X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdt%2Fmdt_reint.c;h=505b2541ad01ac3325b1c66d9b6f94ff36fda217;hp=a520d0a8ccb17b45f5dba952feeeb6f89f81f50c;hb=c6f3d533542b9462b5b8df95183d80321d4d9c34;hpb=a9e48fa940c45e51fd792c5d9fc4c647f64ed94e

diff --git a/lustre/mdt/mdt_reint.c b/lustre/mdt/mdt_reint.c
index a520d0a..505b254 100644
--- a/lustre/mdt/mdt_reint.c
+++ b/lustre/mdt/mdt_reint.c
@@ -217,78 +217,6 @@ static int mdt_lookup_version_check(struct mdt_thread_info *info,
 
 }
 
-static inline int mdt_remote_permission_check(struct mdt_thread_info *info)
-{
-	struct lu_ucred	*uc  = mdt_ucred(info);
-	struct mdt_device *mdt = info->mti_mdt;
-
-	if (!md_capable(uc, CFS_CAP_SYS_ADMIN)) {
-		if (uc->uc_gid != mdt->mdt_enable_remote_dir_gid &&
-		    mdt->mdt_enable_remote_dir_gid != -1)
-			return -EPERM;
-	}
-
-	return 0;
-}
-
-/**
- * mdt_remote_permission: Check whether the remote operation is permitted,
- *
- * Only sysadmin can create remote directory / striped directory,
- * migrate directory and set default stripedEA on directory, unless
- *
- * lctl set_param mdt.*.enable_remote_dir_gid=allow_gid.
- *
- * param[in] info: mdt_thread_info.
- *
- * retval	= 0 remote operation is allowed.
- *              < 0 remote operation is denied.
- */
-int mdt_remote_permission(struct mdt_thread_info *info)
-{
-	struct md_op_spec *spec = &info->mti_spec;
-	struct lu_attr *attr = &info->mti_attr.ma_attr;
-	struct obd_export *exp = mdt_info_req(info)->rq_export;
-	int rc;
-
-	if (info->mti_rr.rr_opcode == REINT_MIGRATE) {
-		rc = mdt_remote_permission_check(info);
-		if (rc != 0)
-			return rc;
-	}
-
-	if (info->mti_rr.rr_opcode == REINT_CREATE &&
-	    (S_ISDIR(attr->la_mode) && spec->u.sp_ea.eadata != NULL &&
-	     spec->u.sp_ea.eadatalen != 0)) {
-		const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
-
-		/* Only new clients can create remote dir( >= 2.4) and
-		 * striped dir(>= 2.6), old client will return -ENOTSUPP */
-		if (!mdt_is_dne_client(exp))
-			return -ENOTSUPP;
-
-		if (le32_to_cpu(lum->lum_stripe_count) > 1 &&
-		    !mdt_is_striped_client(exp))
-			return -ENOTSUPP;
-
-		rc = mdt_remote_permission_check(info);
-		if (rc != 0)
-			return rc;
-	}
-
-	if (info->mti_rr.rr_opcode == REINT_SETATTR) {
-		struct md_attr *ma = &info->mti_attr;
-
-		if ((ma->ma_valid & MA_LMV)) {
-			rc = mdt_remote_permission_check(info);
-			if (rc != 0)
-				return rc;
-		}
-	}
-
-	return 0;
-}
-
 static int mdt_unlock_slaves(struct mdt_thread_info *mti,
 			     struct mdt_object *obj,
 			     struct ldlm_enqueue_info *einfo,
@@ -423,6 +351,7 @@ static int mdt_create(struct mdt_thread_info *info)
 	struct mdt_body         *repbody;
 	struct md_attr          *ma = &info->mti_attr;
 	struct mdt_reint_record *rr = &info->mti_rr;
+	struct md_op_spec	*spec = &info->mti_spec;
 	int rc;
 	ENTRY;
 
@@ -433,6 +362,33 @@ static int mdt_create(struct mdt_thread_info *info)
 	if (!fid_is_md_operative(rr->rr_fid1))
 		RETURN(-EPERM);
 
