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