scrub->os_pos_current, rc);
 727                 /* Continue, as long as the scrub itself can go ahead. */
 728         }
 729
 730         if (scrub->os_in_prior) {
 731                 scrub->os_in_prior = 0;
 732                 return 0;
 733         }
 734
 735 next:
 736         scrub->os_pos_current = param->gbase + ++(param->offset);
 737         if (it != NULL && it->ooi_waiting &&
 738             ooc->ooc_pos_preload < scrub->os_pos_current) {
 739                 it->ooi_waiting = 0;
 740                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
 741         }
 742
 743         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
 744                 return 0;
 745
 746 wait:
 747         if (osd_scrub_has_window(scrub, ooc)) {
 748                 *noslot = 0;
 749                 return 0;
 750         }
 751
 752         scrub->os_waiting = 1;
 753         l_wait_event(thread->t_ctl_waitq,
 754                      osd_scrub_has_window(scrub, ooc) ||
 755                      !cfs_list_empty(&scrub->os_inconsistent_items) ||
 756                      !thread_is_running(thread),
 757                      &lwi);
 758         scrub->os_waiting = 0;
 759
 760         if (osd_scrub_has_window(scrub, ooc))
 761                 *noslot = 0;
 762         else
 763                 *noslot = 1;
 764         return 0;
 765 }
 766
 767 static int osd_preload_exec(struct osd_thread_info *info,
 768                             struct osd_device *dev, struct osd_iit_param *param,
 769                             struct osd_idmap_cache *oic, int *noslot, int rc)
 770 {
 771         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
 772
 773         if (rc == 0) {
 774                 ooc->ooc_cached_items++;
 775                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
 776                                         ~OSD_OTABLE_IT_CACHE_MASK;
 777         }
 778         return rc > 0 ? 0 : rc;
 779 }
 780
 781 #define SCRUB_IT_ALL    1
 782 #define SCRUB_IT_CRASH  2
 783
 784 static int osd_inode_iteration(struct osd_thread_info *info,
 785                                struct osd_device *dev, __u32 max, int preload)
 786 {
 787         osd_iit_next_policy   next;
 788         osd_iit_exec_policy   exec;
 789         __u32                *pos;
 790         __u32                *count;
 791         struct osd_iit_param  param;
 792         __u32                 limit;
 793         int                   noslot = 0;
 794         int                   rc;
 795         ENTRY;
 796
 797         if (preload == 0) {
 798                 struct osd_scrub *scrub = &dev->od_scrub;
 799
 800                 next = osd_scrub_next;
 801                 exec = osd_scrub_exec;
 802                 pos = &scrub->os_pos_current;
 803                 count = &scrub->os_new_checked;
 804         } else {
 805                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
 806
 807                 next = osd_preload_next;
 808                 exec = osd_preload_exec;
 809                 pos = &ooc->ooc_pos_preload;
 810                 count = &ooc->ooc_cached_items;
 811         }
 812         param.sb = osd_sb(dev);
 813         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
 814
 815         while (*pos <= limit && *count < max) {
 816                 struct osd_idmap_cache *oic = NULL;
 817
 818                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
 819                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
 820                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
 821                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
 822                 if (param.bitmap == NULL) {
 823                         CERROR("%.16s: fail to read bitmap for %u, "
 824                                "scrub will stop, urgent mode\n",
 825                                LDISKFS_SB(param.sb)->s_es->s_volume_name,
 826                                (__u32)param.bg);
 827                         RETURN(-EIO);
 828                 }
 829
 830                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
 831                        *count < max) {
 832                         rc = next(info, dev, &param, &oic, noslot);
 833                         switch (rc) {
 834                         case SCRUB_NEXT_BREAK:
 835                                 goto next_group;
 836                         case SCRUB_NEXT_EXIT:
 837                                 brelse(param.bitmap);
