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, 2015, 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 <linux/kthread.h>
  41 #include <lustre/lustre_idl.h>
  42 #include <lustre_disk.h>
  43 #include <dt_object.h>
  44 #include <linux/xattr.h>
  45
  46 #include "osd_internal.h"
  47 #include "osd_oi.h"
  48 #include "osd_scrub.h"
  49
  50 #define HALF_SEC        msecs_to_jiffies(MSEC_PER_SEC >> 1)
  51
  52 #define OSD_OTABLE_MAX_HASH             0x00000000ffffffffULL
  53
  54 #define SCRUB_NEXT_BREAK        1 /* exit current loop and process next group */
  55 #define SCRUB_NEXT_CONTINUE     2 /* skip current object and process next bit */
  56 #define SCRUB_NEXT_EXIT         3 /* exit all the loops */
  57 #define SCRUB_NEXT_WAIT         4 /* wait for free cache slot */
  58 #define SCRUB_NEXT_CRASH        5 /* simulate system crash during OI scrub */
  59 #define SCRUB_NEXT_FATAL        6 /* simulate failure during OI scrub */
  60 #define SCRUB_NEXT_NOSCRUB      7 /* new created object, no scrub on it */
  61 #define SCRUB_NEXT_NOLMA        8 /* the inode has no FID-in-LMA */
  62 #define SCRUB_NEXT_OSTOBJ       9 /* for OST-object */
  63 #define SCRUB_NEXT_OSTOBJ_OLD   10 /* old OST-object, no LMA or no FID-on-OST
  64                                     * flags in LMA */
  65
  66 /* misc functions */
  67
  68 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
  69 {
  70         return container_of0(scrub, struct osd_device, od_scrub);
  71 }
  72
  73 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
  74 {
  75         return osd_sb(osd_scrub2dev(scrub));
  76 }
  77
  78 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
  79                                        struct osd_otable_cache *ooc)
  80 {
  81         return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
  82 }
  83
  84 static inline const char *osd_scrub2name(struct osd_scrub *scrub)
  85 {
  86         return LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
  87 }
  88
  89 /**
  90  * update/insert/delete the specified OI mapping (@fid @id) according to the ops
  91  *
  92  * \retval   1, changed nothing
  93  * \retval   0, changed successfully
  94  * \retval -ve, on error
  95  */
  96 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
  97                                      struct osd_device *dev,
  98                                      const struct lu_fid *fid,
  99                                      const struct osd_inode_id *id,
 100                                      int ops, bool force,
 101                                      enum oi_check_flags flags, bool *exist)
 102 {
 103         handle_t *th;
 104         int       rc;
 105         ENTRY;
 106
 107         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
 108                 RETURN(0);
 109
 110         /* DTO_INDEX_INSERT is enough for other two ops:
 111          * delete/update, but save stack. */
 112         th = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 113                                 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
 114         if (IS_ERR(th)) {
 115                 rc = PTR_ERR(th);
 116                 CDEBUG(D_LFSCK, "%s: fail to start trans for scrub op %d "
 117                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 118                        PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1,
 119                        rc);
 120                 RETURN(rc);
 121         }
 122
 123         switch (ops) {
 124         case DTO_INDEX_UPDATE:
 125                 rc = osd_oi_update(info, dev, fid, id, th, flags);
 126                 if (unlikely(rc == -ENOENT)) {
 127                         /* Some unlink thread may removed the OI mapping. */
 128                         rc = 1;
 129                 }
 130                 break;
 131         case DTO_INDEX_INSERT:
 132                 rc = osd_oi_insert(info, dev, fid, id, th, flags, exist);
 133                 if (unlikely(rc == -EEXIST)) {
 134                         rc = 1;
 135                         /* XXX: There are trouble things when adding OI
 136                          *      mapping for IGIF object, which may cause
 137                          *      multiple objects to be mapped to the same
 138                          *      IGIF formatted FID. Consider the following
 139                          *      situations:
 140                          *
 141                          *      1) The MDT is upgrading from 1.8 device.
 142                          *      The OI scrub generates IGIF FID1 for the
 143                          *      OBJ1 and adds the OI mapping.
 144                          *
 145                          *      2) For some reason, the OI scrub does not
 146                          *      process all the IGIF objects completely.
 147                          *
 148                          *      3) The MDT is backuped and restored against
 149                          *      this device.
 150                          *
 151                          *      4) When the MDT mounts up, the OI scrub will
 152                          *      try to rebuild the OI files. For some IGIF
 153                          *      object, OBJ2, which was not processed by the
 154                          *      OI scrub before the backup/restore, and the
 155                          *      new generated IGIF formatted FID may be just
 156                          *      the FID1, the same as OBJ1.
 157                          *
 158                          *      Under such case, the OI scrub cannot know how
 159                          *      to generate new FID for the OBJ2.
 160                          *
 161                          *      Currently, we do nothing for that. One possible
 162                          *      solution is to generate new normal FID for the
 163                          *      conflict object.
 164                          *
 165                          *      Anyway, it is rare, only exists in theory. */
 166                 }
 167                 break;
 168         case DTO_INDEX_DELETE:
 169                 rc = osd_oi_delete(info, dev, fid, th, flags);
 170                 if (rc == -ENOENT) {
 171                         /* It is normal that the unlink thread has removed the
 172                          * OI mapping already. */
 173                         rc = 1;
 174                 }
 175                 break;
 176         default:
 177                 LASSERTF(0, "Unexpected ops %d\n", ops);
 178                 break;
 179         }
 180
 181         ldiskfs_journal_stop(th);
 182         if (rc < 0)
 183                 CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d "
 184                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 185                        PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1,
 186                        rc);
 187
 188         RETURN(rc);
 189 }
 190
 191 /* OI_scrub file ops */
 192
 193 static void osd_scrub_file_to_cpu(struct scrub_file *des,
 194                                   struct scrub_file *src)
 195 {
 196         memcpy(des->sf_uuid, src->sf_uuid, 16);
 197         des->sf_flags   = le64_to_cpu(src->sf_flags);
 198         des->sf_magic   = le32_to_cpu(src->sf_magic);
 199         des->sf_status  = le16_to_cpu(src->sf_status);
 200         des->sf_param   = le16_to_cpu(src->sf_param);
 201         des->sf_time_last_complete      =
 202                                 le64_to_cpu(src->sf_time_last_complete);
 203         des->sf_time_latest_start       =
 204                                 le64_to_cpu(src->sf_time_latest_start);
 205         des->sf_time_last_checkpoint    =
 206                                 le64_to_cpu(src->sf_time_last_checkpoint);
 207         des->sf_pos_latest_start        =
 208                                 le64_to_cpu(src->sf_pos_latest_start);
 209         des->sf_pos_last_checkpoint     =
 210                                 le64_to_cpu(src->sf_pos_last_checkpoint);
 211         des->sf_pos_first_inconsistent  =
 212                                 le64_to_cpu(src->sf_pos_first_inconsistent);
 213         des->sf_items_checked           =
 214                                 le64_to_cpu(src->sf_items_checked);
 215         des->sf_items_updated           =
 216                                 le64_to_cpu(src->sf_items_updated);
 217         des->sf_items_failed            =
 218                                 le64_to_cpu(src->sf_items_failed);
 219         des->sf_items_updated_prior     =
 220                                 le64_to_cpu(src->sf_items_updated_prior);
 221         des->sf_run_time        = le32_to_cpu(src->sf_run_time);
 222         des->sf_success_count   = le32_to_cpu(src->sf_success_count);
 223         des->sf_oi_count        = le16_to_cpu(src->sf_oi_count);
 224         des->sf_internal_flags  = le16_to_cpu(src->sf_internal_flags);
 225         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 226 }
 227
 228 static void osd_scrub_file_to_le(struct scrub_file *des,
 229                                  struct scrub_file *src)
 230 {
 231         memcpy(des->sf_uuid, src->sf_uuid, 16);
 232         des->sf_flags   = cpu_to_le64(src->sf_flags);
 233         des->sf_magic   = cpu_to_le32(src->sf_magic);
 234         des->sf_status  = cpu_to_le16(src->sf_status);
 235         des->sf_param   = cpu_to_le16(src->sf_param);
 236         des->sf_time_last_complete      =
 237                                 cpu_to_le64(src->sf_time_last_complete);
 238         des->sf_time_latest_start       =
 239                                 cpu_to_le64(src->sf_time_latest_start);
 240         des->sf_time_last_checkpoint    =
 241                                 cpu_to_le64(src->sf_time_last_checkpoint);
 242         des->sf_pos_latest_start        =
 243                                 cpu_to_le64(src->sf_pos_latest_start);
 244         des->sf_pos_last_checkpoint     =
 245                                 cpu_to_le64(src->sf_pos_last_checkpoint);
 246         des->sf_pos_first_inconsistent  =
 247                                 cpu_to_le64(src->sf_pos_first_inconsistent);
 248         des->sf_items_checked           =
 249                                 cpu_to_le64(src->sf_items_checked);
 250         des->sf_items_updated           =
 251                                 cpu_to_le64(src->sf_items_updated);
 252         des->sf_items_failed            =
 253                                 cpu_to_le64(src->sf_items_failed);
 254         des->sf_items_updated_prior     =
 255                                 cpu_to_le64(src->sf_items_updated_prior);
 256         des->sf_run_time        = cpu_to_le32(src->sf_run_time);
 257         des->sf_success_count   = cpu_to_le32(src->sf_success_count);
 258         des->sf_oi_count        = cpu_to_le16(src->sf_oi_count);
 259         des->sf_internal_flags  = cpu_to_le16(src->sf_internal_flags);
 260         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 261 }
 262
 263 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
 264 {
 265         struct scrub_file *sf = &scrub->os_file;
 266
 267         memset(sf, 0, sizeof(*sf));
 268         memcpy(sf->sf_uuid, uuid, 16);
 269         sf->sf_magic = SCRUB_MAGIC_V1;
 270         sf->sf_status = SS_INIT;
 271 }
 272
 273 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
 274 {
 275         struct scrub_file *sf = &scrub->os_file;
 276
 277         CDEBUG(D_LFSCK, "%.16s: reset OI scrub file, old flags = "
 278                LPX64", add flags = "LPX64"\n",
 279                osd_scrub2name(scrub), sf->sf_flags, flags);
 280
 281         memcpy(sf->sf_uuid, uuid, 16);
 282         sf->sf_status = SS_INIT;
 283         sf->sf_flags |= flags;
 284         sf->sf_flags &= ~SF_AUTO;
 285         sf->sf_run_time = 0;
 286         sf->sf_time_latest_start = 0;
 287         sf->sf_time_last_checkpoint = 0;
 288         sf->sf_pos_latest_start = 0;
 289         sf->sf_pos_last_checkpoint = 0;
 290         sf->sf_pos_first_inconsistent = 0;
 291         sf->sf_items_checked = 0;
 292         sf->sf_items_updated = 0;
 293         sf->sf_items_failed = 0;
 294         if (!scrub->os_in_join)
 295                 sf->sf_items_updated_prior = 0;
 296
 297         sf->sf_items_noscrub = 0;
 298         sf->sf_items_igif = 0;
 299 }
 300
 301 static int osd_scrub_file_load(struct osd_scrub *scrub)
 302 {
 303         loff_t  pos  = 0;
 304         int     len  = sizeof(scrub->os_file_disk);
 305         int     rc;
 306
 307         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
 308         if (rc == len) {
 309                 struct scrub_file *sf = &scrub->os_file;
 310
 311                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
 312                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
 313                         CDEBUG(D_LFSCK, "%.16s: invalid scrub magic "
 314                                "0x%x != 0x%x\n", osd_scrub2name(scrub),
 315                                sf->sf_magic, SCRUB_MAGIC_V1);
 316                         /* Process it as new scrub file. */
 317                         rc = -ENOENT;
 318                 } else {
 319                         rc = 0;
 320                 }
 321         } else if (rc != 0) {
 322                 CDEBUG(D_LFSCK, "%.16s: fail to load scrub file, "
 323                        "expected = %d: rc = %d\n",
 324                        osd_scrub2name(scrub), len, rc);
 325                 if (rc > 0)
 326                         rc = -EFAULT;
 327         } else {
 328                 /* return -ENOENT for empty scrub file case. */
 329                 rc = -ENOENT;
 330         }
 331
 332         return rc;
 333 }
 334
 335 int osd_scrub_file_store(struct osd_scrub *scrub)
 336 {
 337         struct osd_device *dev;
 338         handle_t          *jh;
 339         loff_t             pos     = 0;
 340         int                len     = sizeof(scrub->os_file_disk);
 341         int                credits;
 342         int                rc;
 343
 344         dev = container_of0(scrub, struct osd_device, od_scrub);
 345         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
 346                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
 347         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits);
 348         if (IS_ERR(jh)) {
 349                 rc = PTR_ERR(jh);
 350                 CDEBUG(D_LFSCK, "%.16s: fail to start trans for scrub store: "
 351                        "rc = %d\n", osd_scrub2name(scrub), rc);
 352                 return rc;
 353         }
 354
 355         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
 356         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
 357                                       len, 0, &pos, jh);
 358         ldiskfs_journal_stop(jh);
 359         if (rc != 0)
 360                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file, "
 361                        "expected = %d: rc = %d\n",
 362                        osd_scrub2name(scrub), len, rc);
 363
 364         scrub->os_time_last_checkpoint = cfs_time_current();
 365         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
 366                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
 367         return rc;
 368 }
 369
 370 static int
 371 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
 372                      struct inode *inode, const struct lu_fid *fid)
 373 {
 374         struct filter_fid_old   *ff      = &info->oti_ff;
 375         struct dentry           *dentry  = &info->oti_obj_dentry;
 376         struct lu_fid           *tfid    = &info->oti_fid;
 377         handle_t                *jh;
 378         int                      size    = 0;
 379         int                      rc;
 380         bool                     removed = false;
 381         bool                     reset   = true;
 382         ENTRY;
 383
 384         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
 385                 RETURN(0);
 386
 387         if (fid_is_idif(fid) && dev->od_index_in_idif == 0) {
 388                 struct ost_id *oi = &info->oti_ostid;
 389
 390                 fid_to_ostid(fid, oi);
 391                 ostid_to_fid(tfid, oi, 0);
 392         } else {
 393                 *tfid = *fid;
 394         }
 395
 396         /* We want the LMA to fit into the 256-byte OST inode, so operate
 397          * as following:
 398          * 1) read old XATTR_NAME_FID and save the parent FID;
 399          * 2) delete the old XATTR_NAME_FID;
 400          * 3) make new LMA and add it;
 401          * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
 402          *
 403          * Making the LMA to fit into the 256-byte OST inode can save time for
 404          * normal osd_check_lma() and for other OI scrub scanning in future.
