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, 2014, 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 #define DEBUG_SUBSYSTEM S_LFSCK
  39
  40 #include <lustre/lustre_idl.h>
  41 #include <lustre_disk.h>
  42 #include <dt_object.h>
  43 #include <linux/xattr.h>
  44
  45 #include "osd_internal.h"
  46 #include "osd_oi.h"
  47 #include "osd_scrub.h"
  48
  49 #define HALF_SEC        msecs_to_jiffies(MSEC_PER_SEC >> 1)
  50
  51 #define OSD_OTABLE_MAX_HASH             0x00000000ffffffffULL
  52
  53 #define SCRUB_NEXT_BREAK        1 /* exit current loop and process next group */
  54 #define SCRUB_NEXT_CONTINUE     2 /* skip current object and process next bit */
  55 #define SCRUB_NEXT_EXIT         3 /* exit all the loops */
  56 #define SCRUB_NEXT_WAIT         4 /* wait for free cache slot */
  57 #define SCRUB_NEXT_CRASH        5 /* simulate system crash during OI scrub */
  58 #define SCRUB_NEXT_FATAL        6 /* simulate failure during OI scrub */
  59 #define SCRUB_NEXT_NOSCRUB      7 /* new created object, no scrub on it */
  60 #define SCRUB_NEXT_NOLMA        8 /* the inode has no FID-in-LMA */
  61 #define SCRUB_NEXT_OSTOBJ       9 /* for OST-object */
  62 #define SCRUB_NEXT_OSTOBJ_OLD   10 /* old OST-object, no LMA or no FID-on-OST
  63                                     * flags in LMA */
  64
  65 /* misc functions */
  66
  67 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
  68 {
  69         return container_of0(scrub, struct osd_device, od_scrub);
  70 }
  71
  72 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
  73 {
  74         return osd_sb(osd_scrub2dev(scrub));
  75 }
  76
  77 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
  78                                        struct osd_otable_cache *ooc)
  79 {
  80         return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
  81 }
  82
  83 static inline const char *osd_scrub2name(struct osd_scrub *scrub)
  84 {
  85         return LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
  86 }
  87
  88 /**
  89  * update/insert/delete the specified OI mapping (@fid @id) according to the ops
  90  *
  91  * \retval   1, changed nothing
  92  * \retval   0, changed successfully
  93  * \retval -ve, on error
  94  */
  95 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
  96                                      struct osd_device *dev,
  97                                      const struct lu_fid *fid,
  98                                      const struct osd_inode_id *id,
  99                                      int ops, bool force,
 100                                      enum oi_check_flags flags)
 101 {
 102         handle_t *th;
 103         int       rc;
 104         ENTRY;
 105
 106         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
 107                 RETURN(0);
 108
 109         /* DTO_INDEX_INSERT is enough for other two ops:
 110          * delete/update, but save stack. */
 111         th = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 112                                 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
 113         if (IS_ERR(th)) {
 114                 rc = PTR_ERR(th);
 115                 CDEBUG(D_LFSCK, "%s: fail to start trans for scrub op %d "
 116                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 117                        PFID(fid), id->oii_ino, id->oii_gen, rc);
 118                 RETURN(rc);
 119         }
 120
 121         switch (ops) {
 122         case DTO_INDEX_UPDATE:
 123                 rc = osd_oi_update(info, dev, fid, id, th, flags);
 124                 if (unlikely(rc == -ENOENT)) {
 125                         /* Some unlink thread may removed the OI mapping. */
 126                         rc = 1;
 127                 }
 128                 break;
 129         case DTO_INDEX_INSERT:
 130                 rc = osd_oi_insert(info, dev, fid, id, th, flags);
 131                 if (unlikely(rc == -EEXIST)) {
 132                         rc = 1;
 133                         /* XXX: There are trouble things when adding OI
 134                          *      mapping for IGIF object, which may cause
 135                          *      multiple objects to be mapped to the same
 136                          *      IGIF formatted FID. Consider the following
 137                          *      situations:
 138                          *
 139                          *      1) The MDT is upgrading from 1.8 device.
 140                          *      The OI scrub generates IGIF FID1 for the
 141                          *      OBJ1 and adds the OI mapping.
 142                          *
 143                          *      2) For some reason, the OI scrub does not
 144                          *      process all the IGIF objects completely.
 145                          *
 146                          *      3) The MDT is backuped and restored against
 147                          *      this device.
 148                          *
 149                          *      4) When the MDT mounts up, the OI scrub will
 150                          *      try to rebuild the OI files. For some IGIF
 151                          *      object, OBJ2, which was not processed by the
 152                          *      OI scrub before the backup/restore, and the
 153                          *      new generated IGIF formatted FID may be just
 154                          *      the FID1, the same as OBJ1.
 155                          *
 156                          *      Under such case, the OI scrub cannot know how
 157                          *      to generate new FID for the OBJ2.
 158                          *
 159                          *      Currently, we do nothing for that. One possible
 160                          *      solution is to generate new normal FID for the
 161                          *      conflict object.
 162                          *
 163                          *      Anyway, it is rare, only exists in theory. */
 164                 }
 165                 break;
 166         case DTO_INDEX_DELETE:
 167                 rc = osd_oi_delete(info, dev, fid, th, flags);
 168                 if (rc == -ENOENT) {
 169                         /* It is normal that the unlink thread has removed the
 170                          * OI mapping already. */
 171                         rc = 1;
 172                 }
 173                 break;
 174         default:
 175                 LASSERTF(0, "Unexpected ops %d\n", ops);
 176                 break;
 177         }
 178
 179         ldiskfs_journal_stop(th);
 180         if (rc < 0)
 181                 CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d "
 182                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 183                        PFID(fid), id->oii_ino, id->oii_gen, rc);
 184
 185         RETURN(rc);
 186 }
 187
 188 /* OI_scrub file ops */
 189
 190 static void osd_scrub_file_to_cpu(struct scrub_file *des,
 191                                   struct scrub_file *src)
 192 {
 193         memcpy(des->sf_uuid, src->sf_uuid, 16);
 194         des->sf_flags   = le64_to_cpu(src->sf_flags);
 195         des->sf_magic   = le32_to_cpu(src->sf_magic);
 196         des->sf_status  = le16_to_cpu(src->sf_status);
 197         des->sf_param   = le16_to_cpu(src->sf_param);
 198         des->sf_time_last_complete      =
 199                                 le64_to_cpu(src->sf_time_last_complete);
 200         des->sf_time_latest_start       =
 201                                 le64_to_cpu(src->sf_time_latest_start);
 202         des->sf_time_last_checkpoint    =
 203                                 le64_to_cpu(src->sf_time_last_checkpoint);
 204         des->sf_pos_latest_start        =
 205                                 le64_to_cpu(src->sf_pos_latest_start);
 206         des->sf_pos_last_checkpoint     =
 207                                 le64_to_cpu(src->sf_pos_last_checkpoint);
 208         des->sf_pos_first_inconsistent  =
 209                                 le64_to_cpu(src->sf_pos_first_inconsistent);
 210         des->sf_items_checked           =
 211                                 le64_to_cpu(src->sf_items_checked);
 212         des->sf_items_updated           =
 213                                 le64_to_cpu(src->sf_items_updated);
 214         des->sf_items_failed            =
 215                                 le64_to_cpu(src->sf_items_failed);
 216         des->sf_items_updated_prior     =
 217                                 le64_to_cpu(src->sf_items_updated_prior);
 218         des->sf_run_time        = le32_to_cpu(src->sf_run_time);
 219         des->sf_success_count   = le32_to_cpu(src->sf_success_count);
 220         des->sf_oi_count        = le16_to_cpu(src->sf_oi_count);
 221         des->sf_internal_flags  = le16_to_cpu(src->sf_internal_flags);
 222         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 223 }
 224
 225 static void osd_scrub_file_to_le(struct scrub_file *des,
 226                                  struct scrub_file *src)
 227 {
 228         memcpy(des->sf_uuid, src->sf_uuid, 16);
 229         des->sf_flags   = cpu_to_le64(src->sf_flags);
 230         des->sf_magic   = cpu_to_le32(src->sf_magic);
 231         des->sf_status  = cpu_to_le16(src->sf_status);
 232         des->sf_param   = cpu_to_le16(src->sf_param);
 233         des->sf_time_last_complete      =
 234                                 cpu_to_le64(src->sf_time_last_complete);
 235         des->sf_time_latest_start       =
 236                                 cpu_to_le64(src->sf_time_latest_start);
 237         des->sf_time_last_checkpoint    =
 238                                 cpu_to_le64(src->sf_time_last_checkpoint);
 239         des->sf_pos_latest_start        =
 240                                 cpu_to_le64(src->sf_pos_latest_start);
 241         des->sf_pos_last_checkpoint     =
 242                                 cpu_to_le64(src->sf_pos_last_checkpoint);
 243         des->sf_pos_first_inconsistent  =
 244                                 cpu_to_le64(src->sf_pos_first_inconsistent);
 245         des->sf_items_checked           =
 246                                 cpu_to_le64(src->sf_items_checked);
 247         des->sf_items_updated           =
 248                                 cpu_to_le64(src->sf_items_updated);
 249         des->sf_items_failed            =
 250                                 cpu_to_le64(src->sf_items_failed);
 251         des->sf_items_updated_prior     =
 252                                 cpu_to_le64(src->sf_items_updated_prior);
 253         des->sf_run_time        = cpu_to_le32(src->sf_run_time);
 254         des->sf_success_count   = cpu_to_le32(src->sf_success_count);
 255         des->sf_oi_count        = cpu_to_le16(src->sf_oi_count);
 256         des->sf_internal_flags  = cpu_to_le16(src->sf_internal_flags);
 257         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 258 }
 259
 260 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
 261 {
 262         struct scrub_file *sf = &scrub->os_file;
 263
 264         memset(sf, 0, sizeof(*sf));
 265         memcpy(sf->sf_uuid, uuid, 16);
 266         sf->sf_magic = SCRUB_MAGIC_V1;
 267         sf->sf_status = SS_INIT;
 268 }
 269
 270 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
 271 {
 272         struct scrub_file *sf = &scrub->os_file;
 273
 274         CDEBUG(D_LFSCK, "%.16s: reset OI scrub file, old flags = "
 275                LPX64", add flags = "LPX64"\n",
 276                osd_scrub2name(scrub), sf->sf_flags, flags);
 277
 278         memcpy(sf->sf_uuid, uuid, 16);
 279         sf->sf_status = SS_INIT;
 280         sf->sf_flags |= flags;
 281         sf->sf_flags &= ~SF_AUTO;
 282         sf->sf_run_time = 0;
 283         sf->sf_time_latest_start = 0;
 284         sf->sf_time_last_checkpoint = 0;
 285         sf->sf_pos_latest_start = 0;
 286         sf->sf_pos_last_checkpoint = 0;
 287         sf->sf_pos_first_inconsistent = 0;
 288         sf->sf_items_checked = 0;
 289         sf->sf_items_updated = 0;
 290         sf->sf_items_failed = 0;
 291         if (!scrub->os_in_join)
 292                 sf->sf_items_updated_prior = 0;
 293
 294         sf->sf_items_noscrub = 0;
 295         sf->sf_items_igif = 0;
 296 }
 297
 298 static int osd_scrub_file_load(struct osd_scrub *scrub)
 299 {
 300         loff_t  pos  = 0;
 301         int     len  = sizeof(scrub->os_file_disk);
 302         int     rc;
 303
 304         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
 305         if (rc == len) {
 306                 struct scrub_file *sf = &scrub->os_file;
 307
 308                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
 309                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
 310                         CDEBUG(D_LFSCK, "%.16s: invalid scrub magic "
 311                                "0x%x != 0x%x\n", osd_scrub2name(scrub),
 312                                sf->sf_magic, SCRUB_MAGIC_V1);
 313                         /* Process it as new scrub file. */
 314                         rc = -ENOENT;
 315                 } else {
 316                         rc = 0;
 317                 }
 318         } else if (rc != 0) {
 319                 CDEBUG(D_LFSCK, "%.16s: fail to load scrub file, "
 320                        "expected = %d: rc = %d\n",
 321                        osd_scrub2name(scrub), len, rc);
 322                 if (rc > 0)
 323                         rc = -EFAULT;
 324         } else {
 325                 /* return -ENOENT for empty scrub file case. */
 326                 rc = -ENOENT;
 327         }
 328
 329         return rc;
 330 }
 331
 332 int osd_scrub_file_store(struct osd_scrub *scrub)
 333 {
 334         struct osd_device *dev;
 335         handle_t          *jh;
 336         loff_t             pos     = 0;
 337         int                len     = sizeof(scrub->os_file_disk);
 338         int                credits;
 339         int                rc;
 340
 341         dev = container_of0(scrub, struct osd_device, od_scrub);
 342         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
 343                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
 344         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits);
 345         if (IS_ERR(jh)) {
 346                 rc = PTR_ERR(jh);
 347                 CDEBUG(D_LFSCK, "%.16s: fail to start trans for scrub store: "
 348                        "rc = %d\n", osd_scrub2name(scrub), rc);
 349                 return rc;
 350         }
 351
 352         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
 353         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
 354                                       len, 0, &pos, jh);
 355         ldiskfs_journal_stop(jh);
 356         if (rc != 0)
 357                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file, "
 358                        "expected = %d: rc = %d\n",
 359                        osd_scrub2name(scrub), len, rc);
 360
 361         scrub->os_time_last_checkpoint = cfs_time_current();
 362         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
 363                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
 364         return rc;
 365 }
 366
 367 static int
 368 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
 369                      struct inode *inode, const struct lu_fid *fid)
 370 {
 371         struct filter_fid_old   *ff      = &info->oti_ff;
 372         struct dentry           *dentry  = &info->oti_obj_dentry;
 373         handle_t                *jh;
 374         int                      size    = 0;
 375         int                      rc;
 376         bool                     removed = false;
 377         bool                     reset   = true;
 378         ENTRY;
 379
 380         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
 381                 RETURN(0);
 382
 383         /* We want the LMA to fit into the 256-byte OST inode, so operate
 384          * as following:
 385          * 1) read old XATTR_NAME_FID and save the parent FID;
 386          * 2) delete the old XATTR_NAME_FID;
 387          * 3) make new LMA and add it;
 388          * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
 389          *
 390          * Making the LMA to fit into the 256-byte OST inode can save time for
 391          * normal osd_check_lma() and for other OI scrub scanning in future.