 838                                 RETURN(0);
 839                         case SCRUB_NEXT_CRASH:
 840                                 brelse(param.bitmap);
 841                                 RETURN(SCRUB_IT_CRASH);
 842                         case SCRUB_NEXT_FATAL:
 843                                 brelse(param.bitmap);
 844                                 RETURN(-EINVAL);
 845                         }
 846
 847                         rc = exec(info, dev, &param, oic, &noslot, rc);
 848                         if (rc != 0) {
 849                                 brelse(param.bitmap);
 850                                 RETURN(rc);
 851                         }
 852                 }
 853
 854 next_group:
 855                 brelse(param.bitmap);
 856         }
 857
 858         if (*pos > limit)
 859                 RETURN(SCRUB_IT_ALL);
 860         RETURN(0);
 861 }
 862
 863 static int osd_scrub_main(void *args)
 864 {
 865         struct lu_env         env;
 866         struct osd_device    *dev    = (struct osd_device *)args;
 867         struct osd_scrub     *scrub  = &dev->od_scrub;
 868         struct ptlrpc_thread *thread = &scrub->os_thread;
 869         struct super_block   *sb     = osd_sb(dev);
 870         int                   rc;
 871         ENTRY;
 872
 873         cfs_daemonize("OI_scrub");
 874         rc = lu_env_init(&env, LCT_DT_THREAD);
 875         if (rc != 0) {
 876                 CERROR("%.16s: OI scrub, fail to init env, rc = %d\n",
 877                        LDISKFS_SB(sb)->s_es->s_volume_name, rc);
 878                 GOTO(noenv, rc);
 879         }
 880
 881         rc = osd_scrub_prep(dev);
 882         if (rc != 0) {
 883                 CERROR("%.16s: OI scrub, fail to scrub prep, rc = %d\n",
 884                        LDISKFS_SB(sb)->s_es->s_volume_name, rc);
 885                 GOTO(out, rc);
 886         }
 887
 888         if (!scrub->os_full_speed) {
 889                 struct l_wait_info lwi = { 0 };
 890                 struct osd_otable_it *it = dev->od_otable_it;
 891                 struct osd_otable_cache *ooc = &it->ooi_cache;
 892
 893                 l_wait_event(thread->t_ctl_waitq,
 894                              it->ooi_user_ready || !thread_is_running(thread),
 895                              &lwi);
 896                 if (unlikely(!thread_is_running(thread)))
 897                         GOTO(post, rc = 0);
 898
 899                 LASSERT(scrub->os_pos_current >= ooc->ooc_pos_preload);
 900                 scrub->os_pos_current = ooc->ooc_pos_preload;
 901         }
 902
 903         CDEBUG(D_LFSCK, "OI scrub: flags = 0x%x, pos = %u\n",
 904                scrub->os_start_flags, scrub->os_pos_current);
 905
 906         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, 0);
 907         if (unlikely(rc == SCRUB_IT_CRASH))
 908                 GOTO(out, rc = -EINVAL);
 909         GOTO(post, rc);
 910
 911 post:
 912         osd_scrub_post(scrub, rc);
 913         CDEBUG(D_LFSCK, "OI scrub: stop, rc = %d, pos = %u\n",
 914                rc, scrub->os_pos_current);
 915
 916 out:
 917         while (!cfs_list_empty(&scrub->os_inconsistent_items)) {
 918                 struct osd_inconsistent_item *oii;
 919
 920                 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
 921                                      struct osd_inconsistent_item, oii_list);
 922                 cfs_list_del_init(&oii->oii_list);
 923                 OBD_FREE_PTR(oii);
 924         }
 925         lu_env_fini(&env);
 926
 927 noenv:
 928         cfs_spin_lock(&scrub->os_lock);
 929         thread_set_flags(thread, SVC_STOPPED);
 930         cfs_waitq_broadcast(&thread->t_ctl_waitq);
 931         cfs_spin_unlock(&scrub->os_lock);
 932         return rc;
 933 }
 934
 935 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
 936 {
 937         struct osd_scrub     *scrub  = &dev->od_scrub;
 938         struct ptlrpc_thread *thread = &scrub->os_thread;
 939         struct l_wait_info    lwi    = { 0 };
 940         int                   rc;
 941         ENTRY;
 942
 943 again:
 944         /* os_lock: sync status between stop and scrub thread */
 945         cfs_spin_lock(&scrub->os_lock);
 946         if (thread_is_running(thread)) {
 947                 cfs_spin_unlock(&scrub->os_lock);
 948                 RETURN(-EALREADY);
 949         } else if (unlikely(thread_is_stopping(thread))) {
