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, 2013, Intel Corporation.
  24  */
  25 /*
  26  * lustre/osd-ldiskfs/osd_scrub.c
  27  *
  28  * Top-level entry points into osd module
  29  *
  30  * The OI scrub is used for rebuilding Object Index files when restores MDT from
  31  * file-level backup.
  32  *
  33  * The otable based iterator scans ldiskfs inode table to feed up layer LFSCK.
  34  *
  35  * Author: Fan Yong <yong.fan@whamcloud.com>
  36  */
  37
  38 #ifndef EXPORT_SYMTAB
  39 # define EXPORT_SYMTAB
  40 #endif
  41 #define DEBUG_SUBSYSTEM S_MDS
  42
  43 #include <lustre/lustre_idl.h>
  44 #include <lustre_disk.h>
  45 #include <dt_object.h>
  46
  47 #include "osd_internal.h"
  48 #include "osd_oi.h"
  49 #include "osd_scrub.h"
  50
  51 #define HALF_SEC        (CFS_HZ >> 1)
  52
  53 #define OSD_OTABLE_MAX_HASH             0x00000000ffffffffULL
  54
  55 #define SCRUB_NEXT_BREAK        1 /* exit current loop and process next group */
  56 #define SCRUB_NEXT_CONTINUE     2 /* skip current object and process next bit */
  57 #define SCRUB_NEXT_EXIT         3 /* exit all the loops */
  58 #define SCRUB_NEXT_WAIT         4 /* wait for free cache slot */
  59 #define SCRUB_NEXT_CRASH        5 /* simulate system crash during OI scrub */
  60 #define SCRUB_NEXT_FATAL        6 /* simulate failure during OI scrub */
  61 #define SCRUB_NEXT_NOSCRUB      7 /* new created object, no scrub on it */
  62 #define SCRUB_NEXT_NOLMA        8 /* the inode has no FID-in-LMA */
  63
  64 /* misc functions */
  65
  66 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
  67 {
  68         return container_of0(scrub, struct osd_device, od_scrub);
  69 }
  70
  71 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
  72 {
  73         return osd_sb(osd_scrub2dev(scrub));
  74 }
  75
  76 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
  77                                        struct osd_otable_cache *ooc)
  78 {
  79         return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
  80 }
  81
  82 /**
  83  * update/insert/delete the specified OI mapping (@fid @id) according to the ops
  84  *
  85  * \retval   1, changed nothing
  86  * \retval   0, changed successfully
  87  * \retval -ve, on error
  88  */
  89 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
  90                                      struct osd_device *dev,
  91                                      const struct lu_fid *fid,
  92                                      const struct osd_inode_id *id, int ops)
  93 {
  94         struct lu_fid         *oi_fid = &info->oti_fid2;
  95         struct osd_inode_id   *oi_id  = &info->oti_id2;
  96         struct iam_container  *bag;
  97         struct iam_path_descr *ipd;
  98         handle_t              *jh;
  99         int                    rc;
 100         ENTRY;
 101
 102         fid_cpu_to_be(oi_fid, fid);
 103         osd_id_pack(oi_id, id);
 104         jh = ldiskfs_journal_start_sb(osd_sb(dev),
 105                                       osd_dto_credits_noquota[ops]);
 106         if (IS_ERR(jh)) {
 107                 rc = PTR_ERR(jh);
 108                 CERROR("%.16s: fail to start trans for scrub store: rc = %d\n",
 109                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
 110                 RETURN(rc);
 111         }
 112
 113         bag = &osd_fid2oi(dev, fid)->oi_dir.od_container;
 114         ipd = osd_idx_ipd_get(info->oti_env, bag);
 115         if (unlikely(ipd == NULL)) {
 116                 ldiskfs_journal_stop(jh);
 117                 CERROR("%.16s: fail to get ipd for scrub store\n",
 118                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name);
 119                 RETURN(-ENOMEM);
 120         }
 121
 122         switch (ops) {
 123         case DTO_INDEX_UPDATE:
 124                 rc = iam_update(jh, bag, (const struct iam_key *)oi_fid,
 125                                 (struct iam_rec *)oi_id, ipd);
 126                 if (unlikely(rc == -ENOENT)) {
 127                         /* Some unlink thread may removed the OI mapping. */
 128                         rc = 1;
 129                 }
 130                 break;
 131         case DTO_INDEX_INSERT:
 132                 rc = iam_insert(jh, bag, (const struct iam_key *)oi_fid,
 133                                 (struct iam_rec *)oi_id, ipd);
 134                 if (unlikely(rc == -EEXIST)) {
 135                         rc = 1;
 136                         /* XXX: There are trouble things when adding OI
 137                          *      mapping for IGIF object, which may cause
 138                          *      multiple objects to be mapped to the same
 139                          *      IGIF formatted FID. Consider the following
 140                          *      situations:
 141                          *
 142                          *      1) The MDT is upgrading from 1.8 device.
 143                          *      The OI scrub generates IGIF FID1 for the
 144                          *      OBJ1 and adds the OI mapping.
 145                          *
 146                          *      2) For some reason, the OI scrub does not
 147                          *      process all the IGIF objects completely.
 148                          *
 149                          *      3) The MDT is backuped and restored against
 150                          *      this device.
 151                          *
 152                          *      4) When the MDT mounts up, the OI scrub will
 153                          *      try to rebuild the OI files. For some IGIF
 154                          *      object, OBJ2, which was not processed by the
 155                          *      OI scrub before the backup/restore, and the
 156                          *      new generated IGIF formatted FID may be just
 157                          *      the FID1, the same as OBJ1.
 158                          *
 159                          *      Under such case, the OI scrub cannot know how
 160                          *      to generate new FID for the OBJ2.
 161                          *
 162                          *      Currently, we do nothing for that. One possible
 163                          *      solution is to generate new normal FID for the
 164                          *      conflict object.
 165                          *
 166                          *      Anyway, it is rare, only exists in theory. */
 167                 }
 168                 break;
 169         case DTO_INDEX_DELETE:
 170                 rc = iam_delete(jh, bag, (const struct iam_key *)oi_fid, ipd);
 171                 if (rc == -ENOENT) {
 172                         /* It is normal that the unlink thread has removed the
 173                          * OI mapping already. */
 174                         rc = 1;
 175                 }
 176                 break;
 177         default:
 178                 LASSERTF(0, "Unexpected ops %d\n", ops);
 179                 break;
 180         }
 181         osd_ipd_put(info->oti_env, bag, ipd);
 182         ldiskfs_journal_stop(jh);
 183         RETURN(rc);
 184 }
 185
 186 /* OI_scrub file ops */
 187
 188 static void osd_scrub_file_to_cpu(struct scrub_file *des,
 189                                   struct scrub_file *src)
 190 {
 191         memcpy(des->sf_uuid, src->sf_uuid, 16);
 192         des->sf_flags   = le64_to_cpu(src->sf_flags);
 193         des->sf_magic   = le32_to_cpu(src->sf_magic);
 194         des->sf_status  = le16_to_cpu(src->sf_status);
 195         des->sf_param   = le16_to_cpu(src->sf_param);
 196         des->sf_time_last_complete      =
 197                                 le64_to_cpu(src->sf_time_last_complete);
 198         des->sf_time_latest_start       =
 199                                 le64_to_cpu(src->sf_time_latest_start);
 200         des->sf_time_last_checkpoint    =
 201                                 le64_to_cpu(src->sf_time_last_checkpoint);
 202         des->sf_pos_latest_start        =
 203                                 le64_to_cpu(src->sf_pos_latest_start);
 204         des->sf_pos_last_checkpoint     =
 205                                 le64_to_cpu(src->sf_pos_last_checkpoint);
 206         des->sf_pos_first_inconsistent  =
 207                                 le64_to_cpu(src->sf_pos_first_inconsistent);
 208         des->sf_items_checked           =
 209                                 le64_to_cpu(src->sf_items_checked);
 210         des->sf_items_updated           =
 211                                 le64_to_cpu(src->sf_items_updated);
 212         des->sf_items_failed            =
 213                                 le64_to_cpu(src->sf_items_failed);
 214         des->sf_items_updated_prior     =
 215                                 le64_to_cpu(src->sf_items_updated_prior);
 216         des->sf_run_time        = le32_to_cpu(src->sf_run_time);
 217         des->sf_success_count   = le32_to_cpu(src->sf_success_count);
 218         des->sf_oi_count        = le16_to_cpu(src->sf_oi_count);
 219         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 220 }
 221
 222 static void osd_scrub_file_to_le(struct scrub_file *des,
 223                                  struct scrub_file *src)
 224 {
 225         memcpy(des->sf_uuid, src->sf_uuid, 16);
 226         des->sf_flags   = cpu_to_le64(src->sf_flags);
 227         des->sf_magic   = cpu_to_le32(src->sf_magic);
 228         des->sf_status  = cpu_to_le16(src->sf_status);
 229         des->sf_param   = cpu_to_le16(src->sf_param);
 230         des->sf_time_last_complete      =
 231                                 cpu_to_le64(src->sf_time_last_complete);
 232         des->sf_time_latest_start       =
 233                                 cpu_to_le64(src->sf_time_latest_start);
 234         des->sf_time_last_checkpoint    =
 235                                 cpu_to_le64(src->sf_time_last_checkpoint);
 236         des->sf_pos_latest_start        =
 237                                 cpu_to_le64(src->sf_pos_latest_start);
 238         des->sf_pos_last_checkpoint     =
 239                                 cpu_to_le64(src->sf_pos_last_checkpoint);
 240         des->sf_pos_first_inconsistent  =
 241                                 cpu_to_le64(src->sf_pos_first_inconsistent);
 242         des->sf_items_checked           =
 243                                 cpu_to_le64(src->sf_items_checked);
 244         des->sf_items_updated           =
 245                                 cpu_to_le64(src->sf_items_updated);
 246         des->sf_items_failed            =
 247                                 cpu_to_le64(src->sf_items_failed);
 248         des->sf_items_updated_prior     =
 249                                 cpu_to_le64(src->sf_items_updated_prior);
 250         des->sf_run_time        = cpu_to_le32(src->sf_run_time);
 251         des->sf_success_count   = cpu_to_le32(src->sf_success_count);
 252         des->sf_oi_count        = cpu_to_le16(src->sf_oi_count);
