lustre/osd-ldiskfs/osd_scrub.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2012, 2013, Intel Corporation.
  24  */
  25 /*
  26  * lustre/osd-ldiskfs/osd_scrub.c
  27  *
  28  * Top-level entry points into osd module
  29  *
  30  * The OI scrub is used for rebuilding Object Index files when restores MDT from
  31  * file-level backup.
  32  *
  33  * The otable based iterator scans ldiskfs inode table to feed up layer LFSCK.
  34  *
  35  * Author: Fan Yong <yong.fan@whamcloud.com>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_MDS
  39
  40 #include <lustre/lustre_idl.h>
  41 #include <lustre_disk.h>
  42 #include <dt_object.h>
  43 #include <linux/xattr.h>
  44
  45 #include "osd_internal.h"
  46 #include "osd_oi.h"
  47 #include "osd_scrub.h"
  48
  49 #define HALF_SEC        (HZ >> 1)
  50
  51 #define OSD_OTABLE_MAX_HASH             0x00000000ffffffffULL
  52
  53 #define SCRUB_NEXT_BREAK        1 /* exit current loop and process next group */
  54 #define SCRUB_NEXT_CONTINUE     2 /* skip current object and process next bit */
  55 #define SCRUB_NEXT_EXIT         3 /* exit all the loops */
  56 #define SCRUB_NEXT_WAIT         4 /* wait for free cache slot */
  57 #define SCRUB_NEXT_CRASH        5 /* simulate system crash during OI scrub */
  58 #define SCRUB_NEXT_FATAL        6 /* simulate failure during OI scrub */
  59 #define SCRUB_NEXT_NOSCRUB      7 /* new created object, no scrub on it */
  60 #define SCRUB_NEXT_NOLMA        8 /* the inode has no FID-in-LMA */
  61 #define SCRUB_NEXT_OSTOBJ       9 /* for OST-object */
  62 #define SCRUB_NEXT_OSTOBJ_OLD   10 /* old OST-object, no LMA or no FID-on-OST
  63                                     * flags in LMA */
  64
  65 /* misc functions */
  66
  67 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
  68 {
  69         return container_of0(scrub, struct osd_device, od_scrub);
  70 }
  71
  72 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
  73 {
  74         return osd_sb(osd_scrub2dev(scrub));
  75 }
  76
  77 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
  78                                        struct osd_otable_cache *ooc)
  79 {
  80         return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
  81 }
  82
  83 /**
  84  * update/insert/delete the specified OI mapping (@fid @id) according to the ops
  85  *
  86  * \retval   1, changed nothing
  87  * \retval   0, changed successfully
  88  * \retval -ve, on error
  89  */
  90 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
  91                                      struct osd_device *dev,
  92                                      const struct lu_fid *fid,
  93                                      const struct osd_inode_id *id,
  94                                      int ops, bool force,
  95                                      enum oi_check_flags flags)
  96 {
  97         handle_t *th;
  98         int       rc;
  99         ENTRY;
 100
 101         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
 102                 RETURN(0);
 103
 104         /* DTO_INDEX_INSERT is enough for other two ops:
 105          * delete/update, but save stack. */
 106         th = ldiskfs_journal_start_sb(osd_sb(dev),
 107                                 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
 108         if (IS_ERR(th)) {
 109                 rc = PTR_ERR(th);
 110                 CERROR("%s: fail to start trans for scrub %d: rc = %d\n",
 111                        osd_name(dev), ops, rc);
 112                 RETURN(rc);
 113         }
 114
 115         switch (ops) {
 116         case DTO_INDEX_UPDATE:
 117                 rc = osd_oi_update(info, dev, fid, id, th, flags);
 118                 if (unlikely(rc == -ENOENT)) {
 119                         /* Some unlink thread may removed the OI mapping. */
 120                         rc = 1;
 121                 }
 122                 break;
 123         case DTO_INDEX_INSERT:
 124                 rc = osd_oi_insert(info, dev, fid, id, th, flags);
 125                 if (unlikely(rc == -EEXIST)) {
 126                         rc = 1;
 127                         /* XXX: There are trouble things when adding OI
 128                          *      mapping for IGIF object, which may cause
 129                          *      multiple objects to be mapped to the same
 130                          *      IGIF formatted FID. Consider the following
 131                          *      situations:
 132                          *
 133                          *      1) The MDT is upgrading from 1.8 device.
 134                          *      The OI scrub generates IGIF FID1 for the
 135                          *      OBJ1 and adds the OI mapping.
 136                          *
 137                          *      2) For some reason, the OI scrub does not
 138                          *      process all the IGIF objects completely.
 139                          *
 140                          *      3) The MDT is backuped and restored against
 141                          *      this device.
 142                          *
 143                          *      4) When the MDT mounts up, the OI scrub will
 144                          *      try to rebuild the OI files. For some IGIF
 145                          *      object, OBJ2, which was not processed by the
 146                          *      OI scrub before the backup/restore, and the
 147                          *      new generated IGIF formatted FID may be just
 148                          *      the FID1, the same as OBJ1.
 149                          *
 150                          *      Under such case, the OI scrub cannot know how
 151                          *      to generate new FID for the OBJ2.
 152                          *
 153                          *      Currently, we do nothing for that. One possible
 154                          *      solution is to generate new normal FID for the
 155                          *      conflict object.
 156                          *
 157                          *      Anyway, it is rare, only exists in theory. */
 158                 }
 159                 break;
 160         case DTO_INDEX_DELETE:
 161                 rc = osd_oi_delete(info, dev, fid, th, flags);
 162                 if (rc == -ENOENT) {
 163                         /* It is normal that the unlink thread has removed the
 164                          * OI mapping already. */
 165                         rc = 1;
 166                 }
 167                 break;
 168         default:
 169                 LASSERTF(0, "Unexpected ops %d\n", ops);
 170                 break;
 171         }
 172
 173         ldiskfs_journal_stop(th);
 174         return rc;
 175 }
 176
 177 /* OI_scrub file ops */
 178
 179 static void osd_scrub_file_to_cpu(struct scrub_file *des,
 180                                   struct scrub_file *src)
 181 {
 182         memcpy(des->sf_uuid, src->sf_uuid, 16);
 183         des->sf_flags   = le64_to_cpu(src->sf_flags);
 184         des->sf_magic   = le32_to_cpu(src->sf_magic);
 185         des->sf_status  = le16_to_cpu(src->sf_status);
 186         des->sf_param   = le16_to_cpu(src->sf_param);
 187         des->sf_time_last_complete      =
 188                                 le64_to_cpu(src->sf_time_last_complete);
 189         des->sf_time_latest_start       =
 190                                 le64_to_cpu(src->sf_time_latest_start);
 191         des->sf_time_last_checkpoint    =
 192                                 le64_to_cpu(src->sf_time_last_checkpoint);
 193         des->sf_pos_latest_start        =
 194                                 le64_to_cpu(src->sf_pos_latest_start);
 195         des->sf_pos_last_checkpoint     =
 196                                 le64_to_cpu(src->sf_pos_last_checkpoint);
 197         des->sf_pos_first_inconsistent  =
 198                                 le64_to_cpu(src->sf_pos_first_inconsistent);
 199         des->sf_items_checked           =
 200                                 le64_to_cpu(src->sf_items_checked);
 201         des->sf_items_updated           =
 202                                 le64_to_cpu(src->sf_items_updated);
 203         des->sf_items_failed            =
 204                                 le64_to_cpu(src->sf_items_failed);
 205         des->sf_items_updated_prior     =
 206                                 le64_to_cpu(src->sf_items_updated_prior);
 207         des->sf_run_time        = le32_to_cpu(src->sf_run_time);
 208         des->sf_success_count   = le32_to_cpu(src->sf_success_count);
 209         des->sf_oi_count        = le16_to_cpu(src->sf_oi_count);
 210         des->sf_internal_flags  = le16_to_cpu(src->sf_internal_flags);
 211         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 212 }
 213
 214 static void osd_scrub_file_to_le(struct scrub_file *des,
 215                                  struct scrub_file *src)
 216 {
 217         memcpy(des->sf_uuid, src->sf_uuid, 16);
 218         des->sf_flags   = cpu_to_le64(src->sf_flags);
 219         des->sf_magic   = cpu_to_le32(src->sf_magic);
 220         des->sf_status  = cpu_to_le16(src->sf_status);
 221         des->sf_param   = cpu_to_le16(src->sf_param);
 222         des->sf_time_last_complete      =
 223                                 cpu_to_le64(src->sf_time_last_complete);
 224         des->sf_time_latest_start       =
 225                                 cpu_to_le64(src->sf_time_latest_start);
 226         des->sf_time_last_checkpoint    =
 227                                 cpu_to_le64(src->sf_time_last_checkpoint);
 228         des->sf_pos_latest_start        =
 229                                 cpu_to_le64(src->sf_pos_latest_start);
 230         des->sf_pos_last_checkpoint     =
 231                                 cpu_to_le64(src->sf_pos_last_checkpoint);
 232         des->sf_pos_first_inconsistent  =
 233                                 cpu_to_le64(src->sf_pos_first_inconsistent);
 234         des->sf_items_checked           =
 235                                 cpu_to_le64(src->sf_items_checked);
 236         des->sf_items_updated           =
 237                                 cpu_to_le64(src->sf_items_updated);
 238         des->sf_items_failed            =
 239                                 cpu_to_le64(src->sf_items_failed);
 240         des->sf_items_updated_prior     =
 241                                 cpu_to_le64(src->sf_items_updated_prior);
 242         des->sf_run_time        = cpu_to_le32(src->sf_run_time);
 243         des->sf_success_count   = cpu_to_le32(src->sf_success_count);
 244         des->sf_oi_count        = cpu_to_le16(src->sf_oi_count);
 245         des->sf_internal_flags  = cpu_to_le16(src->sf_internal_flags);
 246         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 247 }
 248
 249 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
 250 {
 251         struct scrub_file *sf = &scrub->os_file;
 252
 253         memset(sf, 0, sizeof(*sf));
 254         memcpy(sf->sf_uuid, uuid, 16);
 255         sf->sf_magic = SCRUB_MAGIC_V1;
 256         sf->sf_status = SS_INIT;
 257 }
 258
 259 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
 260 {
 261         struct scrub_file *sf = &scrub->os_file;
 262
 263         CDEBUG(D_LFSCK, "Reset OI scrub file, flags = "LPX64"\n", flags);
 264         memcpy(sf->sf_uuid, uuid, 16);
 265         sf->sf_status = SS_INIT;
 266         sf->sf_flags |= flags;
 267         sf->sf_run_time = 0;
 268         sf->sf_time_latest_start = 0;
 269         sf->sf_time_last_checkpoint = 0;
 270         sf->sf_pos_latest_start = 0;
 271         sf->sf_pos_last_checkpoint = 0;
 272         sf->sf_pos_first_inconsistent = 0;
 273         sf->sf_items_checked = 0;
 274         sf->sf_items_updated = 0;
 275         sf->sf_items_failed = 0;
 276         sf->sf_items_updated_prior = 0;
 277         sf->sf_items_noscrub = 0;
 278         sf->sf_items_igif = 0;
 279 }
 280
 281 static int osd_scrub_file_load(struct osd_scrub *scrub)
 282 {
 283         loff_t  pos  = 0;
 284         char   *name = LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
 285         int     len  = sizeof(scrub->os_file_disk);
 286         int     rc;
 287
 288         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
 289         if (rc == len) {
 290                 struct scrub_file *sf = &scrub->os_file;
 291
 292                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
 293                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
 294                         CWARN("%.16s: invalid scrub magic 0x%x != 0x%x\n,",
 295                               name, sf->sf_magic, SCRUB_MAGIC_V1);
 296                         /* Process it as new scrub file. */
 297                         rc = -ENOENT;
 298                 } else {
 299                         rc = 0;
 300                 }
 301         } else if (rc != 0) {
 302                 CERROR("%.16s: fail to load scrub file, expected = %d, "
 303                        "rc = %d\n", name, len, rc);
 304                 if (rc > 0)
 305                         rc = -EFAULT;
 306         } else {
 307                 /* return -ENOENT for empty scrub file case. */
 308                 rc = -ENOENT;
 309         }
 310
 311         return rc;
 312 }
 313
 314 int osd_scrub_file_store(struct osd_scrub *scrub)
 315 {
 316         struct osd_device *dev;
 317         handle_t          *jh;
 318         loff_t             pos     = 0;
 319         int                len     = sizeof(scrub->os_file_disk);
 320         int                credits;
 321         int                rc;
 322
 323         dev = container_of0(scrub, struct osd_device, od_scrub);
 324         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
 325                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
 326         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits);
 327         if (IS_ERR(jh)) {
 328                 rc = PTR_ERR(jh);
 329                 CERROR("%.16s: fail to start trans for scrub store, rc = %d\n",
 330                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,rc);
 331                 return rc;
 332         }
 333
 334         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
 335         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
 336                                       len, 0, &pos, jh);
 337         ldiskfs_journal_stop(jh);
 338         if (rc != 0)
 339                 CERROR("%.16s: fail to store scrub file, expected = %d, "
 340                        "rc = %d\n",
 341                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
 342                        len, rc);
 343         scrub->os_time_last_checkpoint = cfs_time_current();
 344         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
 345                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
 346         return rc;
 347 }
 348
 349 static int
 350 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
 351                      struct inode *inode, const struct lu_fid *fid)
 352 {
 353         struct filter_fid_old   *ff      = &info->oti_ff;
 354         struct dentry           *dentry  = &info->oti_obj_dentry;
 355         handle_t                *jh;
 356         int                      size    = 0;
 357         int                      rc;
 358         bool                     removed = false;
 359         bool                     reset   = true;
 360         ENTRY;
 361
 362         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
 363                 RETURN(0);
 364
 365         /* We want the LMA to fit into the 256-byte OST inode, so operate
 366          * as following:
 367          * 1) read old XATTR_NAME_FID and save the parent FID;
 368          * 2) delete the old XATTR_NAME_FID;
 369          * 3) make new LMA and add it;
 370          * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
 371          *
 372          * Making the LMA to fit into the 256-byte OST inode can save time for
 373          * normal osd_check_lma() and for other OI scrub scanning in future.