 950                 cfs_spin_unlock(&scrub->os_lock);
 951                 l_wait_event(thread->t_ctl_waitq,
 952                              thread_is_stopped(thread),
 953                              &lwi);
 954                 goto again;
 955         }
 956         cfs_spin_unlock(&scrub->os_lock);
 957
 958         scrub->os_start_flags = flags;
 959         thread_set_flags(thread, 0);
 960         rc = cfs_create_thread(osd_scrub_main, dev, 0);
 961         if (rc < 0) {
 962                 CERROR("%.16s: cannot start iteration thread, rc = %d\n",
 963                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
 964                 RETURN(rc);
 965         }
 966
 967         l_wait_event(thread->t_ctl_waitq,
 968                      thread_is_running(thread) || thread_is_stopped(thread),
 969                      &lwi);
 970
 971         RETURN(0);
 972 }
 973
 974 int osd_scrub_start(struct osd_device *dev)
 975 {
 976         __u32 flags = SS_AUTO;
 977         int   rc;
 978         ENTRY;
 979
 980         if (dev->od_scrub.os_file.sf_status == SS_COMPLETED)
 981                 flags |= SS_RESET;
 982
 983         /* od_otable_mutex: prevent curcurrent start/stop */
 984         cfs_mutex_lock(&dev->od_otable_mutex);
 985         rc = do_osd_scrub_start(dev, flags);
 986         cfs_mutex_unlock(&dev->od_otable_mutex);
 987
 988         RETURN(rc == -EALREADY ? 0 : rc);
 989 }
 990
 991 static void do_osd_scrub_stop(struct osd_scrub *scrub)
 992 {
 993         struct ptlrpc_thread *thread = &scrub->os_thread;
 994         struct l_wait_info    lwi    = { 0 };
 995
 996         /* os_lock: sync status between stop and scrub thread */
 997         cfs_spin_lock(&scrub->os_lock);
 998         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
 999                 thread_set_flags(thread, SVC_STOPPING);
1000                 cfs_spin_unlock(&scrub->os_lock);
1001                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
1002                 l_wait_event(thread->t_ctl_waitq,
1003                              thread_is_stopped(thread),
1004                              &lwi);
1005                 /* Do not skip the last lock/unlock, which can guarantee that
1006                  * the caller cannot return until the OI scrub thread exit. */
1007                 cfs_spin_lock(&scrub->os_lock);
1008         }
1009         cfs_spin_unlock(&scrub->os_lock);
1010 }
1011
1012 static void osd_scrub_stop(struct osd_device *dev)
1013 {
1014         /* od_otable_mutex: prevent curcurrent start/stop */
1015         cfs_mutex_lock(&dev->od_otable_mutex);
1016         dev->od_scrub.os_paused = 1;
1017         do_osd_scrub_stop(&dev->od_scrub);
1018         cfs_mutex_unlock(&dev->od_otable_mutex);
1019 }
1020
1021 static const char osd_scrub_name[] = "OI_scrub";
1022
1023 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
1024 {
1025         struct osd_thread_info     *info   = osd_oti_get(env);
1026         struct osd_scrub           *scrub  = &dev->od_scrub;
1027         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
1028         struct scrub_file          *sf     = &scrub->os_file;
1029         struct osd_inode_id        *id     = &scrub->os_oic.oic_lid;
1030         struct super_block         *sb     = osd_sb(dev);
1031         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
1032         struct inode               *inode;
1033         struct lvfs_run_ctxt        saved;
1034         struct file                *filp;
1035         int                         dirty  = 0;
1036         int                         init   = 0;
1037         int                         rc     = 0;
1038         ENTRY;
1039
1040         memset(scrub, 0, sizeof(*scrub));
1041         OBD_SET_CTXT_MAGIC(ctxt);
1042         ctxt->pwdmnt = dev->od_mnt;
1043         ctxt->pwd = dev->od_mnt->mnt_root;
1044         ctxt->fs = get_ds();
1045
1046         cfs_waitq_init(&scrub->os_thread.t_ctl_waitq);
1047         cfs_init_rwsem(&scrub->os_rwsem);
1048         cfs_spin_lock_init(&scrub->os_lock);
1049         CFS_INIT_LIST_HEAD(&scrub->os_inconsistent_items);
1050
1051         push_ctxt(&saved, ctxt, NULL);
1052         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
1053         if (IS_ERR(filp))
1054                 RETURN(PTR_ERR(filp));
1055
1056         scrub->os_inode = igrab(filp->f_dentry->d_inode);
1057         filp_close(filp, 0);
1058         pop_ctxt(&saved, ctxt, NULL);
1059
1060         rc = osd_scrub_file_load(scrub);
