lustre/osd-ldiskfs/osd_scrub.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2012, 2013, Intel Corporation.
  24  */
  25 /*
  26  * lustre/osd-ldiskfs/osd_scrub.c
  27  *
  28  * Top-level entry points into osd module
  29  *
  30  * The OI scrub is used for rebuilding Object Index files when restores MDT from
  31  * file-level backup.
  32  *
  33  * The otable based iterator scans ldiskfs inode table to feed up layer LFSCK.
  34  *
  35  * Author: Fan Yong <yong.fan@whamcloud.com>
  36  */
  37
  38 #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        (HZ >> 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, flags = "LPX64"\n",
 275                osd_scrub2name(scrub), flags);
 276         memcpy(sf->sf_uuid, uuid, 16);
 277         sf->sf_status = SS_INIT;
 278         sf->sf_flags |= flags;
 279         sf->sf_run_time = 0;
 280         sf->sf_time_latest_start = 0;
 281         sf->sf_time_last_checkpoint = 0;
 282         sf->sf_pos_latest_start = 0;
 283         sf->sf_pos_last_checkpoint = 0;
 284         sf->sf_pos_first_inconsistent = 0;
 285         sf->sf_items_checked = 0;
 286         sf->sf_items_updated = 0;
 287         sf->sf_items_failed = 0;
 288         sf->sf_items_updated_prior = 0;
 289         sf->sf_items_noscrub = 0;
 290         sf->sf_items_igif = 0;
 291 }
 292
 293 static int osd_scrub_file_load(struct osd_scrub *scrub)
 294 {
 295         loff_t  pos  = 0;
 296         int     len  = sizeof(scrub->os_file_disk);
 297         int     rc;
 298
 299         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
 300         if (rc == len) {
 301                 struct scrub_file *sf = &scrub->os_file;
 302
 303                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
 304                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
 305                         CDEBUG(D_LFSCK, "%.16s: invalid scrub magic "
 306                                "0x%x != 0x%x\n", osd_scrub2name(scrub),
 307                                sf->sf_magic, SCRUB_MAGIC_V1);
 308                         /* Process it as new scrub file. */
 309                         rc = -ENOENT;
 310                 } else {
 311                         rc = 0;
 312                 }
 313         } else if (rc != 0) {
 314                 CDEBUG(D_LFSCK, "%.16s: fail to load scrub file, "
 315                        "expected = %d: rc = %d\n",
 316                        osd_scrub2name(scrub), len, rc);
 317                 if (rc > 0)
 318                         rc = -EFAULT;
 319         } else {
 320                 /* return -ENOENT for empty scrub file case. */
 321                 rc = -ENOENT;
 322         }
 323
 324         return rc;
 325 }
 326
 327 int osd_scrub_file_store(struct osd_scrub *scrub)
 328 {
 329         struct osd_device *dev;
 330         handle_t          *jh;
 331         loff_t             pos     = 0;
 332         int                len     = sizeof(scrub->os_file_disk);
 333         int                credits;
 334         int                rc;
 335
 336         dev = container_of0(scrub, struct osd_device, od_scrub);
 337         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
 338                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
 339         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits);
 340         if (IS_ERR(jh)) {
 341                 rc = PTR_ERR(jh);
 342                 CDEBUG(D_LFSCK, "%.16s: fail to start trans for scrub store: "
 343                        "rc = %d\n", osd_scrub2name(scrub), rc);
 344                 return rc;
 345         }
 346
 347         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
 348         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
 349                                       len, 0, &pos, jh);
 350         ldiskfs_journal_stop(jh);
 351         if (rc != 0)
 352                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file, "
 353                        "expected = %d: rc = %d\n",
 354                        osd_scrub2name(scrub), len, rc);
 355
 356         scrub->os_time_last_checkpoint = cfs_time_current();
 357         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
 358                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
 359         return rc;
 360 }
 361
 362 static int
 363 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
 364                      struct inode *inode, const struct lu_fid *fid)
 365 {
 366         struct filter_fid_old   *ff      = &info->oti_ff;
 367         struct dentry           *dentry  = &info->oti_obj_dentry;
 368         handle_t                *jh;
 369         int                      size    = 0;
 370         int                      rc;
 371         bool                     removed = false;
 372         bool                     reset   = true;
 373         ENTRY;
 374
 375         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
 376                 RETURN(0);
 377
 378         /* We want the LMA to fit into the 256-byte OST inode, so operate
 379          * as following:
 380          * 1) read old XATTR_NAME_FID and save the parent FID;
 381          * 2) delete the old XATTR_NAME_FID;
 382          * 3) make new LMA and add it;
 383          * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
 384          *
 385          * Making the LMA to fit into the 256-byte OST inode can save time for
 386          * normal osd_check_lma() and for other OI scrub scanning in future.
 387          * So it is worth to make some slow conversion here. */
 388         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 389                                 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
 390         if (IS_ERR(jh)) {
 391                 rc = PTR_ERR(jh);
 392                 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
 393                        DFID": rc = %d\n", osd_name(dev), PFID(fid), rc);
 394                 RETURN(rc);
 395         }
 396
 397         /* 1) read old XATTR_NAME_FID and save the parent FID */
 398         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 399         if (rc == sizeof(*ff)) {
 400                 /* 2) delete the old XATTR_NAME_FID */
 401                 ll_vfs_dq_init(inode);
 402                 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
 403                 if (rc != 0)
 404                         GOTO(stop, rc);
 405
 406                 removed = true;
 407         } else if (unlikely(rc == -ENODATA)) {
 408                 reset = false;
 409         } else if (rc != sizeof(struct filter_fid)) {
 410                 GOTO(stop, rc = -EINVAL);
 411         }
 412
 413         /* 3) make new LMA and add it */
 414         rc = osd_ea_fid_set(info, inode, fid, LMAC_FID_ON_OST, 0);
 415         if (rc == 0 && reset)
 416                 size = sizeof(struct filter_fid);
 417         else if (rc != 0 && removed)
 418                 /* If failed, we should try to add the old back. */
 419                 size = sizeof(struct filter_fid_old);
 420
 421         /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
 422         if (size > 0) {
 423                 int rc1;
 424
 425                 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
 426                                       XATTR_CREATE);
 427                 if (rc1 != 0 && rc == 0)
 428                         rc = rc1;
 429         }
 430
 431         GOTO(stop, rc);
 432
 433 stop:
 434         ldiskfs_journal_stop(jh);
 435         if (rc < 0)
 436                 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
 437                        osd_name(dev), PFID(fid), rc);
 438         return rc;
 439 }
 440
 441 static int
 442 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
 443                        struct osd_idmap_cache *oic, int val)
 444 {
 445         struct osd_scrub             *scrub  = &dev->od_scrub;
 446         struct scrub_file            *sf     = &scrub->os_file;
 447         struct lu_fid                *fid    = &oic->oic_fid;
 448         struct osd_inode_id          *lid    = &oic->oic_lid;
 449         struct osd_inode_id          *lid2   = &info->oti_id;
 450         struct osd_inconsistent_item *oii    = NULL;
 451         struct inode                 *inode  = NULL;
 452         int                           ops    = DTO_INDEX_UPDATE;
 453         int                           idx;
 454         int                           rc;
 455         bool                          converted = false;
 456         ENTRY;
 457
 458         down_write(&scrub->os_rwsem);
 459         scrub->os_new_checked++;
 460         if (val < 0)
 461                 GOTO(out, rc = val);
 462
 463         if (scrub->os_in_prior)
 464                 oii = list_entry(oic, struct osd_inconsistent_item,
 465                                  oii_cache);
 466
 467         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
 468                 GOTO(out, rc = 0);
 469
 470         if (fid_is_igif(fid))
 471                 sf->sf_items_igif++;
 472
 473         if (val == SCRUB_NEXT_OSTOBJ_OLD) {
 474                 inode = osd_iget(info, dev, lid);
 475                 if (IS_ERR(inode)) {
 476                         rc = PTR_ERR(inode);
 477                         /* Someone removed the inode. */
 478                         if (rc == -ENOENT || rc == -ESTALE)
 479                                 rc = 0;
 480                         GOTO(out, rc);
 481                 }
 482
 483                 sf->sf_flags |= SF_UPGRADE;
 484                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
 485                 dev->od_check_ff = 1;
 486                 rc = osd_scrub_convert_ff(info, dev, inode, fid);
 487                 if (rc != 0)
 488                         GOTO(out, rc);
 489
 490                 converted = true;
 491         }
 492
 493         if ((val == SCRUB_NEXT_NOLMA) &&
 494             (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
 495                 GOTO(out, rc = 0);
 496
 497         if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA))
 498                 goto iget;
 499
 500         rc = osd_oi_lookup(info, dev, fid, lid2,
 501                 (val == SCRUB_NEXT_OSTOBJ ||
 502                  val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 503         if (rc != 0) {
 504                 if (rc != -ENOENT && rc != -ESTALE)
 505                         GOTO(out, rc);
 506
 507 iget:
 508                 if (inode == NULL) {
 509                         inode = osd_iget(info, dev, lid);
 510                         if (IS_ERR(inode)) {
 511                                 rc = PTR_ERR(inode);
 512                                 /* Someone removed the inode. */
 513                                 if (rc == -ENOENT || rc == -ESTALE)
 514                                         rc = 0;
 515                                 GOTO(out, rc);
 516                         }
 517                 }
 518
 519                 scrub->os_full_speed = 1;
 520                 ops = DTO_INDEX_INSERT;
 521                 idx = osd_oi_fid2idx(dev, fid);
 522                 switch (val) {
 523                 case SCRUB_NEXT_NOLMA:
 524                         sf->sf_flags |= SF_UPGRADE;
 525                         if (!(sf->sf_param & SP_DRYRUN)) {
 526                                 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
 527                                 if (rc != 0)
 528                                         GOTO(out, rc);
 529                         }
 530
 531                         if (!(sf->sf_flags & SF_INCONSISTENT))
 532                                 dev->od_igif_inoi = 0;
 533                         break;
 534                 case SCRUB_NEXT_OSTOBJ:
 535                         sf->sf_flags |= SF_INCONSISTENT;
 536                 case SCRUB_NEXT_OSTOBJ_OLD:
 537                         break;
 538                 default:
 539                         sf->sf_flags |= SF_RECREATED;
 540                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 541                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 542                         break;
 543                 }
 544         } else if (osd_id_eq(lid, lid2)) {
 545                 if (converted)
 546                         sf->sf_items_updated++;
 547
 548                 GOTO(out, rc = 0);
 549         } else {
 550                 scrub->os_full_speed = 1;
 551                 sf->sf_flags |= SF_INCONSISTENT;
 552
 553                 /* XXX: If the device is restored from file-level backup, then
 554                  *      some IGIFs may have been already in OI files, and some
 555                  *      may be not yet. Means upgrading from 1.8 may be partly
 556                  *      processed, but some clients may hold some immobilized
 557                  *      IGIFs, and use them to access related objects. Under
 558                  *      such case, OSD does not know whether an given IGIF has
 559                  *      been processed or to be processed, and it also cannot
 560                  *      generate local ino#/gen# directly from the immobilized
 561                  *      IGIF because of the backup/restore. Then force OSD to
 562                  *      lookup the given IGIF in OI files, and if no entry,
 563                  *      then ask the client to retry after upgrading completed.
