lustre/osd-ldiskfs/osd_scrub.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License version 2 for more details.  A copy is
  14  * included in the COPYING file that accompanied this code.
  15
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2012, 2014, Intel Corporation.
  24  */
  25 /*
  26  * lustre/osd-ldiskfs/osd_scrub.c
  27  *
  28  * Top-level entry points into osd module
  29  *
  30  * The OI scrub is used for rebuilding Object Index files when restores MDT from
  31  * file-level backup.
  32  *
  33  * The otable based iterator scans ldiskfs inode table to feed up layer LFSCK.
  34  *
  35  * Author: Fan Yong <yong.fan@whamcloud.com>
  36  */
  37
  38 #define DEBUG_SUBSYSTEM S_LFSCK
  39
  40 #include <linux/kthread.h>
  41 #include <lustre/lustre_idl.h>
  42 #include <lustre_disk.h>
  43 #include <dt_object.h>
  44 #include <linux/xattr.h>
  45
  46 #include "osd_internal.h"
  47 #include "osd_oi.h"
  48 #include "osd_scrub.h"
  49
  50 #define HALF_SEC        msecs_to_jiffies(MSEC_PER_SEC >> 1)
  51
  52 #define OSD_OTABLE_MAX_HASH             0x00000000ffffffffULL
  53
  54 #define SCRUB_NEXT_BREAK        1 /* exit current loop and process next group */
  55 #define SCRUB_NEXT_CONTINUE     2 /* skip current object and process next bit */
  56 #define SCRUB_NEXT_EXIT         3 /* exit all the loops */
  57 #define SCRUB_NEXT_WAIT         4 /* wait for free cache slot */
  58 #define SCRUB_NEXT_CRASH        5 /* simulate system crash during OI scrub */
  59 #define SCRUB_NEXT_FATAL        6 /* simulate failure during OI scrub */
  60 #define SCRUB_NEXT_NOSCRUB      7 /* new created object, no scrub on it */
  61 #define SCRUB_NEXT_NOLMA        8 /* the inode has no FID-in-LMA */
  62 #define SCRUB_NEXT_OSTOBJ       9 /* for OST-object */
  63 #define SCRUB_NEXT_OSTOBJ_OLD   10 /* old OST-object, no LMA or no FID-on-OST
  64                                     * flags in LMA */
  65
  66 /* misc functions */
  67
  68 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
  69 {
  70         return container_of0(scrub, struct osd_device, od_scrub);
  71 }
  72
  73 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
  74 {
  75         return osd_sb(osd_scrub2dev(scrub));
  76 }
  77
  78 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
  79                                        struct osd_otable_cache *ooc)
  80 {
  81         return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
  82 }
  83
  84 static inline const char *osd_scrub2name(struct osd_scrub *scrub)
  85 {
  86         return LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
  87 }
  88
  89 /**
  90  * update/insert/delete the specified OI mapping (@fid @id) according to the ops
  91  *
  92  * \retval   1, changed nothing
  93  * \retval   0, changed successfully
  94  * \retval -ve, on error
  95  */
  96 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
  97                                      struct osd_device *dev,
  98                                      const struct lu_fid *fid,
  99                                      const struct osd_inode_id *id,
 100                                      int ops, bool force,
 101                                      enum oi_check_flags flags)
 102 {
 103         handle_t *th;
 104         int       rc;
 105         ENTRY;
 106
 107         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
 108                 RETURN(0);
 109
 110         /* DTO_INDEX_INSERT is enough for other two ops:
 111          * delete/update, but save stack. */
 112         th = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 113                                 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
 114         if (IS_ERR(th)) {
 115                 rc = PTR_ERR(th);
 116                 CDEBUG(D_LFSCK, "%s: fail to start trans for scrub op %d "
 117                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 118                        PFID(fid), id->oii_ino, id->oii_gen, rc);
 119                 RETURN(rc);
 120         }
 121
 122         switch (ops) {
 123         case DTO_INDEX_UPDATE:
 124                 rc = osd_oi_update(info, dev, fid, id, th, flags);
 125                 if (unlikely(rc == -ENOENT)) {
 126                         /* Some unlink thread may removed the OI mapping. */
 127                         rc = 1;
 128                 }
 129                 break;
 130         case DTO_INDEX_INSERT:
 131                 rc = osd_oi_insert(info, dev, fid, id, th, flags);
 132                 if (unlikely(rc == -EEXIST)) {
 133                         rc = 1;
 134                         /* XXX: There are trouble things when adding OI
 135                          *      mapping for IGIF object, which may cause
 136                          *      multiple objects to be mapped to the same
 137                          *      IGIF formatted FID. Consider the following
 138                          *      situations:
 139                          *
 140                          *      1) The MDT is upgrading from 1.8 device.
 141                          *      The OI scrub generates IGIF FID1 for the
 142                          *      OBJ1 and adds the OI mapping.
 143                          *
 144                          *      2) For some reason, the OI scrub does not
 145                          *      process all the IGIF objects completely.
 146                          *
 147                          *      3) The MDT is backuped and restored against
 148                          *      this device.
 149                          *
 150                          *      4) When the MDT mounts up, the OI scrub will
 151                          *      try to rebuild the OI files. For some IGIF
 152                          *      object, OBJ2, which was not processed by the
 153                          *      OI scrub before the backup/restore, and the
 154                          *      new generated IGIF formatted FID may be just
 155                          *      the FID1, the same as OBJ1.
 156                          *
 157                          *      Under such case, the OI scrub cannot know how
 158                          *      to generate new FID for the OBJ2.
 159                          *
 160                          *      Currently, we do nothing for that. One possible
 161                          *      solution is to generate new normal FID for the
 162                          *      conflict object.
 163                          *
 164                          *      Anyway, it is rare, only exists in theory. */
 165                 }
 166                 break;
 167         case DTO_INDEX_DELETE:
 168                 rc = osd_oi_delete(info, dev, fid, th, flags);
 169                 if (rc == -ENOENT) {
 170                         /* It is normal that the unlink thread has removed the
 171                          * OI mapping already. */
 172                         rc = 1;
 173                 }
 174                 break;
 175         default:
 176                 LASSERTF(0, "Unexpected ops %d\n", ops);
 177                 break;
 178         }
 179
 180         ldiskfs_journal_stop(th);
 181         if (rc < 0)
 182                 CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d "
 183                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
 184                        PFID(fid), id->oii_ino, id->oii_gen, rc);
 185
 186         RETURN(rc);
 187 }
 188
 189 /* OI_scrub file ops */
 190
 191 static void osd_scrub_file_to_cpu(struct scrub_file *des,
 192                                   struct scrub_file *src)
 193 {
 194         memcpy(des->sf_uuid, src->sf_uuid, 16);
 195         des->sf_flags   = le64_to_cpu(src->sf_flags);
 196         des->sf_magic   = le32_to_cpu(src->sf_magic);
 197         des->sf_status  = le16_to_cpu(src->sf_status);
 198         des->sf_param   = le16_to_cpu(src->sf_param);
 199         des->sf_time_last_complete      =
 200                                 le64_to_cpu(src->sf_time_last_complete);
 201         des->sf_time_latest_start       =
 202                                 le64_to_cpu(src->sf_time_latest_start);
 203         des->sf_time_last_checkpoint    =
 204                                 le64_to_cpu(src->sf_time_last_checkpoint);
 205         des->sf_pos_latest_start        =
 206                                 le64_to_cpu(src->sf_pos_latest_start);
 207         des->sf_pos_last_checkpoint     =
 208                                 le64_to_cpu(src->sf_pos_last_checkpoint);
 209         des->sf_pos_first_inconsistent  =
 210                                 le64_to_cpu(src->sf_pos_first_inconsistent);
 211         des->sf_items_checked           =
 212                                 le64_to_cpu(src->sf_items_checked);
 213         des->sf_items_updated           =
 214                                 le64_to_cpu(src->sf_items_updated);
 215         des->sf_items_failed            =
 216                                 le64_to_cpu(src->sf_items_failed);
 217         des->sf_items_updated_prior     =
 218                                 le64_to_cpu(src->sf_items_updated_prior);
 219         des->sf_run_time        = le32_to_cpu(src->sf_run_time);
 220         des->sf_success_count   = le32_to_cpu(src->sf_success_count);
 221         des->sf_oi_count        = le16_to_cpu(src->sf_oi_count);
 222         des->sf_internal_flags  = le16_to_cpu(src->sf_internal_flags);
 223         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 224 }
 225
 226 static void osd_scrub_file_to_le(struct scrub_file *des,
 227                                  struct scrub_file *src)
 228 {
 229         memcpy(des->sf_uuid, src->sf_uuid, 16);
 230         des->sf_flags   = cpu_to_le64(src->sf_flags);
 231         des->sf_magic   = cpu_to_le32(src->sf_magic);
 232         des->sf_status  = cpu_to_le16(src->sf_status);
 233         des->sf_param   = cpu_to_le16(src->sf_param);
 234         des->sf_time_last_complete      =
 235                                 cpu_to_le64(src->sf_time_last_complete);
 236         des->sf_time_latest_start       =
 237                                 cpu_to_le64(src->sf_time_latest_start);
 238         des->sf_time_last_checkpoint    =
 239                                 cpu_to_le64(src->sf_time_last_checkpoint);
 240         des->sf_pos_latest_start        =
 241                                 cpu_to_le64(src->sf_pos_latest_start);
 242         des->sf_pos_last_checkpoint     =
 243                                 cpu_to_le64(src->sf_pos_last_checkpoint);
 244         des->sf_pos_first_inconsistent  =
 245                                 cpu_to_le64(src->sf_pos_first_inconsistent);
 246         des->sf_items_checked           =
 247                                 cpu_to_le64(src->sf_items_checked);
 248         des->sf_items_updated           =
 249                                 cpu_to_le64(src->sf_items_updated);
 250         des->sf_items_failed            =
 251                                 cpu_to_le64(src->sf_items_failed);
 252         des->sf_items_updated_prior     =
 253                                 cpu_to_le64(src->sf_items_updated_prior);
 254         des->sf_run_time        = cpu_to_le32(src->sf_run_time);
 255         des->sf_success_count   = cpu_to_le32(src->sf_success_count);
 256         des->sf_oi_count        = cpu_to_le16(src->sf_oi_count);
 257         des->sf_internal_flags  = cpu_to_le16(src->sf_internal_flags);
 258         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
 259 }
 260
 261 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
 262 {
 263         struct scrub_file *sf = &scrub->os_file;
 264
 265         memset(sf, 0, sizeof(*sf));
 266         memcpy(sf->sf_uuid, uuid, 16);
 267         sf->sf_magic = SCRUB_MAGIC_V1;
 268         sf->sf_status = SS_INIT;
 269 }
 270
 271 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
 272 {
 273         struct scrub_file *sf = &scrub->os_file;
 274
 275         CDEBUG(D_LFSCK, "%.16s: reset OI scrub file, old flags = "
 276                LPX64", add flags = "LPX64"\n",
 277                osd_scrub2name(scrub), sf->sf_flags, flags);
 278
 279         memcpy(sf->sf_uuid, uuid, 16);
 280         sf->sf_status = SS_INIT;
 281         sf->sf_flags |= flags;
 282         sf->sf_flags &= ~SF_AUTO;
 283         sf->sf_run_time = 0;
 284         sf->sf_time_latest_start = 0;
 285         sf->sf_time_last_checkpoint = 0;
 286         sf->sf_pos_latest_start = 0;
 287         sf->sf_pos_last_checkpoint = 0;
 288         sf->sf_pos_first_inconsistent = 0;
 289         sf->sf_items_checked = 0;
 290         sf->sf_items_updated = 0;
 291         sf->sf_items_failed = 0;
 292         if (!scrub->os_in_join)
 293                 sf->sf_items_updated_prior = 0;
 294
 295         sf->sf_items_noscrub = 0;
 296         sf->sf_items_igif = 0;
 297 }
 298
 299 static int osd_scrub_file_load(struct osd_scrub *scrub)
 300 {
 301         loff_t  pos  = 0;
 302         int     len  = sizeof(scrub->os_file_disk);
 303         int     rc;
 304
 305         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
 306         if (rc == len) {
 307                 struct scrub_file *sf = &scrub->os_file;
 308
 309                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
 310                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
 311                         CDEBUG(D_LFSCK, "%.16s: invalid scrub magic "
 312                                "0x%x != 0x%x\n", osd_scrub2name(scrub),
 313                                sf->sf_magic, SCRUB_MAGIC_V1);
 314                         /* Process it as new scrub file. */
 315                         rc = -ENOENT;
 316                 } else {
 317                         rc = 0;
 318                 }
 319         } else if (rc != 0) {
 320                 CDEBUG(D_LFSCK, "%.16s: fail to load scrub file, "
 321                        "expected = %d: rc = %d\n",
 322                        osd_scrub2name(scrub), len, rc);
 323                 if (rc > 0)
 324                         rc = -EFAULT;
 325         } else {
 326                 /* return -ENOENT for empty scrub file case. */
 327                 rc = -ENOENT;
 328         }
 329
 330         return rc;
 331 }
 332
 333 int osd_scrub_file_store(struct osd_scrub *scrub)
 334 {
 335         struct osd_device *dev;
 336         handle_t          *jh;
 337         loff_t             pos     = 0;
 338         int                len     = sizeof(scrub->os_file_disk);
 339         int                credits;
 340         int                rc;
 341
 342         dev = container_of0(scrub, struct osd_device, od_scrub);
 343         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
 344                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
 345         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits);
 346         if (IS_ERR(jh)) {
 347                 rc = PTR_ERR(jh);
 348                 CDEBUG(D_LFSCK, "%.16s: fail to start trans for scrub store: "
 349                        "rc = %d\n", osd_scrub2name(scrub), rc);
 350                 return rc;
 351         }
 352
 353         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
 354         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
 355                                       len, 0, &pos, jh);
 356         ldiskfs_journal_stop(jh);
 357         if (rc != 0)
 358                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file, "
 359                        "expected = %d: rc = %d\n",
 360                        osd_scrub2name(scrub), len, rc);
 361
 362         scrub->os_time_last_checkpoint = cfs_time_current();
 363         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
 364                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
 365         return rc;
 366 }
 367
 368 static int
 369 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
 370                      struct inode *inode, const struct lu_fid *fid)
 371 {
 372         struct filter_fid_old   *ff      = &info->oti_ff;
 373         struct dentry           *dentry  = &info->oti_obj_dentry;
 374         struct lu_fid           *tfid    = &info->oti_fid;
 375         handle_t                *jh;
 376         int                      size    = 0;
 377         int                      rc;
 378         bool                     removed = false;
 379         bool                     reset   = true;
 380         ENTRY;
 381
 382         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
 383                 RETURN(0);
 384
 385         if (fid_is_idif(fid) && dev->od_index_in_idif == 0) {
 386                 struct ost_id *oi = &info->oti_ostid;
 387
 388                 fid_to_ostid(fid, oi);
 389                 ostid_to_fid(tfid, oi, 0);
 390         } else {
 391                 *tfid = *fid;
 392         }
 393
 394         /* We want the LMA to fit into the 256-byte OST inode, so operate
 395          * as following:
 396          * 1) read old XATTR_NAME_FID and save the parent FID;
 397          * 2) delete the old XATTR_NAME_FID;
 398          * 3) make new LMA and add it;
 399          * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
 400          *
 401          * Making the LMA to fit into the 256-byte OST inode can save time for
 402          * normal osd_check_lma() and for other OI scrub scanning in future.