1061         if (rc == -ENOENT) {
1062                 osd_scrub_file_init(scrub, es->s_uuid);
1063                 dirty = 1;
1064                 init = 1;
1065         } else if (rc != 0) {
1066                 RETURN(rc);
1067         } else {
1068                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
1069                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
1070                         dirty = 1;
1071                 } else if (sf->sf_status == SS_SCANNING) {
1072                         sf->sf_status = SS_CRASHED;
1073                         dirty = 1;
1074                 }
1075         }
1076
1077         if (sf->sf_pos_last_checkpoint != 0)
1078                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
1079         else
1080                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb);
1081
1082         if (dirty != 0) {
1083                 rc = osd_scrub_file_store(scrub);
1084                 if (rc != 0)
1085                         RETURN(rc);
1086         }
1087
1088         /* Initialize OI files. */
1089         rc = osd_oi_init(info, dev);
1090         if (rc < 0)
1091                 RETURN(rc);
1092
1093         if (init != 0) {
1094                 rc = __osd_oi_lookup(info, dev, &LU_DOT_LUSTRE_FID, id);
1095                 if (rc == 0) {
1096                         inode = osd_iget(info, dev, id);
1097                         if (IS_ERR(inode)) {
1098                                 rc = PTR_ERR(inode);
1099                                 /* It is restored from old 2.x backup. */
1100                                 if (rc == -ENOENT || rc == -ESTALE) {
1101                                         osd_scrub_file_reset(scrub, es->s_uuid,
1102                                                              SF_INCONSISTENT);
1103                                         rc = osd_scrub_file_store(scrub);
1104                                 }
1105                         } else {
1106                                 iput(inode);
1107                         }
1108                 } else if (rc == -ENOENT) {
1109                         rc = 0;
1110                 }
1111         }
1112
1113         if (rc == 0 && !dev->od_noscrub &&
1114             ((sf->sf_status == SS_PAUSED) ||
1115              (sf->sf_status == SS_CRASHED &&
1116               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_AUTO)) ||
1117              (sf->sf_status == SS_INIT &&
1118               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT))))
1119                 rc = osd_scrub_start(dev);
1120
1121         RETURN(rc);
1122 }
1123
1124 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
1125 {
1126         struct osd_scrub *scrub = &dev->od_scrub;
1127
1128         LASSERT(dev->od_otable_it == NULL);
1129
1130         if (scrub->os_inode != NULL) {
1131                 osd_scrub_stop(dev);
1132                 iput(scrub->os_inode);
1133                 scrub->os_inode = NULL;
1134         }
1135         if (dev->od_oi_table != NULL)
1136                 osd_oi_fini(osd_oti_get(env), dev);
1137 }
1138
1139 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
1140                                        struct dt_object *dt, __u32 attr,
1141                                        struct lustre_capa *capa)
1142 {
1143         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
1144         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
1145         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
1146         struct osd_scrub       *scrub = &dev->od_scrub;
1147         struct osd_otable_it   *it;
1148         __u32                   start = 0;
1149         int                     rc;
1150         ENTRY;
1151
1152         /* od_otable_mutex: prevent curcurrent init/fini */
1153         cfs_mutex_lock(&dev->od_otable_mutex);
1154         if (dev->od_otable_it != NULL)
1155                 GOTO(out, it = ERR_PTR(-EALREADY));
1156
1157         OBD_ALLOC_PTR(it);
1158         if (it == NULL)
1159                 GOTO(out, it = ERR_PTR(-ENOMEM));
1160
1161         dev->od_otable_it = it;
1162         it->ooi_dev = dev;
1163         it->ooi_cache.ooc_consumer_idx = -1;
1164         if (flags & DOIF_OUTUSED)
1165                 it->ooi_used_outside = 1;
1166
1167         if (flags & DOIF_RESET)
1168                 start |= SS_RESET;
1169
1170         if (valid & DOIV_ERROR_HANDLE) {
1171                 if (flags & DOIF_FAILOUT)
