Whamcloud - gitweb
LU-709 build: clean up percpu_counter and sb.s_time_gran
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_handler.c
index 0bf11de..c2e0e4f 100644 (file)
@@ -1,6 +1,4 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
  * GPL HEADER START
  *
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
@@ -43,9 +41,6 @@
  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  */
 
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
 #define DEBUG_SUBSYSTEM S_MDS
 
 #include <linux/module.h>
@@ -77,6 +72,8 @@
 
 /* llo_* api support */
 #include <md_object.h>
+/* dt_acct_features */
+#include <lquota.h>
 
 #ifdef HAVE_LDISKFS_PDO
 int ldiskfs_pdo = 1;
@@ -93,6 +90,7 @@ static const char remote_obj_dir[] = "REM_OBJ_DIR";
 static const struct lu_object_operations      osd_lu_obj_ops;
 static const struct dt_object_operations      osd_obj_ops;
 static const struct dt_object_operations      osd_obj_ea_ops;
+static const struct dt_object_operations      osd_obj_otable_it_ops;
 static const struct dt_index_operations       osd_index_iam_ops;
 static const struct dt_index_operations       osd_index_ea_ops;
 
@@ -108,11 +106,17 @@ static int osd_object_invariant(const struct lu_object *l)
 
 #ifdef HAVE_QUOTA_SUPPORT
 static inline void
-osd_push_ctxt(const struct lu_env *env, struct osd_ctxt *save)
+osd_push_ctxt(const struct lu_env *env, struct osd_ctxt *save, bool is_md)
 {
-        struct md_ucred *uc = md_ucred(env);
+       struct md_ucred *uc;
         struct cred     *tc;
 
+       if (!is_md)
+               /* OFD support */
+               return;
+
+       uc = md_ucred(env);
+
         LASSERT(uc != NULL);
 
         save->oc_uid = current_fsuid();
@@ -128,10 +132,14 @@ osd_push_ctxt(const struct lu_env *env, struct osd_ctxt *save)
 }
 
 static inline void
-osd_pop_ctxt(struct osd_ctxt *save)
+osd_pop_ctxt(struct osd_ctxt *save, bool is_md)
 {
         struct cred *tc;
 
+       if (!is_md)
+               /* OFD support */
+               return;
+
         if ((tc = prepare_creds())) {
                 tc->fsuid         = save->oc_uid;
                 tc->fsgid         = save->oc_gid;
@@ -248,92 +256,234 @@ static struct lu_object *osd_object_alloc(const struct lu_env *env,
         }
 }
 
-/*
- * retrieve object from backend ext fs.
- **/
-struct inode *osd_iget(struct osd_thread_info *info,
-                       struct osd_device *dev,
-                       const struct osd_inode_id *id)
+static int osd_get_lma(struct inode *inode, struct dentry *dentry,
+                      struct lustre_mdt_attrs *lma)
 {
-        struct inode *inode = NULL;
-
-        inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
-        if (IS_ERR(inode)) {
-                CERROR("Cannot get inode, rc = %li\n", PTR_ERR(inode));
-        } else if (id->oii_gen != OSD_OII_NOGEN &&
-                   inode->i_generation != id->oii_gen) {
-                iput(inode);
-                inode = ERR_PTR(-ESTALE);
-        } else if (inode->i_nlink == 0) {
-                /* due to parallel readdir and unlink,
-                * we can have dead inode here. */
-                CWARN("stale inode\n");
-                make_bad_inode(inode);
-                iput(inode);
-                inode = ERR_PTR(-ESTALE);
-        } else if (is_bad_inode(inode)) {
-                CERROR("bad inode %lx\n",inode->i_ino);
-                iput(inode);
-                inode = ERR_PTR(-ENOENT);
-        } else {
-                /* Do not update file c/mtime in ldiskfs.
-                 * NB: we don't have any lock to protect this because we don't
-                 * have reference on osd_object now, but contention with
-                 * another lookup + attr_set can't happen in the tiny window
-                 * between if (...) and set S_NOCMTIME. */
-                if (!(inode->i_flags & S_NOCMTIME))
-                        inode->i_flags |= S_NOCMTIME;
-        }
-        return inode;
-}
-
-static int osd_fid_lookup(const struct lu_env *env,
-                          struct osd_object *obj, const struct lu_fid *fid)
-{
-        struct osd_thread_info *info;
-        struct lu_device       *ldev = obj->oo_dt.do_lu.lo_dev;
-        struct osd_device      *dev;
-        struct osd_inode_id    *id;
-        struct inode           *inode;
-        int                     result;
+       int rc;
 
-        LINVRNT(osd_invariant(obj));
-        LASSERT(obj->oo_inode == NULL);
-        LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
-        /*
-         * This assertion checks that osd layer sees only local
-         * fids. Unfortunately it is somewhat expensive (does a
-         * cache-lookup). Disabling it for production/acceptance-testing.
-         */
-        LASSERT(1 || fid_is_local(env, ldev->ld_site, fid));
-
-        ENTRY;
+       dentry->d_inode = inode;
+       rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)lma,
+                                  sizeof(*lma));
+       if (rc > 0) {
+               /* Check LMA compatibility */
+               if (lma->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP)) {
+                       CWARN("%.16s: unsupported incompat LMA feature(s) "
+                             "%lx/%#x\n",
+                             LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
+                             inode->i_ino, le32_to_cpu(lma->lma_incompat) &
+                                                       ~LMA_INCOMPAT_SUPP);
+                       rc = -ENOSYS;
+               } else {
+                       lustre_lma_swab(lma);
+                       rc = 0;
+               }
+       } else if (rc == 0) {
+               rc = -ENODATA;
+       }
 
-        info = osd_oti_get(env);
-        LASSERT(info);
-        dev  = osd_dev(ldev);
-        id   = &info->oti_id;
-
-        if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
-                RETURN(-ENOENT);
+       return rc;
+}
 