 405          * So it is worth to make some slow conversion here. */
 406         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 407                                 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
 408         if (IS_ERR(jh)) {
 409                 rc = PTR_ERR(jh);
 410                 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
 411                        DFID": rc = %d\n", osd_name(dev), PFID(tfid), rc);
 412                 RETURN(rc);
 413         }
 414
 415         /* 1) read old XATTR_NAME_FID and save the parent FID */
 416         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 417         if (rc == sizeof(*ff)) {
 418                 /* 2) delete the old XATTR_NAME_FID */
 419                 ll_vfs_dq_init(inode);
 420                 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
 421                 if (rc != 0)
 422                         GOTO(stop, rc);
 423
 424                 removed = true;
 425         } else if (unlikely(rc == -ENODATA)) {
 426                 reset = false;
 427         } else if (rc != sizeof(struct filter_fid)) {
 428                 GOTO(stop, rc = -EINVAL);
 429         }
 430
 431         /* 3) make new LMA and add it */
 432         rc = osd_ea_fid_set(info, inode, tfid, LMAC_FID_ON_OST, 0);
 433         if (rc == 0 && reset)
 434                 size = sizeof(struct filter_fid);
 435         else if (rc != 0 && removed)
 436                 /* If failed, we should try to add the old back. */
 437                 size = sizeof(struct filter_fid_old);
 438
 439         /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
 440         if (size > 0) {
 441                 int rc1;
 442
 443                 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
 444                                       XATTR_CREATE);
 445                 if (rc1 != 0 && rc == 0)
 446                         rc = rc1;
 447         }
 448
 449         GOTO(stop, rc);
 450
 451 stop:
 452         ldiskfs_journal_stop(jh);
 453         if (rc < 0)
 454                 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
 455                        osd_name(dev), PFID(tfid), rc);
 456         return rc;
 457 }
 458
 459 static int
 460 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
 461                        struct osd_idmap_cache *oic, int val)
 462 {
 463         struct osd_scrub             *scrub  = &dev->od_scrub;
 464         struct scrub_file            *sf     = &scrub->os_file;
 465         struct lu_fid                *fid    = &oic->oic_fid;
 466         struct osd_inode_id          *lid    = &oic->oic_lid;
 467         struct osd_inode_id          *lid2   = &info->oti_id;
 468         struct osd_inconsistent_item *oii    = NULL;
 469         struct inode                 *inode  = NULL;
 470         int                           ops    = DTO_INDEX_UPDATE;
 471         int                           rc;
 472         bool                          converted = false;
 473         bool                          exist     = false;
 474         ENTRY;
 475
 476         down_write(&scrub->os_rwsem);
 477         scrub->os_new_checked++;
 478         if (val < 0)
 479                 GOTO(out, rc = val);
 480
 481         if (scrub->os_in_prior)
 482                 oii = list_entry(oic, struct osd_inconsistent_item,
 483                                  oii_cache);
 484
 485         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
 486                 GOTO(out, rc = 0);
 487
 488         if (fid_is_igif(fid))
 489                 sf->sf_items_igif++;
 490
 491         if (val == SCRUB_NEXT_OSTOBJ_OLD) {
 492                 inode = osd_iget(info, dev, lid);
 493                 if (IS_ERR(inode)) {
 494                         rc = PTR_ERR(inode);
 495                         /* Someone removed the inode. */
 496                         if (rc == -ENOENT || rc == -ESTALE)
 497                                 rc = 0;
 498                         GOTO(out, rc);
 499                 }
 500
 501                 /* The inode has been reused as EA inode, ignore it. */
 502                 if (unlikely(osd_is_ea_inode(inode)))
 503                         GOTO(out, rc = 0);
 504
 505                 sf->sf_flags |= SF_UPGRADE;
 506                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
 507                 dev->od_check_ff = 1;
 508                 rc = osd_scrub_convert_ff(info, dev, inode, fid);
 509                 if (rc != 0)
 510                         GOTO(out, rc);
 511
 512                 converted = true;
 513         }
 514
 515         if ((val == SCRUB_NEXT_NOLMA) &&
 516             (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
 517                 GOTO(out, rc = 0);
 518
 519         if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA)) {
 520                 ops = DTO_INDEX_INSERT;
 521
 522                 goto iget;
 523         }
 524
 525         rc = osd_oi_lookup(info, dev, fid, lid2,
 526                 (val == SCRUB_NEXT_OSTOBJ ||
 527                  val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 528         if (rc != 0) {
 529                 if (rc == -ENOENT)
 530                         ops = DTO_INDEX_INSERT;
 531                 else if (rc != -ESTALE)
 532                         GOTO(out, rc);
 533
 534 iget:
 535                 if (inode == NULL) {
 536                         inode = osd_iget(info, dev, lid);
 537                         if (IS_ERR(inode)) {
 538                                 rc = PTR_ERR(inode);
 539                                 /* Someone removed the inode. */
 540                                 if (rc == -ENOENT || rc == -ESTALE)
 541                                         rc = 0;
 542                                 GOTO(out, rc);
 543                         }
 544
 545                         /* The inode has been reused as EA inode, ignore it. */
 546                         if (unlikely(osd_is_ea_inode(inode)))
 547                                 GOTO(out, rc = 0);
 548                 }
 549
 550                 if (!scrub->os_partial_scan)
 551                         scrub->os_full_speed = 1;
 552
 553                 switch (val) {
 554                 case SCRUB_NEXT_NOLMA:
 555                         sf->sf_flags |= SF_UPGRADE;
 556                         if (!(sf->sf_param & SP_DRYRUN)) {
 557                                 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
 558                                 if (rc != 0)
 559                                         GOTO(out, rc);
 560                         }
 561
 562                         if (!(sf->sf_flags & SF_INCONSISTENT))
 563                                 dev->od_igif_inoi = 0;
 564                         break;
 565                 case SCRUB_NEXT_OSTOBJ:
 566                         sf->sf_flags |= SF_INCONSISTENT;
 567                 case SCRUB_NEXT_OSTOBJ_OLD:
 568                         break;
 569                 default:
 570                         break;
 571                 }
 572         } else if (osd_id_eq(lid, lid2)) {
 573                 if (converted)
 574                         sf->sf_items_updated++;
 575
 576                 GOTO(out, rc = 0);
 577         } else {
 578                 if (!scrub->os_partial_scan)
 579                         scrub->os_full_speed = 1;
 580
 581                 sf->sf_flags |= SF_INCONSISTENT;
 582
 583                 /* XXX: If the device is restored from file-level backup, then
 584                  *      some IGIFs may have been already in OI files, and some
 585                  *      may be not yet. Means upgrading from 1.8 may be partly
 586                  *      processed, but some clients may hold some immobilized
 587                  *      IGIFs, and use them to access related objects. Under
 588                  *      such case, OSD does not know whether an given IGIF has
 589                  *      been processed or to be processed, and it also cannot
 590                  *      generate local ino#/gen# directly from the immobilized
 591                  *      IGIF because of the backup/restore. Then force OSD to
 592                  *      lookup the given IGIF in OI files, and if no entry,
 593                  *      then ask the client to retry after upgrading completed.
 594                  *      No better choice. */
 595                 dev->od_igif_inoi = 1;
 596         }
 597
 598         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
 599                         (val == SCRUB_NEXT_OSTOBJ ||
 600                          val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0,
 601                         &exist);
 602         if (rc == 0) {
 603                 if (scrub->os_in_prior)
 604                         sf->sf_items_updated_prior++;
 605                 else
 606                         sf->sf_items_updated++;
 607
 608                 if (ops == DTO_INDEX_INSERT && val == 0 && !exist) {
 609                         int idx = osd_oi_fid2idx(dev, fid);
 610
 611                         sf->sf_flags |= SF_RECREATED;
 612                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 613                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 614                 }
 615         }
 616
 617         GOTO(out, rc);
 618
 619 out:
 620         if (rc < 0) {
 621                 sf->sf_items_failed++;
 622                 if (sf->sf_pos_first_inconsistent == 0 ||
 623                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 624                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 625         } else {
 626                 rc = 0;
 627         }
 628
 629         /* There may be conflict unlink during the OI scrub,
 630          * if happend, then remove the new added OI mapping. */
 631         if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
 632             unlikely(ldiskfs_test_inode_state(inode,
 633                                               LDISKFS_STATE_LUSTRE_DESTROY)))
 634                 osd_scrub_refresh_mapping(info, dev, fid, lid,
 635                                 DTO_INDEX_DELETE, false,
 636                                 (val == SCRUB_NEXT_OSTOBJ ||
 637                                  val == SCRUB_NEXT_OSTOBJ_OLD) ?
 638                                 OI_KNOWN_ON_OST : 0, NULL);
 639         up_write(&scrub->os_rwsem);
 640
 641         if (inode != NULL && !IS_ERR(inode))
 642                 iput(inode);
 643
 644         if (oii != NULL) {
 645                 LASSERT(list_empty(&oii->oii_list));
 646
 647                 OBD_FREE_PTR(oii);
 648         }
 649
 650         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 651 }
 652
 653 /* OI scrub APIs */
 654
 655 static int osd_scrub_prep(struct osd_device *dev)
 656 {
 657         struct osd_scrub     *scrub  = &dev->od_scrub;
 658         struct ptlrpc_thread *thread = &scrub->os_thread;
 659         struct scrub_file    *sf     = &scrub->os_file;
 660         __u32                 flags  = scrub->os_start_flags;
 661         int                   rc;
 662         bool                  drop_dryrun = false;
 663         ENTRY;
 664
 665         CDEBUG(D_LFSCK, "%.16s: OI scrub prep, flags = 0x%x\n",
 666                osd_scrub2name(scrub), flags);
 667
 668         down_write(&scrub->os_rwsem);
 669         if (flags & SS_SET_FAILOUT)
 670                 sf->sf_param |= SP_FAILOUT;
 671         else if (flags & SS_CLEAR_FAILOUT)
 672                 sf->sf_param &= ~SP_FAILOUT;
 673
 674         if (flags & SS_SET_DRYRUN) {
 675                 sf->sf_param |= SP_DRYRUN;
 676         } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
 677                 sf->sf_param &= ~SP_DRYRUN;
 678                 drop_dryrun = true;
 679         }
 680
 681         if (flags & SS_RESET)
 682                 osd_scrub_file_reset(scrub,
 683                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
 684
 685         if (flags & SS_AUTO_FULL) {
 686                 scrub->os_full_speed = 1;
 687                 scrub->os_partial_scan = 0;
 688                 sf->sf_flags |= SF_AUTO;
 689         } else if (flags & SS_AUTO_PARTIAL) {
 690                 scrub->os_full_speed = 0;
 691                 scrub->os_partial_scan = 1;
 692                 sf->sf_flags |= SF_AUTO;
 693         } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
 694                                    SF_UPGRADE)) {
 695                 scrub->os_full_speed = 1;
 696                 scrub->os_partial_scan = 0;
 697         } else {
 698                 scrub->os_full_speed = 0;
 699                 scrub->os_partial_scan = 0;
 700         }
 701
 702         spin_lock(&scrub->os_lock);
 703         scrub->os_in_prior = 0;
 704         scrub->os_waiting = 0;
 705         scrub->os_paused = 0;
 706         scrub->os_in_join = 0;
 707         scrub->os_full_scrub = 0;
 708         spin_unlock(&scrub->os_lock);
 709         scrub->os_new_checked = 0;
 710         if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
 711                 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
 712         else if (sf->sf_pos_last_checkpoint != 0)
 713                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 714         else
 715                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 716
 717         scrub->os_pos_current = sf->sf_pos_latest_start;
 718         sf->sf_status = SS_SCANNING;
 719         sf->sf_time_latest_start = cfs_time_current_sec();
 720         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 721         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
 722         rc = osd_scrub_file_store(scrub);
 723         if (rc == 0) {
 724                 spin_lock(&scrub->os_lock);
 725                 thread_set_flags(thread, SVC_RUNNING);
 726                 spin_unlock(&scrub->os_lock);
 727                 wake_up_all(&thread->t_ctl_waitq);
 728         }
 729         up_write(&scrub->os_rwsem);
 730
 731         RETURN(rc);
 732 }
 733
 734 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
 735 {
 736         struct scrub_file *sf = &scrub->os_file;
 737         int                rc;
 738
 739         if (likely(cfs_time_before(cfs_time_current(),
 740                                    scrub->os_time_next_checkpoint) ||
 741                    scrub->os_new_checked == 0))
 742                 return 0;
 743
 744         down_write(&scrub->os_rwsem);
 745         sf->sf_items_checked += scrub->os_new_checked;
 746         scrub->os_new_checked = 0;
 747         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 748         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 749         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 750                                             scrub->os_time_last_checkpoint);
 751         rc = osd_scrub_file_store(scrub);
 752         up_write(&scrub->os_rwsem);
 753
 754         return rc;
 755 }
 756
 757 static int osd_scrub_post(struct osd_scrub *scrub, int result)
 758 {
 759         struct scrub_file *sf = &scrub->os_file;
 760         int rc;
 761         ENTRY;
 762
 763         CDEBUG(D_LFSCK, "%.16s: OI scrub post, result = %d\n",
 764                osd_scrub2name(scrub), result);
 765
 766         down_write(&scrub->os_rwsem);
 767         spin_lock(&scrub->os_lock);
 768         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 769         spin_unlock(&scrub->os_lock);
 770         if (scrub->os_new_checked > 0) {
 771                 sf->sf_items_checked += scrub->os_new_checked;
 772                 scrub->os_new_checked = 0;
 773                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 774         }
 775         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 776         if (result > 0) {
 777                 struct osd_device *dev =
 778                         container_of0(scrub, struct osd_device, od_scrub);
