Whamcloud - gitweb
LU-12635 build: Support for gcc -Wimplicit-fallthrough
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_scrub.c
index 358edc8..dfdd87c 100644 (file)
@@ -20,7 +20,7 @@
  * GPL HEADER END
  */
 /*
- * Copyright (c) 2012, 2016, Intel Corporation.
+ * Copyright (c) 2012, 2017, Intel Corporation.
  */
 /*
  * lustre/osd-ldiskfs/osd_scrub.c
@@ -163,13 +163,13 @@ static int
 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
                     struct inode *inode, const struct lu_fid *fid)
 {
-       struct filter_fid_old   *ff      = &info->oti_ff;
-       struct dentry           *dentry  = &info->oti_obj_dentry;
-       struct lu_fid           *tfid    = &info->oti_fid;
-       handle_t                *jh;
-       int                      size    = 0;
-       int                      rc;
-       bool                     reset   = false;
+       struct filter_fid_18_23 *ff = &info->oti_ff_old;
+       struct dentry *dentry = &info->oti_obj_dentry;
+       struct lu_fid *tfid = &info->oti_fid;
+       bool fid_18_23 = false;
+       handle_t *jh;
+       int size = 0;
+       int rc;
        ENTRY;
 
        if (dev->od_scrub.os_scrub.os_file.sf_param & SP_DRYRUN)
@@ -207,19 +207,19 @@ osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
        rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
        if (rc == sizeof(*ff)) {
                /* 2) delete the old XATTR_NAME_FID */
-               ll_vfs_dq_init(inode);
-               rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
+               dquot_initialize(inode);
+               rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
                if (rc)
                        GOTO(stop, rc);
 
-               reset = true;
-       } else if (rc != -ENODATA && rc != sizeof(struct filter_fid)) {
+               fid_18_23 = true;
+       } else if (rc != -ENODATA && rc < (int)sizeof(struct filter_fid_24_29)) {
                GOTO(stop, rc = -EINVAL);
        }
 
        /* 3) make new LMA and add it */
        rc = osd_ea_fid_set(info, inode, tfid, LMAC_FID_ON_OST, 0);
-       if (reset) {
+       if (fid_18_23) {
                if (rc)
                        /* If failed, we should try to add the old back. */
                        size = sizeof(*ff);
@@ -472,8 +472,7 @@ static int osd_scrub_prep(const struct lu_env *env, struct osd_device *dev)
        }
 
        if (flags & SS_RESET)
-               scrub_file_reset(scrub,
-                       LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
+               scrub_file_reset(scrub, dev->od_uuid, 0);
 
        if (flags & SS_AUTO_FULL) {
                scrub->os_full_speed = 1;
@@ -509,7 +508,7 @@ static int osd_scrub_prep(const struct lu_env *env, struct osd_device *dev)
 
        scrub->os_pos_current = sf->sf_pos_latest_start;
        sf->sf_status = SS_SCANNING;
-       sf->sf_time_latest_start = cfs_time_current_sec();
+       sf->sf_time_latest_start = ktime_get_real_seconds();
        sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
        sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
        rc = scrub_file_store(env, scrub);
@@ -544,7 +543,7 @@ static int osd_scrub_post(const struct lu_env *env, struct osd_device *dev,
                scrub->os_new_checked = 0;
                sf->sf_pos_last_checkpoint = scrub->os_pos_current;
        }
-       sf->sf_time_last_checkpoint = cfs_time_current_sec();
+       sf->sf_time_last_checkpoint = ktime_get_real_seconds();
        if (result > 0) {
                dev->od_igif_inoi = 1;
                dev->od_check_ff = 0;
@@ -564,8 +563,9 @@ static int osd_scrub_post(const struct lu_env *env, struct osd_device *dev,
        } else {
                sf->sf_status = SS_FAILED;
        }
-       sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
-                                           scrub->os_time_last_checkpoint);
+       sf->sf_run_time += ktime_get_seconds() -
+                          scrub->os_time_last_checkpoint;
+
        rc = scrub_file_store(env, scrub);
        up_write(&scrub->os_rwsem);
 