-        result = osd_oi_lookup(info, dev, fid, id);
-        if (result != 0) {
-                if (result == -ENOENT)
-                        result = 0;
-                GOTO(out, result);
-        }
+/*
+ * retrieve object from backend ext fs.
+ **/
+struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
+                      struct osd_inode_id *id)
+{
+       struct inode *inode = NULL;
+
+       inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
+       if (IS_ERR(inode)) {
+               CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
+                      id->oii_ino, PTR_ERR(inode));
+       } else if (id->oii_gen != OSD_OII_NOGEN &&
+                  inode->i_generation != id->oii_gen) {
+               CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, "
+                      "gen1 = %u\n",
+                      id->oii_ino, id->oii_gen, inode->i_generation);
+               iput(inode);
+               inode = ERR_PTR(-ESTALE);
+       } else if (inode->i_nlink == 0) {
+               /* due to parallel readdir and unlink,
+               * we can have dead inode here. */
+               CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
+               make_bad_inode(inode);
+               iput(inode);
+               inode = ERR_PTR(-ESTALE);
+       } else if (is_bad_inode(inode)) {
+               CWARN("%.16s: bad inode: ino = %u\n",
+               LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
+               iput(inode);
+               inode = ERR_PTR(-ENOENT);
+       } else {
+               if (id->oii_gen == OSD_OII_NOGEN)
+                       osd_id_gen(id, inode->i_ino, inode->i_generation);
+
+               /* Do not update file c/mtime in ldiskfs.
+                * NB: we don't have any lock to protect this because we don't
+                * have reference on osd_object now, but contention with
+                * another lookup + attr_set can't happen in the tiny window
+                * between if (...) and set S_NOCMTIME. */
+               if (!(inode->i_flags & S_NOCMTIME))
+                       inode->i_flags |= S_NOCMTIME;
+       }
+       return inode;
+}
+
+struct inode *osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
+                          struct osd_inode_id *id, struct lu_fid *fid)
+{
+       struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
+       struct inode            *inode;
+       int                      rc;
+
+       inode = osd_iget(info, dev, id);
+       if (IS_ERR(inode))
+               return inode;
+
+       rc = osd_get_lma(inode, &info->oti_obj_dentry, lma);
+       if (rc == 0) {
+               *fid = lma->lma_self_fid;
+       } else if (rc == -ENODATA) {
+               LU_IGIF_BUILD(fid, inode->i_ino, inode->i_generation);
+       } else {
+               iput(inode);
+               inode = ERR_PTR(rc);
+       }
+       return inode;
+}
+
+static struct inode *
+osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev,
+               struct osd_inode_id *id, const struct lu_fid *fid)
+{
+       struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
+       struct inode            *inode;
+       int                      rc;
+
+       inode = osd_iget(info, dev, id);
+       if (IS_ERR(inode))
+               return inode;
+
+       rc = osd_get_lma(inode, &info->oti_obj_dentry, lma);
+       if (rc != 0) {
+               if (rc == -ENODATA) {
+                       CDEBUG(D_LFSCK, "inconsistent obj: NULL, %lu, "DFID"\n",
+                              inode->i_ino, PFID(fid));
+                       rc = -EREMCHG;
+               }
+               iput(inode);
+               return ERR_PTR(rc);
+       }
+
+       if (!lu_fid_eq(fid, &lma->lma_self_fid)) {
+               CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n",
+                      PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid));
+               iput(inode);
+               return ERR_PTR(EREMCHG);
+       }
+       return inode;
+}
+
+static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
+                         const struct lu_fid *fid,
+                         const struct lu_object_conf *conf)
+{
+       struct osd_thread_info *info;
+       struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
+       struct osd_device      *dev;
+       struct osd_idmap_cache *oic;
+       struct osd_inode_id    *id;
+       struct inode           *inode;
+       struct osd_scrub       *scrub;
+       struct scrub_file      *sf;
+       int                     result;
+       int                     verify = 0;
+       ENTRY;
+
+       LINVRNT(osd_invariant(obj));
+       LASSERT(obj->oo_inode == NULL);
+       LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
+
+       dev = osd_dev(ldev);
+       scrub = &dev->od_scrub;
+       sf = &scrub->os_file;
+       info = osd_oti_get(env);
+       LASSERT(info);
+       oic = &info->oti_cache;
+       id  = &oic->oic_lid;
+
+       if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
+               RETURN(-ENOENT);
+
+       if (fid_is_norm(fid)) {
+               /* Search order: 1. per-thread cache. */
+               if (lu_fid_eq(fid, &oic->oic_fid)) {
+                       goto iget;
+               } else if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
+                       /* Search order: 2. OI scrub pending list. */
+                       result = osd_oii_lookup(dev, fid, id);
+                       if (result == 0)
+                               goto iget;
+               }
+
+               if (sf->sf_flags & SF_INCONSISTENT)
+                       verify = 1;
+       }
+
+       fid_zero(&oic->oic_fid);
+
+       /*
+        * Objects are created as locking anchors or place holders for objects
+        * yet to be created. No need to osd_oi_lookup() at here because FID
+        * shouldn't never be re-used, if it's really a duplicate FID from
+        * unexpected reason, we should be able to detect it later by calling
+        * do_create->osd_oi_insert()
+        */
+       if (conf != NULL && (conf->loc_flags & LOC_F_NEW) != 0)
+               GOTO(out, result = 0);
+
+       /* Search order: 3. OI files. */
+       result = osd_oi_lookup(info, dev, fid, id);
+       if (result == -ENOENT) {
+               if (!fid_is_norm(fid) ||
+                   !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
+                                     sf->sf_oi_bitmap))
+                       GOTO(out, result = 0);
+
+               goto trigger;
+       }
+
+       if (result != 0)
+               GOTO(out, result);
+
+iget:
+       if (verify == 0)
+               inode = osd_iget(info, dev, id);
+       else
+               inode = osd_iget_verify(info, dev, id, fid);
+       if (IS_ERR(inode)) {
+               result = PTR_ERR(inode);
+               if (result == -ENOENT || result == -ESTALE) {
+                       result = 0;
+               } else if (result == -EREMCHG) {
+
+trigger:
+                       if (thread_is_running(&scrub->os_thread)) {
+                               result = -EINPROGRESS;
+                       } else if (!scrub->os_no_scrub) {
+                               result = osd_scrub_start(dev);
+                               LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC "
+                                              "for "DFID", rc = %d [1]\n",
+                                              LDISKFS_SB(osd_sb(dev))->s_es->\
+                                              s_volume_name,PFID(fid), result);
+                               if (result == 0 || result == -EALREADY)
+                                       result = -EINPROGRESS;
+                               else
+                                       result = -EREMCHG;
+                       }
+               }
 
