Whamcloud - gitweb
LU-6109 lfsck: check FID validity before locating object
[fs/lustre-release.git] / lustre / lfsck / lfsck_namespace.c
index 6641db8..6e29717 100644 (file)
@@ -20,7 +20,7 @@
  * GPL HEADER END
  */
 /*
- * Copyright (c) 2012, 2013, Intel Corporation.
+ * Copyright (c) 2013, 2014, Intel Corporation.
  */
 /*
  * lustre/lfsck/lfsck_namespace.c
 
 #include "lfsck_internal.h"
 
-#define LFSCK_NAMESPACE_MAGIC  0xA0629D03
+#define LFSCK_NAMESPACE_MAGIC_V1       0xA0629D03
+#define LFSCK_NAMESPACE_MAGIC_V2       0xA0621A0B
+
+/* For Lustre-2.x (x <= 6), the namespace LFSCK used LFSCK_NAMESPACE_MAGIC_V1
+ * as the trace file magic. When downgrade to such old release, the old LFSCK
+ * will not recognize the new LFSCK_NAMESPACE_MAGIC_V2 in the new trace file,
+ * then it will reset the whole LFSCK, and will not cause start failure. The
+ * similar case will happen when upgrade from such old release. */
+#define LFSCK_NAMESPACE_MAGIC          LFSCK_NAMESPACE_MAGIC_V2
 
 enum lfsck_nameentry_check {
        LFSCK_NAMEENTRY_DEAD            = 1, /* The object has been unlinked. */
@@ -49,8 +57,6 @@ enum lfsck_nameentry_check {
        LFSCK_NAMEENTRY_RECREATED       = 3, /* The entry has been recreated. */
 };
 
-static const char lfsck_namespace_name[] = "lfsck_namespace";
-
 static struct lfsck_namespace_req *
 lfsck_namespace_assistant_req_init(struct lfsck_instance *lfsck,
                                   struct lu_dirent *ent, __u16 type)
@@ -319,9 +325,19 @@ static int lfsck_namespace_load_bitmap(const struct lu_env *env,
 }
 
 /**
- * \retval +ve: the lfsck_namespace is broken, the caller should reset it.
- * \retval 0: succeed.
- * \retval -ve: failed cases.
+ * Load namespace LFSCK statistics information from the trace file.
+ *
+ * For old release (Lustre-2.6 or older), the statistics information was
+ * stored as XATTR_NAME_LFSCK_NAMESPACE_OLD EA. But in Lustre-2.7, we need
+ * more statistics information. To avoid confusing old MDT when downgrade,
+ * Lustre-2.7 stores the namespace LFSCK statistics information as new
+ * XATTR_NAME_LFSCK_NAMESPACE EA.
+ *
+ * \param[in] env      pointer to the thread context
+ * \param[in] com      pointer to the lfsck component
+ *
+ * \retval             0 for success
+ * \retval             negative error number on failure
  */
 static int lfsck_namespace_load(const struct lu_env *env,
                                struct lfsck_component *com)
@@ -341,7 +357,7 @@ static int lfsck_namespace_load(const struct lu_env *env,
                        CDEBUG(D_LFSCK, "%s: invalid lfsck_namespace magic "
                               "%#x != %#x\n", lfsck_lfsck2name(com->lc_lfsck),
                               ns->ln_magic, LFSCK_NAMESPACE_MAGIC);
-                       rc = 1;
+                       rc = -ESTALE;
                } else {
                        rc = 0;
                }
@@ -350,13 +366,22 @@ static int lfsck_namespace_load(const struct lu_env *env,
                       "expected = %d: rc = %d\n",
                       lfsck_lfsck2name(com->lc_lfsck), len, rc);
                if (rc >= 0)
-                       rc = 1;
+                       rc = -ESTALE;
+       } else {
+               /* Check whether it is old trace file or not.
+                * If yes, it should be reset via returning -ESTALE. */
+               rc = dt_xattr_get(env, com->lc_obj,
+                                 lfsck_buf_get(env, com->lc_file_disk, len),
+                                 XATTR_NAME_LFSCK_NAMESPACE_OLD, BYPASS_CAPA);
+               if (rc >= 0)
+                       rc = -ESTALE;
        }
+
        return rc;
 }
 
 static int lfsck_namespace_store(const struct lu_env *env,
-                                struct lfsck_component *com)
+                                struct lfsck_component *com, bool init)
 {
        struct dt_object                *obj    = com->lc_obj;
        struct lfsck_instance           *lfsck  = com->lc_lfsck;
@@ -367,6 +392,9 @@ static int lfsck_namespace_store(const struct lu_env *env,
        __u32                            nbits  = 0;
        int                              len    = com->lc_file_size;
        int                              rc;
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
+       struct lu_buf            tbuf   = { &len, sizeof(len) };
+#endif
        ENTRY;
 
        if (lad != NULL) {
@@ -398,6 +426,20 @@ static int lfsck_namespace_store(const struct lu_env *env,
                        GOTO(out, rc);
        }
 
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
+       /* To be compatible with old Lustre-2.x MDT (x <= 6), generate dummy
+        * XATTR_NAME_LFSCK_NAMESPACE_OLD EA, then when downgrade to Lustre-2.x,
+        * the old LFSCK will find "invalid" XATTR_NAME_LFSCK_NAMESPACE_OLD EA,
+        * then reset the namespace LFSCK trace file. */
+       if (init) {
+               rc = dt_declare_xattr_set(env, obj, &tbuf,
+                                         XATTR_NAME_LFSCK_NAMESPACE_OLD,
+                                         LU_XATTR_CREATE, handle);
+               if (rc != 0)
+                       GOTO(out, rc);
+       }
+#endif
+
        rc = dt_trans_start_local(env, lfsck->li_bottom, handle);
        if (rc != 0)
                GOTO(out, rc);
@@ -411,6 +453,13 @@ static int lfsck_namespace_store(const struct lu_env *env,
                                  XATTR_NAME_LFSCK_BITMAP, 0, handle,
                                  BYPASS_CAPA);
 
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 8, 53, 0)
+       if (rc == 0 && init)
+               rc = dt_xattr_set(env, obj, &tbuf,
+                                 XATTR_NAME_LFSCK_NAMESPACE_OLD,
+                                 LU_XATTR_CREATE, handle, BYPASS_CAPA);
+#endif
+
        GOTO(out, rc);
 
 out:
@@ -423,6 +472,70 @@ log:
        return rc;
 }
 
+static struct dt_object *
+lfsck_namespace_load_one_trace_file(const struct lu_env *env,
+                                   struct lfsck_component *com,
+                                   struct dt_object *parent,
+                                   const char *name,
+                                   const struct dt_index_features *ft,
+                                   bool reset)
+{
+       struct lfsck_instance   *lfsck = com->lc_lfsck;
+       struct dt_object        *obj;
+       int                      rc;
+
+       if (reset) {
+               rc = local_object_unlink(env, lfsck->li_bottom, parent, name);
+               if (rc != 0 && rc != -ENOENT)
+                       return ERR_PTR(rc);
+       }
+
+       if (ft != NULL)
+               obj = local_index_find_or_create(env, lfsck->li_los, parent,
+                                       name, S_IFREG | S_IRUGO | S_IWUSR, ft);
+       else
+               obj = local_file_find_or_create(env, lfsck->li_los, parent,
+                                       name, S_IFREG | S_IRUGO | S_IWUSR);
+
+       return obj;
+}
+
+static int lfsck_namespace_load_sub_trace_files(const struct lu_env *env,
+                                               struct lfsck_component *com,
+                                               bool reset)
+{
+       char                            *name = lfsck_env_info(env)->lti_key;
+       struct lfsck_sub_trace_obj      *lsto;
+       struct dt_object                *obj;
+       int                              rc;
+       int                              i;
+
+       for (i = 0, lsto = &com->lc_sub_trace_objs[0];
+            i < LFSCK_STF_COUNT; i++, lsto++) {
+               snprintf(name, NAME_MAX, "%s_%02d", LFSCK_NAMESPACE, i);
+               if (lsto->lsto_obj != NULL) {
+                       if (!reset)
+                               continue;
+
+                       lu_object_put(env, &lsto->lsto_obj->do_lu);
+                       lsto->lsto_obj = NULL;
+               }
+
+               obj = lfsck_namespace_load_one_trace_file(env, com,
+                                       com->lc_lfsck->li_lfsck_dir,
+                                       name, &dt_lfsck_features, reset);
+               if (IS_ERR(obj))
+                       return PTR_ERR(obj);
+
+               lsto->lsto_obj = obj;
+               rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_features);
+               if (rc != 0)
+                       return rc;
+       }
+
+       return 0;
+}
+
 static int lfsck_namespace_init(const struct lu_env *env,
                                struct lfsck_component *com)
 {
@@ -433,8 +546,11 @@ static int lfsck_namespace_init(const struct lu_env *env,
        ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
        ns->ln_status = LS_INIT;
        down_write(&com->lc_sem);
-       rc = lfsck_namespace_store(env, com);
+       rc = lfsck_namespace_store(env, com, true);
        up_write(&com->lc_sem);
+       if (rc == 0)
+               rc = lfsck_namespace_load_sub_trace_files(env, com, true);
+
        return rc;
 }
 