 253         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 254 }
 255
 256 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
 257 {
 258         struct scrub_file *sf = &scrub->os_file;
 259
 260         memset(sf, 0, sizeof(*sf));
 261         memcpy(sf->sf_uuid, uuid, 16);
 262         sf->sf_magic = SCRUB_MAGIC_V1;
 263         sf->sf_status = SS_INIT;
 264 }
 265
 266 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
 267 {
 268         struct scrub_file *sf = &scrub->os_file;
 269
 270         CDEBUG(D_LFSCK, "Reset OI scrub file, flags = "LPX64"\n", flags);
 271         memcpy(sf->sf_uuid, uuid, 16);
 272         sf->sf_status = SS_INIT;
 273         sf->sf_flags |= flags;
 274         sf->sf_param = 0;
 275         sf->sf_run_time = 0;
 276         sf->sf_time_latest_start = 0;
 277         sf->sf_time_last_checkpoint = 0;
 278         sf->sf_pos_latest_start = 0;
 279         sf->sf_pos_last_checkpoint = 0;
 280         sf->sf_pos_first_inconsistent = 0;
 281         sf->sf_items_checked = 0;
 282         sf->sf_items_updated = 0;
 283         sf->sf_items_failed = 0;
 284         sf->sf_items_updated_prior = 0;
 285         sf->sf_items_noscrub = 0;
 286         sf->sf_items_igif = 0;
 287 }
 288
 289 static int osd_scrub_file_load(struct osd_scrub *scrub)
 290 {
 291         loff_t  pos  = 0;
 292         char   *name = LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
 293         int     len  = sizeof(scrub->os_file_disk);
 294         int     rc;
 295
 296         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
 297         if (rc == len) {
 298                 struct scrub_file *sf = &scrub->os_file;
 299
 300                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
 301                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
 302                         CWARN("%.16s: invalid scrub magic 0x%x != 0x%x\n,",
 303                               name, sf->sf_magic, SCRUB_MAGIC_V1);
 304                         /* Process it as new scrub file. */
 305                         rc = -ENOENT;
 306                 } else {
 307                         rc = 0;
 308                 }
 309         } else if (rc != 0) {
 310                 CERROR("%.16s: fail to load scrub file, expected = %d, "
 311                        "rc = %d\n", name, len, rc);
 312                 if (rc > 0)
 313                         rc = -EFAULT;
 314         } else {
 315                 /* return -ENOENT for empty scrub file case. */
 316                 rc = -ENOENT;
 317         }
 318
 319         return rc;
 320 }
 321
 322 int osd_scrub_file_store(struct osd_scrub *scrub)
 323 {
 324         struct osd_device *dev;
 325         handle_t          *jh;
 326         loff_t             pos     = 0;
 327         int                len     = sizeof(scrub->os_file_disk);
 328         int                credits;
 329         int                rc;
 330
 331         dev = container_of0(scrub, struct osd_device, od_scrub);
 332         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
 333                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
 334         jh = ldiskfs_journal_start_sb(osd_sb(dev), credits);
 335         if (IS_ERR(jh)) {
 336                 rc = PTR_ERR(jh);
 337                 CERROR("%.16s: fail to start trans for scrub store, rc = %d\n",
 338                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,rc);
 339                 return rc;
 340         }
 341
 342         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
 343         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
 344                                       len, 0, &pos, jh);
 345         ldiskfs_journal_stop(jh);
 346         if (rc != 0)
 347                 CERROR("%.16s: fail to store scrub file, expected = %d, "
 348                        "rc = %d\n",
 349                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
 350                        len, rc);
 351         scrub->os_time_last_checkpoint = cfs_time_current();
 352         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
 353                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
 354         return rc;
 355 }
 356
 357 /* OI scrub APIs */
 358
 359 static int osd_scrub_prep(struct osd_device *dev)
 360 {
 361         struct osd_scrub     *scrub  = &dev->od_scrub;
 362         struct ptlrpc_thread *thread = &scrub->os_thread;
 363         struct scrub_file    *sf     = &scrub->os_file;
 364         __u32                 flags  = scrub->os_start_flags;
 365         int                   rc;
 366         ENTRY;
 367
 368         down_write(&scrub->os_rwsem);
 369         if (flags & SS_SET_FAILOUT)
 370                 sf->sf_param |= SP_FAILOUT;
 371
 372         if (flags & SS_CLEAR_FAILOUT)
 373                 sf->sf_param &= ~SP_FAILOUT;
 374
 375         if (flags & SS_RESET)
 376                 osd_scrub_file_reset(scrub,
 377                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
 378
 379         if (flags & SS_AUTO) {
 380                 scrub->os_full_speed = 1;
 381                 sf->sf_flags |= SF_AUTO;
 382         } else {
 383                 scrub->os_full_speed = 0;
 384         }
 385
 386         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
 387                 scrub->os_full_speed = 1;
 388
 389         scrub->os_in_prior = 0;
 390         scrub->os_waiting = 0;
 391         scrub->os_paused = 0;
 392         scrub->os_new_checked = 0;
 393         if (sf->sf_pos_last_checkpoint != 0)
 394                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 395         else
 396                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 397
 398         scrub->os_pos_current = sf->sf_pos_latest_start;
 399         sf->sf_status = SS_SCANNING;
 400         sf->sf_time_latest_start = cfs_time_current_sec();
 401         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 402         rc = osd_scrub_file_store(scrub);
 403         if (rc == 0) {
 404                 spin_lock(&scrub->os_lock);
 405                 thread_set_flags(thread, SVC_RUNNING);
 406                 spin_unlock(&scrub->os_lock);
 407                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
 408         }
 409         up_write(&scrub->os_rwsem);
 410
 411         RETURN(rc);
 412 }
 413
 414 static int
 415 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
 416                        struct osd_idmap_cache *oic, int val)
 417 {
 418         struct osd_scrub             *scrub  = &dev->od_scrub;
 419         struct scrub_file            *sf     = &scrub->os_file;
 420         struct lu_fid                *fid    = &oic->oic_fid;
 421         struct osd_inode_id          *lid    = &oic->oic_lid;
 422         struct osd_inode_id          *lid2   = &info->oti_id;
 423         struct osd_inconsistent_item *oii    = NULL;
 424         struct inode                 *inode  = NULL;
 425         int                           ops    = DTO_INDEX_UPDATE;
 426         int                           idx;
 427         int                           rc;
 428         ENTRY;
 429
 430         down_write(&scrub->os_rwsem);
 431         scrub->os_new_checked++;
 432         if (val < 0)
 433                 GOTO(out, rc = val);
 434
 435         if (scrub->os_in_prior)
 436                 oii = cfs_list_entry(oic, struct osd_inconsistent_item,
 437                                      oii_cache);
 438
 439         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
 440                 GOTO(out, rc = 0);
 441
 442         if (fid_is_igif(fid))
 443                 sf->sf_items_igif++;
 444
 445         if ((val == SCRUB_NEXT_NOLMA) &&
 446             (!dev->od_handle_nolma || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
 447                 GOTO(out, rc = 0);
 448
 449         if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA))
 450                 goto iget;
 451
 452         /* XXX: Currently, no FID-in-LMA for OST object, so osd_oi_lookup()
 453          *      without checking FLD is enough.
 454          *
 455          *      It should be updated if FID-in-LMA for OSD object introduced
 456          *      in the future. */
 457         rc = osd_oi_lookup(info, dev, fid, lid2, false);
 458         if (rc != 0) {
 459                 if (rc != -ENOENT)
 460                         GOTO(out, rc);
 461
 462 iget:
 463                 inode = osd_iget(info, dev, lid);
 464                 if (IS_ERR(inode)) {
 465                         rc = PTR_ERR(inode);
 466                         /* Someone removed the inode. */
 467                         if (rc == -ENOENT || rc == -ESTALE)
 468                                 rc = 0;
 469                         GOTO(out, rc);
 470                 }
 471
 472                 /* Check whether the inode to be unlinked during OI scrub. */
 473                 if (unlikely(inode->i_nlink == 0)) {
 474                         iput(inode);
 475                         GOTO(out, rc = 0);
 476                 }
 477
 478                 ops = DTO_INDEX_INSERT;
 479                 idx = osd_oi_fid2idx(dev, fid);
 480                 if (val == SCRUB_NEXT_NOLMA) {
 481                         sf->sf_flags |= SF_UPGRADE;
 482                         scrub->os_full_speed = 1;
 483                         rc = osd_ea_fid_set(info, inode, fid, 0);
 484                         if (rc != 0)
 485                                 GOTO(out, rc);
 486
 487                         if (!(sf->sf_flags & SF_INCONSISTENT))
 488                                 dev->od_igif_inoi = 0;
 489                 } else {
 490                         sf->sf_flags |= SF_RECREATED;
 491                         scrub->os_full_speed = 1;
 492                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 493                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 494                 }
 495         } else if (osd_id_eq(lid, lid2)) {
 496                 GOTO(out, rc = 0);
 497         } else {
 498                 sf->sf_flags |= SF_INCONSISTENT;
 499                 scrub->os_full_speed = 1;
 500
 501                 /* XXX: If the device is restored from file-level backup, then
 502                  *      some IGIFs may have been already in OI files, and some
 503                  *      may be not yet. Means upgrading from 1.8 may be partly
 504                  *      processed, but some clients may hold some immobilized
 505                  *      IGIFs, and use them to access related objects. Under
 506                  *      such case, OSD does not know whether an given IGIF has
 507                  *      been processed or to be processed, and it also cannot
 508                  *      generate local ino#/gen# directly from the immobilized
 509                  *      IGIF because of the backup/restore. Then force OSD to
 510                  *      lookup the given IGIF in OI files, and if no entry,
 511                  *      then ask the client to retry after upgrading completed.