 403          * So it is worth to make some slow conversion here. */
 404         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
 405                                 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
 406         if (IS_ERR(jh)) {
 407                 rc = PTR_ERR(jh);
 408                 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
 409                        DFID": rc = %d\n", osd_name(dev), PFID(tfid), rc);
 410                 RETURN(rc);
 411         }
 412
 413         /* 1) read old XATTR_NAME_FID and save the parent FID */
 414         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 415         if (rc == sizeof(*ff)) {
 416                 /* 2) delete the old XATTR_NAME_FID */
 417                 ll_vfs_dq_init(inode);
 418                 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
 419                 if (rc != 0)
 420                         GOTO(stop, rc);
 421
 422                 removed = true;
 423         } else if (unlikely(rc == -ENODATA)) {
 424                 reset = false;
 425         } else if (rc != sizeof(struct filter_fid)) {
 426                 GOTO(stop, rc = -EINVAL);
 427         }
 428
 429         /* 3) make new LMA and add it */
 430         rc = osd_ea_fid_set(info, inode, tfid, LMAC_FID_ON_OST, 0);
 431         if (rc == 0 && reset)
 432                 size = sizeof(struct filter_fid);
 433         else if (rc != 0 && removed)
 434                 /* If failed, we should try to add the old back. */
 435                 size = sizeof(struct filter_fid_old);
 436
 437         /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
 438         if (size > 0) {
 439                 int rc1;
 440
 441                 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
 442                                       XATTR_CREATE);
 443                 if (rc1 != 0 && rc == 0)
 444                         rc = rc1;
 445         }
 446
 447         GOTO(stop, rc);
 448
 449 stop:
 450         ldiskfs_journal_stop(jh);
 451         if (rc < 0)
 452                 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
 453                        osd_name(dev), PFID(tfid), rc);
 454         return rc;
 455 }
 456
 457 static int
 458 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
 459                        struct osd_idmap_cache *oic, int val)
 460 {
 461         struct osd_scrub             *scrub  = &dev->od_scrub;
 462         struct scrub_file            *sf     = &scrub->os_file;
 463         struct lu_fid                *fid    = &oic->oic_fid;
 464         struct osd_inode_id          *lid    = &oic->oic_lid;
 465         struct osd_inode_id          *lid2   = &info->oti_id;
 466         struct osd_inconsistent_item *oii    = NULL;
 467         struct inode                 *inode  = NULL;
 468         int                           ops    = DTO_INDEX_UPDATE;
 469         int                           idx;
 470         int                           rc;
 471         bool                          converted = false;
 472         ENTRY;
 473
 474         down_write(&scrub->os_rwsem);
 475         scrub->os_new_checked++;
 476         if (val < 0)
 477                 GOTO(out, rc = val);
 478
 479         if (scrub->os_in_prior)
 480                 oii = list_entry(oic, struct osd_inconsistent_item,
 481                                  oii_cache);
 482
 483         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
 484                 GOTO(out, rc = 0);
 485
 486         if (fid_is_igif(fid))
 487                 sf->sf_items_igif++;
 488
 489         if (val == SCRUB_NEXT_OSTOBJ_OLD) {
 490                 inode = osd_iget(info, dev, lid);
 491                 if (IS_ERR(inode)) {
 492                         rc = PTR_ERR(inode);
 493                         /* Someone removed the inode. */
 494                         if (rc == -ENOENT || rc == -ESTALE)
 495                                 rc = 0;
 496                         GOTO(out, rc);
 497                 }
 498
 499                 sf->sf_flags |= SF_UPGRADE;
 500                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
 501                 dev->od_check_ff = 1;
 502                 rc = osd_scrub_convert_ff(info, dev, inode, fid);
 503                 if (rc != 0)
 504                         GOTO(out, rc);
 505
 506                 converted = true;
 507         }
 508
 509         if ((val == SCRUB_NEXT_NOLMA) &&
 510             (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
 511                 GOTO(out, rc = 0);
 512
 513         if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA)) {
 514                 ops = DTO_INDEX_INSERT;
 515
 516                 goto iget;
 517         }
 518
 519         rc = osd_oi_lookup(info, dev, fid, lid2,
 520                 (val == SCRUB_NEXT_OSTOBJ ||
 521                  val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 522         if (rc != 0) {
 523                 if (rc == -ENOENT)
 524                         ops = DTO_INDEX_INSERT;
 525                 else if (rc != -ESTALE)
 526                         GOTO(out, rc);
 527
 528 iget:
 529                 if (inode == NULL) {
 530                         inode = osd_iget(info, dev, lid);
 531                         if (IS_ERR(inode)) {
 532                                 rc = PTR_ERR(inode);
 533                                 /* Someone removed the inode. */
 534                                 if (rc == -ENOENT || rc == -ESTALE)
 535                                         rc = 0;
 536                                 GOTO(out, rc);
 537                         }
 538                 }
 539
 540                 if (!scrub->os_partial_scan)
 541                         scrub->os_full_speed = 1;
 542
 543                 idx = osd_oi_fid2idx(dev, fid);
 544                 switch (val) {
 545                 case SCRUB_NEXT_NOLMA:
 546                         sf->sf_flags |= SF_UPGRADE;
 547                         if (!(sf->sf_param & SP_DRYRUN)) {
 548                                 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
 549                                 if (rc != 0)
 550                                         GOTO(out, rc);
 551                         }
 552
 553                         if (!(sf->sf_flags & SF_INCONSISTENT))
 554                                 dev->od_igif_inoi = 0;
 555                         break;
 556                 case SCRUB_NEXT_OSTOBJ:
 557                         sf->sf_flags |= SF_INCONSISTENT;
 558                 case SCRUB_NEXT_OSTOBJ_OLD:
 559                         break;
 560                 default:
 561                         sf->sf_flags |= SF_RECREATED;
 562                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
 563                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
 564                         break;
 565                 }
 566         } else if (osd_id_eq(lid, lid2)) {
 567                 if (converted)
 568                         sf->sf_items_updated++;
 569
 570                 GOTO(out, rc = 0);
 571         } else {
 572                 if (!scrub->os_partial_scan)
 573                         scrub->os_full_speed = 1;
 574
 575                 sf->sf_flags |= SF_INCONSISTENT;
 576
 577                 /* XXX: If the device is restored from file-level backup, then
 578                  *      some IGIFs may have been already in OI files, and some
 579                  *      may be not yet. Means upgrading from 1.8 may be partly
 580                  *      processed, but some clients may hold some immobilized
 581                  *      IGIFs, and use them to access related objects. Under
 582                  *      such case, OSD does not know whether an given IGIF has
 583                  *      been processed or to be processed, and it also cannot
 584                  *      generate local ino#/gen# directly from the immobilized
 585                  *      IGIF because of the backup/restore. Then force OSD to
 586                  *      lookup the given IGIF in OI files, and if no entry,
 587                  *      then ask the client to retry after upgrading completed.
 588                  *      No better choice. */
 589                 dev->od_igif_inoi = 1;
 590         }
 591
 592         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
 593                         (val == SCRUB_NEXT_OSTOBJ ||
 594                          val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
 595         if (rc == 0) {
 596                 if (scrub->os_in_prior)
 597                         sf->sf_items_updated_prior++;
 598                 else
 599                         sf->sf_items_updated++;
 600         }
 601
 602         GOTO(out, rc);
 603
 604 out:
 605         if (rc < 0) {
 606                 sf->sf_items_failed++;
 607                 if (sf->sf_pos_first_inconsistent == 0 ||
 608                     sf->sf_pos_first_inconsistent > lid->oii_ino)
 609                         sf->sf_pos_first_inconsistent = lid->oii_ino;
 610         } else {
 611                 rc = 0;
 612         }
 613
 614         /* There may be conflict unlink during the OI scrub,
 615          * if happend, then remove the new added OI mapping. */
 616         if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
 617             unlikely(inode->i_nlink == 0))
 618                 osd_scrub_refresh_mapping(info, dev, fid, lid,
 619                                 DTO_INDEX_DELETE, false,
 620                                 (val == SCRUB_NEXT_OSTOBJ ||
 621                                  val == SCRUB_NEXT_OSTOBJ_OLD) ?