 564                  *      No better choice. */
 565                 dev->od_igif_inoi = 1;
 566         }
 567
 568         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
 569                         (val == SCRUB_NEXT_OSTOBJ ||
 570                          val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 571         if (rc == 0) {
 572                 if (scrub->os_in_prior)
 573                         sf->sf_items_updated_prior++;
 574                 else
 575                         sf->sf_items_updated++;
 576
 577                 /* The target has been changed, need to be re-loaded. */
 578                 lu_object_purge(info->oti_env, osd2lu_dev(dev), fid);
 579         }
 580
 581         GOTO(out, rc);
 582
 583 out:
 584         if (rc < 0) {
 585                 sf->sf_items_failed++;
 586                 if (sf->sf_pos_first_inconsistent == 0 ||
 587                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 588                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 589         } else {
 590                 rc = 0;
 591         }
 592
 593         /* There may be conflict unlink during the OI scrub,
 594          * if happend, then remove the new added OI mapping. */
 595         if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
 596             unlikely(inode->i_nlink == 0))
 597                 osd_scrub_refresh_mapping(info, dev, fid, lid,
 598                                 DTO_INDEX_DELETE, false,
 599                                 (val == SCRUB_NEXT_OSTOBJ ||
 600                                  val == SCRUB_NEXT_OSTOBJ_OLD) ?
 601                                 OI_KNOWN_ON_OST : 0);
 602         up_write(&scrub->os_rwsem);
 603
 604         if (inode != NULL && !IS_ERR(inode))
 605                 iput(inode);
 606
 607         if (oii != NULL) {
 608                 LASSERT(!list_empty(&oii->oii_list));
 609
 610                 spin_lock(&scrub->os_lock);
 611                 list_del_init(&oii->oii_list);
 612                 spin_unlock(&scrub->os_lock);
 613                 OBD_FREE_PTR(oii);
 614         }
 615         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 616 }
 617
 618 /* OI scrub APIs */
 619
 620 static int osd_scrub_prep(struct osd_device *dev)
 621 {
 622         struct osd_scrub     *scrub  = &dev->od_scrub;
 623         struct ptlrpc_thread *thread = &scrub->os_thread;
 624         struct scrub_file    *sf     = &scrub->os_file;
 625         __u32                 flags  = scrub->os_start_flags;
 626         int                   rc;
 627         bool                  drop_dryrun = false;
 628         ENTRY;
 629
 630         down_write(&scrub->os_rwsem);
 631         if (flags & SS_SET_FAILOUT)
 632                 sf->sf_param |= SP_FAILOUT;
 633
 634         if (flags & SS_CLEAR_FAILOUT)
 635                 sf->sf_param &= ~SP_FAILOUT;
 636
 637         if (flags & SS_SET_DRYRUN)
 638                 sf->sf_param |= SP_DRYRUN;
 639
 640         if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
 641                 sf->sf_param &= ~SP_DRYRUN;
 642                 drop_dryrun = true;
 643         }
 644
 645         if (flags & SS_RESET)
 646                 osd_scrub_file_reset(scrub,
 647                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
 648
 649         if (flags & SS_AUTO) {
 650                 scrub->os_full_speed = 1;
 651                 sf->sf_flags |= SF_AUTO;
 652                 /* For the case of OI scrub auto triggered, NOT dryrun. */
 653                 sf->sf_param &= ~SP_FAILOUT;
 654         } else {
 655                 scrub->os_full_speed = 0;
 656         }
 657
 658         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
 659                 scrub->os_full_speed = 1;
 660
 661         scrub->os_in_prior = 0;
 662         spin_lock(&scrub->os_lock);
 663         scrub->os_waiting = 0;
 664         scrub->os_paused = 0;
 665         spin_unlock(&scrub->os_lock);
 666         scrub->os_new_checked = 0;
 667         if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
 668                 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
 669         else if (sf->sf_pos_last_checkpoint != 0)
 670                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 671         else
 672                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 673
 674         scrub->os_pos_current = sf->sf_pos_latest_start;
 675         sf->sf_status = SS_SCANNING;
 676         sf->sf_time_latest_start = cfs_time_current_sec();
 677         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 678         rc = osd_scrub_file_store(scrub);
 679         if (rc == 0) {
 680                 spin_lock(&scrub->os_lock);
 681                 thread_set_flags(thread, SVC_RUNNING);
 682                 spin_unlock(&scrub->os_lock);
 683                 wake_up_all(&thread->t_ctl_waitq);
 684         }
 685         up_write(&scrub->os_rwsem);
 686
 687         RETURN(rc);
 688 }
 689
 690 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
 691 {
 692         struct scrub_file *sf = &scrub->os_file;
 693         int                rc;
 694
 695         if (likely(cfs_time_before(cfs_time_current(),
 696                                    scrub->os_time_next_checkpoint) ||
 697                    scrub->os_new_checked == 0))
 698                 return 0;
 699
 700         down_write(&scrub->os_rwsem);
 701         sf->sf_items_checked += scrub->os_new_checked;
 702         scrub->os_new_checked = 0;
 703         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 704         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 705         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 706                                             scrub->os_time_last_checkpoint);
 707         rc = osd_scrub_file_store(scrub);
 708         up_write(&scrub->os_rwsem);
 709
 710         return rc;
 711 }
 712
 713 static void osd_scrub_post(struct osd_scrub *scrub, int result)
 714 {
 715         struct scrub_file *sf = &scrub->os_file;
 716         ENTRY;
 717
 718         down_write(&scrub->os_rwsem);
 719         spin_lock(&scrub->os_lock);
 720         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 721         spin_unlock(&scrub->os_lock);
 722         if (scrub->os_new_checked > 0) {
 723                 sf->sf_items_checked += scrub->os_new_checked;
 724                 scrub->os_new_checked = 0;
 725                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 726         }
 727         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 728         if (result > 0) {
 729                 struct osd_device *dev =
 730                         container_of0(scrub, struct osd_device, od_scrub);
 731
 732                 dev->od_igif_inoi = 1;
 733                 dev->od_check_ff = 0;
 734                 sf->sf_status = SS_COMPLETED;
 735                 if (!(sf->sf_param & SP_DRYRUN)) {
 736                         memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 737                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 738                                           SF_UPGRADE | SF_AUTO);
 739                 }
 740                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 741                 sf->sf_success_count++;
 742         } else if (result == 0) {
 743                 if (scrub->os_paused)
 744                         sf->sf_status = SS_PAUSED;
 745                 else
 746                         sf->sf_status = SS_STOPPED;
 747         } else {
 748                 sf->sf_status = SS_FAILED;
 749         }
 750         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 751                                             scrub->os_time_last_checkpoint);
 752         result = osd_scrub_file_store(scrub);
 753         up_write(&scrub->os_rwsem);
 754
 755         EXIT;
 756 }
 757
 758 /* iteration engine */
 759
 760 struct osd_iit_param {
 761         struct super_block *sb;
 762         struct buffer_head *bitmap;
 763         ldiskfs_group_t bg;
 764         __u32 gbase;
 765         __u32 offset;
 766 };
 767
 768 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 769                                    struct osd_device *dev,
 770                                    struct osd_iit_param *param,
 771                                    struct osd_idmap_cache **oic,
 772                                    int noslot);
 773
 774 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 775                                    struct osd_device *dev,
 776                                    struct osd_iit_param *param,
 777                                    struct osd_idmap_cache *oic,
 778                                    int *noslot, int rc);
 779
 780 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 781 {
 782         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 783                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 784         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 785                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 786                 return SCRUB_NEXT_BREAK;
 787         } else {
 788                 *pos = param->gbase + param->offset;
 789                 return 0;
 790         }
 791 }
 792
 793 /**
 794  * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
 795  * \retval 0: FID-on-MDT
 796  */
 797 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
 798                                       struct osd_device *dev,
 799                                       struct lu_fid *fid)
 800 {
 801         /* XXX: The initial OI scrub will scan the top level /O to generate
 802          *      a small local FLDB according to the <seq>. If the given FID
 803          *      is in the local FLDB, then it is FID-on-OST; otherwise it's
 804          *      quite possible for FID-on-MDT. */
 805         if (dev->od_is_ost)
 806                 return SCRUB_NEXT_OSTOBJ_OLD;
 807         else
 808                 return 0;
 809 }
 810
 811 static int osd_scrub_get_fid(struct osd_thread_info *info,
 812                              struct osd_device *dev, struct inode *inode,
 813                              struct lu_fid *fid, bool scrub)
 814 {
 815         struct lustre_mdt_attrs *lma     = &info->oti_mdt_attrs;
 816         int                      rc;
 817         bool                     has_lma = false;
 818
 819         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 820         if (rc == 0) {
 821                 has_lma = true;
 822                 if (lma->lma_compat & LMAC_NOT_IN_OI ||
 823                     lma->lma_incompat & LMAI_AGENT)
 824                         return SCRUB_NEXT_CONTINUE;
 825
 826                 *fid = lma->lma_self_fid;
 827                 if (!scrub)
 828                         return 0;
 829
 830                 if (lma->lma_compat & LMAC_FID_ON_OST)
 831                         return SCRUB_NEXT_OSTOBJ;
 832
 833                 if (fid_is_idif(fid))
 834                         return SCRUB_NEXT_OSTOBJ_OLD;
 835
 836                 /* For local object. */
 837                 if (fid_is_internal(fid))
 838                         return 0;
 839
 840                 /* For external visible MDT-object with non-normal FID. */
 841                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
 842                         return 0;
 843
 844                 /* For the object with normal FID, it may be MDT-object,
 845                  * or may be 2.4 OST-object, need further distinguish.