 512                  *      No better choice. */
 513                 dev->od_igif_inoi = 1;
 514         }
 515
 516         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops);
 517         if (rc == 0) {
 518                 if (scrub->os_in_prior)
 519                         sf->sf_items_updated_prior++;
 520                 else
 521                         sf->sf_items_updated++;
 522         }
 523
 524         GOTO(out, rc);
 525
 526 out:
 527         if (rc < 0) {
 528                 sf->sf_items_failed++;
 529                 if (sf->sf_pos_first_inconsistent == 0 ||
 530                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 531                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 532         } else {
 533                 rc = 0;
 534         }
 535
 536         if (ops == DTO_INDEX_INSERT) {
 537                 /* There may be conflict unlink during the OI scrub,
 538                  * if happend, then remove the new added OI mapping. */
 539                 if (unlikely(inode->i_nlink == 0))
 540                         osd_scrub_refresh_mapping(info, dev, fid, lid,
 541                                                   DTO_INDEX_DELETE);
 542                 iput(inode);
 543         }
 544         up_write(&scrub->os_rwsem);
 545
 546         if (oii != NULL) {
 547                 LASSERT(!cfs_list_empty(&oii->oii_list));
 548
 549                 spin_lock(&scrub->os_lock);
 550                 cfs_list_del_init(&oii->oii_list);
 551                 spin_unlock(&scrub->os_lock);
 552                 OBD_FREE_PTR(oii);
 553         }
 554         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 555 }
 556
 557 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
 558 {
 559         struct scrub_file *sf = &scrub->os_file;
 560         int                rc;
 561
 562         if (likely(cfs_time_before(cfs_time_current(),
 563                                    scrub->os_time_next_checkpoint) ||
 564                    scrub->os_new_checked == 0))
 565                 return 0;
 566
 567         down_write(&scrub->os_rwsem);
 568         sf->sf_items_checked += scrub->os_new_checked;
 569         scrub->os_new_checked = 0;
 570         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 571         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 572         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 573                                             scrub->os_time_last_checkpoint);
 574         rc = osd_scrub_file_store(scrub);
 575         up_write(&scrub->os_rwsem);
 576
 577         return rc;
 578 }
 579
 580 static void osd_scrub_post(struct osd_scrub *scrub, int result)
 581 {
 582         struct scrub_file *sf = &scrub->os_file;
 583         ENTRY;
 584
 585         down_write(&scrub->os_rwsem);
 586         spin_lock(&scrub->os_lock);
 587         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 588         spin_unlock(&scrub->os_lock);
 589         if (scrub->os_new_checked > 0) {
 590                 sf->sf_items_checked += scrub->os_new_checked;
 591                 scrub->os_new_checked = 0;
 592                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 593         }
 594         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 595         if (result > 0) {
 596                 struct osd_device *dev =
 597                         container_of0(scrub, struct osd_device, od_scrub);
 598
 599                 dev->od_igif_inoi = 1;
 600                 sf->sf_status = SS_COMPLETED;
 601                 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 602                 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 603                                   SF_UPGRADE | SF_AUTO);
 604                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 605                 sf->sf_success_count++;
 606         } else if (result == 0) {
 607                 if (scrub->os_paused)
 608                         sf->sf_status = SS_PAUSED;
 609                 else
 610                         sf->sf_status = SS_STOPPED;
 611         } else {
 612                 sf->sf_status = SS_FAILED;
 613         }
 614         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 615                                             scrub->os_time_last_checkpoint);
 616         result = osd_scrub_file_store(scrub);
 617         if (result < 0)
 618                 CERROR("%.16s: fail to osd_scrub_post, rc = %d\n",
 619                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
 620                        result);
 621         up_write(&scrub->os_rwsem);
 622
 623         EXIT;
 624 }
 625
 626 /* iteration engine */
 627
 628 struct osd_iit_param {
 629         struct super_block *sb;
 630         struct buffer_head *bitmap;
 631         ldiskfs_group_t bg;
 632         __u32 gbase;
 633         __u32 offset;
 634 };
 635
 636 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 637                                    struct osd_device *dev,
 638                                    struct osd_iit_param *param,
 639                                    struct osd_idmap_cache **oic,
 640                                    int noslot);
 641
 642 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 643                                    struct osd_device *dev,
 644                                    struct osd_iit_param *param,
 645                                    struct osd_idmap_cache *oic,
 646                                    int *noslot, int rc);
 647
 648 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 649 {
 650         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 651                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 652         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 653                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 654                 return SCRUB_NEXT_BREAK;
 655         } else {
 656                 *pos = param->gbase + param->offset;
 657                 return 0;
 658         }
 659 }
 660
 661 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 662                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 663                         struct super_block *sb, bool scrub)
 664 {
 665         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 666         struct inode            *inode;
 667         int                      rc;
 668
 669         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 670         inode = osd_iget(info, dev, lid);
 671         if (IS_ERR(inode)) {
 672                 rc = PTR_ERR(inode);
 673                 /* The inode may be removed after bitmap searching, or the
 674                  * file is new created without inode initialized yet. */
 675                 if (rc == -ENOENT || rc == -ESTALE)
 676                         return SCRUB_NEXT_CONTINUE;
 677
 678                 CERROR("%.16s: fail to read inode, ino# = %u, rc = %d\n",
 679                        LDISKFS_SB(sb)->s_es->s_volume_name, pos, rc);
 680                 return rc;
 681         }
 682
 683         /* If the inode has no OI mapping, then it is special locally used,
 684          * should be invisible to OI scrub or up layer LFSCK. */
 685         if (ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI)) {
 686                 iput(inode);
 687                 return SCRUB_NEXT_CONTINUE;
 688         }
 689
 690         if (scrub &&
 691             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 692                 /* Only skip it for the first OI scrub accessing. */
 693                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 694                 iput(inode);
 695                 return SCRUB_NEXT_NOSCRUB;
 696         }
 697
 698         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 699         if (rc == 0) {
 700                 if (fid_is_llog(&lma->lma_self_fid) ||
 701                     (!scrub && fid_is_internal(&lma->lma_self_fid)))
 702                         rc = SCRUB_NEXT_CONTINUE;
 703                 else
 704                         *fid = lma->lma_self_fid;
 705         } else if (rc == -ENODATA) {
 706                 lu_igif_build(fid, inode->i_ino, inode->i_generation);
 707                 if (scrub)
 708                         rc = SCRUB_NEXT_NOLMA;
 709                 else
 710                         rc = 0;
 711         }
 712         iput(inode);
 713         return rc;
 714 }
 715
 716 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 717                           struct osd_iit_param *param,
 718                           struct osd_idmap_cache **oic, int noslot)
 719 {
 720         struct osd_scrub     *scrub  = &dev->od_scrub;
 721         struct ptlrpc_thread *thread = &scrub->os_thread;
 722         struct lu_fid        *fid;
 723         struct osd_inode_id  *lid;
 724         int                   rc;
 725
 726         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
 727                 struct l_wait_info lwi;
 728
 729                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
 730                 l_wait_event(thread->t_ctl_waitq,
 731                              !cfs_list_empty(&scrub->os_inconsistent_items) ||
 732                              !thread_is_running(thread),
 733                              &lwi);
 734         }
 735
 736         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 737                 spin_lock(&scrub->os_lock);
 738                 thread_set_flags(thread, SVC_STOPPING);
 739                 spin_unlock(&scrub->os_lock);
 740                 return SCRUB_NEXT_CRASH;
 741         }
 742
 743         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
 744                 return SCRUB_NEXT_FATAL;
 745
 746         if (unlikely(!thread_is_running(thread)))
 747                 return SCRUB_NEXT_EXIT;
 748
 749         if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
 750                 struct osd_inconsistent_item *oii;
 751
 752                 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
 753                                      struct osd_inconsistent_item, oii_list);
 754                 *oic = &oii->oii_cache;
 755                 scrub->os_in_prior = 1;
 756                 return 0;
 757         }
 758
 759         if (noslot != 0)
 760                 return SCRUB_NEXT_WAIT;
 761
 762         rc = osd_iit_next(param, &scrub->os_pos_current);
 763         if (rc != 0)
 764                 return rc;
 765
 766         *oic = &scrub->os_oic;
 767         fid = &(*oic)->oic_fid;
 768         lid = &(*oic)->oic_lid;
 769         rc = osd_iit_iget(info, dev, fid, lid,
 770                           scrub->os_pos_current, param->sb, true);
 771         return rc;
 772 }
 773
 774 static int osd_preload_next(struct osd_thread_info *info,
 775                             struct osd_device *dev, struct osd_iit_param *param,
 776                             struct osd_idmap_cache **oic, int noslot)
 777 {
 778         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
 779         struct osd_scrub        *scrub;
 780         struct ptlrpc_thread    *thread;
 781         int                      rc;
 782
 783         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
 784         if (rc != 0)
 785                 return rc;
 786
 787         scrub = &dev->od_scrub;
 788         thread = &scrub->os_thread;
 789         if (thread_is_running(thread) &&
 790             ooc->ooc_pos_preload >= scrub->os_pos_current)
 791                 return SCRUB_NEXT_EXIT;
 792
 793         rc = osd_iit_iget(info, dev,
 794                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
 795                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
 796                           ooc->ooc_pos_preload, param->sb, false);
 797         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
 798          * ignore the failure, so it still need to skip the inode next time. */
 799         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
 800         return rc;
 801 }
 802
 803 static inline int
 804 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
 805 {
 806         spin_lock(&scrub->os_lock);
 807         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
 808             !cfs_list_empty(&scrub->os_inconsistent_items) ||
 809             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
 810                 scrub->os_waiting = 0;
 811         else
 812                 scrub->os_waiting = 1;