+	if (S_ISDIR(ma->ma_attr.la_mode) &&
+	    spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) {
+		const struct lmv_user_md *lum = spec->u.sp_ea.eadata;
+		struct lu_ucred	*uc = mdt_ucred(info);
+		struct obd_export *exp = mdt_info_req(info)->rq_export;
+
+		/* Only new clients can create remote dir( >= 2.4) and
+		 * striped dir(>= 2.6), old client will return -ENOTSUPP */
+		if (!mdt_is_dne_client(exp))
+			RETURN(-ENOTSUPP);
+
+		if (le32_to_cpu(lum->lum_stripe_count) > 1) {
+			if (!mdt_is_striped_client(exp))
+				RETURN(-ENOTSUPP);
+
+			if (!mdt->mdt_enable_striped_dir)
+				RETURN(-EPERM);
+		} else if (!mdt->mdt_enable_remote_dir) {
+			RETURN(-EPERM);
+		}
+
+		if (!md_capable(uc, CFS_CAP_SYS_ADMIN) &&
+		    uc->uc_gid != mdt->mdt_enable_remote_dir_gid &&
+		    mdt->mdt_enable_remote_dir_gid != -1)
+			RETURN(-EPERM);
+	}
+
 	repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 
 	parent = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
@@ -474,10 +430,6 @@ static int mdt_create(struct mdt_thread_info *info)
 	if (unlikely(IS_ERR(child)))
 		GOTO(unlock_parent, rc = PTR_ERR(child));
 
-	rc = mdt_remote_permission(info);
-	if (rc != 0)
-		GOTO(put_child, rc);
-
 	ma->ma_need = MA_INODE;
 	ma->ma_valid = 0;
 
@@ -672,26 +624,27 @@ int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
 }
 
 static int mdt_reint_setattr(struct mdt_thread_info *info,
-                             struct mdt_lock_handle *lhc)
+			     struct mdt_lock_handle *lhc)
 {
-        struct md_attr          *ma = &info->mti_attr;
-        struct mdt_reint_record *rr = &info->mti_rr;
-        struct ptlrpc_request   *req = mdt_info_req(info);
-        struct mdt_object       *mo;
-        struct mdt_body         *repbody;
-	int			 rc, rc2;
-        ENTRY;
+	struct mdt_device *mdt = info->mti_mdt;
+	struct md_attr *ma = &info->mti_attr;
+	struct mdt_reint_record *rr = &info->mti_rr;
+	struct ptlrpc_request *req = mdt_info_req(info);
+	struct mdt_object *mo;
+	struct mdt_body *repbody;
+	int rc, rc2;
+	ENTRY;
 
-        DEBUG_REQ(D_INODE, req, "setattr "DFID" %x", PFID(rr->rr_fid1),
-                  (unsigned int)ma->ma_attr.la_valid);
+	DEBUG_REQ(D_INODE, req, "setattr "DFID" %x", PFID(rr->rr_fid1),
+		  (unsigned int)ma->ma_attr.la_valid);
 
 	if (info->mti_dlm_req)
 		ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
 
 	repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
-        mo = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid1);
-        if (IS_ERR(mo))
-                GOTO(out, rc = PTR_ERR(mo));
+	mo = mdt_object_find(info->mti_env, mdt, rr->rr_fid1);
+	if (IS_ERR(mo))
+		GOTO(out, rc = PTR_ERR(mo));
 
 	if (!mdt_object_exists(mo))
 		GOTO(out_put, rc = -ENOENT);
@@ -751,17 +704,32 @@ static int mdt_reint_setattr(struct mdt_thread_info *info,
 		if (ma->ma_valid & MA_LOV)
 			GOTO(out_put, rc = -EPROTO);
 
+		/* MDT supports FMD for regular files due to Data-on-MDT */
+		if (S_ISREG(lu_object_attr(&mo->mot_obj)) &&
+		    ma->ma_attr.la_valid & (LA_ATIME | LA_MTIME | LA_CTIME))
+			tgt_fmd_update(info->mti_exp, mdt_object_fid(mo),
+				       req->rq_xid);
+
 		rc = mdt_attr_set(info, mo, ma);
 		if (rc)
 			GOTO(out_put, rc);
 	} else if ((ma->ma_valid & (MA_LOV | MA_LMV)) &&
 		   (ma->ma_valid & MA_INODE)) {
-		struct lu_buf *buf  = &info->mti_buf;
-		struct mdt_lock_handle  *lh;
+		struct lu_buf *buf = &info->mti_buf;
+		struct lu_ucred *uc = mdt_ucred(info);
+		struct mdt_lock_handle *lh;
 