 374          * So it is worth to make some slow conversion here. */
 375         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 376                                 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
 377         if (IS_ERR(jh)) {
 378                 rc = PTR_ERR(jh);
 379                 CERROR("%s: fail to start trans for convert ff: "DFID
 380                        ": rc = %d\n",
 381                        osd_name(dev), PFID(fid), rc);
 382                 RETURN(rc);
 383         }
 384
 385         /* 1) read old XATTR_NAME_FID and save the parent FID */
 386         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 387         if (rc == sizeof(*ff)) {
 388                 /* 2) delete the old XATTR_NAME_FID */
 389                 ll_vfs_dq_init(inode);
 390                 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
 391                 if (rc != 0)
 392                         GOTO(stop, rc);
 393
 394                 removed = true;
 395         } else if (unlikely(rc == -ENODATA)) {
 396                 reset = false;
 397         } else if (rc != sizeof(struct filter_fid)) {
 398                 GOTO(stop, rc = -EINVAL);
 399         }
 400
 401         /* 3) make new LMA and add it */
 402         rc = osd_ea_fid_set(info, inode, fid, LMAC_FID_ON_OST, 0);
 403         if (rc == 0 && reset)
 404                 size = sizeof(struct filter_fid);
 405         else if (rc != 0 && removed)
 406                 /* If failed, we should try to add the old back. */
 407                 size = sizeof(struct filter_fid_old);
 408
 409         /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
 410         if (size > 0) {
 411                 int rc1;
 412
 413                 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
 414                                       XATTR_CREATE);
 415                 if (rc1 != 0 && rc == 0)
 416                         rc = rc1;
 417         }
 418
 419         GOTO(stop, rc);
 420
 421 stop:
 422         ldiskfs_journal_stop(jh);
 423         return rc;
 424 }
 425
 426 static int
 427 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
 428                        struct osd_idmap_cache *oic, int val)
 429 {
 430         struct osd_scrub             *scrub  = &dev->od_scrub;
 431         struct scrub_file            *sf     = &scrub->os_file;
 432         struct lu_fid                *fid    = &oic->oic_fid;
 433         struct osd_inode_id          *lid    = &oic->oic_lid;
 434         struct osd_inode_id          *lid2   = &info->oti_id;
 435         struct osd_inconsistent_item *oii    = NULL;
 436         struct inode                 *inode  = NULL;
 437         int                           ops    = DTO_INDEX_UPDATE;
 438         int                           idx;
 439         int                           rc;
 440         bool                          converted = false;
 441         ENTRY;
 442
 443         down_write(&scrub->os_rwsem);
 444         scrub->os_new_checked++;
 445         if (val < 0)
 446                 GOTO(out, rc = val);
 447
 448         if (scrub->os_in_prior)
 449                 oii = cfs_list_entry(oic, struct osd_inconsistent_item,
 450                                      oii_cache);
 451
 452         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
 453                 GOTO(out, rc = 0);
 454
 455         if (fid_is_igif(fid))
 456                 sf->sf_items_igif++;
 457
 458         if (val == SCRUB_NEXT_OSTOBJ_OLD) {
 459                 inode = osd_iget(info, dev, lid);
 460                 if (IS_ERR(inode)) {
 461                         rc = PTR_ERR(inode);
 462                         /* Someone removed the inode. */
 463                         if (rc == -ENOENT || rc == -ESTALE)
 464                                 rc = 0;
 465                         GOTO(out, rc);
 466                 }
 467
 468                 sf->sf_flags |= SF_UPGRADE;
 469                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
 470                 dev->od_check_ff = 1;
 471                 rc = osd_scrub_convert_ff(info, dev, inode, fid);
 472                 if (rc != 0)
 473                         GOTO(out, rc);
 474
 475                 converted = true;
 476         }
 477
 478         if ((val == SCRUB_NEXT_NOLMA) &&
 479             (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
 480                 GOTO(out, rc = 0);
 481
 482         if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA))
 483                 goto iget;
 484
 485         rc = osd_oi_lookup(info, dev, fid, lid2,
 486                 (val == SCRUB_NEXT_OSTOBJ ||
 487                  val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 488         if (rc != 0) {
 489                 if (rc != -ENOENT)
 490                         GOTO(out, rc);
 491
 492 iget:
 493                 if (inode == NULL) {
 494                         inode = osd_iget(info, dev, lid);
 495                         if (IS_ERR(inode)) {
 496                                 rc = PTR_ERR(inode);
 497                                 /* Someone removed the inode. */
 498                                 if (rc == -ENOENT || rc == -ESTALE)
 499                                         rc = 0;
 500                                 GOTO(out, rc);
 501                         }
 502                 }
 503
 504                 scrub->os_full_speed = 1;
 505                 ops = DTO_INDEX_INSERT;
 506                 idx = osd_oi_fid2idx(dev, fid);
 507                 switch (val) {
 508                 case SCRUB_NEXT_NOLMA:
 509                         sf->sf_flags |= SF_UPGRADE;
 510                         if (!(sf->sf_param & SP_DRYRUN)) {
 511                                 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
 512                                 if (rc != 0)
 513                                         GOTO(out, rc);
 514                         }
 515
 516                         if (!(sf->sf_flags & SF_INCONSISTENT))
 517                                 dev->od_igif_inoi = 0;
 518                         break;
 519                 case SCRUB_NEXT_OSTOBJ:
 520                         sf->sf_flags |= SF_INCONSISTENT;
 521                 case SCRUB_NEXT_OSTOBJ_OLD:
 522                         break;
 523                 default:
 524                         sf->sf_flags |= SF_RECREATED;
 525                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 526                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 527                         break;
 528                 }
 529         } else if (osd_id_eq(lid, lid2)) {
 530                 if (converted)
 531                         sf->sf_items_updated++;
 532
 533                 GOTO(out, rc = 0);
 534         } else {
 535                 scrub->os_full_speed = 1;
 536                 sf->sf_flags |= SF_INCONSISTENT;
 537
 538                 /* XXX: If the device is restored from file-level backup, then
 539                  *      some IGIFs may have been already in OI files, and some
 540                  *      may be not yet. Means upgrading from 1.8 may be partly
 541                  *      processed, but some clients may hold some immobilized
 542                  *      IGIFs, and use them to access related objects. Under
 543                  *      such case, OSD does not know whether an given IGIF has
 544                  *      been processed or to be processed, and it also cannot
 545                  *      generate local ino#/gen# directly from the immobilized
 546                  *      IGIF because of the backup/restore. Then force OSD to
 547                  *      lookup the given IGIF in OI files, and if no entry,
 548                  *      then ask the client to retry after upgrading completed.
 549                  *      No better choice. */
 550                 dev->od_igif_inoi = 1;
 551         }
 552
 553         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
 554                         (val == SCRUB_NEXT_OSTOBJ ||
 555                          val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 556         if (rc == 0) {
 557                 if (scrub->os_in_prior)
 558                         sf->sf_items_updated_prior++;
 559                 else
 560                         sf->sf_items_updated++;
 561
 562                 /* The target has been changed, need to be re-loaded. */
 563                 lu_object_purge(info->oti_env, osd2lu_dev(dev), fid);
 564         }
 565
 566         GOTO(out, rc);
 567
 568 out:
 569         if (rc < 0) {
 570                 sf->sf_items_failed++;
 571                 if (sf->sf_pos_first_inconsistent == 0 ||
 572                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 573                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 574         } else {
 575                 rc = 0;
 576         }
 577
 578         /* There may be conflict unlink during the OI scrub,
 579          * if happend, then remove the new added OI mapping. */
 580         if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
 581             unlikely(inode->i_nlink == 0))
 582                 osd_scrub_refresh_mapping(info, dev, fid, lid,
 583                                 DTO_INDEX_DELETE, false,
 584                                 (val == SCRUB_NEXT_OSTOBJ ||
 585                                  val == SCRUB_NEXT_OSTOBJ_OLD) ?
