X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_compat.c;h=ae42d2c86da7ec53432d092c55bdbf2f78bbad3e;hp=e2d6025e67914186f45cd865cd2bfe201b117779;hb=750fc5bc98606df0d3d25e8ddda8288ec186a258;hpb=11b08d4ee69a242b763e5f4a34c511cb857f7868

diff --git a/lustre/osd-ldiskfs/osd_compat.c b/lustre/osd-ldiskfs/osd_compat.c
index e2d6025..ae42d2c 100644
--- a/lustre/osd-ldiskfs/osd_compat.c
+++ b/lustre/osd-ldiskfs/osd_compat.c
@@ -15,11 +15,7 @@
  *
  * You should have received a copy of the GNU General Public License
  * version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
  *
  * GPL HEADER END
  */
@@ -27,7 +23,7 @@
  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2012, Intel Corporation.
+ * Copyright (c) 2012, 2017, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -40,65 +36,181 @@
  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  */
 
-/* LUSTRE_VERSION_CODE */
-#include <lustre_ver.h>
 /* prerequisite for linux/xattr.h */
 #include <linux/types.h>
 /* prerequisite for linux/xattr.h */
 #include <linux/fs.h>
+/* XATTR_{REPLACE,CREATE} */
+#include <linux/xattr.h>
 
 /*
  * struct OBD_{ALLOC,FREE}*()
  * OBD_FAIL_CHECK
  */
 #include <obd_support.h>
-#include <lvfs.h>
 
 #include "osd_internal.h"
 #include "osd_oi.h"
 
 static void osd_push_ctxt(const struct osd_device *dev,
-                          struct lvfs_run_ctxt *newctxt,
-                          struct lvfs_run_ctxt *save)
+			  struct lvfs_run_ctxt *newctxt,
+			  struct lvfs_run_ctxt *save)
 {
-        OBD_SET_CTXT_MAGIC(newctxt);
-        newctxt->pwdmnt = dev->od_mnt;
-        newctxt->pwd = dev->od_mnt->mnt_root;
-        newctxt->fs = get_ds();
-
-        push_ctxt(save, newctxt, NULL);
+	OBD_SET_CTXT_MAGIC(newctxt);
+	newctxt->pwdmnt = dev->od_mnt;
+	newctxt->pwd = dev->od_mnt->mnt_root;
+	newctxt->fs = KERNEL_DS;
+	newctxt->umask = current_umask();
+	newctxt->dt = NULL;
+
+	push_ctxt(save, newctxt);
 }
 
-static void osd_pop_ctxt(const struct osd_device *dev,
-			 struct lvfs_run_ctxt *new,
-			 struct lvfs_run_ctxt *save)
+/**
+ * osd_ios_lookup_one_len - lookup single pathname component
+ *
+ * @name:	pathname component to lookup
+ * @base:	base directory to lookup from
+ * @len:	maximum length @len should be interpreted to
+ *
+ * Treat found dentry with NULL d_inode as an -ENOENT error so LFSCK
+ * can repair the file.
+ */
+struct dentry *osd_ios_lookup_one_len(const char *name, struct dentry *base,
+				      int len)
 {
-	pop_ctxt(save, new, NULL);
+	struct dentry *dentry;
+
+	dentry = ll_lookup_one_len(name, base, len);
+	if (IS_ERR(dentry)) {
+		int rc = PTR_ERR(dentry);
+
+		if (rc != -ENOENT)
+			CERROR("Fail to find %.*s in %.*s (%lu/%u): rc = %d\n",
+			       len, name, base->d_name.len,
+			       base->d_name.name, base->d_inode->i_ino,
+			       base->d_inode->i_generation, rc);
+
+		return dentry;
+	}
+
+	if (dentry->d_inode == NULL) {
+		dput(dentry);
+		return ERR_PTR(-ENOENT);
+	}
+
+	return dentry;
 }
 
-int osd_last_rcvd_subdir_count(struct osd_device *osd)
+/* utility to make a directory */
+static struct dentry *
+simple_mkdir(const struct lu_env *env, struct osd_device *osd,
+	     struct dentry *dir, const struct lu_fid *fid,
+	     const char *name, __u32 compat, int mode, bool *created)
 {
-        struct lr_server_data lsd;
-        struct dentry        *dlast;
-        loff_t                off;
-        int                   rc = 0;
-	int                   count = FILTER_SUBDIR_COUNT;
+	struct osd_thread_info *info = osd_oti_get(env);
+	struct lu_fid *tfid = &info->oti_fid3;
+	struct inode *inode;
+	struct dentry *dchild;
+	int err = 0;
+
+	ENTRY;
+
+	// ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
+	CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
+	dchild = ll_lookup_one_len(name, dir, strlen(name));
+	if (IS_ERR(dchild))
+		RETURN(dchild);
 
-        ENTRY;
+	inode = dchild->d_inode;
+	if (inode) {
+		struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
+		int old_mode = inode->i_mode;
 
-        dlast = ll_lookup_one_len(LAST_RCVD, osd_sb(osd)->s_root,
-                                  strlen(LAST_RCVD));
-        if (IS_ERR(dlast))
-                return PTR_ERR(dlast);
-        else if (dlast->d_inode == NULL)
-                goto out;
+		if (created)
+			*created = false;
 
-        off = 0;
-        rc = osd_ldiskfs_read(dlast->d_inode, &lsd, sizeof(lsd), &off);
-        if (rc == sizeof(lsd)) {
-                CDEBUG(D_INFO, "read last_rcvd header, uuid = %s, "
-                       "subdir count = %d\n", lsd.lsd_uuid,
-                       lsd.lsd_subdir_count);
+		if (!S_ISDIR(old_mode)) {
+			CERROR("found %s (%lu/%u) is mode %o\n", name,
+			       inode->i_ino, inode->i_generation, old_mode);
+			GOTO(out_err, err = -ENOTDIR);
+		}
+
+		if (unlikely(osd->od_dt_dev.dd_rdonly))
+			RETURN(dchild);
+
+		/* Fixup directory permissions if necessary */
+		if ((old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
+			CDEBUG(D_CONFIG,
+			       "fixing permissions on %s from %o to %o\n",
+			       name, old_mode, mode);
+			inode->i_mode = (mode & S_IALLUGO) |
+					(old_mode & ~S_IALLUGO);
+			mark_inode_dirty(inode);
+		}
+
+		err = osd_get_lma(info, inode, &info->oti_obj_dentry,
+				  &info->oti_ost_attrs);
+		if (err == -ENODATA)
+			goto set_fid;
+
+		if (err)
+			GOTO(out_err, err);
+
+		if ((fid && !lu_fid_eq(fid, &lma->lma_self_fid)) ||
+		    lma->lma_compat != compat)
+			goto set_fid;
+
+		RETURN(dchild);
+	}
+
+	err = vfs_mkdir(dir->d_inode, dchild, mode);
+	if (err)
+		GOTO(out_err, err);
+
+	inode = dchild->d_inode;
+	if (created)
+		*created = true;
+
+set_fid:
+	if (fid)
+		*tfid = *fid;
+	else
+		lu_igif_build(tfid, inode->i_ino, inode->i_generation);
+	err = osd_ea_fid_set(info, inode, tfid, compat, 0);
+	if (err)
+		GOTO(out_err, err);
+
+	RETURN(dchild);
+
+out_err:
+	dput(dchild);
+	return ERR_PTR(err);
+}
+
+static int osd_last_rcvd_subdir_count(struct osd_device *osd)
+{
+	struct lr_server_data lsd;
+	struct dentry *dlast;
+	loff_t off;
+	int rc = 0;
+	int count = OBJ_SUBDIR_COUNT;
+
+	ENTRY;
+
+	dlast = ll_lookup_one_len(LAST_RCVD, osd_sb(osd)->s_root,
+				  strlen(LAST_RCVD));
+	if (IS_ERR(dlast))
+		return PTR_ERR(dlast);
+	else if (dlast->d_inode == NULL)
+		goto out;
+
+	off = 0;
+	rc = osd_ldiskfs_read(dlast->d_inode, &lsd, sizeof(lsd), &off);
+	if (rc == sizeof(lsd)) {
+		CDEBUG(D_INFO,
+		      "read last_rcvd header, uuid = %s, subdir count = %d\n",
+		      lsd.lsd_uuid, lsd.lsd_subdir_count);
 		if (le16_to_cpu(lsd.lsd_subdir_count) > 0)
 			count = le16_to_cpu(lsd.lsd_subdir_count);
 	} else if (rc != 0) {
@@ -114,23 +226,17 @@ out:
 	return count;
 }
 