-		rc = mdt_remote_permission(info);
-		if (rc < 0)
-			GOTO(out_put, rc);
+		/* reject if either remote or striped dir is disabled */
+		if (ma->ma_valid & MA_LMV) {
+			if (!mdt->mdt_enable_remote_dir ||
+			    !mdt->mdt_enable_striped_dir)
+				GOTO(out_put, rc = -EPERM);
+
+			if (!md_capable(uc, CFS_CAP_SYS_ADMIN) &&
+			    uc->uc_gid != mdt->mdt_enable_remote_dir_gid &&
+			    mdt->mdt_enable_remote_dir_gid != -1)
+				GOTO(out_put, rc = -EPERM);
+		}
 
 		if (ma->ma_attr.la_valid != 0)
 			GOTO(out_put, rc = -EPROTO);
@@ -953,7 +921,7 @@ relock:
 	if (!cos_incompat) {
 		rc = mdt_object_striped(info, mc);
 		if (rc < 0)
-			GOTO(unlock_parent, rc = PTR_ERR(mc));
+			GOTO(unlock_parent, rc);
 
 		cos_incompat = rc;
 		if (cos_incompat) {
@@ -1313,7 +1281,7 @@ static void mdt_rename_unlock(struct lustre_handle *lh)
 	EXIT;
 }
 
-static struct mdt_object *mdt_object_find_check(struct mdt_thread_info *info,
+static struct mdt_object *mdt_parent_find_check(struct mdt_thread_info *info,
 						const struct lu_fid *fid,
 						int idx)
 {
@@ -1331,6 +1299,12 @@ static struct mdt_object *mdt_object_find_check(struct mdt_thread_info *info,
 	if (rc)
 		GOTO(out_put, rc);
 
+	if (!mdt_object_exists(dir))
+		GOTO(out_put, rc = -ENOENT);
+
+	if (!S_ISDIR(lu_object_attr(&dir->mot_obj)))
+		GOTO(out_put, rc = -ENOTDIR);
+
 	RETURN(dir);
 out_put:
 	mdt_object_put(info->mti_env, dir);
@@ -1341,9 +1315,9 @@ out_put:
  * in case obj is remote obj on its parent, revoke LOOKUP lock,
  * herein we don't really check it, just do revoke.
  */
-static int mdt_revoke_remote_lookup_lock(struct mdt_thread_info *info,
-					 struct mdt_object *pobj,
-					 struct mdt_object *obj)
+int mdt_revoke_remote_lookup_lock(struct mdt_thread_info *info,
+				  struct mdt_object *pobj,
+				  struct mdt_object *obj)
 {
 	struct mdt_lock_handle *lh = &info->mti_lh[MDT_LH_LOCAL];
 	int rc;
@@ -1352,10 +1326,11 @@ static int mdt_revoke_remote_lookup_lock(struct mdt_thread_info *info,
 	mdt_lock_reg_init(lh, LCK_EX);
 
 	if (mdt_object_remote(pobj)) {
+		/* don't bother to check if pobj and obj are on the same MDT. */
 		rc = mdt_remote_object_lock(info, pobj, mdt_object_fid(obj),
 					    &lh->mlh_rreg_lh, LCK_EX,
 					    MDS_INODELOCK_LOOKUP, false);
-	} else {
+	} else if (mdt_object_remote(obj)) {
 		struct ldlm_res_id *res = &info->mti_res_id;
 		union ldlm_policy_data *policy = &info->mti_policy;
 		__u64 dlmflags = LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB |
@@ -1367,6 +1342,9 @@ static int mdt_revoke_remote_lookup_lock(struct mdt_thread_info *info,
 		rc = mdt_fid_lock(info->mti_env, info->mti_mdt->mdt_namespace,
 				  &lh->mlh_reg_lh, LCK_EX, policy, res,
 				  dlmflags, NULL);
+	} else {
+		/* do nothing if both are local */
+		return 0;
 	}
 
 	if (rc != ELDLM_OK)
@@ -1404,6 +1382,21 @@ static void mdt_unlock_list(struct mdt_thread_info *info,
 	}
 }
 