@@ -626,8 +626,8 @@ static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
         *      quite possible for FID-on-MDT. */
        if (dev->od_is_ost)
                return SCRUB_NEXT_OSTOBJ_OLD;
-       else
-               return 0;
+
+       return 0;
 }
 
 static int osd_scrub_get_fid(struct osd_thread_info *info,
@@ -635,8 +635,8 @@ static int osd_scrub_get_fid(struct osd_thread_info *info,
                             struct lu_fid *fid, bool scrub)
 {
        struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
-       int rc;
        bool has_lma = false;
+       int rc;
 
        rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
                         &info->oti_ost_attrs);
@@ -673,7 +673,7 @@ static int osd_scrub_get_fid(struct osd_thread_info *info,
                rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
                if (rc == 0) {
                        if (scrub)
-                               /* It is old 2.x (x <= 3) or 1.8 OST-object. */
+                               /* It is 2.3 or older OST-object. */
                                rc = SCRUB_NEXT_OSTOBJ_OLD;
                        return rc;
                }
@@ -684,7 +684,7 @@ static int osd_scrub_get_fid(struct osd_thread_info *info,
                                 * to generate its FID, ignore it directly. */
                                rc = SCRUB_NEXT_CONTINUE;
                        else
-                               /* It is 2.4 OST-object. */
+                               /* It is 2.4 or newer OST-object. */
                                rc = SCRUB_NEXT_OSTOBJ_OLD;
                        return rc;
                }
@@ -729,9 +729,16 @@ static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
        /* Not handle the backend root object and agent parent object.
         * They are neither visible to namespace nor have OI mappings. */
        if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
-                    pos == osd_remote_parent_ino(dev)))
+                    is_remote_parent_ino(dev, pos)))
                RETURN(SCRUB_NEXT_CONTINUE);
 
+        /* Skip project quota inode since it is greater than s_first_ino. */
+#ifdef HAVE_PROJECT_QUOTA
+       if (ldiskfs_has_feature_project(sb) &&
+           pos == le32_to_cpu(LDISKFS_SB(sb)->s_es->s_prj_quota_inum))
+               RETURN(SCRUB_NEXT_CONTINUE);
+#endif
+
        osd_id_gen(lid, pos, OSD_OII_NOGEN);
        inode = osd_iget(info, dev, lid);
        if (IS_ERR(inode)) {
@@ -981,8 +988,8 @@ static void osd_scrub_join(const struct lu_env *env, struct osd_device *dev,
                sf->sf_param &= ~SP_DRYRUN;
 
        if (flags & SS_RESET) {
-               scrub_file_reset(scrub, LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
-                       inconsistent ? SF_INCONSISTENT : 0);
+               scrub_file_reset(scrub, dev->od_uuid,
+                                inconsistent ? SF_INCONSISTENT : 0);
                sf->sf_status = SS_SCANNING;
        }
 
@@ -1003,7 +1010,7 @@ static void osd_scrub_join(const struct lu_env *env, struct osd_device *dev,
                sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
 
        scrub->os_pos_current = sf->sf_pos_latest_start;
-       sf->sf_time_latest_start = cfs_time_current_sec();
+       sf->sf_time_latest_start = ktime_get_real_seconds();
        sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
        sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
        rc = scrub_file_store(env, scrub);
@@ -1418,8 +1425,8 @@ static const struct osd_lf_map osd_lf_maps[] = {
 
        /* PENDING */
        {
-               .olm_name       = "PENDING",
-               .olm_namelen    = sizeof("PENDING") - 1,
+               .olm_name       = MDT_ORPHAN_DIR,
+               .olm_namelen    = sizeof(MDT_ORPHAN_DIR) - 1,
        },
 
        /* ROOT */
@@ -1671,6 +1678,19 @@ static const struct osd_lf_map osd_lf_maps[] = {
                .olm_name       = LUSTRE_NODEMAP_NAME,
        },
 
+       /* index_backup */
+       {
+               .olm_name       = INDEX_BACKUP_DIR,
+               .olm_fid        = {
+                       .f_seq  = FID_SEQ_LOCAL_FILE,
+                       .f_oid  = INDEX_BACKUP_OID,
+               },
+               .olm_flags      = OLF_SCAN_SUBITEMS | OLF_NOT_BACKUP,
+               .olm_namelen    = sizeof(INDEX_BACKUP_DIR) - 1,
+               .olm_scandir    = osd_ios_general_scan,
+               .olm_filldir    = osd_ios_varfid_fill,
+       },
+
        {
                .olm_name       = NULL
        }
@@ -1718,34 +1738,9 @@ struct osd_ios_filldir_buf {
        struct osd_thread_info  *oifb_info;
        struct osd_device       *oifb_dev;
        struct dentry           *oifb_dentry;
+       int                      oifb_items;
 };
 