-        inode = osd_iget(info, dev, id);
-        if (IS_ERR(inode)) {
-                /*
-                 * If fid wasn't found in oi, inode-less object is
-                 * created, for which lu_object_exists() returns
-                 * false. This is used in a (frequent) case when
-                 * objects are created as locking anchors or
-                 * place holders for objects yet to be created.
-                 */
-                result = PTR_ERR(inode);
                 GOTO(out, result);
         }
 
@@ -345,18 +495,20 @@ static int osd_fid_lookup(const struct lu_env *env,
         }
 
         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
-                goto out;
+               GOTO(out, result = 0);
+
+       LASSERT(obj->oo_hl_head == NULL);
+       obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
+       if (obj->oo_hl_head == NULL) {
+               obj->oo_inode = NULL;
+               iput(inode);
+               GOTO(out, result = -ENOMEM);
+       }
+       GOTO(out, result = 0);
 
-        LASSERT(obj->oo_hl_head == NULL);
-        obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
-        if (obj->oo_hl_head == NULL) {
-                obj->oo_inode = NULL;
-                iput(inode);
-                result = -ENOMEM;
-        }
 out:
-        LINVRNT(osd_invariant(obj));
-        RETURN(result);
+       LINVRNT(osd_invariant(obj));
+       return result;
 }
 
 /*
@@ -375,21 +527,26 @@ static void osd_object_init0(struct osd_object *obj)
  * life-cycle.
  */
 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
-                           const struct lu_object_conf *unused)
+                          const struct lu_object_conf *conf)
 {
-        struct osd_object *obj = osd_obj(l);
-        int result;
+       struct osd_object *obj = osd_obj(l);
+       int result;
 
-        LINVRNT(osd_invariant(obj));
+       LINVRNT(osd_invariant(obj));
 
-        result = osd_fid_lookup(env, obj, lu_object_fid(l));
-        obj->oo_dt.do_body_ops = &osd_body_ops_new;
-        if (result == 0) {
-                if (obj->oo_inode != NULL)
-                        osd_object_init0(obj);
-        }
-        LINVRNT(osd_invariant(obj));
-        return result;
+       result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
+       obj->oo_dt.do_body_ops = &osd_body_ops_new;
+       if (result == 0) {
+               if (obj->oo_inode != NULL) {
+                       osd_object_init0(obj);
+               } else if (fid_is_otable_it(&l->lo_header->loh_fid)) {
+                       obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
+                       /* LFSCK iterator object is special without inode */
+                       l->lo_header->loh_attr |= LOHA_EXISTS;
+               }
+       }
+       LINVRNT(osd_invariant(obj));
+       return result;
 }
 
 /*
@@ -556,9 +713,11 @@ static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
 
         dt_txn_hook_commit(th);
 
-        /* call per-transaction callbacks if any */
-        cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage)
-                dcb->dcb_func(NULL, th, dcb, error);
+       /* call per-transaction callbacks if any */
+       cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
+               cfs_list_del_init(&dcb->dcb_linkage);
+               dcb->dcb_func(NULL, th, dcb, error);
+       }
 
         lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
         lu_device_put(lud);
@@ -621,9 +780,10 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d,
                 GOTO(out, rc);
 
         if (!osd_param_is_sane(dev, th)) {
-                CWARN("%s: too many transaction credits (%d > %d)\n",
-                      d->dd_lu_dev.ld_obd->obd_name, oh->ot_credits,
-                      osd_journal(dev)->j_max_transaction_buffers);
+               CWARN("%.16s: too many transaction credits (%d > %d)\n",
+                     LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
+                     oh->ot_credits,
+                     osd_journal(dev)->j_max_transaction_buffers);
                 /* XXX Limit the credits to 'max_transaction_buffers', and
                  *     let the underlying filesystem to catch the error if
                  *     we really need so many credits.
@@ -670,7 +830,7 @@ int osd_trans_start(const struct lu_env *env, struct dt_device *d,
                  *      data (ofd) and reverse ordering for metadata
                  *      (mdd). then at some point we'll fix the latter
                  */
-                if (lu_device_is_md(&d->dd_lu_dev)) {
+               if (dev->od_is_md) {
                         LASSERT(oti->oti_r_locks == 0);
                         LASSERT(oti->oti_w_locks == 0);
                 }
@@ -723,7 +883,7 @@ static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
                  *      data (ofd) and reverse ordering for metadata
                  *      (mdd). then at some point we'll fix the latter
                  */
-                if (lu_device_is_md(&th->th_dev->dd_lu_dev)) {
+               if (osd_dt_dev(th->th_dev)->od_is_md) {
                         LASSERT(oti->oti_r_locks == 0);
                         LASSERT(oti->oti_w_locks == 0);
                 }
@@ -830,6 +990,9 @@ int osd_statfs(const struct lu_env *env, struct dt_device *d,
         struct kstatfs     *ksfs;
         int result = 0;
 
+       if (unlikely(osd->od_mount == NULL))
+               return -EINPROGRESS;
+
         /* osd_lproc.c call this without env, allocate ksfs for that case */
         if (unlikely(env == NULL)) {
                 OBD_ALLOC_PTR(ksfs);
@@ -873,7 +1036,7 @@ static void osd_conf_get(const struct lu_env *env,
          */
         param->ddp_max_name_len = LDISKFS_NAME_LEN;
         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
-        param->ddp_block_shift  = osd_sb(osd_dt_dev(dev))->s_blocksize_bits;
+       param->ddp_block_shift  = sb->s_blocksize_bits;
         param->ddp_mntopts      = 0;
         if (test_opt(sb, XATTR_USER))
                 param->ddp_mntopts |= MNTOPT_USERXATTR;
@@ -1000,7 +1163,7 @@ const int osd_dto_credits_noquota[DTO_NR] = {
         [DTO_INDEX_INSERT]  = 16,
         [DTO_INDEX_DELETE]  = 16,
         /**
-         * Unused now
+        * Used for OI scrub
          */
         [DTO_INDEX_UPDATE]  = 16,
         /**
@@ -1226,15 +1389,15 @@ int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
 }
 
 static struct timespec *osd_inode_time(const struct lu_env *env,
-                                       struct inode *inode, __u64 seconds)
+                                      struct inode *inode, __u64 seconds)
 {
-        struct osd_thread_info *oti = osd_oti_get(env);
-        struct timespec        *t   = &oti->oti_time;
+       struct osd_thread_info  *oti = osd_oti_get(env);
+       struct timespec         *t   = &oti->oti_time;
 