+static inline void mdt_migrate_object_unlock(struct mdt_thread_info *info,
+					     struct mdt_object *obj,
+					     struct mdt_lock_handle *lh,
+					     struct ldlm_enqueue_info *einfo,
+					     struct list_head *slave_locks,
+					     int decref)
+{
+	if (mdt_object_remote(obj)) {
+		mdt_unlock_list(info, slave_locks, decref);
+		mdt_object_unlock(info, obj, lh, decref);
+	} else {
+		mdt_reint_striped_unlock(info, obj, lh, einfo, decref);
+	}
+}
+
 /*
  * lock parents of links, and also check whether total locks don't exceed
  * RS_MAX_LOCKS.
@@ -1412,18 +1405,20 @@ static void mdt_unlock_list(struct mdt_thread_info *info,
  * \retval	1 on success, but total lock count may exceed RS_MAX_LOCKS
  * \retval	-ev negative errno upon error
  */
-static int mdt_lock_links(struct mdt_thread_info *info,
-			  struct mdt_object *pobj,
-			  const struct md_attr *ma,
-			  struct mdt_object *obj,
-			  struct list_head *link_locks)
+static int mdt_link_parents_lock(struct mdt_thread_info *info,
+				 struct mdt_object *pobj,
+				 const struct md_attr *ma,
+				 struct mdt_object *obj,
+				 struct mdt_lock_handle *lhp,
+				 struct ldlm_enqueue_info *peinfo,
+				 struct list_head *parent_slave_locks,
+				 struct list_head *link_locks)
 {
 	struct mdt_device *mdt = info->mti_mdt;
 	struct lu_buf *buf = &info->mti_big_buf;
 	struct lu_name *lname = &info->mti_name;
 	struct linkea_data ldata = { NULL };
 	bool blocked = false;
-	int retries = 5;
 	int local_lnkp_cnt = 0;
 	int rc;
 
@@ -1444,7 +1439,6 @@ static int mdt_lock_links(struct mdt_thread_info *info,
 		RETURN(rc);
 	}
 
-repeat:
 	for (linkea_first_entry(&ldata); ldata.ld_lee && !rc;
 	     linkea_next_entry(&ldata)) {
 		struct mdt_object *lnkp;
@@ -1547,12 +1541,19 @@ repeat:
 		rc = mdt_object_lock_try(info, lnkp, &msl->msl_lh, &ibits,
 					 MDS_INODELOCK_UPDATE, true);
 		if (!(ibits & MDS_INODELOCK_UPDATE)) {
-			blocked = true;
 
-			CDEBUG(D_INFO, "busy lock on "DFID" "DNAME" retry %d\n",
-			       PFID(&fid), PNAME(lname), retries);
+			CDEBUG(D_INFO, "busy lock on "DFID" "DNAME"\n",
+			       PFID(&fid), PNAME(lname));
 
 			mdt_unlock_list(info, link_locks, 1);
+			/* also unlock parent locks to avoid deadlock */
+			if (!blocked)
+				mdt_migrate_object_unlock(info, pobj, lhp,
+							  peinfo,
+							  parent_slave_locks,
+							  1);
+
+			blocked = true;
 
 			mdt_lock_pdo_init(&msl->msl_lh, LCK_PW, lname);
 			rc = mdt_object_lock(info, lnkp, &msl->msl_lh,
@@ -1593,15 +1594,8 @@ repeat:
 		rc = mdt_revoke_remote_lookup_lock(info, lnkp, obj);
 	}
 
-	if (blocked) {
-		rc = -EBUSY;
-		if (--retries > 0) {
-			mdt_unlock_list(info, link_locks, rc);
-			blocked = false;
-			local_lnkp_cnt = 0;
-			goto repeat;
-		}
-	}
+	if (blocked)
+		GOTO(out, rc = -EBUSY);
 
 	EXIT;
 out:
@@ -1667,7 +1661,6 @@ static int mdt_lock_remote_slaves(struct mdt_thread_info *info,
 		INIT_LIST_HEAD(&msl->msl_linkage);
 		msl->msl_obj = slave;
 		list_add_tail(&msl->msl_linkage, slave_locks);
-
 	}
 	EXIT;
 