 392          * So it is worth to make some slow conversion here. */
 393         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 394                                 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
 395         if (IS_ERR(jh)) {
 396                 rc = PTR_ERR(jh);
 397                 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
 398                        DFID": rc = %d\n", osd_name(dev), PFID(fid), rc);
 399                 RETURN(rc);
 400         }
 401
 402         /* 1) read old XATTR_NAME_FID and save the parent FID */
 403         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 404         if (rc == sizeof(*ff)) {
 405                 /* 2) delete the old XATTR_NAME_FID */
 406                 ll_vfs_dq_init(inode);
 407                 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
 408                 if (rc != 0)
 409                         GOTO(stop, rc);
 410
 411                 removed = true;
 412         } else if (unlikely(rc == -ENODATA)) {
 413                 reset = false;
 414         } else if (rc != sizeof(struct filter_fid)) {
 415                 GOTO(stop, rc = -EINVAL);
 416         }
 417
 418         /* 3) make new LMA and add it */
 419         rc = osd_ea_fid_set(info, inode, fid, LMAC_FID_ON_OST, 0);
 420         if (rc == 0 && reset)
 421                 size = sizeof(struct filter_fid);
 422         else if (rc != 0 && removed)
 423                 /* If failed, we should try to add the old back. */
 424                 size = sizeof(struct filter_fid_old);
 425
 426         /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
 427         if (size > 0) {
 428                 int rc1;
 429
 430                 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
 431                                       XATTR_CREATE);
 432                 if (rc1 != 0 && rc == 0)
 433                         rc = rc1;
 434         }
 435
 436         GOTO(stop, rc);
 437
 438 stop:
 439         ldiskfs_journal_stop(jh);
 440         if (rc < 0)
 441                 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
 442                        osd_name(dev), PFID(fid), rc);
 443         return rc;
 444 }
 445
 446 static int
 447 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
 448                        struct osd_idmap_cache *oic, int val)
 449 {
 450         struct osd_scrub             *scrub  = &dev->od_scrub;
 451         struct scrub_file            *sf     = &scrub->os_file;
 452         struct lu_fid                *fid    = &oic->oic_fid;
 453         struct osd_inode_id          *lid    = &oic->oic_lid;
 454         struct osd_inode_id          *lid2   = &info->oti_id;
 455         struct osd_inconsistent_item *oii    = NULL;
 456         struct inode                 *inode  = NULL;
 457         int                           ops    = DTO_INDEX_UPDATE;
 458         int                           idx;
 459         int                           rc;
 460         bool                          converted = false;
 461         ENTRY;
 462
 463         down_write(&scrub->os_rwsem);
 464         scrub->os_new_checked++;
 465         if (val < 0)
 466                 GOTO(out, rc = val);
 467
 468         if (scrub->os_in_prior)
 469                 oii = list_entry(oic, struct osd_inconsistent_item,
 470                                  oii_cache);
 471
 472         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
 473                 GOTO(out, rc = 0);
 474
 475         if (fid_is_igif(fid))
 476                 sf->sf_items_igif++;
 477
 478         if (val == SCRUB_NEXT_OSTOBJ_OLD) {
 479                 inode = osd_iget(info, dev, lid);
 480                 if (IS_ERR(inode)) {
 481                         rc = PTR_ERR(inode);
 482                         /* Someone removed the inode. */
 483                         if (rc == -ENOENT || rc == -ESTALE)
 484                                 rc = 0;
 485                         GOTO(out, rc);
 486                 }
 487
 488                 sf->sf_flags |= SF_UPGRADE;
 489                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
 490                 dev->od_check_ff = 1;
 491                 rc = osd_scrub_convert_ff(info, dev, inode, fid);
 492                 if (rc != 0)
 493                         GOTO(out, rc);
 494
 495                 converted = true;
 496         }
 497
 498         if ((val == SCRUB_NEXT_NOLMA) &&
 499             (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
 500                 GOTO(out, rc = 0);
 501
 502         if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA))
 503                 goto iget;
 504
 505         rc = osd_oi_lookup(info, dev, fid, lid2,
 506                 (val == SCRUB_NEXT_OSTOBJ ||
 507                  val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 508         if (rc != 0) {
 509                 if (rc != -ENOENT && rc != -ESTALE)
 510                         GOTO(out, rc);
 511
 512 iget:
 513                 if (inode == NULL) {
 514                         inode = osd_iget(info, dev, lid);
 515                         if (IS_ERR(inode)) {
 516                                 rc = PTR_ERR(inode);
 517                                 /* Someone removed the inode. */
 518                                 if (rc == -ENOENT || rc == -ESTALE)
 519                                         rc = 0;
 520                                 GOTO(out, rc);
 521                         }
 522                 }
 523
 524                 if (!scrub->os_partial_scan)
 525                         scrub->os_full_speed = 1;
 526
 527                 ops = DTO_INDEX_INSERT;
 528                 idx = osd_oi_fid2idx(dev, fid);
 529                 switch (val) {
 530                 case SCRUB_NEXT_NOLMA:
 531                         sf->sf_flags |= SF_UPGRADE;
 532                         if (!(sf->sf_param & SP_DRYRUN)) {
 533                                 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
 534                                 if (rc != 0)
 535                                         GOTO(out, rc);
 536                         }
 537
 538                         if (!(sf->sf_flags & SF_INCONSISTENT))
 539                                 dev->od_igif_inoi = 0;
 540                         break;
 541                 case SCRUB_NEXT_OSTOBJ:
 542                         sf->sf_flags |= SF_INCONSISTENT;
 543                 case SCRUB_NEXT_OSTOBJ_OLD:
 544                         break;
 545                 default:
 546                         sf->sf_flags |= SF_RECREATED;
 547                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 548                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 549                         break;
 550                 }
 551         } else if (osd_id_eq(lid, lid2)) {
 552                 if (converted)
 553                         sf->sf_items_updated++;
 554
 555                 GOTO(out, rc = 0);
 556         } else {
 557                 if (!scrub->os_partial_scan)
 558                         scrub->os_full_speed = 1;
 559
 560                 sf->sf_flags |= SF_INCONSISTENT;
 561
 562                 /* XXX: If the device is restored from file-level backup, then
 563                  *      some IGIFs may have been already in OI files, and some
 564                  *      may be not yet. Means upgrading from 1.8 may be partly
 565                  *      processed, but some clients may hold some immobilized
 566                  *      IGIFs, and use them to access related objects. Under
 567                  *      such case, OSD does not know whether an given IGIF has
 568                  *      been processed or to be processed, and it also cannot
 569                  *      generate local ino#/gen# directly from the immobilized
 570                  *      IGIF because of the backup/restore. Then force OSD to
 571                  *      lookup the given IGIF in OI files, and if no entry,
 572                  *      then ask the client to retry after upgrading completed.
 573                  *      No better choice. */
 574                 dev->od_igif_inoi = 1;
 575         }
 576
 577         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
 578                         (val == SCRUB_NEXT_OSTOBJ ||
 579                          val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 580         if (rc == 0) {
 581                 if (scrub->os_in_prior)
 582                         sf->sf_items_updated_prior++;
 583                 else
 584                         sf->sf_items_updated++;
 585         }
 586
 587         GOTO(out, rc);
 588
 589 out:
 590         if (rc < 0) {
 591                 sf->sf_items_failed++;
 592                 if (sf->sf_pos_first_inconsistent == 0 ||
 593                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 594                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 595         } else {
 596                 rc = 0;
 597         }
 598
 599         /* There may be conflict unlink during the OI scrub,
 600          * if happend, then remove the new added OI mapping. */
 601         if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
 602             unlikely(inode->i_nlink == 0))
 603                 osd_scrub_refresh_mapping(info, dev, fid, lid,
 604                                 DTO_INDEX_DELETE, false,
 605                                 (val == SCRUB_NEXT_OSTOBJ ||
 606                                  val == SCRUB_NEXT_OSTOBJ_OLD) ?