@@ -456,10 +572,11 @@ int lfsck_namespace_trace_update(const struct lu_env *env,
                                 const __u8 flags, bool add)
 {
        struct lfsck_instance   *lfsck  = com->lc_lfsck;
-       struct dt_object        *obj    = com->lc_obj;
+       struct dt_object        *obj;
        struct lu_fid           *key    = &lfsck_env_info(env)->lti_fid3;
        struct dt_device        *dev    = lfsck->li_bottom;
        struct thandle          *th     = NULL;
+       int                      idx;
        int                      rc     = 0;
        __u8                     old    = 0;
        __u8                     new    = 0;
@@ -467,7 +584,12 @@ int lfsck_namespace_trace_update(const struct lu_env *env,
 
        LASSERT(flags != 0);
 
-       down_write(&com->lc_sem);
+       if (unlikely(!fid_is_sane(fid)))
+               RETURN(0);
+
+       idx = lfsck_sub_trace_file_fid2idx(fid);
+       obj = com->lc_sub_trace_objs[idx].lsto_obj;
+       mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
        fid_cpu_to_be(key, fid);
        rc = dt_lookup(env, obj, (struct dt_rec *)&old,
                       (const struct dt_key *)key, BYPASS_CAPA);
@@ -542,7 +664,7 @@ log:
               (__u32)flags, (__u32)old, (__u32)new, rc);
 
 unlock:
-       up_write(&com->lc_sem);
+       mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
 
        return rc;
 }
@@ -792,17 +914,17 @@ static int lfsck_namespace_insert_orphan(const struct lu_env *env,
        struct lfsck_thread_info        *info   = lfsck_env_info(env);
        struct lu_name                  *cname  = &info->lti_name;
        struct dt_insert_rec            *rec    = &info->lti_dt_rec;
-       struct lu_fid                   *tfid   = &info->lti_fid5;
        struct lu_attr                  *la     = &info->lti_la3;
        const struct lu_fid             *cfid   = lfsck_dto2fid(orphan);
        const struct lu_fid             *pfid;
+       struct lu_fid                    tfid;
        struct lfsck_instance           *lfsck  = com->lc_lfsck;
        struct dt_device                *dev    = lfsck->li_bottom;
        struct dt_object                *parent;
        struct thandle                  *th     = NULL;
        struct lustre_handle             plh    = { 0 };
        struct lustre_handle             clh    = { 0 };
-       struct linkea_data               ldata  = { 0 };
+       struct linkea_data               ldata  = { NULL };
        struct lu_buf                    linkea_buf;
        int                              namelen;
        int                              idx    = 0;
@@ -811,12 +933,8 @@ static int lfsck_namespace_insert_orphan(const struct lu_env *env,
        ENTRY;
 
        cname->ln_name = NULL;
-       /* Create .lustre/lost+found/MDTxxxx when needed. */
-       if (unlikely(lfsck->li_lpf_obj == NULL)) {
-               rc = lfsck_create_lpf(env, lfsck);
-               if (rc != 0)
-                       GOTO(log, rc);
-       }
+       if (unlikely(lfsck->li_lpf_obj == NULL))
+               GOTO(log, rc = -ENXIO);
 
        parent = lfsck->li_lpf_obj;
        pfid = lfsck_dto2fid(parent);
@@ -830,13 +948,13 @@ static int lfsck_namespace_insert_orphan(const struct lu_env *env,
        do {
                namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
                                   PFID(cfid), infix, type, idx++);
-               rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
+               rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
                               (const struct dt_key *)info->lti_key,
                               BYPASS_CAPA);
                if (rc != 0 && rc != -ENOENT)
                        GOTO(log, rc);
 
-               if (unlikely(rc == 0 && lu_fid_eq(cfid, tfid)))
+               if (unlikely(rc == 0 && lu_fid_eq(cfid, &tfid)))
                        exist = true;
        } while (rc == 0 && !exist);
 
@@ -954,10 +1072,8 @@ static int lfsck_namespace_insert_orphan(const struct lu_env *env,
                }
        }
 
-       if (rc != 0)
-               GOTO(unlock, rc);
-
-       rc = dt_attr_set(env, orphan, la, th, BYPASS_CAPA);
+       if (rc == 0)
+               rc = dt_attr_set(env, orphan, la, th, BYPASS_CAPA);
 
        GOTO(stop, rc = (rc == 0 ? 1 : rc));
 
@@ -1113,116 +1229,7 @@ log:
 }
 
 /**
- * Create the specified orphan MDT-object on remote MDT.
- *
- * The LFSCK instance on this MDT will send LFSCK RPC to remote MDT to
- * ask the remote LFSCK instance to create the specified orphan object
- * under .lustre/lost+found/MDTxxxx/ directory with the name:
- * ${FID}-P-${conflict_version}.
- *
- * \param[in] env      pointer to the thread context
- * \param[in] com      pointer to the lfsck component
- * \param[in] orphan   pointer to the orphan MDT-object
- * \param[in] type     the orphan's type to be created
- *
- *  type "P":          The orphan object to be created was a parent directory
- *                     of some MDT-object which linkEA shows that the @orphan
- *                     object is missing.
- *
- * \see lfsck_layout_recreate_parent() for more types.
- *
- * \param[in] lmv      pointer to master LMV EA that will be set to the orphan
- *
- * \retval             positive number for repaired cases
- * \retval             0 if needs to repair nothing
- * \retval             negative error number on failure
- */
-static int lfsck_namespace_create_orphan_remote(const struct lu_env *env,
-                                               struct lfsck_component *com,
-                                               struct dt_object *orphan,
-                                               __u32 type,
-                                               struct lmv_mds_md_v1 *lmv)
-{
-       struct lfsck_thread_info        *info   = lfsck_env_info(env);
-       struct lfsck_request            *lr     = &info->lti_lr;
-       struct lu_seq_range             *range  = &info->lti_range;
-       const struct lu_fid             *fid    = lfsck_dto2fid(orphan);
-       struct lfsck_namespace          *ns     = com->lc_file_ram;
-       struct lfsck_instance           *lfsck  = com->lc_lfsck;
-       struct seq_server_site          *ss     =
-                       lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site);
-       struct lfsck_tgt_desc           *ltd    = NULL;
-       struct ptlrpc_request           *req    = NULL;
-       int                              rc;
-       ENTRY;
-
-       if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
-               GOTO(out, rc = 1);
-
-       fld_range_set_mdt(range);
-       rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
-       if (rc != 0)
-               GOTO(out, rc);
-
-       ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, range->lsr_index);
-       if (ltd == NULL) {
-               ns->ln_flags |= LF_INCOMPLETE;
-
-               GOTO(out, rc = -ENODEV);
-       }
-
-       req = ptlrpc_request_alloc(class_exp2cliimp(ltd->ltd_exp),
-                                  &RQF_LFSCK_NOTIFY);
-       if (req == NULL)
-               GOTO(out, rc = -ENOMEM);
-
-       rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
-       if (rc != 0) {
-               ptlrpc_request_free(req);
-
-               GOTO(out, rc);
-       }
-
-       lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
-       memset(lr, 0, sizeof(*lr));
-       lr->lr_event = LE_CREATE_ORPHAN;
-       lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
-       lr->lr_active = LFSCK_TYPE_NAMESPACE;
-       lr->lr_fid = *fid;
-       lr->lr_type = type;
-       if (lmv != NULL) {
-               lr->lr_hash_type = lmv->lmv_hash_type;
-               lr->lr_stripe_count = lmv->lmv_stripe_count;
-               lr->lr_layout_version = lmv->lmv_layout_version;
-               memcpy(lr->lr_pool_name, lmv->lmv_pool_name,
-                      sizeof(lr->lr_pool_name));
-       }
-
-       ptlrpc_request_set_replen(req);
-       rc = ptlrpc_queue_wait(req);
-       ptlrpc_req_finished(req);
-
-       if (rc == 0)
-               rc = 1;
-       else if (rc == -EEXIST)
-               rc = 0;
-
-       GOTO(out, rc);
-
-out:
-       CDEBUG(D_LFSCK, "%s: namespace LFSCK create object "
-              DFID" on the MDT %x remotely: rc = %d\n",
-              lfsck_lfsck2name(lfsck), PFID(fid),
-              ltd != NULL ? ltd->ltd_index : -1, rc);
-
-       if (ltd != NULL)
-               lfsck_tgt_put(ltd);
-
-       return rc;
-}
-
-/**
- * Create the specified orphan MDT-object locally.
+ * Create the specified orphan directory.
  *
  * For the case that the parent MDT-object stored in some MDT-object's
  * linkEA entry is lost, the LFSCK will re-create the parent object as
@@ -1232,24 +1239,15 @@ out:
  * \param[in] env      pointer to the thread context
  * \param[in] com      pointer to the lfsck component
  * \param[in] orphan   pointer to the orphan MDT-object to be created
- * \param[in] type     the orphan's type to be created
- *
- *  type "P":          The orphan object to be created was a parent directory
- *                     of some MDT-object which linkEA shows that the @orphan
- *                     object is missing.
- *
- * \see lfsck_layout_recreate_parent() for more types.
- *
  * \param[in] lmv      pointer to master LMV EA that will be set to the orphan
  *
  * \retval             positive number for repaired cases
  * \retval             negative error number on failure
  */
-static int lfsck_namespace_create_orphan_local(const struct lu_env *env,
-                                              struct lfsck_component *com,
-                                              struct dt_object *orphan,
-                                              __u32 type,
-                                              struct lmv_mds_md_v1 *lmv)
+static int lfsck_namespace_create_orphan_dir(const struct lu_env *env,
+                                            struct lfsck_component *com,
+                                            struct dt_object *orphan,
+                                            struct lmv_mds_md_v1 *lmv)
 {
        struct lfsck_thread_info        *info   = lfsck_env_info(env);
        struct lu_attr                  *la     = &info->lti_la;
@@ -1257,17 +1255,17 @@ static int lfsck_namespace_create_orphan_local(const struct lu_env *env,
        struct dt_object_format         *dof    = &info->lti_dof;
        struct lu_name                  *cname  = &info->lti_name2;
        struct dt_insert_rec            *rec    = &info->lti_dt_rec;
-       struct lu_fid                   *tfid   = &info->lti_fid;
        struct lmv_mds_md_v1            *lmv2   = &info->lti_lmv2;
        const struct lu_fid             *cfid   = lfsck_dto2fid(orphan);
-       const struct lu_fid             *pfid;
+       struct lu_fid                    tfid;
        struct lfsck_instance           *lfsck  = com->lc_lfsck;
-       struct dt_device                *dev    = lfsck->li_bottom;
+       struct lfsck_namespace          *ns     = com->lc_file_ram;
+       struct dt_device                *dev;
        struct dt_object                *parent = NULL;
        struct dt_object                *child  = NULL;
        struct thandle                  *th     = NULL;
        struct lustre_handle             lh     = { 0 };
-       struct linkea_data               ldata  = { 0 };
+       struct linkea_data               ldata  = { NULL };
        struct lu_buf                    linkea_buf;
        struct lu_buf                    lmv_buf;
        char                             name[32];
@@ -1277,26 +1275,45 @@ static int lfsck_namespace_create_orphan_local(const struct lu_env *env,
        ENTRY;
 