1172                         start |= SS_SET_FAILOUT;
1173                 else
1174                         start |= SS_CLEAR_FAILOUT;
1175         }
1176
1177         rc = do_osd_scrub_start(dev, start);
1178         if (rc == -EALREADY) {
1179                 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current - 1;
1180         } else if (rc < 0) {
1181                 dev->od_otable_it = NULL;
1182                 OBD_FREE_PTR(it);
1183                 GOTO(out, it = ERR_PTR(-EALREADY));
1184         } else {
1185                 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
1186         }
1187
1188         GOTO(out, it);
1189
1190 out:
1191         cfs_mutex_unlock(&dev->od_otable_mutex);
1192         return (struct dt_it *)it;
1193 }
1194
1195 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
1196 {
1197         struct osd_otable_it *it  = (struct osd_otable_it *)di;
1198         struct osd_device    *dev = it->ooi_dev;
1199
1200         /* od_otable_mutex: prevent curcurrent init/fini */
1201         cfs_mutex_lock(&dev->od_otable_mutex);
1202         do_osd_scrub_stop(&dev->od_scrub);
1203         LASSERT(dev->od_otable_it == it);
1204
1205         dev->od_otable_it = NULL;
1206         cfs_mutex_unlock(&dev->od_otable_mutex);
1207         OBD_FREE_PTR(it);
1208 }
1209
1210 /**
1211  * XXX: Temporary used to notify otable iteration to be paused.
1212  */
1213 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
1214 {
1215         struct osd_device *dev = ((struct osd_otable_it *)di)->ooi_dev;
1216
1217         /* od_otable_mutex: prevent curcurrent init/fini */
1218         cfs_mutex_lock(&dev->od_otable_mutex);
1219         dev->od_scrub.os_paused = 1;
1220         cfs_mutex_unlock(&dev->od_otable_mutex);
1221 }
1222
1223 /**
1224  * Set the OSD layer iteration start position as the specified key.
1225  *
1226  * The LFSCK out of OSD layer does not know the detail of the key, so if there
1227  * are several keys, they cannot be compared out of OSD, so call "::get()" for
1228  * each key, and OSD will select the smallest one by itself.
1229  */
1230 static int osd_otable_it_get(const struct lu_env *env,
1231                              struct dt_it *di, const struct dt_key *key)
1232 {
1233         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
1234         struct osd_otable_cache *ooc = &it->ooi_cache;
1235         const char              *str = (const char *)key;
1236         __u32                    ino;
1237         ENTRY;
1238
1239         /* Forbid to set iteration position after iteration started. */
1240         if (it->ooi_user_ready)
1241                 RETURN(-EPERM);
1242
1243         if (str[0] == '\0')
1244                 RETURN(-EINVAL);
1245
1246         if (sscanf(str, "%u", &ino) <= 0)
1247                 RETURN(-EINVAL);
1248
1249         /* Skip the one that has been processed last time. */
1250         if (ooc->ooc_pos_preload > ++ino)
1251                 ooc->ooc_pos_preload = ino;
1252
1253         RETURN(0);
1254 }
1255
1256 static int osd_otable_it_preload(const struct lu_env *env,
1257                                  struct osd_otable_it *it)
1258 {
1259         struct osd_device       *dev   = it->ooi_dev;
1260         struct osd_scrub        *scrub = &dev->od_scrub;
1261         struct osd_otable_cache *ooc   = &it->ooi_cache;
1262         int                      rc;
1263         ENTRY;
1264
1265         rc = osd_inode_iteration(osd_oti_get(env), dev,
1266                                  OSD_OTABLE_IT_CACHE_SIZE, 1);
1267         if (rc == SCRUB_IT_ALL)
1268                 it->ooi_all_cached = 1;
1269
1270         CDEBUG(D_LFSCK, "OSD pre-loaded: max = %u, preload = %u, rc = %d\n",
1271                le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count),
1272                ooc->ooc_pos_preload, rc);
1273
1274         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1275                 scrub->os_waiting = 0;
1276                 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
1277         }
1278
1279         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1280 }
1281
1282 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
1283 {
1284         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
1285         struct osd_device       *dev    = it->ooi_dev;
1286         struct osd_scrub        *scrub  = &dev->od_scrub;
1287         struct osd_otable_cache *ooc    = &it->ooi_cache;
1288         struct ptlrpc_thread    *thread = &scrub->os_thread;