 607                                 OI_KNOWN_ON_OST : 0);
 608         up_write(&scrub->os_rwsem);
 609
 610         if (inode != NULL && !IS_ERR(inode))
 611                 iput(inode);
 612
 613         if (oii != NULL) {
 614                 LASSERT(!list_empty(&oii->oii_list));
 615
 616                 spin_lock(&scrub->os_lock);
 617                 list_del_init(&oii->oii_list);
 618                 spin_unlock(&scrub->os_lock);
 619                 OBD_FREE_PTR(oii);
 620         }
 621         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 622 }
 623
 624 /* OI scrub APIs */
 625
 626 static int osd_scrub_prep(struct osd_device *dev)
 627 {
 628         struct osd_scrub     *scrub  = &dev->od_scrub;
 629         struct ptlrpc_thread *thread = &scrub->os_thread;
 630         struct scrub_file    *sf     = &scrub->os_file;
 631         __u32                 flags  = scrub->os_start_flags;
 632         int                   rc;
 633         bool                  drop_dryrun = false;
 634         ENTRY;
 635
 636         CDEBUG(D_LFSCK, "%.16s: OI scrub prep, flags = 0x%x\n",
 637                osd_scrub2name(scrub), flags);
 638
 639         down_write(&scrub->os_rwsem);
 640         if (flags & SS_SET_FAILOUT)
 641                 sf->sf_param |= SP_FAILOUT;
 642         else if (flags & SS_CLEAR_FAILOUT)
 643                 sf->sf_param &= ~SP_FAILOUT;
 644
 645         if (flags & SS_SET_DRYRUN) {
 646                 sf->sf_param |= SP_DRYRUN;
 647         } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
 648                 sf->sf_param &= ~SP_DRYRUN;
 649                 drop_dryrun = true;
 650         }
 651
 652         if (flags & SS_RESET)
 653                 osd_scrub_file_reset(scrub,
 654                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
 655
 656         if (flags & SS_AUTO_FULL) {
 657                 scrub->os_full_speed = 1;
 658                 scrub->os_partial_scan = 0;
 659                 sf->sf_flags |= SF_AUTO;
 660         } else if (flags & SS_AUTO_PARTIAL) {
 661                 scrub->os_full_speed = 0;
 662                 scrub->os_partial_scan = 1;
 663                 sf->sf_flags |= SF_AUTO;
 664         } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
 665                                    SF_UPGRADE)) {
 666                 scrub->os_full_speed = 1;
 667                 scrub->os_partial_scan = 0;
 668         } else {
 669                 scrub->os_full_speed = 0;
 670                 scrub->os_partial_scan = 0;
 671         }
 672
 673         spin_lock(&scrub->os_lock);
 674         scrub->os_in_prior = 0;
 675         scrub->os_waiting = 0;
 676         scrub->os_paused = 0;
 677         scrub->os_in_join = 0;
 678         scrub->os_full_scrub = 0;
 679         spin_unlock(&scrub->os_lock);
 680         scrub->os_new_checked = 0;
 681         if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
 682                 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
 683         else if (sf->sf_pos_last_checkpoint != 0)
 684                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 685         else
 686                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 687
 688         scrub->os_pos_current = sf->sf_pos_latest_start;
 689         sf->sf_status = SS_SCANNING;
 690         sf->sf_time_latest_start = cfs_time_current_sec();
 691         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 692         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
 693         rc = osd_scrub_file_store(scrub);
 694         if (rc == 0) {
 695                 spin_lock(&scrub->os_lock);
 696                 thread_set_flags(thread, SVC_RUNNING);
 697                 spin_unlock(&scrub->os_lock);
 698                 wake_up_all(&thread->t_ctl_waitq);
 699         }
 700         up_write(&scrub->os_rwsem);
 701
 702         RETURN(rc);
 703 }
 704
 705 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
 706 {
 707         struct scrub_file *sf = &scrub->os_file;
 708         int                rc;
 709
 710         if (likely(cfs_time_before(cfs_time_current(),
 711                                    scrub->os_time_next_checkpoint) ||
 712                    scrub->os_new_checked == 0))
 713                 return 0;
 714
 715         down_write(&scrub->os_rwsem);
 716         sf->sf_items_checked += scrub->os_new_checked;
 717         scrub->os_new_checked = 0;
 718         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 719         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 720         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 721                                             scrub->os_time_last_checkpoint);
 722         rc = osd_scrub_file_store(scrub);
 723         up_write(&scrub->os_rwsem);
 724
 725         return rc;
 726 }
 727
 728 static void osd_scrub_post(struct osd_scrub *scrub, int result)
 729 {
 730         struct scrub_file *sf = &scrub->os_file;
 731         ENTRY;
 732
 733         CDEBUG(D_LFSCK, "%.16s: OI scrub post, result = %d\n",
 734                osd_scrub2name(scrub), result);
 735
 736         down_write(&scrub->os_rwsem);
 737         spin_lock(&scrub->os_lock);
 738         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 739         spin_unlock(&scrub->os_lock);
 740         if (scrub->os_new_checked > 0) {
 741                 sf->sf_items_checked += scrub->os_new_checked;
 742                 scrub->os_new_checked = 0;
 743                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 744         }
 745         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 746         if (result > 0) {
 747                 struct osd_device *dev =
 748                         container_of0(scrub, struct osd_device, od_scrub);
 749
 750                 dev->od_igif_inoi = 1;
 751                 dev->od_check_ff = 0;
 752                 sf->sf_status = SS_COMPLETED;
 753                 if (!(sf->sf_param & SP_DRYRUN)) {
 754                         memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 755                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 756                                           SF_UPGRADE | SF_AUTO);
 757                 }
 758                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 759                 sf->sf_success_count++;
 760         } else if (result == 0) {
 761                 if (scrub->os_paused)
 762                         sf->sf_status = SS_PAUSED;
 763                 else
 764                         sf->sf_status = SS_STOPPED;
 765         } else {
 766                 sf->sf_status = SS_FAILED;
 767         }
 768         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 769                                             scrub->os_time_last_checkpoint);
 770         result = osd_scrub_file_store(scrub);
 771         up_write(&scrub->os_rwsem);
 772
 773         EXIT;
 774 }
 775
 776 /* iteration engine */
 777
 778 struct osd_iit_param {
 779         struct super_block *sb;
 780         struct buffer_head *bitmap;
 781         ldiskfs_group_t bg;
 782         __u32 gbase;
 783         __u32 offset;
 784 };
 785
 786 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 787                                    struct osd_device *dev,
 788                                    struct osd_iit_param *param,
 789                                    struct osd_idmap_cache **oic,
 790                                    const bool noslot);
 791
 792 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 793                                    struct osd_device *dev,
 794                                    struct osd_iit_param *param,
 795                                    struct osd_idmap_cache *oic,
 796                                    bool *noslot, int rc);
 797
 798 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 799 {
 800         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 801                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 802         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 803                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 804                 return SCRUB_NEXT_BREAK;
 805         } else {
 806                 *pos = param->gbase + param->offset;
 807                 return 0;
 808         }
 809 }
 810
 811 /**
 812  * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
 813  * \retval 0: FID-on-MDT
 814  */
 815 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
 816                                       struct osd_device *dev,
 817                                       struct lu_fid *fid)
 818 {
 819         /* XXX: The initial OI scrub will scan the top level /O to generate
 820          *      a small local FLDB according to the <seq>. If the given FID
 821          *      is in the local FLDB, then it is FID-on-OST; otherwise it's
 822          *      quite possible for FID-on-MDT. */
 823         if (dev->od_is_ost)
 824                 return SCRUB_NEXT_OSTOBJ_OLD;
 825         else
 826                 return 0;
 827 }
 828
 829 static int osd_scrub_get_fid(struct osd_thread_info *info,
 830                              struct osd_device *dev, struct inode *inode,
 831                              struct lu_fid *fid, bool scrub)
 832 {
 833         struct lustre_mdt_attrs *lma     = &info->oti_mdt_attrs;
 834         int                      rc;
 835         bool                     has_lma = false;
 836
 837         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 838         if (rc == 0) {
 839                 has_lma = true;
 840                 if (lma->lma_compat & LMAC_NOT_IN_OI ||
 841                     lma->lma_incompat & LMAI_AGENT)
 842                         return SCRUB_NEXT_CONTINUE;
 843
 844                 *fid = lma->lma_self_fid;
 845                 if (!scrub)
 846                         return 0;
 847
 848                 if (lma->lma_compat & LMAC_FID_ON_OST)
 849                         return SCRUB_NEXT_OSTOBJ;
 850
 851                 if (fid_is_idif(fid))
 852                         return SCRUB_NEXT_OSTOBJ_OLD;
 853
 854                 /* For local object. */
 855                 if (fid_is_internal(fid))
 856                         return 0;
 857
 858                 /* For external visible MDT-object with non-normal FID. */
 859                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
 860                         return 0;
 861
 862                 /* For the object with normal FID, it may be MDT-object,
 863                  * or may be 2.4 OST-object, need further distinguish.
 864                  * Fall through to next section. */
 865         }
 866
 867         if (rc == -ENODATA || rc == 0) {
 868                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
 869                 if (rc == 0) {
 870                         if (scrub)
 871                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
 872                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 873                         return rc;
 874                 }
 875
 876                 if (rc > 0) {
 877                         if (!has_lma)
 878                                 /* It is FID-on-OST, but we do not know how
 879                                  * to generate its FID, ignore it directly. */
 880                                 rc = SCRUB_NEXT_CONTINUE;
 881                         else
 882                                 /* It is 2.4 OST-object. */
 883                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 884                         return rc;
 885                 }
 886
 887                 if (rc != -ENODATA)
 888                         return rc;
 889
 890                 if (!has_lma) {
 891                         if (dev->od_scrub.os_convert_igif) {
 892                                 lu_igif_build(fid, inode->i_ino,
 893                                               inode->i_generation);
 894                                 if (scrub)
 895                                         rc = SCRUB_NEXT_NOLMA;
 896                                 else
 897                                         rc = 0;
 898                         } else {
 899                                 /* It may be FID-on-OST, or may be FID for
 900                                  * non-MDT0, anyway, we do not know how to
 901                                  * generate its FID, ignore it directly. */
 902                                 rc = SCRUB_NEXT_CONTINUE;
 903                         }
 904                         return rc;
 905                 }
 906
 907                 /* For OI scrub case only: the object has LMA but has no ff
 908                  * (or ff crashed). It may be MDT-object, may be OST-object
 909                  * with crashed ff. The last check is local FLDB. */
 910                 rc = osd_scrub_check_local_fldb(info, dev, fid);
 911         }
 912
 913         return rc;
 914 }
 915
 916 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 917                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 918                         struct super_block *sb, bool scrub)
 919 {
 920         struct inode *inode;
 921         int           rc;
 922         ENTRY;
 923
 924         /* Not handle the backend root object and agent parent object.
 925          * They are neither visible to namespace nor have OI mappings. */
 926         if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
 927                      pos == osd_remote_parent_ino(dev)))
 928                 RETURN(SCRUB_NEXT_CONTINUE);
 929
 930         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 931         inode = osd_iget(info, dev, lid);
 932         if (IS_ERR(inode)) {
 933                 rc = PTR_ERR(inode);
 934                 /* The inode may be removed after bitmap searching, or the
 935                  * file is new created without inode initialized yet. */
 936                 if (rc == -ENOENT || rc == -ESTALE)
 937                         RETURN(SCRUB_NEXT_CONTINUE);
 938
 939                 CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: "
 940                        "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name,
 941                        pos, rc);
 942                 RETURN(rc);
 943         }
 944
 945         if (scrub &&
 946             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 947                 /* Only skip it for the first OI scrub accessing. */
 948                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 949                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
 950         }
 951
 952         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
 953
 954         GOTO(put, rc);
 955
 956 put:
 957         iput(inode);
 958         return rc;
 959 }
 960
 961 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 962                           struct osd_iit_param *param,
 963                           struct osd_idmap_cache **oic, const bool noslot)
 964 {
 965         struct osd_scrub     *scrub  = &dev->od_scrub;
 966         struct ptlrpc_thread *thread = &scrub->os_thread;
 967         struct lu_fid        *fid;
 968         struct osd_inode_id  *lid;
 969         int                   rc;
 970
 971         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
 972                 struct l_wait_info lwi;
 973
 974                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
 975                 if (likely(lwi.lwi_timeout > 0))
 976                         l_wait_event(thread->t_ctl_waitq,
 977                                 !list_empty(&scrub->os_inconsistent_items) ||
 978                                 !thread_is_running(thread),
 979                                 &lwi);
 980         }
 981
 982         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 983                 spin_lock(&scrub->os_lock);
 984                 thread_set_flags(thread, SVC_STOPPING);
 985                 spin_unlock(&scrub->os_lock);
 986                 return SCRUB_NEXT_CRASH;
 987         }
 988
 989         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
 990                 return SCRUB_NEXT_FATAL;
 991
 992         if (unlikely(!thread_is_running(thread)))
 993                 return SCRUB_NEXT_EXIT;
 994
 995         if (!list_empty(&scrub->os_inconsistent_items)) {
 996                 struct osd_inconsistent_item *oii;
 997
 998                 oii = list_entry(scrub->os_inconsistent_items.next,
 999                                  struct osd_inconsistent_item, oii_list);
1000                 *oic = &oii->oii_cache;