 779
 780                 dev->od_igif_inoi = 1;
 781                 dev->od_check_ff = 0;
 782                 sf->sf_status = SS_COMPLETED;
 783                 if (!(sf->sf_param & SP_DRYRUN)) {
 784                         memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 785                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 786                                           SF_UPGRADE | SF_AUTO);
 787                 }
 788                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 789                 sf->sf_success_count++;
 790         } else if (result == 0) {
 791                 if (scrub->os_paused)
 792                         sf->sf_status = SS_PAUSED;
 793                 else
 794                         sf->sf_status = SS_STOPPED;
 795         } else {
 796                 sf->sf_status = SS_FAILED;
 797         }
 798         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 799                                             scrub->os_time_last_checkpoint);
 800         rc = osd_scrub_file_store(scrub);
 801         up_write(&scrub->os_rwsem);
 802
 803         RETURN(rc < 0 ? rc : result);
 804 }
 805
 806 /* iteration engine */
 807
 808 struct osd_iit_param {
 809         struct super_block *sb;
 810         struct buffer_head *bitmap;
 811         ldiskfs_group_t bg;
 812         __u32 gbase;
 813         __u32 offset;
 814 };
 815
 816 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 817                                    struct osd_device *dev,
 818                                    struct osd_iit_param *param,
 819                                    struct osd_idmap_cache **oic,
 820                                    const bool noslot);
 821
 822 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 823                                    struct osd_device *dev,
 824                                    struct osd_iit_param *param,
 825                                    struct osd_idmap_cache *oic,
 826                                    bool *noslot, int rc);
 827
 828 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 829 {
 830         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 831                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 832         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 833                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 834                 return SCRUB_NEXT_BREAK;
 835         } else {
 836                 *pos = param->gbase + param->offset;
 837                 return 0;
 838         }
 839 }
 840
 841 /**
 842  * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
 843  * \retval 0: FID-on-MDT
 844  */
 845 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
 846                                       struct osd_device *dev,
 847                                       struct lu_fid *fid)
 848 {
 849         /* XXX: The initial OI scrub will scan the top level /O to generate
 850          *      a small local FLDB according to the <seq>. If the given FID
 851          *      is in the local FLDB, then it is FID-on-OST; otherwise it's
 852          *      quite possible for FID-on-MDT. */
 853         if (dev->od_is_ost)
 854                 return SCRUB_NEXT_OSTOBJ_OLD;
 855         else
 856                 return 0;
 857 }
 858
 859 static int osd_scrub_get_fid(struct osd_thread_info *info,
 860                              struct osd_device *dev, struct inode *inode,
 861                              struct lu_fid *fid, bool scrub)
 862 {
 863         struct lustre_mdt_attrs *lma     = &info->oti_mdt_attrs;
 864         int                      rc;
 865         bool                     has_lma = false;
 866
 867         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 868         if (rc == 0) {
 869                 has_lma = true;
 870                 if (lma->lma_compat & LMAC_NOT_IN_OI ||
 871                     lma->lma_incompat & LMAI_AGENT)
 872                         return SCRUB_NEXT_CONTINUE;
 873
 874                 *fid = lma->lma_self_fid;
 875                 if (!scrub)
 876                         return 0;
 877
 878                 if (lma->lma_compat & LMAC_FID_ON_OST)
 879                         return SCRUB_NEXT_OSTOBJ;
 880
 881                 if (fid_is_idif(fid))
 882                         return SCRUB_NEXT_OSTOBJ_OLD;
 883
 884                 /* For local object. */
 885                 if (fid_is_internal(fid))
 886                         return 0;
 887
 888                 /* For external visible MDT-object with non-normal FID. */
 889                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
 890                         return 0;
 891
 892                 /* For the object with normal FID, it may be MDT-object,
 893                  * or may be 2.4 OST-object, need further distinguish.
 894                  * Fall through to next section. */
 895         }
 896
 897         if (rc == -ENODATA || rc == 0) {
 898                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
 899                 if (rc == 0) {
 900                         if (scrub)
 901                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
 902                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 903                         return rc;
 904                 }
 905
 906                 if (rc > 0) {
 907                         if (!has_lma)
 908                                 /* It is FID-on-OST, but we do not know how
 909                                  * to generate its FID, ignore it directly. */
 910                                 rc = SCRUB_NEXT_CONTINUE;
 911                         else
 912                                 /* It is 2.4 OST-object. */
 913                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 914                         return rc;
 915                 }
 916
 917                 if (rc != -ENODATA)
 918                         return rc;
 919
 920                 if (!has_lma) {
 921                         if (dev->od_scrub.os_convert_igif) {
 922                                 lu_igif_build(fid, inode->i_ino,
 923                                               inode->i_generation);
 924                                 if (scrub)
 925                                         rc = SCRUB_NEXT_NOLMA;
 926                                 else
 927                                         rc = 0;
 928                         } else {
 929                                 /* It may be FID-on-OST, or may be FID for
 930                                  * non-MDT0, anyway, we do not know how to
 931                                  * generate its FID, ignore it directly. */
 932                                 rc = SCRUB_NEXT_CONTINUE;
 933                         }
 934                         return rc;
 935                 }
 936
 937                 /* For OI scrub case only: the object has LMA but has no ff
 938                  * (or ff crashed). It may be MDT-object, may be OST-object
 939                  * with crashed ff. The last check is local FLDB. */
 940                 rc = osd_scrub_check_local_fldb(info, dev, fid);
 941         }
 942
 943         return rc;
 944 }
 945
 946 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 947                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 948                         struct super_block *sb, bool scrub)
 949 {
 950         struct inode *inode;
 951         int           rc;
 952         ENTRY;
 953
 954         /* Not handle the backend root object and agent parent object.
 955          * They are neither visible to namespace nor have OI mappings. */
 956         if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
 957                      pos == osd_remote_parent_ino(dev)))
 958                 RETURN(SCRUB_NEXT_CONTINUE);
 959
 960         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 961         inode = osd_iget(info, dev, lid);
 962         if (IS_ERR(inode)) {
 963                 rc = PTR_ERR(inode);
 964                 /* The inode may be removed after bitmap searching, or the
 965                  * file is new created without inode initialized yet. */
 966                 if (rc == -ENOENT || rc == -ESTALE)
 967                         RETURN(SCRUB_NEXT_CONTINUE);
 968
 969                 CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: "
 970                        "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name,
 971                        pos, rc);
 972                 RETURN(rc);
 973         }
 974
 975         /* It is an EA inode, no OI mapping for it, skip it. */
 976         if (osd_is_ea_inode(inode))
 977                 GOTO(put, rc = SCRUB_NEXT_CONTINUE);
 978
 979         if (scrub &&
 980             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 981                 /* Only skip it for the first OI scrub accessing. */
 982                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 983                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
 984         }
 985
 986         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
 987
 988         GOTO(put, rc);
 989
 990 put:
 991         iput(inode);
 992         return rc;
 993 }
 994
 995 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 996                           struct osd_iit_param *param,
 997                           struct osd_idmap_cache **oic, const bool noslot)
 998 {
 999         struct osd_scrub     *scrub  = &dev->od_scrub;
1000         struct ptlrpc_thread *thread = &scrub->os_thread;
1001         struct lu_fid        *fid;
1002         struct osd_inode_id  *lid;
1003         int                   rc;
1004
1005         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
1006                 struct l_wait_info lwi;
1007
1008                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
1009                 if (likely(lwi.lwi_timeout > 0))
1010                         l_wait_event(thread->t_ctl_waitq,
1011                                 !list_empty(&scrub->os_inconsistent_items) ||
1012                                 !thread_is_running(thread),
1013                                 &lwi);
1014         }
1015
1016         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
1017                 spin_lock(&scrub->os_lock);
1018                 thread_set_flags(thread, SVC_STOPPING);
1019                 spin_unlock(&scrub->os_lock);
1020                 return SCRUB_NEXT_CRASH;
1021         }
1022
1023         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
1024                 return SCRUB_NEXT_FATAL;
1025
1026         if (unlikely(!thread_is_running(thread)))
1027                 return SCRUB_NEXT_EXIT;
1028
1029         if (!list_empty(&scrub->os_inconsistent_items)) {
1030                 spin_lock(&scrub->os_lock);
1031                 if (likely(!list_empty(&scrub->os_inconsistent_items))) {
1032                         struct osd_inconsistent_item *oii;
1033
1034                         oii = list_entry(scrub->os_inconsistent_items.next,
1035                                 struct osd_inconsistent_item, oii_list);
1036                         list_del_init(&oii->oii_list);
1037                         spin_unlock(&scrub->os_lock);
1038
1039                         *oic = &oii->oii_cache;
1040                         scrub->os_in_prior = 1;
1041
1042                         return 0;
1043                 }
1044                 spin_unlock(&scrub->os_lock);
1045         }
1046
1047         if (noslot)
1048                 return SCRUB_NEXT_WAIT;
1049
1050         rc = osd_iit_next(param, &scrub->os_pos_current);
1051         if (rc != 0)
1052                 return rc;
1053
1054         *oic = &scrub->os_oic;
1055         fid = &(*oic)->oic_fid;
1056         lid = &(*oic)->oic_lid;
1057         rc = osd_iit_iget(info, dev, fid, lid,
1058                           scrub->os_pos_current, param->sb, true);
1059         return rc;
1060 }
1061
1062 static int osd_preload_next(struct osd_thread_info *info,
1063                             struct osd_device *dev, struct osd_iit_param *param,
1064                             struct osd_idmap_cache **oic, const bool noslot)
1065 {
1066         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
1067         struct osd_scrub        *scrub;
1068         struct ptlrpc_thread    *thread;
1069         int                      rc;
1070
1071         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1072         if (rc != 0)
1073                 return rc;
1074
1075         scrub = &dev->od_scrub;
1076         thread = &scrub->os_thread;
1077         if (thread_is_running(thread) &&
1078             ooc->ooc_pos_preload >= scrub->os_pos_current)
1079                 return SCRUB_NEXT_EXIT;
1080
1081         rc = osd_iit_iget(info, dev,
1082                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1083                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1084                           ooc->ooc_pos_preload, param->sb, false);
1085         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1086          * ignore the failure, so it still need to skip the inode next time. */
1087         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1088         return rc;
1089 }
1090
1091 static inline int
1092 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1093 {
1094         spin_lock(&scrub->os_lock);
1095         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1096             !list_empty(&scrub->os_inconsistent_items) ||
1097             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1098                 scrub->os_waiting = 0;
1099         else
1100                 scrub->os_waiting = 1;
1101         spin_unlock(&scrub->os_lock);
1102
1103         return !scrub->os_waiting;
1104 }
1105
1106 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1107                           struct osd_iit_param *param,
1108                           struct osd_idmap_cache *oic, bool *noslot, int rc)
1109 {
1110         struct l_wait_info       lwi    = { 0 };
1111         struct osd_scrub        *scrub  = &dev->od_scrub;
1112         struct scrub_file       *sf     = &scrub->os_file;
1113         struct ptlrpc_thread    *thread = &scrub->os_thread;
1114         struct osd_otable_it    *it     = dev->od_otable_it;
1115         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
1116
1117         switch (rc) {
1118         case SCRUB_NEXT_CONTINUE:
1119                 goto next;
1120         case SCRUB_NEXT_WAIT:
1121                 goto wait;
1122         case SCRUB_NEXT_NOSCRUB:
1123                 down_write(&scrub->os_rwsem);
1124                 scrub->os_new_checked++;
1125                 sf->sf_items_noscrub++;
1126                 up_write(&scrub->os_rwsem);
1127                 goto next;
1128         }
1129
1130         rc = osd_scrub_check_update(info, dev, oic, rc);
1131         if (rc != 0) {
1132                 scrub->os_in_prior = 0;
1133                 return rc;
1134         }
1135
1136         rc = osd_scrub_checkpoint(scrub);
1137         if (rc != 0) {
1138                 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1139                        "rc = %d\n", osd_scrub2name(scrub),
1140                        scrub->os_pos_current, rc);
1141                 /* Continue, as long as the scrub itself can go ahead. */
1142         }
1143
1144         if (scrub->os_in_prior) {
1145                 scrub->os_in_prior = 0;
1146                 return 0;
1147         }
1148
1149 next:
1150         scrub->os_pos_current = param->gbase + ++(param->offset);
1151
1152 wait:
1153         if (it != NULL && it->ooi_waiting && ooc != NULL &&
1154             ooc->ooc_pos_preload < scrub->os_pos_current) {
1155                 spin_lock(&scrub->os_lock);
1156                 it->ooi_waiting = 0;
1157                 wake_up_all(&thread->t_ctl_waitq);