 846                  * Fall through to next section. */
 847         }
 848
 849         if (rc == -ENODATA || rc == 0) {
 850                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
 851                 if (rc == 0) {
 852                         if (scrub)
 853                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
 854                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 855                         return rc;
 856                 }
 857
 858                 if (rc > 0) {
 859                         if (!has_lma)
 860                                 /* It is FID-on-OST, but we do not know how
 861                                  * to generate its FID, ignore it directly. */
 862                                 rc = SCRUB_NEXT_CONTINUE;
 863                         else
 864                                 /* It is 2.4 OST-object. */
 865                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 866                         return rc;
 867                 }
 868
 869                 if (rc != -ENODATA)
 870                         return rc;
 871
 872                 if (!has_lma) {
 873                         if (dev->od_scrub.os_convert_igif) {
 874                                 lu_igif_build(fid, inode->i_ino,
 875                                               inode->i_generation);
 876                                 if (scrub)
 877                                         rc = SCRUB_NEXT_NOLMA;
 878                                 else
 879                                         rc = 0;
 880                         } else {
 881                                 /* It may be FID-on-OST, or may be FID for
 882                                  * non-MDT0, anyway, we do not know how to
 883                                  * generate its FID, ignore it directly. */
 884                                 rc = SCRUB_NEXT_CONTINUE;
 885                         }
 886                         return rc;
 887                 }
 888
 889                 /* For OI scrub case only: the object has LMA but has no ff
 890                  * (or ff crashed). It may be MDT-object, may be OST-object
 891                  * with crashed ff. The last check is local FLDB. */
 892                 rc = osd_scrub_check_local_fldb(info, dev, fid);
 893         }
 894
 895         return rc;
 896 }
 897
 898 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 899                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 900                         struct super_block *sb, bool scrub)
 901 {
 902         struct inode *inode;
 903         int           rc;
 904         ENTRY;
 905
 906         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 907         inode = osd_iget(info, dev, lid);
 908         if (IS_ERR(inode)) {
 909                 rc = PTR_ERR(inode);
 910                 /* The inode may be removed after bitmap searching, or the
 911                  * file is new created without inode initialized yet. */
 912                 if (rc == -ENOENT || rc == -ESTALE)
 913                         RETURN(SCRUB_NEXT_CONTINUE);
 914
 915                 CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: "
 916                        "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name,
 917                        pos, rc);
 918                 RETURN(rc);
 919         }
 920
 921         if (scrub &&
 922             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 923                 /* Only skip it for the first OI scrub accessing. */
 924                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 925                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
 926         }
 927
 928         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
 929
 930         GOTO(put, rc);
 931
 932 put:
 933         iput(inode);
 934         return rc;
 935 }
 936
 937 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 938                           struct osd_iit_param *param,
 939                           struct osd_idmap_cache **oic, int noslot)
 940 {
 941         struct osd_scrub     *scrub  = &dev->od_scrub;
 942         struct ptlrpc_thread *thread = &scrub->os_thread;
 943         struct lu_fid        *fid;
 944         struct osd_inode_id  *lid;
 945         int                   rc;
 946
 947         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
 948                 struct l_wait_info lwi;
 949
 950                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
 951                 l_wait_event(thread->t_ctl_waitq,
 952                              !list_empty(&scrub->os_inconsistent_items) ||
 953                              !thread_is_running(thread),
 954                              &lwi);
 955         }
 956
 957         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 958                 spin_lock(&scrub->os_lock);
 959                 thread_set_flags(thread, SVC_STOPPING);
 960                 spin_unlock(&scrub->os_lock);
 961                 return SCRUB_NEXT_CRASH;
 962         }
 963
 964         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
 965                 return SCRUB_NEXT_FATAL;
 966
 967         if (unlikely(!thread_is_running(thread)))
 968                 return SCRUB_NEXT_EXIT;
 969
 970         if (!list_empty(&scrub->os_inconsistent_items)) {
 971                 struct osd_inconsistent_item *oii;
 972
 973                 oii = list_entry(scrub->os_inconsistent_items.next,
 974                                  struct osd_inconsistent_item, oii_list);
 975                 *oic = &oii->oii_cache;
 976                 scrub->os_in_prior = 1;
 977                 return 0;
 978         }
 979
 980         if (noslot != 0)
 981                 return SCRUB_NEXT_WAIT;
 982
 983         rc = osd_iit_next(param, &scrub->os_pos_current);
 984         if (rc != 0)
 985                 return rc;
 986
 987         *oic = &scrub->os_oic;
 988         fid = &(*oic)->oic_fid;
 989         lid = &(*oic)->oic_lid;
 990         rc = osd_iit_iget(info, dev, fid, lid,
 991                           scrub->os_pos_current, param->sb, true);
 992         return rc;
 993 }
 994
 995 static int osd_preload_next(struct osd_thread_info *info,
 996                             struct osd_device *dev, struct osd_iit_param *param,
 997                             struct osd_idmap_cache **oic, int noslot)
 998 {
 999         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
1000         struct osd_scrub        *scrub;
1001         struct ptlrpc_thread    *thread;
1002         int                      rc;
1003
1004         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1005         if (rc != 0)
1006                 return rc;
1007
1008         scrub = &dev->od_scrub;
1009         thread = &scrub->os_thread;
1010         if (thread_is_running(thread) &&
1011             ooc->ooc_pos_preload >= scrub->os_pos_current)
1012                 return SCRUB_NEXT_EXIT;
1013
1014         rc = osd_iit_iget(info, dev,
1015                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1016                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1017                           ooc->ooc_pos_preload, param->sb, false);
1018         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1019          * ignore the failure, so it still need to skip the inode next time. */
1020         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1021         return rc;
1022 }
1023
1024 static inline int
1025 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1026 {
1027         spin_lock(&scrub->os_lock);
1028         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1029             !list_empty(&scrub->os_inconsistent_items) ||
1030             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1031                 scrub->os_waiting = 0;
1032         else
1033                 scrub->os_waiting = 1;
1034         spin_unlock(&scrub->os_lock);
1035
1036         return !scrub->os_waiting;
1037 }
1038
1039 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1040                           struct osd_iit_param *param,
1041                           struct osd_idmap_cache *oic, int *noslot, int rc)
1042 {
1043         struct l_wait_info       lwi    = { 0 };
1044         struct osd_scrub        *scrub  = &dev->od_scrub;
1045         struct scrub_file       *sf     = &scrub->os_file;
1046         struct ptlrpc_thread    *thread = &scrub->os_thread;
1047         struct osd_otable_it    *it     = dev->od_otable_it;
1048         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
1049
1050         switch (rc) {
1051         case SCRUB_NEXT_CONTINUE:
1052                 goto next;
1053         case SCRUB_NEXT_WAIT:
1054                 goto wait;
1055         case SCRUB_NEXT_NOSCRUB:
1056                 down_write(&scrub->os_rwsem);
1057                 scrub->os_new_checked++;
1058                 sf->sf_items_noscrub++;
1059                 up_write(&scrub->os_rwsem);
1060                 goto next;
1061         }
1062
1063         rc = osd_scrub_check_update(info, dev, oic, rc);
1064         if (rc != 0)
1065                 return rc;
1066
1067         rc = osd_scrub_checkpoint(scrub);
1068         if (rc != 0) {
1069                 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1070                        "rc = %d\n", osd_scrub2name(scrub),
1071                        scrub->os_pos_current, rc);
1072                 /* Continue, as long as the scrub itself can go ahead. */
1073         }
1074
1075         if (scrub->os_in_prior) {
1076                 scrub->os_in_prior = 0;
1077                 return 0;
1078         }
1079
1080 next:
1081         scrub->os_pos_current = param->gbase + ++(param->offset);
1082
1083 wait:
1084         if (it != NULL && it->ooi_waiting && ooc != NULL &&
1085             ooc->ooc_pos_preload < scrub->os_pos_current) {
1086                 spin_lock(&scrub->os_lock);
1087                 it->ooi_waiting = 0;
1088                 wake_up_all(&thread->t_ctl_waitq);
1089                 spin_unlock(&scrub->os_lock);
1090         }
1091
1092         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1093                 return 0;
1094
1095         if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1096                 *noslot = 0;
1097                 return 0;
1098         }
1099
1100         if (it != NULL)
1101                 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1102                              &lwi);
1103
1104         if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1105                 *noslot = 0;
1106         else
1107                 *noslot = 1;
1108         return 0;
1109 }
1110
1111 static int osd_preload_exec(struct osd_thread_info *info,
1112                             struct osd_device *dev, struct osd_iit_param *param,
1113                             struct osd_idmap_cache *oic, int *noslot, int rc)
1114 {
1115         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1116
1117         if (rc == 0) {
1118                 ooc->ooc_cached_items++;
1119                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1120                                         ~OSD_OTABLE_IT_CACHE_MASK;
1121         }
1122         return rc > 0 ? 0 : rc;
1123 }
1124
1125 #define SCRUB_IT_ALL    1
1126 #define SCRUB_IT_CRASH  2
1127
1128 static int osd_inode_iteration(struct osd_thread_info *info,
1129                                struct osd_device *dev, __u32 max, bool preload)
1130 {
1131         osd_iit_next_policy   next;
1132         osd_iit_exec_policy   exec;
1133         __u32                *pos;
1134         __u32                *count;
1135         struct osd_iit_param  param;
1136         __u32                 limit;
1137         int                   noslot = 0;
1138         int                   rc;
1139         ENTRY;
1140
1141         if (!preload) {
1142                 struct osd_scrub *scrub = &dev->od_scrub;
1143
1144                 next = osd_scrub_next;
1145                 exec = osd_scrub_exec;
1146                 pos = &scrub->os_pos_current;
1147                 count = &scrub->os_new_checked;
1148         } else {
1149                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1150
1151                 next = osd_preload_next;
1152                 exec = osd_preload_exec;
1153                 pos = &ooc->ooc_pos_preload;
1154                 count = &ooc->ooc_cached_items;
1155         }
1156         param.sb = osd_sb(dev);
1157         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1158
1159         while (*pos <= limit && *count < max) {
1160                 struct osd_idmap_cache *oic = NULL;
1161
1162                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1163                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1164                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1165                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1166                 if (param.bitmap == NULL) {