-static inline struct dentry *
-osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
-{
-       struct dentry *dentry;
-
-       dentry = ll_lookup_one_len(name, parent, namelen);
-       if (IS_ERR(dentry)) {
-               int rc = PTR_ERR(dentry);
-
-               if (rc != -ENOENT)
-                       CERROR("Fail to find %.*s in %.*s (%lu/%u): rc = %d\n",
-                              namelen, name, parent->d_name.len,
-                              parent->d_name.name, parent->d_inode->i_ino,
-                              parent->d_inode->i_generation, rc);
-
-               return dentry;
-       }
-
-       if (dentry->d_inode == NULL) {
-               dput(dentry);
-               return ERR_PTR(-ENOENT);
-       }
-
-       return dentry;
-}
-
 static int
 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
                 scandir_t scandir, filldir_t filldir)
@@ -1766,6 +1761,119 @@ osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
        RETURN(0);
 }
 
+static bool osd_index_need_recreate(const struct lu_env *env,
+                                   struct osd_device *dev, struct inode *inode)
+{
+       struct osd_directory *iam = &osd_oti_get(env)->oti_iam;
+       struct iam_container *bag = &iam->od_container;
+       int rc;
+       ENTRY;
+
+       rc = iam_container_init(bag, &iam->od_descr, inode);
+       if (rc)
+               RETURN(true);
+
+       rc = iam_container_setup(bag);
+       iam_container_fini(bag);
+       if (rc)
+               RETURN(true);
+
+       RETURN(false);
+}
+
+static void osd_ios_index_register(const struct lu_env *env,
+                                  struct osd_device *osd,
+                                  const struct lu_fid *fid,
+                                  struct inode *inode)
+{
+       struct osd_directory *iam = &osd_oti_get(env)->oti_iam;
+       struct iam_container *bag = &iam->od_container;
+       struct super_block *sb = osd_sb(osd);
+       struct iam_descr *descr;
+       __u32 keysize = 0;
+       __u32 recsize = 0;
+       int rc;
+       ENTRY;
+
+       /* Index must be a regular file. */
+       if (!S_ISREG(inode->i_mode))
+               RETURN_EXIT;
+
+       /* Index's size must be block aligned. */
+       if (inode->i_size < sb->s_blocksize ||
+           (inode->i_size & (sb->s_blocksize - 1)) != 0)
+               RETURN_EXIT;
+
+       iam_container_init(bag, &iam->od_descr, inode);
+       rc = iam_container_setup(bag);
+       if (rc)
+               GOTO(fini, rc = 1);
+
+       descr = bag->ic_descr;
+       /* May be regular file with IAM_LFIX_ROOT_MAGIC matched
+        * coincidentally, or corrupted index object, skip it. */
+       if (descr->id_ptr_size != 4)
+               GOTO(fini, rc = 1);
+
+       keysize = descr->id_key_size;
+       recsize = descr->id_rec_size;
+       rc = osd_index_register(osd, fid, keysize, recsize);
+
+       GOTO(fini, rc);
+
+fini:
+       iam_container_fini(bag);
+       if (!rc)
+               CDEBUG(D_LFSCK, "%s: index object "DFID" (%u/%u) registered\n",
+                      osd_name(osd), PFID(fid), keysize, recsize);
+}
+
+static void osd_index_restore(const struct lu_env *env, struct osd_device *dev,
+                             struct lustre_index_restore_unit *liru,
+                             void *buf, int bufsize)
+{
+       struct osd_thread_info *info = osd_oti_get(env);
+       struct osd_inode_id *id = &info->oti_id;
+       struct lu_fid *tgt_fid = &liru->liru_cfid;
+       struct inode *bak_inode = NULL;
+       struct ldiskfs_dir_entry_2 *de = NULL;
+       struct buffer_head *bh = NULL;
+       struct dentry *dentry;
+       char *name = buf;
+       struct lu_fid bak_fid;
+       int rc;
+       ENTRY;
+
+       lustre_fid2lbx(name, tgt_fid, bufsize);
+       dentry = osd_child_dentry_by_inode(env, dev->od_index_backup_inode,
+                                          name, strlen(name));
+       bh = osd_ldiskfs_find_entry(dev->od_index_backup_inode,
+                                   &dentry->d_name, &de, NULL, NULL);
+       if (IS_ERR(bh))
+               GOTO(log, rc = PTR_ERR(bh));
+
+       osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
+       brelse(bh);
+       bak_inode = osd_iget_fid(info, dev, id, &bak_fid);
+       if (IS_ERR(bak_inode))
+               GOTO(log, rc = PTR_ERR(bak_inode));
+
+       iput(bak_inode);
+       /* The OI mapping for index may be invalid, since it will be
+        * re-created, not update the OI mapping, just cache it in RAM. */
+       osd_id_gen(id, liru->liru_clid, OSD_OII_NOGEN);
+       osd_add_oi_cache(info, dev, id, tgt_fid);
+       rc = lustre_index_restore(env, &dev->od_dt_dev, &liru->liru_pfid,
+                                 tgt_fid, &bak_fid, liru->liru_name,
+                                 &dev->od_index_backup_list, &dev->od_lock,
+                                 buf, bufsize);
+       GOTO(log, rc);
+
+log:
+       CDEBUG(D_WARNING, "%s: restore index '%s' with "DFID": rc = %d\n",
+              osd_name(dev), liru->liru_name, PFID(tgt_fid), rc);
+}
+
 /**
  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
  *
@@ -1775,7 +1883,9 @@ osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
  */
 static int
 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
-                struct inode *inode, const struct lu_fid *fid, int flags)
+                struct inode *parent, struct inode *inode,
+                const struct lu_fid *fid, const char *name,
+                int namelen, int flags)
 {
        struct lustre_mdt_attrs *lma    = &info->oti_ost_attrs.loa_lma;
        struct osd_inode_id     *id     = &info->oti_id;
@@ -1786,6 +1896,12 @@ osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
        int                      rc;
        ENTRY;
 