-        t->tv_sec  = seconds;
-        t->tv_nsec = 0;
-        *t = timespec_trunc(*t, get_sb_time_gran(inode->i_sb));
-        return t;
+       t->tv_sec = seconds;
+       t->tv_nsec = 0;
+       *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
+       return t;
 }
 
 
@@ -1361,6 +1524,32 @@ static int osd_inode_setattr(const struct lu_env *env,
         return 0;
 }
 
+static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
+{
+       if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
+           (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
+               struct iattr    iattr;
+               int             rc;
+
+               iattr.ia_valid = 0;
+               if (attr->la_valid & LA_UID)
+                       iattr.ia_valid |= ATTR_UID;
+               if (attr->la_valid & LA_GID)
+                       iattr.ia_valid |= ATTR_GID;
+               iattr.ia_uid = attr->la_uid;
+               iattr.ia_gid = attr->la_gid;
+
+               rc = ll_vfs_dq_transfer(inode, &iattr);
+               if (rc) {
+                       CERROR("%s: quota transfer failed: rc = %d. Is quota "
+                              "enforcement enabled on the ldiskfs filesystem?",
+                              inode->i_sb->s_id, rc);
+                       return rc;
+               }
+       }
+       return 0;
+}
+
 static int osd_attr_set(const struct lu_env *env,
                         struct dt_object *dt,
                         const struct lu_attr *attr,
@@ -1381,28 +1570,34 @@ static int osd_attr_set(const struct lu_env *env,
         OSD_EXEC_OP(handle, attr_set);
 
         inode = obj->oo_inode;
+       if (!osd_dt_dev(handle->th_dev)->od_is_md) {
+               /* OFD support */
+               rc = osd_quota_transfer(inode, attr);
+               if (rc)
+                       return rc;
+       } else {
 #ifdef HAVE_QUOTA_SUPPORT
-        if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
-            (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
-                struct osd_ctxt *save = &osd_oti_get(env)->oti_ctxt;
-                struct iattr iattr;
-                int rc;
-
-                iattr.ia_valid = 0;
-                if (attr->la_valid & LA_UID)
-                        iattr.ia_valid |= ATTR_UID;
-                if (attr->la_valid & LA_GID)
-                        iattr.ia_valid |= ATTR_GID;
-                iattr.ia_uid = attr->la_uid;
-                iattr.ia_gid = attr->la_gid;
-                osd_push_ctxt(env, save);
-                rc = ll_vfs_dq_transfer(inode, &iattr) ? -EDQUOT : 0;
-                osd_pop_ctxt(save);
-                if (rc != 0)
-                        return rc;
-        }
+               if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
+                   (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
+                       struct osd_ctxt *save = &osd_oti_get(env)->oti_ctxt;
+                       struct           iattr iattr;
+                       int              rc;
+
+                       iattr.ia_valid = 0;
+                       if (attr->la_valid & LA_UID)
+                               iattr.ia_valid |= ATTR_UID;
+                       if (attr->la_valid & LA_GID)
+                               iattr.ia_valid |= ATTR_GID;
+                       iattr.ia_uid = attr->la_uid;
+                       iattr.ia_gid = attr->la_gid;
+                       osd_push_ctxt(env, save, 1);
+                       rc = ll_vfs_dq_transfer(inode, &iattr) ? -EDQUOT : 0;
+                       osd_pop_ctxt(save, 1);
+                       if (rc != 0)
+                               return rc;
+               }
 #endif
-
+       }
         cfs_spin_lock(&obj->oo_guard);
         rc = osd_inode_setattr(env, inode, attr);
         cfs_spin_unlock(&obj->oo_guard);
@@ -1412,26 +1607,6 @@ static int osd_attr_set(const struct lu_env *env,
         return rc;
 }
 
-/*
- * Object creation.
- *
- * XXX temporary solution.
- */
-static int osd_create_pre(struct osd_thread_info *info, struct osd_object *obj,
-                          struct lu_attr *attr, struct thandle *th)
-{
-        return 0;
-}
-
-static int osd_create_post(struct osd_thread_info *info, struct osd_object *obj,
-                           struct lu_attr *attr, struct thandle *th)
-{
-        osd_object_init0(obj);
-        if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
-                unlock_new_inode(obj->oo_inode);
-        return 0;
-}
-
 struct dentry *osd_child_dentry_get(const struct lu_env *env,
                                     struct osd_object *obj,
                                     const char *name, const int namelen)
@@ -1470,20 +1645,21 @@ static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
                 parent = hint->dah_parent;
 
 #ifdef HAVE_QUOTA_SUPPORT
-        osd_push_ctxt(info->oti_env, save);
+       osd_push_ctxt(info->oti_env, save, osd_dt_dev(th->th_dev)->od_is_md);
 #endif
         inode = ldiskfs_create_inode(oth->ot_handle,
                                      parent ? osd_dt_obj(parent)->oo_inode :
                                               osd_sb(osd)->s_root->d_inode,
                                      mode);
 #ifdef HAVE_QUOTA_SUPPORT
-        osd_pop_ctxt(save);
+       osd_pop_ctxt(save, osd_dt_dev(th->th_dev)->od_is_md);
 #endif
         if (!IS_ERR(inode)) {
                 /* Do not update file c/mtime in ldiskfs.
                  * NB: don't need any lock because no contention at this
                  * early stage */
                 inode->i_flags |= S_NOCMTIME;
+               inode->i_state |= I_LUSTRE_NOSCRUB;
                 obj->oo_inode = inode;
                 result = 0;
         } else {
@@ -1540,7 +1716,7 @@ static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
         struct osd_thandle *oth;
         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
 
-        __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
+        __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
 
         LASSERT(S_ISREG(attr->la_mode));
 
@@ -1575,7 +1751,7 @@ static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
 {
         LASSERT(S_ISREG(attr->la_mode));
         return osd_mkfile(info, obj, (attr->la_mode &
-                               (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
+                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
 }
 
 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
@@ -1586,7 +1762,7 @@ static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
 {
         LASSERT(S_ISLNK(attr->la_mode));
         return osd_mkfile(info, obj, (attr->la_mode &
-                              (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
+                              (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
 }
 
 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
@@ -1595,7 +1771,7 @@ static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
                      struct dt_object_format *dof,
                      struct thandle *th)
 {
-        cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX);
+        cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
         int result;
 