1289         struct l_wait_info       lwi    = { 0 };
1290         int                      rc;
1291         ENTRY;
1292
1293         LASSERT(it->ooi_user_ready);
1294
1295 again:
1296         if (!thread_is_running(thread) && !it->ooi_used_outside)
1297                 RETURN(1);
1298
1299         if (ooc->ooc_cached_items > 0) {
1300                 ooc->ooc_cached_items--;
1301                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
1302                                         ~OSD_OTABLE_IT_CACHE_MASK;
1303                 RETURN(0);
1304         }
1305
1306         if (it->ooi_all_cached) {
1307                 l_wait_event(thread->t_ctl_waitq,
1308                              !thread_is_running(thread),
1309                              &lwi);
1310                 RETURN(1);
1311         }
1312
1313         it->ooi_waiting = 1;
1314         l_wait_event(thread->t_ctl_waitq,
1315                      ooc->ooc_pos_preload < scrub->os_pos_current ||
1316                      !thread_is_running(thread),
1317                      &lwi);
1318         it->ooi_waiting = 0;
1319
1320         if (!thread_is_running(thread) && !it->ooi_used_outside)
1321                 RETURN(1);
1322
1323         rc = osd_otable_it_preload(env, it);
1324         if (rc >= 0)
1325                 goto again;
1326
1327         RETURN(rc);
1328 }
1329
1330 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
1331                                         const struct dt_it *di)
1332 {
1333         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
1334         struct osd_otable_cache *ooc = &it->ooi_cache;
1335
1336         sprintf(it->ooi_key, "%u",
1337                 ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino);
1338         return (struct dt_key *)it->ooi_key;
1339 }
1340
1341 static int osd_otable_it_key_size(const struct lu_env *env,
1342                                   const struct dt_it *di)
1343 {
1344         return sizeof(((struct osd_otable_it *)di)->ooi_key);
1345 }
1346
1347 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
1348                              struct dt_rec *rec, __u32 attr)
1349 {
1350         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
1351         struct osd_otable_cache *ooc = &it->ooi_cache;
1352
1353         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
1354         return 0;
1355 }
1356
1357 static int osd_otable_it_load(const struct lu_env *env,
1358                               const struct dt_it *di, __u64 hash)
1359 {
1360         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
1361         struct osd_device       *dev   = it->ooi_dev;
1362         struct osd_otable_cache *ooc   = &it->ooi_cache;
1363         struct osd_scrub        *scrub = &dev->od_scrub;
1364
1365         if (it->ooi_user_ready)
1366                 return 0;
1367
1368         if (ooc->ooc_pos_preload < LDISKFS_FIRST_INO(osd_sb(dev)))
1369                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev));
1370         it->ooi_user_ready = 1;
1371         if (!scrub->os_full_speed)
1372                 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
1373
1374         /* Unplug OSD layer iteration by the first next() call. */
1375         return osd_otable_it_next(env, (struct dt_it *)it);
1376 }
1377
1378 const struct dt_index_operations osd_otable_ops = {
1379         .dio_it = {
1380                 .init     = osd_otable_it_init,
1381                 .fini     = osd_otable_it_fini,
1382                 .put      = osd_otable_it_put,
1383                 .get      = osd_otable_it_get,
1384                 .next     = osd_otable_it_next,
1385                 .key      = osd_otable_it_key,
1386                 .key_size = osd_otable_it_key_size,
1387                 .rec      = osd_otable_it_rec,
1388                 .load     = osd_otable_it_load,
1389         }
1390 };
1391
1392 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
1393                    int insert)
1394 {
1395         struct osd_inconsistent_item *oii;
1396         struct osd_scrub             *scrub  = &dev->od_scrub;
1397         struct ptlrpc_thread         *thread = &scrub->os_thread;
1398         int                           wakeup = 0;
1399         ENTRY;
1400
1401         OBD_ALLOC_PTR(oii);
1402         if (unlikely(oii == NULL))
1403                 RETURN(-ENOMEM);
1404
1405         CFS_INIT_LIST_HEAD(&oii->oii_list);
1406         oii->oii_cache = *oic;