1167                         CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1168                                "scrub will stop, urgent mode\n",
1169                                LDISKFS_SB(param.sb)->s_es->s_volume_name,
1170                                (__u32)param.bg);
1171                         RETURN(-EIO);
1172                 }
1173
1174                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1175                        *count < max) {
1176                         rc = next(info, dev, &param, &oic, noslot);
1177                         switch (rc) {
1178                         case SCRUB_NEXT_BREAK:
1179                                 goto next_group;
1180                         case SCRUB_NEXT_EXIT:
1181                                 brelse(param.bitmap);
1182                                 RETURN(0);
1183                         case SCRUB_NEXT_CRASH:
1184                                 brelse(param.bitmap);
1185                                 RETURN(SCRUB_IT_CRASH);
1186                         case SCRUB_NEXT_FATAL:
1187                                 brelse(param.bitmap);
1188                                 RETURN(-EINVAL);
1189                         }
1190
1191                         rc = exec(info, dev, &param, oic, &noslot, rc);
1192                         if (rc != 0) {
1193                                 brelse(param.bitmap);
1194                                 RETURN(rc);
1195                         }
1196                 }
1197
1198 next_group:
1199                 brelse(param.bitmap);
1200         }
1201
1202         if (*pos > limit)
1203                 RETURN(SCRUB_IT_ALL);
1204         RETURN(0);
1205 }
1206
1207 static int osd_otable_it_preload(const struct lu_env *env,
1208                                  struct osd_otable_it *it)
1209 {
1210         struct osd_device       *dev   = it->ooi_dev;
1211         struct osd_scrub        *scrub = &dev->od_scrub;
1212         struct osd_otable_cache *ooc   = &it->ooi_cache;
1213         int                      rc;
1214         ENTRY;
1215
1216         rc = osd_inode_iteration(osd_oti_get(env), dev,
1217                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1218         if (rc == SCRUB_IT_ALL)
1219                 it->ooi_all_cached = 1;
1220
1221         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1222                 spin_lock(&scrub->os_lock);
1223                 scrub->os_waiting = 0;
1224                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1225                 spin_unlock(&scrub->os_lock);
1226         }
1227
1228         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1229 }
1230
1231 static int osd_scrub_main(void *args)
1232 {
1233         struct lu_env         env;
1234         struct osd_device    *dev    = (struct osd_device *)args;
1235         struct osd_scrub     *scrub  = &dev->od_scrub;
1236         struct ptlrpc_thread *thread = &scrub->os_thread;
1237         int                   rc;
1238         ENTRY;
1239
1240         rc = lu_env_init(&env, LCT_LOCAL);
1241         if (rc != 0) {
1242                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1243                        osd_scrub2name(scrub), rc);
1244                 GOTO(noenv, rc);
1245         }
1246
1247         rc = osd_scrub_prep(dev);
1248         if (rc != 0) {
1249                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1250                        osd_scrub2name(scrub), rc);
1251                 GOTO(out, rc);
1252         }
1253
1254         if (!scrub->os_full_speed) {
1255                 struct l_wait_info lwi = { 0 };
1256                 struct osd_otable_it *it = dev->od_otable_it;
1257                 struct osd_otable_cache *ooc = &it->ooi_cache;
1258
1259                 l_wait_event(thread->t_ctl_waitq,
1260                              it->ooi_user_ready || !thread_is_running(thread),
1261                              &lwi);
1262                 if (unlikely(!thread_is_running(thread)))
1263                         GOTO(post, rc = 0);
1264
1265                 scrub->os_pos_current = ooc->ooc_pos_preload;
1266         }
1267
1268         CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1269                osd_scrub2name(scrub), scrub->os_start_flags,
1270                scrub->os_pos_current);
1271
1272         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1273         if (unlikely(rc == SCRUB_IT_CRASH))
1274                 GOTO(out, rc = -EINVAL);
1275         GOTO(post, rc);
1276
1277 post:
1278         osd_scrub_post(scrub, rc);
1279         CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1280                osd_scrub2name(scrub), scrub->os_pos_current, rc);
1281
1282 out:
1283         while (!list_empty(&scrub->os_inconsistent_items)) {
1284                 struct osd_inconsistent_item *oii;
1285
1286                 oii = list_entry(scrub->os_inconsistent_items.next,
1287                                      struct osd_inconsistent_item, oii_list);
1288                 list_del_init(&oii->oii_list);
1289                 OBD_FREE_PTR(oii);
1290         }
1291         lu_env_fini(&env);
1292
1293 noenv:
1294         spin_lock(&scrub->os_lock);
1295         thread_set_flags(thread, SVC_STOPPED);
1296         wake_up_all(&thread->t_ctl_waitq);
1297         spin_unlock(&scrub->os_lock);
1298         return rc;
1299 }
1300
1301 /* initial OI scrub */
1302
1303 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1304                          struct dentry *, filldir_t filldir);
1305
1306 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1307                                loff_t offset, __u64 ino, unsigned d_type);
1308 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1309                            loff_t offset, __u64 ino, unsigned d_type);
1310 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1311                            loff_t offset, __u64 ino, unsigned d_type);
1312
1313 static int
1314 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1315                      struct dentry *dentry, filldir_t filldir);
1316 static int
1317 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1318                   struct dentry *dentry, filldir_t filldir);
1319
1320 static int
1321 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1322                      struct dentry *dentry, filldir_t filldir);
1323
1324 enum osd_lf_flags {
1325         OLF_SCAN_SUBITEMS       = 0x0001,
1326         OLF_HIDE_FID            = 0x0002,
1327         OLF_SHOW_NAME           = 0x0004,
1328         OLF_NO_OI               = 0x0008,
1329 };
1330
1331 struct osd_lf_map {
1332         char            *olm_name;
1333         struct lu_fid    olm_fid;
1334         __u16            olm_flags;
1335         scandir_t        olm_scandir;
1336         filldir_t        olm_filldir;
1337 };
1338
1339 /* Add the new introduced local files in the list in the future. */
1340 static const struct osd_lf_map osd_lf_maps[] = {
1341         /* CATALOGS */
1342         { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1343                 NULL, NULL },
1344
1345         /* CONFIGS */
1346         { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1347                 OLF_SCAN_SUBITEMS, osd_ios_general_scan,
1348                 osd_ios_varfid_fill },
1349
1350         /* NIDTBL_VERSIONS */
1351         { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1352                 osd_ios_general_scan, osd_ios_varfid_fill },
1353
1354         /* PENDING */
1355         { "PENDING", { 0, 0, 0 }, 0, NULL, NULL },
1356
1357         /* ROOT */
1358         { "ROOT", { FID_SEQ_ROOT, 1, 0 },
1359                 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, osd_ios_ROOT_scan, NULL },
1360
1361         /* changelog_catalog */
1362         { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, NULL, NULL },
1363
1364         /* changelog_users */
1365         { CHANGELOG_USERS, { 0, 0, 0 }, 0, NULL, NULL },
1366
1367         /* fld */
1368         { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1369                 NULL, NULL },
1370
1371         /* last_rcvd */
1372         { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1373                 NULL, NULL },
1374
1375         /* lfsck_bookmark */
1376         { "lfsck_bookmark", { 0, 0, 0 }, 0, NULL, NULL },
1377
1378         /* lov_objid */
1379         { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1380                 NULL, NULL },
1381
1382         /* lov_objseq */
1383         { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1384                 OLF_SHOW_NAME, NULL, NULL },
1385
1386         /* quota_master */
1387         { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1388                 osd_ios_general_scan, osd_ios_varfid_fill },
1389
1390         /* quota_slave */
1391         { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1392                 osd_ios_general_scan, osd_ios_varfid_fill },
1393
1394         /* seq_ctl */
1395         { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1396                 OLF_SHOW_NAME, NULL, NULL },
1397
1398         /* seq_srv */
1399         { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1400                 OLF_SHOW_NAME, NULL, NULL },
1401
1402         /* health_check */
1403         { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1404                 OLF_SHOW_NAME, NULL, NULL },
1405
1406         /* lfsck_namespace */
1407         { "lfsck_namespace", { 0, 0, 0 }, 0, NULL, NULL },
1408
1409         /* OBJECTS, upgrade from old device */
1410         { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, osd_ios_OBJECTS_scan, NULL },
1411
1412         /* lquota_v2.user, upgrade from old device */
1413         { "lquota_v2.user", { 0, 0, 0 }, 0, NULL, NULL },
1414
1415         /* lquota_v2.group, upgrade from old device */
1416         { "lquota_v2.group", { 0, 0, 0 }, 0, NULL, NULL },
1417
1418         /* LAST_GROUP, upgrade from old device */
1419         { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1420                 OLF_SHOW_NAME, NULL, NULL },
1421
1422         /* SLAVE_LOG, llog for destroy slave stripes of striped dir */
1423         { "SLAVE_LOG", { FID_SEQ_LOCAL_FILE, SLAVE_LLOG_CATALOGS_OID, 0 },
1424                OLF_SHOW_NAME, NULL, NULL },
1425
1426         /* lost+found */
1427         { "lost+found", { 0, 0, 0 }, OLF_SCAN_SUBITEMS | OLF_NO_OI,
1428                 osd_ios_general_scan, osd_ios_lf_fill },
1429
1430         { NULL, { 0, 0, 0 }, 0, NULL, NULL }
1431 };
1432
1433 /* Add the new introduced files under .lustre/ in the list in the future. */
1434 static const struct osd_lf_map osd_dl_maps[] = {
1435         /* .lustre/fid */
1436         { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1437                 NULL, NULL },
1438
1439         /* .lustre/lost+found */
1440         { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1441                 NULL, NULL },
1442
1443         { NULL, { 0, 0, 0 }, 0, NULL, NULL }
1444 };
1445
1446 struct osd_ios_item {
1447         struct list_head oii_list;
1448         struct dentry   *oii_dentry;
1449         scandir_t        oii_scandir;
1450         filldir_t        oii_filldir;
1451 };
1452
1453 struct osd_ios_filldir_buf {
1454 #ifdef HAVE_DIR_CONTEXT
1455         /* please keep it as first member */
1456         struct dir_context       ctx;
1457 #endif
1458         struct osd_thread_info  *oifb_info;
1459         struct osd_device       *oifb_dev;
1460         struct dentry           *oifb_dentry;
1461 };
1462
1463 static inline struct dentry *
1464 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1465 {
1466         struct dentry *dentry;
1467
1468         dentry = ll_lookup_one_len(name, parent, namelen);
1469         if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1470                 dput(dentry);
1471                 return ERR_PTR(-ENOENT);
1472         }
1473
1474         return dentry;
1475 }
1476
1477 static int
1478 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1479                  scandir_t scandir, filldir_t filldir)
1480 {
1481         struct osd_ios_item *item;
1482         ENTRY;
1483
1484         OBD_ALLOC_PTR(item);
1485         if (item == NULL)
1486                 RETURN(-ENOMEM);
1487
1488         INIT_LIST_HEAD(&item->oii_list);
1489         item->oii_dentry = dget(dentry);
1490         item->oii_scandir = scandir;
1491         item->oii_filldir = filldir;
1492         list_add_tail(&item->oii_list, &dev->od_ios_list);
1493
1494         RETURN(0);
1495 }
1496
1497 /**
1498  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1499  *
1500  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1501  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1502  * reference the inode, or fixed if it is missing or references another inode.