 813         spin_unlock(&scrub->os_lock);
 814
 815         return !scrub->os_waiting;
 816 }
 817
 818 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
 819                           struct osd_iit_param *param,
 820                           struct osd_idmap_cache *oic, int *noslot, int rc)
 821 {
 822         struct l_wait_info       lwi    = { 0 };
 823         struct osd_scrub        *scrub  = &dev->od_scrub;
 824         struct scrub_file       *sf     = &scrub->os_file;
 825         struct ptlrpc_thread    *thread = &scrub->os_thread;
 826         struct osd_otable_it    *it     = dev->od_otable_it;
 827         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
 828
 829         switch (rc) {
 830         case SCRUB_NEXT_CONTINUE:
 831                 goto next;
 832         case SCRUB_NEXT_WAIT:
 833                 goto wait;
 834         case SCRUB_NEXT_NOSCRUB:
 835                 down_write(&scrub->os_rwsem);
 836                 scrub->os_new_checked++;
 837                 sf->sf_items_noscrub++;
 838                 up_write(&scrub->os_rwsem);
 839                 goto next;
 840         }
 841
 842         rc = osd_scrub_check_update(info, dev, oic, rc);
 843         if (rc != 0)
 844                 return rc;
 845
 846         rc = osd_scrub_checkpoint(scrub);
 847         if (rc != 0) {
 848                 CERROR("%.16s: fail to checkpoint, pos = %u, rc = %d\n",
 849                        LDISKFS_SB(param->sb)->s_es->s_volume_name,
 850                        scrub->os_pos_current, rc);
 851                 /* Continue, as long as the scrub itself can go ahead. */
 852         }
 853
 854         if (scrub->os_in_prior) {
 855                 scrub->os_in_prior = 0;
 856                 return 0;
 857         }
 858
 859 next:
 860         scrub->os_pos_current = param->gbase + ++(param->offset);
 861
 862 wait:
 863         if (it != NULL && it->ooi_waiting &&
 864             ooc->ooc_pos_preload < scrub->os_pos_current) {
 865                 spin_lock(&scrub->os_lock);
 866                 it->ooi_waiting = 0;
 867                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
 868                 spin_unlock(&scrub->os_lock);
 869         }
 870
 871         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
 872                 return 0;
 873
 874         if (osd_scrub_has_window(scrub, ooc)) {
 875                 *noslot = 0;
 876                 return 0;
 877         }
 878
 879         l_wait_event(thread->t_ctl_waitq,
 880                      osd_scrub_wakeup(scrub, it),
 881                      &lwi);
 882
 883         if (osd_scrub_has_window(scrub, ooc))
 884                 *noslot = 0;
 885         else
 886                 *noslot = 1;
 887         return 0;
 888 }
 889
 890 static int osd_preload_exec(struct osd_thread_info *info,
 891                             struct osd_device *dev, struct osd_iit_param *param,
 892                             struct osd_idmap_cache *oic, int *noslot, int rc)
 893 {
 894         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
 895
 896         if (rc == 0) {
 897                 ooc->ooc_cached_items++;
 898                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
 899                                         ~OSD_OTABLE_IT_CACHE_MASK;
 900         }
 901         return rc > 0 ? 0 : rc;
 902 }
 903
 904 #define SCRUB_IT_ALL    1
 905 #define SCRUB_IT_CRASH  2
 906
 907 static int osd_inode_iteration(struct osd_thread_info *info,
 908                                struct osd_device *dev, __u32 max, bool preload)
 909 {
 910         osd_iit_next_policy   next;
 911         osd_iit_exec_policy   exec;
 912         __u32                *pos;
 913         __u32                *count;
 914         struct osd_iit_param  param;
 915         __u32                 limit;
 916         int                   noslot = 0;
 917         int                   rc;
 918         ENTRY;
 919
 920         if (!preload) {
 921                 struct osd_scrub *scrub = &dev->od_scrub;
 922
 923                 next = osd_scrub_next;
 924                 exec = osd_scrub_exec;
 925                 pos = &scrub->os_pos_current;
 926                 count = &scrub->os_new_checked;
 927         } else {
 928                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
 929
 930                 next = osd_preload_next;
 931                 exec = osd_preload_exec;
 932                 pos = &ooc->ooc_pos_preload;
 933                 count = &ooc->ooc_cached_items;
 934         }
 935         param.sb = osd_sb(dev);
 936         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
 937
 938         while (*pos <= limit && *count < max) {
 939                 struct osd_idmap_cache *oic = NULL;
 940
 941                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
 942                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
 943                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
 944                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
 945                 if (param.bitmap == NULL) {
 946                         CERROR("%.16s: fail to read bitmap for %u, "
 947                                "scrub will stop, urgent mode\n",
 948                                LDISKFS_SB(param.sb)->s_es->s_volume_name,
 949                                (__u32)param.bg);
 950                         RETURN(-EIO);
 951                 }
 952
 953                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
 954                        *count < max) {
 955                         rc = next(info, dev, &param, &oic, noslot);
 956                         switch (rc) {
 957                         case SCRUB_NEXT_BREAK:
 958                                 goto next_group;
 959                         case SCRUB_NEXT_EXIT:
 960                                 brelse(param.bitmap);
 961                                 RETURN(0);
 962                         case SCRUB_NEXT_CRASH:
 963                                 brelse(param.bitmap);
 964                                 RETURN(SCRUB_IT_CRASH);
 965                         case SCRUB_NEXT_FATAL:
 966                                 brelse(param.bitmap);
 967                                 RETURN(-EINVAL);
 968                         }
 969
 970                         rc = exec(info, dev, &param, oic, &noslot, rc);
 971                         if (rc != 0) {
 972                                 brelse(param.bitmap);
 973                                 RETURN(rc);
 974                         }
 975
 976                         if (preload && dev->od_otable_it->ooi_stopping) {
 977                                 brelse(param.bitmap);
 978                                 RETURN(0);
 979                         }
 980                 }
 981
 982 next_group:
 983                 brelse(param.bitmap);
 984
 985                 if (preload && dev->od_otable_it->ooi_stopping)
 986                         RETURN(0);
 987         }
 988
 989         if (*pos > limit)
 990                 RETURN(SCRUB_IT_ALL);
 991         RETURN(0);
 992 }
 993
 994 static int osd_otable_it_preload(const struct lu_env *env,
 995                                  struct osd_otable_it *it)
 996 {
 997         struct osd_device       *dev   = it->ooi_dev;
 998         struct osd_scrub        *scrub = &dev->od_scrub;
 999         struct osd_otable_cache *ooc   = &it->ooi_cache;
1000         int                      rc;
1001         ENTRY;
1002
1003         rc = osd_inode_iteration(osd_oti_get(env), dev,
1004                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1005         if (rc == SCRUB_IT_ALL)
1006                 it->ooi_all_cached = 1;
1007
1008         CDEBUG(D_LFSCK, "OSD pre-loaded: max = %u, preload = %u, rc = %d\n",
1009                le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count),
1010                ooc->ooc_pos_preload, rc);
1011
1012         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1013                 scrub->os_waiting = 0;
1014                 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
1015         }
1016
1017         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1018 }
1019
1020 static int osd_scrub_main(void *args)
1021 {
1022         struct lu_env         env;
1023         struct osd_device    *dev    = (struct osd_device *)args;
1024         struct osd_scrub     *scrub  = &dev->od_scrub;
1025         struct ptlrpc_thread *thread = &scrub->os_thread;
1026         struct super_block   *sb     = osd_sb(dev);
1027         int                   rc;
1028         ENTRY;
1029
1030         cfs_daemonize("OI_scrub");
1031         rc = lu_env_init(&env, LCT_DT_THREAD);
1032         if (rc != 0) {
1033                 CERROR("%.16s: OI scrub, fail to init env, rc = %d\n",
1034                        LDISKFS_SB(sb)->s_es->s_volume_name, rc);
1035                 GOTO(noenv, rc);
1036         }
1037
1038         rc = osd_scrub_prep(dev);
1039         if (rc != 0) {
1040                 CERROR("%.16s: OI scrub, fail to scrub prep, rc = %d\n",
1041                        LDISKFS_SB(sb)->s_es->s_volume_name, rc);
1042                 GOTO(out, rc);
1043         }
1044
1045         if (!scrub->os_full_speed) {
1046                 struct l_wait_info lwi = { 0 };
1047                 struct osd_otable_it *it = dev->od_otable_it;
1048                 struct osd_otable_cache *ooc = &it->ooi_cache;
1049
1050                 l_wait_event(thread->t_ctl_waitq,
1051                              it->ooi_user_ready || !thread_is_running(thread),
1052                              &lwi);
1053                 if (unlikely(!thread_is_running(thread)))
1054                         GOTO(post, rc = 0);
1055
1056                 scrub->os_pos_current = ooc->ooc_pos_preload;
1057         }
1058
1059         CDEBUG(D_LFSCK, "OI scrub: flags = 0x%x, pos = %u\n",
1060                scrub->os_start_flags, scrub->os_pos_current);
1061
1062         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1063         if (unlikely(rc == SCRUB_IT_CRASH))
1064                 GOTO(out, rc = -EINVAL);
1065         GOTO(post, rc);
1066
1067 post:
1068         osd_scrub_post(scrub, rc);
1069         CDEBUG(D_LFSCK, "OI scrub: stop, rc = %d, pos = %u\n",
1070                rc, scrub->os_pos_current);
1071
1072 out:
1073         while (!cfs_list_empty(&scrub->os_inconsistent_items)) {
1074                 struct osd_inconsistent_item *oii;
1075
1076                 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
1077                                      struct osd_inconsistent_item, oii_list);
1078                 cfs_list_del_init(&oii->oii_list);
1079                 OBD_FREE_PTR(oii);
1080         }
1081         lu_env_fini(&env);
1082
1083 noenv:
1084         spin_lock(&scrub->os_lock);
1085         thread_set_flags(thread, SVC_STOPPED);
1086         cfs_waitq_broadcast(&thread->t_ctl_waitq);
1087         spin_unlock(&scrub->os_lock);
1088         return rc;
1089 }
1090
1091 /* initial OI scrub */
1092
1093 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1094                          struct dentry *, filldir_t filldir);
1095
1096 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1097                                loff_t offset, __u64 ino, unsigned d_type);
1098
1099 static int
1100 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1101                      struct dentry *dentry, filldir_t filldir);
1102 static int
1103 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1104                   struct dentry *dentry, filldir_t filldir);
1105
1106 static int
1107 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1108                      struct dentry *dentry, filldir_t filldir);
1109
1110 enum osd_lf_flags {
1111         OLF_SCAN_SUBITEMS       = 0x0001,
1112         OLF_HIDE_FID            = 0x0002,
1113         OLF_SHOW_NAME           = 0x0004,
1114 };
1115
1116 struct osd_lf_map {
1117         char            *olm_name;
1118         struct lu_fid    olm_fid;
1119         __u16            olm_flags;
1120         scandir_t        olm_scandir;
1121         filldir_t        olm_filldir;
1122 };
1123
1124 /* Add the new introduced local files in the list in the future. */
1125 static const struct osd_lf_map osd_lf_maps[] = {
1126         /* CATALOGS */
1127         { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1128                 NULL, NULL },
1129
1130         /* CONFIGS */