+       if (!inode) {
+               CDEBUG(D_INODE, "%s: child '%.*s' lacks inode: rc = -2\n",
+                      osd_name(dev), namelen, name);
+               RETURN(-ENOENT);
+       }
+
        rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
                         &info->oti_ost_attrs);
        if (rc != 0 && rc != -ENODATA) {
@@ -1819,6 +1935,28 @@ osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
                        RETURN(0);
 
                tfid = lma->lma_self_fid;
+               if (lma->lma_compat & LMAC_IDX_BACKUP &&
+                   osd_index_need_recreate(info->oti_env, dev, inode)) {
+                       struct lu_fid *pfid = &info->oti_fid3;
+
+                       if (parent == osd_sb(dev)->s_root->d_inode) {
+                               lu_local_obj_fid(pfid, OSD_FS_ROOT_OID);
+                       } else {
+                               rc = osd_scrub_get_fid(info, dev, parent, pfid,
+                                                      false);
+                               if (rc)
+                                       RETURN(rc);
+                       }
+
+                       rc = lustre_liru_new(&dev->od_index_restore_list, pfid,
+                                       &tfid, inode->i_ino, name, namelen);
+
+                       RETURN(rc);
+               }
+
+               if (!(flags & OLF_NOT_BACKUP))
+                       osd_ios_index_register(info->oti_env, dev, &tfid,
+                                              inode);
        }
 
        rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
@@ -1838,8 +1976,7 @@ osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
                RETURN(0);
 
        if (!(sf->sf_flags & SF_INCONSISTENT)) {
-               scrub_file_reset(scrub, LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
-                                SF_INCONSISTENT);
+               scrub_file_reset(scrub, dev->od_uuid, SF_INCONSISTENT);
                rc = scrub_file_store(info->oti_env, scrub);
                if (rc != 0)
                        RETURN(rc);
@@ -1855,7 +1992,7 @@ osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
 
 /**
  * It scans the /lost+found, and for the OST-object (with filter_fid
- * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
+ * or filter_fid_18_23), move them back to its proper /O/<seq>/d<x>.
  */
 #ifdef HAVE_FILLDIR_USE_CTX
 static int osd_ios_lf_fill(struct dir_context *buf,
@@ -1878,6 +2015,8 @@ static int osd_ios_lf_fill(void *buf,
        int                         rc;
        ENTRY;
 
+       fill_buf->oifb_items++;
+
        /* skip any '.' started names */
        if (name[0] == '.')
                RETURN(0);
@@ -1888,6 +2027,11 @@ static int osd_ios_lf_fill(void *buf,
                CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
                      osd_name(dev), namelen, name, (int)PTR_ERR(child));
                RETURN(0);
+       } else if (!child->d_inode) {
+               dput(child);
+               CDEBUG(D_INODE, "%s: child '%.*s' lacks inode\n",
+                      osd_name(dev), namelen, name);
+               RETURN(0);
        }
 
        inode = child->d_inode;
@@ -1946,6 +2090,8 @@ static int osd_ios_varfid_fill(void *buf,
        int                         rc;
        ENTRY;
 