-/*
- * directory structure on legacy MDT:
- *
- * REM_OBJ_DIR/ per mdt
- * AGENT_OBJ_DIR/ per mdt
- *
- */
-static const char remote_obj_dir[] = "REM_OBJ_DIR";
-static const char agent_obj_dir[] = "AGENT_OBJ_DIR";
-int osd_mdt_init(struct osd_device *dev)
+static int osd_mdt_init(const struct lu_env *env, struct osd_device *dev)
 {
-	struct lvfs_run_ctxt  new;
-	struct lvfs_run_ctxt  save;
-	struct dentry	*parent;
+	struct lvfs_run_ctxt new;
+	struct lvfs_run_ctxt save;
+	struct dentry *parent;
 	struct osd_mdobj_map *omm;
-	struct dentry	*d;
-	int		   rc = 0;
+	struct dentry *d;
+	struct osd_thread_info *info = osd_oti_get(env);
+	struct lu_fid *fid = &info->oti_fid3;
+	int rc = 0;
+
 	ENTRY;
 
 	OBD_ALLOC_PTR(dev->od_mdt_map);
@@ -139,27 +245,28 @@ int osd_mdt_init(struct osd_device *dev)
 
 	omm = dev->od_mdt_map;
 
-	LASSERT(dev->od_fsops);
-
 	parent = osd_sb(dev)->s_root;
 	osd_push_ctxt(dev, &new, &save);
 
-	d = simple_mkdir(parent, dev->od_mnt, agent_obj_dir,
-			 0755, 1);
+	lu_local_obj_fid(fid, REMOTE_PARENT_DIR_OID);
+	d = simple_mkdir(env, dev, parent, fid, REMOTE_PARENT_DIR,
+			 LMAC_NOT_IN_OI, 0755, NULL);
 	if (IS_ERR(d))
 		GOTO(cleanup, rc = PTR_ERR(d));
 
-	omm->omm_agent_dentry = d;
+	omm->omm_remote_parent = d;
+
+	GOTO(cleanup, rc = 0);
 
 cleanup:
-	pop_ctxt(&save, &new, NULL);
+	pop_ctxt(&save, &new);
 	if (rc) {
-		if (omm->omm_agent_dentry != NULL)
-			dput(omm->omm_agent_dentry);
+		if (omm->omm_remote_parent != NULL)
+			dput(omm->omm_remote_parent);
 		OBD_FREE_PTR(omm);
 		dev->od_mdt_map = NULL;
 	}
-	RETURN(rc);
+	return rc;
 }
 
 static void osd_mdt_fini(struct osd_device *osd)
@@ -169,66 +276,170 @@ static void osd_mdt_fini(struct osd_device *osd)
 	if (omm == NULL)
 		return;
 
-	if (omm->omm_agent_dentry)
-		dput(omm->omm_agent_dentry);
+	if (omm->omm_remote_parent)
+		dput(omm->omm_remote_parent);
 
 	OBD_FREE_PTR(omm);
 	osd->od_ost_map = NULL;
 }
 
-int osd_add_to_agent(const struct lu_env *env, struct osd_device *osd,
-		     struct osd_object *obj, struct osd_thandle *oh)
+int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
+			     struct osd_object *obj, struct osd_thandle *oh)
 {
-	struct osd_mdobj_map	*omm = osd->od_mdt_map;
-	struct osd_thread_info	*oti = osd_oti_get(env);
-	char			*name = oti->oti_name;
-	struct dentry		*agent;
-	struct dentry		*parent;
-	int			rc;
+	struct osd_mdobj_map *omm = osd->od_mdt_map;
+	struct osd_thread_info *oti = osd_oti_get(env);
+	struct lustre_mdt_attrs *lma = &oti->oti_ost_attrs.loa_lma;
+	char *name = oti->oti_name;
+	struct osd_thread_info *info = osd_oti_get(env);
+	struct dentry *dentry;
+	struct dentry *parent;
+	int rc;
+
+	if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_AGENTENT))
+		RETURN(0);
 
-	parent = omm->omm_agent_dentry;
+	/*
+	 * Set REMOTE_PARENT in lma, so other process like unlink or lfsck
+	 * can identify this object quickly
+	 */
+	rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry,
+			 &oti->oti_ost_attrs);
+	if (rc)
+		RETURN(rc);
+
+	lma->lma_incompat |= LMAI_REMOTE_PARENT;
+	lustre_lma_swab(lma);
+	rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
+			     sizeof(*lma), XATTR_REPLACE);
+	if (rc)
+		RETURN(rc);
+
+	parent = omm->omm_remote_parent;
 	sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
-	agent = osd_child_dentry_by_inode(env, parent->d_inode,
-					  name, strlen(name));
-	mutex_lock(&parent->d_inode->i_mutex);
-	rc = osd_ldiskfs_add_entry(oh->ot_handle, agent, obj->oo_inode, NULL);
-	LASSERTF(parent->d_inode->i_nlink > 1, "%s: agent inode nlink %d",
-		 osd_name(osd), parent->d_inode->i_nlink);
-	parent->d_inode->i_nlink++;
+	dentry = osd_child_dentry_by_inode(env, parent->d_inode,
+					   name, strlen(name));
+	inode_lock(parent->d_inode);
+	rc = osd_ldiskfs_add_entry(info, osd, oh->ot_handle, dentry,
+				   obj->oo_inode, NULL);
+	if (!rc && S_ISDIR(obj->oo_inode->i_mode))
+		ldiskfs_inc_count(oh->ot_handle, parent->d_inode);
+	else if (unlikely(rc == -EEXIST))
+		rc = 0;
+	if (!rc)
+		lu_object_set_agent_entry(&obj->oo_dt.do_lu);
+	CDEBUG(D_INODE, "%s: create agent entry for %s: rc = %d\n",
+	       osd_name(osd), name, rc);
 	mark_inode_dirty(parent->d_inode);
-	mutex_unlock(&parent->d_inode->i_mutex);
+	inode_unlock(parent->d_inode);
 	RETURN(rc);
 }
 
-int osd_delete_from_agent(const struct lu_env *env, struct osd_device *osd,
-			  struct osd_object *obj, struct osd_thandle *oh)
+int osd_delete_from_remote_parent(const struct lu_env *env,
+				  struct osd_device *osd,
+				  struct osd_object *obj,
+				  struct osd_thandle *oh, bool destroy)
 {
-	struct osd_mdobj_map	   *omm = osd->od_mdt_map;
-	struct osd_thread_info	   *oti = osd_oti_get(env);
-	char			   *name = oti->oti_name;
-	struct dentry		   *agent;
-	struct dentry		   *parent;
+	struct osd_mdobj_map *omm = osd->od_mdt_map;
+	struct osd_thread_info *oti = osd_oti_get(env);
+	struct lustre_mdt_attrs *lma = &oti->oti_ost_attrs.loa_lma;
+	char *name = oti->oti_name;
+	struct dentry *dentry;
+	struct dentry *parent;
 	struct ldiskfs_dir_entry_2 *de;
-	struct buffer_head	   *bh;
-	int			   rc;
+	struct buffer_head *bh;
+	int rc;
 