1001                 scrub->os_in_prior = 1;
1002                 return 0;
1003         }
1004
1005         if (noslot)
1006                 return SCRUB_NEXT_WAIT;
1007
1008         rc = osd_iit_next(param, &scrub->os_pos_current);
1009         if (rc != 0)
1010                 return rc;
1011
1012         *oic = &scrub->os_oic;
1013         fid = &(*oic)->oic_fid;
1014         lid = &(*oic)->oic_lid;
1015         rc = osd_iit_iget(info, dev, fid, lid,
1016                           scrub->os_pos_current, param->sb, true);
1017         return rc;
1018 }
1019
1020 static int osd_preload_next(struct osd_thread_info *info,
1021                             struct osd_device *dev, struct osd_iit_param *param,
1022                             struct osd_idmap_cache **oic, const bool noslot)
1023 {
1024         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
1025         struct osd_scrub        *scrub;
1026         struct ptlrpc_thread    *thread;
1027         int                      rc;
1028
1029         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1030         if (rc != 0)
1031                 return rc;
1032
1033         scrub = &dev->od_scrub;
1034         thread = &scrub->os_thread;
1035         if (thread_is_running(thread) &&
1036             ooc->ooc_pos_preload >= scrub->os_pos_current)
1037                 return SCRUB_NEXT_EXIT;
1038
1039         rc = osd_iit_iget(info, dev,
1040                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1041                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1042                           ooc->ooc_pos_preload, param->sb, false);
1043         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1044          * ignore the failure, so it still need to skip the inode next time. */
1045         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1046         return rc;
1047 }
1048
1049 static inline int
1050 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1051 {
1052         spin_lock(&scrub->os_lock);
1053         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1054             !list_empty(&scrub->os_inconsistent_items) ||
1055             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1056                 scrub->os_waiting = 0;
1057         else
1058                 scrub->os_waiting = 1;
1059         spin_unlock(&scrub->os_lock);
1060
1061         return !scrub->os_waiting;
1062 }
1063
1064 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1065                           struct osd_iit_param *param,
1066                           struct osd_idmap_cache *oic, bool *noslot, int rc)
1067 {
1068         struct l_wait_info       lwi    = { 0 };
1069         struct osd_scrub        *scrub  = &dev->od_scrub;
1070         struct scrub_file       *sf     = &scrub->os_file;
1071         struct ptlrpc_thread    *thread = &scrub->os_thread;
1072         struct osd_otable_it    *it     = dev->od_otable_it;
1073         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
1074
1075         switch (rc) {
1076         case SCRUB_NEXT_CONTINUE:
1077                 goto next;
1078         case SCRUB_NEXT_WAIT:
1079                 goto wait;
1080         case SCRUB_NEXT_NOSCRUB:
1081                 down_write(&scrub->os_rwsem);
1082                 scrub->os_new_checked++;
1083                 sf->sf_items_noscrub++;
1084                 up_write(&scrub->os_rwsem);
1085                 goto next;
1086         }
1087
1088         rc = osd_scrub_check_update(info, dev, oic, rc);
1089         if (rc != 0)
1090                 return rc;
1091
1092         rc = osd_scrub_checkpoint(scrub);
1093         if (rc != 0) {
1094                 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1095                        "rc = %d\n", osd_scrub2name(scrub),
1096                        scrub->os_pos_current, rc);
1097                 /* Continue, as long as the scrub itself can go ahead. */
1098         }
1099
1100         if (scrub->os_in_prior) {
1101                 scrub->os_in_prior = 0;
1102                 return 0;
1103         }
1104
1105 next:
1106         scrub->os_pos_current = param->gbase + ++(param->offset);
1107
1108 wait:
1109         if (it != NULL && it->ooi_waiting && ooc != NULL &&
1110             ooc->ooc_pos_preload < scrub->os_pos_current) {
1111                 spin_lock(&scrub->os_lock);
1112                 it->ooi_waiting = 0;
1113                 wake_up_all(&thread->t_ctl_waitq);
1114                 spin_unlock(&scrub->os_lock);
1115         }
1116
1117         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1118                 return 0;
1119
1120         if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1121                 *noslot = false;
1122                 return 0;
1123         }
1124
1125         if (it != NULL)
1126                 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1127                              &lwi);
1128
1129         if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1130                 *noslot = false;
1131         else
1132                 *noslot = true;
1133         return 0;
1134 }
1135
1136 static int osd_preload_exec(struct osd_thread_info *info,
1137                             struct osd_device *dev, struct osd_iit_param *param,
1138                             struct osd_idmap_cache *oic, bool *noslot, int rc)
1139 {
1140         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1141
1142         if (rc == 0) {
1143                 ooc->ooc_cached_items++;
1144                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1145                                         ~OSD_OTABLE_IT_CACHE_MASK;
1146         }
1147         return rc > 0 ? 0 : rc;
1148 }
1149
1150 #define SCRUB_IT_ALL    1
1151 #define SCRUB_IT_CRASH  2
1152
1153 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1154                            bool inconsistent)
1155 {
1156         struct osd_scrub     *scrub  = &dev->od_scrub;
1157         struct ptlrpc_thread *thread = &scrub->os_thread;
1158         struct scrub_file    *sf     = &scrub->os_file;
1159         int                   rc;
1160         ENTRY;
1161
1162         LASSERT(!(flags & SS_AUTO_PARTIAL));
1163
1164         down_write(&scrub->os_rwsem);
1165         scrub->os_in_join = 1;
1166         if (flags & SS_SET_FAILOUT)
1167                 sf->sf_param |= SP_FAILOUT;
1168         else if (flags & SS_CLEAR_FAILOUT)
1169                 sf->sf_param &= ~SP_FAILOUT;
1170
1171         if (flags & SS_SET_DRYRUN)
1172                 sf->sf_param |= SP_DRYRUN;
1173         else if (flags & SS_CLEAR_DRYRUN)
1174                 sf->sf_param &= ~SP_DRYRUN;
1175
1176         if (flags & SS_RESET) {
1177                 osd_scrub_file_reset(scrub,
1178                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1179                         inconsistent ? SF_INCONSISTENT : 0);
1180                 sf->sf_status = SS_SCANNING;
1181         }
1182
1183         if (flags & SS_AUTO_FULL) {
1184                 sf->sf_flags |= SF_AUTO;
1185                 scrub->os_full_speed = 1;
1186         }
1187
1188         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1189                 scrub->os_full_speed = 1;
1190         else
1191                 scrub->os_full_speed = 0;
1192
1193         scrub->os_new_checked = 0;
1194         if (sf->sf_pos_last_checkpoint != 0)
1195                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1196         else
1197                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1198
1199         scrub->os_pos_current = sf->sf_pos_latest_start;
1200         sf->sf_time_latest_start = cfs_time_current_sec();
1201         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1202         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1203         rc = osd_scrub_file_store(scrub);
1204         if (rc != 0)
1205                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
1206                        "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1207
1208         spin_lock(&scrub->os_lock);
1209         scrub->os_waiting = 0;
1210         scrub->os_paused = 0;
1211         scrub->os_partial_scan = 0;
1212         scrub->os_in_join = 0;
1213         scrub->os_full_scrub = 0;
1214         spin_unlock(&scrub->os_lock);
1215         wake_up_all(&thread->t_ctl_waitq);
1216         up_write(&scrub->os_rwsem);
1217
1218         EXIT;
1219 }
1220
1221 static int osd_inode_iteration(struct osd_thread_info *info,
1222                                struct osd_device *dev, __u32 max, bool preload)
1223 {
1224         struct osd_scrub     *scrub  = &dev->od_scrub;
1225         struct ptlrpc_thread *thread = &scrub->os_thread;
1226         struct scrub_file    *sf     = &scrub->os_file;
1227         osd_iit_next_policy   next;
1228         osd_iit_exec_policy   exec;
1229         __u32                *pos;
1230         __u32                *count;
1231         struct osd_iit_param  param  = { NULL };
1232         struct l_wait_info    lwi    = { 0 };
1233         __u32                 limit;
1234         int                   rc;
1235         bool                  noslot = true;
1236         ENTRY;
1237
1238         param.sb = osd_sb(dev);
1239         if (preload)
1240                 goto full;
1241
1242         while (scrub->os_partial_scan && !scrub->os_in_join) {
1243                 struct osd_idmap_cache *oic = NULL;
1244
1245                 rc = osd_scrub_next(info, dev, &param, &oic, noslot);
1246                 switch (rc) {
1247                 case SCRUB_NEXT_EXIT:
1248                         RETURN(0);
1249                 case SCRUB_NEXT_CRASH:
1250                         RETURN(SCRUB_IT_CRASH);
1251                 case SCRUB_NEXT_FATAL:
1252                         RETURN(-EINVAL);
1253                 case SCRUB_NEXT_WAIT: {
1254                         struct kstatfs *ksfs = &info->oti_ksfs;
1255                         __u64 saved_flags;
1256
1257                         if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1258                             unlikely(sf->sf_items_updated_prior == 0))
1259                                 goto wait;
1260
1261                         if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1262                             scrub->os_full_scrub) {
1263                                 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1264                                                true);
1265                                 goto full;
1266                         }
1267
1268                         rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1269                         if (rc == 0) {
1270                                 __u64 used = ksfs->f_files - ksfs->f_ffree;
1271
1272                                 do_div(used, sf->sf_items_updated_prior);
1273                                 /* If we hit too much inconsistent OI
1274                                  * mappings during the partial scan,
1275                                  * then scan the device completely. */
1276                                 if (used < dev->od_full_scrub_ratio) {
1277                                         osd_scrub_join(dev,
1278                                                 SS_AUTO_FULL | SS_RESET, true);
1279                                         goto full;
1280                                 }
1281                         }
1282
1283 wait:
1284                         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1285                             cfs_fail_val > 0)
1286                                 continue;
1287
1288                         saved_flags = sf->sf_flags;
1289                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1290                                           SF_UPGRADE | SF_AUTO);
1291                         sf->sf_status = SS_COMPLETED;
1292                         l_wait_event(thread->t_ctl_waitq,
1293                                      !thread_is_running(thread) ||
1294                                      !scrub->os_partial_scan ||
1295                                      scrub->os_in_join ||
1296                                      !list_empty(&scrub->os_inconsistent_items),
1297                                      &lwi);
1298                         sf->sf_flags = saved_flags;
1299                         sf->sf_status = SS_SCANNING;
1300
1301                         if (unlikely(!thread_is_running(thread)))
1302                                 RETURN(0);
1303
1304                         if (!scrub->os_partial_scan || scrub->os_in_join)
1305                                 goto full;
1306
1307                         continue;
1308                 }
1309                 default:
1310                         LASSERTF(rc == 0, "rc = %d\n", rc);
1311
1312                         osd_scrub_exec(info, dev, &param, oic, &noslot, rc);
1313                         break;
1314                 }
1315         }
1316
1317 full:
1318         if (!preload) {
1319                 l_wait_event(thread->t_ctl_waitq,
1320                              !thread_is_running(thread) || !scrub->os_in_join,
1321                              &lwi);
1322
1323                 if (unlikely(!thread_is_running(thread)))
1324                         RETURN(0);
1325         }
1326
1327         noslot = false;
1328         if (!preload) {
1329                 next = osd_scrub_next;
1330                 exec = osd_scrub_exec;
1331                 pos = &scrub->os_pos_current;
1332                 count = &scrub->os_new_checked;
1333         } else {
1334                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1335
1336                 next = osd_preload_next;
1337                 exec = osd_preload_exec;
1338                 pos = &ooc->ooc_pos_preload;
1339                 count = &ooc->ooc_cached_items;
1340         }
1341         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1342
1343         while (*pos <= limit && *count < max) {
1344                 struct osd_idmap_cache *oic = NULL;
1345                 struct ldiskfs_group_desc *desc;
1346
1347                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1348                 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1349                 if (desc == NULL)
1350                         RETURN(-EIO);
1351
1352                 ldiskfs_lock_group(param.sb, param.bg);
1353                 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1354                         ldiskfs_unlock_group(param.sb, param.bg);
1355                         *pos = 1 + (param.bg + 1) *
1356                                 LDISKFS_INODES_PER_GROUP(param.sb);
1357                         continue;
1358                 }
1359                 ldiskfs_unlock_group(param.sb, param.bg);
1360
1361                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1362                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1363                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1364                 if (param.bitmap == NULL) {
1365                         CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1366                                "scrub will stop, urgent mode\n",
1367                                osd_scrub2name(scrub), (__u32)param.bg);
1368                         RETURN(-EIO);
1369                 }
1370
1371                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1372                        *count < max) {
1373                         if (param.offset +
1374                                 ldiskfs_itable_unused_count(param.sb, desc) >
1375                             LDISKFS_INODES_PER_GROUP(param.sb))
1376                                 goto next_group;
1377
1378                         rc = next(info, dev, &param, &oic, noslot);
1379                         switch (rc) {
1380                         case SCRUB_NEXT_BREAK:
1381                                 goto next_group;
1382                         case SCRUB_NEXT_EXIT:
1383                                 brelse(param.bitmap);
1384                                 RETURN(0);
1385                         case SCRUB_NEXT_CRASH:
1386                                 brelse(param.bitmap);
1387                                 RETURN(SCRUB_IT_CRASH);
1388                         case SCRUB_NEXT_FATAL:
1389                                 brelse(param.bitmap);
1390                                 RETURN(-EINVAL);
1391                         }
1392
1393                         rc = exec(info, dev, &param, oic, &noslot, rc);
1394                         if (rc != 0) {
1395                                 brelse(param.bitmap);