        LASSERT(!dt_object_exists(orphan));
-       LASSERT(!dt_object_remote(orphan));
-
-       /* @orphan maybe not attached to lfsck->li_bottom */
-       child = lfsck_object_find_by_dev(env, dev, cfid);
-       if (IS_ERR(child))
-               GOTO(log, rc = PTR_ERR(child));
 
        cname->ln_name = NULL;
        if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
                GOTO(log, rc = 1);
 
-       /* Create .lustre/lost+found/MDTxxxx when needed. */
-       if (unlikely(lfsck->li_lpf_obj == NULL)) {
-               rc = lfsck_create_lpf(env, lfsck);
+       if (dt_object_remote(orphan)) {
+               LASSERT(lfsck->li_lpf_root_obj != NULL);
+
+               idx = lfsck_find_mdt_idx_by_fid(env, lfsck, cfid);
+               if (idx < 0)
+                       GOTO(log, rc = idx);
+
+               snprintf(name, 8, "MDT%04x", idx);
+               rc = dt_lookup(env, lfsck->li_lpf_root_obj,
+                              (struct dt_rec *)&tfid,
+                              (const struct dt_key *)name, BYPASS_CAPA);
                if (rc != 0)
-                       GOTO(log, rc);
+                       GOTO(log, rc = (rc == -ENOENT ? -ENXIO : rc));
+
+               parent = lfsck_object_find_bottom(env, lfsck, &tfid);
+               if (IS_ERR(parent))
+                       GOTO(log, rc = PTR_ERR(parent));
+
+               if (unlikely(!dt_try_as_dir(env, parent)))
+                       GOTO(log, rc = -ENOTDIR);
+       } else {
+               if (unlikely(lfsck->li_lpf_obj == NULL))
+                       GOTO(log, rc = -ENXIO);
+
+               parent = lfsck->li_lpf_obj;
        }
 
-       parent = lfsck->li_lpf_obj;
-       pfid = lfsck_dto2fid(parent);
+       dev = lfsck_find_dev_by_fid(env, lfsck, cfid);
+       if (IS_ERR(dev))
+               GOTO(log, rc = PTR_ERR(dev));
+
+       child = lfsck_object_find_by_dev(env, dev, cfid);
+       if (IS_ERR(child))
+               GOTO(log, rc = PTR_ERR(child));
 
        /* Hold update lock on the parent to prevent others to access. */
        rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
@@ -1304,10 +1321,11 @@ static int lfsck_namespace_create_orphan_local(const struct lu_env *env,
        if (rc != 0)
                GOTO(log, rc);
 
+       idx = 0;
        do {
                namelen = snprintf(name, 31, DFID"-P-%d",
                                   PFID(cfid), idx++);
-               rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
+               rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
                               (const struct dt_key *)name, BYPASS_CAPA);
                if (rc != 0 && rc != -ENOENT)
                        GOTO(unlock1, rc);
@@ -1317,7 +1335,7 @@ static int lfsck_namespace_create_orphan_local(const struct lu_env *env,
        cname->ln_namelen = namelen;
 
        memset(la, 0, sizeof(*la));
-       la->la_mode = type | (S_ISDIR(type) ? 0700 : 0600);
+       la->la_mode = S_IFDIR | 0700;
        la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
                       LA_ATIME | LA_MTIME | LA_CTIME;
 
@@ -1325,13 +1343,13 @@ static int lfsck_namespace_create_orphan_local(const struct lu_env *env,
                                  la->la_mode & S_IFMT);
 
        memset(dof, 0, sizeof(*dof));
-       dof->dof_type = dt_mode_to_dft(type);
+       dof->dof_type = dt_mode_to_dft(S_IFDIR);
 
        rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
        if (rc != 0)
                GOTO(unlock1, rc);
 
-       rc = linkea_add_buf(&ldata, cname, pfid);
+       rc = linkea_add_buf(&ldata, cname, lfsck_dto2fid(parent));
        if (rc != 0)
                GOTO(unlock1, rc);
 
@@ -1339,13 +1357,38 @@ static int lfsck_namespace_create_orphan_local(const struct lu_env *env,
        if (IS_ERR(th))
                GOTO(unlock1, rc = PTR_ERR(th));
 
+       /* Sync the remote transaction to guarantee that the subsequent
+        * lock against the @orphan can find the @orphan in time. */
+       if (dt_object_remote(orphan))
+               th->th_sync = 1;
+
        rc = dt_declare_create(env, child, la, hint, dof, th);
-       if (rc == 0 && S_ISDIR(type))
+       if (rc != 0)
+               GOTO(stop, rc);
+
+       if (unlikely(!dt_try_as_dir(env, child)))
+               GOTO(stop, rc = -ENOTDIR);
+
+       rec->rec_type = S_IFDIR;
+       rec->rec_fid = cfid;
+       rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
+                              (const struct dt_key *)dot, th);
+       if (rc != 0)
+               GOTO(stop, rc);
+
+       rec->rec_fid = lfsck_dto2fid(parent);
+       rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
+                              (const struct dt_key *)dotdot, th);
+       if (rc == 0)
                rc = dt_declare_ref_add(env, child, th);
 
        if (rc != 0)
                GOTO(stop, rc);
 
+       rc = dt_declare_ref_add(env, child, th);
+       if (rc != 0)
+               GOTO(stop, rc);
+
        if (lmv != NULL) {
                lmv->lmv_magic = LMV_MAGIC;
                lmv->lmv_master_mdt_index = lfsck_dev_idx(dev);
@@ -1364,11 +1407,10 @@ static int lfsck_namespace_create_orphan_local(const struct lu_env *env,
        if (rc != 0)
                GOTO(stop, rc);
 
-       rec->rec_type = type;
        rec->rec_fid = cfid;
        rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
                               (const struct dt_key *)name, th);
-       if (rc == 0 && S_ISDIR(type))
+       if (rc == 0)
                rc = dt_declare_ref_add(env, parent, th);
 
        if (rc != 0)
@@ -1383,28 +1425,22 @@ static int lfsck_namespace_create_orphan_local(const struct lu_env *env,
        if (rc != 0)
                GOTO(unlock2, rc);
 
-       if (S_ISDIR(type)) {
-               if (unlikely(!dt_try_as_dir(env, child)))
-                       GOTO(unlock2, rc = -ENOTDIR);
-
-               rec->rec_type = S_IFDIR;
-               rec->rec_fid = cfid;
-               rc = dt_insert(env, child, (const struct dt_rec *)rec,
-                              (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
-               if (rc != 0)
-                       GOTO(unlock2, rc);
+       rec->rec_fid = cfid;
+       rc = dt_insert(env, child, (const struct dt_rec *)rec,
+                      (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
+       if (rc != 0)
+               GOTO(unlock2, rc);
 
-               rec->rec_fid = pfid;
-               rc = dt_insert(env, child, (const struct dt_rec *)rec,
-                              (const struct dt_key *)dotdot, th,
-                              BYPASS_CAPA, 1);
-               if (rc != 0)
-                       GOTO(unlock2, rc);
+       rec->rec_fid = lfsck_dto2fid(parent);
+       rc = dt_insert(env, child, (const struct dt_rec *)rec,
+                      (const struct dt_key *)dotdot, th,
+                      BYPASS_CAPA, 1);
+       if (rc != 0)
+               GOTO(unlock2, rc);
 
-               rc = dt_ref_add(env, child, th);
-               if (rc != 0)
-                       GOTO(unlock2, rc);
-       }
+       rc = dt_ref_add(env, child, th);
+       if (rc != 0)
+               GOTO(unlock2, rc);
 
        if (lmv != NULL) {
                rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV, 0,
@@ -1419,11 +1455,10 @@ static int lfsck_namespace_create_orphan_local(const struct lu_env *env,
        if (rc != 0)
                GOTO(stop, rc);
 