1503  */
1504 static int
1505 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1506                  struct inode *inode, const struct lu_fid *fid, int flags)
1507 {
1508         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
1509         struct osd_inode_id     *id     = &info->oti_id;
1510         struct osd_inode_id     *id2    = &info->oti_id2;
1511         struct osd_scrub        *scrub  = &dev->od_scrub;
1512         struct scrub_file       *sf     = &scrub->os_file;
1513         struct lu_fid            tfid;
1514         int                      rc;
1515         ENTRY;
1516
1517         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1518         if (rc != 0 && rc != -ENODATA) {
1519                 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1520                        "rc = %d\n", osd_name(dev), rc);
1521
1522                 RETURN(rc);
1523         }
1524
1525         osd_id_gen(id, inode->i_ino, inode->i_generation);
1526         if (rc == -ENODATA) {
1527                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID)
1528                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1529                 else
1530                         tfid = *fid;
1531                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1532                 if (rc != 0) {
1533                         CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1534                               "scrub: rc = %d\n", osd_name(dev), rc);
1535
1536                         RETURN(rc);
1537                 }
1538         } else {
1539                 if (lma->lma_compat & LMAC_NOT_IN_OI)
1540                         RETURN(0);
1541
1542                 tfid = lma->lma_self_fid;
1543         }
1544
1545         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1546         if (rc != 0) {
1547                 if (rc != -ENOENT)
1548                         RETURN(rc);
1549
1550                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1551                                                DTO_INDEX_INSERT, true, 0);
1552                 if (rc > 0)
1553                         rc = 0;
1554
1555                 RETURN(rc);
1556         }
1557
1558         if (osd_id_eq_strict(id, id2))
1559                 RETURN(0);
1560
1561         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1562                 osd_scrub_file_reset(scrub,
1563                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1564                                      SF_INCONSISTENT);
1565                 rc = osd_scrub_file_store(scrub);
1566                 if (rc != 0)
1567                         RETURN(rc);
1568         }
1569
1570         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1571                                        DTO_INDEX_UPDATE, true, 0);
1572         if (rc > 0)
1573                 rc = 0;
1574
1575         RETURN(rc);
1576 }
1577
1578 /**
1579  * It scans the /lost+found, and for the OST-object (with filter_fid
1580  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1581  */
1582 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1583                            loff_t offset, __u64 ino, unsigned d_type)
1584 {
1585         struct osd_ios_filldir_buf *fill_buf = buf;
1586         struct osd_thread_info     *info     = fill_buf->oifb_info;
1587         struct osd_device          *dev      = fill_buf->oifb_dev;
1588         struct lu_fid              *fid      = &info->oti_fid;
1589         struct osd_scrub           *scrub    = &dev->od_scrub;
1590         struct dentry              *parent   = fill_buf->oifb_dentry;
1591         struct dentry              *child;
1592         struct inode               *dir      = parent->d_inode;
1593         struct inode               *inode;
1594         int                         rc;
1595         ENTRY;
1596
1597         /* skip any '.' started names */
1598         if (name[0] == '.')
1599                 RETURN(0);
1600
1601         scrub->os_lf_scanned++;
1602         child = osd_ios_lookup_one_len(name, parent, namelen);
1603         if (IS_ERR(child)) {
1604                 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1605                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
1606                 RETURN(0);
1607         }
1608
1609         inode = child->d_inode;
1610         if (S_ISDIR(inode->i_mode)) {
1611                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1612                                       osd_ios_lf_fill);
1613                 if (rc != 0)
1614                         CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1615                               "rc = %d\n", osd_name(dev), namelen, name, rc);
1616                 GOTO(put, rc);
1617         }
1618
1619         if (!S_ISREG(inode->i_mode))
1620                 GOTO(put, rc = 0);
1621
1622         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1623         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1624                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1625                 if (rc == 0) {
1626                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1627                                "/lost+found.\n", namelen, name, PFID(fid));
1628                         scrub->os_lf_repaired++;
1629                 } else {
1630                         CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1631                                DFID": rc = %d\n",
1632                                osd_name(dev), namelen, name, PFID(fid), rc);
1633                 }
1634         }
1635
1636         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1637          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1638          *      can process them in furtuer. */
1639
1640         GOTO(put, rc);
1641
1642 put:
1643         if (rc < 0)
1644                 scrub->os_lf_failed++;
1645         dput(child);
1646         /* skip the failure to make the scanning to continue. */
1647         return 0;
1648 }
1649
1650 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1651                                loff_t offset, __u64 ino, unsigned d_type)
1652 {
1653         struct osd_ios_filldir_buf *fill_buf = buf;
1654         struct osd_device          *dev      = fill_buf->oifb_dev;
1655         struct dentry              *child;
1656         int                         rc;
1657         ENTRY;
1658
1659         /* skip any '.' started names */
1660         if (name[0] == '.')
1661                 RETURN(0);
1662
1663         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1664         if (IS_ERR(child))
1665                 RETURN(PTR_ERR(child));
1666
1667         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1668                               NULL, 0);
1669         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1670                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1671                                       osd_ios_varfid_fill);
1672         dput(child);
1673
1674         RETURN(rc);
1675 }
1676
1677 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1678                            loff_t offset, __u64 ino, unsigned d_type)
1679 {
1680         struct osd_ios_filldir_buf *fill_buf = buf;
1681         struct osd_device          *dev      = fill_buf->oifb_dev;
1682         const struct osd_lf_map    *map;
1683         struct dentry              *child;
1684         int                         rc       = 0;
1685         ENTRY;
1686
1687         /* skip any '.' started names */
1688         if (name[0] == '.')
1689                 RETURN(0);
1690
1691         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
1692                 if (strlen(map->olm_name) != namelen)
1693                         continue;
1694
1695                 if (strncmp(map->olm_name, name, namelen) == 0)
1696                         break;
1697         }
1698
1699         if (map->olm_name == NULL)
1700                 RETURN(0);
1701
1702         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1703         if (IS_ERR(child))
1704                 RETURN(PTR_ERR(child));
1705
1706         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1707                               &map->olm_fid, map->olm_flags);
1708         dput(child);
1709
1710         RETURN(rc);
1711 }
1712
1713 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
1714                              loff_t offset, __u64 ino, unsigned d_type)
1715 {
1716         struct osd_ios_filldir_buf *fill_buf = buf;
1717         struct osd_device          *dev      = fill_buf->oifb_dev;
1718         const struct osd_lf_map    *map;
1719         struct dentry              *child;
1720         int                         rc       = 0;
1721         ENTRY;
1722
1723         /* skip any '.' started names */
1724         if (name[0] == '.')
1725                 RETURN(0);
1726
1727         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
1728                 if (strlen(map->olm_name) != namelen)
1729                         continue;
1730
1731                 if (strncmp(map->olm_name, name, namelen) == 0)
1732                         break;
1733         }
1734
1735         if (map->olm_name == NULL)
1736                 RETURN(0);
1737
1738         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1739         if (IS_ERR(child))
1740                 RETURN(PTR_ERR(child));
1741
1742         if (!(map->olm_flags & OLF_NO_OI))
1743                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1744                                       &map->olm_fid, map->olm_flags);
1745         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
1746                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
1747                                       map->olm_filldir);
1748         dput(child);
1749
1750         RETURN(rc);
1751 }
1752
1753 static int
1754 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1755                      struct dentry *dentry, filldir_t filldir)
1756 {
1757         struct osd_ios_filldir_buf    buf   = {
1758 #ifdef HAVE_DIR_CONTEXT
1759                                                 .ctx.actor = filldir,
1760 #endif
1761                                                 .oifb_info = info,
1762                                                 .oifb_dev = dev,
1763                                                 .oifb_dentry = dentry };
1764         struct file                  *filp  = &info->oti_it_ea.oie_file;
1765         struct inode                 *inode = dentry->d_inode;
1766         const struct file_operations *fops  = inode->i_fop;
1767         int                           rc;
1768         ENTRY;
1769
1770         LASSERT(filldir != NULL);
1771
1772         filp->f_pos = 0;
1773         filp->f_dentry = dentry;
1774         filp->f_mode = FMODE_64BITHASH;
1775         filp->f_mapping = inode->i_mapping;
1776         filp->f_op = fops;
1777         filp->private_data = NULL;
1778         set_file_inode(filp, inode);
1779
1780 #ifdef HAVE_DIR_CONTEXT
1781         buf.ctx.pos = filp->f_pos;
1782         rc = fops->iterate(filp, &buf.ctx);
1783         filp->f_pos = buf.ctx.pos;
1784 #else
1785         rc = fops->readdir(filp, &buf, filldir);
1786 #endif
1787         fops->release(inode, filp);
1788
1789         RETURN(rc);
1790 }
1791
1792 static int
1793 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1794                   struct dentry *dentry, filldir_t filldir)
1795 {
1796         struct osd_scrub  *scrub  = &dev->od_scrub;
1797         struct scrub_file *sf     = &scrub->os_file;
1798         struct dentry     *child;
1799         int                rc;
1800         ENTRY;
1801
1802         /* It is existing MDT0 device. We only allow the case of object without
1803          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
1804          * can generate IGIF mode FID for the object and related OI mapping. If
1805          * it is on other MDTs, then becuase file-level backup/restore, related
1806          * OI mapping may be invalid already, we do not know which is the right
1807          * FID for the object. We only allow IGIF objects to reside on the MDT0.