1396                                 RETURN(rc);
1397                         }
1398                 }
1399
1400 next_group:
1401                 brelse(param.bitmap);
1402         }
1403
1404         if (*pos > limit)
1405                 RETURN(SCRUB_IT_ALL);
1406         RETURN(0);
1407 }
1408
1409 static int osd_otable_it_preload(const struct lu_env *env,
1410                                  struct osd_otable_it *it)
1411 {
1412         struct osd_device       *dev   = it->ooi_dev;
1413         struct osd_scrub        *scrub = &dev->od_scrub;
1414         struct osd_otable_cache *ooc   = &it->ooi_cache;
1415         int                      rc;
1416         ENTRY;
1417
1418         rc = osd_inode_iteration(osd_oti_get(env), dev,
1419                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1420         if (rc == SCRUB_IT_ALL)
1421                 it->ooi_all_cached = 1;
1422
1423         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1424                 spin_lock(&scrub->os_lock);
1425                 scrub->os_waiting = 0;
1426                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1427                 spin_unlock(&scrub->os_lock);
1428         }
1429
1430         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1431 }
1432
1433 static int osd_scrub_main(void *args)
1434 {
1435         struct lu_env         env;
1436         struct osd_device    *dev    = (struct osd_device *)args;
1437         struct osd_scrub     *scrub  = &dev->od_scrub;
1438         struct ptlrpc_thread *thread = &scrub->os_thread;
1439         int                   rc;
1440         ENTRY;
1441
1442         rc = lu_env_init(&env, LCT_LOCAL);
1443         if (rc != 0) {
1444                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1445                        osd_scrub2name(scrub), rc);
1446                 GOTO(noenv, rc);
1447         }
1448
1449         rc = osd_scrub_prep(dev);
1450         if (rc != 0) {
1451                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1452                        osd_scrub2name(scrub), rc);
1453                 GOTO(out, rc);
1454         }
1455
1456         if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1457                 struct l_wait_info lwi = { 0 };
1458                 struct osd_otable_it *it = dev->od_otable_it;
1459                 struct osd_otable_cache *ooc = &it->ooi_cache;
1460
1461                 l_wait_event(thread->t_ctl_waitq,
1462                              it->ooi_user_ready || !thread_is_running(thread),
1463                              &lwi);
1464                 if (unlikely(!thread_is_running(thread)))
1465                         GOTO(post, rc = 0);
1466
1467                 scrub->os_pos_current = ooc->ooc_pos_preload;
1468         }
1469
1470         CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1471                osd_scrub2name(scrub), scrub->os_start_flags,
1472                scrub->os_pos_current);
1473
1474         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1475         if (unlikely(rc == SCRUB_IT_CRASH))
1476                 GOTO(out, rc = -EINVAL);
1477         GOTO(post, rc);
1478
1479 post:
1480         osd_scrub_post(scrub, rc);
1481         CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1482                osd_scrub2name(scrub), scrub->os_pos_current, rc);
1483
1484 out:
1485         while (!list_empty(&scrub->os_inconsistent_items)) {
1486                 struct osd_inconsistent_item *oii;
1487
1488                 oii = list_entry(scrub->os_inconsistent_items.next,
1489                                      struct osd_inconsistent_item, oii_list);
1490                 list_del_init(&oii->oii_list);
1491                 OBD_FREE_PTR(oii);
1492         }
1493         lu_env_fini(&env);
1494
1495 noenv:
1496         spin_lock(&scrub->os_lock);
1497         thread_set_flags(thread, SVC_STOPPED);
1498         wake_up_all(&thread->t_ctl_waitq);
1499         spin_unlock(&scrub->os_lock);
1500         return rc;
1501 }
1502
1503 /* initial OI scrub */
1504
1505 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1506                          struct dentry *, filldir_t filldir);
1507
1508 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1509                                loff_t offset, __u64 ino, unsigned d_type);
1510 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1511                            loff_t offset, __u64 ino, unsigned d_type);
1512 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1513                            loff_t offset, __u64 ino, unsigned d_type);
1514
1515 static int
1516 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1517                      struct dentry *dentry, filldir_t filldir);
1518 static int
1519 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1520                   struct dentry *dentry, filldir_t filldir);
1521
1522 static int
1523 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1524                      struct dentry *dentry, filldir_t filldir);
1525
1526 enum osd_lf_flags {
1527         OLF_SCAN_SUBITEMS       = 0x0001,
1528         OLF_HIDE_FID            = 0x0002,
1529         OLF_SHOW_NAME           = 0x0004,
1530         OLF_NO_OI               = 0x0008,
1531 };
1532
1533 struct osd_lf_map {
1534         char            *olm_name;
1535         struct lu_fid    olm_fid;
1536         __u16            olm_flags;
1537         __u16            olm_namelen;
1538         scandir_t        olm_scandir;
1539         filldir_t        olm_filldir;
1540 };
1541
1542 /* Add the new introduced local files in the list in the future. */
1543 static const struct osd_lf_map osd_lf_maps[] = {
1544         /* CATALOGS */
1545         { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1546                 sizeof(CATLIST) - 1, NULL, NULL },
1547
1548         /* CONFIGS */
1549         { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1550                 OLF_SCAN_SUBITEMS, sizeof(MOUNT_CONFIGS_DIR) - 1,
1551                 osd_ios_general_scan, osd_ios_varfid_fill },
1552
1553         /* NIDTBL_VERSIONS */
1554         { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1555                 sizeof(MGS_NIDTBL_DIR) - 1, osd_ios_general_scan,
1556                 osd_ios_varfid_fill },
1557
1558         /* PENDING */
1559         { "PENDING", { 0, 0, 0 }, 0, sizeof("PENDING") - 1, NULL, NULL },
1560
1561         /* ROOT */
1562         { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 },
1563                 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, sizeof("ROOT") - 1,
1564                 osd_ios_ROOT_scan, NULL },
1565
1566         /* changelog_catalog */
1567         { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, sizeof(CHANGELOG_CATALOG) - 1,
1568                 NULL, NULL },
1569
1570         /* changelog_users */
1571         { CHANGELOG_USERS, { 0, 0, 0 }, 0, sizeof(CHANGELOG_USERS) - 1,
1572                 NULL, NULL },
1573
1574         /* fld */
1575         { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1576                 sizeof("fld") - 1, NULL, NULL },
1577
1578         /* last_rcvd */
1579         { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1580                 sizeof(LAST_RCVD) - 1, NULL, NULL },
1581
1582         /* lov_objid */
1583         { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1584                 sizeof(LOV_OBJID) - 1, NULL, NULL },
1585
1586         /* lov_objseq */
1587         { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1588                 OLF_SHOW_NAME, sizeof(LOV_OBJSEQ) - 1, NULL, NULL },
1589
1590         /* quota_master */
1591         { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QMT_DIR) - 1,
1592                 osd_ios_general_scan, osd_ios_varfid_fill },
1593
1594         /* quota_slave */
1595         { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QSD_DIR) - 1,
1596                 osd_ios_general_scan, osd_ios_varfid_fill },
1597
1598         /* seq_ctl */
1599         { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1600                 OLF_SHOW_NAME, sizeof("seq_ctl") - 1, NULL, NULL },
1601
1602         /* seq_srv */
1603         { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1604                 OLF_SHOW_NAME, sizeof("seq_srv") - 1, NULL, NULL },
1605
1606         /* health_check */
1607         { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1608                 OLF_SHOW_NAME, sizeof(HEALTH_CHECK) - 1, NULL, NULL },
1609
1610         /* LFSCK */
1611         { LFSCK_DIR, { 0, 0, 0 }, 0, sizeof(LFSCK_DIR) - 1,
1612                 osd_ios_general_scan, osd_ios_varfid_fill },
1613
1614         /* lfsck_bookmark */
1615         { LFSCK_BOOKMARK, { 0, 0, 0 }, 0, sizeof(LFSCK_BOOKMARK) - 1,
1616                 NULL, NULL },
1617
1618         /* lfsck_layout */
1619         { LFSCK_LAYOUT, { 0, 0, 0 }, 0, sizeof(LFSCK_LAYOUT) - 1,
1620                 NULL, NULL },
1621
1622         /* lfsck_namespace */
1623         { LFSCK_NAMESPACE, { 0, 0, 0 }, 0, sizeof(LFSCK_NAMESPACE) - 1,
1624                 NULL, NULL },
1625
1626         /* OBJECTS, upgrade from old device */
1627         { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(OBJECTS) - 1,
1628                 osd_ios_OBJECTS_scan, NULL },
1629
1630         /* lquota_v2.user, upgrade from old device */
1631         { "lquota_v2.user", { 0, 0, 0 }, 0, sizeof("lquota_v2.user") - 1,
1632                 NULL, NULL },
1633
1634         /* lquota_v2.group, upgrade from old device */
1635         { "lquota_v2.group", { 0, 0, 0 }, 0, sizeof("lquota_v2.group") - 1,
1636                 NULL, NULL },
1637
1638         /* LAST_GROUP, upgrade from old device */
1639         { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1640                 OLF_SHOW_NAME, sizeof("LAST_GROUP") - 1, NULL, NULL },
1641
1642         /* SLAVE_LOG, llog for destroy slave stripes of striped dir */
1643         { "SLAVE_LOG", { FID_SEQ_LOCAL_FILE, SLAVE_LLOG_CATALOGS_OID, 0 },
1644                OLF_SHOW_NAME, sizeof("SLAVE_LOG") - 1, NULL, NULL },
1645
1646         /* lost+found */
1647         { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 },
1648                 OLF_SCAN_SUBITEMS, sizeof("lost+found") - 1,
1649                 osd_ios_general_scan, osd_ios_lf_fill },
1650
1651         { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1652 };
1653
1654 /* Add the new introduced files under .lustre/ in the list in the future. */
1655 static const struct osd_lf_map osd_dl_maps[] = {
1656         /* .lustre/fid */
1657         { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1658                 sizeof("fid") - 1, NULL, NULL },
1659
1660         /* .lustre/lost+found */
1661         { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1662                 sizeof("lost+found") - 1, NULL, NULL },
1663
1664         { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1665 };
1666
1667 struct osd_ios_item {
1668         struct list_head oii_list;
1669         struct dentry   *oii_dentry;
1670         scandir_t        oii_scandir;
1671         filldir_t        oii_filldir;
1672 };
1673
1674 struct osd_ios_filldir_buf {
1675 #ifdef HAVE_DIR_CONTEXT
1676         /* please keep it as first member */
1677         struct dir_context       ctx;
1678 #endif
1679         struct osd_thread_info  *oifb_info;
1680         struct osd_device       *oifb_dev;
1681         struct dentry           *oifb_dentry;
1682 };
1683
1684 static inline struct dentry *
1685 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1686 {
1687         struct dentry *dentry;
1688
1689         dentry = ll_lookup_one_len(name, parent, namelen);
1690         if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1691                 dput(dentry);
1692                 return ERR_PTR(-ENOENT);
1693         }
1694
1695         return dentry;
1696 }
1697
1698 static int
1699 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1700                  scandir_t scandir, filldir_t filldir)
1701 {
1702         struct osd_ios_item *item;
1703         ENTRY;
1704
1705         OBD_ALLOC_PTR(item);
1706         if (item == NULL)
1707                 RETURN(-ENOMEM);
1708
1709         INIT_LIST_HEAD(&item->oii_list);
1710         item->oii_dentry = dget(dentry);
1711         item->oii_scandir = scandir;
1712         item->oii_filldir = filldir;
1713         list_add_tail(&item->oii_list, &dev->od_ios_list);
1714
1715         RETURN(0);
1716 }
1717
1718 /**
1719  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1720  *
1721  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1722  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1723  * reference the inode, or fixed if it is missing or references another inode.
1724  */
1725 static int
1726 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1727                  struct inode *inode, const struct lu_fid *fid, int flags)
1728 {
1729         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
1730         struct osd_inode_id     *id     = &info->oti_id;
1731         struct osd_inode_id     *id2    = &info->oti_id2;
1732         struct osd_scrub        *scrub  = &dev->od_scrub;
1733         struct scrub_file       *sf     = &scrub->os_file;
1734         struct lu_fid            tfid;
1735         int                      rc;
1736         ENTRY;
1737
1738         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1739         if (rc != 0 && rc != -ENODATA) {
1740                 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1741                        "rc = %d\n", osd_name(dev), rc);
1742
1743                 RETURN(rc);
1744         }
1745
1746         osd_id_gen(id, inode->i_ino, inode->i_generation);
1747         if (rc == -ENODATA) {
1748                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID)
1749                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1750                 else
1751                         tfid = *fid;
1752                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1753                 if (rc != 0) {
1754                         CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1755                               "scrub: rc = %d\n", osd_name(dev), rc);
1756
1757                         RETURN(rc);
1758                 }
1759         } else {
1760                 if (lma->lma_compat & LMAC_NOT_IN_OI)
1761                         RETURN(0);
1762
1763                 tfid = lma->lma_self_fid;
1764         }
1765
1766         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1767         if (rc != 0) {
1768                 if (rc != -ENOENT)
1769                         RETURN(rc);
1770
1771                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1772                                                DTO_INDEX_INSERT, true, 0);
1773                 if (rc > 0)
1774                         rc = 0;
1775
1776                 RETURN(rc);
1777         }
1778
1779         if (osd_id_eq_strict(id, id2))
1780                 RETURN(0);
1781
1782         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1783                 osd_scrub_file_reset(scrub,
1784                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1785                                      SF_INCONSISTENT);
1786                 rc = osd_scrub_file_store(scrub);
1787                 if (rc != 0)
1788                         RETURN(rc);
1789         }
1790
1791         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1792                                        DTO_INDEX_UPDATE, true, 0);
1793         if (rc > 0)
1794                 rc = 0;
1795
1796         RETURN(rc);
1797 }
1798
1799 /**
1800  * It scans the /lost+found, and for the OST-object (with filter_fid
1801  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1802  */
1803 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1804                            loff_t offset, __u64 ino, unsigned d_type)
1805 {
1806         struct osd_ios_filldir_buf *fill_buf = buf;
1807         struct osd_thread_info     *info     = fill_buf->oifb_info;
1808         struct osd_device          *dev      = fill_buf->oifb_dev;
1809         struct lu_fid              *fid      = &info->oti_fid;
1810         struct osd_scrub           *scrub    = &dev->od_scrub;
1811         struct dentry              *parent   = fill_buf->oifb_dentry;
1812         struct dentry              *child;
1813         struct inode               *dir      = parent->d_inode;
1814         struct inode               *inode;
1815         int                         rc;
1816         ENTRY;
1817
1818         /* skip any '.' started names */
1819         if (name[0] == '.')