 622                                 OI_KNOWN_ON_OST : 0);
 623         up_write(&scrub->os_rwsem);
 624
 625         if (inode != NULL && !IS_ERR(inode))
 626                 iput(inode);
 627
 628         if (oii != NULL) {
 629                 LASSERT(!list_empty(&oii->oii_list));
 630
 631                 spin_lock(&scrub->os_lock);
 632                 list_del_init(&oii->oii_list);
 633                 spin_unlock(&scrub->os_lock);
 634                 OBD_FREE_PTR(oii);
 635         }
 636         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
 637 }
 638
 639 /* OI scrub APIs */
 640
 641 static int osd_scrub_prep(struct osd_device *dev)
 642 {
 643         struct osd_scrub     *scrub  = &dev->od_scrub;
 644         struct ptlrpc_thread *thread = &scrub->os_thread;
 645         struct scrub_file    *sf     = &scrub->os_file;
 646         __u32                 flags  = scrub->os_start_flags;
 647         int                   rc;
 648         bool                  drop_dryrun = false;
 649         ENTRY;
 650
 651         CDEBUG(D_LFSCK, "%.16s: OI scrub prep, flags = 0x%x\n",
 652                osd_scrub2name(scrub), flags);
 653
 654         down_write(&scrub->os_rwsem);
 655         if (flags & SS_SET_FAILOUT)
 656                 sf->sf_param |= SP_FAILOUT;
 657         else if (flags & SS_CLEAR_FAILOUT)
 658                 sf->sf_param &= ~SP_FAILOUT;
 659
 660         if (flags & SS_SET_DRYRUN) {
 661                 sf->sf_param |= SP_DRYRUN;
 662         } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
 663                 sf->sf_param &= ~SP_DRYRUN;
 664                 drop_dryrun = true;
 665         }
 666
 667         if (flags & SS_RESET)
 668                 osd_scrub_file_reset(scrub,
 669                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
 670
 671         if (flags & SS_AUTO_FULL) {
 672                 scrub->os_full_speed = 1;
 673                 scrub->os_partial_scan = 0;
 674                 sf->sf_flags |= SF_AUTO;
 675         } else if (flags & SS_AUTO_PARTIAL) {
 676                 scrub->os_full_speed = 0;
 677                 scrub->os_partial_scan = 1;
 678                 sf->sf_flags |= SF_AUTO;
 679         } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
 680                                    SF_UPGRADE)) {
 681                 scrub->os_full_speed = 1;
 682                 scrub->os_partial_scan = 0;
 683         } else {
 684                 scrub->os_full_speed = 0;
 685                 scrub->os_partial_scan = 0;
 686         }
 687
 688         spin_lock(&scrub->os_lock);
 689         scrub->os_in_prior = 0;
 690         scrub->os_waiting = 0;
 691         scrub->os_paused = 0;
 692         scrub->os_in_join = 0;
 693         scrub->os_full_scrub = 0;
 694         spin_unlock(&scrub->os_lock);
 695         scrub->os_new_checked = 0;
 696         if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
 697                 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
 698         else if (sf->sf_pos_last_checkpoint != 0)
 699                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
 700         else
 701                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 702
 703         scrub->os_pos_current = sf->sf_pos_latest_start;
 704         sf->sf_status = SS_SCANNING;
 705         sf->sf_time_latest_start = cfs_time_current_sec();
 706         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
 707         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
 708         rc = osd_scrub_file_store(scrub);
 709         if (rc == 0) {
 710                 spin_lock(&scrub->os_lock);
 711                 thread_set_flags(thread, SVC_RUNNING);
 712                 spin_unlock(&scrub->os_lock);
 713                 wake_up_all(&thread->t_ctl_waitq);
 714         }
 715         up_write(&scrub->os_rwsem);
 716
 717         RETURN(rc);
 718 }
 719
 720 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
 721 {
 722         struct scrub_file *sf = &scrub->os_file;
 723         int                rc;
 724
 725         if (likely(cfs_time_before(cfs_time_current(),
 726                                    scrub->os_time_next_checkpoint) ||
 727                    scrub->os_new_checked == 0))
 728                 return 0;
 729
 730         down_write(&scrub->os_rwsem);
 731         sf->sf_items_checked += scrub->os_new_checked;
 732         scrub->os_new_checked = 0;
 733         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 734         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 735         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 736                                             scrub->os_time_last_checkpoint);
 737         rc = osd_scrub_file_store(scrub);
 738         up_write(&scrub->os_rwsem);
 739
 740         return rc;
 741 }
 742
 743 static void osd_scrub_post(struct osd_scrub *scrub, int result)
 744 {
 745         struct scrub_file *sf = &scrub->os_file;
 746         ENTRY;
 747
 748         CDEBUG(D_LFSCK, "%.16s: OI scrub post, result = %d\n",
 749                osd_scrub2name(scrub), result);
 750
 751         down_write(&scrub->os_rwsem);
 752         spin_lock(&scrub->os_lock);
 753         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
 754         spin_unlock(&scrub->os_lock);
 755         if (scrub->os_new_checked > 0) {
 756                 sf->sf_items_checked += scrub->os_new_checked;
 757                 scrub->os_new_checked = 0;
 758                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
 759         }
 760         sf->sf_time_last_checkpoint = cfs_time_current_sec();
 761         if (result > 0) {
 762                 struct osd_device *dev =
 763                         container_of0(scrub, struct osd_device, od_scrub);
 764
 765                 dev->od_igif_inoi = 1;
 766                 dev->od_check_ff = 0;
 767                 sf->sf_status = SS_COMPLETED;
 768                 if (!(sf->sf_param & SP_DRYRUN)) {
 769                         memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
 770                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
 771                                           SF_UPGRADE | SF_AUTO);
 772                 }
 773                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
 774                 sf->sf_success_count++;
 775         } else if (result == 0) {
 776                 if (scrub->os_paused)
 777                         sf->sf_status = SS_PAUSED;
 778                 else
 779                         sf->sf_status = SS_STOPPED;
 780         } else {
 781                 sf->sf_status = SS_FAILED;
 782         }
 783         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
 784                                             scrub->os_time_last_checkpoint);
 785         result = osd_scrub_file_store(scrub);
 786         up_write(&scrub->os_rwsem);
 787
 788         EXIT;
 789 }
 790
 791 /* iteration engine */
 792
 793 struct osd_iit_param {
 794         struct super_block *sb;
 795         struct buffer_head *bitmap;
 796         ldiskfs_group_t bg;
 797         __u32 gbase;
 798         __u32 offset;
 799 };
 800
 801 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
 802                                    struct osd_device *dev,
 803                                    struct osd_iit_param *param,
 804                                    struct osd_idmap_cache **oic,
 805                                    const bool noslot);
 806
 807 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
 808                                    struct osd_device *dev,
 809                                    struct osd_iit_param *param,
 810                                    struct osd_idmap_cache *oic,
 811                                    bool *noslot, int rc);
 812
 813 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
 814 {
 815         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
 816                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
 817         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
 818                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
 819                 return SCRUB_NEXT_BREAK;
 820         } else {
 821                 *pos = param->gbase + param->offset;
 822                 return 0;
 823         }
 824 }
 825
 826 /**
 827  * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
 828  * \retval 0: FID-on-MDT
 829  */
 830 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
 831                                       struct osd_device *dev,
 832                                       struct lu_fid *fid)
 833 {
 834         /* XXX: The initial OI scrub will scan the top level /O to generate
 835          *      a small local FLDB according to the <seq>. If the given FID
 836          *      is in the local FLDB, then it is FID-on-OST; otherwise it's
 837          *      quite possible for FID-on-MDT. */
 838         if (dev->od_is_ost)
 839                 return SCRUB_NEXT_OSTOBJ_OLD;
 840         else
 841                 return 0;
 842 }
 843
 844 static int osd_scrub_get_fid(struct osd_thread_info *info,
 845                              struct osd_device *dev, struct inode *inode,
 846                              struct lu_fid *fid, bool scrub)
 847 {
 848         struct lustre_mdt_attrs *lma     = &info->oti_mdt_attrs;
 849         int                      rc;
 850         bool                     has_lma = false;
 851
 852         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 853         if (rc == 0) {
 854                 has_lma = true;
 855                 if (lma->lma_compat & LMAC_NOT_IN_OI ||
 856                     lma->lma_incompat & LMAI_AGENT)
 857                         return SCRUB_NEXT_CONTINUE;
 858
 859                 *fid = lma->lma_self_fid;
 860                 if (!scrub)
 861                         return 0;
 862
 863                 if (lma->lma_compat & LMAC_FID_ON_OST)
 864                         return SCRUB_NEXT_OSTOBJ;
 865
 866                 if (fid_is_idif(fid))
 867                         return SCRUB_NEXT_OSTOBJ_OLD;
 868
 869                 /* For local object. */
 870                 if (fid_is_internal(fid))
 871                         return 0;
 872
 873                 /* For external visible MDT-object with non-normal FID. */
 874                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
 875                         return 0;
 876
 877                 /* For the object with normal FID, it may be MDT-object,
 878                  * or may be 2.4 OST-object, need further distinguish.
 879                  * Fall through to next section. */
 880         }
 881
 882         if (rc == -ENODATA || rc == 0) {
 883                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
 884                 if (rc == 0) {
 885                         if (scrub)
 886                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
 887                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 888                         return rc;
 889                 }
 890
 891                 if (rc > 0) {
 892                         if (!has_lma)
 893                                 /* It is FID-on-OST, but we do not know how
 894                                  * to generate its FID, ignore it directly. */
 895                                 rc = SCRUB_NEXT_CONTINUE;
 896                         else
 897                                 /* It is 2.4 OST-object. */
 898                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
 899                         return rc;
 900                 }
 901
 902                 if (rc != -ENODATA)
 903                         return rc;
 904
 905                 if (!has_lma) {
 906                         if (dev->od_scrub.os_convert_igif) {
 907                                 lu_igif_build(fid, inode->i_ino,
 908                                               inode->i_generation);
 909                                 if (scrub)
 910                                         rc = SCRUB_NEXT_NOLMA;
 911                                 else
 912                                         rc = 0;
 913                         } else {
 914                                 /* It may be FID-on-OST, or may be FID for
 915                                  * non-MDT0, anyway, we do not know how to
 916                                  * generate its FID, ignore it directly. */
 917                                 rc = SCRUB_NEXT_CONTINUE;
 918                         }
 919                         return rc;
 920                 }
 921
 922                 /* For OI scrub case only: the object has LMA but has no ff
 923                  * (or ff crashed). It may be MDT-object, may be OST-object
 924                  * with crashed ff. The last check is local FLDB. */
 925                 rc = osd_scrub_check_local_fldb(info, dev, fid);
 926         }
 927
 928         return rc;
 929 }
 930
 931 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
 932                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
 933                         struct super_block *sb, bool scrub)
 934 {
 935         struct inode *inode;
 936         int           rc;
 937         ENTRY;
 938
 939         /* Not handle the backend root object and agent parent object.
 940          * They are neither visible to namespace nor have OI mappings. */
 941         if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
 942                      pos == osd_remote_parent_ino(dev)))
 943                 RETURN(SCRUB_NEXT_CONTINUE);
 944
 945         osd_id_gen(lid, pos, OSD_OII_NOGEN);
 946         inode = osd_iget(info, dev, lid);
 947         if (IS_ERR(inode)) {
 948                 rc = PTR_ERR(inode);
 949                 /* The inode may be removed after bitmap searching, or the
 950                  * file is new created without inode initialized yet. */
 951                 if (rc == -ENOENT || rc == -ESTALE)
 952                         RETURN(SCRUB_NEXT_CONTINUE);
 953
 954                 CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: "
 955                        "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name,
 956                        pos, rc);
 957                 RETURN(rc);
 958         }
 959
 960         if (scrub &&
 961             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
 962                 /* Only skip it for the first OI scrub accessing. */
 963                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
 964                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
 965         }
 966
 967         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
 968
 969         GOTO(put, rc);
 970
 971 put:
 972         iput(inode);
 973         return rc;
 974 }
 975
 976 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
 977                           struct osd_iit_param *param,
 978                           struct osd_idmap_cache **oic, const bool noslot)
 979 {
 980         struct osd_scrub     *scrub  = &dev->od_scrub;
 981         struct ptlrpc_thread *thread = &scrub->os_thread;
 982         struct lu_fid        *fid;
 983         struct osd_inode_id  *lid;
 984         int                   rc;
 985
 986         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
 987                 struct l_wait_info lwi;
 988
 989                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
 990                 if (likely(lwi.lwi_timeout > 0))
 991                         l_wait_event(thread->t_ctl_waitq,
 992                                 !list_empty(&scrub->os_inconsistent_items) ||
 993                                 !thread_is_running(thread),
 994                                 &lwi);
 995         }
 996
 997         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
 998                 spin_lock(&scrub->os_lock);
 999                 thread_set_flags(thread, SVC_STOPPING);
1000                 spin_unlock(&scrub->os_lock);
1001                 return SCRUB_NEXT_CRASH;
1002         }
1003
1004         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
1005                 return SCRUB_NEXT_FATAL;
1006
1007         if (unlikely(!thread_is_running(thread)))
1008                 return SCRUB_NEXT_EXIT;
1009
1010         if (!list_empty(&scrub->os_inconsistent_items)) {
1011                 struct osd_inconsistent_item *oii;
1012
1013                 oii = list_entry(scrub->os_inconsistent_items.next,
1014                                  struct osd_inconsistent_item, oii_list);