-	parent = omm->omm_agent_dentry;
+	parent = omm->omm_remote_parent;
 	sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
-	agent = osd_child_dentry_by_inode(env, parent->d_inode,
-					  name, strlen(name));
-	mutex_lock(&parent->d_inode->i_mutex);
-	bh = osd_ldiskfs_find_entry(parent->d_inode, agent, &de, NULL);
-	if (bh == NULL) {
-		mutex_unlock(&parent->d_inode->i_mutex);
+	dentry = osd_child_dentry_by_inode(env, parent->d_inode,
+					   name, strlen(name));
+	inode_lock(parent->d_inode);
+	bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
+				    NULL, NULL);
+	if (IS_ERR(bh)) {
+		inode_unlock(parent->d_inode);
+		rc = PTR_ERR(bh);
+		if (unlikely(rc == -ENOENT))
+			rc = 0;
+	} else {
+		rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode,
+					  de, bh);
+		if (!rc && S_ISDIR(obj->oo_inode->i_mode))
+			ldiskfs_dec_count(oh->ot_handle, parent->d_inode);
+		mark_inode_dirty(parent->d_inode);
+		inode_unlock(parent->d_inode);
+		brelse(bh);
+		CDEBUG(D_INODE, "%s: remove agent entry for %s: rc = %d\n",
+		       osd_name(osd), name, rc);
+	}
+
+	if (destroy || rc) {
+		if (!rc)
+			lu_object_clear_agent_entry(&obj->oo_dt.do_lu);
+
+		RETURN(rc);
+	}
+
+	rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry,
+			 &oti->oti_ost_attrs);
+	if (rc)
+		RETURN(rc);
+
+	/* Get rid of REMOTE_PARENT flag from incompat */
+	lma->lma_incompat &= ~LMAI_REMOTE_PARENT;
+	lustre_lma_swab(lma);
+	rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
+			     sizeof(*lma), XATTR_REPLACE);
+	if (!rc)
+		lu_object_clear_agent_entry(&obj->oo_dt.do_lu);
+	RETURN(rc);
+}
+
+int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
+				struct osd_device *osd,
+				const struct lu_fid *fid,
+				struct osd_inode_id *id)
+{
+	struct osd_mdobj_map *omm = osd->od_mdt_map;
+	char *name = oti->oti_name;
+	struct dentry *parent;
+	struct dentry *dentry;
+	struct ldiskfs_dir_entry_2 *de;
+	struct buffer_head *bh;
+	int rc;
+
+	ENTRY;
+
+	if (unlikely(osd->od_is_ost))
 		RETURN(-ENOENT);
+
+	parent = omm->omm_remote_parent;
+	sprintf(name, DFID_NOBRACE, PFID(fid));
+	dentry = osd_child_dentry_by_inode(oti->oti_env, parent->d_inode,
+					   name, strlen(name));
+	inode_lock(parent->d_inode);
+	bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
+				    NULL, NULL);
+	if (IS_ERR(bh)) {
+		rc = PTR_ERR(bh);
+	} else {
+		struct inode *inode;
+
+		osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
+		brelse(bh);
+		inode = osd_iget(oti, osd, id);
+		if (IS_ERR(inode)) {
+			rc = PTR_ERR(inode);
+			if (rc == -ESTALE)
+				rc = -ENOENT;
+		} else {
+			iput(inode);
+			rc = 0;
+		}
 	}
-	rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode, de, bh);
-	LASSERTF(parent->d_inode->i_nlink > 1, "%s: agent inode nlink %d",
-		 osd_name(osd), parent->d_inode->i_nlink);
-	parent->d_inode->i_nlink--;
-	mark_inode_dirty(parent->d_inode);
-	mutex_unlock(&parent->d_inode->i_mutex);
-	brelse(bh);
+	inode_unlock(parent->d_inode);
+	if (rc == 0)
+		osd_add_oi_cache(oti, osd, id, fid);
 	RETURN(rc);
 }
 
@@ -242,13 +453,14 @@ int osd_delete_from_agent(const struct lu_env *env, struct osd_device *osd,
  * CONFIGS
  *
  */
-int osd_ost_init(struct osd_device *dev)
+static int osd_ost_init(const struct lu_env *env, struct osd_device *dev)
 {
-	struct lvfs_run_ctxt  new;
-	struct lvfs_run_ctxt  save;
-	struct dentry	     *rootd = osd_sb(dev)->s_root;
-	struct dentry	     *d;
-	int		      rc;
+	struct lvfs_run_ctxt new;
+	struct lvfs_run_ctxt save;
+	struct dentry *d;
+	int rc;
+	bool created = false;
+
 	ENTRY;
 
 	OBD_ALLOC_PTR(dev->od_ost_map);
@@ -257,53 +469,48 @@ int osd_ost_init(struct osd_device *dev)
 
 	/* to get subdir count from last_rcvd */
 	rc = osd_last_rcvd_subdir_count(dev);
-	if (rc < 0) {
-		OBD_FREE_PTR(dev->od_ost_map);
-		RETURN(rc);
-	}
+	if (rc < 0)
+		GOTO(cleanup_alloc, rc);
 
 	dev->od_ost_map->om_subdir_count = rc;
-        rc = 0;
-
-	CFS_INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
+	INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
 	rwlock_init(&dev->od_ost_map->om_seq_list_lock);
-	sema_init(&dev->od_ost_map->om_dir_init_sem, 1);
+	mutex_init(&dev->od_ost_map->om_dir_init_mutex);
 
-        LASSERT(dev->od_fsops);
-        osd_push_ctxt(dev, &new, &save);
-
-        d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
+	osd_push_ctxt(dev, &new, &save);
+	d = simple_mkdir(env, dev, osd_sb(dev)->s_root, NULL, "O",
+			 LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0755, &created);
 	if (IS_ERR(d))
-		GOTO(cleanup, rc = PTR_ERR(d));
+		GOTO(cleanup_ctxt, rc = PTR_ERR(d));
+
+	if (created)
+		/* It is quite probably that the device is new formatted. */
+		dev->od_maybe_new = 1;
 
-	ldiskfs_set_inode_state(d->d_inode, LDISKFS_STATE_LUSTRE_NO_OI);
 	dev->od_ost_map->om_root = d;
 
-cleanup:
-	osd_pop_ctxt(dev, &new, &save);
-        if (IS_ERR(d)) {
-                OBD_FREE_PTR(dev->od_ost_map);
-                RETURN(PTR_ERR(d));
-        }
+	pop_ctxt(&save, &new);
+	RETURN(0);
 
-	RETURN(rc);
+cleanup_ctxt:
+	pop_ctxt(&save, &new);
+cleanup_alloc:
+	OBD_FREE_PTR(dev->od_ost_map);
+	return rc;
 }
 
-static void osd_seq_free(struct osd_obj_map *map,
-			 struct osd_obj_seq *osd_seq)
+static void osd_seq_free(struct osd_obj_seq *osd_seq)
 {
 	int j;
 