         LINVRNT(osd_invariant(obj));
@@ -1606,6 +1782,10 @@ static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
         result = osd_mkfile(info, obj, mode, hint, th);
         if (result == 0) {
                 LASSERT(obj->oo_inode != NULL);
+               /*
+                * This inode should be marked dirty for i_rdev.  Currently
+                * that is done in the osd_attr_init().
+                */
                 init_special_inode(obj->oo_inode, mode, attr->la_rdev);
         }
         LINVRNT(osd_invariant(obj));
@@ -1657,6 +1837,51 @@ static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
         ah->dah_mode = child_mode;
 }
 
+static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
+                         struct lu_attr *attr, struct dt_object_format *dof)
+{
+       struct inode   *inode = obj->oo_inode;
+       __u64           valid = attr->la_valid;
+       int             result;
+
+       attr->la_valid &= ~(LA_TYPE | LA_MODE);
+
+        if (dof->dof_type != DFT_NODE)
+                attr->la_valid &= ~LA_RDEV;
+        if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
+                attr->la_valid &= ~LA_ATIME;
+        if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
+                attr->la_valid &= ~LA_CTIME;
+        if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
+                attr->la_valid &= ~LA_MTIME;
+
+       if (!osd_obj2dev(obj)->od_is_md) {
+               /* OFD support */
+               result = osd_quota_transfer(inode, attr);
+               if (result)
+                       return;
+       } else {
+#ifdef HAVE_QUOTA_SUPPORT
+               attr->la_valid &= ~(LA_UID | LA_GID);
+#endif
+       }
+
+        if (attr->la_valid != 0) {
+                result = osd_inode_setattr(info->oti_env, inode, attr);
+                /*
+                 * The osd_inode_setattr() should always succeed here.  The
+                 * only error that could be returned is EDQUOT when we are
+                 * trying to change the UID or GID of the inode. However, this
+                 * should not happen since quota enforcement is no longer
+                 * enabled on ldiskfs (lquota takes care of it).
+                 */
+                LASSERTF(result == 0, "%d", result);
+                inode->i_sb->s_op->dirty_inode(inode);
+        }
+
+        attr->la_valid = valid;
+}
+
 /**
  * Helper function for osd_object_create()
  *
@@ -1668,16 +1893,26 @@ static int __osd_object_create(struct osd_thread_info *info,
                                struct dt_object_format *dof,
                                struct thandle *th)
 {
+       int     result;
+       __u32   umask;
 
-        int result;
+       /* we drop umask so that permissions we pass are not affected */
+       umask = current->fs->umask;
+       current->fs->umask = 0;
 
-        result = osd_create_pre(info, obj, attr, th);
+       result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
+                                                 th);
         if (result == 0) {
-                result = osd_create_type_f(dof->dof_type)(info, obj,
-                                           attr, hint, dof, th);
-                if (result == 0)
-                        result = osd_create_post(info, obj, attr, th);
+               osd_attr_init(info, obj, attr, dof);
+               osd_object_init0(obj);
+               /* bz 24037 */
+               if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
+                       unlock_new_inode(obj->oo_inode);
         }
+
+       /* restore previous umask value */
+       current->fs->umask = umask;
+
         return result;
 }
 
@@ -1692,16 +1927,16 @@ static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
         struct osd_thread_info *info = osd_oti_get(env);
         struct osd_inode_id    *id   = &info->oti_id;
         struct osd_device      *osd  = osd_obj2dev(obj);
-        struct md_ucred        *uc   = md_ucred(env);
 
         LASSERT(obj->oo_inode != NULL);
-        LASSERT(uc != NULL);
 
-        id->oii_ino = obj->oo_inode->i_ino;
-        id->oii_gen = obj->oo_inode->i_generation;
+       if (osd->od_is_md) {
+               struct md_ucred *uc = md_ucred(env);
+               LASSERT(uc != NULL);
+       }
 
-        return osd_oi_insert(info, osd, fid, id, th,
-                             uc->mu_cap & CFS_CAP_SYS_RESOURCE_MASK);
+       osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
+       return osd_oi_insert(info, osd, fid, id, th);
 }
 
 static int osd_declare_object_create(const struct lu_env *env,
@@ -1762,6 +1997,11 @@ static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
         LASSERT(osd_write_locked(env, obj));
         LASSERT(th != NULL);
 
+       if (unlikely(fid_is_acct(fid)))
+               /* Quota files can't be created from the kernel any more,
+                * 'tune2fs -O quota' will take care of creating them */
+               RETURN(-EPERM);
+
         OSD_EXEC_OP(th, create);
 
         result = __osd_object_create(info, obj, attr, hint, dof, th);
@@ -1820,6 +2060,12 @@ static int osd_object_destroy(const struct lu_env *env,
         LASSERT(inode);
         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
 
+       if (unlikely(fid_is_acct(fid)))
+               RETURN(-EPERM);
+
+       /* Parallel control for OI scrub. For most of cases, there is no
+        * lock contention. So it will not affect unlink performance. */
+       cfs_mutex_lock(&inode->i_mutex);
         if (S_ISDIR(inode->i_mode)) {
                 LASSERT(osd_inode_unlinked(inode) ||
                         inode->i_nlink == 1);
@@ -1834,6 +2080,7 @@ static int osd_object_destroy(const struct lu_env *env,
         OSD_EXEC_OP(th, destroy);
 
         result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
+       cfs_mutex_unlock(&inode->i_mutex);
 
         /* XXX: add to ext3 orphan list */
         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
@@ -1859,7 +2106,6 @@ static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
 
         LASSERT(dt_object_exists(dt));
         LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
-        LASSERT(osd_write_locked(env, obj));
 
         if (fl & LU_XATTR_REPLACE)
                 fs_flags |= XATTR_REPLACE;
@@ -1898,15 +2144,6 @@ static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt,
 }
 
 /**
- * Helper function to form igif
- */
-static inline void osd_igif_get(const struct lu_env *env, struct inode  *inode,
-                                struct lu_fid *fid)
-{
-        LU_IGIF_BUILD(fid, inode->i_ino, inode->i_generation);
-}
-
-/**
  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
  * To have compatilibility with 1.8 ldiskfs driver we need to have
@@ -1933,54 +2170,20 @@ void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
  * \retval 0 on success
  */
 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
-                          __u32 ino, struct lu_fid *fid)
+                         __u32 ino, struct lu_fid *fid,
+                         struct osd_inode_id *id)
 {
-        struct osd_thread_info  *info      = osd_oti_get(env);
-        struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
-        struct lu_device        *ldev   = obj->oo_dt.do_lu.lo_dev;
-        struct dentry           *dentry = &info->oti_child_dentry;
-        struct osd_inode_id     *id     = &info->oti_id;
-        struct osd_device       *dev;
-        struct inode            *inode;
-        int                      rc;
+       struct osd_thread_info *info  = osd_oti_get(env);
+       struct inode           *inode;
+       ENTRY;
 