 586                                 OI_KNOWN_ON_OST : 0);
 587         up_write(&scrub->os_rwsem);
 588
 589         if (inode != NULL && !IS_ERR(inode))
 590                 iput(inode);
 591
 592         if (oii != NULL) {
 593                 LASSERT(!cfs_list_empty(&oii->oii_list));
 594
 595                 spin_lock(&scrub->os_lock);
 596                 cfs_list_del_init(&oii->oii_list);
 597                 spin_unlock(&scrub->os_lock);
 598                 OBD_FREE_PTR(oii);
 599         }
 600         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 601 }
 602
 603 /* OI scrub APIs */
 604
 605 static int osd_scrub_prep(struct osd_device *dev)
 606 {
 607         struct osd_scrub     *scrub  = &dev->od_scrub;
 608         struct ptlrpc_thread *thread = &scrub->os_thread;
 609         struct scrub_file    *sf     = &scrub->os_file;
 610         __u32                 flags  = scrub->os_start_flags;
 611         int                   rc;
 612         bool                  drop_dryrun = false;
 613         ENTRY;
 614
 615         down_write(&scrub->os_rwsem);
 616         if (flags & SS_SET_FAILOUT)
 617                 sf->sf_param |= SP_FAILOUT;
 618
 619         if (flags & SS_CLEAR_FAILOUT)
 620                 sf->sf_param &= ~SP_FAILOUT;
 621
 622         if (flags & SS_SET_DRYRUN)
 623                 sf->sf_param |= SP_DRYRUN;
 624
 625         if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
 626                 sf->sf_param &= ~SP_DRYRUN;
 627                 drop_dryrun = true;
 628         }
 629
 630         if (flags & SS_RESET)
 631                 osd_scrub_file_reset(scrub,
 632                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
 633
 634         if (flags & SS_AUTO) {
 635                 scrub->os_full_speed = 1;
 636                 sf->sf_flags |= SF_AUTO;
 637                 /* For the case of OI scrub auto triggered, NOT dryrun. */
 638                 sf->sf_param &= ~SP_FAILOUT;
 639         } else {
 640                 scrub->os_full_speed = 0;
 641         }
 642
 643         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
 644                 scrub->os_full_speed = 1;
 645
 646         scrub->os_in_prior = 0;
 647         spin_lock(&scrub->os_lock);
 648         scrub->os_waiting = 0;
 649         scrub->os_paused = 0;
 650         spin_unlock(&scrub->os_lock);
 651         scrub->os_new_checked = 0;
 652         if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
 653                 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
 654         else if (sf->sf_pos_last_checkpoint != 0)
 655                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 656         else
 657                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 658
 659         scrub->os_pos_current = sf->sf_pos_latest_start;
 660         sf->sf_status = SS_SCANNING;
 661         sf->sf_time_latest_start = cfs_time_current_sec();
 662         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 663         rc = osd_scrub_file_store(scrub);
 664         if (rc == 0) {
 665                 spin_lock(&scrub->os_lock);
 666                 thread_set_flags(thread, SVC_RUNNING);
 667                 spin_unlock(&scrub->os_lock);
 668                 wake_up_all(&thread->t_ctl_waitq);
 669         }
 670         up_write(&scrub->os_rwsem);
 671
 672         RETURN(rc);
 673 }
 674
 675 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
 676 {
 677         struct scrub_file *sf = &scrub->os_file;
 678         int                rc;
 679
 680         if (likely(cfs_time_before(cfs_time_current(),
 681                                    scrub->os_time_next_checkpoint) ||
 682                    scrub->os_new_checked == 0))
 683                 return 0;
 684
 685         down_write(&scrub->os_rwsem);
 686         sf->sf_items_checked += scrub->os_new_checked;
 687         scrub->os_new_checked = 0;
 688         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 689         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 690         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 691                                             scrub->os_time_last_checkpoint);
 692         rc = osd_scrub_file_store(scrub);
 693         up_write(&scrub->os_rwsem);
 694
 695         return rc;
 696 }
 697
 698 static void osd_scrub_post(struct osd_scrub *scrub, int result)
 699 {
 700         struct scrub_file *sf = &scrub->os_file;
 701         ENTRY;
 702
 703         down_write(&scrub->os_rwsem);
 704         spin_lock(&scrub->os_lock);
 705         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 706         spin_unlock(&scrub->os_lock);
 707         if (scrub->os_new_checked > 0) {
 708                 sf->sf_items_checked += scrub->os_new_checked;
 709                 scrub->os_new_checked = 0;
 710                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 711         }
 712         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 713         if (result > 0) {
 714                 struct osd_device *dev =
 715                         container_of0(scrub, struct osd_device, od_scrub);
 716
 717                 dev->od_igif_inoi = 1;
 718                 dev->od_check_ff = 0;
 719                 sf->sf_status = SS_COMPLETED;
 720                 if (!(sf->sf_param & SP_DRYRUN)) {
 721                         memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 722                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 723                                           SF_UPGRADE | SF_AUTO);
 724                 }
 725                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 726                 sf->sf_success_count++;
 727         } else if (result == 0) {
 728                 if (scrub->os_paused)
 729                         sf->sf_status = SS_PAUSED;
 730                 else
 731                         sf->sf_status = SS_STOPPED;
 732         } else {
 733                 sf->sf_status = SS_FAILED;
 734         }
 735         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 736                                             scrub->os_time_last_checkpoint);
 737         result = osd_scrub_file_store(scrub);
 738         if (result < 0)
 739                 CERROR("%.16s: fail to osd_scrub_post, rc = %d\n",
 740                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
 741                        result);
 742         up_write(&scrub->os_rwsem);
 743
 744         EXIT;
 745 }
 746
 747 /* iteration engine */
 748
 749 struct osd_iit_param {
 750         struct super_block *sb;
 751         struct buffer_head *bitmap;
 752         ldiskfs_group_t bg;
 753         __u32 gbase;
 754         __u32 offset;
 755 };
 756
 757 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 758                                    struct osd_device *dev,
 759                                    struct osd_iit_param *param,
 760                                    struct osd_idmap_cache **oic,
 761                                    int noslot);
 762
 763 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 764                                    struct osd_device *dev,
 765                                    struct osd_iit_param *param,
 766                                    struct osd_idmap_cache *oic,
 767                                    int *noslot, int rc);
 768
 769 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 770 {
 771         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 772                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 773         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 774                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 775                 return SCRUB_NEXT_BREAK;
 776         } else {
 777                 *pos = param->gbase + param->offset;
 778                 return 0;
 779         }
 780 }
 781
 782 /**
 783  * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
 784  * \retval 0: FID-on-MDT
 785  */
 786 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
 787                                       struct osd_device *dev,
 788                                       struct lu_fid *fid)
 789 {
 790         /* XXX: The initial OI scrub will scan the top level /O to generate
 791          *      a small local FLDB according to the <seq>. If the given FID
 792          *      is in the local FLDB, then it is FID-on-OST; otherwise it's
 793          *      quite possible for FID-on-MDT. */
 794         if (dev->od_is_ost)
 795                 return SCRUB_NEXT_OSTOBJ_OLD;
 796         else
 797                 return 0;
 798 }
 799
 800 static int osd_scrub_get_fid(struct osd_thread_info *info,
 801                              struct osd_device *dev, struct inode *inode,
 802                              struct lu_fid *fid, bool scrub)
 803 {
 804         struct lustre_mdt_attrs *lma     = &info->oti_mdt_attrs;
 805         int                      rc;
 806         bool                     has_lma = false;
 807
 808         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 809         if (rc == 0) {
 810                 has_lma = true;
 811                 if (lma->lma_compat & LMAC_NOT_IN_OI ||
 812                     lma->lma_incompat & LMAI_AGENT)
 813                         return SCRUB_NEXT_CONTINUE;
 814
 815                 *fid = lma->lma_self_fid;
 816                 if (!scrub)
 817                         return 0;
 818
 819                 if (lma->lma_compat & LMAC_FID_ON_OST)
 820                         return SCRUB_NEXT_OSTOBJ;
 821
 822                 if (fid_is_idif(fid))
 823                         return SCRUB_NEXT_OSTOBJ_OLD;
 824
 825                 /* For local object. */
 826                 if (fid_is_internal(fid))
 827                         return 0;
 828
 829                 /* For external visible MDT-object with non-normal FID. */
 830                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
 831                         return 0;
 832
 833                 /* For the object with normal FID, it may be MDT-object,
 834                  * or may be 2.4 OST-object, need further distinguish.
 835                  * Fall through to next section. */
 836         }
 837
 838         if (rc == -ENODATA || rc == 0) {
 839                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
 840                 if (rc == 0) {
 841                         if (scrub)
 842                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
 843                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 844                         return rc;
 845                 }
 846
 847                 if (rc > 0) {
 848                         if (!has_lma)
 849                                 /* It is FID-on-OST, but we do not know how
 850                                  * to generate its FID, ignore it directly. */
 851                                 rc = SCRUB_NEXT_CONTINUE;
 852                         else
 853                                 /* It is 2.4 OST-object. */
 854                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 855                         return rc;
 856                 }
 857
 858                 if (rc != -ENODATA)
 859                         return rc;
 860
 861                 if (!has_lma) {
 862                         if (dev->od_scrub.os_convert_igif) {
 863                                 lu_igif_build(fid, inode->i_ino,
 864                                               inode->i_generation);
 865                                 if (scrub)
 866                                         rc = SCRUB_NEXT_NOLMA;
 867                                 else
 868                                         rc = 0;
 869                         } else {
 870                                 /* It may be FID-on-OST, or may be FID for
 871                                  * non-MDT0, anyway, we do not know how to
 872                                  * generate its FID, ignore it directly. */
 873                                 rc = SCRUB_NEXT_CONTINUE;
 874                         }
 875                         return rc;
 876                 }
 877
 878                 /* For OI scrub case only: the object has LMA but has no ff
 879                  * (or ff crashed). It may be MDT-object, may be OST-object
 880                  * with crashed ff. The last check is local FLDB. */
 881                 rc = osd_scrub_check_local_fldb(info, dev, fid);
 882         }
 883
 884         return rc;
 885 }
 886
 887 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 888                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 889                         struct super_block *sb, bool scrub)
 890 {
 891         struct inode *inode;
 892         int           rc;
 893         ENTRY;
 894
 895         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 896         inode = osd_iget(info, dev, lid);
 897         if (IS_ERR(inode)) {
 898                 rc = PTR_ERR(inode);
 899                 /* The inode may be removed after bitmap searching, or the
 900                  * file is new created without inode initialized yet. */
 901                 if (rc == -ENOENT || rc == -ESTALE)
 902                         RETURN(SCRUB_NEXT_CONTINUE);
 903
 904                 CERROR("%.16s: fail to read inode, ino# = %u, rc = %d\n",
 905                        LDISKFS_SB(sb)->s_es->s_volume_name, pos, rc);
 906                 RETURN(rc);
 907         }
 908
 909         if (scrub &&
 910             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 911                 /* Only skip it for the first OI scrub accessing. */
 912                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 913                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
 914         }
 915
 916         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
 917
 918         GOTO(put, rc);
 919
 920 put:
 921         iput(inode);
 922         return rc;
 923 }
 924
 925 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 926                           struct osd_iit_param *param,
 927                           struct osd_idmap_cache **oic, int noslot)
 928 {
 929         struct osd_scrub     *scrub  = &dev->od_scrub;
 930         struct ptlrpc_thread *thread = &scrub->os_thread;
 931         struct lu_fid        *fid;
 932         struct osd_inode_id  *lid;
 933         int                   rc;
 934
 935         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
 936                 struct l_wait_info lwi;
 937
 938                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
 939                 l_wait_event(thread->t_ctl_waitq,
 940                              !cfs_list_empty(&scrub->os_inconsistent_items) ||
 941                              !thread_is_running(thread),
 942                              &lwi);