1131         { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1132                 OLF_SCAN_SUBITEMS, osd_ios_general_scan,
1133                 osd_ios_varfid_fill },
1134
1135         /* NIDTBL_VERSIONS */
1136         { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1137                 osd_ios_general_scan, osd_ios_varfid_fill },
1138
1139         /* PENDING */
1140         { "PENDING", { 0, 0, 0 }, 0, NULL, NULL },
1141
1142         /* ROOT */
1143         { "ROOT", { FID_SEQ_ROOT, 1, 0 },
1144                 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, osd_ios_ROOT_scan, NULL },
1145
1146         /* changelog_catalog */
1147         { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, NULL, NULL },
1148
1149         /* changelog_users */
1150         { CHANGELOG_USERS, { 0, 0, 0 }, 0, NULL, NULL },
1151
1152         /* fld */
1153         { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1154                 NULL, NULL },
1155
1156         /* last_rcvd */
1157         { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1158                 NULL, NULL },
1159
1160         /* lfsck_bookmark */
1161         { "lfsck_bookmark", { 0, 0, 0 }, 0, NULL, NULL },
1162
1163         /* lov_objid */
1164         { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1165                 NULL, NULL },
1166
1167         /* lov_objseq */
1168         { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1169                 OLF_SHOW_NAME, NULL, NULL },
1170
1171         /* quota_master */
1172         { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1173                 osd_ios_general_scan, osd_ios_varfid_fill },
1174
1175         /* quota_slave */
1176         { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1177                 osd_ios_general_scan, osd_ios_varfid_fill },
1178
1179         /* seq_ctl */
1180         { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1181                 OLF_SHOW_NAME, NULL, NULL },
1182
1183         /* seq_srv */
1184         { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1185                 OLF_SHOW_NAME, NULL, NULL },
1186
1187         /* health_check */
1188         { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1189                 OLF_SHOW_NAME, NULL, NULL },
1190
1191         /* lfsck_namespace */
1192         { "lfsck_namespace", { 0, 0, 0 }, 0, NULL, NULL },
1193
1194         /* OBJECTS, upgrade from old device */
1195         { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, osd_ios_OBJECTS_scan, NULL },
1196
1197         /* lquota_v2.user, upgrade from old device */
1198         { "lquota_v2.user", { 0, 0, 0 }, 0, NULL, NULL },
1199
1200         /* lquota_v2.group, upgrade from old device */
1201         { "lquota_v2.group", { 0, 0, 0 }, 0, NULL, NULL },
1202
1203         /* LAST_GROUP, upgrade from old device */
1204         { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1205                 OLF_SHOW_NAME, NULL, NULL },
1206
1207         { NULL, { 0, 0, 0 }, 0, NULL, NULL }
1208 };
1209
1210 struct osd_ios_item {
1211         cfs_list_t       oii_list;
1212         struct dentry   *oii_dentry;
1213         scandir_t        oii_scandir;
1214         filldir_t        oii_filldir;
1215 };
1216
1217 struct osd_ios_filldir_buf {
1218         struct osd_thread_info  *oifb_info;
1219         struct osd_device       *oifb_dev;
1220         struct dentry           *oifb_dentry;
1221 };
1222
1223 static inline struct dentry *
1224 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1225 {
1226         struct dentry *dentry;
1227
1228         dentry = ll_lookup_one_len(name, parent, namelen);
1229         if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1230                 dput(dentry);
1231                 return ERR_PTR(-ENOENT);
1232         }
1233
1234         return dentry;
1235 }
1236
1237 static inline void
1238 osd_ios_llogname2fid(struct lu_fid *fid, const char *name, int namelen)
1239 {
1240         obd_id id = 0;
1241         int    i  = 0;
1242
1243         fid->f_seq = FID_SEQ_LLOG;
1244         while (i < namelen)
1245                 id = id * 10 + name[i++] - '0';
1246
1247         fid->f_oid = id & 0x00000000ffffffffULL;
1248         fid->f_ver = id >> 32;
1249 }
1250
1251 static inline void
1252 osd_ios_Oname2fid(struct lu_fid *fid, const char *name, int namelen)
1253 {
1254         __u64 seq = 0;
1255         int   i   = 0;
1256
1257         while (i < namelen)
1258                 seq = seq * 10 + name[i++] - '0';
1259
1260         lu_last_id_fid(fid, seq);
1261 }
1262
1263 static int
1264 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1265                  scandir_t scandir, filldir_t filldir)
1266 {
1267         struct osd_ios_item *item;
1268
1269         OBD_ALLOC_PTR(item);
1270         if (item == NULL)
1271                 return -ENOMEM;
1272
1273         CFS_INIT_LIST_HEAD(&item->oii_list);
1274         item->oii_dentry = dget(dentry);
1275         item->oii_scandir = scandir;
1276         item->oii_filldir = filldir;
1277         cfs_list_add_tail(&item->oii_list, &dev->od_ios_list);
1278         return 0;
1279 }
1280
1281 /**
1282  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1283  *
1284  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1285  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1286  * reference the inode, or fixed if it is missing or references another inode.
1287  */
1288 static int
1289 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1290                  struct inode *inode, const struct lu_fid *fid, int flags)
1291 {
1292         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
1293         struct osd_inode_id     *id     = &info->oti_id;
1294         struct osd_inode_id     *id2    = &info->oti_id2;
1295         struct osd_scrub        *scrub  = &dev->od_scrub;
1296         struct scrub_file       *sf     = &scrub->os_file;
1297         struct lu_fid            tfid;
1298         int                      rc;
1299         ENTRY;
1300
1301         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1302         if (rc != 0 && rc != -ENODATA)
1303                 RETURN(rc);
1304
1305         osd_id_gen(id, inode->i_ino, inode->i_generation);
1306         if (rc == -ENODATA) {
1307                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID)
1308                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1309                 else
1310                         tfid = *fid;
1311                 rc = osd_ea_fid_set(info, inode, &tfid, 0);
1312                 if (rc != 0)
1313                         RETURN(rc);
1314         } else {
1315                 tfid = lma->lma_self_fid;
1316         }
1317
1318         rc = __osd_oi_lookup(info, dev, &tfid, id2);
1319         if (rc != 0) {
1320                 if (rc != -ENOENT)
1321                         RETURN(rc);
1322
1323                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1324                                                DTO_INDEX_INSERT);
1325                 RETURN(rc);
1326         }
1327
1328         if (osd_id_eq_strict(id, id2))
1329                 RETURN(0);
1330
1331         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1332                 osd_scrub_file_reset(scrub,
1333                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1334                                      SF_INCONSISTENT);
1335                 rc = osd_scrub_file_store(scrub);
1336                 if (rc != 0)
1337                         RETURN(rc);
1338         }
1339
1340         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id, DTO_INDEX_UPDATE);
1341
1342         RETURN(rc);
1343 }
1344
1345 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1346                                loff_t offset, __u64 ino, unsigned d_type)
1347 {
1348         struct osd_ios_filldir_buf *fill_buf = buf;
1349         struct osd_device          *dev      = fill_buf->oifb_dev;
1350         struct dentry              *child;
1351         int                         rc;
1352         ENTRY;
1353
1354         /* skip any '.' started names */
1355         if (name[0] == '.')
1356                 RETURN(0);
1357
1358         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1359         if (IS_ERR(child))
1360                 RETURN(PTR_ERR(child));
1361
1362         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1363                               NULL, 0);
1364         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1365                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1366                                       osd_ios_varfid_fill);
1367         dput(child);
1368
1369         RETURN(rc);
1370 }
1371
1372 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
1373                              loff_t offset, __u64 ino, unsigned d_type)
1374 {
1375         struct osd_ios_filldir_buf *fill_buf = buf;
1376         struct osd_device          *dev      = fill_buf->oifb_dev;
1377         const struct osd_lf_map    *map;
1378         struct dentry              *child;
1379         int                         rc       = 0;
1380         ENTRY;
1381
1382         /* skip any '.' started names */
1383         if (name[0] == '.')
1384                 RETURN(0);
1385
1386         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
1387                 if (strlen(map->olm_name) != namelen)
1388                         continue;
1389
1390                 if (strncmp(map->olm_name, name, namelen) == 0)
1391                         break;
1392         }
1393
1394         if (map->olm_name == NULL)
1395                 RETURN(0);
1396
1397         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1398         if (IS_ERR(child))
1399                 RETURN(PTR_ERR(child));
1400
1401         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1402                               &map->olm_fid, map->olm_flags);
1403         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
1404                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
1405                                       map->olm_filldir);
1406         dput(child);
1407
1408         RETURN(rc);
1409 }
1410
1411 static int
1412 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1413                      struct dentry *dentry, filldir_t filldir)
1414 {
1415         struct osd_ios_filldir_buf    buf   = { info, dev, dentry };
1416         struct file                  *filp  = &info->oti_it_ea.oie_file;
1417         struct inode                 *inode = dentry->d_inode;
1418         const struct file_operations *fops  = inode->i_fop;
1419         int                           rc;
1420         ENTRY;
1421
1422         LASSERT(filldir != NULL);
1423
1424         filp->f_pos = 0;
1425         filp->f_dentry = dentry;
1426         filp->f_mode = FMODE_64BITHASH;
1427         filp->f_mapping = inode->i_mapping;
1428         filp->f_op = fops;
1429         filp->private_data = NULL;
1430
1431         rc = fops->readdir(filp, &buf, filldir);
1432         fops->release(inode, filp);
1433
1434         RETURN(rc);
1435 }
1436
1437 static int
1438 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1439                   struct dentry *dentry, filldir_t filldir)
1440 {
1441         struct osd_scrub  *scrub  = &dev->od_scrub;
1442         struct scrub_file *sf     = &scrub->os_file;
1443         struct dentry     *child;
1444         int                rc;
1445         ENTRY;
1446
1447         /* It is existing MDT device. */
1448         dev->od_handle_nolma = 1;
1449         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
1450                                        strlen(dot_lustre_name));
1451         if (IS_ERR(child)) {
1452                 rc = PTR_ERR(child);
1453                 if (rc == -ENOENT) {
1454                         /* It is 1.8 MDT device. */
1455                         if (!(sf->sf_flags & SF_UPGRADE)) {
1456                                 osd_scrub_file_reset(scrub,
1457                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1458                                         SF_UPGRADE);
1459                                 rc = osd_scrub_file_store(scrub);
1460                         } else {
1461                                 rc = 0;
1462                         }
1463                 }
1464         } else {
1465                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
1466                  * so the client will get IGIF for the ".lustre" object when
1467                  * the MDT restart.