-       rec->rec_type = type;
        rec->rec_fid = cfid;
        rc = dt_insert(env, parent, (const struct dt_rec *)rec,
                       (const struct dt_key *)name, th, BYPASS_CAPA, 1);
-       if (rc == 0 && S_ISDIR(type)) {
+       if (rc == 0) {
                dt_write_lock(env, parent, 0);
                rc = dt_ref_add(env, parent, th);
                dt_write_unlock(env, parent);
@@ -1441,55 +1476,16 @@ unlock1:
        lfsck_ibits_unlock(&lh, LCK_EX);
 
 log:
-       CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan locally for "
-              "the object "DFID", name = %s, type %o: rc = %d\n",
+       CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan dir for "
+              "the object "DFID", name = %s: rc = %d\n",
               lfsck_lfsck2name(lfsck), PFID(cfid),
-              cname->ln_name != NULL ? cname->ln_name : "<NULL>", type, rc);
+              cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
 
        if (child != NULL && !IS_ERR(child))
                lfsck_object_put(env, child);
 
-       return rc;
-}
-
-/**
- * Create the specified orphan MDT-object.
- *
- * For the case that the parent MDT-object stored in some MDT-object's
- * linkEA entry is lost, the LFSCK will re-create the parent object as
- * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
- * with the name: ${FID}-P-${conflict_version}.
- *
- * \param[in] env      pointer to the thread context
- * \param[in] com      pointer to the lfsck component
- * \param[in] orphan   pointer to the orphan MDT-object
- *
- *  type "P":          The orphan object to be created was a parent directory
- *                     of some MDT-object which linkEA shows that the @orphan
- *                     object is missing.
- *
- * \see lfsck_layout_recreate_parent() for more types.
- *
- * \param[in] lmv      pointer to master LMV EA that will be set to the orphan
- *
- * \retval             positive number for repaired cases
- * \retval             0 if needs to repair nothing
- * \retval             negative error number on failure
- */
-static int lfsck_namespace_create_orphan(const struct lu_env *env,
-                                        struct lfsck_component *com,
-                                        struct dt_object *orphan,
-                                        struct lmv_mds_md_v1 *lmv)
-{
-       struct lfsck_namespace *ns = com->lc_file_ram;
-       int                     rc;
-
-       if (dt_object_remote(orphan))
-               rc = lfsck_namespace_create_orphan_remote(env, com, orphan,
-                                                         S_IFDIR, lmv);
-       else
-               rc = lfsck_namespace_create_orphan_local(env, com, orphan,
-                                                        S_IFDIR, lmv);
+       if (parent != NULL && !IS_ERR(parent) && parent != lfsck->li_lpf_obj)
+               lfsck_object_put(env, parent);
 
        if (rc != 0)
                ns->ln_flags |= LF_INCONSISTENT;
@@ -1530,7 +1526,7 @@ static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
        struct lfsck_bookmark           *bk        = &lfsck->li_bookmark_ram;
        struct thandle                  *th        = NULL;
        struct lustre_handle             lh        = { 0 };
-       struct linkea_data               ldata_new = { 0 };
+       struct linkea_data               ldata_new = { NULL };
        struct lu_buf                    linkea_buf;
        int                              rc        = 0;
        ENTRY;
@@ -1739,16 +1735,16 @@ static int lfsck_namespace_replace_cond(const struct lu_env *env,
                                        const struct lu_name *cname)
 {
        struct lfsck_thread_info        *info   = lfsck_env_info(env);
-       struct lu_fid                   *tfid   = &info->lti_fid5;
        struct lu_attr                  *la     = &info->lti_la;
        struct dt_insert_rec            *rec    = &info->lti_dt_rec;
+       struct lu_fid                    tfid;
        struct lfsck_instance           *lfsck  = com->lc_lfsck;
        struct dt_device                *dev    = lfsck->li_next;
        const char                      *name   = cname->ln_name;
        struct dt_object                *obj    = NULL;
        struct lustre_handle             plh    = { 0 };
        struct lustre_handle             clh    = { 0 };
-       struct linkea_data               ldata  = { 0 };
+       struct linkea_data               ldata  = { NULL };
        struct thandle                  *th     = NULL;
        bool                             exist  = true;
        int                              rc     = 0;
@@ -1780,7 +1776,7 @@ static int lfsck_namespace_replace_cond(const struct lu_env *env,
                goto replace;
        }
 
-       rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
+       rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
                       (const struct dt_key *)name, BYPASS_CAPA);
        if (rc == -ENOENT) {
                exist = false;
@@ -1791,7 +1787,7 @@ static int lfsck_namespace_replace_cond(const struct lu_env *env,
                GOTO(log, rc);
 
        /* Someone changed the name entry, cannot replace it. */
-       if (!lu_fid_eq(cfid, tfid))
+       if (!lu_fid_eq(cfid, &tfid))
                GOTO(log, rc = 0);
 
        /* lock the object to be destroyed. */
@@ -1929,8 +1925,6 @@ int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
        int                              rc     = 0;
        ENTRY;
 
-       LASSERT(!dt_object_remote(obj));
-
        th = dt_trans_create(env, dev);
        if (IS_ERR(th))
                GOTO(log, rc = PTR_ERR(th));
@@ -2006,15 +2000,14 @@ int lfsck_namespace_repair_dirent(const struct lu_env *env,
                                  const char *name, const char *name2,
                                  __u16 type, bool update, bool dec)
 {
-       struct lfsck_thread_info        *info   = lfsck_env_info(env);
-       struct dt_insert_rec            *rec    = &info->lti_dt_rec;
-       const struct lu_fid             *cfid   = lfsck_dto2fid(child);
-       struct lu_fid                   *tfid   = &info->lti_fid5;
-       struct lfsck_instance           *lfsck  = com->lc_lfsck;
-       struct dt_device                *dev    = lfsck->li_next;
-       struct thandle                  *th     = NULL;
-       struct lustre_handle             lh     = { 0 };
-       int                              rc     = 0;
+       struct dt_insert_rec    *rec    = &lfsck_env_info(env)->lti_dt_rec;
+       const struct lu_fid     *cfid   = lfsck_dto2fid(child);
+       struct lu_fid            tfid;
+       struct lfsck_instance   *lfsck  = com->lc_lfsck;
+       struct dt_device        *dev    = lfsck->li_next;
+       struct thandle          *th     = NULL;
+       struct lustre_handle     lh     = { 0 };
+       int                      rc     = 0;
        ENTRY;
 
        if (unlikely(!dt_try_as_dir(env, parent)))
@@ -2054,7 +2047,7 @@ int lfsck_namespace_repair_dirent(const struct lu_env *env,
                GOTO(stop, rc);
 
        dt_write_lock(env, parent, 0);
-       rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
+       rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
                       (const struct dt_key *)name, BYPASS_CAPA);
        /* Someone has removed the bad name entry by race. */
        if (rc == -ENOENT)
@@ -2065,7 +2058,7 @@ int lfsck_namespace_repair_dirent(const struct lu_env *env,
 
        /* Someone has removed the bad name entry and reused it for other
         * object by race. */
-       if (!lu_fid_eq(tfid, cfid))
+       if (!lu_fid_eq(&tfid, cfid))
                GOTO(unlock2, rc = 0);
 
        if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
@@ -2156,7 +2149,7 @@ static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
        struct lfsck_instance           *lfsck  = com->lc_lfsck;
        struct dt_device                *dev    = lfsck->li_bottom;
        struct thandle                  *th     = NULL;
-       struct linkea_data               ldata  = { 0 };
+       struct linkea_data               ldata  = { NULL };
        struct lu_buf                    linkea_buf;
        int                              rc     = 0;
        ENTRY;
@@ -2339,7 +2332,7 @@ lfsck_namespace_dsd_single(const struct lu_env *env,
        struct lfsck_thread_info *info          = lfsck_env_info(env);
        struct lu_name           *cname         = &info->lti_name;
        const struct lu_fid      *cfid          = lfsck_dto2fid(child);
-       struct lu_fid            *tfid          = &info->lti_fid3;
+       struct lu_fid             tfid;
        struct lfsck_namespace   *ns            = com->lc_file_ram;
        struct lfsck_instance    *lfsck         = com->lc_lfsck;
        struct dt_object         *parent        = NULL;
@@ -2347,9 +2340,9 @@ lfsck_namespace_dsd_single(const struct lu_env *env,
        int                       rc            = 0;
        ENTRY;
 
-       lfsck_namespace_unpack_linkea_entry(ldata, cname, tfid, info->lti_key);
+       lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid, info->lti_key);
        /* The unique linkEA entry with bad parent will be handled as orphan. */
-       if (!fid_is_sane(tfid)) {
+       if (!fid_is_sane(&tfid)) {
                if (!lustre_handle_is_used(lh) && retry != NULL)
                        *retry = true;
                else
@@ -2359,7 +2352,7 @@ lfsck_namespace_dsd_single(const struct lu_env *env,
                GOTO(out, rc);
        }
 
-       parent = lfsck_object_find_bottom(env, lfsck, tfid);
+       parent = lfsck_object_find_bottom(env, lfsck, &tfid);
        if (IS_ERR(parent))
                GOTO(out, rc = PTR_ERR(parent));
 
@@ -2401,7 +2394,7 @@ lost_parent:
                        /* It is an invalid name entry, we
                         * cannot trust the parent also. */
                        rc = lfsck_namespace_shrink_linkea(env, com, child,
-                                               ldata, cname, tfid, true);
+                                               ldata, cname, &tfid, true);
                        if (rc < 0)
                                GOTO(out, rc);
 