1158                 spin_unlock(&scrub->os_lock);
1159         }
1160
1161         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1162                 return 0;
1163
1164         if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1165                 *noslot = false;
1166                 return 0;
1167         }
1168
1169         if (it != NULL)
1170                 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1171                              &lwi);
1172
1173         if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1174                 *noslot = false;
1175         else
1176                 *noslot = true;
1177         return 0;
1178 }
1179
1180 static int osd_preload_exec(struct osd_thread_info *info,
1181                             struct osd_device *dev, struct osd_iit_param *param,
1182                             struct osd_idmap_cache *oic, bool *noslot, int rc)
1183 {
1184         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1185
1186         if (rc == 0) {
1187                 ooc->ooc_cached_items++;
1188                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1189                                         ~OSD_OTABLE_IT_CACHE_MASK;
1190         }
1191         return rc > 0 ? 0 : rc;
1192 }
1193
1194 #define SCRUB_IT_ALL    1
1195 #define SCRUB_IT_CRASH  2
1196
1197 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1198                            bool inconsistent)
1199 {
1200         struct osd_scrub     *scrub  = &dev->od_scrub;
1201         struct ptlrpc_thread *thread = &scrub->os_thread;
1202         struct scrub_file    *sf     = &scrub->os_file;
1203         int                   rc;
1204         ENTRY;
1205
1206         LASSERT(!(flags & SS_AUTO_PARTIAL));
1207
1208         down_write(&scrub->os_rwsem);
1209         scrub->os_in_join = 1;
1210         if (flags & SS_SET_FAILOUT)
1211                 sf->sf_param |= SP_FAILOUT;
1212         else if (flags & SS_CLEAR_FAILOUT)
1213                 sf->sf_param &= ~SP_FAILOUT;
1214
1215         if (flags & SS_SET_DRYRUN)
1216                 sf->sf_param |= SP_DRYRUN;
1217         else if (flags & SS_CLEAR_DRYRUN)
1218                 sf->sf_param &= ~SP_DRYRUN;
1219
1220         if (flags & SS_RESET) {
1221                 osd_scrub_file_reset(scrub,
1222                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1223                         inconsistent ? SF_INCONSISTENT : 0);
1224                 sf->sf_status = SS_SCANNING;
1225         }
1226
1227         if (flags & SS_AUTO_FULL) {
1228                 sf->sf_flags |= SF_AUTO;
1229                 scrub->os_full_speed = 1;
1230         }
1231
1232         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1233                 scrub->os_full_speed = 1;
1234         else
1235                 scrub->os_full_speed = 0;
1236
1237         scrub->os_new_checked = 0;
1238         if (sf->sf_pos_last_checkpoint != 0)
1239                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1240         else
1241                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1242
1243         scrub->os_pos_current = sf->sf_pos_latest_start;
1244         sf->sf_time_latest_start = cfs_time_current_sec();
1245         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1246         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1247         rc = osd_scrub_file_store(scrub);
1248         if (rc != 0)
1249                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
1250                        "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1251
1252         spin_lock(&scrub->os_lock);
1253         scrub->os_waiting = 0;
1254         scrub->os_paused = 0;
1255         scrub->os_partial_scan = 0;
1256         scrub->os_in_join = 0;
1257         scrub->os_full_scrub = 0;
1258         spin_unlock(&scrub->os_lock);
1259         wake_up_all(&thread->t_ctl_waitq);
1260         up_write(&scrub->os_rwsem);
1261
1262         EXIT;
1263 }
1264
1265 static int osd_inode_iteration(struct osd_thread_info *info,
1266                                struct osd_device *dev, __u32 max, bool preload)
1267 {
1268         struct osd_scrub     *scrub  = &dev->od_scrub;
1269         struct ptlrpc_thread *thread = &scrub->os_thread;
1270         struct scrub_file    *sf     = &scrub->os_file;
1271         osd_iit_next_policy   next;
1272         osd_iit_exec_policy   exec;
1273         __u32                *pos;
1274         __u32                *count;
1275         struct osd_iit_param  param  = { NULL };
1276         struct l_wait_info    lwi    = { 0 };
1277         __u32                 limit;
1278         int                   rc;
1279         bool                  noslot = true;
1280         ENTRY;
1281
1282         param.sb = osd_sb(dev);
1283         if (preload)
1284                 goto full;
1285
1286         while (scrub->os_partial_scan && !scrub->os_in_join) {
1287                 struct osd_idmap_cache *oic = NULL;
1288
1289                 rc = osd_scrub_next(info, dev, &param, &oic, noslot);
1290                 switch (rc) {
1291                 case SCRUB_NEXT_EXIT:
1292                         RETURN(0);
1293                 case SCRUB_NEXT_CRASH:
1294                         RETURN(SCRUB_IT_CRASH);
1295                 case SCRUB_NEXT_FATAL:
1296                         RETURN(-EINVAL);
1297                 case SCRUB_NEXT_WAIT: {
1298                         struct kstatfs *ksfs = &info->oti_ksfs;
1299                         __u64 saved_flags;
1300
1301                         if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1302                             unlikely(sf->sf_items_updated_prior == 0))
1303                                 goto wait;
1304
1305                         if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1306                             scrub->os_full_scrub) {
1307                                 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1308                                                true);
1309                                 goto full;
1310                         }
1311
1312                         rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1313                         if (rc == 0) {
1314                                 __u64 used = ksfs->f_files - ksfs->f_ffree;
1315
1316                                 do_div(used, sf->sf_items_updated_prior);
1317                                 /* If we hit too much inconsistent OI
1318                                  * mappings during the partial scan,
1319                                  * then scan the device completely. */
1320                                 if (used < dev->od_full_scrub_ratio) {
1321                                         osd_scrub_join(dev,
1322                                                 SS_AUTO_FULL | SS_RESET, true);
1323                                         goto full;
1324                                 }
1325                         }
1326
1327 wait:
1328                         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1329                             cfs_fail_val > 0)
1330                                 continue;
1331
1332                         saved_flags = sf->sf_flags;
1333                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1334                                           SF_UPGRADE | SF_AUTO);
1335                         sf->sf_status = SS_COMPLETED;
1336                         l_wait_event(thread->t_ctl_waitq,
1337                                      !thread_is_running(thread) ||
1338                                      !scrub->os_partial_scan ||
1339                                      scrub->os_in_join ||
1340                                      !list_empty(&scrub->os_inconsistent_items),
1341                                      &lwi);
1342                         sf->sf_flags = saved_flags;
1343                         sf->sf_status = SS_SCANNING;
1344
1345                         if (unlikely(!thread_is_running(thread)))
1346                                 RETURN(0);
1347
1348                         if (!scrub->os_partial_scan || scrub->os_in_join)
1349                                 goto full;
1350
1351                         continue;
1352                 }
1353                 default:
1354                         LASSERTF(rc == 0, "rc = %d\n", rc);
1355
1356                         osd_scrub_exec(info, dev, &param, oic, &noslot, rc);
1357                         break;
1358                 }
1359         }
1360
1361 full:
1362         if (!preload) {
1363                 l_wait_event(thread->t_ctl_waitq,
1364                              !thread_is_running(thread) || !scrub->os_in_join,
1365                              &lwi);
1366
1367                 if (unlikely(!thread_is_running(thread)))
1368                         RETURN(0);
1369         }
1370
1371         noslot = false;
1372         if (!preload) {
1373                 next = osd_scrub_next;
1374                 exec = osd_scrub_exec;
1375                 pos = &scrub->os_pos_current;
1376                 count = &scrub->os_new_checked;
1377         } else {
1378                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1379
1380                 next = osd_preload_next;
1381                 exec = osd_preload_exec;
1382                 pos = &ooc->ooc_pos_preload;
1383                 count = &ooc->ooc_cached_items;
1384         }
1385         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1386
1387         while (*pos <= limit && *count < max) {
1388                 struct osd_idmap_cache *oic = NULL;
1389                 struct ldiskfs_group_desc *desc;
1390
1391                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1392                 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1393                 if (desc == NULL)
1394                         RETURN(-EIO);
1395
1396                 ldiskfs_lock_group(param.sb, param.bg);
1397                 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1398                         ldiskfs_unlock_group(param.sb, param.bg);
1399                         *pos = 1 + (param.bg + 1) *
1400                                 LDISKFS_INODES_PER_GROUP(param.sb);
1401                         continue;
1402                 }
1403                 ldiskfs_unlock_group(param.sb, param.bg);
1404
1405                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1406                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1407                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1408                 if (param.bitmap == NULL) {
1409                         CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1410                                "scrub will stop, urgent mode\n",
1411                                osd_scrub2name(scrub), (__u32)param.bg);
1412                         RETURN(-EIO);
1413                 }
1414
1415                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1416                        *count < max) {
1417                         if (param.offset +
1418                                 ldiskfs_itable_unused_count(param.sb, desc) >
1419                             LDISKFS_INODES_PER_GROUP(param.sb)) {
1420                                 *pos = 1 + (param.bg + 1) *
1421                                         LDISKFS_INODES_PER_GROUP(param.sb);
1422                                 goto next_group;
1423                         }
1424
1425                         rc = next(info, dev, &param, &oic, noslot);
1426                         switch (rc) {
1427                         case SCRUB_NEXT_BREAK:
1428                                 goto next_group;
1429                         case SCRUB_NEXT_EXIT:
1430                                 brelse(param.bitmap);
1431                                 RETURN(0);
1432                         case SCRUB_NEXT_CRASH:
1433                                 brelse(param.bitmap);
1434                                 RETURN(SCRUB_IT_CRASH);
1435                         case SCRUB_NEXT_FATAL:
1436                                 brelse(param.bitmap);
1437                                 RETURN(-EINVAL);
1438                         }
1439
1440                         rc = exec(info, dev, &param, oic, &noslot, rc);
1441                         if (rc != 0) {
1442                                 brelse(param.bitmap);
1443                                 RETURN(rc);
1444                         }
1445                 }
1446
1447 next_group:
1448                 brelse(param.bitmap);
1449         }
1450
1451         if (*pos > limit)
1452                 RETURN(SCRUB_IT_ALL);
1453         RETURN(0);
1454 }
1455
1456 static int osd_otable_it_preload(const struct lu_env *env,
1457                                  struct osd_otable_it *it)
1458 {
1459         struct osd_device       *dev   = it->ooi_dev;
1460         struct osd_scrub        *scrub = &dev->od_scrub;
1461         struct osd_otable_cache *ooc   = &it->ooi_cache;
1462         int                      rc;
1463         ENTRY;
1464
1465         rc = osd_inode_iteration(osd_oti_get(env), dev,
1466                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1467         if (rc == SCRUB_IT_ALL)
1468                 it->ooi_all_cached = 1;
1469
1470         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1471                 spin_lock(&scrub->os_lock);
1472                 scrub->os_waiting = 0;
1473                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1474                 spin_unlock(&scrub->os_lock);
1475         }
1476
1477         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1478 }
1479
1480 static int osd_scrub_main(void *args)
1481 {
1482         struct lu_env         env;
1483         struct osd_device    *dev    = (struct osd_device *)args;
1484         struct osd_scrub     *scrub  = &dev->od_scrub;
1485         struct ptlrpc_thread *thread = &scrub->os_thread;
1486         int                   rc;
1487         ENTRY;
1488
1489         rc = lu_env_init(&env, LCT_LOCAL);
1490         if (rc != 0) {
1491                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1492                        osd_scrub2name(scrub), rc);
1493                 GOTO(noenv, rc);
1494         }
1495
1496         rc = osd_scrub_prep(dev);
1497         if (rc != 0) {
1498                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1499                        osd_scrub2name(scrub), rc);
1500                 GOTO(out, rc);
1501         }
1502
1503         if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1504                 struct l_wait_info lwi = { 0 };
1505                 struct osd_otable_it *it = dev->od_otable_it;
1506                 struct osd_otable_cache *ooc = &it->ooi_cache;
1507
1508                 l_wait_event(thread->t_ctl_waitq,
1509                              it->ooi_user_ready || !thread_is_running(thread),
1510                              &lwi);
1511                 if (unlikely(!thread_is_running(thread)))
1512                         GOTO(post, rc = 0);
1513
1514                 scrub->os_pos_current = ooc->ooc_pos_preload;
1515         }
1516
1517         CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1518                osd_scrub2name(scrub), scrub->os_start_flags,
1519                scrub->os_pos_current);
1520
1521         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1522         if (unlikely(rc == SCRUB_IT_CRASH))
1523                 GOTO(out, rc = -EINVAL);
1524         GOTO(post, rc);
1525
1526 post:
1527         rc = osd_scrub_post(scrub, rc);
1528         CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1529                osd_scrub2name(scrub), scrub->os_pos_current, rc);
1530
1531 out:
1532         while (!list_empty(&scrub->os_inconsistent_items)) {
1533                 struct osd_inconsistent_item *oii;
1534
1535                 oii = list_entry(scrub->os_inconsistent_items.next,