1808          *
1809          * XXX: For the case of object on non-MDT0 device with neither LMA nor
1810          *      "fid" xattr, then something crashed. We cannot re-generate the
1811          *      FID directly, instead, the OI scrub will scan the OI structure
1812          *      and try to re-generate the LMA from the OI mapping. But if the
1813          *      OI mapping crashed or lost also, then we have to give up under
1814          *      double failure cases. */
1815         scrub->os_convert_igif = 1;
1816         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
1817                                        strlen(dot_lustre_name));
1818         if (IS_ERR(child)) {
1819                 rc = PTR_ERR(child);
1820                 if (rc == -ENOENT) {
1821                         /* It is 1.8 MDT device. */
1822                         if (!(sf->sf_flags & SF_UPGRADE)) {
1823                                 osd_scrub_file_reset(scrub,
1824                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1825                                         SF_UPGRADE);
1826                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
1827                                 rc = osd_scrub_file_store(scrub);
1828                         } else {
1829                                 rc = 0;
1830                         }
1831                 }
1832         } else {
1833                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
1834                  * so the client will get IGIF for the ".lustre" object when
1835                  * the MDT restart.
1836                  *
1837                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
1838                  * it does not know whether there are some old clients cached
1839                  * the ".lustre" IGIF during the upgrading. Two choices:
1840                  *
1841                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
1842                  *    It will allow the old connected clients to access the
1843                  *    ".lustre" with cached IGIF. But it will cause others
1844                  *    on the MDT failed to check "fid_is_dot_lustre()".
1845                  *
1846                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
1847                  *    for ".lustre" in spite of whether there are some clients
1848                  *    cached the ".lustre" IGIF or not. It enables the check
1849                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
1850                  *    that the old connected clients cannot access the ".lustre"
1851                  *    with the cached IGIF.
1852                  *
1853                  * Usually, it is rare case for the old connected clients
1854                  * to access the ".lustre" with cached IGIF. So we prefer
1855                  * to the solution 2). */
1856                 rc = osd_ios_scan_one(info, dev, child->d_inode,
1857                                       &LU_DOT_LUSTRE_FID, 0);
1858                 if (rc == 0)
1859                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1860                                               osd_ios_dl_fill);
1861                 dput(child);
1862         }
1863
1864         RETURN(rc);
1865 }
1866
1867 static int
1868 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1869                      struct dentry *dentry, filldir_t filldir)
1870 {
1871         struct osd_scrub  *scrub  = &dev->od_scrub;
1872         struct scrub_file *sf     = &scrub->os_file;
1873         struct dentry     *child;
1874         int                rc;
1875         ENTRY;
1876
1877         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
1878                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
1879                 rc = osd_scrub_file_store(scrub);
1880                 if (rc != 0)
1881                         RETURN(rc);
1882         }
1883
1884         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
1885         if (!IS_ERR(child)) {
1886                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1887                 dput(child);
1888         } else {
1889                 rc = PTR_ERR(child);
1890         }
1891
1892         if (rc != 0 && rc != -ENOENT)
1893                 RETURN(rc);
1894
1895         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
1896         if (!IS_ERR(child)) {
1897                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1898                 dput(child);
1899         } else {
1900                 rc = PTR_ERR(child);
1901         }
1902
1903         if (rc == -ENOENT)
1904                 rc = 0;
1905
1906         RETURN(rc);
1907 }
1908
1909 static int osd_initial_OI_scrub(struct osd_thread_info *info,
1910                                 struct osd_device *dev)
1911 {
1912         struct osd_ios_item     *item    = NULL;
1913         scandir_t                scandir = osd_ios_general_scan;
1914         filldir_t                filldir = osd_ios_root_fill;
1915         struct dentry           *dentry  = osd_sb(dev)->s_root;
1916         const struct osd_lf_map *map     = osd_lf_maps;
1917         int                      rc;
1918         ENTRY;
1919
1920         /* Lookup IGIF in OI by force for initial OI scrub. */
1921         dev->od_igif_inoi = 1;
1922
1923         while (1) {
1924                 rc = scandir(info, dev, dentry, filldir);
1925                 if (item != NULL) {
1926                         dput(item->oii_dentry);
1927                         OBD_FREE_PTR(item);
1928                 }
1929
1930                 if (rc != 0)
1931                         break;
1932
1933                 if (list_empty(&dev->od_ios_list))
1934                         break;
1935
1936                 item = list_entry(dev->od_ios_list.next,
1937                                   struct osd_ios_item, oii_list);
1938                 list_del_init(&item->oii_list);
1939
1940                 LASSERT(item->oii_scandir != NULL);
1941                 scandir = item->oii_scandir;
1942                 filldir = item->oii_filldir;
1943                 dentry = item->oii_dentry;
1944         }
1945
1946         while (!list_empty(&dev->od_ios_list)) {
1947                 item = list_entry(dev->od_ios_list.next,
1948                                   struct osd_ios_item, oii_list);
1949                 list_del_init(&item->oii_list);
1950                 dput(item->oii_dentry);
1951                 OBD_FREE_PTR(item);
1952         }
1953
1954         if (rc != 0)
1955                 RETURN(rc);
1956
1957         /* There maybe the case that the object has been removed, but its OI
1958          * mapping is still in the OI file, such as the "CATALOGS" after MDT
1959          * file-level backup/restore. So here cleanup the stale OI mappings. */
1960         while (map->olm_name != NULL) {
1961                 struct dentry *child;
1962
1963                 if (fid_is_zero(&map->olm_fid)) {
1964                         map++;
1965                         continue;
1966                 }
1967
1968                 child = osd_ios_lookup_one_len(map->olm_name,
1969                                                osd_sb(dev)->s_root,
1970                                                strlen(map->olm_name));
1971                 if (!IS_ERR(child))
1972                         dput(child);
1973                 else if (PTR_ERR(child) == -ENOENT)
1974                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
1975                                                   NULL, DTO_INDEX_DELETE,
1976                                                   true, 0);
1977                 map++;
1978         }
1979
1980         RETURN(0);
1981 }
1982
1983 char *osd_lf_fid2name(const struct lu_fid *fid)
1984 {
1985         const struct osd_lf_map *map = osd_lf_maps;
1986
1987         while (map->olm_name != NULL) {
1988                 if (!lu_fid_eq(fid, &map->olm_fid)) {
1989                         map++;
1990                         continue;
1991                 }
1992
1993                 if (map->olm_flags & OLF_SHOW_NAME)
1994                         return map->olm_name;
1995                 else
1996                         return "";
1997         }
1998
1999         return NULL;
2000 }
2001
2002 /* OI scrub start/stop */
2003
2004 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2005 {
2006         struct osd_scrub     *scrub  = &dev->od_scrub;
2007         struct ptlrpc_thread *thread = &scrub->os_thread;
2008         struct l_wait_info    lwi    = { 0 };
2009         struct task_struct   *task;
2010         int                   rc;
2011         ENTRY;
2012
2013 again:
2014         /* os_lock: sync status between stop and scrub thread */
2015         spin_lock(&scrub->os_lock);
2016         if (thread_is_running(thread)) {
2017                 spin_unlock(&scrub->os_lock);
2018                 RETURN(-EALREADY);
2019         } else if (unlikely(thread_is_stopping(thread))) {
2020                 spin_unlock(&scrub->os_lock);
2021                 l_wait_event(thread->t_ctl_waitq,
2022                              thread_is_stopped(thread),
2023                              &lwi);
2024                 goto again;
2025         }
2026         spin_unlock(&scrub->os_lock);
2027
2028         if (scrub->os_file.sf_status == SS_COMPLETED) {
2029                 if (!(flags & SS_SET_FAILOUT))
2030                         flags |= SS_CLEAR_FAILOUT;
2031
2032                 if (!(flags & SS_SET_DRYRUN))
2033                         flags |= SS_CLEAR_DRYRUN;
2034
2035                 flags |= SS_RESET;
2036         }
2037
2038         scrub->os_start_flags = flags;
2039         thread_set_flags(thread, 0);
2040         task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2041         if (IS_ERR(task)) {
2042                 rc = PTR_ERR(task);
2043                 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2044                        osd_scrub2name(scrub), rc);
2045                 RETURN(rc);
2046         }
2047
2048         l_wait_event(thread->t_ctl_waitq,
2049                      thread_is_running(thread) || thread_is_stopped(thread),
2050                      &lwi);
2051
2052         RETURN(0);
2053 }
2054
2055 int osd_scrub_start(struct osd_device *dev)
2056 {
2057         int rc;
2058         ENTRY;
2059
2060         /* od_otable_mutex: prevent curcurrent start/stop */
2061         mutex_lock(&dev->od_otable_mutex);
2062         rc = do_osd_scrub_start(dev, SS_AUTO);
2063         mutex_unlock(&dev->od_otable_mutex);
2064
2065         RETURN(rc == -EALREADY ? 0 : rc);
2066 }
2067
2068 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2069 {
2070         struct ptlrpc_thread *thread = &scrub->os_thread;
2071         struct l_wait_info    lwi    = { 0 };
2072
2073         /* os_lock: sync status between stop and scrub thread */
2074         spin_lock(&scrub->os_lock);
2075         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2076                 thread_set_flags(thread, SVC_STOPPING);
2077                 spin_unlock(&scrub->os_lock);
2078                 wake_up_all(&thread->t_ctl_waitq);
2079                 l_wait_event(thread->t_ctl_waitq,
2080                              thread_is_stopped(thread),
2081                              &lwi);
2082                 /* Do not skip the last lock/unlock, which can guarantee that
2083                  * the caller cannot return until the OI scrub thread exit. */
2084                 spin_lock(&scrub->os_lock);
2085         }
2086         spin_unlock(&scrub->os_lock);
2087 }
2088
2089 static void osd_scrub_stop(struct osd_device *dev)
2090 {
2091         /* od_otable_mutex: prevent curcurrent start/stop */
2092         mutex_lock(&dev->od_otable_mutex);
2093         dev->od_scrub.os_paused = 1;
2094         do_osd_scrub_stop(&dev->od_scrub);
2095         mutex_unlock(&dev->od_otable_mutex);
2096 }
2097
2098 /* OI scrub setup/cleanup */
2099
2100 static const char osd_scrub_name[] = "OI_scrub";
2101
2102 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2103 {
2104         struct osd_thread_info     *info   = osd_oti_get(env);
2105         struct osd_scrub           *scrub  = &dev->od_scrub;
2106         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
2107         struct scrub_file          *sf     = &scrub->os_file;
2108         struct super_block         *sb     = osd_sb(dev);
2109         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
2110         struct lvfs_run_ctxt        saved;
2111         struct file                *filp;
2112         struct inode               *inode;
2113         struct lu_fid              *fid    = &info->oti_fid;
2114         int                         dirty  = 0;
2115         int                         rc     = 0;
2116         ENTRY;
2117
2118         memset(scrub, 0, sizeof(*scrub));
2119         OBD_SET_CTXT_MAGIC(ctxt);
2120         ctxt->pwdmnt = dev->od_mnt;
2121         ctxt->pwd = dev->od_mnt->mnt_root;
2122         ctxt->fs = get_ds();
2123
2124         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2125         init_rwsem(&scrub->os_rwsem);
2126         spin_lock_init(&scrub->os_lock);
2127         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2128
2129         push_ctxt(&saved, ctxt);
2130         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2131         if (IS_ERR(filp)) {
2132                 pop_ctxt(&saved, ctxt);
2133                 RETURN(PTR_ERR(filp));
2134         }
2135
2136         inode = filp->f_dentry->d_inode;
2137         /* 'What the @fid is' is not imporatant, because the object
2138          * has no OI mapping, and only is visible inside the OSD.*/
2139         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2140         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2141         if (rc != 0) {
2142                 filp_close(filp, 0);
2143                 pop_ctxt(&saved, ctxt);
2144                 RETURN(rc);
2145         }
2146
2147         scrub->os_inode = igrab(inode);
2148         filp_close(filp, 0);
2149         pop_ctxt(&saved, ctxt);
2150
2151         rc = osd_scrub_file_load(scrub);
2152         if (rc == -ENOENT) {
2153                 osd_scrub_file_init(scrub, es->s_uuid);
2154                 /* If the "/O" dir does not exist when mount (indicated by
2155                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2156                  * then it is quite probably that the device is a new one,
2157                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2158                  *
2159                  * For the rare case that "/O" and "OI_scrub" both lost on
2160                  * an old device, it can be found and cleared later.