+       fill_buf->oifb_items++;
+
        /* skip any '.' started names */
        if (name[0] == '.')
                RETURN(0);
@@ -1954,8 +2100,9 @@ static int osd_ios_varfid_fill(void *buf,
        if (IS_ERR(child))
                RETURN(PTR_ERR(child));
 
-       rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
-                             NULL, 0);
+       rc = osd_ios_scan_one(fill_buf->oifb_info, dev,
+                             fill_buf->oifb_dentry->d_inode, child->d_inode,
+                             NULL, name, namelen, 0);
        if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
                rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
                                      osd_ios_varfid_fill);
@@ -1980,6 +2127,8 @@ static int osd_ios_dl_fill(void *buf,
        int                         rc       = 0;
        ENTRY;
 
+       fill_buf->oifb_items++;
+
        /* skip any '.' started names */
        if (name[0] == '.')
                RETURN(0);
@@ -1999,8 +2148,9 @@ static int osd_ios_dl_fill(void *buf,
        if (IS_ERR(child))
                RETURN(PTR_ERR(child));
 
-       rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
-                             &map->olm_fid, map->olm_flags);
+       rc = osd_ios_scan_one(fill_buf->oifb_info, dev,
+                             fill_buf->oifb_dentry->d_inode, child->d_inode,
+                             &map->olm_fid, name, namelen, map->olm_flags);
        dput(child);
 
        RETURN(rc);
@@ -2021,6 +2171,8 @@ static int osd_ios_uld_fill(void *buf,
        int                         rc       = 0;
        ENTRY;
 
+       fill_buf->oifb_items++;
+
        /* skip any non-DFID format name */
        if (name[0] != '[')
                RETURN(0);
@@ -2033,7 +2185,8 @@ static int osd_ios_uld_fill(void *buf,
        sscanf(&name[1], SFID, RFID(&tfid));
        if (fid_is_sane(&tfid))
                rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
-                                     child->d_inode, &tfid, 0);
+                                     fill_buf->oifb_dentry->d_inode,
+                                     child->d_inode, &tfid, name, namelen, 0);
        else
                rc = -EIO;
        dput(child);
@@ -2057,6 +2210,8 @@ static int osd_ios_root_fill(void *buf,
        int                         rc       = 0;
        ENTRY;
 
+       fill_buf->oifb_items++;
+
        /* skip any '.' started names */
        if (name[0] == '.')
                RETURN(0);
@@ -2075,13 +2230,17 @@ static int osd_ios_root_fill(void *buf,
        child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
        if (IS_ERR(child))
                RETURN(PTR_ERR(child));
+       else if (!child->d_inode)
+               GOTO(out_put, rc = -ENOENT);
 
        if (!(map->olm_flags & OLF_NO_OI))
-               rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
-                                     &map->olm_fid, map->olm_flags);
+               rc = osd_ios_scan_one(fill_buf->oifb_info, dev,
+                               fill_buf->oifb_dentry->d_inode, child->d_inode,
+                               &map->olm_fid, name, namelen, map->olm_flags);
        if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
                rc = osd_ios_new_item(dev, child, map->olm_scandir,
                                      map->olm_filldir);
+out_put:
        dput(child);
 
        RETURN(rc);
@@ -2114,13 +2273,21 @@ osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
        filp->private_data = NULL;
        set_file_inode(filp, inode);
 
+       do {
+               buf.oifb_items = 0;
 #ifdef HAVE_DIR_CONTEXT
-       buf.ctx.pos = filp->f_pos;
-       rc = fops->iterate(filp, &buf.ctx);
-       filp->f_pos = buf.ctx.pos;
+               buf.ctx.pos = filp->f_pos;
+#ifdef HAVE_ITERATE_SHARED
+               rc = fops->iterate_shared(filp, &buf.ctx);
 #else
-       rc = fops->readdir(filp, &buf, filldir);
+               rc = fops->iterate(filp, &buf.ctx);
 #endif
+               filp->f_pos = buf.ctx.pos;
+#else
+               rc = fops->readdir(filp, &buf, filldir);
+#endif
+       } while (rc >= 0 && buf.oifb_items > 0 &&
+                filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
        fops->release(inode, filp);
 