1015                 *oic = &oii->oii_cache;
1016                 scrub->os_in_prior = 1;
1017                 return 0;
1018         }
1019
1020         if (noslot)
1021                 return SCRUB_NEXT_WAIT;
1022
1023         rc = osd_iit_next(param, &scrub->os_pos_current);
1024         if (rc != 0)
1025                 return rc;
1026
1027         *oic = &scrub->os_oic;
1028         fid = &(*oic)->oic_fid;
1029         lid = &(*oic)->oic_lid;
1030         rc = osd_iit_iget(info, dev, fid, lid,
1031                           scrub->os_pos_current, param->sb, true);
1032         return rc;
1033 }
1034
1035 static int osd_preload_next(struct osd_thread_info *info,
1036                             struct osd_device *dev, struct osd_iit_param *param,
1037                             struct osd_idmap_cache **oic, const bool noslot)
1038 {
1039         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
1040         struct osd_scrub        *scrub;
1041         struct ptlrpc_thread    *thread;
1042         int                      rc;
1043
1044         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1045         if (rc != 0)
1046                 return rc;
1047
1048         scrub = &dev->od_scrub;
1049         thread = &scrub->os_thread;
1050         if (thread_is_running(thread) &&
1051             ooc->ooc_pos_preload >= scrub->os_pos_current)
1052                 return SCRUB_NEXT_EXIT;
1053
1054         rc = osd_iit_iget(info, dev,
1055                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1056                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1057                           ooc->ooc_pos_preload, param->sb, false);
1058         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1059          * ignore the failure, so it still need to skip the inode next time. */
1060         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1061         return rc;
1062 }
1063
1064 static inline int
1065 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1066 {
1067         spin_lock(&scrub->os_lock);
1068         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1069             !list_empty(&scrub->os_inconsistent_items) ||
1070             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1071                 scrub->os_waiting = 0;
1072         else
1073                 scrub->os_waiting = 1;
1074         spin_unlock(&scrub->os_lock);
1075
1076         return !scrub->os_waiting;
1077 }
1078
1079 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1080                           struct osd_iit_param *param,
1081                           struct osd_idmap_cache *oic, bool *noslot, int rc)
1082 {
1083         struct l_wait_info       lwi    = { 0 };
1084         struct osd_scrub        *scrub  = &dev->od_scrub;
1085         struct scrub_file       *sf     = &scrub->os_file;
1086         struct ptlrpc_thread    *thread = &scrub->os_thread;
1087         struct osd_otable_it    *it     = dev->od_otable_it;
1088         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
1089
1090         switch (rc) {
1091         case SCRUB_NEXT_CONTINUE:
1092                 goto next;
1093         case SCRUB_NEXT_WAIT:
1094                 goto wait;
1095         case SCRUB_NEXT_NOSCRUB:
1096                 down_write(&scrub->os_rwsem);
1097                 scrub->os_new_checked++;
1098                 sf->sf_items_noscrub++;
1099                 up_write(&scrub->os_rwsem);
1100                 goto next;
1101         }
1102
1103         rc = osd_scrub_check_update(info, dev, oic, rc);
1104         if (rc != 0)
1105                 return rc;
1106
1107         rc = osd_scrub_checkpoint(scrub);
1108         if (rc != 0) {
1109                 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1110                        "rc = %d\n", osd_scrub2name(scrub),
1111                        scrub->os_pos_current, rc);
1112                 /* Continue, as long as the scrub itself can go ahead. */
1113         }
1114
1115         if (scrub->os_in_prior) {
1116                 scrub->os_in_prior = 0;
1117                 return 0;
1118         }
1119
1120 next:
1121         scrub->os_pos_current = param->gbase + ++(param->offset);
1122
1123 wait:
1124         if (it != NULL && it->ooi_waiting && ooc != NULL &&
1125             ooc->ooc_pos_preload < scrub->os_pos_current) {
1126                 spin_lock(&scrub->os_lock);
1127                 it->ooi_waiting = 0;
1128                 wake_up_all(&thread->t_ctl_waitq);
1129                 spin_unlock(&scrub->os_lock);
1130         }
1131
1132         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1133                 return 0;
1134
1135         if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1136                 *noslot = false;
1137                 return 0;
1138         }
1139
1140         if (it != NULL)
1141                 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1142                              &lwi);
1143
1144         if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1145                 *noslot = false;
1146         else
1147                 *noslot = true;
1148         return 0;
1149 }
1150
1151 static int osd_preload_exec(struct osd_thread_info *info,
1152                             struct osd_device *dev, struct osd_iit_param *param,
1153                             struct osd_idmap_cache *oic, bool *noslot, int rc)
1154 {
1155         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1156
1157         if (rc == 0) {
1158                 ooc->ooc_cached_items++;
1159                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1160                                         ~OSD_OTABLE_IT_CACHE_MASK;
1161         }
1162         return rc > 0 ? 0 : rc;
1163 }
1164
1165 #define SCRUB_IT_ALL    1
1166 #define SCRUB_IT_CRASH  2
1167
1168 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1169                            bool inconsistent)
1170 {
1171         struct osd_scrub     *scrub  = &dev->od_scrub;
1172         struct ptlrpc_thread *thread = &scrub->os_thread;
1173         struct scrub_file    *sf     = &scrub->os_file;
1174         int                   rc;
1175         ENTRY;
1176
1177         LASSERT(!(flags & SS_AUTO_PARTIAL));
1178
1179         down_write(&scrub->os_rwsem);
1180         scrub->os_in_join = 1;
1181         if (flags & SS_SET_FAILOUT)
1182                 sf->sf_param |= SP_FAILOUT;
1183         else if (flags & SS_CLEAR_FAILOUT)
1184                 sf->sf_param &= ~SP_FAILOUT;
1185
1186         if (flags & SS_SET_DRYRUN)
1187                 sf->sf_param |= SP_DRYRUN;
1188         else if (flags & SS_CLEAR_DRYRUN)
1189                 sf->sf_param &= ~SP_DRYRUN;
1190
1191         if (flags & SS_RESET) {
1192                 osd_scrub_file_reset(scrub,
1193                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1194                         inconsistent ? SF_INCONSISTENT : 0);
1195                 sf->sf_status = SS_SCANNING;
1196         }
1197
1198         if (flags & SS_AUTO_FULL) {
1199                 sf->sf_flags |= SF_AUTO;
1200                 scrub->os_full_speed = 1;
1201         }
1202
1203         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1204                 scrub->os_full_speed = 1;
1205         else
1206                 scrub->os_full_speed = 0;
1207
1208         scrub->os_new_checked = 0;
1209         if (sf->sf_pos_last_checkpoint != 0)
1210                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1211         else
1212                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1213
1214         scrub->os_pos_current = sf->sf_pos_latest_start;
1215         sf->sf_time_latest_start = cfs_time_current_sec();
1216         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1217         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1218         rc = osd_scrub_file_store(scrub);
1219         if (rc != 0)
1220                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
1221                        "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1222
1223         spin_lock(&scrub->os_lock);
1224         scrub->os_waiting = 0;
1225         scrub->os_paused = 0;
1226         scrub->os_partial_scan = 0;
1227         scrub->os_in_join = 0;
1228         scrub->os_full_scrub = 0;
1229         spin_unlock(&scrub->os_lock);
1230         wake_up_all(&thread->t_ctl_waitq);
1231         up_write(&scrub->os_rwsem);
1232
1233         EXIT;
1234 }
1235
1236 static int osd_inode_iteration(struct osd_thread_info *info,
1237                                struct osd_device *dev, __u32 max, bool preload)
1238 {
1239         struct osd_scrub     *scrub  = &dev->od_scrub;
1240         struct ptlrpc_thread *thread = &scrub->os_thread;
1241         struct scrub_file    *sf     = &scrub->os_file;
1242         osd_iit_next_policy   next;
1243         osd_iit_exec_policy   exec;
1244         __u32                *pos;
1245         __u32                *count;
1246         struct osd_iit_param  param  = { NULL };
1247         struct l_wait_info    lwi    = { 0 };
1248         __u32                 limit;
1249         int                   rc;
1250         bool                  noslot = true;
1251         ENTRY;
1252
1253         param.sb = osd_sb(dev);
1254         if (preload)
1255                 goto full;
1256
1257         while (scrub->os_partial_scan && !scrub->os_in_join) {
1258                 struct osd_idmap_cache *oic = NULL;
1259
1260                 rc = osd_scrub_next(info, dev, &param, &oic, noslot);
1261                 switch (rc) {
1262                 case SCRUB_NEXT_EXIT:
1263                         RETURN(0);
1264                 case SCRUB_NEXT_CRASH:
1265                         RETURN(SCRUB_IT_CRASH);
1266                 case SCRUB_NEXT_FATAL:
1267                         RETURN(-EINVAL);
1268                 case SCRUB_NEXT_WAIT: {
1269                         struct kstatfs *ksfs = &info->oti_ksfs;
1270                         __u64 saved_flags;
1271
1272                         if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1273                             unlikely(sf->sf_items_updated_prior == 0))
1274                                 goto wait;
1275
1276                         if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1277                             scrub->os_full_scrub) {
1278                                 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1279                                                true);
1280                                 goto full;
1281                         }
1282
1283                         rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1284                         if (rc == 0) {
1285                                 __u64 used = ksfs->f_files - ksfs->f_ffree;
1286
1287                                 do_div(used, sf->sf_items_updated_prior);
1288                                 /* If we hit too much inconsistent OI
1289                                  * mappings during the partial scan,
1290                                  * then scan the device completely. */
1291                                 if (used < dev->od_full_scrub_ratio) {
1292                                         osd_scrub_join(dev,
1293                                                 SS_AUTO_FULL | SS_RESET, true);
1294                                         goto full;
1295                                 }
1296                         }
1297
1298 wait:
1299                         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1300                             cfs_fail_val > 0)
1301                                 continue;
1302
1303                         saved_flags = sf->sf_flags;
1304                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1305                                           SF_UPGRADE | SF_AUTO);
1306                         sf->sf_status = SS_COMPLETED;
1307                         l_wait_event(thread->t_ctl_waitq,
1308                                      !thread_is_running(thread) ||
1309                                      !scrub->os_partial_scan ||
1310                                      scrub->os_in_join ||
1311                                      !list_empty(&scrub->os_inconsistent_items),
1312                                      &lwi);
1313                         sf->sf_flags = saved_flags;
1314                         sf->sf_status = SS_SCANNING;
1315
1316                         if (unlikely(!thread_is_running(thread)))
1317                                 RETURN(0);
1318
1319                         if (!scrub->os_partial_scan || scrub->os_in_join)
1320                                 goto full;
1321
1322                         continue;
1323                 }
1324                 default:
1325                         LASSERTF(rc == 0, "rc = %d\n", rc);
1326
1327                         osd_scrub_exec(info, dev, &param, oic, &noslot, rc);
1328                         break;
1329                 }
1330         }
1331
1332 full:
1333         if (!preload) {
1334                 l_wait_event(thread->t_ctl_waitq,
1335                              !thread_is_running(thread) || !scrub->os_in_join,
1336                              &lwi);
1337
1338                 if (unlikely(!thread_is_running(thread)))
1339                         RETURN(0);
1340         }
1341
1342         noslot = false;
1343         if (!preload) {
1344                 next = osd_scrub_next;
1345                 exec = osd_scrub_exec;
1346                 pos = &scrub->os_pos_current;
1347                 count = &scrub->os_new_checked;
1348         } else {
1349                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1350
1351                 next = osd_preload_next;
1352                 exec = osd_preload_exec;
1353                 pos = &ooc->ooc_pos_preload;
1354                 count = &ooc->ooc_cached_items;
1355         }
1356         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1357
1358         while (*pos <= limit && *count < max) {
1359                 struct osd_idmap_cache *oic = NULL;
1360                 struct ldiskfs_group_desc *desc;
1361
1362                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1363                 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1364                 if (desc == NULL)
1365                         RETURN(-EIO);
1366
1367                 ldiskfs_lock_group(param.sb, param.bg);
1368                 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1369                         ldiskfs_unlock_group(param.sb, param.bg);
1370                         *pos = 1 + (param.bg + 1) *
1371                                 LDISKFS_INODES_PER_GROUP(param.sb);
1372                         continue;
1373                 }
1374                 ldiskfs_unlock_group(param.sb, param.bg);
1375
1376                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1377                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1378                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1379                 if (param.bitmap == NULL) {
1380                         CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1381                                "scrub will stop, urgent mode\n",
1382                                osd_scrub2name(scrub), (__u32)param.bg);
1383                         RETURN(-EIO);
1384                 }
1385
1386                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1387                        *count < max) {
1388                         if (param.offset +
1389                                 ldiskfs_itable_unused_count(param.sb, desc) >
1390                             LDISKFS_INODES_PER_GROUP(param.sb))
1391                                 goto next_group;
1392
1393                         rc = next(info, dev, &param, &oic, noslot);
1394                         switch (rc) {
1395                         case SCRUB_NEXT_BREAK:
1396                                 goto next_group;
1397                         case SCRUB_NEXT_EXIT:
1398                                 brelse(param.bitmap);
1399                                 RETURN(0);
1400                         case SCRUB_NEXT_CRASH:
1401                                 brelse(param.bitmap);
1402                                 RETURN(SCRUB_IT_CRASH);
1403                         case SCRUB_NEXT_FATAL:
1404                                 brelse(param.bitmap);
1405                                 RETURN(-EINVAL);
1406                         }
1407