1820                 RETURN(0);
1821
1822         scrub->os_lf_scanned++;
1823         child = osd_ios_lookup_one_len(name, parent, namelen);
1824         if (IS_ERR(child)) {
1825                 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1826                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
1827                 RETURN(0);
1828         }
1829
1830         inode = child->d_inode;
1831         if (S_ISDIR(inode->i_mode)) {
1832                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1833                                       osd_ios_lf_fill);
1834                 if (rc != 0)
1835                         CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1836                               "rc = %d\n", osd_name(dev), namelen, name, rc);
1837                 GOTO(put, rc);
1838         }
1839
1840         if (!S_ISREG(inode->i_mode))
1841                 GOTO(put, rc = 0);
1842
1843         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1844         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1845                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1846                 if (rc == 0) {
1847                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1848                                "/lost+found.\n", namelen, name, PFID(fid));
1849                         scrub->os_lf_repaired++;
1850                 } else {
1851                         CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1852                                DFID": rc = %d\n",
1853                                osd_name(dev), namelen, name, PFID(fid), rc);
1854                 }
1855         }
1856
1857         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1858          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1859          *      can process them in furtuer. */
1860
1861         GOTO(put, rc);
1862
1863 put:
1864         if (rc < 0)
1865                 scrub->os_lf_failed++;
1866         dput(child);
1867         /* skip the failure to make the scanning to continue. */
1868         return 0;
1869 }
1870
1871 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1872                                loff_t offset, __u64 ino, unsigned d_type)
1873 {
1874         struct osd_ios_filldir_buf *fill_buf = buf;
1875         struct osd_device          *dev      = fill_buf->oifb_dev;
1876         struct dentry              *child;
1877         int                         rc;
1878         ENTRY;
1879
1880         /* skip any '.' started names */
1881         if (name[0] == '.')
1882                 RETURN(0);
1883
1884         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1885         if (IS_ERR(child))
1886                 RETURN(PTR_ERR(child));
1887
1888         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1889                               NULL, 0);
1890         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1891                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1892                                       osd_ios_varfid_fill);
1893         dput(child);
1894
1895         RETURN(rc);
1896 }
1897
1898 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1899                            loff_t offset, __u64 ino, unsigned d_type)
1900 {
1901         struct osd_ios_filldir_buf *fill_buf = buf;
1902         struct osd_device          *dev      = fill_buf->oifb_dev;
1903         const struct osd_lf_map    *map;
1904         struct dentry              *child;
1905         int                         rc       = 0;
1906         ENTRY;
1907
1908         /* skip any '.' started names */
1909         if (name[0] == '.')
1910                 RETURN(0);
1911
1912         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
1913                 if (map->olm_namelen != namelen)
1914                         continue;
1915
1916                 if (strncmp(map->olm_name, name, namelen) == 0)
1917                         break;
1918         }
1919
1920         if (map->olm_name == NULL)
1921                 RETURN(0);
1922
1923         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1924         if (IS_ERR(child))
1925                 RETURN(PTR_ERR(child));
1926
1927         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1928                               &map->olm_fid, map->olm_flags);
1929         dput(child);
1930
1931         RETURN(rc);
1932 }
1933
1934 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
1935                              loff_t offset, __u64 ino, unsigned d_type)
1936 {
1937         struct osd_ios_filldir_buf *fill_buf = buf;
1938         struct osd_device          *dev      = fill_buf->oifb_dev;
1939         const struct osd_lf_map    *map;
1940         struct dentry              *child;
1941         int                         rc       = 0;
1942         ENTRY;
1943
1944         /* skip any '.' started names */
1945         if (name[0] == '.')
1946                 RETURN(0);
1947
1948         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
1949                 if (map->olm_namelen != namelen)
1950                         continue;
1951
1952                 if (strncmp(map->olm_name, name, namelen) == 0)
1953                         break;
1954         }
1955
1956         if (map->olm_name == NULL)
1957                 RETURN(0);
1958
1959         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1960         if (IS_ERR(child))
1961                 RETURN(PTR_ERR(child));
1962
1963         if (!(map->olm_flags & OLF_NO_OI))
1964                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1965                                       &map->olm_fid, map->olm_flags);
1966         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
1967                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
1968                                       map->olm_filldir);
1969         dput(child);
1970
1971         RETURN(rc);
1972 }
1973
1974 static int
1975 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1976                      struct dentry *dentry, filldir_t filldir)
1977 {
1978         struct osd_ios_filldir_buf    buf   = {
1979 #ifdef HAVE_DIR_CONTEXT
1980                                                 .ctx.actor = filldir,
1981 #endif
1982                                                 .oifb_info = info,
1983                                                 .oifb_dev = dev,
1984                                                 .oifb_dentry = dentry };
1985         struct file                  *filp  = &info->oti_file;
1986         struct inode                 *inode = dentry->d_inode;
1987         const struct file_operations *fops  = inode->i_fop;
1988         int                           rc;
1989         ENTRY;
1990
1991         LASSERT(filldir != NULL);
1992
1993         filp->f_pos = 0;
1994         filp->f_dentry = dentry;
1995         filp->f_mode = FMODE_64BITHASH;
1996         filp->f_mapping = inode->i_mapping;
1997         filp->f_op = fops;
1998         filp->private_data = NULL;
1999         set_file_inode(filp, inode);
2000
2001 #ifdef HAVE_DIR_CONTEXT
2002         buf.ctx.pos = filp->f_pos;
2003         rc = fops->iterate(filp, &buf.ctx);
2004         filp->f_pos = buf.ctx.pos;
2005 #else
2006         rc = fops->readdir(filp, &buf, filldir);
2007 #endif
2008         fops->release(inode, filp);
2009
2010         RETURN(rc);
2011 }
2012
2013 static int
2014 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2015                   struct dentry *dentry, filldir_t filldir)
2016 {
2017         struct osd_scrub  *scrub  = &dev->od_scrub;
2018         struct scrub_file *sf     = &scrub->os_file;
2019         struct dentry     *child;
2020         int                rc;
2021         ENTRY;
2022
2023         /* It is existing MDT0 device. We only allow the case of object without
2024          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2025          * can generate IGIF mode FID for the object and related OI mapping. If
2026          * it is on other MDTs, then becuase file-level backup/restore, related
2027          * OI mapping may be invalid already, we do not know which is the right
2028          * FID for the object. We only allow IGIF objects to reside on the MDT0.
2029          *
2030          * XXX: For the case of object on non-MDT0 device with neither LMA nor
2031          *      "fid" xattr, then something crashed. We cannot re-generate the
2032          *      FID directly, instead, the OI scrub will scan the OI structure
2033          *      and try to re-generate the LMA from the OI mapping. But if the
2034          *      OI mapping crashed or lost also, then we have to give up under
2035          *      double failure cases. */
2036         scrub->os_convert_igif = 1;
2037         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2038                                        strlen(dot_lustre_name));
2039         if (IS_ERR(child)) {
2040                 rc = PTR_ERR(child);
2041                 if (rc == -ENOENT) {
2042                         /* It is 1.8 MDT device. */
2043                         if (!(sf->sf_flags & SF_UPGRADE)) {
2044                                 osd_scrub_file_reset(scrub,
2045                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2046                                         SF_UPGRADE);
2047                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2048                                 rc = osd_scrub_file_store(scrub);
2049                         } else {
2050                                 rc = 0;
2051                         }
2052                 }
2053         } else {
2054                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2055                  * so the client will get IGIF for the ".lustre" object when
2056                  * the MDT restart.
2057                  *
2058                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2059                  * it does not know whether there are some old clients cached
2060                  * the ".lustre" IGIF during the upgrading. Two choices:
2061                  *
2062                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2063                  *    It will allow the old connected clients to access the
2064                  *    ".lustre" with cached IGIF. But it will cause others
2065                  *    on the MDT failed to check "fid_is_dot_lustre()".
2066                  *
2067                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2068                  *    for ".lustre" in spite of whether there are some clients
2069                  *    cached the ".lustre" IGIF or not. It enables the check
2070                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
2071                  *    that the old connected clients cannot access the ".lustre"
2072                  *    with the cached IGIF.
2073                  *
2074                  * Usually, it is rare case for the old connected clients
2075                  * to access the ".lustre" with cached IGIF. So we prefer
2076                  * to the solution 2). */
2077                 rc = osd_ios_scan_one(info, dev, child->d_inode,
2078                                       &LU_DOT_LUSTRE_FID, 0);
2079                 if (rc == 0)
2080                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2081                                               osd_ios_dl_fill);
2082                 dput(child);
2083         }
2084
2085         RETURN(rc);
2086 }
2087
2088 static int
2089 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2090                      struct dentry *dentry, filldir_t filldir)
2091 {
2092         struct osd_scrub  *scrub  = &dev->od_scrub;
2093         struct scrub_file *sf     = &scrub->os_file;
2094         struct dentry     *child;
2095         int                rc;
2096         ENTRY;
2097
2098         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2099                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2100                 rc = osd_scrub_file_store(scrub);
2101                 if (rc != 0)
2102                         RETURN(rc);
2103         }
2104
2105         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2106         if (!IS_ERR(child)) {
2107                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2108                 dput(child);
2109         } else {
2110                 rc = PTR_ERR(child);
2111         }
2112
2113         if (rc != 0 && rc != -ENOENT)
2114                 RETURN(rc);
2115
2116         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2117         if (!IS_ERR(child)) {
2118                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2119                 dput(child);
2120         } else {
2121                 rc = PTR_ERR(child);
2122         }
2123
2124         if (rc == -ENOENT)
2125                 rc = 0;
2126
2127         RETURN(rc);
2128 }
2129
2130 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2131                                 struct osd_device *dev)
2132 {
2133         struct osd_ios_item     *item    = NULL;
2134         scandir_t                scandir = osd_ios_general_scan;
2135         filldir_t                filldir = osd_ios_root_fill;
2136         struct dentry           *dentry  = osd_sb(dev)->s_root;
2137         const struct osd_lf_map *map     = osd_lf_maps;
2138         int                      rc;
2139         ENTRY;
2140
2141         /* Lookup IGIF in OI by force for initial OI scrub. */
2142         dev->od_igif_inoi = 1;
2143
2144         while (1) {
2145                 rc = scandir(info, dev, dentry, filldir);
2146                 if (item != NULL) {
2147                         dput(item->oii_dentry);
2148                         OBD_FREE_PTR(item);
2149                 }
2150
2151                 if (rc != 0)
2152                         break;
2153
2154                 if (list_empty(&dev->od_ios_list))
2155                         break;
2156
2157                 item = list_entry(dev->od_ios_list.next,
2158                                   struct osd_ios_item, oii_list);
2159                 list_del_init(&item->oii_list);
2160
2161                 LASSERT(item->oii_scandir != NULL);
2162                 scandir = item->oii_scandir;
2163                 filldir = item->oii_filldir;
2164                 dentry = item->oii_dentry;
2165         }
2166
2167         while (!list_empty(&dev->od_ios_list)) {
2168                 item = list_entry(dev->od_ios_list.next,
2169                                   struct osd_ios_item, oii_list);
2170                 list_del_init(&item->oii_list);
2171                 dput(item->oii_dentry);
2172                 OBD_FREE_PTR(item);
2173         }
2174
2175         if (rc != 0)
2176                 RETURN(rc);
2177
2178         /* There maybe the case that the object has been removed, but its OI
2179          * mapping is still in the OI file, such as the "CATALOGS" after MDT
2180          * file-level backup/restore. So here cleanup the stale OI mappings. */
2181         while (map->olm_name != NULL) {
2182                 struct dentry *child;
2183
2184                 if (fid_is_zero(&map->olm_fid)) {
2185                         map++;
2186                         continue;
2187                 }
2188
2189                 child = osd_ios_lookup_one_len(map->olm_name,
2190                                                osd_sb(dev)->s_root,
2191                                                map->olm_namelen);
2192                 if (!IS_ERR(child))
2193                         dput(child);
2194                 else if (PTR_ERR(child) == -ENOENT)
2195                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2196                                                   NULL, DTO_INDEX_DELETE,
2197                                                   true, 0);
2198                 map++;
2199         }
2200
2201         RETURN(0);
2202 }
2203
2204 char *osd_lf_fid2name(const struct lu_fid *fid)
2205 {
2206         const struct osd_lf_map *map = osd_lf_maps;
2207
2208         while (map->olm_name != NULL) {
2209                 if (!lu_fid_eq(fid, &map->olm_fid)) {
2210                         map++;
2211                         continue;
2212                 }
2213
2214                 if (map->olm_flags & OLF_SHOW_NAME)
2215                         return map->olm_name;
2216                 else
2217                         return "";
2218         }
2219
2220         return NULL;
2221 }
2222
2223 /* OI scrub start/stop */
2224
2225 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2226 {
2227         struct osd_scrub     *scrub  = &dev->od_scrub;
2228         struct ptlrpc_thread *thread = &scrub->os_thread;
2229         struct l_wait_info    lwi    = { 0 };
2230         struct task_struct   *task;
2231         int                   rc;
2232         ENTRY;
2233
2234         /* os_lock: sync status between stop and scrub thread */
2235         spin_lock(&scrub->os_lock);
2236
2237 again:
2238         if (thread_is_running(thread)) {
2239                 spin_unlock(&scrub->os_lock);
2240                 if (!(scrub->os_file.sf_flags & SF_AUTO) ||
2241                      (flags & (SS_AUTO_FULL | SS_AUTO_PARTIAL)))
2242                         RETURN(-EALREADY);
2243
2244                 osd_scrub_join(dev, flags, false);
2245                 spin_lock(&scrub->os_lock);
2246                 if (!thread_is_running(thread))
2247                         goto again;