1468                  *
1469                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
1470                  * it does not know whether there are some old clients cached
1471                  * the ".lustre" IGIF during the upgrading. Two choices:
1472                  *
1473                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
1474                  *    It will allow the old connected clients to access the
1475                  *    ".lustre" with cached IGIF. But it will cause others
1476                  *    on the MDT failed to check "fid_is_dot_lustre()".
1477                  *
1478                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
1479                  *    for ".lustre" in spite of whether there are some clients
1480                  *    cached the ".lustre" IGIF or not. It enables the check
1481                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
1482                  *    that the old connected clients cannot access the ".lustre"
1483                  *    with the cached IGIF.
1484                  *
1485                  * Usually, it is rare case for the old connected clients
1486                  * to access the ".lustre" with cached IGIF. So we prefer
1487                  * to the solution 2). */
1488                 rc = osd_ios_scan_one(info, dev, child->d_inode,
1489                                       &LU_DOT_LUSTRE_FID, 0);
1490                 dput(child);
1491         }
1492
1493         RETURN(rc);
1494 }
1495
1496 static int
1497 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1498                      struct dentry *dentry, filldir_t filldir)
1499 {
1500         struct dentry *child;
1501         int            rc;
1502         ENTRY;
1503
1504         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
1505         if (!IS_ERR(child)) {
1506                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1507                 dput(child);
1508         } else {
1509                 rc = PTR_ERR(child);
1510         }
1511
1512         if (rc != 0 && rc != -ENOENT)
1513                 RETURN(rc);
1514
1515         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
1516         if (!IS_ERR(child)) {
1517                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1518                 dput(child);
1519         } else {
1520                 rc = PTR_ERR(child);
1521         }
1522
1523         if (rc == -ENOENT)
1524                 rc = 0;
1525
1526         RETURN(rc);
1527 }
1528
1529 static int osd_initial_OI_scrub(struct osd_thread_info *info,
1530                                 struct osd_device *dev)
1531 {
1532         struct osd_ios_item *item    = NULL;
1533         scandir_t            scandir = osd_ios_general_scan;
1534         filldir_t            filldir = osd_ios_root_fill;
1535         struct dentry       *dentry  = osd_sb(dev)->s_root;
1536         int                  rc;
1537         ENTRY;
1538
1539         while (1) {
1540                 rc = scandir(info, dev, dentry, filldir);
1541                 if (item != NULL) {
1542                         dput(item->oii_dentry);
1543                         OBD_FREE_PTR(item);
1544                 }
1545
1546                 if (rc != 0)
1547                         break;
1548
1549                 if (cfs_list_empty(&dev->od_ios_list))
1550                         break;
1551
1552                 item = cfs_list_entry(dev->od_ios_list.next,
1553                                       struct osd_ios_item, oii_list);
1554                 cfs_list_del_init(&item->oii_list);
1555
1556                 LASSERT(item->oii_scandir != NULL);
1557                 scandir = item->oii_scandir;
1558                 filldir = item->oii_filldir;
1559                 dentry = item->oii_dentry;
1560         }
1561
1562         while (!cfs_list_empty(&dev->od_ios_list)) {
1563                 item = cfs_list_entry(dev->od_ios_list.next,
1564                                       struct osd_ios_item, oii_list);
1565                 cfs_list_del_init(&item->oii_list);
1566                 dput(item->oii_dentry);
1567                 OBD_FREE_PTR(item);
1568         }
1569
1570         RETURN(rc);
1571 }
1572
1573 char *osd_lf_fid2name(const struct lu_fid *fid)
1574 {
1575         const struct osd_lf_map *map = osd_lf_maps;
1576
1577         while (map->olm_name != NULL) {
1578                 if (!lu_fid_eq(fid, &map->olm_fid)) {
1579                         map++;
1580                         continue;
1581                 }
1582
1583                 if (map->olm_flags & OLF_SHOW_NAME)
1584                         return map->olm_name;
1585                 else
1586                         return "";
1587         }
1588
1589         return NULL;
1590 }
1591
1592 /* OI scrub start/stop */
1593
1594 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
1595 {
1596         struct osd_scrub     *scrub  = &dev->od_scrub;
1597         struct ptlrpc_thread *thread = &scrub->os_thread;
1598         struct l_wait_info    lwi    = { 0 };
1599         int                   rc;
1600         ENTRY;
1601
1602 again:
1603         /* os_lock: sync status between stop and scrub thread */
1604         spin_lock(&scrub->os_lock);
1605         if (thread_is_running(thread)) {
1606                 spin_unlock(&scrub->os_lock);
1607                 RETURN(-EALREADY);
1608         } else if (unlikely(thread_is_stopping(thread))) {
1609                 spin_unlock(&scrub->os_lock);
1610                 l_wait_event(thread->t_ctl_waitq,
1611                              thread_is_stopped(thread),
1612                              &lwi);
1613                 goto again;
1614         }
1615         spin_unlock(&scrub->os_lock);
1616
1617         if (scrub->os_file.sf_status == SS_COMPLETED)
1618                 flags |= SS_RESET;
1619
1620         scrub->os_start_flags = flags;
1621         thread_set_flags(thread, 0);
1622         rc = cfs_create_thread(osd_scrub_main, dev, 0);
1623         if (rc < 0) {
1624                 CERROR("%.16s: cannot start iteration thread, rc = %d\n",
1625                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
1626                 RETURN(rc);
1627         }
1628
1629         l_wait_event(thread->t_ctl_waitq,
1630                      thread_is_running(thread) || thread_is_stopped(thread),
1631                      &lwi);
1632
1633         RETURN(0);
1634 }
1635
1636 int osd_scrub_start(struct osd_device *dev)
1637 {
1638         int rc;
1639         ENTRY;
1640
1641         /* od_otable_mutex: prevent curcurrent start/stop */
1642         mutex_lock(&dev->od_otable_mutex);
1643         rc = do_osd_scrub_start(dev, SS_AUTO);
1644         mutex_unlock(&dev->od_otable_mutex);
1645
1646         RETURN(rc == -EALREADY ? 0 : rc);
1647 }
1648
1649 static void do_osd_scrub_stop(struct osd_scrub *scrub)
1650 {
1651         struct ptlrpc_thread *thread = &scrub->os_thread;
1652         struct l_wait_info    lwi    = { 0 };
1653
1654         /* os_lock: sync status between stop and scrub thread */
1655         spin_lock(&scrub->os_lock);
1656         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
1657                 thread_set_flags(thread, SVC_STOPPING);
1658                 spin_unlock(&scrub->os_lock);
1659                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
1660                 l_wait_event(thread->t_ctl_waitq,
1661                              thread_is_stopped(thread),
1662                              &lwi);
1663                 /* Do not skip the last lock/unlock, which can guarantee that
1664                  * the caller cannot return until the OI scrub thread exit. */
1665                 spin_lock(&scrub->os_lock);
1666         }
1667         spin_unlock(&scrub->os_lock);
1668 }
1669
1670 static void osd_scrub_stop(struct osd_device *dev)
1671 {
1672         /* od_otable_mutex: prevent curcurrent start/stop */
1673         mutex_lock(&dev->od_otable_mutex);
1674         dev->od_scrub.os_paused = 1;
1675         do_osd_scrub_stop(&dev->od_scrub);
1676         mutex_unlock(&dev->od_otable_mutex);
1677 }
1678
1679 /* OI scrub setup/cleanup */
1680
1681 static const char osd_scrub_name[] = "OI_scrub";
1682
1683 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
1684 {
1685         struct osd_thread_info     *info   = osd_oti_get(env);
1686         struct osd_scrub           *scrub  = &dev->od_scrub;
1687         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
1688         struct scrub_file          *sf     = &scrub->os_file;
1689         struct super_block         *sb     = osd_sb(dev);
1690         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
1691         struct lvfs_run_ctxt        saved;
1692         struct file                *filp;
1693         int                         dirty  = 0;
1694         int                         rc     = 0;
1695         ENTRY;
1696
1697         memset(scrub, 0, sizeof(*scrub));
1698         OBD_SET_CTXT_MAGIC(ctxt);
1699         ctxt->pwdmnt = dev->od_mnt;
1700         ctxt->pwd = dev->od_mnt->mnt_root;
1701         ctxt->fs = get_ds();
1702
1703         cfs_waitq_init(&scrub->os_thread.t_ctl_waitq);
1704         init_rwsem(&scrub->os_rwsem);
1705         spin_lock_init(&scrub->os_lock);
1706         CFS_INIT_LIST_HEAD(&scrub->os_inconsistent_items);
1707
1708         push_ctxt(&saved, ctxt, NULL);
1709         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
1710         if (IS_ERR(filp))
1711                 RETURN(PTR_ERR(filp));
1712
1713         scrub->os_inode = igrab(filp->f_dentry->d_inode);
1714         filp_close(filp, 0);
1715         pop_ctxt(&saved, ctxt, NULL);
1716         ldiskfs_set_inode_state(scrub->os_inode,
1717                                 LDISKFS_STATE_LUSTRE_NO_OI);
1718
1719         rc = osd_scrub_file_load(scrub);
1720         if (rc == -ENOENT) {
1721                 osd_scrub_file_init(scrub, es->s_uuid);
1722                 dirty = 1;
1723         } else if (rc != 0) {
1724                 RETURN(rc);
1725         } else {
1726                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
1727                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
1728                         dirty = 1;
1729                 } else if (sf->sf_status == SS_SCANNING) {
1730                         sf->sf_status = SS_CRASHED;
1731                         dirty = 1;
1732                 }
1733         }
1734
1735         if (sf->sf_pos_last_checkpoint != 0)
1736                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
1737         else
1738                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
1739
1740         if (dirty != 0) {
1741                 rc = osd_scrub_file_store(scrub);
1742                 if (rc != 0)
1743                         RETURN(rc);
1744         }
1745
1746         /* Initialize OI files. */
1747         rc = osd_oi_init(info, dev);
1748         if (rc < 0)
1749                 RETURN(rc);
1750
1751         rc = osd_initial_OI_scrub(info, dev);
1752         if (rc == 0) {
1753                 if ((sf->sf_flags & SF_UPGRADE) &&
1754                    !(sf->sf_flags & SF_INCONSISTENT))
1755                         /* The 'od_igif_inoi' will be set after the
1756                          * upgrading completed, needs NOT remount. */
1757                         dev->od_igif_inoi = 0;
1758                 else
1759                         /* The 'od_igif_inoi' will be set under the
1760                          * following cases:
1761                          * 1) new created system, or
1762                          * 2) restored from file-level backup, or
1763                          * 3) the upgrading completed.