@@ -1677,21 +1670,6 @@ out:
 	return rc;
 }
 
-static inline void mdt_migrate_object_unlock(struct mdt_thread_info *info,
-					     struct mdt_object *obj,
-					     struct mdt_lock_handle *lh,
-					     struct ldlm_enqueue_info *einfo,
-					     struct list_head *slave_locks,
-					     int decref)
-{
-	if (mdt_object_remote(obj)) {
-		mdt_unlock_list(info, slave_locks, decref);
-		mdt_object_unlock(info, obj, lh, decref);
-	} else {
-		mdt_reint_striped_unlock(info, obj, lh, einfo, decref);
-	}
-}
-
 /* lock parent and its stripes */
 static int mdt_migrate_parent_lock(struct mdt_thread_info *info,
 				   struct mdt_object *obj,
@@ -1741,7 +1719,6 @@ static int mdt_migrate_object_lock(struct mdt_thread_info *info,
 	int rc;
 
 	if (mdt_object_remote(obj)) {
-		/* don't bother to check if pobj and obj are on the same MDT. */
 		rc = mdt_revoke_remote_lookup_lock(info, pobj, obj);
 		if (rc)
 			return rc;
@@ -1962,11 +1939,14 @@ close:
  *  9. unlock above locks
  * 10. sync device if source has links
  */
-static int mdt_reint_migrate_internal(struct mdt_thread_info *info)
+static int mdt_reint_migrate(struct mdt_thread_info *info,
+			     struct mdt_lock_handle *unused)
 {
 	const struct lu_env *env = info->mti_env;
 	struct mdt_device *mdt = info->mti_mdt;
+	struct ptlrpc_request *req = mdt_info_req(info);
 	struct mdt_reint_record *rr = &info->mti_rr;
+	struct lu_ucred *uc = mdt_ucred(info);
 	struct md_attr *ma = &info->mti_attr;
 	struct ldlm_enqueue_info *peinfo = &info->mti_einfo[0];
 	struct ldlm_enqueue_info *seinfo = &info->mti_einfo[1];
@@ -1974,12 +1954,14 @@ static int mdt_reint_migrate_internal(struct mdt_thread_info *info)
 	struct mdt_object *spobj = NULL;
 	struct mdt_object *sobj = NULL;
 	struct mdt_object *tobj;
+	struct lustre_handle rename_lh = { 0 };
 	struct mdt_lock_handle *lhp;
 	struct mdt_lock_handle *lhs;
 	struct mdt_lock_handle *lht;
 	LIST_HEAD(parent_slave_locks);
 	LIST_HEAD(child_slave_locks);
 	LIST_HEAD(link_locks);
+	int lock_retries = 5;
 	bool open_sem_locked = false;
 	bool do_sync = false;
 	int rc;
@@ -1988,18 +1970,44 @@ static int mdt_reint_migrate_internal(struct mdt_thread_info *info)
 	CDEBUG(D_INODE, "migrate "DFID"/"DNAME" to "DFID"\n", PFID(rr->rr_fid1),
 	       PNAME(&rr->rr_name), PFID(rr->rr_fid2));
 
+	if (info->mti_dlm_req)
+		ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
+
+	if (!fid_is_md_operative(rr->rr_fid1) ||
+	    !fid_is_md_operative(rr->rr_fid2))
+		RETURN(-EPERM);
+
 	/* don't allow migrate . or .. */
 	if (lu_name_is_dot_or_dotdot(&rr->rr_name))
 		RETURN(-EBUSY);
 
-	rc = mdt_remote_permission(info);
-	if (rc)
-		RETURN(rc);
+	if (!mdt->mdt_enable_remote_dir || !mdt->mdt_enable_dir_migration)
+		RETURN(-EPERM);
+
+	if (!md_capable(uc, CFS_CAP_SYS_ADMIN) &&
+	    uc->uc_gid != mdt->mdt_enable_remote_dir_gid &&
+	    mdt->mdt_enable_remote_dir_gid != -1)
+		RETURN(-EPERM);
+
+	/*
+	 * Note: do not enqueue rename lock for replay request, because
+	 * if other MDT holds rename lock, but being blocked to wait for
+	 * this MDT to finish its recovery, and the failover MDT can not
+	 * get rename lock, which will cause deadlock.
+	 */
+	if (!req_is_replay(req)) {
+		rc = mdt_rename_lock(info, &rename_lh);
+		if (rc != 0) {
+			CERROR("%s: can't lock FS for rename: rc = %d\n",
+			       mdt_obd_name(info->mti_mdt), rc);
+			RETURN(rc);
+		}
+	}
 