1407         oii->oii_insert = insert;
1408
1409         cfs_spin_lock(&scrub->os_lock);
1410         if (unlikely(!thread_is_running(thread))) {
1411                 cfs_spin_unlock(&scrub->os_lock);
1412                 OBD_FREE_PTR(oii);
1413                 RETURN(-EAGAIN);
1414         }
1415
1416         if (cfs_list_empty(&scrub->os_inconsistent_items))
1417                 wakeup = 1;
1418         cfs_list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
1419         cfs_spin_unlock(&scrub->os_lock);
1420
1421         if (wakeup != 0)
1422                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
1423
1424         RETURN(0);
1425 }
1426
1427 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
1428                    struct osd_inode_id *id)
1429 {
1430         struct osd_scrub             *scrub = &dev->od_scrub;
1431         struct osd_inconsistent_item *oii;
1432         ENTRY;
1433
1434         cfs_spin_lock(&scrub->os_lock);
1435         cfs_list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
1436                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
1437                         *id = oii->oii_cache.oic_lid;
1438                         cfs_spin_unlock(&scrub->os_lock);
1439                         RETURN(0);
1440                 }
1441         }
1442         cfs_spin_unlock(&scrub->os_lock);
1443
1444         RETURN(-ENOENT);
1445 }
1446
1447 static const char *scrub_status_names[] = {
1448         "init",
1449         "scanning",
1450         "completed",
1451         "failed",
1452         "stopped",
1453         "paused",
1454         "crashed",
1455         NULL
1456 };
1457
1458 static const char *scrub_flags_names[] = {
1459         "recreated",
1460         "inconsistent",
1461         "auto",
1462         NULL
1463 };
1464
1465 static const char *scrub_param_names[] = {
1466         "failout",
1467         NULL
1468 };
1469
1470 static int scrub_bits_dump(char **buf, int *len, int bits, const char *names[],
1471                            const char *prefix)
1472 {
1473         int save = *len;
1474         int flag;
1475         int rc;
1476         int i;
1477
1478         rc = snprintf(*buf, *len, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
1479         if (rc <= 0)
1480                 return -ENOSPC;
1481
1482         *buf += rc;
1483         *len -= rc;
1484         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
1485                 if (flag & bits) {
1486                         bits &= ~flag;
1487                         rc = snprintf(*buf, *len, "%s%c", names[i],
1488                                       bits != 0 ? ',' : '\n');
1489                         if (rc <= 0)
1490                                 return -ENOSPC;
1491
1492                         *buf += rc;
1493                         *len -= rc;
1494                 }
1495         }
1496         return save - *len;
1497 }
1498
1499 static int scrub_time_dump(char **buf, int *len, __u64 time, const char *prefix)
1500 {
1501         int rc;
1502
1503         if (time != 0)
1504                 rc = snprintf(*buf, *len, "%s: "LPU64" seconds\n", prefix,
1505                               cfs_time_current_sec() - time);
1506         else
1507                 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
1508         if (rc <= 0)
1509                 return -ENOSPC;
1510
1511         *buf += rc;
1512         *len -= rc;
1513         return rc;
1514 }
1515
1516 static int scrub_pos_dump(char **buf, int *len, __u64 pos, const char *prefix)
1517 {
1518         int rc;
1519
1520         if (pos != 0)
1521                 rc = snprintf(*buf, *len, "%s: "LPU64"\n", prefix, pos);
1522         else
1523                 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
1524         if (rc <= 0)
1525                 return -ENOSPC;
1526
1527         *buf += rc;
1528         *len -= rc;
1529         return rc;
1530 }
1531
1532 int osd_scrub_dump(struct osd_device *dev, char *buf, int len)
1533 {
1534         struct osd_scrub  *scrub   = &dev->od_scrub;
1535         struct scrub_file *sf      = &scrub->os_file;
1536         __u64              checked;
1537         __u64              speed;
1538         int                save    = len;
1539         int                ret     = -ENOSPC;
1540         int                rc;
1541
1542         cfs_down_read(&scrub->os_rwsem);
1543         rc = snprintf(buf, len,
1544                       "name: OI scrub\n"