 943         }
 944
 945         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 946                 spin_lock(&scrub->os_lock);
 947                 thread_set_flags(thread, SVC_STOPPING);
 948                 spin_unlock(&scrub->os_lock);
 949                 return SCRUB_NEXT_CRASH;
 950         }
 951
 952         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
 953                 return SCRUB_NEXT_FATAL;
 954
 955         if (unlikely(!thread_is_running(thread)))
 956                 return SCRUB_NEXT_EXIT;
 957
 958         if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
 959                 struct osd_inconsistent_item *oii;
 960
 961                 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
 962                                      struct osd_inconsistent_item, oii_list);
 963                 *oic = &oii->oii_cache;
 964                 scrub->os_in_prior = 1;
 965                 return 0;
 966         }
 967
 968         if (noslot != 0)
 969                 return SCRUB_NEXT_WAIT;
 970
 971         rc = osd_iit_next(param, &scrub->os_pos_current);
 972         if (rc != 0)
 973                 return rc;
 974
 975         *oic = &scrub->os_oic;
 976         fid = &(*oic)->oic_fid;
 977         lid = &(*oic)->oic_lid;
 978         rc = osd_iit_iget(info, dev, fid, lid,
 979                           scrub->os_pos_current, param->sb, true);
 980         return rc;
 981 }
 982
 983 static int osd_preload_next(struct osd_thread_info *info,
 984                             struct osd_device *dev, struct osd_iit_param *param,
 985                             struct osd_idmap_cache **oic, int noslot)
 986 {
 987         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
 988         struct osd_scrub        *scrub;
 989         struct ptlrpc_thread    *thread;
 990         int                      rc;
 991
 992         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
 993         if (rc != 0)
 994                 return rc;
 995
 996         scrub = &dev->od_scrub;
 997         thread = &scrub->os_thread;
 998         if (thread_is_running(thread) &&
 999             ooc->ooc_pos_preload >= scrub->os_pos_current)
1000                 return SCRUB_NEXT_EXIT;
1001
1002         rc = osd_iit_iget(info, dev,
1003                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1004                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1005                           ooc->ooc_pos_preload, param->sb, false);
1006         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1007          * ignore the failure, so it still need to skip the inode next time. */
1008         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1009         return rc;
1010 }
1011
1012 static inline int
1013 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1014 {
1015         spin_lock(&scrub->os_lock);
1016         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1017             !cfs_list_empty(&scrub->os_inconsistent_items) ||
1018             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1019                 scrub->os_waiting = 0;
1020         else
1021                 scrub->os_waiting = 1;
1022         spin_unlock(&scrub->os_lock);
1023
1024         return !scrub->os_waiting;
1025 }
1026
1027 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1028                           struct osd_iit_param *param,
1029                           struct osd_idmap_cache *oic, int *noslot, int rc)
1030 {
1031         struct l_wait_info       lwi    = { 0 };
1032         struct osd_scrub        *scrub  = &dev->od_scrub;
1033         struct scrub_file       *sf     = &scrub->os_file;
1034         struct ptlrpc_thread    *thread = &scrub->os_thread;
1035         struct osd_otable_it    *it     = dev->od_otable_it;
1036         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
1037
1038         switch (rc) {
1039         case SCRUB_NEXT_CONTINUE:
1040                 goto next;
1041         case SCRUB_NEXT_WAIT:
1042                 goto wait;
1043         case SCRUB_NEXT_NOSCRUB:
1044                 down_write(&scrub->os_rwsem);
1045                 scrub->os_new_checked++;
1046                 sf->sf_items_noscrub++;
1047                 up_write(&scrub->os_rwsem);
1048                 goto next;
1049         }
1050
1051         rc = osd_scrub_check_update(info, dev, oic, rc);
1052         if (rc != 0)
1053                 return rc;
1054
1055         rc = osd_scrub_checkpoint(scrub);
1056         if (rc != 0) {
1057                 CERROR("%.16s: fail to checkpoint, pos = %u, rc = %d\n",
1058                        LDISKFS_SB(param->sb)->s_es->s_volume_name,
1059                        scrub->os_pos_current, rc);
1060                 /* Continue, as long as the scrub itself can go ahead. */
1061         }
1062
1063         if (scrub->os_in_prior) {
1064                 scrub->os_in_prior = 0;
1065                 return 0;
1066         }
1067
1068 next:
1069         scrub->os_pos_current = param->gbase + ++(param->offset);
1070
1071 wait:
1072         if (it != NULL && it->ooi_waiting && ooc != NULL &&
1073             ooc->ooc_pos_preload < scrub->os_pos_current) {
1074                 spin_lock(&scrub->os_lock);
1075                 it->ooi_waiting = 0;
1076                 wake_up_all(&thread->t_ctl_waitq);
1077                 spin_unlock(&scrub->os_lock);
1078         }
1079
1080         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1081                 return 0;
1082
1083         if (osd_scrub_has_window(scrub, ooc)) {
1084                 *noslot = 0;
1085                 return 0;
1086         }
1087
1088         l_wait_event(thread->t_ctl_waitq,
1089                      osd_scrub_wakeup(scrub, it),
1090                      &lwi);
1091
1092         if (osd_scrub_has_window(scrub, ooc))
1093                 *noslot = 0;
1094         else
1095                 *noslot = 1;
1096         return 0;
1097 }
1098
1099 static int osd_preload_exec(struct osd_thread_info *info,
1100                             struct osd_device *dev, struct osd_iit_param *param,
1101                             struct osd_idmap_cache *oic, int *noslot, int rc)
1102 {
1103         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1104
1105         if (rc == 0) {
1106                 ooc->ooc_cached_items++;
1107                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1108                                         ~OSD_OTABLE_IT_CACHE_MASK;
1109         }
1110         return rc > 0 ? 0 : rc;
1111 }
1112
1113 #define SCRUB_IT_ALL    1
1114 #define SCRUB_IT_CRASH  2
1115
1116 static int osd_inode_iteration(struct osd_thread_info *info,
1117                                struct osd_device *dev, __u32 max, bool preload)
1118 {
1119         osd_iit_next_policy   next;
1120         osd_iit_exec_policy   exec;
1121         __u32                *pos;
1122         __u32                *count;
1123         struct osd_iit_param  param;
1124         __u32                 limit;
1125         int                   noslot = 0;
1126         int                   rc;
1127         ENTRY;
1128
1129         if (!preload) {
1130                 struct osd_scrub *scrub = &dev->od_scrub;
1131
1132                 next = osd_scrub_next;
1133                 exec = osd_scrub_exec;
1134                 pos = &scrub->os_pos_current;
1135                 count = &scrub->os_new_checked;
1136         } else {
1137                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1138
1139                 next = osd_preload_next;
1140                 exec = osd_preload_exec;
1141                 pos = &ooc->ooc_pos_preload;
1142                 count = &ooc->ooc_cached_items;
1143         }
1144         param.sb = osd_sb(dev);
1145         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1146
1147         while (*pos <= limit && *count < max) {
1148                 struct osd_idmap_cache *oic = NULL;
1149
1150                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1151                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1152                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1153                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1154                 if (param.bitmap == NULL) {
1155                         CERROR("%.16s: fail to read bitmap for %u, "
1156                                "scrub will stop, urgent mode\n",
1157                                LDISKFS_SB(param.sb)->s_es->s_volume_name,
1158                                (__u32)param.bg);
1159                         RETURN(-EIO);
1160                 }
1161
1162                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1163                        *count < max) {
1164                         rc = next(info, dev, &param, &oic, noslot);
1165                         switch (rc) {
1166                         case SCRUB_NEXT_BREAK:
1167                                 goto next_group;
1168                         case SCRUB_NEXT_EXIT:
1169                                 brelse(param.bitmap);
1170                                 RETURN(0);
1171                         case SCRUB_NEXT_CRASH:
1172                                 brelse(param.bitmap);
1173                                 RETURN(SCRUB_IT_CRASH);
1174                         case SCRUB_NEXT_FATAL:
1175                                 brelse(param.bitmap);
1176                                 RETURN(-EINVAL);
1177                         }
1178
1179                         rc = exec(info, dev, &param, oic, &noslot, rc);
1180                         if (rc != 0) {
1181                                 brelse(param.bitmap);
1182                                 RETURN(rc);
1183                         }
1184                 }
1185
1186 next_group:
1187                 brelse(param.bitmap);
1188         }
1189
1190         if (*pos > limit)
1191                 RETURN(SCRUB_IT_ALL);
1192         RETURN(0);
1193 }
1194
1195 static int osd_otable_it_preload(const struct lu_env *env,
1196                                  struct osd_otable_it *it)
1197 {
1198         struct osd_device       *dev   = it->ooi_dev;
1199         struct osd_scrub        *scrub = &dev->od_scrub;
1200         struct osd_otable_cache *ooc   = &it->ooi_cache;
1201         int                      rc;
1202         ENTRY;
1203
1204         rc = osd_inode_iteration(osd_oti_get(env), dev,
1205                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1206         if (rc == SCRUB_IT_ALL)
1207                 it->ooi_all_cached = 1;
1208
1209         CDEBUG(D_LFSCK, "OSD pre-loaded: max = %u, preload = %u, rc = %d\n",
1210                le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count),
1211                ooc->ooc_pos_preload, rc);
1212
1213         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1214                 spin_lock(&scrub->os_lock);
1215                 scrub->os_waiting = 0;
1216                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1217                 spin_unlock(&scrub->os_lock);
1218         }
1219
1220         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1221 }
1222
1223 static int osd_scrub_main(void *args)
1224 {
1225         struct lu_env         env;
1226         struct osd_device    *dev    = (struct osd_device *)args;
1227         struct osd_scrub     *scrub  = &dev->od_scrub;
1228         struct ptlrpc_thread *thread = &scrub->os_thread;
1229         struct super_block   *sb     = osd_sb(dev);
1230         int                   rc;
1231         ENTRY;
1232
1233         rc = lu_env_init(&env, LCT_LOCAL);
1234         if (rc != 0) {
1235                 CERROR("%.16s: OI scrub, fail to init env, rc = %d\n",
1236                        LDISKFS_SB(sb)->s_es->s_volume_name, rc);
1237                 GOTO(noenv, rc);
1238         }
1239
1240         rc = osd_scrub_prep(dev);
1241         if (rc != 0) {
1242                 CERROR("%.16s: OI scrub, fail to scrub prep, rc = %d\n",
1243                        LDISKFS_SB(sb)->s_es->s_volume_name, rc);
1244                 GOTO(out, rc);
1245         }
1246
1247         if (!scrub->os_full_speed) {
1248                 struct l_wait_info lwi = { 0 };
1249                 struct osd_otable_it *it = dev->od_otable_it;
1250                 struct osd_otable_cache *ooc = &it->ooi_cache;
1251
1252                 l_wait_event(thread->t_ctl_waitq,
1253                              it->ooi_user_ready || !thread_is_running(thread),
1254                              &lwi);
1255                 if (unlikely(!thread_is_running(thread)))
1256                         GOTO(post, rc = 0);
1257
1258                 scrub->os_pos_current = ooc->ooc_pos_preload;
1259         }
1260
1261         CDEBUG(D_LFSCK, "OI scrub: flags = 0x%x, pos = %u\n",
1262                scrub->os_start_flags, scrub->os_pos_current);
1263
1264         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1265         if (unlikely(rc == SCRUB_IT_CRASH))
1266                 GOTO(out, rc = -EINVAL);
1267         GOTO(post, rc);
1268
1269 post:
1270         osd_scrub_post(scrub, rc);
1271         CDEBUG(D_LFSCK, "OI scrub: stop, rc = %d, pos = %u\n",
1272                rc, scrub->os_pos_current);
1273
1274 out:
1275         while (!cfs_list_empty(&scrub->os_inconsistent_items)) {
1276                 struct osd_inconsistent_item *oii;
1277
1278                 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
1279                                      struct osd_inconsistent_item, oii_list);
1280                 cfs_list_del_init(&oii->oii_list);
1281                 OBD_FREE_PTR(oii);
1282         }
1283         lu_env_fini(&env);
1284
1285 noenv:
1286         spin_lock(&scrub->os_lock);
1287         thread_set_flags(thread, SVC_STOPPED);
1288         wake_up_all(&thread->t_ctl_waitq);
1289         spin_unlock(&scrub->os_lock);
1290         return rc;
1291 }
1292
1293 /* initial OI scrub */
1294
1295 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1296                          struct dentry *, filldir_t filldir);
1297
1298 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1299                                loff_t offset, __u64 ino, unsigned d_type);
1300 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1301                            loff_t offset, __u64 ino, unsigned d_type);
1302 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1303                            loff_t offset, __u64 ino, unsigned d_type);
1304
1305 static int
1306 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1307                      struct dentry *dentry, filldir_t filldir);
1308 static int
1309 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1310                   struct dentry *dentry, filldir_t filldir);
1311
1312 static int
1313 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1314                      struct dentry *dentry, filldir_t filldir);