        RETURN(rc);
@@ -2152,20 +2319,20 @@ osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
        scrub->os_convert_igif = 1;
        child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
                                       strlen(dot_lustre_name));
-       if (IS_ERR(child)) {
+       if (IS_ERR(child) && PTR_ERR(child) != -ENOENT) {
                rc = PTR_ERR(child);
-               if (rc == -ENOENT) {
-                       /* It is 1.8 MDT device. */
-                       if (!(sf->sf_flags & SF_UPGRADE)) {
-                               scrub_file_reset(scrub,
-                                       LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
-                                       SF_UPGRADE);
-                               sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
-                               rc = scrub_file_store(info->oti_env, scrub);
-                       } else {
-                               rc = 0;
-                       }
+       } else if (IS_ERR(child) || !child->d_inode) {
+               /* It is 1.8 MDT device. */
+               if (!(sf->sf_flags & SF_UPGRADE)) {
+                       scrub_file_reset(scrub, dev->od_uuid,
+                                        SF_UPGRADE);
+                       sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
+                       rc = scrub_file_store(info->oti_env, scrub);
+               } else {
+                       rc = 0;
                }
+               if (!IS_ERR(child))
+                       dput(child);
        } else {
                /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
                 * so the client will get IGIF for the ".lustre" object when
@@ -2190,8 +2357,10 @@ osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
                 * Usually, it is rare case for the old connected clients
                 * to access the ".lustre" with cached IGIF. So we prefer
                 * to the solution 2). */
-               rc = osd_ios_scan_one(info, dev, child->d_inode,
-                                     &LU_DOT_LUSTRE_FID, 0);
+               rc = osd_ios_scan_one(info, dev, dentry->d_inode,
+                                     child->d_inode, &LU_DOT_LUSTRE_FID,
+                                     dot_lustre_name,
+                                     strlen(dot_lustre_name), 0);
                if (rc == 0)
                        rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
                                              osd_ios_dl_fill);
@@ -2219,54 +2388,50 @@ osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
        }
 
        child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
-       if (!IS_ERR(child)) {
-               rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
-               dput(child);
-       } else {
+       if (IS_ERR(child)) {
                rc = PTR_ERR(child);
+       } else {
+               rc = osd_ios_scan_one(info, dev, dentry->d_inode,
+                                     child->d_inode, NULL, ADMIN_USR,
+                                     strlen(ADMIN_USR), 0);
+               dput(child);
        }
 
        if (rc != 0 && rc != -ENOENT)
-               RETURN(rc);
+               GOTO(out, rc);
 
        child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
-       if (!IS_ERR(child)) {
-               rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
-               dput(child);
-       } else {
-               rc = PTR_ERR(child);
-       }
-
-       if (rc == -ENOENT)
-               rc = 0;
+       if (IS_ERR(child))
+               GOTO(out, rc = PTR_ERR(child));
 
-       RETURN(rc);
+       rc = osd_ios_scan_one(info, dev, dentry->d_inode,
+                             child->d_inode, NULL, ADMIN_GRP,
+                             strlen(ADMIN_GRP), 0);
+       dput(child);
+out:
+       RETURN(rc == -ENOENT ? 0 : rc);
 }
 
-static int osd_initial_OI_scrub(struct osd_thread_info *info,
-                               struct osd_device *dev)
+static void osd_initial_OI_scrub(struct osd_thread_info *info,
+                                struct osd_device *dev)
 {
        struct osd_ios_item     *item    = NULL;
        scandir_t                scandir = osd_ios_general_scan;
        filldir_t                filldir = osd_ios_root_fill;
        struct dentry           *dentry  = osd_sb(dev)->s_root;
        const struct osd_lf_map *map     = osd_lf_maps;
-       int                      rc;
        ENTRY;
 
        /* Lookup IGIF in OI by force for initial OI scrub. */
        dev->od_igif_inoi = 1;
 
        while (1) {
-               rc = scandir(info, dev, dentry, filldir);
+               scandir(info, dev, dentry, filldir);
                if (item != NULL) {
                        dput(item->oii_dentry);
                        OBD_FREE_PTR(item);
                }
 
-               if (rc != 0)
-                       break;
-
                if (list_empty(&dev->od_ios_list))
                        break;
 
@@ -2280,17 +2445,6 @@ static int osd_initial_OI_scrub(struct osd_thread_info *info,
                dentry = item->oii_dentry;
        }
 