1408                         rc = exec(info, dev, &param, oic, &noslot, rc);
1409                         if (rc != 0) {
1410                                 brelse(param.bitmap);
1411                                 RETURN(rc);
1412                         }
1413                 }
1414
1415 next_group:
1416                 brelse(param.bitmap);
1417         }
1418
1419         if (*pos > limit)
1420                 RETURN(SCRUB_IT_ALL);
1421         RETURN(0);
1422 }
1423
1424 static int osd_otable_it_preload(const struct lu_env *env,
1425                                  struct osd_otable_it *it)
1426 {
1427         struct osd_device       *dev   = it->ooi_dev;
1428         struct osd_scrub        *scrub = &dev->od_scrub;
1429         struct osd_otable_cache *ooc   = &it->ooi_cache;
1430         int                      rc;
1431         ENTRY;
1432
1433         rc = osd_inode_iteration(osd_oti_get(env), dev,
1434                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1435         if (rc == SCRUB_IT_ALL)
1436                 it->ooi_all_cached = 1;
1437
1438         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1439                 spin_lock(&scrub->os_lock);
1440                 scrub->os_waiting = 0;
1441                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1442                 spin_unlock(&scrub->os_lock);
1443         }
1444
1445         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1446 }
1447
1448 static int osd_scrub_main(void *args)
1449 {
1450         struct lu_env         env;
1451         struct osd_device    *dev    = (struct osd_device *)args;
1452         struct osd_scrub     *scrub  = &dev->od_scrub;
1453         struct ptlrpc_thread *thread = &scrub->os_thread;
1454         int                   rc;
1455         ENTRY;
1456
1457         rc = lu_env_init(&env, LCT_LOCAL);
1458         if (rc != 0) {
1459                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1460                        osd_scrub2name(scrub), rc);
1461                 GOTO(noenv, rc);
1462         }
1463
1464         rc = osd_scrub_prep(dev);
1465         if (rc != 0) {
1466                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1467                        osd_scrub2name(scrub), rc);
1468                 GOTO(out, rc);
1469         }
1470
1471         if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1472                 struct l_wait_info lwi = { 0 };
1473                 struct osd_otable_it *it = dev->od_otable_it;
1474                 struct osd_otable_cache *ooc = &it->ooi_cache;
1475
1476                 l_wait_event(thread->t_ctl_waitq,
1477                              it->ooi_user_ready || !thread_is_running(thread),
1478                              &lwi);
1479                 if (unlikely(!thread_is_running(thread)))
1480                         GOTO(post, rc = 0);
1481
1482                 scrub->os_pos_current = ooc->ooc_pos_preload;
1483         }
1484
1485         CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1486                osd_scrub2name(scrub), scrub->os_start_flags,
1487                scrub->os_pos_current);
1488
1489         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1490         if (unlikely(rc == SCRUB_IT_CRASH))
1491                 GOTO(out, rc = -EINVAL);
1492         GOTO(post, rc);
1493
1494 post:
1495         osd_scrub_post(scrub, rc);
1496         CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1497                osd_scrub2name(scrub), scrub->os_pos_current, rc);
1498
1499 out:
1500         while (!list_empty(&scrub->os_inconsistent_items)) {
1501                 struct osd_inconsistent_item *oii;
1502
1503                 oii = list_entry(scrub->os_inconsistent_items.next,
1504                                      struct osd_inconsistent_item, oii_list);
1505                 list_del_init(&oii->oii_list);
1506                 OBD_FREE_PTR(oii);
1507         }
1508         lu_env_fini(&env);
1509
1510 noenv:
1511         spin_lock(&scrub->os_lock);
1512         thread_set_flags(thread, SVC_STOPPED);
1513         wake_up_all(&thread->t_ctl_waitq);
1514         spin_unlock(&scrub->os_lock);
1515         return rc;
1516 }
1517
1518 /* initial OI scrub */
1519
1520 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1521                          struct dentry *, filldir_t filldir);
1522
1523 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1524                                loff_t offset, __u64 ino, unsigned d_type);
1525 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1526                            loff_t offset, __u64 ino, unsigned d_type);
1527 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1528                            loff_t offset, __u64 ino, unsigned d_type);
1529
1530 static int
1531 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1532                      struct dentry *dentry, filldir_t filldir);
1533 static int
1534 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1535                   struct dentry *dentry, filldir_t filldir);
1536
1537 static int
1538 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1539                      struct dentry *dentry, filldir_t filldir);
1540
1541 enum osd_lf_flags {
1542         OLF_SCAN_SUBITEMS       = 0x0001,
1543         OLF_HIDE_FID            = 0x0002,
1544         OLF_SHOW_NAME           = 0x0004,
1545         OLF_NO_OI               = 0x0008,
1546 };
1547
1548 struct osd_lf_map {
1549         char            *olm_name;
1550         struct lu_fid    olm_fid;
1551         __u16            olm_flags;
1552         __u16            olm_namelen;
1553         scandir_t        olm_scandir;
1554         filldir_t        olm_filldir;
1555 };
1556
1557 /* Add the new introduced local files in the list in the future. */
1558 static const struct osd_lf_map osd_lf_maps[] = {
1559         /* CATALOGS */
1560         { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1561                 sizeof(CATLIST) - 1, NULL, NULL },
1562
1563         /* CONFIGS */
1564         { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1565                 OLF_SCAN_SUBITEMS, sizeof(MOUNT_CONFIGS_DIR) - 1,
1566                 osd_ios_general_scan, osd_ios_varfid_fill },
1567
1568         /* NIDTBL_VERSIONS */
1569         { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1570                 sizeof(MGS_NIDTBL_DIR) - 1, osd_ios_general_scan,
1571                 osd_ios_varfid_fill },
1572
1573         /* PENDING */
1574         { "PENDING", { 0, 0, 0 }, 0, sizeof("PENDING") - 1, NULL, NULL },
1575
1576         /* ROOT */
1577         { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 },
1578                 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, sizeof("ROOT") - 1,
1579                 osd_ios_ROOT_scan, NULL },
1580
1581         /* changelog_catalog */
1582         { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, sizeof(CHANGELOG_CATALOG) - 1,
1583                 NULL, NULL },
1584
1585         /* changelog_users */
1586         { CHANGELOG_USERS, { 0, 0, 0 }, 0, sizeof(CHANGELOG_USERS) - 1,
1587                 NULL, NULL },
1588
1589         /* fld */
1590         { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1591                 sizeof("fld") - 1, NULL, NULL },
1592
1593         /* last_rcvd */
1594         { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1595                 sizeof(LAST_RCVD) - 1, NULL, NULL },
1596
1597         /* lov_objid */
1598         { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1599                 sizeof(LOV_OBJID) - 1, NULL, NULL },
1600
1601         /* lov_objseq */
1602         { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1603                 OLF_SHOW_NAME, sizeof(LOV_OBJSEQ) - 1, NULL, NULL },
1604
1605         /* quota_master */
1606         { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QMT_DIR) - 1,
1607                 osd_ios_general_scan, osd_ios_varfid_fill },
1608
1609         /* quota_slave */
1610         { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QSD_DIR) - 1,
1611                 osd_ios_general_scan, osd_ios_varfid_fill },
1612
1613         /* seq_ctl */
1614         { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1615                 OLF_SHOW_NAME, sizeof("seq_ctl") - 1, NULL, NULL },
1616
1617         /* seq_srv */
1618         { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1619                 OLF_SHOW_NAME, sizeof("seq_srv") - 1, NULL, NULL },
1620
1621         /* health_check */
1622         { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1623                 OLF_SHOW_NAME, sizeof(HEALTH_CHECK) - 1, NULL, NULL },
1624
1625         /* LFSCK */
1626         { LFSCK_DIR, { 0, 0, 0 }, 0, sizeof(LFSCK_DIR) - 1,
1627                 osd_ios_general_scan, osd_ios_varfid_fill },
1628
1629         /* lfsck_bookmark */
1630         { LFSCK_BOOKMARK, { 0, 0, 0 }, 0, sizeof(LFSCK_BOOKMARK) - 1,
1631                 NULL, NULL },
1632
1633         /* lfsck_layout */
1634         { LFSCK_LAYOUT, { 0, 0, 0 }, 0, sizeof(LFSCK_LAYOUT) - 1,
1635                 NULL, NULL },
1636
1637         /* lfsck_namespace */
1638         { LFSCK_NAMESPACE, { 0, 0, 0 }, 0, sizeof(LFSCK_NAMESPACE) - 1,
1639                 NULL, NULL },
1640
1641         /* OBJECTS, upgrade from old device */
1642         { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(OBJECTS) - 1,
1643                 osd_ios_OBJECTS_scan, NULL },
1644
1645         /* lquota_v2.user, upgrade from old device */
1646         { "lquota_v2.user", { 0, 0, 0 }, 0, sizeof("lquota_v2.user") - 1,
1647                 NULL, NULL },
1648
1649         /* lquota_v2.group, upgrade from old device */
1650         { "lquota_v2.group", { 0, 0, 0 }, 0, sizeof("lquota_v2.group") - 1,
1651                 NULL, NULL },
1652
1653         /* LAST_GROUP, upgrade from old device */
1654         { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1655                 OLF_SHOW_NAME, sizeof("LAST_GROUP") - 1, NULL, NULL },
1656
1657         /* lost+found */
1658         { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 },
1659                 OLF_SCAN_SUBITEMS, sizeof("lost+found") - 1,
1660                 osd_ios_general_scan, osd_ios_lf_fill },
1661
1662         { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1663 };
1664
1665 /* Add the new introduced files under .lustre/ in the list in the future. */
1666 static const struct osd_lf_map osd_dl_maps[] = {
1667         /* .lustre/fid */
1668         { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1669                 sizeof("fid") - 1, NULL, NULL },
1670
1671         /* .lustre/lost+found */
1672         { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1673                 sizeof("lost+found") - 1, NULL, NULL },
1674
1675         { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1676 };
1677
1678 struct osd_ios_item {
1679         struct list_head oii_list;
1680         struct dentry   *oii_dentry;
1681         scandir_t        oii_scandir;
1682         filldir_t        oii_filldir;
1683 };
1684
1685 struct osd_ios_filldir_buf {
1686 #ifdef HAVE_DIR_CONTEXT
1687         /* please keep it as first member */
1688         struct dir_context       ctx;
1689 #endif
1690         struct osd_thread_info  *oifb_info;
1691         struct osd_device       *oifb_dev;
1692         struct dentry           *oifb_dentry;
1693 };
1694
1695 static inline struct dentry *
1696 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1697 {
1698         struct dentry *dentry;
1699
1700         dentry = ll_lookup_one_len(name, parent, namelen);
1701         if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1702                 dput(dentry);
1703                 return ERR_PTR(-ENOENT);
1704         }
1705
1706         return dentry;
1707 }
1708
1709 static int
1710 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1711                  scandir_t scandir, filldir_t filldir)
1712 {
1713         struct osd_ios_item *item;
1714         ENTRY;
1715
1716         OBD_ALLOC_PTR(item);
1717         if (item == NULL)
1718                 RETURN(-ENOMEM);
1719
1720         INIT_LIST_HEAD(&item->oii_list);
1721         item->oii_dentry = dget(dentry);
1722         item->oii_scandir = scandir;
1723         item->oii_filldir = filldir;
1724         list_add_tail(&item->oii_list, &dev->od_ios_list);
1725
1726         RETURN(0);
1727 }
1728
1729 /**
1730  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1731  *
1732  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1733  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1734  * reference the inode, or fixed if it is missing or references another inode.
1735  */
1736 static int
1737 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1738                  struct inode *inode, const struct lu_fid *fid, int flags)
1739 {
1740         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
1741         struct osd_inode_id     *id     = &info->oti_id;
1742         struct osd_inode_id     *id2    = &info->oti_id2;
1743         struct osd_scrub        *scrub  = &dev->od_scrub;
1744         struct scrub_file       *sf     = &scrub->os_file;
1745         struct lu_fid            tfid;
1746         int                      rc;
1747         ENTRY;
1748
1749         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1750         if (rc != 0 && rc != -ENODATA) {
1751                 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1752                        "rc = %d\n", osd_name(dev), rc);
1753
1754                 RETURN(rc);
1755         }
1756
1757         osd_id_gen(id, inode->i_ino, inode->i_generation);
1758         if (rc == -ENODATA) {
1759                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID)
1760                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1761                 else
1762                         tfid = *fid;
1763                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1764                 if (rc != 0) {
1765                         CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1766                               "scrub: rc = %d\n", osd_name(dev), rc);
1767
1768                         RETURN(rc);
1769                 }
1770         } else {
1771                 if (lma->lma_compat & LMAC_NOT_IN_OI)
1772                         RETURN(0);
1773
1774                 tfid = lma->lma_self_fid;
1775         }
1776
1777         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1778         if (rc != 0) {
1779                 if (rc != -ENOENT)
1780                         RETURN(rc);
1781
1782                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1783                                                DTO_INDEX_INSERT, true, 0);
1784                 if (rc > 0)
1785                         rc = 0;
1786
1787                 RETURN(rc);
1788         }
1789
1790         if (osd_id_eq_strict(id, id2))
1791                 RETURN(0);
1792
1793         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1794                 osd_scrub_file_reset(scrub,
1795                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1796                                      SF_INCONSISTENT);
1797                 rc = osd_scrub_file_store(scrub);
1798                 if (rc != 0)
1799                         RETURN(rc);
1800         }
1801
1802         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1803                                        DTO_INDEX_UPDATE, true, 0);
1804         if (rc > 0)
1805                 rc = 0;
1806
1807         RETURN(rc);
1808 }
1809
1810 /**
1811  * It scans the /lost+found, and for the OST-object (with filter_fid
1812  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1813  */
1814 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1815                            loff_t offset, __u64 ino, unsigned d_type)
1816 {
1817         struct osd_ios_filldir_buf *fill_buf = buf;
1818         struct osd_thread_info     *info     = fill_buf->oifb_info;
1819         struct osd_device          *dev      = fill_buf->oifb_dev;
1820         struct lu_fid              *fid      = &info->oti_fid;
1821         struct osd_scrub           *scrub    = &dev->od_scrub;
1822         struct dentry              *parent   = fill_buf->oifb_dentry;
1823         struct dentry              *child;
1824         struct inode               *dir      = parent->d_inode;
1825         struct inode               *inode;
1826         int                         rc;
1827         ENTRY;
1828
1829         /* skip any '.' started names */
1830         if (name[0] == '.')