2248
2249                 spin_unlock(&scrub->os_lock);
2250                 RETURN(0);
2251         }
2252
2253         if (unlikely(thread_is_stopping(thread))) {
2254                 spin_unlock(&scrub->os_lock);
2255                 l_wait_event(thread->t_ctl_waitq,
2256                              thread_is_stopped(thread),
2257                              &lwi);
2258                 spin_lock(&scrub->os_lock);
2259                 goto again;
2260         }
2261         spin_unlock(&scrub->os_lock);
2262
2263         if (scrub->os_file.sf_status == SS_COMPLETED) {
2264                 if (!(flags & SS_SET_FAILOUT))
2265                         flags |= SS_CLEAR_FAILOUT;
2266
2267                 if (!(flags & SS_SET_DRYRUN))
2268                         flags |= SS_CLEAR_DRYRUN;
2269
2270                 flags |= SS_RESET;
2271         }
2272
2273         scrub->os_start_flags = flags;
2274         thread_set_flags(thread, 0);
2275         task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2276         if (IS_ERR(task)) {
2277                 rc = PTR_ERR(task);
2278                 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2279                        osd_scrub2name(scrub), rc);
2280                 RETURN(rc);
2281         }
2282
2283         l_wait_event(thread->t_ctl_waitq,
2284                      thread_is_running(thread) || thread_is_stopped(thread),
2285                      &lwi);
2286
2287         RETURN(0);
2288 }
2289
2290 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2291 {
2292         int rc;
2293         ENTRY;
2294
2295         /* od_otable_mutex: prevent curcurrent start/stop */
2296         mutex_lock(&dev->od_otable_mutex);
2297         rc = do_osd_scrub_start(dev, flags);
2298         mutex_unlock(&dev->od_otable_mutex);
2299
2300         RETURN(rc == -EALREADY ? 0 : rc);
2301 }
2302
2303 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2304 {
2305         struct ptlrpc_thread *thread = &scrub->os_thread;
2306         struct l_wait_info    lwi    = { 0 };
2307
2308         /* os_lock: sync status between stop and scrub thread */
2309         spin_lock(&scrub->os_lock);
2310         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2311                 thread_set_flags(thread, SVC_STOPPING);
2312                 spin_unlock(&scrub->os_lock);
2313                 wake_up_all(&thread->t_ctl_waitq);
2314                 l_wait_event(thread->t_ctl_waitq,
2315                              thread_is_stopped(thread),
2316                              &lwi);
2317                 /* Do not skip the last lock/unlock, which can guarantee that
2318                  * the caller cannot return until the OI scrub thread exit. */
2319                 spin_lock(&scrub->os_lock);
2320         }
2321         spin_unlock(&scrub->os_lock);
2322 }
2323
2324 static void osd_scrub_stop(struct osd_device *dev)
2325 {
2326         /* od_otable_mutex: prevent curcurrent start/stop */
2327         mutex_lock(&dev->od_otable_mutex);
2328         dev->od_scrub.os_paused = 1;
2329         do_osd_scrub_stop(&dev->od_scrub);
2330         mutex_unlock(&dev->od_otable_mutex);
2331 }
2332
2333 /* OI scrub setup/cleanup */
2334
2335 static const char osd_scrub_name[] = "OI_scrub";
2336
2337 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2338 {
2339         struct osd_thread_info     *info   = osd_oti_get(env);
2340         struct osd_scrub           *scrub  = &dev->od_scrub;
2341         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
2342         struct scrub_file          *sf     = &scrub->os_file;
2343         struct super_block         *sb     = osd_sb(dev);
2344         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
2345         struct lvfs_run_ctxt        saved;
2346         struct file                *filp;
2347         struct inode               *inode;
2348         struct lu_fid              *fid    = &info->oti_fid;
2349         int                         dirty  = 0;
2350         int                         rc     = 0;
2351         ENTRY;
2352
2353         memset(scrub, 0, sizeof(*scrub));
2354         OBD_SET_CTXT_MAGIC(ctxt);
2355         ctxt->pwdmnt = dev->od_mnt;
2356         ctxt->pwd = dev->od_mnt->mnt_root;
2357         ctxt->fs = get_ds();
2358
2359         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2360         init_rwsem(&scrub->os_rwsem);
2361         spin_lock_init(&scrub->os_lock);
2362         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2363
2364         push_ctxt(&saved, ctxt);
2365         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2366         if (IS_ERR(filp)) {
2367                 pop_ctxt(&saved, ctxt);
2368                 RETURN(PTR_ERR(filp));
2369         }
2370
2371         inode = filp->f_dentry->d_inode;
2372         /* 'What the @fid is' is not imporatant, because the object
2373          * has no OI mapping, and only is visible inside the OSD.*/
2374         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2375         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2376         if (rc != 0) {
2377                 filp_close(filp, NULL);
2378                 pop_ctxt(&saved, ctxt);
2379                 RETURN(rc);
2380         }
2381
2382         scrub->os_inode = igrab(inode);
2383         filp_close(filp, NULL);
2384         pop_ctxt(&saved, ctxt);
2385
2386         rc = osd_scrub_file_load(scrub);
2387         if (rc == -ENOENT) {
2388                 osd_scrub_file_init(scrub, es->s_uuid);
2389                 /* If the "/O" dir does not exist when mount (indicated by
2390                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2391                  * then it is quite probably that the device is a new one,
2392                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2393                  *
2394                  * For the rare case that "/O" and "OI_scrub" both lost on
2395                  * an old device, it can be found and cleared later.
2396                  *
2397                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2398                  * need to check "filter_fid_old" and to convert it to
2399                  * "filter_fid" for each object, and all the IGIF should
2400                  * have their FID mapping in OI files already. */
2401                 if (dev->od_maybe_new)
2402                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2403                 dirty = 1;
2404         } else if (rc != 0) {
2405                 GOTO(cleanup_inode, rc);
2406         } else {
2407                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2408                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2409                         dirty = 1;
2410                 } else if (sf->sf_status == SS_SCANNING) {
2411                         sf->sf_status = SS_CRASHED;
2412                         dirty = 1;
2413                 }
2414         }
2415
2416         if (sf->sf_pos_last_checkpoint != 0)
2417                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2418         else
2419                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2420
2421         if (dirty != 0) {
2422                 rc = osd_scrub_file_store(scrub);
2423                 if (rc != 0)
2424                         GOTO(cleanup_inode, rc);
2425         }
2426
2427         /* Initialize OI files. */
2428         rc = osd_oi_init(info, dev);
2429         if (rc < 0)
2430                 GOTO(cleanup_inode, rc);
2431
2432         rc = osd_initial_OI_scrub(info, dev);
2433         if (rc != 0)
2434                 GOTO(cleanup_oi, rc);
2435
2436         if (sf->sf_flags & SF_UPGRADE ||
2437             !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2438               sf->sf_success_count > 0)) {
2439                 dev->od_igif_inoi = 0;
2440                 dev->od_check_ff = dev->od_is_ost;
2441         } else {
2442                 dev->od_igif_inoi = 1;
2443                 dev->od_check_ff = 0;
2444         }
2445
2446         if (sf->sf_flags & SF_INCONSISTENT)
2447                 /* The 'od_igif_inoi' will be set under the
2448                  * following cases:
2449                  * 1) new created system, or
2450                  * 2) restored from file-level backup, or
2451                  * 3) the upgrading completed.
2452                  *
2453                  * The 'od_igif_inoi' may be cleared by OI scrub
2454                  * later if found that the system is upgrading. */
2455                 dev->od_igif_inoi = 1;
2456
2457         if (!dev->od_noscrub &&
2458             ((sf->sf_status == SS_PAUSED) ||
2459              (sf->sf_status == SS_CRASHED &&
2460               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2461                               SF_UPGRADE | SF_AUTO)) ||
2462              (sf->sf_status == SS_INIT &&
2463               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2464                               SF_UPGRADE))))
2465                 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2466
2467         if (rc != 0)
2468                 GOTO(cleanup_oi, rc);
2469
2470         /* it is possible that dcache entries may keep objects after they are
2471          * deleted by OSD. While it looks safe this can cause object data to
2472          * stay until umount causing failures in tests calculating free space,
2473          * e.g. replay-ost-single. Since those dcache entries are not used
2474          * anymore let's just free them after use here */
2475         shrink_dcache_sb(sb);
2476
2477         RETURN(0);
2478 cleanup_oi:
2479         osd_oi_fini(info, dev);
2480 cleanup_inode:
2481         iput(scrub->os_inode);
2482         scrub->os_inode = NULL;
2483
2484         return rc;
2485 }
2486
2487 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2488 {
2489         struct osd_scrub *scrub = &dev->od_scrub;
2490
2491         LASSERT(dev->od_otable_it == NULL);
2492
2493         if (scrub->os_inode != NULL) {
2494                 osd_scrub_stop(dev);
2495                 iput(scrub->os_inode);
2496                 scrub->os_inode = NULL;
2497         }
2498         if (dev->od_oi_table != NULL)
2499                 osd_oi_fini(osd_oti_get(env), dev);
2500 }
2501
2502 /* object table based iteration APIs */
2503
2504 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2505                                        struct dt_object *dt, __u32 attr,
2506                                        struct lustre_capa *capa)
2507 {
2508         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2509         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2510         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2511         struct osd_scrub       *scrub = &dev->od_scrub;
2512         struct osd_otable_it   *it;
2513         __u32                   start = 0;
2514         int                     rc;
2515         ENTRY;
2516
2517         /* od_otable_mutex: prevent curcurrent init/fini */
2518         mutex_lock(&dev->od_otable_mutex);
2519         if (dev->od_otable_it != NULL)
2520                 GOTO(out, it = ERR_PTR(-EALREADY));
2521
2522         OBD_ALLOC_PTR(it);
2523         if (it == NULL)
2524                 GOTO(out, it = ERR_PTR(-ENOMEM));
2525
2526         dev->od_otable_it = it;
2527         it->ooi_dev = dev;
2528         it->ooi_cache.ooc_consumer_idx = -1;
2529         if (flags & DOIF_OUTUSED)
2530                 it->ooi_used_outside = 1;
2531
2532         if (flags & DOIF_RESET)
2533                 start |= SS_RESET;
2534
2535         if (valid & DOIV_ERROR_HANDLE) {
2536                 if (flags & DOIF_FAILOUT)
2537                         start |= SS_SET_FAILOUT;
2538                 else
2539                         start |= SS_CLEAR_FAILOUT;
2540         }
2541
2542         if (valid & DOIV_DRYRUN) {
2543                 if (flags & DOIF_DRYRUN)
2544                         start |= SS_SET_DRYRUN;
2545                 else
2546                         start |= SS_CLEAR_DRYRUN;
2547         }
2548
2549         rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2550         if (rc < 0 && rc != -EALREADY) {
2551                 dev->od_otable_it = NULL;
2552                 OBD_FREE_PTR(it);
2553                 GOTO(out, it = ERR_PTR(rc));
2554         }
2555
2556         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2557
2558         GOTO(out, it);
2559
2560 out:
2561         mutex_unlock(&dev->od_otable_mutex);
2562         return (struct dt_it *)it;
2563 }
2564
2565 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2566 {
2567         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2568         struct osd_device    *dev = it->ooi_dev;
2569
2570         /* od_otable_mutex: prevent curcurrent init/fini */
2571         mutex_lock(&dev->od_otable_mutex);
2572         do_osd_scrub_stop(&dev->od_scrub);
2573         LASSERT(dev->od_otable_it == it);
2574
2575         dev->od_otable_it = NULL;
2576         mutex_unlock(&dev->od_otable_mutex);
2577         OBD_FREE_PTR(it);
2578 }
2579
2580 static int osd_otable_it_get(const struct lu_env *env,
2581                              struct dt_it *di, const struct dt_key *key)
2582 {
2583         return 0;
2584 }
2585
2586 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2587 {
2588 }
2589
2590 static inline int
2591 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2592 {
2593         spin_lock(&scrub->os_lock);
2594         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2595             scrub->os_waiting ||
2596             !thread_is_running(&scrub->os_thread))
2597                 it->ooi_waiting = 0;
2598         else
2599                 it->ooi_waiting = 1;
2600         spin_unlock(&scrub->os_lock);
2601
2602         return !it->ooi_waiting;
2603 }
2604
2605 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2606 {
2607         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2608         struct osd_device       *dev    = it->ooi_dev;
2609         struct osd_scrub        *scrub  = &dev->od_scrub;
2610         struct osd_otable_cache *ooc    = &it->ooi_cache;
2611         struct ptlrpc_thread    *thread = &scrub->os_thread;
2612         struct l_wait_info       lwi    = { 0 };
2613         int                      rc;
2614         ENTRY;
2615
2616         LASSERT(it->ooi_user_ready);
2617
2618 again:
2619         if (!thread_is_running(thread) && !it->ooi_used_outside)
2620                 RETURN(1);
2621
2622         if (ooc->ooc_cached_items > 0) {
2623                 ooc->ooc_cached_items--;
2624                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2625                                         ~OSD_OTABLE_IT_CACHE_MASK;
2626                 RETURN(0);
2627         }
2628
2629         if (it->ooi_all_cached) {
2630                 l_wait_event(thread->t_ctl_waitq,
2631                              !thread_is_running(thread),
2632                              &lwi);
2633                 RETURN(1);
2634         }
2635
2636         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2637                 spin_lock(&scrub->os_lock);
2638                 scrub->os_waiting = 0;
2639                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2640                 spin_unlock(&scrub->os_lock);
2641         }
2642
2643         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2644                 l_wait_event(thread->t_ctl_waitq,
2645                              osd_otable_it_wakeup(scrub, it),
2646                              &lwi);
2647
2648         if (!thread_is_running(thread) && !it->ooi_used_outside)
2649                 RETURN(1);
2650
2651         rc = osd_otable_it_preload(env, it);
2652         if (rc >= 0)
2653                 goto again;
2654
2655         RETURN(rc);
2656 }
2657
2658 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2659                                         const struct dt_it *di)
2660 {
2661         return NULL;
2662 }
2663
2664 static int osd_otable_it_key_size(const struct lu_env *env,
2665                                   const struct dt_it *di)
2666 {
2667         return sizeof(__u64);
2668 }
2669
2670 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2671                              struct dt_rec *rec, __u32 attr)
2672 {
2673         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2674         struct osd_otable_cache *ooc = &it->ooi_cache;
2675
2676         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2677
2678         /* Filter out Invald FID already. */
2679         LASSERTF(fid_is_sane((struct lu_fid *)rec),
2680                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2681                  PFID((struct lu_fid *)rec),
2682                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2683
2684         return 0;
2685 }
2686
2687 static __u64 osd_otable_it_store(const struct lu_env *env,
2688                                  const struct dt_it *di)
2689 {
2690         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2691         struct osd_otable_cache *ooc = &it->ooi_cache;
2692         __u64                    hash;
2693
2694         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2695                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2696         else
2697                 hash = ooc->ooc_pos_preload;
2698         return hash;
2699 }
2700
2701 /**
2702  * Set the OSD layer iteration start position as the specified hash.