-       while (!list_empty(&dev->od_ios_list)) {
-               item = list_entry(dev->od_ios_list.next,
-                                 struct osd_ios_item, oii_list);
-               list_del_init(&item->oii_list);
-               dput(item->oii_dentry);
-               OBD_FREE_PTR(item);
-       }
-
-       if (rc != 0)
-               RETURN(rc);
-
        /* There maybe the case that the object has been removed, but its OI
         * mapping is still in the OI file, such as the "CATALOGS" after MDT
         * file-level backup/restore. So here cleanup the stale OI mappings. */
@@ -2305,16 +2459,42 @@ static int osd_initial_OI_scrub(struct osd_thread_info *info,
                child = osd_ios_lookup_one_len(map->olm_name,
                                               osd_sb(dev)->s_root,
                                               map->olm_namelen);
-               if (!IS_ERR(child))
-                       dput(child);
-               else if (PTR_ERR(child) == -ENOENT)
+               if (PTR_ERR(child) == -ENOENT ||
+                   (!IS_ERR(child) && !child->d_inode))
                        osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
                                                  NULL, DTO_INDEX_DELETE,
                                                  true, 0, NULL);
+               if (!IS_ERR(child))
+                       dput(child);
                map++;
        }
 
-       RETURN(0);
+       if (!list_empty(&dev->od_index_restore_list)) {
+               char *buf;
+
+               OBD_ALLOC_LARGE(buf, INDEX_BACKUP_BUFSIZE);
+               if (!buf)
+                       CERROR("%s: not enough RAM for rebuild index\n",
+                              osd_name(dev));
+
+               while (!list_empty(&dev->od_index_restore_list)) {
+                       struct lustre_index_restore_unit *liru;
+
+                       liru = list_entry(dev->od_index_restore_list.next,
+                                         struct lustre_index_restore_unit,
+                                         liru_link);
+                       list_del(&liru->liru_link);
+                       if (buf)
+                               osd_index_restore(info->oti_env, dev, liru,
+                                                 buf, INDEX_BACKUP_BUFSIZE);
+                       OBD_FREE(liru, liru->liru_len);
+               }
+
+               if (buf)
+                       OBD_FREE_LARGE(buf, INDEX_BACKUP_BUFSIZE);
+       }
+
+       EXIT;
 }
 
 char *osd_lf_fid2name(const struct lu_fid *fid)
@@ -2363,7 +2543,7 @@ int osd_scrub_start(const struct lu_env *env, struct osd_device *dev,
        RETURN(rc);
 }
 
-static void osd_scrub_stop(struct osd_device *dev)
+void osd_scrub_stop(struct osd_device *dev)
 {
        struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
 
@@ -2385,12 +2565,11 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
        struct lvfs_run_ctxt *ctxt = &dev->od_scrub.os_ctxt;
        struct scrub_file *sf = &scrub->os_file;
        struct super_block *sb = osd_sb(dev);
-       struct ldiskfs_super_block *es = LDISKFS_SB(sb)->s_es;
        struct lvfs_run_ctxt saved;
        struct file *filp;
        struct inode *inode;
        struct lu_fid *fid = &info->oti_fid;
-       struct lu_object_conf conf;
+       struct osd_inode_id *id = &info->oti_id;
        struct dt_object *obj;
        bool dirty = false;
        bool restored = false;
@@ -2401,7 +2580,7 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
        OBD_SET_CTXT_MAGIC(ctxt);
        ctxt->pwdmnt = dev->od_mnt;
        ctxt->pwd = dev->od_mnt->mnt_root;
-       ctxt->fs = get_ds();
+       ctxt->fs = KERNEL_DS;
 
        init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
        init_rwsem(&scrub->os_rwsem);
@@ -2430,22 +2609,24 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
                }
        }
 
-       igrab(inode);
+       osd_id_gen(id, inode->i_ino, inode->i_generation);
+       osd_add_oi_cache(info, dev, id, fid);
        filp_close(filp, NULL);
        pop_ctxt(&saved, ctxt);
 
-       conf.loc_flags = LOC_F_NEW;
-       obj = lu2dt(lu_object_find_slice(env, osd2lu_dev(dev), fid, &conf));
-       if (IS_ERR_OR_NULL(obj)) {
-               iput(inode);
-               RETURN(obj == NULL ? -ENOENT : PTR_ERR(obj));
-       }
+       obj = lu2dt(lu_object_find_slice(env, osd2lu_dev(dev), fid, NULL));
+       if (IS_ERR_OR_NULL(obj))
+               RETURN(obj ? PTR_ERR(obj) : -ENOENT);
 