1536                                      struct osd_inconsistent_item, oii_list);
1537                 list_del_init(&oii->oii_list);
1538                 OBD_FREE_PTR(oii);
1539         }
1540         lu_env_fini(&env);
1541
1542 noenv:
1543         spin_lock(&scrub->os_lock);
1544         thread_set_flags(thread, SVC_STOPPED);
1545         wake_up_all(&thread->t_ctl_waitq);
1546         spin_unlock(&scrub->os_lock);
1547         return rc;
1548 }
1549
1550 /* initial OI scrub */
1551
1552 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1553                          struct dentry *, filldir_t filldir);
1554
1555 #ifdef HAVE_FILLDIR_USE_CTX
1556 static int osd_ios_varfid_fill(struct dir_context *buf, const char *name,
1557                                int namelen, loff_t offset, __u64 ino,
1558                                unsigned d_type);
1559 static int osd_ios_lf_fill(struct dir_context *buf, const char *name,
1560                            int namelen, loff_t offset, __u64 ino,
1561                            unsigned d_type);
1562 static int osd_ios_dl_fill(struct dir_context *buf, const char *name,
1563                            int namelen, loff_t offset, __u64 ino,
1564                            unsigned d_type);
1565 static int osd_ios_uld_fill(struct dir_context *buf, const char *name,
1566                             int namelen, loff_t offset, __u64 ino,
1567                             unsigned d_type);
1568 #else
1569 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1570                                loff_t offset, __u64 ino, unsigned d_type);
1571 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1572                            loff_t offset, __u64 ino, unsigned d_type);
1573 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1574                            loff_t offset, __u64 ino, unsigned d_type);
1575 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1576                             loff_t offset, __u64 ino, unsigned d_type);
1577 #endif
1578
1579 static int
1580 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1581                      struct dentry *dentry, filldir_t filldir);
1582 static int
1583 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1584                   struct dentry *dentry, filldir_t filldir);
1585
1586 static int
1587 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1588                      struct dentry *dentry, filldir_t filldir);
1589
1590 enum osd_lf_flags {
1591         OLF_SCAN_SUBITEMS       = 0x0001,
1592         OLF_HIDE_FID            = 0x0002,
1593         OLF_SHOW_NAME           = 0x0004,
1594         OLF_NO_OI               = 0x0008,
1595         OLF_IDX_IN_FID          = 0x0010,
1596 };
1597
1598 struct osd_lf_map {
1599         char            *olm_name;
1600         struct lu_fid    olm_fid;
1601         __u16            olm_flags;
1602         __u16            olm_namelen;
1603         scandir_t        olm_scandir;
1604         filldir_t        olm_filldir;
1605 };
1606
1607 /* Add the new introduced local files in the list in the future. */
1608 static const struct osd_lf_map osd_lf_maps[] = {
1609         /* CATALOGS */
1610         { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1611                 sizeof(CATLIST) - 1, NULL, NULL },
1612
1613         /* CONFIGS */
1614         { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1615                 OLF_SCAN_SUBITEMS, sizeof(MOUNT_CONFIGS_DIR) - 1,
1616                 osd_ios_general_scan, osd_ios_varfid_fill },
1617
1618         /* NIDTBL_VERSIONS */
1619         { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1620                 sizeof(MGS_NIDTBL_DIR) - 1, osd_ios_general_scan,
1621                 osd_ios_varfid_fill },
1622
1623         /* PENDING */
1624         { "PENDING", { 0, 0, 0 }, 0, sizeof("PENDING") - 1, NULL, NULL },
1625
1626         /* ROOT */
1627         { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 },
1628                 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, sizeof("ROOT") - 1,
1629                 osd_ios_ROOT_scan, NULL },
1630
1631         /* changelog_catalog */
1632         { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, sizeof(CHANGELOG_CATALOG) - 1,
1633                 NULL, NULL },
1634
1635         /* changelog_users */
1636         { CHANGELOG_USERS, { 0, 0, 0 }, 0, sizeof(CHANGELOG_USERS) - 1,
1637                 NULL, NULL },
1638
1639         /* fld */
1640         { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1641                 sizeof("fld") - 1, NULL, NULL },
1642
1643         /* last_rcvd */
1644         { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1645                 sizeof(LAST_RCVD) - 1, NULL, NULL },
1646
1647         /* reply_data */
1648         { REPLY_DATA, { FID_SEQ_LOCAL_FILE, REPLY_DATA_OID, 0 }, OLF_SHOW_NAME,
1649                 sizeof(REPLY_DATA) - 1, NULL, NULL },
1650
1651         /* lov_objid */
1652         { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1653                 sizeof(LOV_OBJID) - 1, NULL, NULL },
1654
1655         /* lov_objseq */
1656         { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1657                 OLF_SHOW_NAME, sizeof(LOV_OBJSEQ) - 1, NULL, NULL },
1658
1659         /* quota_master */
1660         { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QMT_DIR) - 1,
1661                 osd_ios_general_scan, osd_ios_varfid_fill },
1662
1663         /* quota_slave */
1664         { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QSD_DIR) - 1,
1665                 osd_ios_general_scan, osd_ios_varfid_fill },
1666
1667         /* seq_ctl */
1668         { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1669                 OLF_SHOW_NAME, sizeof("seq_ctl") - 1, NULL, NULL },
1670
1671         /* seq_srv */
1672         { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1673                 OLF_SHOW_NAME, sizeof("seq_srv") - 1, NULL, NULL },
1674
1675         /* health_check */
1676         { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1677                 OLF_SHOW_NAME, sizeof(HEALTH_CHECK) - 1, NULL, NULL },
1678
1679         /* LFSCK */
1680         { LFSCK_DIR, { 0, 0, 0 }, 0, sizeof(LFSCK_DIR) - 1,
1681                 osd_ios_general_scan, osd_ios_varfid_fill },
1682
1683         /* lfsck_bookmark */
1684         { LFSCK_BOOKMARK, { 0, 0, 0 }, 0, sizeof(LFSCK_BOOKMARK) - 1,
1685                 NULL, NULL },
1686
1687         /* lfsck_layout */
1688         { LFSCK_LAYOUT, { 0, 0, 0 }, 0, sizeof(LFSCK_LAYOUT) - 1,
1689                 NULL, NULL },
1690
1691         /* lfsck_namespace */
1692         { LFSCK_NAMESPACE, { 0, 0, 0 }, 0, sizeof(LFSCK_NAMESPACE) - 1,
1693                 NULL, NULL },
1694
1695         /* OBJECTS, upgrade from old device */
1696         { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(OBJECTS) - 1,
1697                 osd_ios_OBJECTS_scan, NULL },
1698
1699         /* lquota_v2.user, upgrade from old device */
1700         { "lquota_v2.user", { 0, 0, 0 }, 0, sizeof("lquota_v2.user") - 1,
1701                 NULL, NULL },
1702
1703         /* lquota_v2.group, upgrade from old device */
1704         { "lquota_v2.group", { 0, 0, 0 }, 0, sizeof("lquota_v2.group") - 1,
1705                 NULL, NULL },
1706
1707         /* LAST_GROUP, upgrade from old device */
1708         { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1709                 OLF_SHOW_NAME, sizeof("LAST_GROUP") - 1, NULL, NULL },
1710
1711         /* committed batchid for cross-MDT operation */
1712         { "BATCHID", { FID_SEQ_LOCAL_FILE, BATCHID_COMMITTED_OID, 0 },
1713                 OLF_SHOW_NAME, sizeof("BATCHID") - 1, NULL, NULL },
1714
1715         /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1716          * and f_oid = index for their log files.  See lu_update_log{_dir}_fid()
1717          * for more details. */
1718
1719         /* update_log */
1720         { "update_log", { FID_SEQ_UPDATE_LOG, 0, 0 },
1721                 OLF_SHOW_NAME | OLF_IDX_IN_FID, sizeof("update_log") - 1,
1722                 NULL, NULL },
1723
1724         /* update_log_dir */
1725         { "update_log_dir", { FID_SEQ_UPDATE_LOG_DIR, 0, 0 },
1726                 OLF_SHOW_NAME | OLF_SCAN_SUBITEMS | OLF_IDX_IN_FID,
1727                 sizeof("update_log_dir") - 1,
1728                 osd_ios_general_scan, osd_ios_uld_fill },
1729
1730         /* lost+found */
1731         { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 },
1732                 OLF_SCAN_SUBITEMS, sizeof("lost+found") - 1,
1733                 osd_ios_general_scan, osd_ios_lf_fill },
1734
1735         { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1736 };
1737
1738 /* Add the new introduced files under .lustre/ in the list in the future. */
1739 static const struct osd_lf_map osd_dl_maps[] = {
1740         /* .lustre/fid */
1741         { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1742                 sizeof("fid") - 1, NULL, NULL },
1743
1744         /* .lustre/lost+found */
1745         { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1746                 sizeof("lost+found") - 1, NULL, NULL },
1747
1748         { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1749 };
1750
1751 struct osd_ios_item {
1752         struct list_head oii_list;
1753         struct dentry   *oii_dentry;
1754         scandir_t        oii_scandir;
1755         filldir_t        oii_filldir;
1756 };
1757
1758 struct osd_ios_filldir_buf {
1759 #ifdef HAVE_DIR_CONTEXT
1760         /* please keep it as first member */
1761         struct dir_context       ctx;
1762 #endif
1763         struct osd_thread_info  *oifb_info;
1764         struct osd_device       *oifb_dev;
1765         struct dentry           *oifb_dentry;
1766 };
1767
1768 static inline struct dentry *
1769 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1770 {
1771         struct dentry *dentry;
1772
1773         dentry = ll_lookup_one_len(name, parent, namelen);
1774         if (IS_ERR(dentry)) {
1775                 int rc = PTR_ERR(dentry);
1776
1777                 if (rc != -ENOENT)
1778                         CERROR("Fail to find %.*s in %.*s (%lu/%u): rc = %d\n",
1779                                namelen, name, parent->d_name.len,
1780                                parent->d_name.name, parent->d_inode->i_ino,
1781                                parent->d_inode->i_generation, rc);
1782
1783                 return dentry;
1784         }
1785
1786         if (dentry->d_inode == NULL) {
1787                 dput(dentry);
1788                 return ERR_PTR(-ENOENT);
1789         }
1790
1791         return dentry;
1792 }
1793
1794 static int
1795 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1796                  scandir_t scandir, filldir_t filldir)
1797 {
1798         struct osd_ios_item *item;
1799         ENTRY;
1800
1801         OBD_ALLOC_PTR(item);
1802         if (item == NULL)
1803                 RETURN(-ENOMEM);
1804
1805         INIT_LIST_HEAD(&item->oii_list);
1806         item->oii_dentry = dget(dentry);
1807         item->oii_scandir = scandir;
1808         item->oii_filldir = filldir;
1809         list_add_tail(&item->oii_list, &dev->od_ios_list);
1810
1811         RETURN(0);
1812 }
1813
1814 /**
1815  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1816  *
1817  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1818  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1819  * reference the inode, or fixed if it is missing or references another inode.
1820  */
1821 static int
1822 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1823                  struct inode *inode, const struct lu_fid *fid, int flags)
1824 {
1825         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
1826         struct osd_inode_id     *id     = &info->oti_id;
1827         struct osd_inode_id     *id2    = &info->oti_id2;
1828         struct osd_scrub        *scrub  = &dev->od_scrub;
1829         struct scrub_file       *sf     = &scrub->os_file;
1830         struct lu_fid            tfid;
1831         int                      rc;
1832         ENTRY;
1833
1834         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1835         if (rc != 0 && rc != -ENODATA) {
1836                 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1837                        "rc = %d\n", osd_name(dev), rc);
1838
1839                 RETURN(rc);
1840         }
1841
1842         osd_id_gen(id, inode->i_ino, inode->i_generation);
1843         if (rc == -ENODATA) {
1844                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
1845                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1846                 } else {
1847                         tfid = *fid;
1848                         if (flags & OLF_IDX_IN_FID) {
1849                                 LASSERT(dev->od_index >= 0);
1850
1851                                 tfid.f_oid = dev->od_index;
1852                         }
1853                 }
1854                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1855                 if (rc != 0) {
1856                         CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1857                               "scrub: rc = %d\n", osd_name(dev), rc);
1858
1859                         RETURN(rc);
1860                 }
1861         } else {
1862                 if (lma->lma_compat & LMAC_NOT_IN_OI)
1863                         RETURN(0);
1864
1865                 tfid = lma->lma_self_fid;
1866         }
1867
1868         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1869         if (rc != 0) {
1870                 if (rc != -ENOENT)
1871                         RETURN(rc);
1872
1873                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1874                                                DTO_INDEX_INSERT, true, 0, NULL);
1875                 if (rc > 0)
1876                         rc = 0;
1877
1878                 RETURN(rc);
1879         }
1880
1881         if (osd_id_eq_strict(id, id2))
1882                 RETURN(0);
1883
1884         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1885                 osd_scrub_file_reset(scrub,
1886                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1887                                      SF_INCONSISTENT);
1888                 rc = osd_scrub_file_store(scrub);
1889                 if (rc != 0)
1890                         RETURN(rc);
1891         }
1892
1893         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1894                                        DTO_INDEX_UPDATE, true, 0, NULL);
1895         if (rc > 0)
1896                 rc = 0;
1897
1898         RETURN(rc);
1899 }
1900
1901 /**
1902  * It scans the /lost+found, and for the OST-object (with filter_fid
1903  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1904  */
1905 #ifdef HAVE_FILLDIR_USE_CTX
1906 static int osd_ios_lf_fill(struct dir_context *buf, const char *name,
1907                            int namelen,
1908 #else
1909 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1910 #endif
1911                            loff_t offset, __u64 ino, unsigned d_type)
1912 {
1913         struct osd_ios_filldir_buf *fill_buf =
1914                 (struct osd_ios_filldir_buf *)buf;
1915         struct osd_thread_info     *info     = fill_buf->oifb_info;
1916         struct osd_device          *dev      = fill_buf->oifb_dev;
1917         struct lu_fid              *fid      = &info->oti_fid;
1918         struct osd_scrub           *scrub    = &dev->od_scrub;
1919         struct dentry              *parent   = fill_buf->oifb_dentry;
1920         struct dentry              *child;
1921         struct inode               *dir      = parent->d_inode;
1922         struct inode               *inode;
1923         int                         rc;
1924         ENTRY;
1925
1926         /* skip any '.' started names */
1927         if (name[0] == '.')