1315
1316 enum osd_lf_flags {
1317         OLF_SCAN_SUBITEMS       = 0x0001,
1318         OLF_HIDE_FID            = 0x0002,
1319         OLF_SHOW_NAME           = 0x0004,
1320         OLF_NO_OI               = 0x0008,
1321 };
1322
1323 struct osd_lf_map {
1324         char            *olm_name;
1325         struct lu_fid    olm_fid;
1326         __u16            olm_flags;
1327         scandir_t        olm_scandir;
1328         filldir_t        olm_filldir;
1329 };
1330
1331 /* Add the new introduced local files in the list in the future. */
1332 static const struct osd_lf_map osd_lf_maps[] = {
1333         /* CATALOGS */
1334         { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1335                 NULL, NULL },
1336
1337         /* CONFIGS */
1338         { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1339                 OLF_SCAN_SUBITEMS, osd_ios_general_scan,
1340                 osd_ios_varfid_fill },
1341
1342         /* NIDTBL_VERSIONS */
1343         { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1344                 osd_ios_general_scan, osd_ios_varfid_fill },
1345
1346         /* PENDING */
1347         { "PENDING", { 0, 0, 0 }, 0, NULL, NULL },
1348
1349         /* ROOT */
1350         { "ROOT", { FID_SEQ_ROOT, 1, 0 },
1351                 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, osd_ios_ROOT_scan, NULL },
1352
1353         /* changelog_catalog */
1354         { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, NULL, NULL },
1355
1356         /* changelog_users */
1357         { CHANGELOG_USERS, { 0, 0, 0 }, 0, NULL, NULL },
1358
1359         /* fld */
1360         { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1361                 NULL, NULL },
1362
1363         /* last_rcvd */
1364         { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1365                 NULL, NULL },
1366
1367         /* lfsck_bookmark */
1368         { "lfsck_bookmark", { 0, 0, 0 }, 0, NULL, NULL },
1369
1370         /* lov_objid */
1371         { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1372                 NULL, NULL },
1373
1374         /* lov_objseq */
1375         { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1376                 OLF_SHOW_NAME, NULL, NULL },
1377
1378         /* quota_master */
1379         { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1380                 osd_ios_general_scan, osd_ios_varfid_fill },
1381
1382         /* quota_slave */
1383         { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1384                 osd_ios_general_scan, osd_ios_varfid_fill },
1385
1386         /* seq_ctl */
1387         { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1388                 OLF_SHOW_NAME, NULL, NULL },
1389
1390         /* seq_srv */
1391         { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1392                 OLF_SHOW_NAME, NULL, NULL },
1393
1394         /* health_check */
1395         { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1396                 OLF_SHOW_NAME, NULL, NULL },
1397
1398         /* lfsck_namespace */
1399         { "lfsck_namespace", { 0, 0, 0 }, 0, NULL, NULL },
1400
1401         /* OBJECTS, upgrade from old device */
1402         { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, osd_ios_OBJECTS_scan, NULL },
1403
1404         /* lquota_v2.user, upgrade from old device */
1405         { "lquota_v2.user", { 0, 0, 0 }, 0, NULL, NULL },
1406
1407         /* lquota_v2.group, upgrade from old device */
1408         { "lquota_v2.group", { 0, 0, 0 }, 0, NULL, NULL },
1409
1410         /* LAST_GROUP, upgrade from old device */
1411         { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1412                 OLF_SHOW_NAME, NULL, NULL },
1413
1414         /* SLAVE_LOG, llog for destroy slave stripes of striped dir */
1415         { "SLAVE_LOG", { FID_SEQ_LOCAL_FILE, SLAVE_LLOG_CATALOGS_OID, 0 },
1416                OLF_SHOW_NAME, NULL, NULL },
1417
1418         /* lost+found */
1419         { "lost+found", { 0, 0, 0 }, OLF_SCAN_SUBITEMS | OLF_NO_OI,
1420                 osd_ios_general_scan, osd_ios_lf_fill },
1421
1422         { NULL, { 0, 0, 0 }, 0, NULL, NULL }
1423 };
1424
1425 /* Add the new introduced files under .lustre/ in the list in the future. */
1426 static const struct osd_lf_map osd_dl_maps[] = {
1427         /* .lustre/fid */
1428         { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1429                 NULL, NULL },
1430
1431         { NULL, { 0, 0, 0 }, 0, NULL, NULL }
1432 };
1433
1434 struct osd_ios_item {
1435         cfs_list_t       oii_list;
1436         struct dentry   *oii_dentry;
1437         scandir_t        oii_scandir;
1438         filldir_t        oii_filldir;
1439 };
1440
1441 struct osd_ios_filldir_buf {
1442         struct osd_thread_info  *oifb_info;
1443         struct osd_device       *oifb_dev;
1444         struct dentry           *oifb_dentry;
1445 };
1446
1447 static inline struct dentry *
1448 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1449 {
1450         struct dentry *dentry;
1451
1452         CDEBUG(D_LFSCK, "init lookup one: parent = %.*s, name = %.*s\n",
1453                parent->d_name.len, parent->d_name.name, namelen, name);
1454
1455         dentry = ll_lookup_one_len(name, parent, namelen);
1456         if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1457                 dput(dentry);
1458                 return ERR_PTR(-ENOENT);
1459         }
1460
1461         return dentry;
1462 }
1463
1464 static int
1465 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1466                  scandir_t scandir, filldir_t filldir)
1467 {
1468         struct osd_ios_item *item;
1469         ENTRY;
1470
1471         OBD_ALLOC_PTR(item);
1472         if (item == NULL)
1473                 RETURN(-ENOMEM);
1474
1475         CFS_INIT_LIST_HEAD(&item->oii_list);
1476         item->oii_dentry = dget(dentry);
1477         item->oii_scandir = scandir;
1478         item->oii_filldir = filldir;
1479         cfs_list_add_tail(&item->oii_list, &dev->od_ios_list);
1480
1481         RETURN(0);
1482 }
1483
1484 /**
1485  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1486  *
1487  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1488  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1489  * reference the inode, or fixed if it is missing or references another inode.
1490  */
1491 static int
1492 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1493                  struct inode *inode, const struct lu_fid *fid, int flags)
1494 {
1495         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
1496         struct osd_inode_id     *id     = &info->oti_id;
1497         struct osd_inode_id     *id2    = &info->oti_id2;
1498         struct osd_scrub        *scrub  = &dev->od_scrub;
1499         struct scrub_file       *sf     = &scrub->os_file;
1500         struct lu_fid            tfid;
1501         int                      rc;
1502         ENTRY;
1503
1504         CDEBUG(D_LFSCK, "init scan one: ino = %ld\n", inode->i_ino);
1505
1506         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1507         if (rc != 0 && rc != -ENODATA)
1508                 RETURN(rc);
1509
1510         osd_id_gen(id, inode->i_ino, inode->i_generation);
1511         if (rc == -ENODATA) {
1512                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID)
1513                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1514                 else
1515                         tfid = *fid;
1516                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1517                 if (rc != 0)
1518                         RETURN(rc);
1519         } else {
1520                 if (lma->lma_compat & LMAC_NOT_IN_OI)
1521                         RETURN(0);
1522
1523                 tfid = lma->lma_self_fid;
1524         }
1525
1526         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1527         if (rc != 0) {
1528                 if (rc != -ENOENT)
1529                         RETURN(rc);
1530
1531                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1532                                                DTO_INDEX_INSERT, true, 0);
1533                 if (rc > 0)
1534                         rc = 0;
1535
1536                 RETURN(rc);
1537         }
1538
1539         if (osd_id_eq_strict(id, id2))
1540                 RETURN(0);
1541
1542         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1543                 osd_scrub_file_reset(scrub,
1544                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1545                                      SF_INCONSISTENT);
1546                 rc = osd_scrub_file_store(scrub);
1547                 if (rc != 0)
1548                         RETURN(rc);
1549         }
1550
1551         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1552                                        DTO_INDEX_UPDATE, true, 0);
1553         if (rc > 0)
1554                 rc = 0;
1555
1556         RETURN(rc);
1557 }
1558
1559 /**
1560  * It scans the /lost+found, and for the OST-object (with filter_fid
1561  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1562  */
1563 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1564                            loff_t offset, __u64 ino, unsigned d_type)
1565 {
1566         struct osd_ios_filldir_buf *fill_buf = buf;
1567         struct osd_thread_info     *info     = fill_buf->oifb_info;
1568         struct osd_device          *dev      = fill_buf->oifb_dev;
1569         struct lu_fid              *fid      = &info->oti_fid;
1570         struct osd_scrub           *scrub    = &dev->od_scrub;
1571         struct dentry              *parent   = fill_buf->oifb_dentry;
1572         struct dentry              *child;
1573         struct inode               *dir      = parent->d_inode;
1574         struct inode               *inode;
1575         int                         rc;
1576         ENTRY;
1577
1578         /* skip any '.' started names */
1579         if (name[0] == '.')
1580                 RETURN(0);
1581
1582         scrub->os_lf_scanned++;
1583         child = osd_ios_lookup_one_len(name, parent, namelen);
1584         if (IS_ERR(child)) {
1585                 CWARN("%s: cannot lookup child '%.*s': rc = %d\n",
1586                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
1587                 RETURN(0);
1588         }
1589
1590         inode = child->d_inode;
1591         if (S_ISDIR(inode->i_mode)) {
1592                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1593                                       osd_ios_lf_fill);
1594                 if (rc != 0)
1595                         CWARN("%s: cannot add child '%.*s': rc = %d\n",
1596                               osd_name(dev), namelen, name, rc);
1597                 GOTO(put, rc);
1598         }
1599
1600         if (!S_ISREG(inode->i_mode))
1601                 GOTO(put, rc = 0);
1602
1603         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1604         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1605                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1606                 if (rc == 0) {
1607                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1608                                "/lost+found.\n", namelen, name, PFID(fid));
1609                         scrub->os_lf_repaired++;
1610                 } else {
1611                         CWARN("%s: cannot rename for '%.*s' "DFID": rc = %d\n",
1612                               osd_name(dev), namelen, name, PFID(fid), rc);
1613                 }
1614         }
1615
1616         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1617          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1618          *      can process them in furtuer. */
1619
1620         GOTO(put, rc);
1621
1622 put:
1623         if (rc < 0)
1624                 scrub->os_lf_failed++;
1625         dput(child);
1626         /* skip the failure to make the scanning to continue. */
1627         return 0;
1628 }
1629
1630 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1631                                loff_t offset, __u64 ino, unsigned d_type)
1632 {
1633         struct osd_ios_filldir_buf *fill_buf = buf;
1634         struct osd_device          *dev      = fill_buf->oifb_dev;
1635         struct dentry              *child;
1636         int                         rc;
1637         ENTRY;
1638
1639         /* skip any '.' started names */
1640         if (name[0] == '.')
1641                 RETURN(0);
1642
1643         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1644         if (IS_ERR(child))
1645                 RETURN(PTR_ERR(child));
1646
1647         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1648                               NULL, 0);
1649         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1650                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1651                                       osd_ios_varfid_fill);
1652         dput(child);
1653
1654         RETURN(rc);
1655 }
1656
1657 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1658                            loff_t offset, __u64 ino, unsigned d_type)
1659 {
1660         struct osd_ios_filldir_buf *fill_buf = buf;
1661         struct osd_device          *dev      = fill_buf->oifb_dev;
1662         const struct osd_lf_map    *map;
1663         struct dentry              *child;
1664         int                         rc       = 0;
1665         ENTRY;
1666
1667         /* skip any '.' started names */
1668         if (name[0] == '.')
1669                 RETURN(0);
1670
1671         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
1672                 if (strlen(map->olm_name) != namelen)
1673                         continue;
1674
1675                 if (strncmp(map->olm_name, name, namelen) == 0)
1676                         break;
1677         }
1678
1679         if (map->olm_name == NULL)
1680                 RETURN(0);
1681
1682         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1683         if (IS_ERR(child))
1684                 RETURN(PTR_ERR(child));
1685
1686         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1687                               &map->olm_fid, map->olm_flags);
1688         dput(child);
1689
1690         RETURN(rc);
1691 }
1692
1693 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
1694                              loff_t offset, __u64 ino, unsigned d_type)
1695 {
1696         struct osd_ios_filldir_buf *fill_buf = buf;
1697         struct osd_device          *dev      = fill_buf->oifb_dev;
1698         const struct osd_lf_map    *map;
1699         struct dentry              *child;
1700         int                         rc       = 0;
1701         ENTRY;
1702
1703         /* skip any '.' started names */
1704         if (name[0] == '.')