1831                 RETURN(0);
1832
1833         scrub->os_lf_scanned++;
1834         child = osd_ios_lookup_one_len(name, parent, namelen);
1835         if (IS_ERR(child)) {
1836                 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1837                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
1838                 RETURN(0);
1839         }
1840
1841         inode = child->d_inode;
1842         if (S_ISDIR(inode->i_mode)) {
1843                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1844                                       osd_ios_lf_fill);
1845                 if (rc != 0)
1846                         CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1847                               "rc = %d\n", osd_name(dev), namelen, name, rc);
1848                 GOTO(put, rc);
1849         }
1850
1851         if (!S_ISREG(inode->i_mode))
1852                 GOTO(put, rc = 0);
1853
1854         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1855         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1856                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1857                 if (rc == 0) {
1858                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1859                                "/lost+found.\n", namelen, name, PFID(fid));
1860                         scrub->os_lf_repaired++;
1861                 } else {
1862                         CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1863                                DFID": rc = %d\n",
1864                                osd_name(dev), namelen, name, PFID(fid), rc);
1865                 }
1866         }
1867
1868         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1869          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1870          *      can process them in furtuer. */
1871
1872         GOTO(put, rc);
1873
1874 put:
1875         if (rc < 0)
1876                 scrub->os_lf_failed++;
1877         dput(child);
1878         /* skip the failure to make the scanning to continue. */
1879         return 0;
1880 }
1881
1882 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1883                                loff_t offset, __u64 ino, unsigned d_type)
1884 {
1885         struct osd_ios_filldir_buf *fill_buf = buf;
1886         struct osd_device          *dev      = fill_buf->oifb_dev;
1887         struct dentry              *child;
1888         int                         rc;
1889         ENTRY;
1890
1891         /* skip any '.' started names */
1892         if (name[0] == '.')
1893                 RETURN(0);
1894
1895         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1896         if (IS_ERR(child))
1897                 RETURN(PTR_ERR(child));
1898
1899         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1900                               NULL, 0);
1901         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1902                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1903                                       osd_ios_varfid_fill);
1904         dput(child);
1905
1906         RETURN(rc);
1907 }
1908
1909 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1910                            loff_t offset, __u64 ino, unsigned d_type)
1911 {
1912         struct osd_ios_filldir_buf *fill_buf = buf;
1913         struct osd_device          *dev      = fill_buf->oifb_dev;
1914         const struct osd_lf_map    *map;
1915         struct dentry              *child;
1916         int                         rc       = 0;
1917         ENTRY;
1918
1919         /* skip any '.' started names */
1920         if (name[0] == '.')
1921                 RETURN(0);
1922
1923         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
1924                 if (map->olm_namelen != namelen)
1925                         continue;
1926
1927                 if (strncmp(map->olm_name, name, namelen) == 0)
1928                         break;
1929         }
1930
1931         if (map->olm_name == NULL)
1932                 RETURN(0);
1933
1934         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1935         if (IS_ERR(child))
1936                 RETURN(PTR_ERR(child));
1937
1938         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1939                               &map->olm_fid, map->olm_flags);
1940         dput(child);
1941
1942         RETURN(rc);
1943 }
1944
1945 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
1946                              loff_t offset, __u64 ino, unsigned d_type)
1947 {
1948         struct osd_ios_filldir_buf *fill_buf = buf;
1949         struct osd_device          *dev      = fill_buf->oifb_dev;
1950         const struct osd_lf_map    *map;
1951         struct dentry              *child;
1952         int                         rc       = 0;
1953         ENTRY;
1954
1955         /* skip any '.' started names */
1956         if (name[0] == '.')
1957                 RETURN(0);
1958
1959         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
1960                 if (map->olm_namelen != namelen)
1961                         continue;
1962
1963                 if (strncmp(map->olm_name, name, namelen) == 0)
1964                         break;
1965         }
1966
1967         if (map->olm_name == NULL)
1968                 RETURN(0);
1969
1970         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1971         if (IS_ERR(child))
1972                 RETURN(PTR_ERR(child));
1973
1974         if (!(map->olm_flags & OLF_NO_OI))
1975                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1976                                       &map->olm_fid, map->olm_flags);
1977         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
1978                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
1979                                       map->olm_filldir);
1980         dput(child);
1981
1982         RETURN(rc);
1983 }
1984
1985 static int
1986 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1987                      struct dentry *dentry, filldir_t filldir)
1988 {
1989         struct osd_ios_filldir_buf    buf   = {
1990 #ifdef HAVE_DIR_CONTEXT
1991                                                 .ctx.actor = filldir,
1992 #endif
1993                                                 .oifb_info = info,
1994                                                 .oifb_dev = dev,
1995                                                 .oifb_dentry = dentry };
1996         struct file                  *filp  = &info->oti_file;
1997         struct inode                 *inode = dentry->d_inode;
1998         const struct file_operations *fops  = inode->i_fop;
1999         int                           rc;
2000         ENTRY;
2001
2002         LASSERT(filldir != NULL);
2003
2004         filp->f_pos = 0;
2005         filp->f_dentry = dentry;
2006         filp->f_mode = FMODE_64BITHASH;
2007         filp->f_mapping = inode->i_mapping;
2008         filp->f_op = fops;
2009         filp->private_data = NULL;
2010         set_file_inode(filp, inode);
2011
2012 #ifdef HAVE_DIR_CONTEXT
2013         buf.ctx.pos = filp->f_pos;
2014         rc = fops->iterate(filp, &buf.ctx);
2015         filp->f_pos = buf.ctx.pos;
2016 #else
2017         rc = fops->readdir(filp, &buf, filldir);
2018 #endif
2019         fops->release(inode, filp);
2020
2021         RETURN(rc);
2022 }
2023
2024 static int
2025 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2026                   struct dentry *dentry, filldir_t filldir)
2027 {
2028         struct osd_scrub  *scrub  = &dev->od_scrub;
2029         struct scrub_file *sf     = &scrub->os_file;
2030         struct dentry     *child;
2031         int                rc;
2032         ENTRY;
2033
2034         /* It is existing MDT0 device. We only allow the case of object without
2035          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2036          * can generate IGIF mode FID for the object and related OI mapping. If
2037          * it is on other MDTs, then becuase file-level backup/restore, related
2038          * OI mapping may be invalid already, we do not know which is the right
2039          * FID for the object. We only allow IGIF objects to reside on the MDT0.
2040          *
2041          * XXX: For the case of object on non-MDT0 device with neither LMA nor
2042          *      "fid" xattr, then something crashed. We cannot re-generate the
2043          *      FID directly, instead, the OI scrub will scan the OI structure
2044          *      and try to re-generate the LMA from the OI mapping. But if the
2045          *      OI mapping crashed or lost also, then we have to give up under
2046          *      double failure cases. */
2047         scrub->os_convert_igif = 1;
2048         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2049                                        strlen(dot_lustre_name));
2050         if (IS_ERR(child)) {
2051                 rc = PTR_ERR(child);
2052                 if (rc == -ENOENT) {
2053                         /* It is 1.8 MDT device. */
2054                         if (!(sf->sf_flags & SF_UPGRADE)) {
2055                                 osd_scrub_file_reset(scrub,
2056                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2057                                         SF_UPGRADE);
2058                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2059                                 rc = osd_scrub_file_store(scrub);
2060                         } else {
2061                                 rc = 0;
2062                         }
2063                 }
2064         } else {
2065                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2066                  * so the client will get IGIF for the ".lustre" object when
2067                  * the MDT restart.
2068                  *
2069                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2070                  * it does not know whether there are some old clients cached
2071                  * the ".lustre" IGIF during the upgrading. Two choices:
2072                  *
2073                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2074                  *    It will allow the old connected clients to access the
2075                  *    ".lustre" with cached IGIF. But it will cause others
2076                  *    on the MDT failed to check "fid_is_dot_lustre()".
2077                  *
2078                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2079                  *    for ".lustre" in spite of whether there are some clients
2080                  *    cached the ".lustre" IGIF or not. It enables the check
2081                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
2082                  *    that the old connected clients cannot access the ".lustre"
2083                  *    with the cached IGIF.
2084                  *
2085                  * Usually, it is rare case for the old connected clients
2086                  * to access the ".lustre" with cached IGIF. So we prefer
2087                  * to the solution 2). */
2088                 rc = osd_ios_scan_one(info, dev, child->d_inode,
2089                                       &LU_DOT_LUSTRE_FID, 0);
2090                 if (rc == 0)
2091                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2092                                               osd_ios_dl_fill);
2093                 dput(child);
2094         }
2095
2096         RETURN(rc);
2097 }
2098
2099 static int
2100 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2101                      struct dentry *dentry, filldir_t filldir)
2102 {
2103         struct osd_scrub  *scrub  = &dev->od_scrub;
2104         struct scrub_file *sf     = &scrub->os_file;
2105         struct dentry     *child;
2106         int                rc;
2107         ENTRY;
2108
2109         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2110                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2111                 rc = osd_scrub_file_store(scrub);
2112                 if (rc != 0)
2113                         RETURN(rc);
2114         }
2115
2116         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2117         if (!IS_ERR(child)) {
2118                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2119                 dput(child);
2120         } else {
2121                 rc = PTR_ERR(child);
2122         }
2123
2124         if (rc != 0 && rc != -ENOENT)
2125                 RETURN(rc);
2126
2127         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2128         if (!IS_ERR(child)) {
2129                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2130                 dput(child);
2131         } else {
2132                 rc = PTR_ERR(child);
2133         }
2134
2135         if (rc == -ENOENT)
2136                 rc = 0;
2137
2138         RETURN(rc);
2139 }
2140
2141 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2142                                 struct osd_device *dev)
2143 {
2144         struct osd_ios_item     *item    = NULL;
2145         scandir_t                scandir = osd_ios_general_scan;
2146         filldir_t                filldir = osd_ios_root_fill;
2147         struct dentry           *dentry  = osd_sb(dev)->s_root;
2148         const struct osd_lf_map *map     = osd_lf_maps;
2149         int                      rc;
2150         ENTRY;
2151
2152         /* Lookup IGIF in OI by force for initial OI scrub. */
2153         dev->od_igif_inoi = 1;
2154
2155         while (1) {
2156                 rc = scandir(info, dev, dentry, filldir);
2157                 if (item != NULL) {
2158                         dput(item->oii_dentry);
2159                         OBD_FREE_PTR(item);
2160                 }
2161
2162                 if (rc != 0)
2163                         break;
2164
2165                 if (list_empty(&dev->od_ios_list))
2166                         break;
2167
2168                 item = list_entry(dev->od_ios_list.next,
2169                                   struct osd_ios_item, oii_list);
2170                 list_del_init(&item->oii_list);
2171
2172                 LASSERT(item->oii_scandir != NULL);
2173                 scandir = item->oii_scandir;
2174                 filldir = item->oii_filldir;
2175                 dentry = item->oii_dentry;
2176         }
2177
2178         while (!list_empty(&dev->od_ios_list)) {
2179                 item = list_entry(dev->od_ios_list.next,
2180                                   struct osd_ios_item, oii_list);
2181                 list_del_init(&item->oii_list);
2182                 dput(item->oii_dentry);
2183                 OBD_FREE_PTR(item);
2184         }
2185
2186         if (rc != 0)
2187                 RETURN(rc);
2188
2189         /* There maybe the case that the object has been removed, but its OI
2190          * mapping is still in the OI file, such as the "CATALOGS" after MDT
2191          * file-level backup/restore. So here cleanup the stale OI mappings. */
2192         while (map->olm_name != NULL) {
2193                 struct dentry *child;
2194
2195                 if (fid_is_zero(&map->olm_fid)) {
2196                         map++;
2197                         continue;
2198                 }
2199
2200                 child = osd_ios_lookup_one_len(map->olm_name,
2201                                                osd_sb(dev)->s_root,
2202                                                map->olm_namelen);
2203                 if (!IS_ERR(child))
2204                         dput(child);
2205                 else if (PTR_ERR(child) == -ENOENT)
2206                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2207                                                   NULL, DTO_INDEX_DELETE,
2208                                                   true, 0);
2209                 map++;
2210         }
2211
2212         RETURN(0);
2213 }
2214
2215 char *osd_lf_fid2name(const struct lu_fid *fid)
2216 {
2217         const struct osd_lf_map *map = osd_lf_maps;
2218
2219         while (map->olm_name != NULL) {
2220                 if (!lu_fid_eq(fid, &map->olm_fid)) {
2221                         map++;
2222                         continue;
2223                 }
2224
2225                 if (map->olm_flags & OLF_SHOW_NAME)
2226                         return map->olm_name;
2227                 else
2228                         return "";
2229         }
2230
2231         return NULL;
2232 }
2233
2234 /* OI scrub start/stop */
2235
2236 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2237 {
2238         struct osd_scrub     *scrub  = &dev->od_scrub;
2239         struct ptlrpc_thread *thread = &scrub->os_thread;
2240         struct l_wait_info    lwi    = { 0 };
2241         struct task_struct   *task;
2242         int                   rc;
2243         ENTRY;
2244
2245         /* os_lock: sync status between stop and scrub thread */
2246         spin_lock(&scrub->os_lock);
2247
2248 again:
2249         if (thread_is_running(thread)) {
2250                 spin_unlock(&scrub->os_lock);
2251                 if (!(scrub->os_file.sf_flags & SF_AUTO) ||
2252                      (flags & (SS_AUTO_FULL | SS_AUTO_PARTIAL)))
2253                         RETURN(-EALREADY);
2254
2255                 osd_scrub_join(dev, flags, false);
2256                 spin_lock(&scrub->os_lock);
2257                 if (!thread_is_running(thread))