-        ENTRY;
-        dev  = osd_dev(ldev);
-
-        id->oii_ino = ino;
-        id->oii_gen = OSD_OII_NOGEN;
-
-        inode = osd_iget(info, dev, id);
-        if (IS_ERR(inode)) {
-                rc = PTR_ERR(inode);
-                GOTO(out,rc);
-        }
-        dentry->d_inode = inode;
+       osd_id_gen(id, ino, OSD_OII_NOGEN);
+       inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
+       if (IS_ERR(inode))
+               RETURN(PTR_ERR(inode));
 
-        LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
-        rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)mdt_attrs,
-                                   sizeof *mdt_attrs);
-
-        /* Check LMA compatibility */
-        if (rc > 0 &&
-            (mdt_attrs->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP))) {
-                CWARN("Inode %lx: Unsupported incompat LMA feature(s) %#x\n",
-                      inode->i_ino, le32_to_cpu(mdt_attrs->lma_incompat) &
-                      ~LMA_INCOMPAT_SUPP);
-                return -ENOSYS;
-        }
-
-        if (rc > 0) {
-                lustre_lma_swab(mdt_attrs);
-                memcpy(fid, &mdt_attrs->lma_self_fid, sizeof(*fid));
-                rc = 0;
-        } else if (rc == -ENODATA) {
-                osd_igif_get(env, inode, fid);
-                rc = 0;
-        }
-        iput(inode);
-out:
-        RETURN(rc);
+       iput(inode);
+       RETURN(0);
 }
 
 /**
@@ -2009,6 +2212,11 @@ static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
         LASSERT(osd_write_locked(env, obj));
         LASSERT(th != NULL);
 
+       if (unlikely(fid_is_acct(fid)))
+               /* Quota files can't be created from the kernel any more,
+                * 'tune2fs -O quota' will take care of creating them */
+               RETURN(-EPERM);
+
         OSD_EXEC_OP(th, create);
 
         result = __osd_object_create(info, obj, attr, hint, dof, th);
@@ -2075,7 +2283,7 @@ static int osd_object_ref_add(const struct lu_env *env,
                     inode->i_nlink == 2)
                         inode->i_nlink = 1;
         }
-        LASSERT(inode->i_nlink < LDISKFS_LINK_MAX);
+        LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
         cfs_spin_unlock(&obj->oo_guard);
         inode->i_sb->s_op->dirty_inode(inode);
         LINVRNT(osd_invariant(obj));
@@ -2503,10 +2711,10 @@ static int osd_iam_container_init(const struct lu_env *env,
 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
                          const struct dt_index_features *feat)
 {
-        int result;
-        int ea_dir = 0;
-        struct osd_object *obj = osd_dt_obj(dt);
-        struct osd_device *osd = osd_obj2dev(obj);
+       int                      result;
+       int                      skip_iam = 0;
+       struct osd_object       *obj = osd_dt_obj(dt);
+       struct osd_device       *osd = osd_obj2dev(obj);
 
         LINVRNT(osd_invariant(obj));
         LASSERT(dt_object_exists(dt));
@@ -2520,7 +2728,14 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
                         result = 0;
                 else
                         result = -ENOTDIR;
-                ea_dir = 1;
+               skip_iam = 1;
+       } else if (unlikely(feat == &dt_otable_features)) {
+               dt->do_index_ops = &osd_otable_ops;
+               return 0;
+       } else if (feat == &dt_acct_features) {
+               dt->do_index_ops = &osd_acct_index_ops;
+               result = 0;
+               skip_iam = 1;
         } else if (!osd_has_index(obj)) {
                 struct osd_directory *dir;
 
@@ -2556,7 +2771,7 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
                 result = 0;
         }
 
-        if (result == 0 && ea_dir == 0) {
+       if (result == 0 && skip_iam == 0) {
                 if (!osd_iam_index_probe(env, obj, feat))
                         result = -ENOTDIR;
         }
@@ -2565,6 +2780,15 @@ static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
         return result;
 }
 
+static int osd_otable_it_attr_get(const struct lu_env *env,
+                                struct dt_object *dt,
+                                struct lu_attr *attr,
+                                struct lustre_capa *capa)
+{
+       attr->la_valid = 0;
+       return 0;
+}
+
 static const struct dt_object_operations osd_obj_ops = {
         .do_read_lock         = osd_object_read_lock,
         .do_write_lock        = osd_object_write_lock,
@@ -2629,6 +2853,11 @@ static const struct dt_object_operations osd_obj_ea_ops = {
         .do_data_get          = osd_data_get,
 };
 
+static const struct dt_object_operations osd_obj_otable_it_ops = {
+       .do_attr_get    = osd_otable_it_attr_get,
+       .do_index_try   = osd_index_try,
+};
+
 static int osd_index_declare_iam_delete(const struct lu_env *env,
                                         struct dt_object *dt,
                                         const struct dt_key *key,
@@ -2914,7 +3143,7 @@ static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
         LASSERT(th != NULL);
 
         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
-                return -EACCES;
+               RETURN(-EACCES);
 
         OSD_EXEC_OP(th, insert);
 
@@ -3097,6 +3326,52 @@ static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
         return rc;
 }
 