1705                 RETURN(0);
1706
1707         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
1708                 if (strlen(map->olm_name) != namelen)
1709                         continue;
1710
1711                 if (strncmp(map->olm_name, name, namelen) == 0)
1712                         break;
1713         }
1714
1715         if (map->olm_name == NULL)
1716                 RETURN(0);
1717
1718         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1719         if (IS_ERR(child))
1720                 RETURN(PTR_ERR(child));
1721
1722         if (!(map->olm_flags & OLF_NO_OI))
1723                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1724                                       &map->olm_fid, map->olm_flags);
1725         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
1726                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
1727                                       map->olm_filldir);
1728         dput(child);
1729
1730         RETURN(rc);
1731 }
1732
1733 static int
1734 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1735                      struct dentry *dentry, filldir_t filldir)
1736 {
1737         struct osd_ios_filldir_buf    buf   = { info, dev, dentry };
1738         struct file                  *filp  = &info->oti_it_ea.oie_file;
1739         struct inode                 *inode = dentry->d_inode;
1740         const struct file_operations *fops  = inode->i_fop;
1741         int                           rc;
1742         ENTRY;
1743
1744         LASSERT(filldir != NULL);
1745
1746         filp->f_pos = 0;
1747         filp->f_dentry = dentry;
1748         filp->f_mode = FMODE_64BITHASH;
1749         filp->f_mapping = inode->i_mapping;
1750         filp->f_op = fops;
1751         filp->private_data = NULL;
1752         set_file_inode(filp, inode);
1753
1754         rc = fops->readdir(filp, &buf, filldir);
1755         fops->release(inode, filp);
1756
1757         RETURN(rc);
1758 }
1759
1760 static int
1761 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1762                   struct dentry *dentry, filldir_t filldir)
1763 {
1764         struct osd_scrub  *scrub  = &dev->od_scrub;
1765         struct scrub_file *sf     = &scrub->os_file;
1766         struct dentry     *child;
1767         int                rc;
1768         ENTRY;
1769
1770         /* It is existing MDT0 device. We only allow the case of object without
1771          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
1772          * can generate IGIF mode FID for the object and related OI mapping. If
1773          * it is on other MDTs, then becuase file-level backup/restore, related
1774          * OI mapping may be invalid already, we do not know which is the right
1775          * FID for the object. We only allow IGIF objects to reside on the MDT0.
1776          *
1777          * XXX: For the case of object on non-MDT0 device with neither LMA nor
1778          *      "fid" xattr, then something crashed. We cannot re-generate the
1779          *      FID directly, instead, the OI scrub will scan the OI structure
1780          *      and try to re-generate the LMA from the OI mapping. But if the
1781          *      OI mapping crashed or lost also, then we have to give up under
1782          *      double failure cases. */
1783         scrub->os_convert_igif = 1;
1784         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
1785                                        strlen(dot_lustre_name));
1786         if (IS_ERR(child)) {
1787                 rc = PTR_ERR(child);
1788                 if (rc == -ENOENT) {
1789                         /* It is 1.8 MDT device. */
1790                         if (!(sf->sf_flags & SF_UPGRADE)) {
1791                                 osd_scrub_file_reset(scrub,
1792                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1793                                         SF_UPGRADE);
1794                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
1795                                 rc = osd_scrub_file_store(scrub);
1796                         } else {
1797                                 rc = 0;
1798                         }
1799                 }
1800         } else {
1801                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
1802                  * so the client will get IGIF for the ".lustre" object when
1803                  * the MDT restart.
1804                  *
1805                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
1806                  * it does not know whether there are some old clients cached
1807                  * the ".lustre" IGIF during the upgrading. Two choices:
1808                  *
1809                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
1810                  *    It will allow the old connected clients to access the
1811                  *    ".lustre" with cached IGIF. But it will cause others
1812                  *    on the MDT failed to check "fid_is_dot_lustre()".
1813                  *
1814                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
1815                  *    for ".lustre" in spite of whether there are some clients
1816                  *    cached the ".lustre" IGIF or not. It enables the check
1817                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
1818                  *    that the old connected clients cannot access the ".lustre"
1819                  *    with the cached IGIF.
1820                  *
1821                  * Usually, it is rare case for the old connected clients
1822                  * to access the ".lustre" with cached IGIF. So we prefer
1823                  * to the solution 2). */
1824                 rc = osd_ios_scan_one(info, dev, child->d_inode,
1825                                       &LU_DOT_LUSTRE_FID, 0);
1826                 if (rc == 0)
1827                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1828                                               osd_ios_dl_fill);
1829                 dput(child);
1830         }
1831
1832         RETURN(rc);
1833 }
1834
1835 static int
1836 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1837                      struct dentry *dentry, filldir_t filldir)
1838 {
1839         struct osd_scrub  *scrub  = &dev->od_scrub;
1840         struct scrub_file *sf     = &scrub->os_file;
1841         struct dentry     *child;
1842         int                rc;
1843         ENTRY;
1844
1845         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
1846                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
1847                 rc = osd_scrub_file_store(scrub);
1848                 if (rc != 0)
1849                         RETURN(rc);
1850         }
1851
1852         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
1853         if (!IS_ERR(child)) {
1854                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1855                 dput(child);
1856         } else {
1857                 rc = PTR_ERR(child);
1858         }
1859
1860         if (rc != 0 && rc != -ENOENT)
1861                 RETURN(rc);
1862
1863         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
1864         if (!IS_ERR(child)) {
1865                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1866                 dput(child);
1867         } else {
1868                 rc = PTR_ERR(child);
1869         }
1870
1871         if (rc == -ENOENT)
1872                 rc = 0;
1873
1874         RETURN(rc);
1875 }
1876
1877 static int osd_initial_OI_scrub(struct osd_thread_info *info,
1878                                 struct osd_device *dev)
1879 {
1880         struct osd_ios_item     *item    = NULL;
1881         scandir_t                scandir = osd_ios_general_scan;
1882         filldir_t                filldir = osd_ios_root_fill;
1883         struct dentry           *dentry  = osd_sb(dev)->s_root;
1884         const struct osd_lf_map *map     = osd_lf_maps;
1885         int                      rc;
1886         ENTRY;
1887
1888         /* Lookup IGIF in OI by force for initial OI scrub. */
1889         dev->od_igif_inoi = 1;
1890
1891         while (1) {
1892                 rc = scandir(info, dev, dentry, filldir);
1893                 if (item != NULL) {
1894                         dput(item->oii_dentry);
1895                         OBD_FREE_PTR(item);
1896                 }
1897
1898                 if (rc != 0)
1899                         break;
1900
1901                 if (cfs_list_empty(&dev->od_ios_list))
1902                         break;
1903
1904                 item = cfs_list_entry(dev->od_ios_list.next,
1905                                       struct osd_ios_item, oii_list);
1906                 cfs_list_del_init(&item->oii_list);
1907
1908                 LASSERT(item->oii_scandir != NULL);
1909                 scandir = item->oii_scandir;
1910                 filldir = item->oii_filldir;
1911                 dentry = item->oii_dentry;
1912         }
1913
1914         while (!cfs_list_empty(&dev->od_ios_list)) {
1915                 item = cfs_list_entry(dev->od_ios_list.next,
1916                                       struct osd_ios_item, oii_list);
1917                 cfs_list_del_init(&item->oii_list);
1918                 dput(item->oii_dentry);
1919                 OBD_FREE_PTR(item);
1920         }
1921
1922         if (rc != 0)
1923                 RETURN(rc);
1924
1925         /* There maybe the case that the object has been removed, but its OI
1926          * mapping is still in the OI file, such as the "CATALOGS" after MDT
1927          * file-level backup/restore. So here cleanup the stale OI mappings. */
1928         while (map->olm_name != NULL) {
1929                 struct dentry *child;
1930
1931                 if (fid_is_zero(&map->olm_fid)) {
1932                         map++;
1933                         continue;
1934                 }
1935
1936                 child = osd_ios_lookup_one_len(map->olm_name,
1937                                                osd_sb(dev)->s_root,
1938                                                strlen(map->olm_name));
1939                 if (!IS_ERR(child))
1940                         dput(child);
1941                 else if (PTR_ERR(child) == -ENOENT)
1942                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
1943                                                   NULL, DTO_INDEX_DELETE,
1944                                                   true, 0);
1945                 map++;
1946         }
1947
1948         RETURN(0);
1949 }
1950
1951 char *osd_lf_fid2name(const struct lu_fid *fid)
1952 {
1953         const struct osd_lf_map *map = osd_lf_maps;
1954
1955         while (map->olm_name != NULL) {
1956                 if (!lu_fid_eq(fid, &map->olm_fid)) {
1957                         map++;
1958                         continue;
1959                 }
1960
1961                 if (map->olm_flags & OLF_SHOW_NAME)
1962                         return map->olm_name;
1963                 else
1964                         return "";
1965         }
1966
1967         return NULL;
1968 }
1969
1970 /* OI scrub start/stop */
1971
1972 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
1973 {
1974         struct osd_scrub     *scrub  = &dev->od_scrub;
1975         struct ptlrpc_thread *thread = &scrub->os_thread;
1976         struct l_wait_info    lwi    = { 0 };
1977         int                   rc;
1978         ENTRY;
1979
1980 again:
1981         /* os_lock: sync status between stop and scrub thread */
1982         spin_lock(&scrub->os_lock);
1983         if (thread_is_running(thread)) {
1984                 spin_unlock(&scrub->os_lock);
1985                 RETURN(-EALREADY);
1986         } else if (unlikely(thread_is_stopping(thread))) {
1987                 spin_unlock(&scrub->os_lock);
1988                 l_wait_event(thread->t_ctl_waitq,
1989                              thread_is_stopped(thread),
1990                              &lwi);
1991                 goto again;
1992         }
1993         spin_unlock(&scrub->os_lock);
1994
1995         if (scrub->os_file.sf_status == SS_COMPLETED)
1996                 flags |= SS_RESET;
1997
1998         scrub->os_start_flags = flags;
1999         thread_set_flags(thread, 0);
2000         rc = PTR_ERR(kthread_run(osd_scrub_main, dev, "OI_scrub"));
2001         if (IS_ERR_VALUE(rc)) {
2002                 CERROR("%.16s: cannot start iteration thread, rc = %d\n",
2003                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
2004                 RETURN(rc);
2005         }
2006
2007         l_wait_event(thread->t_ctl_waitq,
2008                      thread_is_running(thread) || thread_is_stopped(thread),
2009                      &lwi);
2010
2011         RETURN(0);
2012 }
2013
2014 int osd_scrub_start(struct osd_device *dev)
2015 {
2016         int rc;
2017         ENTRY;
2018
2019         /* od_otable_mutex: prevent curcurrent start/stop */
2020         mutex_lock(&dev->od_otable_mutex);
2021         rc = do_osd_scrub_start(dev, SS_AUTO);
2022         mutex_unlock(&dev->od_otable_mutex);
2023
2024         RETURN(rc == -EALREADY ? 0 : rc);
2025 }
2026
2027 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2028 {
2029         struct ptlrpc_thread *thread = &scrub->os_thread;
2030         struct l_wait_info    lwi    = { 0 };
2031
2032         /* os_lock: sync status between stop and scrub thread */
2033         spin_lock(&scrub->os_lock);
2034         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2035                 thread_set_flags(thread, SVC_STOPPING);
2036                 spin_unlock(&scrub->os_lock);
2037                 wake_up_all(&thread->t_ctl_waitq);
2038                 l_wait_event(thread->t_ctl_waitq,
2039                              thread_is_stopped(thread),
2040                              &lwi);
2041                 /* Do not skip the last lock/unlock, which can guarantee that
2042                  * the caller cannot return until the OI scrub thread exit. */
2043                 spin_lock(&scrub->os_lock);
2044         }
2045         spin_unlock(&scrub->os_lock);
2046 }
2047
2048 static void osd_scrub_stop(struct osd_device *dev)
2049 {
2050         /* od_otable_mutex: prevent curcurrent start/stop */
2051         mutex_lock(&dev->od_otable_mutex);
2052         dev->od_scrub.os_paused = 1;
2053         do_osd_scrub_stop(&dev->od_scrub);
2054         mutex_unlock(&dev->od_otable_mutex);
2055 }
2056
2057 /* OI scrub setup/cleanup */
2058
2059 static const char osd_scrub_name[] = "OI_scrub";
2060
2061 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2062 {
2063         struct osd_thread_info     *info   = osd_oti_get(env);
2064         struct osd_scrub           *scrub  = &dev->od_scrub;
2065         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
2066         struct scrub_file          *sf     = &scrub->os_file;
2067         struct super_block         *sb     = osd_sb(dev);
2068         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
2069         struct lvfs_run_ctxt        saved;
2070         struct file                *filp;
2071         struct inode               *inode;
2072         struct lu_fid              *fid    = &info->oti_fid;
2073         int                         dirty  = 0;
2074         int                         rc     = 0;
2075         ENTRY;
2076
2077         memset(scrub, 0, sizeof(*scrub));
2078         OBD_SET_CTXT_MAGIC(ctxt);
2079         ctxt->pwdmnt = dev->od_mnt;
2080         ctxt->pwd = dev->od_mnt->mnt_root;
2081         ctxt->fs = get_ds();
2082
2083         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2084         init_rwsem(&scrub->os_rwsem);
2085         spin_lock_init(&scrub->os_lock);
2086         CFS_INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2087
2088         push_ctxt(&saved, ctxt);
2089         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2090         if (IS_ERR(filp)) {
2091                 pop_ctxt(&saved, ctxt);
2092                 RETURN(PTR_ERR(filp));
2093         }
2094
2095         inode = filp->f_dentry->d_inode;
2096         /* 'What the @fid is' is not imporatant, because the object
2097          * has no OI mapping, and only is visible inside the OSD.*/
2098         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2099         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2100         if (rc != 0) {
2101                 filp_close(filp, 0);
2102                 pop_ctxt(&saved, ctxt);
2103                 RETURN(rc);
2104         }
2105
2106         scrub->os_inode = igrab(inode);
2107         filp_close(filp, 0);
2108         pop_ctxt(&saved, ctxt);
2109
2110         rc = osd_scrub_file_load(scrub);
2111         if (rc == -ENOENT) {
2112                 osd_scrub_file_init(scrub, es->s_uuid);
2113                 /* If the "/O" dir does not exist when mount (indicated by
2114                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2115                  * then it is quite probably that the device is a new one,
2116                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2117                  *
2118                  * For the rare case that "/O" and "OI_scrub" both lost on
2119                  * an old device, it can be found and cleared later.