-	cfs_list_del_init(&osd_seq->oos_seq_list);
-
 	if (osd_seq->oos_dirs) {
 		for (j = 0; j < osd_seq->oos_subdir_count; j++) {
 			if (osd_seq->oos_dirs[j])
 				dput(osd_seq->oos_dirs[j]);
-                }
+		}
 		OBD_FREE(osd_seq->oos_dirs,
 			 sizeof(struct dentry *) * osd_seq->oos_subdir_count);
-        }
+	}
 
 	if (osd_seq->oos_root)
 		dput(osd_seq->oos_root);
@@ -313,19 +520,22 @@ static void osd_seq_free(struct osd_obj_map *map,
 
 static void osd_ost_fini(struct osd_device *osd)
 {
-	struct osd_obj_seq    *osd_seq;
-	struct osd_obj_seq    *tmp;
-	struct osd_obj_map    *map = osd->od_ost_map;
+	struct osd_obj_seq *osd_seq;
+	struct osd_obj_seq *tmp;
+	struct osd_obj_map *map = osd->od_ost_map;
+
 	ENTRY;
 
 	if (map == NULL)
 		return;
 
 	write_lock(&map->om_seq_list_lock);
-	cfs_list_for_each_entry_safe(osd_seq, tmp,
-				     &map->om_seq_list,
-				     oos_seq_list) {
-		osd_seq_free(map, osd_seq);
+	list_for_each_entry_safe(osd_seq, tmp, &map->om_seq_list,
+				 oos_seq_list) {
+		list_del_init(&osd_seq->oos_seq_list);
+		write_unlock(&map->om_seq_list_lock);
+		osd_seq_free(osd_seq);
+		write_lock(&map->om_seq_list_lock);
 	}
 	write_unlock(&map->om_seq_list_lock);
 	if (map->om_root)
@@ -335,34 +545,83 @@ static void osd_ost_fini(struct osd_device *osd)
 	EXIT;
 }
 
-int osd_obj_map_init(struct osd_device *dev)
+static int osd_index_backup_dir_init(const struct lu_env *env,
+				     struct osd_device *dev)
+{
+	struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
+	struct lvfs_run_ctxt new;
+	struct lvfs_run_ctxt save;
+	struct dentry *dentry;
+	int rc = 0;
+
+	ENTRY;
+
+	lu_local_obj_fid(fid, INDEX_BACKUP_OID);
+	osd_push_ctxt(dev, &new, &save);
+	dentry = simple_mkdir(env, dev, osd_sb(dev)->s_root, fid,
+			      INDEX_BACKUP_DIR, LMAC_NOT_IN_OI, 0755, NULL);
+	if (IS_ERR(dentry)) {
+		rc = PTR_ERR(dentry);
+	} else {
+		dev->od_index_backup_inode = igrab(dentry->d_inode);
+		dput(dentry);
+	}
+	pop_ctxt(&save, &new);
+
+	RETURN(rc);
+}
+
+static void osd_index_backup_dir_fini(struct osd_device *dev)
+{
+	if (dev->od_index_backup_inode) {
+		iput(dev->od_index_backup_inode);
+		dev->od_index_backup_inode = NULL;
+	}
+}
+
+int osd_obj_map_init(const struct lu_env *env, struct osd_device *dev)
 {
 	int rc;
+	bool mdt_init = false;
+
 	ENTRY;
 
-	/* prepare structures for OST */
-	rc = osd_ost_init(dev);
+	rc = osd_ost_init(env, dev);
 	if (rc)
 		RETURN(rc);
 
-	/* prepare structures for MDS */
-	rc = osd_mdt_init(dev);
+	if (!dev->od_is_ost) {
+		rc = osd_mdt_init(env, dev);
+		if (rc) {
+			osd_ost_fini(dev);
+			RETURN(rc);
+		}
+
+		mdt_init = true;
+	}
 
-        RETURN(rc);
+	rc = osd_index_backup_dir_init(env, dev);
+	if (rc) {
+		osd_ost_fini(dev);
+		if (mdt_init)
+			osd_mdt_fini(dev);
+	}
+
+	RETURN(rc);
 }
 
-struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, obd_seq seq)
+static struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, u64 seq)
 {
 	struct osd_obj_seq *osd_seq;
 
-	cfs_list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
+	list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
 		if (osd_seq->oos_seq == seq)
 			return osd_seq;
 	}
 	return NULL;
 }
 
-struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, obd_seq seq)
+static struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, u64 seq)
 {
 	struct osd_obj_seq *osd_seq;
 
@@ -374,27 +633,180 @@ struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, obd_seq seq)
 
 void osd_obj_map_fini(struct osd_device *dev)
 {
+	osd_index_backup_dir_fini(dev);
 	osd_ost_fini(dev);
 	osd_mdt_fini(dev);
 }
 
+/**
+ * Update the specified OI mapping.
+ *
+ * \retval   1, changed nothing
+ * \retval   0, changed successfully
+ * \retval -ve, on error
+ */
+static int osd_obj_update_entry(struct osd_thread_info *info,
+				struct osd_device *osd,
+				struct dentry *dir, const char *name,
+				const struct lu_fid *fid,
+				const struct osd_inode_id *id,
+				handle_t *th)
+{
+	struct inode *parent = dir->d_inode;
+	struct dentry *child;
+	struct ldiskfs_dir_entry_2 *de;
+	struct buffer_head *bh;
+	struct inode *inode;
+	struct dentry *dentry = &info->oti_obj_dentry;
+	struct osd_inode_id *oi_id = &info->oti_id3;
+	struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
+	struct lu_fid *oi_fid = &lma->lma_self_fid;
+	int rc;
+
+	ENTRY;
+
+	LASSERT(th != NULL);
+	LASSERT(th->h_transaction != NULL);
+
+	child = &info->oti_child_dentry;
+	child->d_parent = dir;
+	child->d_name.hash = 0;
+	child->d_name.name = name;
+	child->d_name.len = strlen(name);
+
+	ll_vfs_dq_init(parent);
+	inode_lock(parent);
+	bh = osd_ldiskfs_find_entry(parent, &child->d_name, &de, NULL, NULL);
+	if (IS_ERR(bh))
+		GOTO(out, rc = PTR_ERR(bh));
+
+	if (le32_to_cpu(de->inode) == id->oii_ino)
+		GOTO(out, rc = 1);
+
+	osd_id_gen(oi_id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
+	inode = osd_iget(info, osd, oi_id);
+	if (IS_ERR(inode)) {
+		rc = PTR_ERR(inode);
+		if (rc == -ENOENT || rc == -ESTALE)
+			goto update;
+		GOTO(out, rc);
+	}
+
+	/*
+	 * The EA inode should NOT be in OI, old OI scrub may added
+	 * such OI mapping by wrong, replace it.
+	 */
+	if (unlikely(osd_is_ea_inode(inode))) {
+		iput(inode);
+		goto update;
+	}
+
+	rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
+	if (rc == -ENODATA) {
+		rc = osd_get_idif(info, inode, dentry, oi_fid);
+		if (rc > 0 || rc == -ENODATA) {
+			oi_fid = NULL;
+			rc = 0;
+		}
+	}
+	iput(inode);
+
+	if (rc != 0)
+		GOTO(out, rc);
+
+	/*
+	 * If the OST-object has neither FID-in-LMA nor FID-in-ff, it is
+	 * either a crashed object or a uninitialized one. Replace it.
+	 */
+	if (oi_fid != NULL && lu_fid_eq(fid, oi_fid)) {
+		CERROR("%s: the FID "DFID" is used by two objects: "
+		       "%u/%u %u/%u\n", osd_name(osd), PFID(fid),
+		       oi_id->oii_ino, oi_id->oii_gen,
+		       id->oii_ino, id->oii_gen);
+		GOTO(out, rc = -EEXIST);
+	}
+
+	if (fid_is_idif(fid) && oi_fid != NULL && fid_is_idif(oi_fid)) {
+		__u32 idx1 = fid_idif_ost_idx(fid);
+		__u32 idx2 = fid_idif_ost_idx(oi_fid);
+		struct ost_id *ostid = &info->oti_ostid;
+		struct lu_fid *tfid = &info->oti_fid3;
+
+		LASSERTF(idx1 == 0 || idx1 == osd->od_index,
+			 "invalid given FID "DFID", not match the "
+			 "device index %u\n", PFID(fid), osd->od_index);
+
+		if (idx1 != idx2) {
+			if (idx1 == 0 && idx2 == osd->od_index) {
+				fid_to_ostid(fid, ostid);
+				ostid_to_fid(tfid, ostid, idx2);
+				if (lu_fid_eq(tfid, oi_fid)) {
+					CERROR("%s: the FID "DFID" is used by "
+					       "two objects(2): %u/%u %u/%u\n",
+					       osd_name(osd), PFID(fid),
+					       oi_id->oii_ino, oi_id->oii_gen,
+					       id->oii_ino, id->oii_gen);
+
+					GOTO(out, rc = -EEXIST);
+				}
+			} else if (idx2 == 0 && idx1 == osd->od_index) {
+				fid_to_ostid(oi_fid, ostid);
+				ostid_to_fid(tfid, ostid, idx1);
+				if (lu_fid_eq(tfid, fid)) {
+					CERROR("%s: the FID "DFID" is used by "
+					       "two objects(2): %u/%u %u/%u\n",
+					       osd_name(osd), PFID(fid),
+					       oi_id->oii_ino, oi_id->oii_gen,
+					       id->oii_ino, id->oii_gen);
+
+					GOTO(out, rc = -EEXIST);
+				}
+			}
+		}
+	}
+
+update:
+	/*
+	 * There may be temporary inconsistency: On one hand, the new
+	 * object may be referenced by multiple entries, which is out
+	 * of our control unless we traverse the whole /O completely,
+	 * which is non-flat order and inefficient, should be avoided;
+	 * On the other hand, the old object may become orphan if it
+	 * is still valid. Since it was referenced by an invalid entry,
+	 * making it as invisible temporary may be not worse. OI scrub
+	 * will process it later.
+	 */
+	rc = ldiskfs_journal_get_write_access(th, bh);
+	if (rc != 0)
+		GOTO(out, rc);
+
+	de->inode = cpu_to_le32(id->oii_ino);
+	rc = ldiskfs_handle_dirty_metadata(th, NULL, bh);
+
+	GOTO(out, rc);
+
+out:
+	if (!IS_ERR(bh))
+		brelse(bh);
+	inode_unlock(parent);
+	return rc;
+}
+
 static int osd_obj_del_entry(struct osd_thread_info *info,
 			     struct osd_device *osd,
 			     struct dentry *dird, char *name,
-			     struct thandle *th)
+			     handle_t *th)
 {
 	struct ldiskfs_dir_entry_2 *de;
-	struct buffer_head         *bh;
-	struct osd_thandle         *oh;
-	struct dentry              *child;
-	struct inode               *dir = dird->d_inode;
-	int                         rc;
-	ENTRY;
+	struct buffer_head *bh;
+	struct dentry *child;
+	struct inode *dir = dird->d_inode;
+	int rc;
 