1928                 RETURN(0);
1929
1930         scrub->os_lf_scanned++;
1931         child = osd_ios_lookup_one_len(name, parent, namelen);
1932         if (IS_ERR(child)) {
1933                 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1934                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
1935                 RETURN(0);
1936         }
1937
1938         inode = child->d_inode;
1939         if (S_ISDIR(inode->i_mode)) {
1940                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1941                                       osd_ios_lf_fill);
1942                 if (rc != 0)
1943                         CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1944                               "rc = %d\n", osd_name(dev), namelen, name, rc);
1945                 GOTO(put, rc);
1946         }
1947
1948         if (!S_ISREG(inode->i_mode))
1949                 GOTO(put, rc = 0);
1950
1951         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1952         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1953                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1954                 if (rc == 0) {
1955                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1956                                "/lost+found.\n", namelen, name, PFID(fid));
1957                         scrub->os_lf_repaired++;
1958                 } else {
1959                         CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1960                                DFID": rc = %d\n",
1961                                osd_name(dev), namelen, name, PFID(fid), rc);
1962                 }
1963         }
1964
1965         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1966          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1967          *      can process them in furtuer. */
1968
1969         GOTO(put, rc);
1970
1971 put:
1972         if (rc < 0)
1973                 scrub->os_lf_failed++;
1974         dput(child);
1975         /* skip the failure to make the scanning to continue. */
1976         return 0;
1977 }
1978
1979 #ifdef HAVE_FILLDIR_USE_CTX
1980 static int osd_ios_varfid_fill(struct dir_context  *buf, const char *name,
1981                                int namelen,
1982 #else
1983 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1984 #endif
1985                                loff_t offset, __u64 ino, unsigned d_type)
1986 {
1987         struct osd_ios_filldir_buf *fill_buf =
1988                 (struct osd_ios_filldir_buf *)buf;
1989         struct osd_device          *dev      = fill_buf->oifb_dev;
1990         struct dentry              *child;
1991         int                         rc;
1992         ENTRY;
1993
1994         /* skip any '.' started names */
1995         if (name[0] == '.')
1996                 RETURN(0);
1997
1998         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1999         if (IS_ERR(child))
2000                 RETURN(PTR_ERR(child));
2001
2002         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2003                               NULL, 0);
2004         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
2005                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2006                                       osd_ios_varfid_fill);
2007         dput(child);
2008
2009         RETURN(rc);
2010 }
2011
2012 #ifdef HAVE_FILLDIR_USE_CTX
2013 static int osd_ios_dl_fill(struct dir_context  *buf, const char *name,
2014                            int namelen,
2015 #else
2016 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
2017 #endif
2018                            loff_t offset, __u64 ino, unsigned d_type)
2019 {
2020         struct osd_ios_filldir_buf *fill_buf =
2021                 (struct osd_ios_filldir_buf *)buf;
2022         struct osd_device          *dev      = fill_buf->oifb_dev;
2023         const struct osd_lf_map    *map;
2024         struct dentry              *child;
2025         int                         rc       = 0;
2026         ENTRY;
2027
2028         /* skip any '.' started names */
2029         if (name[0] == '.')
2030                 RETURN(0);
2031
2032         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
2033                 if (map->olm_namelen != namelen)
2034                         continue;
2035
2036                 if (strncmp(map->olm_name, name, namelen) == 0)
2037                         break;
2038         }
2039
2040         if (map->olm_name == NULL)
2041                 RETURN(0);
2042
2043         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2044         if (IS_ERR(child))
2045                 RETURN(PTR_ERR(child));
2046
2047         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2048                               &map->olm_fid, map->olm_flags);
2049         dput(child);
2050
2051         RETURN(rc);
2052 }
2053
2054 #ifdef HAVE_FILLDIR_USE_CTX
2055 static int osd_ios_uld_fill(struct dir_context *buf, const char *name,
2056                             int namelen,
2057 #else
2058 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
2059 #endif
2060                             loff_t offset, __u64 ino, unsigned d_type)
2061 {
2062         struct osd_ios_filldir_buf *fill_buf =
2063                 (struct osd_ios_filldir_buf *)buf;
2064         struct dentry              *child;
2065         struct lu_fid               tfid;
2066         int                         rc       = 0;
2067         ENTRY;
2068
2069         /* skip any non-DFID format name */
2070         if (name[0] != '[')
2071                 RETURN(0);
2072
2073         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2074         if (IS_ERR(child))
2075                 RETURN(PTR_ERR(child));
2076
2077         /* skip the start '[' */
2078         sscanf(&name[1], SFID, RFID(&tfid));
2079         if (fid_is_sane(&tfid))
2080                 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
2081                                       child->d_inode, &tfid, 0);
2082         else
2083                 rc = -EIO;
2084         dput(child);
2085
2086         RETURN(rc);
2087 }
2088
2089 #ifdef HAVE_FILLDIR_USE_CTX
2090 static int osd_ios_root_fill(struct dir_context *buf, const char *name,
2091                              int namelen,
2092 #else
2093 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
2094 #endif
2095                              loff_t offset, __u64 ino, unsigned d_type)
2096 {
2097         struct osd_ios_filldir_buf *fill_buf =
2098                 (struct osd_ios_filldir_buf *)buf;
2099         struct osd_device          *dev      = fill_buf->oifb_dev;
2100         const struct osd_lf_map    *map;
2101         struct dentry              *child;
2102         int                         rc       = 0;
2103         ENTRY;
2104
2105         /* skip any '.' started names */
2106         if (name[0] == '.')
2107                 RETURN(0);
2108
2109         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2110                 if (map->olm_namelen != namelen)
2111                         continue;
2112
2113                 if (strncmp(map->olm_name, name, namelen) == 0)
2114                         break;
2115         }
2116
2117         if (map->olm_name == NULL)
2118                 RETURN(0);
2119
2120         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2121         if (IS_ERR(child))
2122                 RETURN(PTR_ERR(child));
2123
2124         if (!(map->olm_flags & OLF_NO_OI))
2125                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2126                                       &map->olm_fid, map->olm_flags);
2127         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2128                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2129                                       map->olm_filldir);
2130         dput(child);
2131
2132         RETURN(rc);
2133 }
2134
2135 static int
2136 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2137                      struct dentry *dentry, filldir_t filldir)
2138 {
2139         struct osd_ios_filldir_buf    buf   = {
2140 #ifdef HAVE_DIR_CONTEXT
2141                                                 .ctx.actor = filldir,
2142 #endif
2143                                                 .oifb_info = info,
2144                                                 .oifb_dev = dev,
2145                                                 .oifb_dentry = dentry };
2146         struct file                  *filp  = &info->oti_file;
2147         struct inode                 *inode = dentry->d_inode;
2148         const struct file_operations *fops  = inode->i_fop;
2149         int                           rc;
2150         ENTRY;
2151
2152         LASSERT(filldir != NULL);
2153
2154         filp->f_pos = 0;
2155         filp->f_path.dentry = dentry;
2156         filp->f_mode = FMODE_64BITHASH;
2157         filp->f_mapping = inode->i_mapping;
2158         filp->f_op = fops;
2159         filp->private_data = NULL;
2160         set_file_inode(filp, inode);
2161
2162 #ifdef HAVE_DIR_CONTEXT
2163         buf.ctx.pos = filp->f_pos;
2164         rc = fops->iterate(filp, &buf.ctx);
2165         filp->f_pos = buf.ctx.pos;
2166 #else
2167         rc = fops->readdir(filp, &buf, filldir);
2168 #endif
2169         fops->release(inode, filp);
2170
2171         RETURN(rc);
2172 }
2173
2174 static int
2175 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2176                   struct dentry *dentry, filldir_t filldir)
2177 {
2178         struct osd_scrub  *scrub  = &dev->od_scrub;
2179         struct scrub_file *sf     = &scrub->os_file;
2180         struct dentry     *child;
2181         int                rc;
2182         ENTRY;
2183
2184         /* It is existing MDT0 device. We only allow the case of object without
2185          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2186          * can generate IGIF mode FID for the object and related OI mapping. If
2187          * it is on other MDTs, then becuase file-level backup/restore, related
2188          * OI mapping may be invalid already, we do not know which is the right
2189          * FID for the object. We only allow IGIF objects to reside on the MDT0.
2190          *
2191          * XXX: For the case of object on non-MDT0 device with neither LMA nor
2192          *      "fid" xattr, then something crashed. We cannot re-generate the
2193          *      FID directly, instead, the OI scrub will scan the OI structure
2194          *      and try to re-generate the LMA from the OI mapping. But if the
2195          *      OI mapping crashed or lost also, then we have to give up under
2196          *      double failure cases. */
2197         scrub->os_convert_igif = 1;
2198         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2199                                        strlen(dot_lustre_name));
2200         if (IS_ERR(child)) {
2201                 rc = PTR_ERR(child);
2202                 if (rc == -ENOENT) {
2203                         /* It is 1.8 MDT device. */
2204                         if (!(sf->sf_flags & SF_UPGRADE)) {
2205                                 osd_scrub_file_reset(scrub,
2206                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2207                                         SF_UPGRADE);
2208                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2209                                 rc = osd_scrub_file_store(scrub);
2210                         } else {
2211                                 rc = 0;
2212                         }
2213                 }
2214         } else {
2215                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2216                  * so the client will get IGIF for the ".lustre" object when
2217                  * the MDT restart.
2218                  *
2219                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2220                  * it does not know whether there are some old clients cached
2221                  * the ".lustre" IGIF during the upgrading. Two choices:
2222                  *
2223                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2224                  *    It will allow the old connected clients to access the
2225                  *    ".lustre" with cached IGIF. But it will cause others
2226                  *    on the MDT failed to check "fid_is_dot_lustre()".
2227                  *
2228                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2229                  *    for ".lustre" in spite of whether there are some clients
2230                  *    cached the ".lustre" IGIF or not. It enables the check
2231                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
2232                  *    that the old connected clients cannot access the ".lustre"
2233                  *    with the cached IGIF.
2234                  *
2235                  * Usually, it is rare case for the old connected clients
2236                  * to access the ".lustre" with cached IGIF. So we prefer
2237                  * to the solution 2). */
2238                 rc = osd_ios_scan_one(info, dev, child->d_inode,
2239                                       &LU_DOT_LUSTRE_FID, 0);
2240                 if (rc == 0)
2241                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2242                                               osd_ios_dl_fill);
2243                 dput(child);
2244         }
2245
2246         RETURN(rc);
2247 }
2248
2249 static int
2250 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2251                      struct dentry *dentry, filldir_t filldir)
2252 {
2253         struct osd_scrub  *scrub  = &dev->od_scrub;
2254         struct scrub_file *sf     = &scrub->os_file;
2255         struct dentry     *child;
2256         int                rc;
2257         ENTRY;
2258
2259         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2260                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2261                 rc = osd_scrub_file_store(scrub);
2262                 if (rc != 0)
2263                         RETURN(rc);
2264         }
2265
2266         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2267         if (!IS_ERR(child)) {
2268                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2269                 dput(child);
2270         } else {
2271                 rc = PTR_ERR(child);
2272         }
2273
2274         if (rc != 0 && rc != -ENOENT)
2275                 RETURN(rc);
2276
2277         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2278         if (!IS_ERR(child)) {
2279                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2280                 dput(child);
2281         } else {
2282                 rc = PTR_ERR(child);
2283         }
2284
2285         if (rc == -ENOENT)
2286                 rc = 0;
2287
2288         RETURN(rc);
2289 }
2290
2291 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2292                                 struct osd_device *dev)
2293 {
2294         struct osd_ios_item     *item    = NULL;
2295         scandir_t                scandir = osd_ios_general_scan;
2296         filldir_t                filldir = osd_ios_root_fill;
2297         struct dentry           *dentry  = osd_sb(dev)->s_root;
2298         const struct osd_lf_map *map     = osd_lf_maps;
2299         int                      rc;
2300         ENTRY;
2301
2302         /* Lookup IGIF in OI by force for initial OI scrub. */
2303         dev->od_igif_inoi = 1;
2304
2305         while (1) {
2306                 rc = scandir(info, dev, dentry, filldir);
2307                 if (item != NULL) {
2308                         dput(item->oii_dentry);
2309                         OBD_FREE_PTR(item);
2310                 }
2311
2312                 if (rc != 0)
2313                         break;
2314
2315                 if (list_empty(&dev->od_ios_list))
2316                         break;
2317
2318                 item = list_entry(dev->od_ios_list.next,
2319                                   struct osd_ios_item, oii_list);
2320                 list_del_init(&item->oii_list);
2321
2322                 LASSERT(item->oii_scandir != NULL);
2323                 scandir = item->oii_scandir;
2324                 filldir = item->oii_filldir;
2325                 dentry = item->oii_dentry;
2326         }
2327
2328         while (!list_empty(&dev->od_ios_list)) {
2329                 item = list_entry(dev->od_ios_list.next,
2330                                   struct osd_ios_item, oii_list);
2331                 list_del_init(&item->oii_list);
2332                 dput(item->oii_dentry);
2333                 OBD_FREE_PTR(item);
2334         }
2335
2336         if (rc != 0)
2337                 RETURN(rc);
2338
2339         /* There maybe the case that the object has been removed, but its OI
2340          * mapping is still in the OI file, such as the "CATALOGS" after MDT
2341          * file-level backup/restore. So here cleanup the stale OI mappings. */
2342         while (map->olm_name != NULL) {
2343                 struct dentry *child;
2344
2345                 if (fid_is_zero(&map->olm_fid)) {
2346                         map++;
2347                         continue;
2348                 }
2349
2350                 child = osd_ios_lookup_one_len(map->olm_name,
2351                                                osd_sb(dev)->s_root,
2352                                                map->olm_namelen);
2353                 if (!IS_ERR(child))
2354                         dput(child);
2355                 else if (PTR_ERR(child) == -ENOENT)
2356                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2357                                                   NULL, DTO_INDEX_DELETE,
2358                                                   true, 0, NULL);
2359                 map++;
2360         }
2361
2362         RETURN(0);
2363 }
2364
2365 char *osd_lf_fid2name(const struct lu_fid *fid)
2366 {
2367         const struct osd_lf_map *map = osd_lf_maps;
2368
2369         while (map->olm_name != NULL) {
2370                 if (!lu_fid_eq(fid, &map->olm_fid)) {
2371                         map++;
2372                         continue;
2373                 }
2374
2375                 if (map->olm_flags & OLF_SHOW_NAME)
2376                         return map->olm_name;
2377                 else
2378                         return "";
2379         }
2380
2381         return NULL;
2382 }
2383
2384 /* OI scrub start/stop */
2385
2386 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2387 {
2388         struct osd_scrub     *scrub  = &dev->od_scrub;
2389         struct ptlrpc_thread *thread = &scrub->os_thread;
2390         struct l_wait_info    lwi    = { 0 };
2391         struct task_struct   *task;
2392         int                   rc;
2393         ENTRY;
2394
2395         /* os_lock: sync status between stop and scrub thread */
2396         spin_lock(&scrub->os_lock);
2397
2398 again:
2399         if (thread_is_running(thread)) {
2400                 spin_unlock(&scrub->os_lock);
2401                 if (!(scrub->os_file.sf_flags & SF_AUTO ||