@@ -2414,7 +2407,7 @@ lost_parent:
                }
 
                /* Create the lost parent as an orphan. */
-               rc = lfsck_namespace_create_orphan(env, com, parent, lmv);
+               rc = lfsck_namespace_create_orphan_dir(env, com, parent, lmv);
                if (rc >= 0) {
                        /* Add the missing name entry to the parent. */
                        rc = lfsck_namespace_insert_normal(env, com, parent,
@@ -2430,7 +2423,7 @@ lost_parent:
                                 * current system to be consistent. */
                                rc = lfsck_namespace_shrink_linkea(env,
                                                com, child, ldata,
-                                               cname, tfid, true);
+                                               cname, &tfid, true);
                                if (rc >= 0) {
                                        snprintf(info->lti_tmpbuf,
                                                 sizeof(info->lti_tmpbuf),
@@ -2456,7 +2449,7 @@ lost_parent:
                GOTO(out, rc);
        }
 
-       rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
+       rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
                       (const struct dt_key *)cname->ln_name, BYPASS_CAPA);
        if (rc == -ENOENT) {
                /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
@@ -2486,7 +2479,7 @@ lost_parent:
                /* It is an invalid name entry, drop it. */
                if (unlikely(rc > 0)) {
                        rc = lfsck_namespace_shrink_linkea(env, com, child,
-                                               ldata, cname, tfid, true);
+                                               ldata, cname, &tfid, true);
                        if (rc >= 0) {
                                snprintf(info->lti_tmpbuf,
                                         sizeof(info->lti_tmpbuf),
@@ -2518,7 +2511,7 @@ lost_parent:
                         * internal status of create operation. Under such
                         * case, nothing to be done. */
                        rc = lfsck_namespace_shrink_linkea_cond(env, com,
-                                       parent, child, ldata, cname, tfid);
+                                       parent, child, ldata, cname, &tfid);
                        if (rc >= 0) {
                                snprintf(info->lti_tmpbuf,
                                         sizeof(info->lti_tmpbuf),
@@ -2534,7 +2527,7 @@ lost_parent:
        if (rc != 0)
                GOTO(out, rc);
 
-       if (!lu_fid_eq(tfid, cfid)) {
+       if (!lu_fid_eq(&tfid, cfid)) {
                if (!lustre_handle_is_used(lh) && retry != NULL) {
                        *retry = true;
 
@@ -2546,7 +2539,7 @@ lost_parent:
                 * may be created by the LFSCK for repairing dangling
                 * name entry. Try to replace it. */
                rc = lfsck_namespace_replace_cond(env, com, parent, child,
-                                                 tfid, cname);
+                                                 &tfid, cname);
                if (rc == 0)
                        rc = lfsck_namespace_dsd_orphan(env, com, child,
                                                        pfid, lh, type);
@@ -2554,6 +2547,9 @@ lost_parent:
                GOTO(out, rc);
        }
 
+       if (fid_is_zero(pfid))
+               GOTO(out, rc = 0);
+
        /* The ".." name entry is wrong, update it. */
        if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
                if (!lustre_handle_is_used(lh) && retry != NULL) {
@@ -2612,27 +2608,29 @@ lfsck_namespace_dsd_multiple(const struct lu_env *env,
        struct lfsck_thread_info *info          = lfsck_env_info(env);
        struct lu_name           *cname         = &info->lti_name;
        const struct lu_fid      *cfid          = lfsck_dto2fid(child);
-       struct lu_fid            *tfid          = &info->lti_fid3;
-       struct lu_fid            *pfid2         = &info->lti_fid4;
+       struct lu_fid            *pfid2         = &info->lti_fid3;
+       struct lu_fid             tfid;
        struct lfsck_namespace   *ns            = com->lc_file_ram;
        struct lfsck_instance    *lfsck         = com->lc_lfsck;
        struct lfsck_bookmark    *bk            = &lfsck->li_bookmark_ram;
        struct dt_object         *parent        = NULL;
-       struct linkea_data        ldata_new     = { 0 };
-       int                       count         = 0;
+       struct linkea_data        ldata_new     = { NULL };
+       int                       dirent_count  = 0;
+       int                       linkea_count  = 0;
        int                       rc            = 0;
        bool                      once          = true;
        ENTRY;
 
 again:
        while (ldata->ld_lee != NULL) {
-               lfsck_namespace_unpack_linkea_entry(ldata, cname, tfid,
+               lfsck_namespace_unpack_linkea_entry(ldata, cname, &tfid,
                                                    info->lti_key);
                /* Drop repeated linkEA entries. */
-               lfsck_namespace_filter_linkea_entry(ldata, cname, tfid, true);
+               lfsck_namespace_filter_linkea_entry(ldata, cname, &tfid, true);
                /* Drop invalid linkEA entry. */
-               if (!fid_is_sane(tfid)) {
+               if (!fid_is_sane(&tfid)) {
                        linkea_del_buf(ldata, cname);
+                       linkea_count++;
                        continue;
                }
 
@@ -2649,12 +2647,12 @@ again:
                 * When the LFSCK runs again, if the dangling name is still
                 * there, the LFSCK should move the orphan directory object
                 * back to the normal namespace. */
-               if (!lpf && !lu_fid_eq(pfid, tfid) && once) {
+               if (!lpf && !lu_fid_eq(pfid, &tfid) && once) {
                        linkea_next_entry(ldata);
                        continue;
                }
 
-               parent = lfsck_object_find_bottom(env, lfsck, tfid);
+               parent = lfsck_object_find_bottom(env, lfsck, &tfid);
                if (IS_ERR(parent))
                        RETURN(PTR_ERR(parent));
 
@@ -2666,6 +2664,7 @@ again:
                                 * child to be visible via other parent, then
                                 * remove this linkEA entry. */
                                linkea_del_buf(ldata, cname);
+                               linkea_count++;
                                continue;
                        }
 
@@ -2676,10 +2675,11 @@ again:
                if (unlikely(!dt_try_as_dir(env, parent))) {
                        lfsck_object_put(env, parent);
                        linkea_del_buf(ldata, cname);
+                       linkea_count++;
                        continue;
                }
 
-               rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
+               rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
                               (const struct dt_key *)cname->ln_name,
                               BYPASS_CAPA);
                *pfid2 = *lfsck_dto2fid(parent);
@@ -2695,7 +2695,7 @@ again:
                        RETURN(rc);
                }
 
-               if (lu_fid_eq(tfid, cfid)) {
+               if (lu_fid_eq(&tfid, cfid)) {
                        lfsck_object_put(env, parent);
                        if (!lu_fid_eq(pfid, pfid2)) {
                                *type = LNIT_UNMATCHED_PAIRS;
@@ -2723,17 +2723,18 @@ rebuild:
                                RETURN(rc);
 
                        linkea_del_buf(ldata, cname);
+                       linkea_count++;
                        linkea_first_entry(ldata);
                        /* There may be some invalid dangling name entries under
                         * other parent directories, remove all of them. */
                        while (ldata->ld_lee != NULL) {
                                lfsck_namespace_unpack_linkea_entry(ldata,
-                                               cname, tfid, info->lti_key);
-                               if (!fid_is_sane(tfid))
+                                               cname, &tfid, info->lti_key);
+                               if (!fid_is_sane(&tfid))
                                        goto next;
 
                                parent = lfsck_object_find_bottom(env, lfsck,
-                                                                 tfid);
+                                                                 &tfid);
                                if (IS_ERR(parent)) {
                                        rc = PTR_ERR(parent);
                                        if (rc != -ENOENT &&
@@ -2759,13 +2760,13 @@ rebuild:
                                        goto next;
                                }
 
-                               count += rc;
+                               dirent_count += rc;
 
 next:
                                linkea_del_buf(ldata, cname);
                        }
 
-                       ns->ln_dirent_repaired += count;
+                       ns->ln_dirent_repaired += dirent_count;
 
                        RETURN(rc);
                }
@@ -2775,7 +2776,7 @@ next:
                 * created by the LFSCK for repairing dangling name entry.
                 * Try to replace it. */
                rc = lfsck_namespace_replace_cond(env, com, parent, child,
-                                                 tfid, cname);
+                                                 &tfid, cname);
                lfsck_object_put(env, parent);
                if (rc < 0)
                        RETURN(rc);
@@ -2786,10 +2787,15 @@ next:
                linkea_del_buf(ldata, cname);
        }
 
+       linkea_first_entry(ldata);
        if (ldata->ld_leh->leh_reccount == 1) {
                rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
                                                lh, type, NULL);
 
+               if (rc == 0 && fid_is_zero(pfid) && linkea_count > 0)
+                       rc = lfsck_namespace_rebuild_linkea(env, com, child,
+                                                           ldata);
+
                RETURN(rc);
        }
 