-	oh = container_of(th, struct osd_thandle, ot_super);
-	LASSERT(oh->ot_handle != NULL);
-	LASSERT(oh->ot_handle->h_transaction != NULL);
+	ENTRY;
 
+	LASSERT(th != NULL);
+	LASSERT(th->h_transaction != NULL);
 
 	child = &info->oti_child_dentry;
 	child->d_name.hash = 0;
@@ -404,85 +816,104 @@ static int osd_obj_del_entry(struct osd_thread_info *info,
 	child->d_inode = NULL;
 
 	ll_vfs_dq_init(dir);
-	mutex_lock(&dir->i_mutex);
-	rc = -ENOENT;
-	bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
-	if (bh) {
-		rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
+	inode_lock(dir);
+	bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
+	if (IS_ERR(bh)) {
+		rc = PTR_ERR(bh);
+	} else {
+		rc = ldiskfs_delete_entry(th, dir, de, bh);
 		brelse(bh);
 	}
-	mutex_unlock(&dir->i_mutex);
+	inode_unlock(dir);
 
 	RETURN(rc);
 }
 
-int osd_obj_add_entry(struct osd_thread_info *info,
-		      struct osd_device *osd,
-		      struct dentry *dir, char *name,
-		      const struct osd_inode_id *id,
-		      struct thandle *th)
+static int osd_obj_add_entry(struct osd_thread_info *info,
+			     struct osd_device *osd,
+			     struct dentry *dir, char *name,
+			     const struct osd_inode_id *id,
+			     handle_t *th)
 {
-        struct osd_thandle *oh;
-        struct dentry *child;
-        struct inode *inode;
-        int rc;
+	struct dentry *child;
+	struct inode *inode;
+	int rc;
 
-        ENTRY;
+	ENTRY;
 
-        oh = container_of(th, struct osd_thandle, ot_super);
-        LASSERT(oh->ot_handle != NULL);
-        LASSERT(oh->ot_handle->h_transaction != NULL);
+	if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY))
+		RETURN(0);
+
+	LASSERT(th != NULL);
+	LASSERT(th->h_transaction != NULL);
+
+	inode = info->oti_inode;
+	if (unlikely(inode == NULL)) {
+		struct ldiskfs_inode_info *lii;
+
+		OBD_ALLOC_PTR(lii);
+		if (lii == NULL)
+			RETURN(-ENOMEM);
+		inode = info->oti_inode = &lii->vfs_inode;
+	}
 
-        inode = &info->oti_inode;
-        inode->i_sb = osd_sb(osd);
+	inode->i_sb = osd_sb(osd);
 	osd_id_to_inode(inode, id);
 	inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
 
-        child = &info->oti_child_dentry;
-        child->d_name.hash = 0;
-        child->d_name.name = name;
-        child->d_name.len = strlen(name);
-        child->d_parent = dir;
-        child->d_inode = inode;
+	child = &info->oti_child_dentry;
+	child->d_name.hash = 0;
+	child->d_name.name = name;
+	child->d_name.len = strlen(name);
+	child->d_parent = dir;
+	child->d_inode = inode;
+
+	if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
+		inode->i_ino++;
 
 	ll_vfs_dq_init(dir->d_inode);
-	mutex_lock(&dir->d_inode->i_mutex);
-	rc = osd_ldiskfs_add_entry(oh->ot_handle, child, inode, NULL);
-	mutex_unlock(&dir->d_inode->i_mutex);
+	inode_lock(dir->d_inode);
+	rc = osd_ldiskfs_add_entry(info, osd, th, child, inode, NULL);
+	inode_unlock(dir->d_inode);
 
 	RETURN(rc);
 }
 
 /**
- * Use LPU64 for legacy OST sequences, but use LPX64i for new
+ * Use %llu for legacy OST sequences, but use %llx for new
  * sequences names, so that the O/{seq}/dN/{oid} more closely
  * follows the DFID/PFID format. This makes it easier to map from
  * debug messages to objects in the future, and the legacy space
  * of FID_SEQ_OST_MDT0 will be unused in the future.
  **/
-static inline void osd_seq_name(char *seq_name, obd_seq seq)
+static inline void osd_seq_name(char *seq_name, size_t name_size, u64 seq)
 {
-	sprintf(seq_name, (fid_seq_is_rsvd(seq) ||
-			   fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
-		fid_seq_is_idif(seq) ? 0 : seq);
+	snprintf(seq_name, name_size,
+		 (fid_seq_is_rsvd(seq) ||
+		  fid_seq_is_mdt0(seq)) ? "%llu" : "%llx",
+		 fid_seq_is_idif(seq) ? 0 : seq);
 }
 
-static inline void osd_oid_name(char *name, const struct lu_fid *fid, obd_id id)
+static inline void osd_oid_name(char *name, size_t name_size,
+				const struct lu_fid *fid, u64 id)
 {
-	sprintf(name, (fid_seq_is_rsvd(fid_seq(fid)) ||
-		fid_seq_is_mdt0(fid_seq(fid)) ||
-		fid_seq_is_idif(fid_seq(fid))) ? LPU64 : LPX64i, id);
+	snprintf(name, name_size,
+		 (fid_seq_is_rsvd(fid_seq(fid)) ||
+		  fid_seq_is_mdt0(fid_seq(fid)) ||
+		  fid_seq_is_idif(fid_seq(fid))) ? "%llu" : "%llx", id);
 }
 