1764                          *
1765                          * The 'od_igif_inoi' may be cleared by OI scrub
1766                          * later if found that the system is upgrading. */
1767                         dev->od_igif_inoi = 1;
1768
1769                 if (!dev->od_noscrub &&
1770                     ((sf->sf_status == SS_PAUSED) ||
1771                      (sf->sf_status == SS_CRASHED &&
1772                       sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
1773                                       SF_UPGRADE | SF_AUTO)) ||
1774                      (sf->sf_status == SS_INIT &&
1775                       sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
1776                                       SF_UPGRADE))))
1777                         rc = osd_scrub_start(dev);
1778         }
1779
1780         RETURN(rc);
1781 }
1782
1783 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
1784 {
1785         struct osd_scrub *scrub = &dev->od_scrub;
1786
1787         LASSERT(dev->od_otable_it == NULL);
1788
1789         if (scrub->os_inode != NULL) {
1790                 osd_scrub_stop(dev);
1791                 iput(scrub->os_inode);
1792                 scrub->os_inode = NULL;
1793         }
1794         if (dev->od_oi_table != NULL)
1795                 osd_oi_fini(osd_oti_get(env), dev);
1796 }
1797
1798 /* object table based iteration APIs */
1799
1800 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
1801                                        struct dt_object *dt, __u32 attr,
1802                                        struct lustre_capa *capa)
1803 {
1804         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
1805         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
1806         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
1807         struct osd_scrub       *scrub = &dev->od_scrub;
1808         struct osd_otable_it   *it;
1809         __u32                   start = 0;
1810         int                     rc;
1811         ENTRY;
1812
1813         /* od_otable_mutex: prevent curcurrent init/fini */
1814         mutex_lock(&dev->od_otable_mutex);
1815         if (dev->od_otable_it != NULL)
1816                 GOTO(out, it = ERR_PTR(-EALREADY));
1817
1818         OBD_ALLOC_PTR(it);
1819         if (it == NULL)
1820                 GOTO(out, it = ERR_PTR(-ENOMEM));
1821
1822         dev->od_otable_it = it;
1823         it->ooi_dev = dev;
1824         it->ooi_pid = cfs_curproc_pid();
1825         it->ooi_cache.ooc_consumer_idx = -1;
1826         if (flags & DOIF_OUTUSED)
1827                 it->ooi_used_outside = 1;
1828
1829         if (flags & DOIF_RESET)
1830                 start |= SS_RESET;
1831
1832         if (valid & DOIV_ERROR_HANDLE) {
1833                 if (flags & DOIF_FAILOUT)
1834                         start |= SS_SET_FAILOUT;
1835                 else
1836                         start |= SS_CLEAR_FAILOUT;
1837         }
1838
1839         rc = do_osd_scrub_start(dev, start);
1840         if (rc < 0 && rc != -EALREADY) {
1841                 dev->od_otable_it = NULL;
1842                 OBD_FREE_PTR(it);
1843                 GOTO(out, it = ERR_PTR(rc));
1844         }
1845
1846         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
1847
1848         GOTO(out, it);
1849
1850 out:
1851         mutex_unlock(&dev->od_otable_mutex);
1852         return (struct dt_it *)it;
1853 }
1854
1855 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
1856 {
1857         struct osd_otable_it *it  = (struct osd_otable_it *)di;
1858         struct osd_device    *dev = it->ooi_dev;
1859
1860         /* od_otable_mutex: prevent curcurrent init/fini */
1861         mutex_lock(&dev->od_otable_mutex);
1862         do_osd_scrub_stop(&dev->od_scrub);
1863         LASSERT(dev->od_otable_it == it);
1864
1865         dev->od_otable_it = NULL;
1866         mutex_unlock(&dev->od_otable_mutex);
1867         OBD_FREE_PTR(it);
1868 }
1869
1870 static int osd_otable_it_get(const struct lu_env *env,
1871                              struct dt_it *di, const struct dt_key *key)
1872 {
1873         return 0;
1874 }
1875
1876 /**
1877  * It is hack here:
1878  *
1879  * Sometimes the otable-based iteration driver (LFSCK) may be blocked in OSD
1880  * layer when someone wants to stop/pause the iteration. Under such case, we
1881  * need some mechanism to notify the event and wakeup the blocker.
1882  */
1883 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
1884 {
1885         struct osd_otable_it *it  = (struct osd_otable_it *)di;
1886         struct osd_device    *dev = it->ooi_dev;
1887
1888         /* od_otable_mutex: prevent curcurrent init/fini */
1889         mutex_lock(&dev->od_otable_mutex);
1890         if (it->ooi_pid == cfs_curproc_pid()) {
1891                 dev->od_scrub.os_paused = 1;
1892         } else {
1893                 struct ptlrpc_thread *thread = &dev->od_scrub.os_thread;
1894
1895                 it->ooi_stopping = 1;
1896                 if (it->ooi_waiting)
1897                         cfs_waitq_broadcast(&thread->t_ctl_waitq);
1898         }
1899         mutex_unlock(&dev->od_otable_mutex);
1900 }
1901
1902 static inline int
1903 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1904 {
1905         spin_lock(&scrub->os_lock);
1906         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
1907             scrub->os_waiting || it->ooi_stopping ||
1908             !thread_is_running(&scrub->os_thread))
1909                 it->ooi_waiting = 0;
1910         else
1911                 it->ooi_waiting = 1;
1912         spin_unlock(&scrub->os_lock);
1913
1914         return !it->ooi_waiting;
1915 }
1916
1917 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
1918 {
1919         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
1920         struct osd_device       *dev    = it->ooi_dev;
1921         struct osd_scrub        *scrub  = &dev->od_scrub;
1922         struct osd_otable_cache *ooc    = &it->ooi_cache;
1923         struct ptlrpc_thread    *thread = &scrub->os_thread;
1924         struct l_wait_info       lwi    = { 0 };
1925         int                      rc;
1926         ENTRY;
1927
1928         LASSERT(it->ooi_user_ready);
1929
1930 again:
1931         if (!thread_is_running(thread) && !it->ooi_used_outside)
1932                 RETURN(1);
1933
1934         if (ooc->ooc_cached_items > 0) {
1935                 ooc->ooc_cached_items--;
1936                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
1937                                         ~OSD_OTABLE_IT_CACHE_MASK;
1938                 RETURN(0);
1939         }
1940
1941         if (it->ooi_all_cached) {
1942                 l_wait_event(thread->t_ctl_waitq,
1943                              !thread_is_running(thread),
1944                              &lwi);
1945                 RETURN(1);
1946         }
1947
1948         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1949                 spin_lock(&scrub->os_lock);
1950                 scrub->os_waiting = 0;
1951                 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
1952                 spin_unlock(&scrub->os_lock);
1953         }
1954
1955         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
1956                 l_wait_event(thread->t_ctl_waitq,
1957                              osd_otable_it_wakeup(scrub, it),
1958                              &lwi);
1959
1960         if (!thread_is_running(thread) && !it->ooi_used_outside)
1961                 RETURN(1);
1962
1963         if (it->ooi_stopping)
1964                 RETURN(0);
1965
1966         rc = osd_otable_it_preload(env, it);
1967         if (rc >= 0)
1968                 goto again;
1969
1970         RETURN(rc);
1971 }
1972
1973 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
1974                                         const struct dt_it *di)
1975 {
1976         return NULL;
1977 }
1978
1979 static int osd_otable_it_key_size(const struct lu_env *env,
1980                                   const struct dt_it *di)
1981 {
1982         return sizeof(__u64);
1983 }
1984
1985 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
1986                              struct dt_rec *rec, __u32 attr)
1987 {
1988         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
1989         struct osd_otable_cache *ooc = &it->ooi_cache;
1990
1991         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
1992
1993         /* Filter out Invald FID already. */
1994         LASSERTF(fid_is_sane((struct lu_fid *)rec),
1995                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
1996                  PFID((struct lu_fid *)rec),
1997                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
1998
1999         return 0;
2000 }
2001
2002 static __u64 osd_otable_it_store(const struct lu_env *env,
2003                                  const struct dt_it *di)
2004 {
2005         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2006         struct osd_otable_cache *ooc = &it->ooi_cache;
2007         __u64                    hash;
2008
2009         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2010                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2011         else
2012                 hash = ooc->ooc_pos_preload;
2013         return hash;
2014 }
2015
2016 /**
2017  * Set the OSD layer iteration start position as the specified hash.