@@ -2802,7 +2808,6 @@ next:
                RETURN(rc);
        }
 
-       linkea_first_entry(ldata);
        /* If the dangling name entry for the orphan directory object has
         * been remvoed, then just check whether the directory object is
         * still under the .lustre/lost+found/MDTxxxx/ or not. */
@@ -2834,7 +2839,7 @@ next:
  * \param[in] env      pointer to the thread context
  * \param[in] com      pointer to the lfsck component
  * \param[in] obj      pointer to the dt_object to be handled
- * \param[in,out] nlink        pointer to buffer to object's hard lock count before
+ * \param[in,out] la   pointer to buffer to object's attribute before
  *                     and after the repairing
  *
  * \retval             positive number for repaired cases
@@ -2843,10 +2848,10 @@ next:
  */
 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
                                        struct lfsck_component *com,
-                                       struct dt_object *obj, __u32 *nlink)
+                                       struct dt_object *obj,
+                                       struct lu_attr *la)
 {
        struct lfsck_thread_info        *info   = lfsck_env_info(env);
-       struct lu_attr                  *la     = &info->lti_la3;
        struct lu_fid                   *tfid   = &info->lti_fid3;
        struct lfsck_namespace          *ns     = com->lc_file_ram;
        struct lfsck_instance           *lfsck  = com->lc_lfsck;
@@ -2854,9 +2859,10 @@ static int lfsck_namespace_repair_nlink(const struct lu_env *env,
        const struct lu_fid             *cfid   = lfsck_dto2fid(obj);
        struct dt_object                *child  = NULL;
        struct thandle                  *th     = NULL;
-       struct linkea_data               ldata  = { 0 };
+       struct linkea_data               ldata  = { NULL };
        struct lustre_handle             lh     = { 0 };
-       __u32                            old    = *nlink;
+       __u32                            old    = la->la_nlink;
+       int                              idx;
        int                              rc     = 0;
        __u8                             flags;
        ENTRY;
@@ -2899,25 +2905,29 @@ static int lfsck_namespace_repair_nlink(const struct lu_env *env,
                GOTO(unlock, rc = 0);
 
        fid_cpu_to_be(tfid, cfid);
-       rc = dt_lookup(env, com->lc_obj, (struct dt_rec *)&flags,
-                      (const struct dt_key *)tfid, BYPASS_CAPA);
+       idx = lfsck_sub_trace_file_fid2idx(cfid);
+       rc = dt_lookup(env, com->lc_sub_trace_objs[idx].lsto_obj,
+                      (struct dt_rec *)&flags, (const struct dt_key *)tfid,
+                      BYPASS_CAPA);
        if (rc != 0)
                GOTO(unlock, rc);
 
        if (flags & LNTF_SKIP_NLINK)
                GOTO(unlock, rc = 0);
 
-       rc = lfsck_links_read2(env, child, &ldata);
-       if (rc == -ENODATA)
-               GOTO(unlock, rc = 0);
+       rc = dt_attr_get(env, child, la, BYPASS_CAPA);
+       if (rc != 0)
+               GOTO(unlock, rc = (rc == -ENOENT ? 0 : rc));
 
+       rc = lfsck_links_read2(env, child, &ldata);
        if (rc != 0)
-               GOTO(unlock, rc);
+               GOTO(unlock, rc = (rc == -ENODATA ? 0 : rc));
 
-       if (*nlink == ldata.ld_leh->leh_reccount)
+       if (la->la_nlink == ldata.ld_leh->leh_reccount ||
+           unlikely(la->la_nlink == 0))
                GOTO(unlock, rc = 0);
 
-       la->la_nlink = *nlink = ldata.ld_leh->leh_reccount;
+       la->la_nlink = ldata.ld_leh->leh_reccount;
        if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
                GOTO(unlock, rc = 1);
 
@@ -2938,7 +2948,7 @@ log:
 
        CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
               "nlink count from %u to %u: rc = %d\n",
-              lfsck_lfsck2name(lfsck), PFID(cfid), old, *nlink, rc);
+              lfsck_lfsck2name(lfsck), PFID(cfid), old, la->la_nlink, rc);
 
        if (rc != 0)
                ns->ln_flags |= LF_INCONSISTENT;
@@ -2993,7 +3003,7 @@ static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
        struct lfsck_namespace   *ns            = com->lc_file_ram;
        struct lfsck_instance    *lfsck         = com->lc_lfsck;
        struct lustre_handle      lh            = { 0 };
-       struct linkea_data        ldata         = { 0 };
+       struct linkea_data        ldata         = { NULL };
        bool                      unknown       = false;
        bool                      lpf           = false;
        bool                      retry         = false;
@@ -3071,6 +3081,8 @@ lock:
        } else if (lfsck->li_lpf_obj != NULL &&
                   lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
                lpf = true;
+       } else if (unlikely(!fid_is_sane(pfid))) {
+               fid_zero(pfid);
        }
 
        rc = lfsck_links_read(env, child, &ldata);
@@ -3223,7 +3235,7 @@ static int lfsck_namespace_double_scan_one(const struct lu_env *env,
        struct lfsck_instance    *lfsck    = com->lc_lfsck;
        struct lfsck_namespace   *ns       = com->lc_file_ram;
        struct dt_object         *parent   = NULL;
-       struct linkea_data        ldata    = { 0 };
+       struct linkea_data        ldata    = { NULL };
        bool                      repaired = false;
        int                       count    = 0;
        int                       rc;
@@ -3314,7 +3326,7 @@ lost_parent:
                                }
 
                                /* Create the lost parent as an orphan. */
-                               rc = lfsck_namespace_create_orphan(env, com,
+                               rc = lfsck_namespace_create_orphan_dir(env, com,
                                                                parent, NULL);
                                if (rc < 0) {
                                        lfsck_object_put(env, parent);
@@ -3514,37 +3526,38 @@ out:
                count = ldata.ld_leh->leh_reccount;
        }
 
-       /* If the LFSCK is marked as LF_INCOMPLETE, then means some
-        * MDT has ever tried to verify some remote MDT-object that
-        * resides on this MDT, but this MDT failed to respond such
-        * request. So means there may be some remote name entry on
-        * other MDT that references this object with another name,
-        * so we cannot know whether this linkEA is valid or not.
-        * So keep it there and maybe resolved when next LFSCK run. */
-       if (count == 0 && !(ns->ln_flags & LF_INCOMPLETE)) {
-               /* If the child becomes orphan, then insert it into
-                * the global .lustre/lost+found/MDTxxxx directory. */
-               rc = lfsck_namespace_insert_orphan(env, com, child, "", "O",
-                                                  &count);
-               if (rc < 0)
-                       return rc;
+       if (count == 0) {
+               /* If the LFSCK is marked as LF_INCOMPLETE, then means some
+                * MDT has ever tried to verify some remote MDT-object that
+                * resides on this MDT, but this MDT failed to respond such
+                * request. So means there may be some remote name entry on
+                * other MDT that references this object with another name,
+                * so we cannot know whether this linkEA is valid or not.
+                * So keep it there and maybe resolved when next LFSCK run. */
+               if (!(ns->ln_flags & LF_INCOMPLETE)) {
+                       /* If the child becomes orphan, then insert it into
+                        * the global .lustre/lost+found/MDTxxxx directory. */
+                       rc = lfsck_namespace_insert_orphan(env, com, child,
+                                                          "", "O", &count);
+                       if (rc < 0)
+                               return rc;
 
-               if (rc > 0) {
-                       ns->ln_mul_ref_repaired++;
-                       repaired = true;
+                       if (rc > 0) {
+                               ns->ln_mul_ref_repaired++;
+                               repaired = true;
+                       }
                }
-       }
-
-       rc = dt_attr_get(env, child, la, BYPASS_CAPA);
-       if (rc != 0)
-               return rc;
+       } else {
+               rc = dt_attr_get(env, child, la, BYPASS_CAPA);
+               if (rc != 0)
+                       return rc;
 
-       if (la->la_nlink != count) {
-               rc = lfsck_namespace_repair_nlink(env, com, child,
-                                                 &la->la_nlink);
-               if (rc > 0) {
-                       ns->ln_objs_nlink_repaired++;
-                       rc = 0;
+               if (la->la_nlink != 0 && la->la_nlink != count) {
+                       rc = lfsck_namespace_repair_nlink(env, com, child, la);
+                       if (rc > 0) {
+                               ns->ln_objs_nlink_repaired++;
+                               rc = 0;
+                       }
                }
        }
 
@@ -3661,6 +3674,33 @@ static void lfsck_namespace_release_lmv(const struct lu_env *env,
        }
 }
 
+static int lfsck_namespace_check_for_double_scan(const struct lu_env *env,
+                                                struct lfsck_component *com,
+                                                struct dt_object *obj)
+{
+       struct lu_attr *la = &lfsck_env_info(env)->lti_la;
+       int             rc;
+
+       rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
+       if (rc != 0)
+               return rc;
+
+       /* zero-linkEA object may be orphan, but it also maybe because
+        * of upgrading. Currently, we cannot record it for double scan.
+        * Because it may cause the LFSCK trace file to be too large. */
+
+       /* "la_ctime" == 1 means that it has ever been removed from
+        * backend /lost+found directory but not been added back to
+        * the normal namespace yet. */
+
+       if ((S_ISREG(lfsck_object_type(obj)) && la->la_nlink > 1) ||
+           unlikely(la->la_ctime == 1))
+               rc = lfsck_namespace_trace_update(env, com, lfsck_dto2fid(obj),
+                                                 LNTF_CHECK_LINKEA, true);
+
+       return rc;
+}
+
 /* namespace APIs */
 
 static int lfsck_namespace_reset(const struct lu_env *env,
@@ -3695,29 +3735,22 @@ static int lfsck_namespace_reset(const struct lu_env *env,
        ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
        ns->ln_status = LS_INIT;
 
-       rc = local_object_unlink(env, lfsck->li_bottom, root,
-                                lfsck_namespace_name);
-       if (rc != 0)
-               GOTO(out, rc);
-
        lfsck_object_put(env, com->lc_obj);
        com->lc_obj = NULL;
-       dto = local_index_find_or_create(env, lfsck->li_los, root,
-                                        lfsck_namespace_name,
-                                        S_IFREG | S_IRUGO | S_IWUSR,
-                                        &dt_lfsck_features);
+       dto = lfsck_namespace_load_one_trace_file(env, com, root,
+                               LFSCK_NAMESPACE, NULL, true);
        if (IS_ERR(dto))
                GOTO(out, rc = PTR_ERR(dto));
 
        com->lc_obj = dto;
-       rc = dto->do_ops->do_index_try(env, dto, &dt_lfsck_features);
+       rc = lfsck_namespace_load_sub_trace_files(env, com, true);
        if (rc != 0)
                GOTO(out, rc);
 
        lad->lad_incomplete = 0;
        CFS_RESET_BITMAP(lad->lad_bitmap);
 