2120                  *
2121                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2122                  * need to check "filter_fid_old" and to convert it to
2123                  * "filter_fid" for each object, and all the IGIF should
2124                  * have their FID mapping in OI files already. */
2125                 if (dev->od_maybe_new)
2126                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2127                 dirty = 1;
2128         } else if (rc != 0) {
2129                 RETURN(rc);
2130         } else {
2131                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2132                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2133                         dirty = 1;
2134                 } else if (sf->sf_status == SS_SCANNING) {
2135                         sf->sf_status = SS_CRASHED;
2136                         dirty = 1;
2137                 }
2138         }
2139
2140         if (sf->sf_pos_last_checkpoint != 0)
2141                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2142         else
2143                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2144
2145         if (dirty != 0) {
2146                 rc = osd_scrub_file_store(scrub);
2147                 if (rc != 0)
2148                         RETURN(rc);
2149         }
2150
2151         /* Initialize OI files. */
2152         rc = osd_oi_init(info, dev);
2153         if (rc < 0)
2154                 RETURN(rc);
2155
2156         rc = osd_initial_OI_scrub(info, dev);
2157         if (rc == 0) {
2158                 if (sf->sf_flags & SF_UPGRADE ||
2159                     !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2160                       sf->sf_success_count > 0)) {
2161                         dev->od_igif_inoi = 0;
2162                         dev->od_check_ff = dev->od_is_ost;
2163                 } else {
2164                         dev->od_igif_inoi = 1;
2165                         dev->od_check_ff = 0;
2166                 }
2167
2168                 if (sf->sf_flags & SF_INCONSISTENT)
2169                         /* The 'od_igif_inoi' will be set under the
2170                          * following cases:
2171                          * 1) new created system, or
2172                          * 2) restored from file-level backup, or
2173                          * 3) the upgrading completed.
2174                          *
2175                          * The 'od_igif_inoi' may be cleared by OI scrub
2176                          * later if found that the system is upgrading. */
2177                         dev->od_igif_inoi = 1;
2178
2179                 if (!dev->od_noscrub &&
2180                     ((sf->sf_status == SS_PAUSED) ||
2181                      (sf->sf_status == SS_CRASHED &&
2182                       sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2183                                       SF_UPGRADE | SF_AUTO)) ||
2184                      (sf->sf_status == SS_INIT &&
2185                       sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2186                                       SF_UPGRADE))))
2187                         rc = osd_scrub_start(dev);
2188         }
2189
2190         /* it is possible that dcache entries may keep objects after they are
2191          * deleted by OSD. While it looks safe this can cause object data to
2192          * stay until umount causing failures in tests calculating free space,
2193          * e.g. replay-ost-single. Since those dcache entries are not used
2194          * anymore let's just free them after use here */
2195         shrink_dcache_sb(sb);
2196
2197         RETURN(rc);
2198 }
2199
2200 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2201 {
2202         struct osd_scrub *scrub = &dev->od_scrub;
2203
2204         LASSERT(dev->od_otable_it == NULL);
2205
2206         if (scrub->os_inode != NULL) {
2207                 osd_scrub_stop(dev);
2208                 iput(scrub->os_inode);
2209                 scrub->os_inode = NULL;
2210         }
2211         if (dev->od_oi_table != NULL)
2212                 osd_oi_fini(osd_oti_get(env), dev);
2213 }
2214
2215 /* object table based iteration APIs */
2216
2217 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2218                                        struct dt_object *dt, __u32 attr,
2219                                        struct lustre_capa *capa)
2220 {
2221         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2222         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2223         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2224         struct osd_scrub       *scrub = &dev->od_scrub;
2225         struct osd_otable_it   *it;
2226         __u32                   start = 0;
2227         int                     rc;
2228         ENTRY;
2229
2230         /* od_otable_mutex: prevent curcurrent init/fini */
2231         mutex_lock(&dev->od_otable_mutex);
2232         if (dev->od_otable_it != NULL)
2233                 GOTO(out, it = ERR_PTR(-EALREADY));
2234
2235         OBD_ALLOC_PTR(it);
2236         if (it == NULL)
2237                 GOTO(out, it = ERR_PTR(-ENOMEM));
2238
2239         dev->od_otable_it = it;
2240         it->ooi_dev = dev;
2241         it->ooi_cache.ooc_consumer_idx = -1;
2242         if (flags & DOIF_OUTUSED)
2243                 it->ooi_used_outside = 1;
2244
2245         if (flags & DOIF_RESET)
2246                 start |= SS_RESET;
2247
2248         if (valid & DOIV_ERROR_HANDLE) {
2249                 if (flags & DOIF_FAILOUT)
2250                         start |= SS_SET_FAILOUT;
2251                 else
2252                         start |= SS_CLEAR_FAILOUT;
2253         }
2254
2255         if (valid & DOIV_DRYRUN) {
2256                 if (flags & DOIF_DRYRUN)
2257                         start |= SS_SET_DRYRUN;
2258                 else
2259                         start |= SS_CLEAR_DRYRUN;
2260         }
2261
2262         rc = do_osd_scrub_start(dev, start);
2263         if (rc < 0 && rc != -EALREADY) {
2264                 dev->od_otable_it = NULL;
2265                 OBD_FREE_PTR(it);
2266                 GOTO(out, it = ERR_PTR(rc));
2267         }
2268
2269         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2270
2271         GOTO(out, it);
2272
2273 out:
2274         mutex_unlock(&dev->od_otable_mutex);
2275         return (struct dt_it *)it;
2276 }
2277
2278 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2279 {
2280         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2281         struct osd_device    *dev = it->ooi_dev;
2282
2283         /* od_otable_mutex: prevent curcurrent init/fini */
2284         mutex_lock(&dev->od_otable_mutex);
2285         do_osd_scrub_stop(&dev->od_scrub);
2286         LASSERT(dev->od_otable_it == it);
2287
2288         dev->od_otable_it = NULL;
2289         mutex_unlock(&dev->od_otable_mutex);
2290         OBD_FREE_PTR(it);
2291 }
2292
2293 static int osd_otable_it_get(const struct lu_env *env,
2294                              struct dt_it *di, const struct dt_key *key)
2295 {
2296         return 0;
2297 }
2298
2299 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2300 {
2301 }
2302
2303 static inline int
2304 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2305 {
2306         spin_lock(&scrub->os_lock);
2307         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2308             scrub->os_waiting ||
2309             !thread_is_running(&scrub->os_thread))
2310                 it->ooi_waiting = 0;
2311         else
2312                 it->ooi_waiting = 1;
2313         spin_unlock(&scrub->os_lock);
2314
2315         return !it->ooi_waiting;
2316 }
2317
2318 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2319 {
2320         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2321         struct osd_device       *dev    = it->ooi_dev;
2322         struct osd_scrub        *scrub  = &dev->od_scrub;
2323         struct osd_otable_cache *ooc    = &it->ooi_cache;
2324         struct ptlrpc_thread    *thread = &scrub->os_thread;
2325         struct l_wait_info       lwi    = { 0 };
2326         int                      rc;
2327         ENTRY;
2328
2329         LASSERT(it->ooi_user_ready);
2330
2331 again:
2332         if (!thread_is_running(thread) && !it->ooi_used_outside)
2333                 RETURN(1);
2334
2335         if (ooc->ooc_cached_items > 0) {
2336                 ooc->ooc_cached_items--;
2337                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2338                                         ~OSD_OTABLE_IT_CACHE_MASK;
2339                 RETURN(0);
2340         }
2341
2342         if (it->ooi_all_cached) {
2343                 l_wait_event(thread->t_ctl_waitq,
2344                              !thread_is_running(thread),
2345                              &lwi);
2346                 RETURN(1);
2347         }
2348
2349         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2350                 spin_lock(&scrub->os_lock);
2351                 scrub->os_waiting = 0;
2352                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2353                 spin_unlock(&scrub->os_lock);
2354         }
2355
2356         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2357                 l_wait_event(thread->t_ctl_waitq,
2358                              osd_otable_it_wakeup(scrub, it),
2359                              &lwi);
2360
2361         if (!thread_is_running(thread) && !it->ooi_used_outside)
2362                 RETURN(1);
2363
2364         rc = osd_otable_it_preload(env, it);
2365         if (rc >= 0)
2366                 goto again;
2367
2368         RETURN(rc);
2369 }
2370
2371 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2372                                         const struct dt_it *di)
2373 {
2374         return NULL;
2375 }
2376
2377 static int osd_otable_it_key_size(const struct lu_env *env,
2378                                   const struct dt_it *di)
2379 {
2380         return sizeof(__u64);
2381 }
2382
2383 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2384                              struct dt_rec *rec, __u32 attr)
2385 {
2386         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2387         struct osd_otable_cache *ooc = &it->ooi_cache;
2388
2389         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2390
2391         /* Filter out Invald FID already. */
2392         LASSERTF(fid_is_sane((struct lu_fid *)rec),
2393                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2394                  PFID((struct lu_fid *)rec),
2395                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2396
2397         return 0;
2398 }
2399
2400 static __u64 osd_otable_it_store(const struct lu_env *env,
2401                                  const struct dt_it *di)
2402 {
2403         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2404         struct osd_otable_cache *ooc = &it->ooi_cache;
2405         __u64                    hash;
2406
2407         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2408                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2409         else
2410                 hash = ooc->ooc_pos_preload;
2411         return hash;
2412 }
2413
2414 /**
2415  * Set the OSD layer iteration start position as the specified hash.