2018  */
2019 static int osd_otable_it_load(const struct lu_env *env,
2020                               const struct dt_it *di, __u64 hash)
2021 {
2022         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
2023         struct osd_device       *dev   = it->ooi_dev;
2024         struct osd_otable_cache *ooc   = &it->ooi_cache;
2025         struct osd_scrub        *scrub = &dev->od_scrub;
2026         int                      rc;
2027         ENTRY;
2028
2029         /* Forbid to set iteration position after iteration started. */
2030         if (it->ooi_user_ready)
2031                 RETURN(-EPERM);
2032
2033         if (hash > OSD_OTABLE_MAX_HASH)
2034                 hash = OSD_OTABLE_MAX_HASH;
2035
2036         ooc->ooc_pos_preload = hash;
2037         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2038                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2039
2040         it->ooi_user_ready = 1;
2041         if (!scrub->os_full_speed)
2042                 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
2043
2044         /* Unplug OSD layer iteration by the first next() call. */
2045         rc = osd_otable_it_next(env, (struct dt_it *)it);
2046
2047         RETURN(rc);
2048 }
2049
2050 static int osd_otable_it_key_rec(const struct lu_env *env,
2051                                  const struct dt_it *di, void *key_rec)
2052 {
2053         return 0;
2054 }
2055
2056 const struct dt_index_operations osd_otable_ops = {
2057         .dio_it = {
2058                 .init     = osd_otable_it_init,
2059                 .fini     = osd_otable_it_fini,
2060                 .get      = osd_otable_it_get,
2061                 .put      = osd_otable_it_put,
2062                 .next     = osd_otable_it_next,
2063                 .key      = osd_otable_it_key,
2064                 .key_size = osd_otable_it_key_size,
2065                 .rec      = osd_otable_it_rec,
2066                 .store    = osd_otable_it_store,
2067                 .load     = osd_otable_it_load,
2068                 .key_rec  = osd_otable_it_key_rec,
2069         }
2070 };
2071
2072 /* high priority inconsistent items list APIs */
2073
2074 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2075                    int insert)
2076 {
2077         struct osd_inconsistent_item *oii;
2078         struct osd_scrub             *scrub  = &dev->od_scrub;
2079         struct ptlrpc_thread         *thread = &scrub->os_thread;
2080         int                           wakeup = 0;
2081         ENTRY;
2082
2083         OBD_ALLOC_PTR(oii);
2084         if (unlikely(oii == NULL))
2085                 RETURN(-ENOMEM);
2086
2087         CFS_INIT_LIST_HEAD(&oii->oii_list);
2088         oii->oii_cache = *oic;
2089         oii->oii_insert = insert;
2090
2091         spin_lock(&scrub->os_lock);
2092         if (unlikely(!thread_is_running(thread))) {
2093                 spin_unlock(&scrub->os_lock);
2094                 OBD_FREE_PTR(oii);
2095                 RETURN(-EAGAIN);
2096         }
2097
2098         if (cfs_list_empty(&scrub->os_inconsistent_items))
2099                 wakeup = 1;
2100         cfs_list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2101         spin_unlock(&scrub->os_lock);
2102
2103         if (wakeup != 0)
2104                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
2105
2106         RETURN(0);
2107 }
2108
2109 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2110                    struct osd_inode_id *id)
2111 {
2112         struct osd_scrub             *scrub = &dev->od_scrub;
2113         struct osd_inconsistent_item *oii;
2114         ENTRY;
2115
2116         spin_lock(&scrub->os_lock);
2117         cfs_list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2118                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2119                         *id = oii->oii_cache.oic_lid;
2120                         spin_unlock(&scrub->os_lock);
2121                         RETURN(0);
2122                 }
2123         }
2124         spin_unlock(&scrub->os_lock);
2125
2126         RETURN(-ENOENT);
2127 }
2128
2129 /* OI scrub dump */
2130
2131 static const char *scrub_status_names[] = {
2132         "init",
2133         "scanning",
2134         "completed",
2135         "failed",
2136         "stopped",
2137         "paused",
2138         "crashed",
2139         NULL
2140 };
2141
2142 static const char *scrub_flags_names[] = {
2143         "recreated",
2144         "inconsistent",
2145         "auto",
2146         "upgrade",
2147         NULL
2148 };
2149
2150 static const char *scrub_param_names[] = {
2151         "failout",
2152         NULL
2153 };
2154
2155 static int scrub_bits_dump(char **buf, int *len, int bits, const char *names[],
2156                            const char *prefix)
2157 {
2158         int save = *len;
2159         int flag;
2160         int rc;
2161         int i;
2162
2163         rc = snprintf(*buf, *len, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2164         if (rc <= 0)
2165                 return -ENOSPC;
2166
2167         *buf += rc;
2168         *len -= rc;
2169         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2170                 if (flag & bits) {
2171                         bits &= ~flag;
2172                         rc = snprintf(*buf, *len, "%s%c", names[i],
2173                                       bits != 0 ? ',' : '\n');
2174                         if (rc <= 0)
2175                                 return -ENOSPC;
2176
2177                         *buf += rc;
2178                         *len -= rc;
2179                 }
2180         }
2181         return save - *len;
2182 }
2183
2184 static int scrub_time_dump(char **buf, int *len, __u64 time, const char *prefix)
2185 {
2186         int rc;
2187
2188         if (time != 0)
2189                 rc = snprintf(*buf, *len, "%s: "LPU64" seconds\n", prefix,
2190                               cfs_time_current_sec() - time);
2191         else
2192                 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
2193         if (rc <= 0)
2194                 return -ENOSPC;
2195
2196         *buf += rc;
2197         *len -= rc;
2198         return rc;
2199 }
2200
2201 static int scrub_pos_dump(char **buf, int *len, __u64 pos, const char *prefix)
2202 {
2203         int rc;
2204
2205         if (pos != 0)
2206                 rc = snprintf(*buf, *len, "%s: "LPU64"\n", prefix, pos);
2207         else
2208                 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
2209         if (rc <= 0)
2210                 return -ENOSPC;
2211
2212         *buf += rc;
2213         *len -= rc;
2214         return rc;
2215 }
2216
2217 int osd_scrub_dump(struct osd_device *dev, char *buf, int len)
2218 {
2219         struct osd_scrub  *scrub   = &dev->od_scrub;
2220         struct scrub_file *sf      = &scrub->os_file;
2221         __u64              checked;
2222         __u64              speed;
2223         int                save    = len;
2224         int                ret     = -ENOSPC;
2225         int                rc;
2226
2227         down_read(&scrub->os_rwsem);
2228         rc = snprintf(buf, len,
2229                       "name: OI_scrub\n"
2230                       "magic: 0x%x\n"
2231                       "oi_files: %d\n"
2232                       "status: %s\n",
2233                       sf->sf_magic, (int)sf->sf_oi_count,
2234                       scrub_status_names[sf->sf_status]);
2235         if (rc <= 0)
2236                 goto out;
2237
2238         buf += rc;
2239         len -= rc;
2240         rc = scrub_bits_dump(&buf, &len, sf->sf_flags, scrub_flags_names,
2241                              "flags");
2242         if (rc < 0)
2243                 goto out;
2244
2245         rc = scrub_bits_dump(&buf, &len, sf->sf_param, scrub_param_names,
2246                              "param");
2247         if (rc < 0)
2248                 goto out;
2249
2250         rc = scrub_time_dump(&buf, &len, sf->sf_time_last_complete,
2251                              "time_since_last_completed");
2252         if (rc < 0)
2253                 goto out;
2254
2255         rc = scrub_time_dump(&buf, &len, sf->sf_time_latest_start,
2256                              "time_since_latest_start");
2257         if (rc < 0)
2258                 goto out;
2259
2260         rc = scrub_time_dump(&buf, &len, sf->sf_time_last_checkpoint,
2261                              "time_since_last_checkpoint");
2262         if (rc < 0)
2263                 goto out;
2264
2265         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_latest_start,
2266                             "latest_start_position");
2267         if (rc < 0)
2268                 goto out;
2269
2270         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_last_checkpoint,
2271                             "last_checkpoint_position");
2272         if (rc < 0)
2273                 goto out;
2274
2275         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_first_inconsistent,
2276                             "first_failure_position");
2277         if (rc < 0)
2278                 goto out;
2279
2280         checked = sf->sf_items_checked + scrub->os_new_checked;
2281         rc = snprintf(buf, len,
2282                       "checked: "LPU64"\n"
2283                       "updated: "LPU64"\n"
2284                       "failed: "LPU64"\n"
2285                       "prior_updated: "LPU64"\n"
2286                       "noscrub: "LPU64"\n"
2287                       "igif: "LPU64"\n"
2288                       "success_count: %u\n",
2289                       checked, sf->sf_items_updated, sf->sf_items_failed,
2290                       sf->sf_items_updated_prior, sf->sf_items_noscrub,
2291                       sf->sf_items_igif, sf->sf_success_count);
2292         if (rc <= 0)
2293                 goto out;
2294
2295         buf += rc;
2296         len -= rc;
2297         speed = checked;
2298         if (thread_is_running(&scrub->os_thread)) {
2299                 cfs_duration_t duration = cfs_time_current() -
2300                                           scrub->os_time_last_checkpoint;
2301                 __u64 new_checked = scrub->os_new_checked * CFS_HZ;
2302                 __u32 rtime = sf->sf_run_time +
2303                               cfs_duration_sec(duration + HALF_SEC);
2304
2305                 if (duration != 0)
2306                         do_div(new_checked, duration);
2307                 if (rtime != 0)
2308                         do_div(speed, rtime);
2309                 rc = snprintf(buf, len,
2310                               "run_time: %u seconds\n"
2311                               "average_speed: "LPU64" objects/sec\n"
2312                               "real-time_speed: "LPU64" objects/sec\n"
2313                               "current_position: %u\n",
2314                               rtime, speed, new_checked, scrub->os_pos_current);
2315         } else {
2316                 if (sf->sf_run_time != 0)
2317                         do_div(speed, sf->sf_run_time);
2318                 rc = snprintf(buf, len,
2319                               "run_time: %u seconds\n"
2320                               "average_speed: "LPU64" objects/sec\n"
2321                               "real-time_speed: N/A\n"
2322                               "current_position: N/A\n",
2323                               sf->sf_run_time, speed);
2324         }
2325         if (rc <= 0)
2326                 goto out;
2327
2328         buf += rc;
2329         len -= rc;
2330         ret = save - len;
2331
2332 out:
2333         up_read(&scrub->os_rwsem);
2334         return ret;
2335 }