 	/* pobj is master object of parent */
-	pobj = mdt_object_find_check(info, rr->rr_fid1, 0);
+	pobj = mdt_parent_find_check(info, rr->rr_fid1, 0);
 	if (IS_ERR(pobj))
-		RETURN(PTR_ERR(pobj));
+		GOTO(unlock_rename, rc = PTR_ERR(pobj));
 
 	if (unlikely(!info->mti_big_lmm)) {
 		info->mti_big_lmmsize = lmv_mds_md_size(64, LMV_MAGIC);
@@ -2015,6 +2023,7 @@ static int mdt_reint_migrate_internal(struct mdt_thread_info *info)
 	if (rc)
 		GOTO(put_parent, rc);
 
+lock_parent:
 	/* lock parent object */
 	lhp = &info->mti_lh[MDT_LH_PARENT];
 	mdt_lock_reg_init(lhp, LCK_PW);
@@ -2033,7 +2042,14 @@ static int mdt_reint_migrate_internal(struct mdt_thread_info *info)
 		GOTO(unlock_parent, rc);
 
 	/* lock parents of source links, and revoke LOOKUP lock of links */
-	rc = mdt_lock_links(info, pobj, ma, sobj, &link_locks);
+	rc = mdt_link_parents_lock(info, pobj, ma, sobj, lhp, peinfo,
+				   &parent_slave_locks, &link_locks);
+	if (rc == -EBUSY && lock_retries-- > 0) {
+		mdt_object_put(env, sobj);
+		mdt_object_put(env, spobj);
+		goto lock_parent;
+	}
+
 	if (rc < 0)
 		GOTO(put_source, rc);
 
@@ -2044,6 +2060,20 @@ static int mdt_reint_migrate_internal(struct mdt_thread_info *info)
 	 */
 	do_sync = rc;
 
+	/* TODO: DoM migration is not supported yet */
+	if (S_ISREG(lu_object_attr(&sobj->mot_obj))) {
+		ma->ma_lmm = info->mti_big_lmm;
+		ma->ma_lmm_size = info->mti_big_lmmsize;
+		ma->ma_valid = 0;
+		rc = mdt_stripe_get(info, sobj, ma, XATTR_NAME_LOV);
+		if (rc)
+			GOTO(put_source, rc);
+
+		if (ma->ma_valid & MA_LOV &&
+		    mdt_lmm_dom_entry(ma->ma_lmm) != LMM_NO_DOM)
+			GOTO(put_source, rc = -EOPNOTSUPP);
+	}
+
 	/* if migration HSM is allowed */
 	if (!mdt->mdt_opts.mo_migrate_hsm_allowed) {
 		ma->ma_need = MA_HSM;
@@ -2109,7 +2139,7 @@ unlock_open_sem:
 	if (open_sem_locked)
 		up_write(&sobj->mot_open_sem);
 unlock_links:
-	mdt_unlock_list(info, &link_locks, rc);
+	mdt_unlock_list(info, &link_locks, do_sync ? 1 : rc);
 put_source:
 	mdt_object_put(env, sobj);
 	mdt_object_put(env, spobj);
@@ -2118,6 +2148,9 @@ unlock_parent:
 				  &parent_slave_locks, rc);
 put_parent:
 	mdt_object_put(env, pobj);
+unlock_rename:
+	if (lustre_handle_is_used(&rename_lh))
+		mdt_rename_unlock(&rename_lh);
 
 	if (!rc && do_sync)
 		mdt_device_sync(env, mdt);
@@ -2240,20 +2273,10 @@ static int mdt_rename_determine_lock_order(struct mdt_thread_info *info,
  * 2 - srcdir child; 3 - tgtdir child.
  * Update on disk version of srcdir child.
  */
-/**
- * For DNE phase I, only these renames are allowed
- *	mv src_p/src_c tgt_p/tgt_c
- * 1. src_p/src_c/tgt_p/tgt_c are in the same MDT.
- * 2. src_p and tgt_p are same directory, and tgt_c does not
- *    exists. In this case, all of modification will happen
- *    in the MDT where ithesource parent is, only one remote
- *    update is needed, i.e. set c_time/m_time on the child.
- *    And tgt_c will be still in the same MDT as the original
- *    src_c.
- */
-static int mdt_reint_rename_internal(struct mdt_thread_info *info,
-				     struct mdt_lock_handle *lhc)
+static int mdt_reint_rename(struct mdt_thread_info *info,
+			    struct mdt_lock_handle *unused)
 {
+	struct mdt_device *mdt = info->mti_mdt;
 	struct mdt_reint_record *rr = &info->mti_rr;
 	struct md_attr *ma = &info->mti_attr;
 	struct ptlrpc_request *req = mdt_info_req(info);
@@ -2261,6 +2284,7 @@ static int mdt_reint_rename_internal(struct mdt_thread_info *info,
 	struct mdt_object *mtgtdir = NULL;
 	struct mdt_object *mold;
 	struct mdt_object *mnew = NULL;
+	struct lustre_handle rename_lh = { 0 };
 	struct mdt_lock_handle *lh_srcdirp;
 	struct mdt_lock_handle *lh_tgtdirp;
 	struct mdt_lock_handle *lh_oldp = NULL;
@@ -2277,8 +2301,15 @@ static int mdt_reint_rename_internal(struct mdt_thread_info *info,
 		  PFID(rr->rr_fid1), PNAME(&rr->rr_name),
 		  PFID(rr->rr_fid2), PNAME(&rr->rr_tgt_name));
 