1545                       "magic: 0x%x\n"
1546                       "oi_files: %d\n"
1547                       "status: %s\n",
1548                       sf->sf_magic, (int)sf->sf_oi_count,
1549                       scrub_status_names[sf->sf_status]);
1550         if (rc <= 0)
1551                 goto out;
1552
1553         buf += rc;
1554         len -= rc;
1555         rc = scrub_bits_dump(&buf, &len, sf->sf_flags, scrub_flags_names,
1556                              "flags");
1557         if (rc < 0)
1558                 goto out;
1559
1560         rc = scrub_bits_dump(&buf, &len, sf->sf_param, scrub_param_names,
1561                              "param");
1562         if (rc < 0)
1563                 goto out;
1564
1565         rc = scrub_time_dump(&buf, &len, sf->sf_time_last_complete,
1566                              "time_since_last_completed");
1567         if (rc < 0)
1568                 goto out;
1569
1570         rc = scrub_time_dump(&buf, &len, sf->sf_time_latest_start,
1571                              "time_since_latest_start");
1572         if (rc < 0)
1573                 goto out;
1574
1575         rc = scrub_time_dump(&buf, &len, sf->sf_time_last_checkpoint,
1576                              "time_since_last_checkpoint");
1577         if (rc < 0)
1578                 goto out;
1579
1580         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_latest_start,
1581                             "latest_start_position");
1582         if (rc < 0)
1583                 goto out;
1584
1585         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_last_checkpoint,
1586                             "last_checkpoint_position");
1587         if (rc < 0)
1588                 goto out;
1589
1590         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_first_inconsistent,
1591                             "first_failure_position");
1592         if (rc < 0)
1593                 goto out;
1594
1595         checked = sf->sf_items_checked + scrub->os_new_checked;
1596         rc = snprintf(buf, len,
1597                       "checked: "LPU64"\n"
1598                       "updated: "LPU64"\n"
1599                       "failed: "LPU64"\n"
1600                       "prior_updated: "LPU64"\n"
1601                       "noscrub: "LPU64"\n"
1602                       "igif: "LPU64"\n"
1603                       "success_count: %u\n",
1604                       checked, sf->sf_items_updated, sf->sf_items_failed,
1605                       sf->sf_items_updated_prior, sf->sf_items_noscrub,
1606                       sf->sf_items_igif, sf->sf_success_count);
1607         if (rc <= 0)
1608                 goto out;
1609
1610         buf += rc;
1611         len -= rc;
1612         speed = checked;
1613         if (thread_is_running(&scrub->os_thread)) {
1614                 cfs_duration_t duration = cfs_time_current() -
1615                                           scrub->os_time_last_checkpoint;
1616                 __u64 new_checked = scrub->os_new_checked * CFS_HZ;
1617                 __u32 rtime = sf->sf_run_time +
1618                               cfs_duration_sec(duration + HALF_SEC);
1619
1620                 if (duration != 0)
1621                         do_div(new_checked, duration);
1622                 if (rtime != 0)
1623                         do_div(speed, rtime);
1624                 rc = snprintf(buf, len,
1625                               "run_time: %u seconds\n"
1626                               "average_speed: "LPU64" objects/sec\n"
1627                               "real-time_speed: "LPU64" objects/sec\n"
1628                               "current_position: %u\n",
1629                               rtime, speed, new_checked, scrub->os_pos_current);
1630         } else {
1631                 if (sf->sf_run_time != 0)
1632                         do_div(speed, sf->sf_run_time);
1633                 rc = snprintf(buf, len,
1634                               "run_time: %u seconds\n"
1635                               "average_speed: "LPU64" objects/sec\n"
1636                               "real-time_speed: N/A\n"
1637                               "current_position: N/A\n",
1638                               sf->sf_run_time, speed);
1639         }
1640         if (rc <= 0)
1641                 goto out;
1642
1643         buf += rc;
1644         len -= rc;
1645         ret = save - len;
1646
1647 out:
1648         cfs_up_read(&scrub->os_rwsem);
1649         return ret;
1650 }