2258                         goto again;
2259
2260                 spin_unlock(&scrub->os_lock);
2261                 RETURN(0);
2262         }
2263
2264         if (unlikely(thread_is_stopping(thread))) {
2265                 spin_unlock(&scrub->os_lock);
2266                 l_wait_event(thread->t_ctl_waitq,
2267                              thread_is_stopped(thread),
2268                              &lwi);
2269                 spin_lock(&scrub->os_lock);
2270                 goto again;
2271         }
2272         spin_unlock(&scrub->os_lock);
2273
2274         if (scrub->os_file.sf_status == SS_COMPLETED) {
2275                 if (!(flags & SS_SET_FAILOUT))
2276                         flags |= SS_CLEAR_FAILOUT;
2277
2278                 if (!(flags & SS_SET_DRYRUN))
2279                         flags |= SS_CLEAR_DRYRUN;
2280
2281                 flags |= SS_RESET;
2282         }
2283
2284         scrub->os_start_flags = flags;
2285         thread_set_flags(thread, 0);
2286         task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2287         if (IS_ERR(task)) {
2288                 rc = PTR_ERR(task);
2289                 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2290                        osd_scrub2name(scrub), rc);
2291                 RETURN(rc);
2292         }
2293
2294         l_wait_event(thread->t_ctl_waitq,
2295                      thread_is_running(thread) || thread_is_stopped(thread),
2296                      &lwi);
2297
2298         RETURN(0);
2299 }
2300
2301 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2302 {
2303         int rc;
2304         ENTRY;
2305
2306         /* od_otable_mutex: prevent curcurrent start/stop */
2307         mutex_lock(&dev->od_otable_mutex);
2308         rc = do_osd_scrub_start(dev, flags);
2309         mutex_unlock(&dev->od_otable_mutex);
2310
2311         RETURN(rc == -EALREADY ? 0 : rc);
2312 }
2313
2314 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2315 {
2316         struct ptlrpc_thread *thread = &scrub->os_thread;
2317         struct l_wait_info    lwi    = { 0 };
2318
2319         /* os_lock: sync status between stop and scrub thread */
2320         spin_lock(&scrub->os_lock);
2321         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2322                 thread_set_flags(thread, SVC_STOPPING);
2323                 spin_unlock(&scrub->os_lock);
2324                 wake_up_all(&thread->t_ctl_waitq);
2325                 l_wait_event(thread->t_ctl_waitq,
2326                              thread_is_stopped(thread),
2327                              &lwi);
2328                 /* Do not skip the last lock/unlock, which can guarantee that
2329                  * the caller cannot return until the OI scrub thread exit. */
2330                 spin_lock(&scrub->os_lock);
2331         }
2332         spin_unlock(&scrub->os_lock);
2333 }
2334
2335 static void osd_scrub_stop(struct osd_device *dev)
2336 {
2337         /* od_otable_mutex: prevent curcurrent start/stop */
2338         mutex_lock(&dev->od_otable_mutex);
2339         dev->od_scrub.os_paused = 1;
2340         do_osd_scrub_stop(&dev->od_scrub);
2341         mutex_unlock(&dev->od_otable_mutex);
2342 }
2343
2344 /* OI scrub setup/cleanup */
2345
2346 static const char osd_scrub_name[] = "OI_scrub";
2347
2348 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2349 {
2350         struct osd_thread_info     *info   = osd_oti_get(env);
2351         struct osd_scrub           *scrub  = &dev->od_scrub;
2352         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
2353         struct scrub_file          *sf     = &scrub->os_file;
2354         struct super_block         *sb     = osd_sb(dev);
2355         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
2356         struct lvfs_run_ctxt        saved;
2357         struct file                *filp;
2358         struct inode               *inode;
2359         struct lu_fid              *fid    = &info->oti_fid;
2360         int                         dirty  = 0;
2361         int                         rc     = 0;
2362         ENTRY;
2363
2364         memset(scrub, 0, sizeof(*scrub));
2365         OBD_SET_CTXT_MAGIC(ctxt);
2366         ctxt->pwdmnt = dev->od_mnt;
2367         ctxt->pwd = dev->od_mnt->mnt_root;
2368         ctxt->fs = get_ds();
2369
2370         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2371         init_rwsem(&scrub->os_rwsem);
2372         spin_lock_init(&scrub->os_lock);
2373         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2374
2375         push_ctxt(&saved, ctxt);
2376         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2377         if (IS_ERR(filp)) {
2378                 pop_ctxt(&saved, ctxt);
2379                 RETURN(PTR_ERR(filp));
2380         }
2381
2382         inode = filp->f_dentry->d_inode;
2383         /* 'What the @fid is' is not imporatant, because the object
2384          * has no OI mapping, and only is visible inside the OSD.*/
2385         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2386         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2387         if (rc != 0) {
2388                 filp_close(filp, NULL);
2389                 pop_ctxt(&saved, ctxt);
2390                 RETURN(rc);
2391         }
2392
2393         scrub->os_inode = igrab(inode);
2394         filp_close(filp, NULL);
2395         pop_ctxt(&saved, ctxt);
2396
2397         rc = osd_scrub_file_load(scrub);
2398         if (rc == -ENOENT) {
2399                 osd_scrub_file_init(scrub, es->s_uuid);
2400                 /* If the "/O" dir does not exist when mount (indicated by
2401                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2402                  * then it is quite probably that the device is a new one,
2403                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2404                  *
2405                  * For the rare case that "/O" and "OI_scrub" both lost on
2406                  * an old device, it can be found and cleared later.
2407                  *
2408                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2409                  * need to check "filter_fid_old" and to convert it to
2410                  * "filter_fid" for each object, and all the IGIF should
2411                  * have their FID mapping in OI files already. */
2412                 if (dev->od_maybe_new)
2413                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2414                 dirty = 1;
2415         } else if (rc != 0) {
2416                 GOTO(cleanup_inode, rc);
2417         } else {
2418                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2419                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2420                         dirty = 1;
2421                 } else if (sf->sf_status == SS_SCANNING) {
2422                         sf->sf_status = SS_CRASHED;
2423                         dirty = 1;
2424                 }
2425         }
2426
2427         if (sf->sf_pos_last_checkpoint != 0)
2428                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2429         else
2430                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2431
2432         if (dirty != 0) {
2433                 rc = osd_scrub_file_store(scrub);
2434                 if (rc != 0)
2435                         GOTO(cleanup_inode, rc);
2436         }
2437
2438         /* Initialize OI files. */
2439         rc = osd_oi_init(info, dev);
2440         if (rc < 0)
2441                 GOTO(cleanup_inode, rc);
2442
2443         rc = osd_initial_OI_scrub(info, dev);
2444         if (rc != 0)
2445                 GOTO(cleanup_oi, rc);
2446
2447         if (sf->sf_flags & SF_UPGRADE ||
2448             !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2449               sf->sf_success_count > 0)) {
2450                 dev->od_igif_inoi = 0;
2451                 dev->od_check_ff = dev->od_is_ost;
2452         } else {
2453                 dev->od_igif_inoi = 1;
2454                 dev->od_check_ff = 0;
2455         }
2456
2457         if (sf->sf_flags & SF_INCONSISTENT)
2458                 /* The 'od_igif_inoi' will be set under the
2459                  * following cases:
2460                  * 1) new created system, or
2461                  * 2) restored from file-level backup, or
2462                  * 3) the upgrading completed.
2463                  *
2464                  * The 'od_igif_inoi' may be cleared by OI scrub
2465                  * later if found that the system is upgrading. */
2466                 dev->od_igif_inoi = 1;
2467
2468         if (!dev->od_noscrub &&
2469             ((sf->sf_status == SS_PAUSED) ||
2470              (sf->sf_status == SS_CRASHED &&
2471               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2472                               SF_UPGRADE | SF_AUTO)) ||
2473              (sf->sf_status == SS_INIT &&
2474               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2475                               SF_UPGRADE))))
2476                 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2477
2478         if (rc != 0)
2479                 GOTO(cleanup_oi, rc);
2480
2481         /* it is possible that dcache entries may keep objects after they are
2482          * deleted by OSD. While it looks safe this can cause object data to
2483          * stay until umount causing failures in tests calculating free space,
2484          * e.g. replay-ost-single. Since those dcache entries are not used
2485          * anymore let's just free them after use here */
2486         shrink_dcache_sb(sb);
2487
2488         RETURN(0);
2489 cleanup_oi:
2490         osd_oi_fini(info, dev);
2491 cleanup_inode:
2492         iput(scrub->os_inode);
2493         scrub->os_inode = NULL;
2494
2495         return rc;
2496 }
2497
2498 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2499 {
2500         struct osd_scrub *scrub = &dev->od_scrub;
2501
2502         LASSERT(dev->od_otable_it == NULL);
2503
2504         if (scrub->os_inode != NULL) {
2505                 osd_scrub_stop(dev);
2506                 iput(scrub->os_inode);
2507                 scrub->os_inode = NULL;
2508         }
2509         if (dev->od_oi_table != NULL)
2510                 osd_oi_fini(osd_oti_get(env), dev);
2511 }
2512
2513 /* object table based iteration APIs */
2514
2515 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2516                                        struct dt_object *dt, __u32 attr)
2517 {
2518         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2519         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2520         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2521         struct osd_scrub       *scrub = &dev->od_scrub;
2522         struct osd_otable_it   *it;
2523         __u32                   start = 0;
2524         int                     rc;
2525         ENTRY;
2526
2527         /* od_otable_mutex: prevent curcurrent init/fini */
2528         mutex_lock(&dev->od_otable_mutex);
2529         if (dev->od_otable_it != NULL)
2530                 GOTO(out, it = ERR_PTR(-EALREADY));
2531
2532         OBD_ALLOC_PTR(it);
2533         if (it == NULL)
2534                 GOTO(out, it = ERR_PTR(-ENOMEM));
2535
2536         dev->od_otable_it = it;
2537         it->ooi_dev = dev;
2538         it->ooi_cache.ooc_consumer_idx = -1;
2539         if (flags & DOIF_OUTUSED)
2540                 it->ooi_used_outside = 1;
2541
2542         if (flags & DOIF_RESET)
2543                 start |= SS_RESET;
2544
2545         if (valid & DOIV_ERROR_HANDLE) {
2546                 if (flags & DOIF_FAILOUT)
2547                         start |= SS_SET_FAILOUT;
2548                 else
2549                         start |= SS_CLEAR_FAILOUT;
2550         }
2551
2552         if (valid & DOIV_DRYRUN) {
2553                 if (flags & DOIF_DRYRUN)
2554                         start |= SS_SET_DRYRUN;
2555                 else
2556                         start |= SS_CLEAR_DRYRUN;
2557         }
2558
2559         rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2560         if (rc < 0 && rc != -EALREADY) {
2561                 dev->od_otable_it = NULL;
2562                 OBD_FREE_PTR(it);
2563                 GOTO(out, it = ERR_PTR(rc));
2564         }
2565
2566         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2567
2568         GOTO(out, it);
2569
2570 out:
2571         mutex_unlock(&dev->od_otable_mutex);
2572         return (struct dt_it *)it;
2573 }
2574
2575 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2576 {
2577         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2578         struct osd_device    *dev = it->ooi_dev;
2579
2580         /* od_otable_mutex: prevent curcurrent init/fini */
2581         mutex_lock(&dev->od_otable_mutex);
2582         do_osd_scrub_stop(&dev->od_scrub);
2583         LASSERT(dev->od_otable_it == it);
2584
2585         dev->od_otable_it = NULL;
2586         mutex_unlock(&dev->od_otable_mutex);
2587         OBD_FREE_PTR(it);
2588 }
2589
2590 static int osd_otable_it_get(const struct lu_env *env,
2591                              struct dt_it *di, const struct dt_key *key)
2592 {
2593         return 0;
2594 }
2595
2596 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2597 {
2598 }
2599
2600 static inline int
2601 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2602 {
2603         spin_lock(&scrub->os_lock);
2604         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2605             scrub->os_waiting ||
2606             !thread_is_running(&scrub->os_thread))
2607                 it->ooi_waiting = 0;
2608         else
2609                 it->ooi_waiting = 1;
2610         spin_unlock(&scrub->os_lock);
2611
2612         return !it->ooi_waiting;
2613 }
2614
2615 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2616 {
2617         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2618         struct osd_device       *dev    = it->ooi_dev;
2619         struct osd_scrub        *scrub  = &dev->od_scrub;
2620         struct osd_otable_cache *ooc    = &it->ooi_cache;
2621         struct ptlrpc_thread    *thread = &scrub->os_thread;
2622         struct l_wait_info       lwi    = { 0 };
2623         int                      rc;
2624         ENTRY;
2625
2626         LASSERT(it->ooi_user_ready);
2627
2628 again:
2629         if (!thread_is_running(thread) && !it->ooi_used_outside)
2630                 RETURN(1);
2631
2632         if (ooc->ooc_cached_items > 0) {
2633                 ooc->ooc_cached_items--;
2634                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2635                                         ~OSD_OTABLE_IT_CACHE_MASK;
2636                 RETURN(0);
2637         }
2638
2639         if (it->ooi_all_cached) {
2640                 l_wait_event(thread->t_ctl_waitq,
2641                              !thread_is_running(thread),
2642                              &lwi);
2643                 RETURN(1);
2644         }
2645
2646         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2647                 spin_lock(&scrub->os_lock);
2648                 scrub->os_waiting = 0;
2649                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2650                 spin_unlock(&scrub->os_lock);
2651         }
2652
2653         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2654                 l_wait_event(thread->t_ctl_waitq,
2655                              osd_otable_it_wakeup(scrub, it),
2656                              &lwi);
2657
2658         if (!thread_is_running(thread) && !it->ooi_used_outside)
2659                 RETURN(1);
2660
2661         rc = osd_otable_it_preload(env, it);
2662         if (rc >= 0)
2663                 goto again;
2664
2665         RETURN(rc);
2666 }
2667
2668 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2669                                         const struct dt_it *di)
2670 {
2671         return NULL;
2672 }
2673
2674 static int osd_otable_it_key_size(const struct lu_env *env,
2675                                   const struct dt_it *di)
2676 {
2677         return sizeof(__u64);
2678 }
2679
2680 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2681                              struct dt_rec *rec, __u32 attr)
2682 {
2683         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2684         struct osd_otable_cache *ooc = &it->ooi_cache;
2685
2686         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2687
2688         /* Filter out Invald FID already. */
2689         LASSERTF(fid_is_sane((struct lu_fid *)rec),
2690                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2691                  PFID((struct lu_fid *)rec),
2692                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2693
2694         return 0;
2695 }
2696
2697 static __u64 osd_otable_it_store(const struct lu_env *env,
2698                                  const struct dt_it *di)
2699 {
2700         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2701         struct osd_otable_cache *ooc = &it->ooi_cache;
2702         __u64                    hash;
2703
2704         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2705                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2706         else
2707                 hash = ooc->ooc_pos_preload;
2708         return hash;
2709 }
2710
2711 /**
2712  * Set the OSD layer iteration start position as the specified hash.