 /* external locking is required */
-static int osd_seq_load_locked(struct osd_device *osd,
+static int osd_seq_load_locked(struct osd_thread_info *info,
+			       struct osd_device *osd,
 			       struct osd_obj_seq *osd_seq)
 {
-	struct osd_obj_map  *map = osd->od_ost_map;
-	struct dentry       *seq_dir;
-	int		    rc = 0;
-	int		    i;
-	char		    seq_name[32];
+	struct osd_obj_map *map = osd->od_ost_map;
+	struct dentry *seq_dir;
+	int rc = 0;
+	int i;
+	char dir_name[32];
+
 	ENTRY;
 
 	if (osd_seq->oos_root != NULL)
@@ -491,15 +922,15 @@ static int osd_seq_load_locked(struct osd_device *osd,
 	LASSERT(map);
 	LASSERT(map->om_root);
 
-	osd_seq_name(seq_name, osd_seq->oos_seq);
+	osd_seq_name(dir_name, sizeof(dir_name), osd_seq->oos_seq);
 
-	seq_dir = simple_mkdir(map->om_root, osd->od_mnt, seq_name, 0755, 1);
+	seq_dir = simple_mkdir(info->oti_env, osd, map->om_root, NULL, dir_name,
+			       LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0755, NULL);
 	if (IS_ERR(seq_dir))
 		GOTO(out_err, rc = PTR_ERR(seq_dir));
 	else if (seq_dir->d_inode == NULL)
 		GOTO(out_put, rc = -EFAULT);
 
-	ldiskfs_set_inode_state(seq_dir->d_inode, LDISKFS_STATE_LUSTRE_NO_OI);
 	osd_seq->oos_root = seq_dir;
 
 	LASSERT(osd_seq->oos_dirs == NULL);
@@ -510,27 +941,30 @@ static int osd_seq_load_locked(struct osd_device *osd,
 
 	for (i = 0; i < osd_seq->oos_subdir_count; i++) {
 		struct dentry   *dir;
-		char	     name[32];
 
-		sprintf(name, "d%u", i);
-		dir = simple_mkdir(osd_seq->oos_root, osd->od_mnt, name,
-				   0700, 1);
+		snprintf(dir_name, sizeof(dir_name), "d%u", i);
+		dir = simple_mkdir(info->oti_env, osd, osd_seq->oos_root, NULL,
+				   dir_name, LMAC_NOT_IN_OI | LMAC_FID_ON_OST,
+				   0700, NULL);
 		if (IS_ERR(dir)) {
-			rc = PTR_ERR(dir);
-		} else if (dir->d_inode) {
-			ldiskfs_set_inode_state(dir->d_inode,
-						LDISKFS_STATE_LUSTRE_NO_OI);
-			osd_seq->oos_dirs[i] = dir;
-			rc = 0;
-		} else {
-			LBUG();
+			GOTO(out_free, rc = PTR_ERR(dir));
+		} else if (dir->d_inode == NULL) {
+			dput(dir);
+			GOTO(out_free, rc = -EFAULT);
 		}
+
+		osd_seq->oos_dirs[i] = dir;
 	}
 
-	if (rc != 0)
-		osd_seq_free(map, osd_seq);
-out_put:
 	if (rc != 0) {
+out_free:
+		for (i = 0; i < osd_seq->oos_subdir_count; i++) {
+			if (osd_seq->oos_dirs[i] != NULL)
+				dput(osd_seq->oos_dirs[i]);
+		}
+		OBD_FREE(osd_seq->oos_dirs,
+			 sizeof(seq_dir) * osd_seq->oos_subdir_count);
+out_put:
 		dput(seq_dir);
 		osd_seq->oos_root = NULL;
 	}
@@ -538,11 +972,13 @@ out_err:
 	RETURN(rc);
 }
 
-struct osd_obj_seq *osd_seq_load(struct osd_device *osd, obd_seq seq)
+static struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info,
+					struct osd_device *osd, u64 seq)
 {
-	struct osd_obj_map	*map;
-	struct osd_obj_seq	*osd_seq;
-	int			rc = 0;
+	struct osd_obj_map *map;
+	struct osd_obj_seq *osd_seq;
+	int rc = 0;
+
 	ENTRY;
 
 	map = osd->od_ost_map;
@@ -554,7 +990,7 @@ struct osd_obj_seq *osd_seq_load(struct osd_device *osd, obd_seq seq)
 		RETURN(osd_seq);
 
 	/* Serializing init process */
-	down(&map->om_dir_init_sem);
+	mutex_lock(&map->om_dir_init_mutex);
 
 	/* Check whether the seq has been added */
 	read_lock(&map->om_seq_list_lock);
@@ -569,21 +1005,23 @@ struct osd_obj_seq *osd_seq_load(struct osd_device *osd, obd_seq seq)
 	if (osd_seq == NULL)
 		GOTO(cleanup, rc = -ENOMEM);
 
-	CFS_INIT_LIST_HEAD(&osd_seq->oos_seq_list);
+	INIT_LIST_HEAD(&osd_seq->oos_seq_list);
 	osd_seq->oos_seq = seq;
-	/* Init subdir count to be 32, but each seq can have
-	 * different subdir count */
+	/*
+	 * Init subdir count to be 32, but each seq can have
+	 * different subdir count
+	 */
 	osd_seq->oos_subdir_count = map->om_subdir_count;
-	rc = osd_seq_load_locked(osd, osd_seq);
+	rc = osd_seq_load_locked(info, osd, osd_seq);
 	if (rc != 0)
 		GOTO(cleanup, rc);
 
 	write_lock(&map->om_seq_list_lock);
-	cfs_list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
+	list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
 	write_unlock(&map->om_seq_list_lock);
 
 cleanup:
-	up(&map->om_dir_init_sem);
+	mutex_unlock(&map->om_dir_init_mutex);
 	if (rc != 0) {
 		if (osd_seq != NULL)
 			OBD_FREE_PTR(osd_seq);
@@ -596,35 +1034,35 @@ cleanup:
 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
 		       const struct lu_fid *fid, struct osd_inode_id *id)
 {
-	struct osd_obj_map		*map;
-	struct osd_obj_seq		*osd_seq;
-	struct dentry			*d_seq;
-	struct dentry			*child;
-	struct ost_id			*ostid = &info->oti_ostid;
-	int				dirn;
-	char				name[32];
-	struct ldiskfs_dir_entry_2	*de;
-	struct buffer_head		*bh;
-	struct inode			*dir;
-	struct inode			*inode;
-        ENTRY;
-
-        /* on the very first lookup we find and open directories */
-
-        map = dev->od_ost_map;
-        LASSERT(map);
+	struct osd_obj_map *map;
+	struct osd_obj_seq *osd_seq;
+	struct dentry *d_seq;
+	struct dentry *child;
+	struct ost_id *ostid = &info->oti_ostid;
+	int dirn;
+	char name[32];
+	struct ldiskfs_dir_entry_2 *de;
+	struct buffer_head *bh;
+	struct inode *dir;
+	struct inode *inode;
+
+	ENTRY;
+
+	/* on the very first lookup we find and open directories */
+	map = dev->od_ost_map;
+	LASSERT(map);
 	LASSERT(map->om_root);
 
-        fid_ostid_pack(fid, ostid);
-	osd_seq = osd_seq_load(dev, ostid->oi_seq);
+	fid_to_ostid(fid, ostid);
+	osd_seq = osd_seq_load(info, dev, ostid_seq(ostid));
 	if (IS_ERR(osd_seq))
 		RETURN(PTR_ERR(osd_seq));
 