2402                       scrub->os_partial_scan) ||
2403                      (flags & SS_AUTO_PARTIAL))
2404                         RETURN(-EALREADY);
2405
2406                 osd_scrub_join(dev, flags, false);
2407                 spin_lock(&scrub->os_lock);
2408                 if (!thread_is_running(thread))
2409                         goto again;
2410
2411                 spin_unlock(&scrub->os_lock);
2412                 RETURN(0);
2413         }
2414
2415         if (unlikely(thread_is_stopping(thread))) {
2416                 spin_unlock(&scrub->os_lock);
2417                 l_wait_event(thread->t_ctl_waitq,
2418                              thread_is_stopped(thread),
2419                              &lwi);
2420                 spin_lock(&scrub->os_lock);
2421                 goto again;
2422         }
2423         spin_unlock(&scrub->os_lock);
2424
2425         if (scrub->os_file.sf_status == SS_COMPLETED) {
2426                 if (!(flags & SS_SET_FAILOUT))
2427                         flags |= SS_CLEAR_FAILOUT;
2428
2429                 if (!(flags & SS_SET_DRYRUN))
2430                         flags |= SS_CLEAR_DRYRUN;
2431
2432                 flags |= SS_RESET;
2433         }
2434
2435         scrub->os_start_flags = flags;
2436         thread_set_flags(thread, 0);
2437         task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2438         if (IS_ERR(task)) {
2439                 rc = PTR_ERR(task);
2440                 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2441                        osd_scrub2name(scrub), rc);
2442                 RETURN(rc);
2443         }
2444
2445         l_wait_event(thread->t_ctl_waitq,
2446                      thread_is_running(thread) || thread_is_stopped(thread),
2447                      &lwi);
2448
2449         RETURN(0);
2450 }
2451
2452 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2453 {
2454         int rc;
2455         ENTRY;
2456
2457         /* od_otable_mutex: prevent curcurrent start/stop */
2458         mutex_lock(&dev->od_otable_mutex);
2459         rc = do_osd_scrub_start(dev, flags);
2460         mutex_unlock(&dev->od_otable_mutex);
2461
2462         RETURN(rc == -EALREADY ? 0 : rc);
2463 }
2464
2465 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2466 {
2467         struct ptlrpc_thread *thread = &scrub->os_thread;
2468         struct l_wait_info    lwi    = { 0 };
2469
2470         /* os_lock: sync status between stop and scrub thread */
2471         spin_lock(&scrub->os_lock);
2472         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2473                 thread_set_flags(thread, SVC_STOPPING);
2474                 spin_unlock(&scrub->os_lock);
2475                 wake_up_all(&thread->t_ctl_waitq);
2476                 l_wait_event(thread->t_ctl_waitq,
2477                              thread_is_stopped(thread),
2478                              &lwi);
2479                 /* Do not skip the last lock/unlock, which can guarantee that
2480                  * the caller cannot return until the OI scrub thread exit. */
2481                 spin_lock(&scrub->os_lock);
2482         }
2483         spin_unlock(&scrub->os_lock);
2484 }
2485
2486 static void osd_scrub_stop(struct osd_device *dev)
2487 {
2488         /* od_otable_mutex: prevent curcurrent start/stop */
2489         mutex_lock(&dev->od_otable_mutex);
2490         dev->od_scrub.os_paused = 1;
2491         do_osd_scrub_stop(&dev->od_scrub);
2492         mutex_unlock(&dev->od_otable_mutex);
2493 }
2494
2495 /* OI scrub setup/cleanup */
2496
2497 static const char osd_scrub_name[] = "OI_scrub";
2498
2499 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2500 {
2501         struct osd_thread_info     *info   = osd_oti_get(env);
2502         struct osd_scrub           *scrub  = &dev->od_scrub;
2503         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
2504         struct scrub_file          *sf     = &scrub->os_file;
2505         struct super_block         *sb     = osd_sb(dev);
2506         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
2507         struct lvfs_run_ctxt        saved;
2508         struct file                *filp;
2509         struct inode               *inode;
2510         struct lu_fid              *fid    = &info->oti_fid;
2511         bool                        dirty  = false;
2512         bool                        restored = false;
2513         int                         rc     = 0;
2514         ENTRY;
2515
2516         memset(scrub, 0, sizeof(*scrub));
2517         OBD_SET_CTXT_MAGIC(ctxt);
2518         ctxt->pwdmnt = dev->od_mnt;
2519         ctxt->pwd = dev->od_mnt->mnt_root;
2520         ctxt->fs = get_ds();
2521
2522         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2523         init_rwsem(&scrub->os_rwsem);
2524         spin_lock_init(&scrub->os_lock);
2525         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2526
2527         push_ctxt(&saved, ctxt);
2528         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2529         if (IS_ERR(filp)) {
2530                 pop_ctxt(&saved, ctxt);
2531                 RETURN(PTR_ERR(filp));
2532         }
2533
2534         inode = file_inode(filp);
2535         /* 'What the @fid is' is not imporatant, because the object
2536          * has no OI mapping, and only is visible inside the OSD.*/
2537         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2538         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2539         if (rc != 0) {
2540                 filp_close(filp, NULL);
2541                 pop_ctxt(&saved, ctxt);
2542                 RETURN(rc);
2543         }
2544
2545         scrub->os_inode = igrab(inode);
2546         filp_close(filp, NULL);
2547         pop_ctxt(&saved, ctxt);
2548
2549         rc = osd_scrub_file_load(scrub);
2550         if (rc == -ENOENT) {
2551                 osd_scrub_file_init(scrub, es->s_uuid);
2552                 /* If the "/O" dir does not exist when mount (indicated by
2553                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2554                  * then it is quite probably that the device is a new one,
2555                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2556                  *
2557                  * For the rare case that "/O" and "OI_scrub" both lost on
2558                  * an old device, it can be found and cleared later.
2559                  *
2560                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2561                  * need to check "filter_fid_old" and to convert it to
2562                  * "filter_fid" for each object, and all the IGIF should
2563                  * have their FID mapping in OI files already. */
2564                 if (dev->od_maybe_new)
2565                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2566                 dirty = true;
2567         } else if (rc != 0) {
2568                 GOTO(cleanup_inode, rc);
2569         } else {
2570                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2571                         struct obd_uuid *old_uuid;
2572                         struct obd_uuid *new_uuid;
2573
2574                         OBD_ALLOC_PTR(old_uuid);
2575                         OBD_ALLOC_PTR(new_uuid);
2576                         if (old_uuid == NULL || new_uuid == NULL) {
2577                                 CERROR("%.16s: UUID has been changed, but"
2578                                        "failed to allocate RAM for report\n",
2579                                        LDISKFS_SB(sb)->s_es->s_volume_name);
2580                         } else {
2581                                 class_uuid_unparse(sf->sf_uuid, old_uuid);
2582                                 class_uuid_unparse(es->s_uuid, new_uuid);
2583                                 CERROR("%.16s: UUID has been changed from "
2584                                        "%s to %s\n",
2585                                        LDISKFS_SB(sb)->s_es->s_volume_name,
2586                                        old_uuid->uuid, new_uuid->uuid);
2587                         }
2588                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2589                         dirty = true;
2590                         restored = true;
2591                         if (old_uuid != NULL)
2592                                 OBD_FREE_PTR(old_uuid);
2593                         if (new_uuid != NULL)
2594                                 OBD_FREE_PTR(new_uuid);
2595                 } else if (sf->sf_status == SS_SCANNING) {
2596                         sf->sf_status = SS_CRASHED;
2597                         dirty = true;
2598                 }
2599         }
2600
2601         if (sf->sf_pos_last_checkpoint != 0)
2602                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2603         else
2604                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2605
2606         if (dirty) {
2607                 rc = osd_scrub_file_store(scrub);
2608                 if (rc != 0)
2609                         GOTO(cleanup_inode, rc);
2610         }
2611
2612         /* Initialize OI files. */
2613         rc = osd_oi_init(info, dev, restored);
2614         if (rc < 0)
2615                 GOTO(cleanup_inode, rc);
2616
2617         rc = osd_initial_OI_scrub(info, dev);
2618         if (rc != 0)
2619                 GOTO(cleanup_oi, rc);
2620
2621         if (sf->sf_flags & SF_UPGRADE ||
2622             !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2623               sf->sf_success_count > 0)) {
2624                 dev->od_igif_inoi = 0;
2625                 dev->od_check_ff = dev->od_is_ost;
2626         } else {
2627                 dev->od_igif_inoi = 1;
2628                 dev->od_check_ff = 0;
2629         }
2630
2631         if (sf->sf_flags & SF_INCONSISTENT)
2632                 /* The 'od_igif_inoi' will be set under the
2633                  * following cases:
2634                  * 1) new created system, or
2635                  * 2) restored from file-level backup, or
2636                  * 3) the upgrading completed.
2637                  *
2638                  * The 'od_igif_inoi' may be cleared by OI scrub
2639                  * later if found that the system is upgrading. */
2640                 dev->od_igif_inoi = 1;
2641
2642         if (!dev->od_noscrub &&
2643             ((sf->sf_status == SS_PAUSED) ||
2644              (sf->sf_status == SS_CRASHED &&
2645               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2646                               SF_UPGRADE | SF_AUTO)) ||
2647              (sf->sf_status == SS_INIT &&
2648               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2649                               SF_UPGRADE))))
2650                 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2651
2652         if (rc != 0)
2653                 GOTO(cleanup_oi, rc);
2654
2655         /* it is possible that dcache entries may keep objects after they are
2656          * deleted by OSD. While it looks safe this can cause object data to
2657          * stay until umount causing failures in tests calculating free space,
2658          * e.g. replay-ost-single. Since those dcache entries are not used
2659          * anymore let's just free them after use here */
2660         shrink_dcache_sb(sb);
2661
2662         RETURN(0);
2663 cleanup_oi:
2664         osd_oi_fini(info, dev);
2665 cleanup_inode:
2666         iput(scrub->os_inode);
2667         scrub->os_inode = NULL;
2668
2669         return rc;
2670 }
2671
2672 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2673 {
2674         struct osd_scrub *scrub = &dev->od_scrub;
2675
2676         LASSERT(dev->od_otable_it == NULL);
2677
2678         if (scrub->os_inode != NULL) {
2679                 osd_scrub_stop(dev);
2680                 iput(scrub->os_inode);
2681                 scrub->os_inode = NULL;
2682         }
2683         if (dev->od_oi_table != NULL)
2684                 osd_oi_fini(osd_oti_get(env), dev);
2685 }
2686
2687 /* object table based iteration APIs */
2688
2689 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2690                                        struct dt_object *dt, __u32 attr)
2691 {
2692         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2693         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2694         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2695         struct osd_scrub       *scrub = &dev->od_scrub;
2696         struct osd_otable_it   *it;
2697         __u32                   start = 0;
2698         int                     rc;
2699         ENTRY;
2700
2701         /* od_otable_mutex: prevent curcurrent init/fini */
2702         mutex_lock(&dev->od_otable_mutex);
2703         if (dev->od_otable_it != NULL)
2704                 GOTO(out, it = ERR_PTR(-EALREADY));
2705
2706         OBD_ALLOC_PTR(it);
2707         if (it == NULL)
2708                 GOTO(out, it = ERR_PTR(-ENOMEM));
2709
2710         dev->od_otable_it = it;
2711         it->ooi_dev = dev;
2712         it->ooi_cache.ooc_consumer_idx = -1;
2713         if (flags & DOIF_OUTUSED)
2714                 it->ooi_used_outside = 1;
2715
2716         if (flags & DOIF_RESET)
2717                 start |= SS_RESET;
2718
2719         if (valid & DOIV_ERROR_HANDLE) {
2720                 if (flags & DOIF_FAILOUT)
2721                         start |= SS_SET_FAILOUT;
2722                 else
2723                         start |= SS_CLEAR_FAILOUT;
2724         }
2725
2726         if (valid & DOIV_DRYRUN) {
2727                 if (flags & DOIF_DRYRUN)
2728                         start |= SS_SET_DRYRUN;
2729                 else
2730                         start |= SS_CLEAR_DRYRUN;
2731         }
2732
2733         rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2734         if (rc < 0 && rc != -EALREADY) {
2735                 dev->od_otable_it = NULL;
2736                 OBD_FREE_PTR(it);
2737                 GOTO(out, it = ERR_PTR(rc));
2738         }
2739
2740         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2741
2742         GOTO(out, it);
2743
2744 out:
2745         mutex_unlock(&dev->od_otable_mutex);
2746         return (struct dt_it *)it;
2747 }
2748
2749 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2750 {
2751         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2752         struct osd_device    *dev = it->ooi_dev;
2753
2754         /* od_otable_mutex: prevent curcurrent init/fini */
2755         mutex_lock(&dev->od_otable_mutex);
2756         do_osd_scrub_stop(&dev->od_scrub);
2757         LASSERT(dev->od_otable_it == it);
2758
2759         dev->od_otable_it = NULL;
2760         mutex_unlock(&dev->od_otable_mutex);
2761         OBD_FREE_PTR(it);
2762 }
2763
2764 static int osd_otable_it_get(const struct lu_env *env,
2765                              struct dt_it *di, const struct dt_key *key)
2766 {
2767         return 0;
2768 }
2769
2770 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2771 {
2772 }
2773
2774 static inline int
2775 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2776 {
2777         spin_lock(&scrub->os_lock);
2778         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2779             scrub->os_waiting ||
2780             !thread_is_running(&scrub->os_thread))
2781                 it->ooi_waiting = 0;
2782         else
2783                 it->ooi_waiting = 1;
2784         spin_unlock(&scrub->os_lock);
2785
2786         return !it->ooi_waiting;
2787 }
2788
2789 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2790 {
2791         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2792         struct osd_device       *dev    = it->ooi_dev;
2793         struct osd_scrub        *scrub  = &dev->od_scrub;
2794         struct osd_otable_cache *ooc    = &it->ooi_cache;
2795         struct ptlrpc_thread    *thread = &scrub->os_thread;
2796         struct l_wait_info       lwi    = { 0 };
2797         int                      rc;
2798         ENTRY;
2799
2800         LASSERT(it->ooi_user_ready);
2801
2802 again:
2803         if (!thread_is_running(thread) && !it->ooi_used_outside)
2804                 RETURN(1);
2805
2806         if (ooc->ooc_cached_items > 0) {
2807                 ooc->ooc_cached_items--;
2808                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2809                                         ~OSD_OTABLE_IT_CACHE_MASK;
2810                 RETURN(0);
2811         }
2812
2813         if (it->ooi_all_cached) {
2814                 l_wait_event(thread->t_ctl_waitq,
2815                              !thread_is_running(thread),
2816                              &lwi);
2817                 RETURN(1);
2818         }
2819
2820         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2821                 spin_lock(&scrub->os_lock);
2822                 scrub->os_waiting = 0;
2823                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2824                 spin_unlock(&scrub->os_lock);
2825         }
2826
2827         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2828                 l_wait_event(thread->t_ctl_waitq,
2829                              osd_otable_it_wakeup(scrub, it),
2830                              &lwi);
2831
2832         if (!thread_is_running(thread) && !it->ooi_used_outside)
2833                 RETURN(1);
2834
2835         rc = osd_otable_it_preload(env, it);
2836         if (rc >= 0)
2837                 goto again;
2838
2839         RETURN(rc);
2840 }
2841
2842 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2843                                         const struct dt_it *di)
2844 {
2845         return NULL;
2846 }
2847
2848 static int osd_otable_it_key_size(const struct lu_env *env,
2849                                   const struct dt_it *di)
2850 {
2851         return sizeof(__u64);
2852 }
2853
2854 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2855                              struct dt_rec *rec, __u32 attr)
2856 {
2857         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2858         struct osd_otable_cache *ooc = &it->ooi_cache;
2859
2860         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2861
2862         /* Filter out Invald FID already. */
2863         LASSERTF(fid_is_sane((struct lu_fid *)rec),
2864                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2865                  PFID((struct lu_fid *)rec),
2866                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2867
2868         return 0;
2869 }
2870
2871 static __u64 osd_otable_it_store(const struct lu_env *env,
2872                                  const struct dt_it *di)
2873 {
2874         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2875         struct osd_otable_cache *ooc = &it->ooi_cache;
2876         __u64                    hash;
2877
2878         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2879                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2880         else
2881                 hash = ooc->ooc_pos_preload;
2882         return hash;
2883 }
2884
2885 /**
2886  * Set the OSD layer iteration start position as the specified hash.