+static int
+osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
+                     struct osd_idmap_cache *oic)
+{
+       struct osd_scrub    *scrub = &dev->od_scrub;
+       struct lu_fid       *fid   = &oic->oic_fid;
+       struct osd_inode_id *id    = &oti->oti_id;
+       int                  once  = 0;
+       int                  rc;
+       ENTRY;
+
+again:
+       rc = osd_oi_lookup(oti, dev, fid, id);
+       if (rc != 0 && rc != -ENOENT)
+               RETURN(rc);
+
+       if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
+               RETURN(0);
+
+       if (thread_is_running(&scrub->os_thread)) {
+               rc = osd_oii_insert(dev, oic, rc == -ENOENT);
+               /* There is race condition between osd_oi_lookup and OI scrub.
+                * The OI scrub finished just after osd_oi_lookup() failure.
+                * Under such case, it is unnecessary to trigger OI scrub again,
+                * but try to call osd_oi_lookup() again. */
+               if (unlikely(rc == -EAGAIN))
+                       goto again;
+
+               RETURN(rc);
+       }
+
+       if (!scrub->os_no_scrub && ++once == 1) {
+               CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n",
+                      PFID(fid));
+               rc = osd_scrub_start(dev);
+               LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID
+                              ", rc = %d [2]\n",
+                              LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
+                              PFID(fid), rc);
+               if (rc == 0)
+                       goto again;
+       }
+
+       RETURN(rc = -EREMCHG);
+}
+
 /**
  * Calls ->lookup() to find dentry. From dentry get inode and
  * read inode's ea to get fid. This is required for  interoperability
@@ -3132,22 +3407,43 @@ static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
 
         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
         if (bh) {
-                ino = le32_to_cpu(de->inode);
-                rc = osd_get_fid_from_dentry(de, rec);
-
-                /* done with de, release bh */
-                brelse(bh);
-                if (rc != 0)
-                        rc = osd_ea_fid_get(env, obj, ino, fid);
-        } else {
-                rc = -ENOENT;
-        }
+               struct osd_thread_info *oti = osd_oti_get(env);
+               struct osd_idmap_cache *oic = &oti->oti_cache;
+               struct osd_device *dev = osd_obj2dev(obj);
+               struct osd_scrub *scrub = &dev->od_scrub;
+               struct scrub_file *sf = &scrub->os_file;
+
+               ino = le32_to_cpu(de->inode);
+               rc = osd_get_fid_from_dentry(de, rec);
+
+               /* done with de, release bh */
+               brelse(bh);
+               if (rc != 0)
+                       rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
+               else
+                       osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
+
+               if (rc != 0 || !fid_is_norm(fid))
+                       GOTO(out, rc);
+
+               oic->oic_fid = *fid;
+               if ((scrub->os_pos_current <= ino) &&
+                   (sf->sf_flags & SF_INCONSISTENT ||
+                    ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
+                                     sf->sf_oi_bitmap)))
+                       rc = osd_consistency_check(oti, dev, oic);
+       } else {
+               rc = -ENOENT;
+       }
+
+       GOTO(out, rc);
 
-        if (hlock != NULL)
-                ldiskfs_htree_unlock(hlock);
-        else
-                cfs_up_read(&obj->oo_ext_idx_sem);
-        RETURN (rc);
+out:
+       if (hlock != NULL)
+               ldiskfs_htree_unlock(hlock);
+       else
+               cfs_up_read(&obj->oo_ext_idx_sem);
+       return rc;
 }
 
 /**
@@ -3176,7 +3472,7 @@ struct osd_object *osd_object_find(const struct lu_env *env,
                         else
                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
                                                 "lu_object can't be located"
-                                                ""DFID"\n", PFID(fid));
+                                               DFID"\n", PFID(fid));
 
                         if (child == NULL) {
                                 lu_object_put(env, luch);
@@ -3187,6 +3483,7 @@ struct osd_object *osd_object_find(const struct lu_env *env,
                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
                                         "lu_object does not exists "DFID"\n",
                                         PFID(fid));
+                       lu_object_put(env, luch);
                         child = ERR_PTR(-ENOENT);
                 }
         } else
@@ -3455,37 +3752,44 @@ static int osd_it_iam_rec(const struct lu_env *env,
                           const struct dt_it *di,
                           struct dt_rec *dtrec, __u32 attr)
 {
-        struct osd_it_iam *it        = (struct osd_it_iam *)di;
-        struct osd_thread_info *info = osd_oti_get(env);
-        struct lu_fid     *fid       = &info->oti_fid;
-        const struct osd_fid_pack *rec;
-        struct lu_dirent *lde = (struct lu_dirent *)dtrec;
-        char *name;
-        int namelen;
-        __u64 hash;
-        int rc;
+       struct osd_it_iam      *it   = (struct osd_it_iam *)di;
+       struct osd_thread_info *info = osd_oti_get(env);
+       ENTRY;
 
-        name = (char *)iam_it_key_get(&it->oi_it);
-        if (IS_ERR(name))
-                RETURN(PTR_ERR(name));
+       if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
+               const struct osd_fid_pack *rec;
+               struct lu_fid             *fid = &info->oti_fid;
+               struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
+               char                      *name;
+               int                        namelen;
+               __u64                      hash;
+               int                        rc;
 
-        namelen = iam_it_key_size(&it->oi_it);
+               name = (char *)iam_it_key_get(&it->oi_it);
+               if (IS_ERR(name))
+                       RETURN(PTR_ERR(name));
 
-        rec = (const struct osd_fid_pack *) iam_it_rec_get(&it->oi_it);
-        if (IS_ERR(rec))
-                RETURN(PTR_ERR(rec));
+               namelen = iam_it_key_size(&it->oi_it);
 
-        rc = osd_fid_unpack(fid, rec);
-        if (rc)
-                RETURN(rc);
+               rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
+               if (IS_ERR(rec))
+                       RETURN(PTR_ERR(rec));
 
-        hash = iam_it_store(&it->oi_it);
+               rc = osd_fid_unpack(fid, rec);
+               if (rc)
+                       RETURN(rc);
 
-        /* IAM does not store object type in IAM index (dir) */
-        osd_it_pack_dirent(lde, fid, hash, name, namelen,
-                           0, LUDA_FID);
+               hash = iam_it_store(&it->oi_it);
 
-        return 0;
+               /* IAM does not store object type in IAM index (dir) */
+               osd_it_pack_dirent(lde, fid, hash, name, namelen,
+                                  0, LUDA_FID);
+       } else {
+               iam_reccpy(&it->oi_it.ii_path.ip_leaf,
+                          (struct iam_rec *)dtrec);
+       }
+
+       RETURN(0);
 }
 