+	if (info->mti_dlm_req)
+		ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
+
+	if (!fid_is_md_operative(rr->rr_fid1) ||
+	    !fid_is_md_operative(rr->rr_fid2))
+		RETURN(-EPERM);
+
 	/* find both parents. */
-	msrcdir = mdt_object_find_check(info, rr->rr_fid1, 0);
+	msrcdir = mdt_parent_find_check(info, rr->rr_fid1, 0);
 	if (IS_ERR(msrcdir))
 		RETURN(PTR_ERR(msrcdir));
 
@@ -2288,14 +2319,41 @@ static int mdt_reint_rename_internal(struct mdt_thread_info *info,
 		mtgtdir = msrcdir;
 		mdt_object_get(info->mti_env, mtgtdir);
 	} else {
-		mtgtdir = mdt_object_find_check(info, rr->rr_fid2, 1);
+		mtgtdir = mdt_parent_find_check(info, rr->rr_fid2, 1);
 		if (IS_ERR(mtgtdir))
 			GOTO(out_put_srcdir, rc = PTR_ERR(mtgtdir));
 	}
 
+	/*
+	 * Note: do not enqueue rename lock for replay request, because
+	 * if other MDT holds rename lock, but being blocked to wait for
+	 * this MDT to finish its recovery, and the failover MDT can not
+	 * get rename lock, which will cause deadlock.
+	 */
+	if (!req_is_replay(req)) {
+		/*
+		 * Normally rename RPC is handled on the MDT with the target
+		 * directory (if target exists, it's on the MDT with the
+		 * target), if the source directory is remote, it's a hint that
+		 * source is remote too (this may not be true, but it won't
+		 * cause any issue), return -EXDEV early to avoid taking
+		 * rename_lock.
+		 */
+		if (!mdt->mdt_enable_remote_rename &&
+		    mdt_object_remote(msrcdir))
+			GOTO(out_put_tgtdir, rc = -EXDEV);
+
+		rc = mdt_rename_lock(info, &rename_lh);
+		if (rc != 0) {
+			CERROR("%s: can't lock FS for rename: rc = %d\n",
+			       mdt_obd_name(mdt), rc);
+			GOTO(out_put_tgtdir, rc);
+		}
+	}
+
 	rc = mdt_rename_determine_lock_order(info, msrcdir, mtgtdir);
 	if (rc < 0)
-		GOTO(out_put_tgtdir, rc);
+		GOTO(out_unlock_rename, rc);
 
 	reverse = rc;
 
@@ -2320,7 +2378,7 @@ relock:
 		rc = mdt_object_lock_save(info, mtgtdir, lh_tgtdirp, 1,
 					  cos_incompat);
 		if (rc)
-			GOTO(out_put_tgtdir, rc);
+			GOTO(out_unlock_rename, rc);
 
 		OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME, 5);
 
@@ -2328,20 +2386,21 @@ relock:
 					  cos_incompat);
 		if (rc != 0) {
 			mdt_object_unlock(info, mtgtdir, lh_tgtdirp, rc);
-			GOTO(out_put_tgtdir, rc);
+			GOTO(out_unlock_rename, rc);
 		}
 	} else {
 		rc = mdt_object_lock_save(info, msrcdir, lh_srcdirp, 0,
 					  cos_incompat);
 		if (rc)
-			GOTO(out_put_tgtdir, rc);
+			GOTO(out_unlock_rename, rc);
 