-	dirn = ostid->oi_id & (osd_seq->oos_subdir_count - 1);
+	dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
 	d_seq = osd_seq->oos_dirs[dirn];
 	LASSERT(d_seq);
 
-	osd_oid_name(name, fid, ostid->oi_id);
+	osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
 
 	child = &info->oti_child_dentry;
 	child->d_parent = d_seq;
@@ -634,19 +1072,22 @@ int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
 	child->d_name.len = strlen(name);
 
 	dir = d_seq->d_inode;
-	mutex_lock(&dir->i_mutex);
-	bh = osd_ldiskfs_find_entry(dir, child, &de, NULL);
-	mutex_unlock(&dir->i_mutex);
+	inode_lock(dir);
+	bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
+	inode_unlock(dir);
 
-	if (bh == NULL)
-		RETURN(-ENOENT);
+	if (IS_ERR(bh))
+		RETURN(PTR_ERR(bh));
 
 	osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
 	brelse(bh);
 
 	inode = osd_iget(info, dev, id);
-	if (IS_ERR(inode))
-		RETURN(PTR_ERR(inode));
+	if (IS_ERR(inode)) {
+		int rc = PTR_ERR(inode);
+
+		RETURN(rc == -ENOENT ? -ESTALE : rc);
+	}
 
 	iput(inode);
 	RETURN(0);
@@ -656,94 +1097,296 @@ int osd_obj_map_insert(struct osd_thread_info *info,
 		       struct osd_device *osd,
 		       const struct lu_fid *fid,
 		       const struct osd_inode_id *id,
-		       struct thandle *th)
+		       handle_t *th)
 {
-	struct osd_obj_map	*map;
-	struct osd_obj_seq	*osd_seq;
-	struct dentry		*d;
-	struct ost_id		*ostid = &info->oti_ostid;
-	int			dirn, rc = 0;
-	char			name[32];
-        ENTRY;
+	struct osd_obj_map *map;
+	struct osd_obj_seq *osd_seq;
+	struct dentry *d;
+	struct ost_id *ostid = &info->oti_ostid;
+	u64 oid;
+	int dirn, rc = 0;
+	char name[32];
 
-        map = osd->od_ost_map;
-        LASSERT(map);
+	ENTRY;
+
+	map = osd->od_ost_map;
+	LASSERT(map);
 
 	/* map fid to seq:objid */
-        fid_ostid_pack(fid, ostid);
+	fid_to_ostid(fid, ostid);
 
-	osd_seq = osd_seq_load(osd, ostid->oi_seq);
+	oid = ostid_id(ostid);
+	osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
 	if (IS_ERR(osd_seq))
 		RETURN(PTR_ERR(osd_seq));
 
-	dirn = ostid->oi_id & (osd_seq->oos_subdir_count - 1);
+	dirn = oid & (osd_seq->oos_subdir_count - 1);
 	d = osd_seq->oos_dirs[dirn];
-        LASSERT(d);
+	LASSERT(d);
 
-	osd_oid_name(name, fid, ostid->oi_id);
+	osd_oid_name(name, sizeof(name), fid, oid);
+
+again:
 	rc = osd_obj_add_entry(info, osd, d, name, id, th);
+	if (rc == -EEXIST) {
+		rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
+		if (unlikely(rc == -ENOENT))
+			goto again;
+
+		if (unlikely(rc == 1))
+			RETURN(0);
+	}
 
-        RETURN(rc);
+	RETURN(rc);
 }
 
 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
-		       const struct lu_fid *fid, struct thandle *th)
+		       const struct lu_fid *fid, handle_t *th)
 {
-	struct osd_obj_map	*map;
-	struct osd_obj_seq	*osd_seq;
-	struct dentry		*d;
-	struct ost_id		*ostid = &info->oti_ostid;
-	int			dirn, rc = 0;
-	char			name[32];
-        ENTRY;
+	struct osd_obj_map *map;
+	struct osd_obj_seq *osd_seq;
+	struct dentry *d;
+	struct ost_id *ostid = &info->oti_ostid;
+	int dirn, rc = 0;
+	char name[32];
+
+	ENTRY;
 
-        map = osd->od_ost_map;
-        LASSERT(map);
+	map = osd->od_ost_map;
+	LASSERT(map);
 
 	/* map fid to seq:objid */
-        fid_ostid_pack(fid, ostid);
+	fid_to_ostid(fid, ostid);
 
-	osd_seq = osd_seq_load(osd, ostid->oi_seq);
+	osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
 	if (IS_ERR(osd_seq))
 		GOTO(cleanup, rc = PTR_ERR(osd_seq));
 
-	dirn = ostid->oi_id & (osd_seq->oos_subdir_count - 1);
+	dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
 	d = osd_seq->oos_dirs[dirn];
 	LASSERT(d);
 
-	osd_oid_name(name, fid, ostid->oi_id);
+	osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
 	rc = osd_obj_del_entry(info, osd, d, name, th);
 cleanup:
-        RETURN(rc);
+	RETURN(rc);
 }
 
-int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
-			const struct lu_fid *fid,
-			const struct osd_inode_id *id,
-			struct thandle *th)
-{
-	struct osd_obj_map	*map = osd->od_ost_map;
-	struct dentry		*root = osd_sb(osd)->s_root;
-	char			*name;
-	int			rc = 0;
+int osd_obj_map_update(struct osd_thread_info *info,
+		       struct osd_device *osd,
+		       const struct lu_fid *fid,
+		       const struct osd_inode_id *id,
+		       handle_t *th)
+{
+	struct osd_obj_seq *osd_seq;
+	struct dentry *d;
+	struct ost_id *ostid = &info->oti_ostid;
+	int dirn, rc = 0;
+	char name[32];
+
+	ENTRY;
+
+	fid_to_ostid(fid, ostid);
+	osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
+	if (IS_ERR(osd_seq))
+		RETURN(PTR_ERR(osd_seq));
+
+	dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
+	d = osd_seq->oos_dirs[dirn];
+	LASSERT(d);
+
+	osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
+	rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
+
+	RETURN(rc);
+}
+
+int osd_obj_map_recover(struct osd_thread_info *info,
+			struct osd_device *osd,
+			struct inode *src_parent,
+			struct dentry *src_child,
+			const struct lu_fid *fid)
+{
+	struct osd_obj_seq *osd_seq;
+	struct dentry *tgt_parent;
+	struct dentry *tgt_child = &info->oti_child_dentry;
+	struct inode *dir;
+	struct inode *inode = src_child->d_inode;
+	struct ost_id *ostid = &info->oti_ostid;
+	handle_t *jh;
+	struct ldiskfs_dir_entry_2 *de;
+	struct buffer_head *bh;
+	char name[32];
+	int dirn;
+	int rc = 0;
+
 	ENTRY;
 
 	if (fid_is_last_id(fid)) {
-		struct osd_obj_seq	*osd_seq;
+		osd_seq = osd_seq_load(info, osd, fid_seq(fid));
+		if (IS_ERR(osd_seq))
+			RETURN(PTR_ERR(osd_seq));
 