-       rc = lfsck_namespace_store(env, com);
+       rc = lfsck_namespace_store(env, com, true);
 
        GOTO(out, rc);
 
@@ -3854,7 +3887,7 @@ static int lfsck_namespace_checkpoint(const struct lu_env *env,
                com->lc_new_checked = 0;
        }
 
-       rc = lfsck_namespace_store(env, com);
+       rc = lfsck_namespace_store(env, com, false);
        up_write(&com->lc_sem);
 
 log:
@@ -3976,14 +4009,13 @@ static int lfsck_namespace_exec_oit(const struct lu_env *env,
        struct lfsck_namespace   *ns    = com->lc_file_ram;
        struct lfsck_instance    *lfsck = com->lc_lfsck;
        const struct lu_fid      *fid   = lfsck_dto2fid(obj);
-       struct lu_attr           *la    = &info->lti_la;
        struct lu_fid            *pfid  = &info->lti_fid2;
        struct lu_name           *cname = &info->lti_name;
        struct lu_seq_range      *range = &info->lti_range;
        struct dt_device         *dev   = lfsck->li_bottom;
        struct seq_server_site   *ss    =
                                lu_site2seq(dev->dd_lu_dev.ld_site);
-       struct linkea_data        ldata = { 0 };
+       struct linkea_data        ldata = { NULL };
        __u32                     idx   = lfsck_dev_idx(dev);
        int                       rc;
        ENTRY;
@@ -4012,23 +4044,8 @@ static int lfsck_namespace_exec_oit(const struct lu_env *env,
                GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
        }
 
-       /* zero-linkEA object may be orphan, but it also maybe because
-        * of upgrading. Currently, we cannot record it for double scan.
-        * Because it may cause the LFSCK trace file to be too large. */
        if (rc == -ENODATA) {
-               if (S_ISDIR(lfsck_object_type(obj)))
-                       GOTO(out, rc = 0);
-
-               rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
-               if (rc != 0)
-                       GOTO(out, rc);
-
-               /* "la_ctime" == 1 means that it has ever been removed from
-                * backend /lost+found directory but not been added back to
-                * the normal namespace yet. */
-               if (la->la_nlink > 1 || unlikely(la->la_ctime == 1))
-                       rc = lfsck_namespace_trace_update(env, com, fid,
-                                               LNTF_CHECK_LINKEA, true);
+               rc = lfsck_namespace_check_for_double_scan(env, com, obj);
 
                GOTO(out, rc);
        }
@@ -4054,24 +4071,12 @@ static int lfsck_namespace_exec_oit(const struct lu_env *env,
                rc = fld_local_lookup(env, ss->ss_server_fld,
                                      fid_seq(pfid), range);
                if ((rc == -ENOENT) ||
-                   (rc == 0 && range->lsr_index != idx)) {
+                   (rc == 0 && range->lsr_index != idx))
                        rc = lfsck_namespace_trace_update(env, com, fid,
                                                LNTF_CHECK_LINKEA, true);
-               } else {
-                       if (S_ISDIR(lfsck_object_type(obj)))
-                               GOTO(out, rc = 0);
-
-                       rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
-                       if (rc != 0)
-                               GOTO(out, rc);
-
-                       /* "la_ctime" == 1 means that it has ever been
-                        * removed from backend /lost+found directory but
-                        * not been added back to the normal namespace yet. */
-                       if (la->la_nlink > 1 || unlikely(la->la_ctime == 1))
-                               rc = lfsck_namespace_trace_update(env, com,
-                                               fid, LNTF_CHECK_LINKEA, true);
-               }
+               else
+                       rc = lfsck_namespace_check_for_double_scan(env, com,
+                                                                  obj);
        }
 
        GOTO(out, rc);
@@ -4092,9 +4097,28 @@ static int lfsck_namespace_exec_dir(const struct lu_env *env,
                                    struct lfsck_component *com,
                                    struct lu_dirent *ent, __u16 type)
 {
-       struct lfsck_assistant_data     *lad    = com->lc_data;
+       struct lfsck_assistant_data     *lad     = com->lc_data;
+       struct lfsck_instance           *lfsck   = com->lc_lfsck;
        struct lfsck_namespace_req      *lnr;
-       bool                             wakeup = false;
+       struct lfsck_bookmark           *bk      = &lfsck->li_bookmark_ram;
+       struct ptlrpc_thread            *mthread = &lfsck->li_thread;
+       struct ptlrpc_thread            *athread = &lad->lad_thread;
+       struct l_wait_info               lwi     = { 0 };
+       bool                             wakeup  = false;
+
+       l_wait_event(mthread->t_ctl_waitq,
+                    bk->lb_async_windows == 0 ||
+                    lad->lad_prefetched < bk->lb_async_windows ||
+                    !thread_is_running(mthread) ||
+                    thread_is_stopped(athread),
+                    &lwi);
+
+       if (unlikely(!thread_is_running(mthread)) ||
+                    thread_is_stopped(athread))
+               return 0;
+
+       if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
+               return 0;
 
        lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, ent, type);
        if (IS_ERR(lnr)) {
@@ -4174,7 +4198,7 @@ static int lfsck_namespace_post(const struct lu_env *env,
                com->lc_new_checked = 0;
        }
 
-       rc = lfsck_namespace_store(env, com);
+       rc = lfsck_namespace_store(env, com, false);
        up_write(&com->lc_sem);
 
        CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
@@ -4246,8 +4270,8 @@ lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
                                          lfsck->li_time_last_checkpoint;
                __u64 checked = ns->ln_items_checked + com->lc_new_checked;
                __u64 speed = checked;
-               __u64 new_checked = msecs_to_jiffies(com->lc_new_checked *
-                                                    MSEC_PER_SEC);
+               __u64 new_checked = com->lc_new_checked *
+                                   msecs_to_jiffies(MSEC_PER_SEC);
                __u32 rtime = ns->ln_run_time_phase1 +
                              cfs_duration_sec(duration + HALF_SEC);
 
@@ -4301,8 +4325,8 @@ lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
                                com->lc_new_checked;
                __u64 speed1 = ns->ln_items_checked;
                __u64 speed2 = checked;
-               __u64 new_checked = msecs_to_jiffies(com->lc_new_checked *
-                                                    MSEC_PER_SEC);
+               __u64 new_checked = com->lc_new_checked *
+                                   msecs_to_jiffies(MSEC_PER_SEC);
                __u32 rtime = ns->ln_run_time_phase2 +
                              cfs_duration_sec(duration + HALF_SEC);
 
@@ -4459,81 +4483,51 @@ static int lfsck_namespace_in_notify(const struct lu_env *env,
        ENTRY;
 
        switch (lr->lr_event) {
-       case LE_CREATE_ORPHAN: {
-               struct dt_object *orphan = NULL;
-               struct lmv_mds_md_v1 *lmv;
-
-               CDEBUG(D_LFSCK, "%s: namespace LFSCK handling notify from "
-                      "MDT %x to create orphan"DFID" with type %o\n",
-                      lfsck_lfsck2name(lfsck), lr->lr_index,
-                      PFID(&lr->lr_fid), lr->lr_type);
-
-               orphan = lfsck_object_find(env, lfsck, &lr->lr_fid);
-               if (IS_ERR(orphan))
-                       GOTO(out_create, rc = PTR_ERR(orphan));
-
-               if (dt_object_exists(orphan))
-                       GOTO(out_create, rc = -EEXIST);
-
-               if (lr->lr_stripe_count > 0) {
-                       lmv = &lfsck_env_info(env)->lti_lmv;
-                       memset(lmv, 0, sizeof(*lmv));
-                       lmv->lmv_hash_type = lr->lr_hash_type;
-                       lmv->lmv_stripe_count = lr->lr_stripe_count;
-                       lmv->lmv_layout_version = lr->lr_layout_version;
-                       memcpy(lmv->lmv_pool_name, lr->lr_pool_name,
-                              sizeof(lmv->lmv_pool_name));
-               } else {
-                       lmv = NULL;
-               }
-
-               rc = lfsck_namespace_create_orphan_local(env, com, orphan,
-                                                        lr->lr_type, lmv);
-
-               GOTO(out_create, rc = (rc == 1) ? 0 : rc);
-
-out_create:
-               CDEBUG(D_LFSCK, "%s: namespace LFSCK handled notify from "
-                      "MDT %x to create orphan"DFID" with type %o: rc = %d\n",
-                      lfsck_lfsck2name(lfsck), lr->lr_index,
-                      PFID(&lr->lr_fid), lr->lr_type, rc);
-
-               if (orphan != NULL && !IS_ERR(orphan))
-                       lfsck_object_put(env, orphan);
-
-               return rc;
-       }
        case LE_SKIP_NLINK_DECLARE: {
-               struct dt_object        *obj   = com->lc_obj;
+               struct dt_object        *obj;
                struct lu_fid           *key   = &lfsck_env_info(env)->lti_fid3;
+               int                      idx;
                __u8                     flags = 0;
 
                LASSERT(th != NULL);
 