2713  */
2714 static int osd_otable_it_load(const struct lu_env *env,
2715                               const struct dt_it *di, __u64 hash)
2716 {
2717         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
2718         struct osd_device       *dev   = it->ooi_dev;
2719         struct osd_otable_cache *ooc   = &it->ooi_cache;
2720         struct osd_scrub        *scrub = &dev->od_scrub;
2721         int                      rc;
2722         ENTRY;
2723
2724         /* Forbid to set iteration position after iteration started. */
2725         if (it->ooi_user_ready)
2726                 RETURN(-EPERM);
2727
2728         if (hash > OSD_OTABLE_MAX_HASH)
2729                 hash = OSD_OTABLE_MAX_HASH;
2730
2731         ooc->ooc_pos_preload = hash;
2732         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2733                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2734
2735         it->ooi_user_ready = 1;
2736         if (!scrub->os_full_speed)
2737                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2738
2739         /* Unplug OSD layer iteration by the first next() call. */
2740         rc = osd_otable_it_next(env, (struct dt_it *)it);
2741
2742         RETURN(rc);
2743 }
2744
2745 static int osd_otable_it_key_rec(const struct lu_env *env,
2746                                  const struct dt_it *di, void *key_rec)
2747 {
2748         return 0;
2749 }
2750
2751 const struct dt_index_operations osd_otable_ops = {
2752         .dio_it = {
2753                 .init     = osd_otable_it_init,
2754                 .fini     = osd_otable_it_fini,
2755                 .get      = osd_otable_it_get,
2756                 .put      = osd_otable_it_put,
2757                 .next     = osd_otable_it_next,
2758                 .key      = osd_otable_it_key,
2759                 .key_size = osd_otable_it_key_size,
2760                 .rec      = osd_otable_it_rec,
2761                 .store    = osd_otable_it_store,
2762                 .load     = osd_otable_it_load,
2763                 .key_rec  = osd_otable_it_key_rec,
2764         }
2765 };
2766
2767 /* high priority inconsistent items list APIs */
2768
2769 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL  60
2770
2771 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2772                    int insert)
2773 {
2774         struct osd_inconsistent_item *oii;
2775         struct osd_scrub             *scrub  = &dev->od_scrub;
2776         struct ptlrpc_thread         *thread = &scrub->os_thread;
2777         int                           wakeup = 0;
2778         ENTRY;
2779
2780         OBD_ALLOC_PTR(oii);
2781         if (unlikely(oii == NULL))
2782                 RETURN(-ENOMEM);
2783
2784         INIT_LIST_HEAD(&oii->oii_list);
2785         oii->oii_cache = *oic;
2786         oii->oii_insert = insert;
2787
2788         if (scrub->os_partial_scan) {
2789                 __u64 now = cfs_time_current_sec();
2790
2791                 /* If there haven't been errors in a long time,
2792                  * decay old count until either the errors are
2793                  * gone or we reach the current interval. */
2794                 while (unlikely(scrub->os_bad_oimap_count > 0 &&
2795                                 scrub->os_bad_oimap_time +
2796                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
2797                         scrub->os_bad_oimap_count >>= 1;
2798                         scrub->os_bad_oimap_time +=
2799                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
2800                 }
2801
2802                 scrub->os_bad_oimap_time = now;
2803                 if (++scrub->os_bad_oimap_count >
2804                     dev->od_full_scrub_threshold_rate)
2805                         scrub->os_full_scrub = 1;
2806         }
2807
2808         spin_lock(&scrub->os_lock);
2809         if (unlikely(!thread_is_running(thread))) {
2810                 spin_unlock(&scrub->os_lock);
2811                 OBD_FREE_PTR(oii);
2812                 RETURN(-EAGAIN);
2813         }
2814
2815         if (list_empty(&scrub->os_inconsistent_items))
2816                 wakeup = 1;
2817         list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2818         spin_unlock(&scrub->os_lock);
2819
2820         if (wakeup != 0)
2821                 wake_up_all(&thread->t_ctl_waitq);
2822
2823         RETURN(0);
2824 }
2825
2826 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2827                    struct osd_inode_id *id)
2828 {
2829         struct osd_scrub             *scrub = &dev->od_scrub;
2830         struct osd_inconsistent_item *oii;
2831         ENTRY;
2832
2833         spin_lock(&scrub->os_lock);
2834         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2835                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2836                         *id = oii->oii_cache.oic_lid;
2837                         spin_unlock(&scrub->os_lock);
2838                         RETURN(0);
2839                 }
2840         }
2841         spin_unlock(&scrub->os_lock);
2842
2843         RETURN(-ENOENT);
2844 }
2845
2846 /* OI scrub dump */
2847
2848 static const char *scrub_status_names[] = {
2849         "init",
2850         "scanning",
2851         "completed",
2852         "failed",
2853         "stopped",
2854         "paused",
2855         "crashed",
2856         NULL
2857 };
2858
2859 static const char *scrub_flags_names[] = {
2860         "recreated",
2861         "inconsistent",
2862         "auto",
2863         "upgrade",
2864         NULL
2865 };
2866
2867 static const char *scrub_param_names[] = {
2868         "failout",
2869         "dryrun",
2870         NULL
2871 };
2872
2873 static int scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
2874                            const char *prefix)
2875 {
2876         int flag;
2877         int rc;
2878         int i;
2879
2880         rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2881         if (rc < 0)
2882                 return rc;
2883
2884         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2885                 if (flag & bits) {
2886                         bits &= ~flag;
2887                         rc = seq_printf(m, "%s%c", names[i],
2888                                         bits != 0 ? ',' : '\n');
2889                         if (rc < 0)
2890                                 return rc;
2891                 }
2892         }
2893         return 0;
2894 }
2895
2896 static int scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
2897 {
2898         int rc;
2899
2900         if (time != 0)
2901                 rc = seq_printf(m, "%s: "LPU64" seconds\n", prefix,
2902                               cfs_time_current_sec() - time);
2903         else
2904                 rc = seq_printf(m, "%s: N/A\n", prefix);
2905         return rc;
2906 }
2907
2908 static int scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
2909 {
2910         int rc;
2911
2912         if (pos != 0)
2913                 rc = seq_printf(m, "%s: "LPU64"\n", prefix, pos);
2914         else
2915                 rc = seq_printf(m, "%s: N/A\n", prefix);
2916         return rc;
2917 }
2918
2919 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
2920 {
2921         struct osd_scrub  *scrub   = &dev->od_scrub;
2922         struct scrub_file *sf      = &scrub->os_file;
2923         __u64              checked;
2924         __u64              speed;
2925         int                rc;
2926
2927         down_read(&scrub->os_rwsem);
2928         rc = seq_printf(m, "name: OI_scrub\n"
2929                         "magic: 0x%x\n"
2930                         "oi_files: %d\n"
2931                         "status: %s\n",
2932                         sf->sf_magic, (int)sf->sf_oi_count,
2933                         scrub_status_names[sf->sf_status]);
2934         if (rc < 0)
2935                 goto out;
2936
2937         rc = scrub_bits_dump(m, sf->sf_flags, scrub_flags_names,
2938                              "flags");
2939         if (rc < 0)
2940                 goto out;
2941
2942         rc = scrub_bits_dump(m, sf->sf_param, scrub_param_names,
2943                              "param");
2944         if (rc < 0)
2945                 goto out;
2946
2947         rc = scrub_time_dump(m, sf->sf_time_last_complete,
2948                              "time_since_last_completed");
2949         if (rc < 0)
2950                 goto out;
2951
2952         rc = scrub_time_dump(m, sf->sf_time_latest_start,
2953                              "time_since_latest_start");
2954         if (rc < 0)
2955                 goto out;
2956
2957         rc = scrub_time_dump(m, sf->sf_time_last_checkpoint,
2958                              "time_since_last_checkpoint");
2959         if (rc < 0)
2960                 goto out;
2961
2962         rc = scrub_pos_dump(m, sf->sf_pos_latest_start,
2963                             "latest_start_position");
2964         if (rc < 0)
2965                 goto out;
2966
2967         rc = scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
2968                             "last_checkpoint_position");
2969         if (rc < 0)
2970                 goto out;
2971
2972         rc = scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
2973                             "first_failure_position");
2974         if (rc < 0)
2975                 goto out;
2976
2977         checked = sf->sf_items_checked + scrub->os_new_checked;
2978         rc = seq_printf(m, "checked: "LPU64"\n"
2979                       "updated: "LPU64"\n"
2980                       "failed: "LPU64"\n"
2981                       "prior_updated: "LPU64"\n"
2982                       "noscrub: "LPU64"\n"
2983                       "igif: "LPU64"\n"
2984                       "success_count: %u\n",
2985                       checked, sf->sf_items_updated, sf->sf_items_failed,
2986                       sf->sf_items_updated_prior, sf->sf_items_noscrub,
2987                       sf->sf_items_igif, sf->sf_success_count);
2988         if (rc < 0)
2989                 goto out;
2990
2991         speed = checked;
2992         if (thread_is_running(&scrub->os_thread)) {
2993                 cfs_duration_t duration = cfs_time_current() -
2994                                           scrub->os_time_last_checkpoint;
2995                 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
2996                                                      MSEC_PER_SEC);
2997                 __u32 rtime = sf->sf_run_time +
2998                               cfs_duration_sec(duration + HALF_SEC);
2999
3000                 if (duration != 0)
3001                         do_div(new_checked, duration);
3002                 if (rtime != 0)
3003                         do_div(speed, rtime);
3004                 rc = seq_printf(m, "run_time: %u seconds\n"
3005                               "average_speed: "LPU64" objects/sec\n"
3006                               "real-time_speed: "LPU64" objects/sec\n"
3007                               "current_position: %u\n"
3008                               "lf_scanned: "LPU64"\n"
3009                               "lf_reparied: "LPU64"\n"
3010                               "lf_failed: "LPU64"\n",
3011                               rtime, speed, new_checked, scrub->os_pos_current,
3012                               scrub->os_lf_scanned, scrub->os_lf_repaired,
3013                               scrub->os_lf_failed);
3014         } else {
3015                 if (sf->sf_run_time != 0)
3016                         do_div(speed, sf->sf_run_time);
3017                 rc = seq_printf(m, "run_time: %u seconds\n"
3018                               "average_speed: "LPU64" objects/sec\n"
3019                               "real-time_speed: N/A\n"
3020                               "current_position: N/A\n"
3021                               "lf_scanned: "LPU64"\n"
3022                               "lf_reparied: "LPU64"\n"
3023                               "lf_failed: "LPU64"\n",
3024                               sf->sf_run_time, speed, scrub->os_lf_scanned,
3025                               scrub->os_lf_repaired, scrub->os_lf_failed);
3026         }
3027
3028 out:
3029         up_read(&scrub->os_rwsem);
3030         return (rc < 0 ? -ENOSPC : 0);
3031 }