2161                  *
2162                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2163                  * need to check "filter_fid_old" and to convert it to
2164                  * "filter_fid" for each object, and all the IGIF should
2165                  * have their FID mapping in OI files already. */
2166                 if (dev->od_maybe_new)
2167                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2168                 dirty = 1;
2169         } else if (rc != 0) {
2170                 RETURN(rc);
2171         } else {
2172                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2173                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2174                         dirty = 1;
2175                 } else if (sf->sf_status == SS_SCANNING) {
2176                         sf->sf_status = SS_CRASHED;
2177                         dirty = 1;
2178                 }
2179         }
2180
2181         if (sf->sf_pos_last_checkpoint != 0)
2182                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2183         else
2184                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2185
2186         if (dirty != 0) {
2187                 rc = osd_scrub_file_store(scrub);
2188                 if (rc != 0)
2189                         RETURN(rc);
2190         }
2191
2192         /* Initialize OI files. */
2193         rc = osd_oi_init(info, dev);
2194         if (rc < 0)
2195                 RETURN(rc);
2196
2197         rc = osd_initial_OI_scrub(info, dev);
2198         if (rc == 0) {
2199                 if (sf->sf_flags & SF_UPGRADE ||
2200                     !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2201                       sf->sf_success_count > 0)) {
2202                         dev->od_igif_inoi = 0;
2203                         dev->od_check_ff = dev->od_is_ost;
2204                 } else {
2205                         dev->od_igif_inoi = 1;
2206                         dev->od_check_ff = 0;
2207                 }
2208
2209                 if (sf->sf_flags & SF_INCONSISTENT)
2210                         /* The 'od_igif_inoi' will be set under the
2211                          * following cases:
2212                          * 1) new created system, or
2213                          * 2) restored from file-level backup, or
2214                          * 3) the upgrading completed.
2215                          *
2216                          * The 'od_igif_inoi' may be cleared by OI scrub
2217                          * later if found that the system is upgrading. */
2218                         dev->od_igif_inoi = 1;
2219
2220                 if (!dev->od_noscrub &&
2221                     ((sf->sf_status == SS_PAUSED) ||
2222                      (sf->sf_status == SS_CRASHED &&
2223                       sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2224                                       SF_UPGRADE | SF_AUTO)) ||
2225                      (sf->sf_status == SS_INIT &&
2226                       sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2227                                       SF_UPGRADE))))
2228                         rc = osd_scrub_start(dev);
2229         }
2230
2231         /* it is possible that dcache entries may keep objects after they are
2232          * deleted by OSD. While it looks safe this can cause object data to
2233          * stay until umount causing failures in tests calculating free space,
2234          * e.g. replay-ost-single. Since those dcache entries are not used
2235          * anymore let's just free them after use here */
2236         shrink_dcache_sb(sb);
2237
2238         RETURN(rc);
2239 }
2240
2241 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2242 {
2243         struct osd_scrub *scrub = &dev->od_scrub;
2244
2245         LASSERT(dev->od_otable_it == NULL);
2246
2247         if (scrub->os_inode != NULL) {
2248                 osd_scrub_stop(dev);
2249                 iput(scrub->os_inode);
2250                 scrub->os_inode = NULL;
2251         }
2252         if (dev->od_oi_table != NULL)
2253                 osd_oi_fini(osd_oti_get(env), dev);
2254 }
2255
2256 /* object table based iteration APIs */
2257
2258 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2259                                        struct dt_object *dt, __u32 attr,
2260                                        struct lustre_capa *capa)
2261 {
2262         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2263         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2264         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2265         struct osd_scrub       *scrub = &dev->od_scrub;
2266         struct osd_otable_it   *it;
2267         __u32                   start = 0;
2268         int                     rc;
2269         ENTRY;
2270
2271         /* od_otable_mutex: prevent curcurrent init/fini */
2272         mutex_lock(&dev->od_otable_mutex);
2273         if (dev->od_otable_it != NULL)
2274                 GOTO(out, it = ERR_PTR(-EALREADY));
2275
2276         OBD_ALLOC_PTR(it);
2277         if (it == NULL)
2278                 GOTO(out, it = ERR_PTR(-ENOMEM));
2279
2280         dev->od_otable_it = it;
2281         it->ooi_dev = dev;
2282         it->ooi_cache.ooc_consumer_idx = -1;
2283         if (flags & DOIF_OUTUSED)
2284                 it->ooi_used_outside = 1;
2285
2286         if (flags & DOIF_RESET)
2287                 start |= SS_RESET;
2288
2289         if (valid & DOIV_ERROR_HANDLE) {
2290                 if (flags & DOIF_FAILOUT)
2291                         start |= SS_SET_FAILOUT;
2292                 else
2293                         start |= SS_CLEAR_FAILOUT;
2294         }
2295
2296         if (valid & DOIV_DRYRUN) {
2297                 if (flags & DOIF_DRYRUN)
2298                         start |= SS_SET_DRYRUN;
2299                 else
2300                         start |= SS_CLEAR_DRYRUN;
2301         }
2302
2303         rc = do_osd_scrub_start(dev, start);
2304         if (rc < 0 && rc != -EALREADY) {
2305                 dev->od_otable_it = NULL;
2306                 OBD_FREE_PTR(it);
2307                 GOTO(out, it = ERR_PTR(rc));
2308         }
2309
2310         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2311
2312         GOTO(out, it);
2313
2314 out:
2315         mutex_unlock(&dev->od_otable_mutex);
2316         return (struct dt_it *)it;
2317 }
2318
2319 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2320 {
2321         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2322         struct osd_device    *dev = it->ooi_dev;
2323
2324         /* od_otable_mutex: prevent curcurrent init/fini */
2325         mutex_lock(&dev->od_otable_mutex);
2326         do_osd_scrub_stop(&dev->od_scrub);
2327         LASSERT(dev->od_otable_it == it);
2328
2329         dev->od_otable_it = NULL;
2330         mutex_unlock(&dev->od_otable_mutex);
2331         OBD_FREE_PTR(it);
2332 }
2333
2334 static int osd_otable_it_get(const struct lu_env *env,
2335                              struct dt_it *di, const struct dt_key *key)
2336 {
2337         return 0;
2338 }
2339
2340 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2341 {
2342 }
2343
2344 static inline int
2345 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2346 {
2347         spin_lock(&scrub->os_lock);
2348         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2349             scrub->os_waiting ||
2350             !thread_is_running(&scrub->os_thread))
2351                 it->ooi_waiting = 0;
2352         else
2353                 it->ooi_waiting = 1;
2354         spin_unlock(&scrub->os_lock);
2355
2356         return !it->ooi_waiting;
2357 }
2358
2359 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2360 {
2361         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2362         struct osd_device       *dev    = it->ooi_dev;
2363         struct osd_scrub        *scrub  = &dev->od_scrub;
2364         struct osd_otable_cache *ooc    = &it->ooi_cache;
2365         struct ptlrpc_thread    *thread = &scrub->os_thread;
2366         struct l_wait_info       lwi    = { 0 };
2367         int                      rc;
2368         ENTRY;
2369
2370         LASSERT(it->ooi_user_ready);
2371
2372 again:
2373         if (!thread_is_running(thread) && !it->ooi_used_outside)
2374                 RETURN(1);
2375
2376         if (ooc->ooc_cached_items > 0) {
2377                 ooc->ooc_cached_items--;
2378                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2379                                         ~OSD_OTABLE_IT_CACHE_MASK;
2380                 RETURN(0);
2381         }
2382
2383         if (it->ooi_all_cached) {
2384                 l_wait_event(thread->t_ctl_waitq,
2385                              !thread_is_running(thread),
2386                              &lwi);
2387                 RETURN(1);
2388         }
2389
2390         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2391                 spin_lock(&scrub->os_lock);
2392                 scrub->os_waiting = 0;
2393                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2394                 spin_unlock(&scrub->os_lock);
2395         }
2396
2397         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2398                 l_wait_event(thread->t_ctl_waitq,
2399                              osd_otable_it_wakeup(scrub, it),
2400                              &lwi);
2401
2402         if (!thread_is_running(thread) && !it->ooi_used_outside)
2403                 RETURN(1);
2404
2405         rc = osd_otable_it_preload(env, it);
2406         if (rc >= 0)
2407                 goto again;
2408
2409         RETURN(rc);
2410 }
2411
2412 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2413                                         const struct dt_it *di)
2414 {
2415         return NULL;
2416 }
2417
2418 static int osd_otable_it_key_size(const struct lu_env *env,
2419                                   const struct dt_it *di)
2420 {
2421         return sizeof(__u64);
2422 }
2423
2424 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2425                              struct dt_rec *rec, __u32 attr)
2426 {
2427         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2428         struct osd_otable_cache *ooc = &it->ooi_cache;
2429
2430         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2431
2432         /* Filter out Invald FID already. */
2433         LASSERTF(fid_is_sane((struct lu_fid *)rec),
2434                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2435                  PFID((struct lu_fid *)rec),
2436                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2437
2438         return 0;
2439 }
2440
2441 static __u64 osd_otable_it_store(const struct lu_env *env,
2442                                  const struct dt_it *di)
2443 {
2444         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2445         struct osd_otable_cache *ooc = &it->ooi_cache;
2446         __u64                    hash;
2447
2448         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2449                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2450         else
2451                 hash = ooc->ooc_pos_preload;
2452         return hash;
2453 }
2454
2455 /**
2456  * Set the OSD layer iteration start position as the specified hash.
