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