2416  */
2417 static int osd_otable_it_load(const struct lu_env *env,
2418                               const struct dt_it *di, __u64 hash)
2419 {
2420         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
2421         struct osd_device       *dev   = it->ooi_dev;
2422         struct osd_otable_cache *ooc   = &it->ooi_cache;
2423         struct osd_scrub        *scrub = &dev->od_scrub;
2424         int                      rc;
2425         ENTRY;
2426
2427         /* Forbid to set iteration position after iteration started. */
2428         if (it->ooi_user_ready)
2429                 RETURN(-EPERM);
2430
2431         if (hash > OSD_OTABLE_MAX_HASH)
2432                 hash = OSD_OTABLE_MAX_HASH;
2433
2434         ooc->ooc_pos_preload = hash;
2435         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2436                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2437
2438         it->ooi_user_ready = 1;
2439         if (!scrub->os_full_speed)
2440                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2441
2442         /* Unplug OSD layer iteration by the first next() call. */
2443         rc = osd_otable_it_next(env, (struct dt_it *)it);
2444
2445         RETURN(rc);
2446 }
2447
2448 static int osd_otable_it_key_rec(const struct lu_env *env,
2449                                  const struct dt_it *di, void *key_rec)
2450 {
2451         return 0;
2452 }
2453
2454 const struct dt_index_operations osd_otable_ops = {
2455         .dio_it = {
2456                 .init     = osd_otable_it_init,
2457                 .fini     = osd_otable_it_fini,
2458                 .get      = osd_otable_it_get,
2459                 .put      = osd_otable_it_put,
2460                 .next     = osd_otable_it_next,
2461                 .key      = osd_otable_it_key,
2462                 .key_size = osd_otable_it_key_size,
2463                 .rec      = osd_otable_it_rec,
2464                 .store    = osd_otable_it_store,
2465                 .load     = osd_otable_it_load,
2466                 .key_rec  = osd_otable_it_key_rec,
2467         }
2468 };
2469
2470 /* high priority inconsistent items list APIs */
2471
2472 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2473                    int insert)
2474 {
2475         struct osd_inconsistent_item *oii;
2476         struct osd_scrub             *scrub  = &dev->od_scrub;
2477         struct ptlrpc_thread         *thread = &scrub->os_thread;
2478         int                           wakeup = 0;
2479         ENTRY;
2480
2481         OBD_ALLOC_PTR(oii);
2482         if (unlikely(oii == NULL))
2483                 RETURN(-ENOMEM);
2484
2485         CFS_INIT_LIST_HEAD(&oii->oii_list);
2486         oii->oii_cache = *oic;
2487         oii->oii_insert = insert;
2488
2489         spin_lock(&scrub->os_lock);
2490         if (unlikely(!thread_is_running(thread))) {
2491                 spin_unlock(&scrub->os_lock);
2492                 OBD_FREE_PTR(oii);
2493                 RETURN(-EAGAIN);
2494         }
2495
2496         if (cfs_list_empty(&scrub->os_inconsistent_items))
2497                 wakeup = 1;
2498         cfs_list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2499         spin_unlock(&scrub->os_lock);
2500
2501         if (wakeup != 0)
2502                 wake_up_all(&thread->t_ctl_waitq);
2503
2504         RETURN(0);
2505 }
2506
2507 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2508                    struct osd_inode_id *id)
2509 {
2510         struct osd_scrub             *scrub = &dev->od_scrub;
2511         struct osd_inconsistent_item *oii;
2512         ENTRY;
2513
2514         spin_lock(&scrub->os_lock);
2515         cfs_list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2516                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2517                         *id = oii->oii_cache.oic_lid;
2518                         spin_unlock(&scrub->os_lock);
2519                         RETURN(0);
2520                 }
2521         }
2522         spin_unlock(&scrub->os_lock);
2523
2524         RETURN(-ENOENT);
2525 }
2526
2527 /* OI scrub dump */
2528
2529 static const char *scrub_status_names[] = {
2530         "init",
2531         "scanning",
2532         "completed",
2533         "failed",
2534         "stopped",
2535         "paused",
2536         "crashed",
2537         NULL
2538 };
2539
2540 static const char *scrub_flags_names[] = {
2541         "recreated",
2542         "inconsistent",
2543         "auto",
2544         "upgrade",
2545         NULL
2546 };
2547
2548 static const char *scrub_param_names[] = {
2549         "failout",
2550         "dryrun",
2551         NULL
2552 };
2553
2554 static int scrub_bits_dump(char **buf, int *len, int bits, const char *names[],
2555                            const char *prefix)
2556 {
2557         int save = *len;
2558         int flag;
2559         int rc;
2560         int i;
2561
2562         rc = snprintf(*buf, *len, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2563         if (rc <= 0)
2564                 return -ENOSPC;
2565
2566         *buf += rc;
2567         *len -= rc;
2568         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2569                 if (flag & bits) {
2570                         bits &= ~flag;
2571                         rc = snprintf(*buf, *len, "%s%c", names[i],
2572                                       bits != 0 ? ',' : '\n');
2573                         if (rc <= 0)
2574                                 return -ENOSPC;
2575
2576                         *buf += rc;
2577                         *len -= rc;
2578                 }
2579         }
2580         return save - *len;
2581 }
2582
2583 static int scrub_time_dump(char **buf, int *len, __u64 time, const char *prefix)
2584 {
2585         int rc;
2586
2587         if (time != 0)
2588                 rc = snprintf(*buf, *len, "%s: "LPU64" seconds\n", prefix,
2589                               cfs_time_current_sec() - time);
2590         else
2591                 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
2592         if (rc <= 0)
2593                 return -ENOSPC;
2594
2595         *buf += rc;
2596         *len -= rc;
2597         return rc;
2598 }
2599
2600 static int scrub_pos_dump(char **buf, int *len, __u64 pos, const char *prefix)
2601 {
2602         int rc;
2603
2604         if (pos != 0)
2605                 rc = snprintf(*buf, *len, "%s: "LPU64"\n", prefix, pos);
2606         else
2607                 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
2608         if (rc <= 0)
2609                 return -ENOSPC;
2610
2611         *buf += rc;
2612         *len -= rc;
2613         return rc;
2614 }
2615
2616 int osd_scrub_dump(struct osd_device *dev, char *buf, int len)
2617 {
2618         struct osd_scrub  *scrub   = &dev->od_scrub;
2619         struct scrub_file *sf      = &scrub->os_file;
2620         __u64              checked;
2621         __u64              speed;
2622         int                save    = len;
2623         int                ret     = -ENOSPC;
2624         int                rc;
2625
2626         down_read(&scrub->os_rwsem);
2627         rc = snprintf(buf, len,
2628                       "name: OI_scrub\n"
2629                       "magic: 0x%x\n"
2630                       "oi_files: %d\n"
2631                       "status: %s\n",
2632                       sf->sf_magic, (int)sf->sf_oi_count,
2633                       scrub_status_names[sf->sf_status]);
2634         if (rc <= 0)
2635                 goto out;
2636
2637         buf += rc;
2638         len -= rc;
2639         rc = scrub_bits_dump(&buf, &len, sf->sf_flags, scrub_flags_names,
2640                              "flags");
2641         if (rc < 0)
2642                 goto out;
2643
2644         rc = scrub_bits_dump(&buf, &len, sf->sf_param, scrub_param_names,
2645                              "param");
2646         if (rc < 0)
2647                 goto out;
2648
2649         rc = scrub_time_dump(&buf, &len, sf->sf_time_last_complete,
2650                              "time_since_last_completed");
2651         if (rc < 0)
2652                 goto out;
2653
2654         rc = scrub_time_dump(&buf, &len, sf->sf_time_latest_start,
2655                              "time_since_latest_start");
2656         if (rc < 0)
2657                 goto out;
2658
2659         rc = scrub_time_dump(&buf, &len, sf->sf_time_last_checkpoint,
2660                              "time_since_last_checkpoint");
2661         if (rc < 0)
2662                 goto out;
2663
2664         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_latest_start,
2665                             "latest_start_position");
2666         if (rc < 0)
2667                 goto out;
2668
2669         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_last_checkpoint,
2670                             "last_checkpoint_position");
2671         if (rc < 0)
2672                 goto out;
2673
2674         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_first_inconsistent,
2675                             "first_failure_position");
2676         if (rc < 0)
2677                 goto out;
2678
2679         checked = sf->sf_items_checked + scrub->os_new_checked;
2680         rc = snprintf(buf, len,
2681                       "checked: "LPU64"\n"
2682                       "updated: "LPU64"\n"
2683                       "failed: "LPU64"\n"
2684                       "prior_updated: "LPU64"\n"
2685                       "noscrub: "LPU64"\n"
2686                       "igif: "LPU64"\n"
2687                       "success_count: %u\n",
2688                       checked, sf->sf_items_updated, sf->sf_items_failed,
2689                       sf->sf_items_updated_prior, sf->sf_items_noscrub,
2690                       sf->sf_items_igif, sf->sf_success_count);
2691         if (rc <= 0)
2692                 goto out;
2693
2694         buf += rc;
2695         len -= rc;
2696         speed = checked;
2697         if (thread_is_running(&scrub->os_thread)) {
2698                 cfs_duration_t duration = cfs_time_current() -
2699                                           scrub->os_time_last_checkpoint;
2700                 __u64 new_checked = scrub->os_new_checked * HZ;
2701                 __u32 rtime = sf->sf_run_time +
2702                               cfs_duration_sec(duration + HALF_SEC);
2703
2704                 if (duration != 0)
2705                         do_div(new_checked, duration);
2706                 if (rtime != 0)
2707                         do_div(speed, rtime);
2708                 rc = snprintf(buf, len,
2709                               "run_time: %u seconds\n"
2710                               "average_speed: "LPU64" objects/sec\n"
2711                               "real-time_speed: "LPU64" objects/sec\n"
2712                               "current_position: %u\n"
2713                               "lf_scanned: "LPU64"\n"
2714                               "lf_reparied: "LPU64"\n"
2715                               "lf_failed: "LPU64"\n",
2716                               rtime, speed, new_checked, scrub->os_pos_current,
2717                               scrub->os_lf_scanned, scrub->os_lf_repaired,
2718                               scrub->os_lf_failed);
2719         } else {
2720                 if (sf->sf_run_time != 0)
2721                         do_div(speed, sf->sf_run_time);
2722                 rc = snprintf(buf, len,
2723                               "run_time: %u seconds\n"
2724                               "average_speed: "LPU64" objects/sec\n"
2725                               "real-time_speed: N/A\n"
2726                               "current_position: N/A\n"
2727                               "lf_scanned: "LPU64"\n"
2728                               "lf_reparied: "LPU64"\n"
2729                               "lf_failed: "LPU64"\n",
2730                               sf->sf_run_time, speed, scrub->os_lf_scanned,
2731                               scrub->os_lf_repaired, scrub->os_lf_failed);
2732         }
2733         if (rc <= 0)
2734                 goto out;
2735
2736         buf += rc;
2737         len -= rc;
2738         ret = save - len;
2739
2740 out:
2741         up_read(&scrub->os_rwsem);
2742         return ret;
2743 }