 		OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RENAME, 5);
 
 		if (mtgtdir != msrcdir) {
 			rc = mdt_object_lock_save(info, mtgtdir, lh_tgtdirp, 1,
 						  cos_incompat);
-		} else if (lh_srcdirp->mlh_pdo_hash !=
+		} else if (!mdt_object_remote(mtgtdir) &&
+			   lh_srcdirp->mlh_pdo_hash !=
 			   lh_tgtdirp->mlh_pdo_hash) {
 			rc = mdt_pdir_hash_lock(info, lh_tgtdirp, mtgtdir,
 						MDS_INODELOCK_UPDATE,
@@ -2350,7 +2409,7 @@ relock:
 		}
 		if (rc != 0) {
 			mdt_object_unlock(info, msrcdir, lh_srcdirp, rc);
-			GOTO(out_put_tgtdir, rc);
+			GOTO(out_unlock_rename, rc);
 		}
 	}
 
@@ -2380,6 +2439,9 @@ relock:
 		GOTO(out_put_old, rc = -ENOENT);
 	}
 
+	if (mdt_object_remote(mold) && !mdt->mdt_enable_remote_rename)
+		GOTO(out_put_old, rc = -EXDEV);
+
 	/* Check if @mtgtdir is subdir of @mold, before locking child
 	 * to avoid reverse locking. */
 	if (mtgtdir != msrcdir) {
@@ -2572,6 +2634,9 @@ out_put_old:
 out_unlock_parents:
 	mdt_object_unlock(info, mtgtdir, lh_tgtdirp, rc);
 	mdt_object_unlock(info, msrcdir, lh_srcdirp, rc);
+out_unlock_rename:
+	if (lustre_handle_is_used(&rename_lh))
+		mdt_rename_unlock(&rename_lh);
 out_put_tgtdir:
 	mdt_object_put(info->mti_env, mtgtdir);
 out_put_srcdir:
@@ -2587,58 +2652,6 @@ out_put_srcdir:
 	return rc;
 }
 
-static int mdt_reint_rename_or_migrate(struct mdt_thread_info *info,
-				       struct mdt_lock_handle *lhc, bool rename)
-{
-	struct mdt_reint_record *rr = &info->mti_rr;
-	struct ptlrpc_request   *req = mdt_info_req(info);
-	struct lustre_handle	rename_lh = { 0 };
-	int			rc;
-	ENTRY;
-
-	if (info->mti_dlm_req)
-		ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
-
-	if (!fid_is_md_operative(rr->rr_fid1) ||
-	    !fid_is_md_operative(rr->rr_fid2))
-		RETURN(-EPERM);
-
-	/* Note: do not enqueue rename lock for replay request, because
-	 * if other MDT holds rename lock, but being blocked to wait for
-	 * this MDT to finish its recovery, and the failover MDT can not
-	 * get rename lock, which will cause deadlock. */
-	if (!req_is_replay(req)) {
-		rc = mdt_rename_lock(info, &rename_lh);
-		if (rc != 0) {
-			CERROR("%s: can't lock FS for rename: rc = %d\n",
-			       mdt_obd_name(info->mti_mdt), rc);
-			RETURN(rc);
-		}
-	}
-
-	if (rename)
-		rc = mdt_reint_rename_internal(info, lhc);
-	else
-		rc = mdt_reint_migrate_internal(info);
-
-	if (lustre_handle_is_used(&rename_lh))
-		mdt_rename_unlock(&rename_lh);
-
-	RETURN(rc);
-}
-
-static int mdt_reint_rename(struct mdt_thread_info *info,
-			    struct mdt_lock_handle *lhc)
-{
-	return mdt_reint_rename_or_migrate(info, lhc, true);
-}
-
-static int mdt_reint_migrate(struct mdt_thread_info *info,
-			    struct mdt_lock_handle *lhc)
-{
-	return mdt_reint_rename_or_migrate(info, lhc, false);
-}
-
 static int mdt_reint_resync(struct mdt_thread_info *info,
 			    struct mdt_lock_handle *lhc)
 {
@@ -2648,7 +2661,7 @@ static int mdt_reint_resync(struct mdt_thread_info *info,
 	struct mdt_object       *mo;
 	struct ldlm_lock	*lease;
 	struct mdt_body         *repbody;
-	struct md_layout_change	 layout = { 0 };
+	struct md_layout_change	 layout = { .mlc_mirror_id = rr->rr_mirror_id };
 	bool			 lease_broken;
 	int			 rc, rc2;
 	ENTRY;