-		/* on creation of LAST_ID we create O/<seq> hierarchy */
-		LASSERT(map);
-		osd_seq = osd_seq_load(osd, fid_seq(fid));
+		tgt_parent = osd_seq->oos_root;
+		tgt_child->d_name.name = "LAST_ID";
+		tgt_child->d_name.len = strlen("LAST_ID");
+	} else {
+		fid_to_ostid(fid, ostid);
+		osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
 		if (IS_ERR(osd_seq))
 			RETURN(PTR_ERR(osd_seq));
-		rc = osd_obj_add_entry(info, osd, osd_seq->oos_root,
-				       "LAST_ID", id, th);
+
+		dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
+		tgt_parent = osd_seq->oos_dirs[dirn];
+		osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
+		tgt_child->d_name.name = name;
+		tgt_child->d_name.len = strlen(name);
+	}
+	LASSERT(tgt_parent != NULL);
+
+	dir = tgt_parent->d_inode;
+	tgt_child->d_name.hash = 0;
+	tgt_child->d_parent = tgt_parent;
+	tgt_child->d_inode = inode;
+
+	/* The non-initialized src_child may be destroyed. */
+	jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
+				osd_dto_credits_noquota[DTO_INDEX_DELETE] +
+				osd_dto_credits_noquota[DTO_INDEX_INSERT] +
+				osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
+	if (IS_ERR(jh))
+		RETURN(PTR_ERR(jh));
+
+	ll_vfs_dq_init(src_parent);
+	ll_vfs_dq_init(dir);
+
+	inode_lock(src_parent);
+	inode_lock(dir);
+	bh = osd_ldiskfs_find_entry(dir, &tgt_child->d_name, &de, NULL, NULL);
+	if (!IS_ERR(bh)) {
+		/*
+		 * XXX: If some other object occupied the same slot. And If such
+		 *	inode is zero-sized and with SUID+SGID, then means it is
+		 *	a new created one. Maybe we can remove it and insert the
+		 *	original one back to the /O/<seq>/d<x>. But there are
+		 *	something to be considered:
+		 *
+		 *	1) The OST-object under /lost+found has crashed LMA.
+		 *	   So it should not conflict with the current one.
+		 *
+		 *	2) There are race conditions that: someone may just want
+		 *	   to modify the current one. Even if the OI scrub takes
+		 *	   the object lock when remove the current one, it still
+		 *	   cause the modification to be lost becasue the target
+		 *	   has been removed when the RPC service thread waiting
+		 *	   for the lock.
+		 *
+		 *	So keep it there before we have suitable solution.
+		 */
+		brelse(bh);
+		inode_unlock(dir);
+		inode_unlock(src_parent);
+		ldiskfs_journal_stop(jh);
+
+		rc = -EEXIST;
+		/* If the src object has never been modified, then remove it. */
+		if (inode->i_size == 0 && inode->i_mode & S_ISUID &&
+		    inode->i_mode & S_ISGID) {
+			rc = ll_vfs_unlink(src_parent, src_child);
+			if (unlikely(rc == -ENOENT))
+				rc = 0;
+		}
+		if (rc)
+			RETURN(rc);
+	}
+
+	bh = osd_ldiskfs_find_entry(src_parent, &src_child->d_name, &de,
+				    NULL, NULL);
+	if (unlikely(IS_ERR(bh)))
+		GOTO(unlock, rc = PTR_ERR(bh));
+
+	rc = ldiskfs_delete_entry(jh, src_parent, de, bh);
+	brelse(bh);
+	if (rc != 0)
+		GOTO(unlock, rc);
+
+	rc = osd_ldiskfs_add_entry(info, osd, jh, tgt_child, inode, NULL);
+
+	GOTO(unlock, rc);
+
+unlock:
+	inode_unlock(dir);
+	inode_unlock(src_parent);
+	ldiskfs_journal_stop(jh);
+	return rc;
+}
+
+static struct dentry *
+osd_object_spec_find(struct osd_thread_info *info, struct osd_device *osd,
+		     const struct lu_fid *fid, char **name)
+{
+	struct dentry *root = ERR_PTR(-ENOENT);
+
+	if (fid_is_last_id(fid)) {
+		struct osd_obj_seq *osd_seq;
+
+		/* on creation of LAST_ID we create O/<seq> hierarchy */
+		osd_seq = osd_seq_load(info, osd, fid_seq(fid));
+		if (IS_ERR(osd_seq))
+			RETURN((struct dentry *)osd_seq);
+
+		*name = "LAST_ID";
+		root = osd_seq->oos_root;
 	} else {
-		name = osd_lf_fid2name(fid);
-		if (name == NULL)
+		*name = osd_lf_fid2name(fid);
+		if (*name == NULL)
 			CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
-		else if (name[0])
-			rc = osd_obj_add_entry(info, osd, root, name, id, th);
+		else if ((*name)[0])
+			root = osd_sb(osd)->s_root;
+	}
+
+	return root;
+}
+
+int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
+			const struct lu_fid *fid, const struct osd_inode_id *id,
+			handle_t *th)
+{
+	struct dentry *root;
+	char *name = NULL;
+	int rc;
+
+	ENTRY;
+
+	root = osd_object_spec_find(info, osd, fid, &name);
+	if (!IS_ERR(root)) {
+		rc = osd_obj_update_entry(info, osd, root, name, fid, id, th);
+	} else {
+		rc = PTR_ERR(root);
+		if (rc == -ENOENT)
+			rc = 1;
+	}
+
+	RETURN(rc);
+}
+
+int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
+			const struct lu_fid *fid, const struct osd_inode_id *id,
+			handle_t *th)
+{
+	struct dentry *root;
+	char *name = NULL;
+	int rc;
+
+	ENTRY;
+
+	root = osd_object_spec_find(info, osd, fid, &name);
+	if (!IS_ERR(root)) {
+		rc = osd_obj_add_entry(info, osd, root, name, id, th);
+	} else {
+		rc = PTR_ERR(root);
+		if (rc == -ENOENT)
+			rc = 0;
 	}
 
 	RETURN(rc);
@@ -752,17 +1395,18 @@ int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
 			const struct lu_fid *fid, struct osd_inode_id *id)
 {
-	struct dentry	*root;
-	struct dentry	*dentry;
-	struct inode	*inode;
-	char		*name;
-	int		rc = -ENOENT;
+	struct dentry *root;
+	struct dentry *dentry;
+	struct inode *inode;
+	char *name = NULL;
+	int rc = -ENOENT;
+
 	ENTRY;
 
 	if (fid_is_last_id(fid)) {
 		struct osd_obj_seq *osd_seq;
 
-		osd_seq = osd_seq_load(osd, fid_seq(fid));
+		osd_seq = osd_seq_load(info, osd, fid_seq(fid));
 		if (IS_ERR(osd_seq))
 			RETURN(PTR_ERR(osd_seq));
 		root = osd_seq->oos_root;
@@ -786,11 +1430,43 @@ int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
 				rc = 0;
 			}
 		}
-		/* if dentry is accessible after osd_compat_spec_insert it
-		 * will still contain NULL inode, so don't keep it in cache */
+		/*
+		 * if dentry is accessible after osd_compat_spec_insert it
+		 * will still contain NULL inode, so don't keep it in cache
+		 */
 		d_invalidate(dentry);
 		dput(dentry);
 	}
 
 	RETURN(rc);
 }
+
+#ifndef HAVE_BIO_INTEGRITY_ENABLED
+bool bio_integrity_enabled(struct bio *bio)
+{
+	struct blk_integrity *bi = blk_get_integrity(bio_get_disk(bio));
+
+	if (bio_op(bio) != REQ_OP_READ && bio_op(bio) != REQ_OP_WRITE)
+		return false;
+
+	if (!bio_sectors(bio))
+		return false;
+
+	 /* Already protected? */
+	if (bio_integrity(bio))
+		return false;
+
+	if (bi == NULL)
+		return false;
+
+	if (bio_data_dir(bio) == READ && bi->profile->verify_fn != NULL &&
+	    (bi->flags & BLK_INTEGRITY_VERIFY))
+		return true;
+
+	if (bio_data_dir(bio) == WRITE && bi->profile->generate_fn != NULL &&
+	    (bi->flags & BLK_INTEGRITY_GENERATE))
+		return true;
+
+	return false;
+}
+#endif