 /**
@@ -3806,27 +4110,45 @@ static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
  * \retval -ve on error
  */
 static inline int osd_it_ea_rec(const struct lu_env *env,
-                                const struct dt_it *di,
-                                struct dt_rec *dtrec, __u32 attr)
-{
-        struct osd_it_ea        *it     = (struct osd_it_ea *)di;
-        struct osd_object       *obj    = it->oie_obj;
-        struct lu_fid           *fid    = &it->oie_dirent->oied_fid;
-        struct lu_dirent        *lde    = (struct lu_dirent *)dtrec;
-        int    rc = 0;
-
-        ENTRY;
-
-        if (!fid_is_sane(fid))
-                rc = osd_ea_fid_get(env, obj, it->oie_dirent->oied_ino, fid);
-
-        if (rc == 0)
-                osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
-                                   it->oie_dirent->oied_name,
-                                   it->oie_dirent->oied_namelen,
-                                   it->oie_dirent->oied_type,
-                                   attr);
-        RETURN(rc);
+                               const struct dt_it *di,
+                               struct dt_rec *dtrec, __u32 attr)
+{
+       struct osd_it_ea       *it    = (struct osd_it_ea *)di;
+       struct osd_object      *obj   = it->oie_obj;
+       struct osd_device      *dev   = osd_obj2dev(obj);
+       struct osd_scrub       *scrub = &dev->od_scrub;
+       struct scrub_file      *sf    = &scrub->os_file;
+       struct osd_thread_info *oti   = osd_oti_get(env);
+       struct osd_idmap_cache *oic   = &oti->oti_cache;
+       struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
+       struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
+       __u32                   ino   = it->oie_dirent->oied_ino;
+       int                     rc    = 0;
+       ENTRY;
+
+       if (!fid_is_sane(fid)) {
+               rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
+               if (rc != 0)
+                       RETURN(rc);
+       } else {
+               osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
+       }
+
+       osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
+                          it->oie_dirent->oied_name,
+                          it->oie_dirent->oied_namelen,
+                          it->oie_dirent->oied_type, attr);
+
+       if (!fid_is_norm(fid))
+               RETURN(0);
+
+       oic->oic_fid = *fid;
+       if ((scrub->os_pos_current <= ino) &&
+           (sf->sf_flags & SF_INCONSISTENT ||
+            ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
+               rc = osd_consistency_check(oti, dev, oic);
+
+       RETURN(rc);
 }
 
 /**
@@ -3987,24 +4309,30 @@ struct lu_context_key osd_key = {
 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
                            const char *name, struct lu_device *next)
 {
-        return osd_procfs_init(osd_dev(d), name);
+       struct osd_device *osd = osd_dev(d);
+
+       strncpy(osd->od_svname, name, MAX_OBD_NAME);
+       return osd_procfs_init(osd, name);
 }
 
 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
 {
-        struct osd_thread_info *info = osd_oti_get(env);
+       ENTRY;
 
-        ENTRY;
+       osd_scrub_cleanup(env, o);
 
-        if (o->od_oi_table != NULL)
-                osd_oi_fini(info, o);
+       if (o->od_fsops) {
+               fsfilt_put_ops(o->od_fsops);
+               o->od_fsops = NULL;
+       }
 
-        if (o->od_fsops) {
-                fsfilt_put_ops(o->od_fsops);
-                o->od_fsops = NULL;
-        }
+       /* shutdown quota slave instance associated with the device */
+       if (o->od_quota_slave != NULL) {
+               qsd_fini(env, o->od_quota_slave);
+               o->od_quota_slave = NULL;
+       }
 
-        RETURN(0);
+       RETURN(0);
 }
 
 static int osd_mount(const struct lu_env *env,
@@ -4101,6 +4429,7 @@ static struct lu_device *osd_device_alloc(const struct lu_env *env,
                         l->ld_ops = &osd_lu_ops;
                         o->od_dt_dev.dd_ops = &osd_dt_ops;
                         cfs_spin_lock_init(&o->od_osfs_lock);
+                       cfs_mutex_init(&o->od_otable_mutex);
                         o->od_osfs_age = cfs_time_shift_64(-1000);
                         o->od_capa_hash = init_capa_hash();
                         if (o->od_capa_hash == NULL) {
@@ -4141,6 +4470,7 @@ static int osd_process_config(const struct lu_env *env,
                 err = osd_mount(env, o, cfg);
                 break;
         case LCFG_CLEANUP:
+               lu_dev_del_linkage(d->ld_site, d);
                 err = osd_shutdown(env, o);
                 break;
         default:
@@ -4159,21 +4489,45 @@ static int osd_recovery_complete(const struct lu_env *env,
 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
                        struct lu_device *dev)
 {
-        struct osd_device      *osd = osd_dev(dev);
-        struct osd_thread_info *oti = osd_oti_get(env);
-        int                     result;
+       struct osd_device *osd = osd_dev(dev);
+       int                result;
+       ENTRY;
 
-        ENTRY;
-
-        /* 1. initialize oi before any file create or file open */
-        result = osd_oi_init(oti, osd);
+       /* 1. setup scrub, including OI files initialization */
+       result = osd_scrub_setup(env, osd);
         if (result < 0)
                 RETURN(result);
 
-        if (!lu_device_is_md(pdev))
+       /* 2. setup quota slave instance */
+       osd->od_quota_slave = qsd_init(env, osd->od_svname, &osd->od_dt_dev,
+                                      osd->od_proc_entry);
+       if (IS_ERR(osd->od_quota_slave)) {
+               result = PTR_ERR(osd->od_quota_slave);
+               osd->od_quota_slave = NULL;
+               RETURN(result);
+       }
+
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,3,50,0)
+       /* Unfortunately, the current MDD implementation relies on some specific
+        * code to be executed in the OSD layer. Since OFD now also uses the OSD
+        * module, we need a way to skip the metadata-specific code when running
+        * with OFD.
+        * The hack here is to check the type of the parent device which is
+        * either MD (i.e. MDD device) with the current MDT stack or DT (i.e.
+        * OFD device) on an OST. As a reminder, obdfilter does not use the OSD
+        * layer and still relies on lvfs. This hack won't work any more when
+        * LOD is landed since LOD is of DT type.
+        * This code should be removed once the orion MDT changes (LOD/OSP, ...)
+        * have been landed */
+       osd->od_is_md = lu_device_is_md(pdev);
+#else
+#warning "all is_md checks must be removed from osd-ldiskfs"
+#endif
+
+        if (!osd->od_is_md)
                 RETURN(0);
 
-        /* 2. setup local objects */
+        /* 3. setup local objects */
         result = llo_local_objects_setup(env, lu2md_dev(pdev), lu2dt_dev(dev));
         RETURN(result);
 }