2457  */
2458 static int osd_otable_it_load(const struct lu_env *env,
2459                               const struct dt_it *di, __u64 hash)
2460 {
2461         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
2462         struct osd_device       *dev   = it->ooi_dev;
2463         struct osd_otable_cache *ooc   = &it->ooi_cache;
2464         struct osd_scrub        *scrub = &dev->od_scrub;
2465         int                      rc;
2466         ENTRY;
2467
2468         /* Forbid to set iteration position after iteration started. */
2469         if (it->ooi_user_ready)
2470                 RETURN(-EPERM);
2471
2472         if (hash > OSD_OTABLE_MAX_HASH)
2473                 hash = OSD_OTABLE_MAX_HASH;
2474
2475         ooc->ooc_pos_preload = hash;
2476         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2477                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2478
2479         it->ooi_user_ready = 1;
2480         if (!scrub->os_full_speed)
2481                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2482
2483         /* Unplug OSD layer iteration by the first next() call. */
2484         rc = osd_otable_it_next(env, (struct dt_it *)it);
2485
2486         RETURN(rc);
2487 }
2488
2489 static int osd_otable_it_key_rec(const struct lu_env *env,
2490                                  const struct dt_it *di, void *key_rec)
2491 {
2492         return 0;
2493 }
2494
2495 const struct dt_index_operations osd_otable_ops = {
2496         .dio_it = {
2497                 .init     = osd_otable_it_init,
2498                 .fini     = osd_otable_it_fini,
2499                 .get      = osd_otable_it_get,
2500                 .put      = osd_otable_it_put,
2501                 .next     = osd_otable_it_next,
2502                 .key      = osd_otable_it_key,
2503                 .key_size = osd_otable_it_key_size,
2504                 .rec      = osd_otable_it_rec,
2505                 .store    = osd_otable_it_store,
2506                 .load     = osd_otable_it_load,
2507                 .key_rec  = osd_otable_it_key_rec,
2508         }
2509 };
2510
2511 /* high priority inconsistent items list APIs */
2512
2513 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2514                    int insert)
2515 {
2516         struct osd_inconsistent_item *oii;
2517         struct osd_scrub             *scrub  = &dev->od_scrub;
2518         struct ptlrpc_thread         *thread = &scrub->os_thread;
2519         int                           wakeup = 0;
2520         ENTRY;
2521
2522         OBD_ALLOC_PTR(oii);
2523         if (unlikely(oii == NULL))
2524                 RETURN(-ENOMEM);
2525
2526         INIT_LIST_HEAD(&oii->oii_list);
2527         oii->oii_cache = *oic;
2528         oii->oii_insert = insert;
2529
2530         spin_lock(&scrub->os_lock);
2531         if (unlikely(!thread_is_running(thread))) {
2532                 spin_unlock(&scrub->os_lock);
2533                 OBD_FREE_PTR(oii);
2534                 RETURN(-EAGAIN);
2535         }
2536
2537         if (list_empty(&scrub->os_inconsistent_items))
2538                 wakeup = 1;
2539         list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2540         spin_unlock(&scrub->os_lock);
2541
2542         if (wakeup != 0)
2543                 wake_up_all(&thread->t_ctl_waitq);
2544
2545         RETURN(0);
2546 }
2547
2548 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2549                    struct osd_inode_id *id)
2550 {
2551         struct osd_scrub             *scrub = &dev->od_scrub;
2552         struct osd_inconsistent_item *oii;
2553         ENTRY;
2554
2555         spin_lock(&scrub->os_lock);
2556         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2557                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2558                         *id = oii->oii_cache.oic_lid;
2559                         spin_unlock(&scrub->os_lock);
2560                         RETURN(0);
2561                 }
2562         }
2563         spin_unlock(&scrub->os_lock);
2564
2565         RETURN(-ENOENT);
2566 }
2567
2568 /* OI scrub dump */
2569
2570 static const char *scrub_status_names[] = {
2571         "init",
2572         "scanning",
2573         "completed",
2574         "failed",
2575         "stopped",
2576         "paused",
2577         "crashed",
2578         NULL
2579 };
2580
2581 static const char *scrub_flags_names[] = {
2582         "recreated",
2583         "inconsistent",
2584         "auto",
2585         "upgrade",
2586         NULL
2587 };
2588
2589 static const char *scrub_param_names[] = {
2590         "failout",
2591         "dryrun",
2592         NULL
2593 };
2594
2595 static int scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
2596                            const char *prefix)
2597 {
2598         int flag;
2599         int rc;
2600         int i;
2601
2602         rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2603         if (rc < 0)
2604                 return rc;
2605
2606         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2607                 if (flag & bits) {
2608                         bits &= ~flag;
2609                         rc = seq_printf(m, "%s%c", names[i],
2610                                         bits != 0 ? ',' : '\n');
2611                         if (rc < 0)
2612                                 return rc;
2613                 }
2614         }
2615         return 0;
2616 }
2617
2618 static int scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
2619 {
2620         int rc;
2621
2622         if (time != 0)
2623                 rc = seq_printf(m, "%s: "LPU64" seconds\n", prefix,
2624                               cfs_time_current_sec() - time);
2625         else
2626                 rc = seq_printf(m, "%s: N/A\n", prefix);
2627         return rc;
2628 }
2629
2630 static int scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
2631 {
2632         int rc;
2633
2634         if (pos != 0)
2635                 rc = seq_printf(m, "%s: "LPU64"\n", prefix, pos);
2636         else
2637                 rc = seq_printf(m, "%s: N/A\n", prefix);
2638         return rc;
2639 }
2640
2641 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
2642 {
2643         struct osd_scrub  *scrub   = &dev->od_scrub;
2644         struct scrub_file *sf      = &scrub->os_file;
2645         __u64              checked;
2646         __u64              speed;
2647         int                rc;
2648
2649         down_read(&scrub->os_rwsem);
2650         rc = seq_printf(m, "name: OI_scrub\n"
2651                         "magic: 0x%x\n"
2652                         "oi_files: %d\n"
2653                         "status: %s\n",
2654                         sf->sf_magic, (int)sf->sf_oi_count,
2655                         scrub_status_names[sf->sf_status]);
2656         if (rc < 0)
2657                 goto out;
2658
2659         rc = scrub_bits_dump(m, sf->sf_flags, scrub_flags_names,
2660                              "flags");
2661         if (rc < 0)
2662                 goto out;
2663
2664         rc = scrub_bits_dump(m, sf->sf_param, scrub_param_names,
2665                              "param");
2666         if (rc < 0)
2667                 goto out;
2668
2669         rc = scrub_time_dump(m, sf->sf_time_last_complete,
2670                              "time_since_last_completed");
2671         if (rc < 0)
2672                 goto out;
2673
2674         rc = scrub_time_dump(m, sf->sf_time_latest_start,
2675                              "time_since_latest_start");
2676         if (rc < 0)
2677                 goto out;
2678
2679         rc = scrub_time_dump(m, sf->sf_time_last_checkpoint,
2680                              "time_since_last_checkpoint");
2681         if (rc < 0)
2682                 goto out;
2683
2684         rc = scrub_pos_dump(m, sf->sf_pos_latest_start,
2685                             "latest_start_position");
2686         if (rc < 0)
2687                 goto out;
2688
2689         rc = scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
2690                             "last_checkpoint_position");
2691         if (rc < 0)
2692                 goto out;
2693
2694         rc = scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
2695                             "first_failure_position");
2696         if (rc < 0)
2697                 goto out;
2698
2699         checked = sf->sf_items_checked + scrub->os_new_checked;
2700         rc = seq_printf(m, "checked: "LPU64"\n"
2701                       "updated: "LPU64"\n"
2702                       "failed: "LPU64"\n"
2703                       "prior_updated: "LPU64"\n"
2704                       "noscrub: "LPU64"\n"
2705                       "igif: "LPU64"\n"
2706                       "success_count: %u\n",
2707                       checked, sf->sf_items_updated, sf->sf_items_failed,
2708                       sf->sf_items_updated_prior, sf->sf_items_noscrub,
2709                       sf->sf_items_igif, sf->sf_success_count);
2710         if (rc < 0)
2711                 goto out;
2712
2713         speed = checked;
2714         if (thread_is_running(&scrub->os_thread)) {
2715                 cfs_duration_t duration = cfs_time_current() -
2716                                           scrub->os_time_last_checkpoint;
2717                 __u64 new_checked = scrub->os_new_checked * HZ;
2718                 __u32 rtime = sf->sf_run_time +
2719                               cfs_duration_sec(duration + HALF_SEC);
2720
2721                 if (duration != 0)
2722                         do_div(new_checked, duration);
2723                 if (rtime != 0)
2724                         do_div(speed, rtime);
2725                 rc = seq_printf(m, "run_time: %u seconds\n"
2726                               "average_speed: "LPU64" objects/sec\n"
2727                               "real-time_speed: "LPU64" objects/sec\n"
2728                               "current_position: %u\n"
2729                               "lf_scanned: "LPU64"\n"
2730                               "lf_reparied: "LPU64"\n"
2731                               "lf_failed: "LPU64"\n",
2732                               rtime, speed, new_checked, scrub->os_pos_current,
2733                               scrub->os_lf_scanned, scrub->os_lf_repaired,
2734                               scrub->os_lf_failed);
2735         } else {
2736                 if (sf->sf_run_time != 0)
2737                         do_div(speed, sf->sf_run_time);
2738                 rc = seq_printf(m, "run_time: %u seconds\n"
2739                               "average_speed: "LPU64" objects/sec\n"
2740                               "real-time_speed: N/A\n"
2741                               "current_position: N/A\n"
2742                               "lf_scanned: "LPU64"\n"
2743                               "lf_reparied: "LPU64"\n"
2744                               "lf_failed: "LPU64"\n",
2745                               sf->sf_run_time, speed, scrub->os_lf_scanned,
2746                               scrub->os_lf_repaired, scrub->os_lf_failed);
2747         }
2748
2749 out:
2750         up_read(&scrub->os_rwsem);
2751         return (rc < 0 ? -ENOSPC : 0);
2752 }