2887  */
2888 static int osd_otable_it_load(const struct lu_env *env,
2889                               const struct dt_it *di, __u64 hash)
2890 {
2891         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
2892         struct osd_device       *dev   = it->ooi_dev;
2893         struct osd_otable_cache *ooc   = &it->ooi_cache;
2894         struct osd_scrub        *scrub = &dev->od_scrub;
2895         int                      rc;
2896         ENTRY;
2897
2898         /* Forbid to set iteration position after iteration started. */
2899         if (it->ooi_user_ready)
2900                 RETURN(-EPERM);
2901
2902         LASSERT(!scrub->os_partial_scan);
2903
2904         if (hash > OSD_OTABLE_MAX_HASH)
2905                 hash = OSD_OTABLE_MAX_HASH;
2906
2907         ooc->ooc_pos_preload = hash;
2908         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2909                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2910
2911         it->ooi_user_ready = 1;
2912         if (!scrub->os_full_speed)
2913                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2914
2915         /* Unplug OSD layer iteration by the first next() call. */
2916         rc = osd_otable_it_next(env, (struct dt_it *)it);
2917
2918         RETURN(rc);
2919 }
2920
2921 static int osd_otable_it_key_rec(const struct lu_env *env,
2922                                  const struct dt_it *di, void *key_rec)
2923 {
2924         return 0;
2925 }
2926
2927 const struct dt_index_operations osd_otable_ops = {
2928         .dio_it = {
2929                 .init     = osd_otable_it_init,
2930                 .fini     = osd_otable_it_fini,
2931                 .get      = osd_otable_it_get,
2932                 .put      = osd_otable_it_put,
2933                 .next     = osd_otable_it_next,
2934                 .key      = osd_otable_it_key,
2935                 .key_size = osd_otable_it_key_size,
2936                 .rec      = osd_otable_it_rec,
2937                 .store    = osd_otable_it_store,
2938                 .load     = osd_otable_it_load,
2939                 .key_rec  = osd_otable_it_key_rec,
2940         }
2941 };
2942
2943 /* high priority inconsistent items list APIs */
2944
2945 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL  60
2946
2947 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2948                    int insert)
2949 {
2950         struct osd_inconsistent_item *oii;
2951         struct osd_scrub             *scrub  = &dev->od_scrub;
2952         struct ptlrpc_thread         *thread = &scrub->os_thread;
2953         int                           wakeup = 0;
2954         ENTRY;
2955
2956         OBD_ALLOC_PTR(oii);
2957         if (unlikely(oii == NULL))
2958                 RETURN(-ENOMEM);
2959
2960         INIT_LIST_HEAD(&oii->oii_list);
2961         oii->oii_cache = *oic;
2962         oii->oii_insert = insert;
2963
2964         if (scrub->os_partial_scan) {
2965                 __u64 now = cfs_time_current_sec();
2966
2967                 /* If there haven't been errors in a long time,
2968                  * decay old count until either the errors are
2969                  * gone or we reach the current interval. */
2970                 while (unlikely(scrub->os_bad_oimap_count > 0 &&
2971                                 scrub->os_bad_oimap_time +
2972                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
2973                         scrub->os_bad_oimap_count >>= 1;
2974                         scrub->os_bad_oimap_time +=
2975                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
2976                 }
2977
2978                 scrub->os_bad_oimap_time = now;
2979                 if (++scrub->os_bad_oimap_count >
2980                     dev->od_full_scrub_threshold_rate)
2981                         scrub->os_full_scrub = 1;
2982         }
2983
2984         spin_lock(&scrub->os_lock);
2985         if (unlikely(!thread_is_running(thread))) {
2986                 spin_unlock(&scrub->os_lock);
2987                 OBD_FREE_PTR(oii);
2988                 RETURN(-EAGAIN);
2989         }
2990
2991         if (list_empty(&scrub->os_inconsistent_items))
2992                 wakeup = 1;
2993         list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2994         spin_unlock(&scrub->os_lock);
2995
2996         if (wakeup != 0)
2997                 wake_up_all(&thread->t_ctl_waitq);
2998
2999         RETURN(0);
3000 }
3001
3002 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
3003                    struct osd_inode_id *id)
3004 {
3005         struct osd_scrub             *scrub = &dev->od_scrub;
3006         struct osd_inconsistent_item *oii;
3007         ENTRY;
3008
3009         spin_lock(&scrub->os_lock);
3010         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
3011                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
3012                         *id = oii->oii_cache.oic_lid;
3013                         spin_unlock(&scrub->os_lock);
3014                         RETURN(0);
3015                 }
3016         }
3017         spin_unlock(&scrub->os_lock);
3018
3019         RETURN(-ENOENT);
3020 }
3021
3022 /* OI scrub dump */
3023
3024 static const char *scrub_status_names[] = {
3025         "init",
3026         "scanning",
3027         "completed",
3028         "failed",
3029         "stopped",
3030         "paused",
3031         "crashed",
3032         NULL
3033 };
3034
3035 static const char *scrub_flags_names[] = {
3036         "recreated",
3037         "inconsistent",
3038         "auto",
3039         "upgrade",
3040         NULL
3041 };
3042
3043 static const char *scrub_param_names[] = {
3044         "failout",
3045         "dryrun",
3046         NULL
3047 };
3048
3049 static void scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
3050                             const char *prefix)
3051 {
3052         int flag;
3053         int i;
3054
3055         seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
3056
3057         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
3058                 if (flag & bits) {
3059                         bits &= ~flag;
3060                         seq_printf(m, "%s%c", names[i],
3061                                    bits != 0 ? ',' : '\n');
3062                 }
3063         }
3064 }
3065
3066 static void scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
3067 {
3068         if (time != 0)
3069                 seq_printf(m, "%s: "LPU64" seconds\n", prefix,
3070                            cfs_time_current_sec() - time);
3071         else
3072                 seq_printf(m, "%s: N/A\n", prefix);
3073 }
3074
3075 static void scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
3076 {
3077         if (pos != 0)
3078                 seq_printf(m, "%s: "LPU64"\n", prefix, pos);
3079         else
3080                 seq_printf(m, "%s: N/A\n", prefix);
3081 }
3082
3083 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
3084 {
3085         struct osd_scrub  *scrub   = &dev->od_scrub;
3086         struct scrub_file *sf      = &scrub->os_file;
3087         __u64              checked;
3088         __u64              speed;
3089
3090         down_read(&scrub->os_rwsem);
3091         seq_printf(m, "name: OI_scrub\n"
3092                    "magic: 0x%x\n"
3093                    "oi_files: %d\n"
3094                    "status: %s\n",
3095                    sf->sf_magic, (int)sf->sf_oi_count,
3096                    scrub_status_names[sf->sf_status]);
3097
3098         scrub_bits_dump(m, sf->sf_flags, scrub_flags_names, "flags");
3099
3100         scrub_bits_dump(m, sf->sf_param, scrub_param_names, "param");
3101
3102         scrub_time_dump(m, sf->sf_time_last_complete,
3103                         "time_since_last_completed");
3104
3105         scrub_time_dump(m, sf->sf_time_latest_start,
3106                         "time_since_latest_start");
3107
3108         scrub_time_dump(m, sf->sf_time_last_checkpoint,
3109                         "time_since_last_checkpoint");
3110
3111         scrub_pos_dump(m, sf->sf_pos_latest_start,
3112                         "latest_start_position");
3113
3114         scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
3115                         "last_checkpoint_position");
3116
3117         scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
3118                         "first_failure_position");
3119
3120         checked = sf->sf_items_checked + scrub->os_new_checked;
3121         seq_printf(m, "checked: "LPU64"\n"
3122                    "updated: "LPU64"\n"
3123                    "failed: "LPU64"\n"
3124                    "prior_updated: "LPU64"\n"
3125                    "noscrub: "LPU64"\n"
3126                    "igif: "LPU64"\n"
3127                    "success_count: %u\n",
3128                    checked, sf->sf_items_updated, sf->sf_items_failed,
3129                    sf->sf_items_updated_prior, sf->sf_items_noscrub,
3130                    sf->sf_items_igif, sf->sf_success_count);
3131
3132         speed = checked;
3133         if (thread_is_running(&scrub->os_thread)) {
3134                 cfs_duration_t duration = cfs_time_current() -
3135                                           scrub->os_time_last_checkpoint;
3136                 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
3137                                                      MSEC_PER_SEC);
3138                 __u32 rtime = sf->sf_run_time +
3139                               cfs_duration_sec(duration + HALF_SEC);
3140
3141                 if (duration != 0)
3142                         do_div(new_checked, duration);
3143                 if (rtime != 0)
3144                         do_div(speed, rtime);
3145                 seq_printf(m, "run_time: %u seconds\n"
3146                            "average_speed: "LPU64" objects/sec\n"
3147                            "real-time_speed: "LPU64" objects/sec\n"
3148                            "current_position: %u\n"
3149                            "lf_scanned: "LPU64"\n"
3150                            "lf_repaired: "LPU64"\n"
3151                            "lf_failed: "LPU64"\n",
3152                            rtime, speed, new_checked, scrub->os_pos_current,
3153                            scrub->os_lf_scanned, scrub->os_lf_repaired,
3154                            scrub->os_lf_failed);
3155         } else {
3156                 if (sf->sf_run_time != 0)
3157                         do_div(speed, sf->sf_run_time);
3158                 seq_printf(m, "run_time: %u seconds\n"
3159                            "average_speed: "LPU64" objects/sec\n"
3160                            "real-time_speed: N/A\n"
3161                            "current_position: N/A\n"
3162                            "lf_scanned: "LPU64"\n"
3163                            "lf_repaired: "LPU64"\n"
3164                            "lf_failed: "LPU64"\n",
3165                            sf->sf_run_time, speed, scrub->os_lf_scanned,
3166                            scrub->os_lf_repaired, scrub->os_lf_failed);
3167         }
3168
3169         up_read(&scrub->os_rwsem);
3170         return 0;
3171 }