2703  */
2704 static int osd_otable_it_load(const struct lu_env *env,
2705                               const struct dt_it *di, __u64 hash)
2706 {
2707         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
2708         struct osd_device       *dev   = it->ooi_dev;
2709         struct osd_otable_cache *ooc   = &it->ooi_cache;
2710         struct osd_scrub        *scrub = &dev->od_scrub;
2711         int                      rc;
2712         ENTRY;
2713
2714         /* Forbid to set iteration position after iteration started. */
2715         if (it->ooi_user_ready)
2716                 RETURN(-EPERM);
2717
2718         if (hash > OSD_OTABLE_MAX_HASH)
2719                 hash = OSD_OTABLE_MAX_HASH;
2720
2721         ooc->ooc_pos_preload = hash;
2722         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2723                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2724
2725         it->ooi_user_ready = 1;
2726         if (!scrub->os_full_speed)
2727                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2728
2729         /* Unplug OSD layer iteration by the first next() call. */
2730         rc = osd_otable_it_next(env, (struct dt_it *)it);
2731
2732         RETURN(rc);
2733 }
2734
2735 static int osd_otable_it_key_rec(const struct lu_env *env,
2736                                  const struct dt_it *di, void *key_rec)
2737 {
2738         return 0;
2739 }
2740
2741 const struct dt_index_operations osd_otable_ops = {
2742         .dio_it = {
2743                 .init     = osd_otable_it_init,
2744                 .fini     = osd_otable_it_fini,
2745                 .get      = osd_otable_it_get,
2746                 .put      = osd_otable_it_put,
2747                 .next     = osd_otable_it_next,
2748                 .key      = osd_otable_it_key,
2749                 .key_size = osd_otable_it_key_size,
2750                 .rec      = osd_otable_it_rec,
2751                 .store    = osd_otable_it_store,
2752                 .load     = osd_otable_it_load,
2753                 .key_rec  = osd_otable_it_key_rec,
2754         }
2755 };
2756
2757 /* high priority inconsistent items list APIs */
2758
2759 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL  60
2760
2761 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2762                    int insert)
2763 {
2764         struct osd_inconsistent_item *oii;
2765         struct osd_scrub             *scrub  = &dev->od_scrub;
2766         struct ptlrpc_thread         *thread = &scrub->os_thread;
2767         int                           wakeup = 0;
2768         ENTRY;
2769
2770         OBD_ALLOC_PTR(oii);
2771         if (unlikely(oii == NULL))
2772                 RETURN(-ENOMEM);
2773
2774         INIT_LIST_HEAD(&oii->oii_list);
2775         oii->oii_cache = *oic;
2776         oii->oii_insert = insert;
2777
2778         if (scrub->os_partial_scan) {
2779                 __u64 now = cfs_time_current_sec();
2780
2781                 /* If there haven't been errors in a long time,
2782                  * decay old count until either the errors are
2783                  * gone or we reach the current interval. */
2784                 while (unlikely(scrub->os_bad_oimap_count > 0 &&
2785                                 scrub->os_bad_oimap_time +
2786                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
2787                         scrub->os_bad_oimap_count >>= 1;
2788                         scrub->os_bad_oimap_time +=
2789                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
2790                 }
2791
2792                 scrub->os_bad_oimap_time = now;
2793                 if (++scrub->os_bad_oimap_count >
2794                     dev->od_full_scrub_threshold_rate)
2795                         scrub->os_full_scrub = 1;
2796         }
2797
2798         spin_lock(&scrub->os_lock);
2799         if (unlikely(!thread_is_running(thread))) {
2800                 spin_unlock(&scrub->os_lock);
2801                 OBD_FREE_PTR(oii);
2802                 RETURN(-EAGAIN);
2803         }
2804
2805         if (list_empty(&scrub->os_inconsistent_items))
2806                 wakeup = 1;
2807         list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2808         spin_unlock(&scrub->os_lock);
2809
2810         if (wakeup != 0)
2811                 wake_up_all(&thread->t_ctl_waitq);
2812
2813         RETURN(0);
2814 }
2815
2816 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2817                    struct osd_inode_id *id)
2818 {
2819         struct osd_scrub             *scrub = &dev->od_scrub;
2820         struct osd_inconsistent_item *oii;
2821         ENTRY;
2822
2823         spin_lock(&scrub->os_lock);
2824         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2825                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2826                         *id = oii->oii_cache.oic_lid;
2827                         spin_unlock(&scrub->os_lock);
2828                         RETURN(0);
2829                 }
2830         }
2831         spin_unlock(&scrub->os_lock);
2832
2833         RETURN(-ENOENT);
2834 }
2835
2836 /* OI scrub dump */
2837
2838 static const char *scrub_status_names[] = {
2839         "init",
2840         "scanning",
2841         "completed",
2842         "failed",
2843         "stopped",
2844         "paused",
2845         "crashed",
2846         NULL
2847 };
2848
2849 static const char *scrub_flags_names[] = {
2850         "recreated",
2851         "inconsistent",
2852         "auto",
2853         "upgrade",
2854         NULL
2855 };
2856
2857 static const char *scrub_param_names[] = {
2858         "failout",
2859         "dryrun",
2860         NULL
2861 };
2862
2863 static int scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
2864                            const char *prefix)
2865 {
2866         int flag;
2867         int rc;
2868         int i;
2869
2870         rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2871         if (rc < 0)
2872                 return rc;
2873
2874         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2875                 if (flag & bits) {
2876                         bits &= ~flag;
2877                         rc = seq_printf(m, "%s%c", names[i],
2878                                         bits != 0 ? ',' : '\n');
2879                         if (rc < 0)
2880                                 return rc;
2881                 }
2882         }
2883         return 0;
2884 }
2885
2886 static int scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
2887 {
2888         int rc;
2889
2890         if (time != 0)
2891                 rc = seq_printf(m, "%s: "LPU64" seconds\n", prefix,
2892                               cfs_time_current_sec() - time);
2893         else
2894                 rc = seq_printf(m, "%s: N/A\n", prefix);
2895         return rc;
2896 }
2897
2898 static int scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
2899 {
2900         int rc;
2901
2902         if (pos != 0)
2903                 rc = seq_printf(m, "%s: "LPU64"\n", prefix, pos);
2904         else
2905                 rc = seq_printf(m, "%s: N/A\n", prefix);
2906         return rc;
2907 }
2908
2909 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
2910 {
2911         struct osd_scrub  *scrub   = &dev->od_scrub;
2912         struct scrub_file *sf      = &scrub->os_file;
2913         __u64              checked;
2914         __u64              speed;
2915         int                rc;
2916
2917         down_read(&scrub->os_rwsem);
2918         rc = seq_printf(m, "name: OI_scrub\n"
2919                         "magic: 0x%x\n"
2920                         "oi_files: %d\n"
2921                         "status: %s\n",
2922                         sf->sf_magic, (int)sf->sf_oi_count,
2923                         scrub_status_names[sf->sf_status]);
2924         if (rc < 0)
2925                 goto out;
2926
2927         rc = scrub_bits_dump(m, sf->sf_flags, scrub_flags_names,
2928                              "flags");
2929         if (rc < 0)
2930                 goto out;
2931
2932         rc = scrub_bits_dump(m, sf->sf_param, scrub_param_names,
2933                              "param");
2934         if (rc < 0)
2935                 goto out;
2936
2937         rc = scrub_time_dump(m, sf->sf_time_last_complete,
2938                              "time_since_last_completed");
2939         if (rc < 0)
2940                 goto out;
2941
2942         rc = scrub_time_dump(m, sf->sf_time_latest_start,
2943                              "time_since_latest_start");
2944         if (rc < 0)
2945                 goto out;
2946
2947         rc = scrub_time_dump(m, sf->sf_time_last_checkpoint,
2948                              "time_since_last_checkpoint");
2949         if (rc < 0)
2950                 goto out;
2951
2952         rc = scrub_pos_dump(m, sf->sf_pos_latest_start,
2953                             "latest_start_position");
2954         if (rc < 0)
2955                 goto out;
2956
2957         rc = scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
2958                             "last_checkpoint_position");
2959         if (rc < 0)
2960                 goto out;
2961
2962         rc = scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
2963                             "first_failure_position");
2964         if (rc < 0)
2965                 goto out;
2966
2967         checked = sf->sf_items_checked + scrub->os_new_checked;
2968         rc = seq_printf(m, "checked: "LPU64"\n"
2969                       "updated: "LPU64"\n"
2970                       "failed: "LPU64"\n"
2971                       "prior_updated: "LPU64"\n"
2972                       "noscrub: "LPU64"\n"
2973                       "igif: "LPU64"\n"
2974                       "success_count: %u\n",
2975                       checked, sf->sf_items_updated, sf->sf_items_failed,
2976                       sf->sf_items_updated_prior, sf->sf_items_noscrub,
2977                       sf->sf_items_igif, sf->sf_success_count);
2978         if (rc < 0)
2979                 goto out;
2980
2981         speed = checked;
2982         if (thread_is_running(&scrub->os_thread)) {
2983                 cfs_duration_t duration = cfs_time_current() -
2984                                           scrub->os_time_last_checkpoint;
2985                 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
2986                                                      MSEC_PER_SEC);
2987                 __u32 rtime = sf->sf_run_time +
2988                               cfs_duration_sec(duration + HALF_SEC);
2989
2990                 if (duration != 0)
2991                         do_div(new_checked, duration);
2992                 if (rtime != 0)
2993                         do_div(speed, rtime);
2994                 rc = seq_printf(m, "run_time: %u seconds\n"
2995                               "average_speed: "LPU64" objects/sec\n"
2996                               "real-time_speed: "LPU64" objects/sec\n"
2997                               "current_position: %u\n"
2998                               "lf_scanned: "LPU64"\n"
2999                               "lf_reparied: "LPU64"\n"
3000                               "lf_failed: "LPU64"\n",
3001                               rtime, speed, new_checked, scrub->os_pos_current,
3002                               scrub->os_lf_scanned, scrub->os_lf_repaired,
3003                               scrub->os_lf_failed);
3004         } else {
3005                 if (sf->sf_run_time != 0)
3006                         do_div(speed, sf->sf_run_time);
3007                 rc = seq_printf(m, "run_time: %u seconds\n"
3008                               "average_speed: "LPU64" objects/sec\n"
3009                               "real-time_speed: N/A\n"
3010                               "current_position: N/A\n"
3011                               "lf_scanned: "LPU64"\n"
3012                               "lf_reparied: "LPU64"\n"
3013                               "lf_failed: "LPU64"\n",
3014                               sf->sf_run_time, speed, scrub->os_lf_scanned,
3015                               scrub->os_lf_repaired, scrub->os_lf_failed);
3016         }
3017
3018 out:
3019         up_read(&scrub->os_rwsem);
3020         return (rc < 0 ? -ENOSPC : 0);
3021 }