-       osd_dt_obj(obj)->oo_inode = inode;
+#ifndef HAVE_S_UUID_AS_UUID_T
+       memcpy(dev->od_uuid.b, sb->s_uuid, UUID_SIZE);
+#else
+       uuid_copy(&dev->od_uuid, &sb->s_uuid);
+#endif
        scrub->os_obj = obj;
        rc = scrub_file_load(env, scrub);
        if (rc == -ENOENT || rc == -EFAULT) {
-               scrub_file_init(scrub, es->s_uuid);
+               scrub_file_init(scrub, dev->od_uuid);
                /* If the "/O" dir does not exist when mount (indicated by
                 * osd_device::od_maybe_new), neither for the "/OI_scrub",
                 * then it is quite probably that the device is a new one,
@@ -2455,7 +2636,7 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
                 * an old device, it can be found and cleared later.
                 *
                 * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
-                * need to check "filter_fid_old" and to convert it to
+                * need to check "filter_fid_18_23" and to convert it to
                 * "filter_fid" for each object, and all the IGIF should
                 * have their FID mapping in OI files already. */
                if (dev->od_maybe_new && rc == -ENOENT)
@@ -2464,30 +2645,13 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
        } else if (rc < 0) {
                GOTO(cleanup_obj, rc);
        } else {
-               if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
-                       struct obd_uuid *old_uuid;
-                       struct obd_uuid *new_uuid;
-
-                       OBD_ALLOC_PTR(old_uuid);
-                       OBD_ALLOC_PTR(new_uuid);
-                       if (old_uuid == NULL || new_uuid == NULL) {
-                               CERROR("%s: UUID has been changed, but"
-                                      "failed to allocate RAM for report\n",
-                                      osd_dev2name(dev));
-                       } else {
-                               class_uuid_unparse(sf->sf_uuid, old_uuid);
-                               class_uuid_unparse(es->s_uuid, new_uuid);
-                               CDEBUG(D_LFSCK, "%s: UUID has been changed "
-                                      "from %s to %s\n", osd_dev2name(dev),
-                                      old_uuid->uuid, new_uuid->uuid);
-                       }
-                       scrub_file_reset(scrub, es->s_uuid, SF_INCONSISTENT);
+               if (!uuid_equal(&sf->sf_uuid, &dev->od_uuid)) {
+                       CDEBUG(D_LFSCK,
+                              "%s: UUID has been changed from %pU to %pU\n",
+                              osd_dev2name(dev), &sf->sf_uuid, &dev->od_uuid);
+                       scrub_file_reset(scrub, dev->od_uuid, SF_INCONSISTENT);
                        dirty = true;
                        restored = true;
-                       if (old_uuid != NULL)
-                               OBD_FREE_PTR(old_uuid);
-                       if (new_uuid != NULL)
-                               OBD_FREE_PTR(new_uuid);
                } else if (sf->sf_status == SS_SCANNING) {
                        sf->sf_status = SS_CRASHED;
                        dirty = true;
@@ -2518,11 +2682,8 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
        if (rc < 0)
                GOTO(cleanup_obj, rc);
 
-       if (!dev->od_dt_dev.dd_rdonly) {
-               rc = osd_initial_OI_scrub(info, dev);
-               if (rc)
-                       GOTO(cleanup_oi, rc);
-       }
+       if (!dev->od_dt_dev.dd_rdonly)
+               osd_initial_OI_scrub(info, dev);
 
        if (sf->sf_flags & SF_UPGRADE ||
            !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
@@ -2545,7 +2706,8 @@ int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
                 * later if found that the system is upgrading. */
                dev->od_igif_inoi = 1;
 
-       if (!dev->od_dt_dev.dd_rdonly && !dev->od_noscrub &&
+       if (!dev->od_dt_dev.dd_rdonly &&
+           dev->od_auto_scrub_interval != AS_NEVER &&
            ((sf->sf_status == SS_PAUSED) ||
             (sf->sf_status == SS_CRASHED &&
              sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
@@ -2885,7 +3047,7 @@ int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
        oii->oii_insert = insert;
 
        if (lscrub->os_partial_scan) {
-               __u64 now = cfs_time_current_sec();
+               __u64 now = ktime_get_real_seconds();
 
                /* If there haven't been errors in a long time,
                 * decay old count until either the errors are