+               idx = lfsck_sub_trace_file_fid2idx(&lr->lr_fid);
+               obj = com->lc_sub_trace_objs[idx].lsto_obj;
+               fid_cpu_to_be(key, &lr->lr_fid);
+               mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
                rc = dt_declare_delete(env, obj,
                                       (const struct dt_key *)key, th);
                if (rc == 0)
                        rc = dt_declare_insert(env, obj,
                                               (const struct dt_rec *)&flags,
                                               (const struct dt_key *)key, th);
+               mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
 
                RETURN(rc);
        }
        case LE_SKIP_NLINK: {
-               struct dt_object        *obj   = com->lc_obj;
+               struct dt_object        *obj;
                struct lu_fid           *key   = &lfsck_env_info(env)->lti_fid3;
+               int                      idx;
                __u8                     flags = 0;
                bool                     exist = false;
                ENTRY;
 
                LASSERT(th != NULL);
 
+               idx = lfsck_sub_trace_file_fid2idx(&lr->lr_fid);
+               obj = com->lc_sub_trace_objs[idx].lsto_obj;
                fid_cpu_to_be(key, &lr->lr_fid);
+               mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
                rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
                               (const struct dt_key *)key, BYPASS_CAPA);
                if (rc == 0) {
-                       if (flags & LNTF_SKIP_NLINK)
+                       if (flags & LNTF_SKIP_NLINK) {
+                               mutex_unlock(
+                               &com->lc_sub_trace_objs[idx].lsto_mutex);
+
                                RETURN(0);
+                       }
 
                        exist = true;
                } else if (rc != -ENOENT) {
@@ -4554,6 +4548,7 @@ out_create:
                GOTO(log, rc);
 
 log:
+               mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
                CDEBUG(D_LFSCK, "%s: RPC service thread mark the "DFID
                       " to be skipped for namespace double scan: rc = %d\n",
                       lfsck_lfsck2name(com->lc_lfsck), PFID(&lr->lr_fid), rc);
@@ -4730,7 +4725,7 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env,
        struct lmv_mds_md_v1            *lmv2   = &info->lti_lmv2;
        struct dt_object                *parent = lnr->lnr_obj;
        const struct lu_name            *cname;
-       struct linkea_data               ldata  = { 0 };
+       struct linkea_data               ldata  = { NULL };
        struct lustre_handle             lh     = { 0 };
        struct lu_buf                    linkea_buf;
        struct lu_buf                    lmv_buf;
@@ -4937,7 +4932,7 @@ static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
        struct lfsck_instance      *lfsck    = com->lc_lfsck;
        struct lfsck_bookmark      *bk       = &lfsck->li_bookmark_ram;
        struct lfsck_namespace     *ns       = com->lc_file_ram;
-       struct linkea_data          ldata    = { 0 };
+       struct linkea_data          ldata    = { NULL };
        const struct lu_name       *cname;
        struct thandle             *handle   = NULL;
        struct lfsck_namespace_req *lnr      =
@@ -4955,10 +4950,11 @@ static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
        bool                        bad_hash = false;
        int                         idx      = 0;
        int                         count    = 0;
-       int                         rc;
+       int                         rc       = 0;
        enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
        ENTRY;
 
+       la->la_nlink = 0;
        if (lnr->lnr_attr & LUDA_UPGRADE) {
                ns->ln_flags |= LF_UPGRADE;
                ns->ln_dirent_repaired++;
@@ -4979,6 +4975,24 @@ static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
                GOTO(out, rc);
        }
 
+       if (unlikely(!fid_is_sane(&lnr->lnr_fid))) {
+               CDEBUG(D_LFSCK, "%s: dir scan find invalid FID "DFID
+                      " for the name entry %.*s under "DFID"\n",
+                      lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
+                      lnr->lnr_namelen, lnr->lnr_name, PFID(pfid));
+
+               if (strcmp(lnr->lnr_name, dotdot) != 0)
+                       /* invalid FID means bad name entry, remove it. */
+                       type = LNIT_BAD_DIRENT;
+               else
+                       /* If the parent FID is invalid, we cannot remove
+                        * the ".." entry directly. */
+                       rc = lfsck_namespace_trace_update(env, com, pfid,
+                                               LNTF_CHECK_PARENT, true);
+
+               GOTO(out, rc);
+       }
+
        if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
                rc = lfsck_namespace_striped_dir_rescan(env, com, lnr);
 
@@ -5112,7 +5126,7 @@ again:
                                type = LNIT_BAD_TYPE;
                        }
 
-                       goto record;
+                       goto stop;
                }
 
                ns->ln_flags |= LF_INCONSISTENT;
@@ -5174,7 +5188,7 @@ nodata:
                        ns->ln_linkea_repaired++;
                        repaired = true;
                        log = true;
-                       goto record;
+                       goto stop;
                }
 
                if (!lustre_handle_is_used(&lh))
@@ -5242,35 +5256,6 @@ nodata:
                GOTO(stop, rc);
        }
 
-record:
-       LASSERT(count > 0);
-
-       rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
-       if (rc != 0)
-               GOTO(stop, rc);
-
-       if ((count == 1 && la->la_nlink == 1) ||
-           S_ISDIR(lfsck_object_type(obj)))
-               /* Usually, it is for single linked object or dir, do nothing.*/
-               GOTO(stop, rc);
-
-       /* Following modification will be in another transaction.  */
-       if (handle != NULL) {
-               dt_write_unlock(env, obj);
-               dtlocked = false;
-
-               dt_trans_stop(env, dev, handle);
-               handle = NULL;
-
-               lfsck_ibits_unlock(&lh, LCK_EX);
-       }
-
-       ns->ln_mul_linked_checked++;
-       rc = lfsck_namespace_trace_update(env, com, &lnr->lnr_fid,
-                                         LNTF_CHECK_LINKEA, true);
-
-       GOTO(out, rc);
-
 stop:
        if (dtlocked)
                dt_write_unlock(env, obj);
@@ -5319,6 +5304,9 @@ out:
                default:
                        break;
                }
+
+               if (count == 1 && S_ISREG(lfsck_object_type(obj)))
+                       dt_attr_get(env, obj, la, BYPASS_CAPA);
        }
 
        down_write(&com->lc_sem);
@@ -5385,9 +5373,12 @@ out:
                                               false);
                }
 
-
                rc = 0;
        }
+
+       if (count > 1 || la->la_nlink > 1)
+               ns->ln_mul_linked_checked++;
+
        up_write(&com->lc_sem);
 
        if (obj != NULL && !IS_ERR(obj))
@@ -5432,10 +5423,11 @@ static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
        struct lu_fid                   *key    = &info->lti_fid;
        struct lu_attr                  *la     = &info->lti_la;
        struct lfsck_instance           *lfsck  = com->lc_lfsck;
-       struct dt_object                *obj    = com->lc_obj;
+       struct dt_object                *obj;
        struct dt_device                *dev    = lfsck->li_bottom;
        struct dt_object                *child  = NULL;
        struct thandle                  *th     = NULL;
+       int                              idx;
        int                              rc     = 0;
        __u8                             flags  = 0;
        bool                             exist  = false;
@@ -5448,6 +5440,8 @@ static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
        LASSERT(dt_object_exists(child));
        LASSERT(!dt_object_remote(child));
 
+       idx = lfsck_sub_trace_file_fid2idx(&ent->lde_fid);
+       obj = com->lc_sub_trace_objs[idx].lsto_obj;
        fid_cpu_to_be(key, &ent->lde_fid);
        rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
                       (const struct dt_key *)key, BYPASS_CAPA);
@@ -5613,19 +5607,9 @@ static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
                rc = 0;
 
        while (rc == 0) {
-               if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY3) &&
-                   cfs_fail_val > 0) {
-                       struct l_wait_info lwi;
-
-                       lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
-                                         NULL, NULL);
-                       l_wait_event(thread->t_ctl_waitq,
-                                    !thread_is_running(thread),
-                                    &lwi);
-
-                       if (unlikely(!thread_is_running(thread)))
-                               break;
-               }
+               if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
+                   unlikely(!thread_is_running(thread)))
+                       break;
 
                rc = iops->rec(env, di, (struct dt_rec *)ent,
                               LUDA_64BITHASH | LUDA_TYPE);
@@ -5832,14 +5816,15 @@ out:
        RETURN(rc == 0 ? 1 : rc);
 }
 
-static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
-                                               struct lfsck_component *com)
+static int
+lfsck_namespace_double_scan_one_trace_file(const struct lu_env *env,
+                                          struct lfsck_component *com,
+                                          struct dt_object *obj, bool first)
 {
        struct lfsck_instance   *lfsck  = com->lc_lfsck;
        struct ptlrpc_thread    *thread = &lfsck->li_thread;
        struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
        struct lfsck_namespace  *ns     = com->lc_file_ram;
-       struct dt_object        *obj    = com->lc_obj;
        const struct dt_it_ops  *iops   = &obj->do_index_ops->dio_it;
        struct dt_object        *target;
        struct dt_it            *di;
@@ -5849,61 +5834,40 @@ static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
        __u8                     flags  = 0;
        ENTRY;
 
-       while (!list_empty(&lfsck->li_list_lmv)) {
-               struct lfsck_lmv_unit *llu;
-
-               spin_lock(&lfsck->li_lock);
-               llu = list_entry(lfsck->li_list_lmv.next,
-                                struct lfsck_lmv_unit, llu_link);
-               list_del_init(&llu->llu_link);
-               spin_unlock(&lfsck->li_lock);
-
-               rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
-               if (rc <= 0)
-                       RETURN(rc);
-       }
-
-       CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
-              lfsck_lfsck2name(lfsck));
-
-       lfsck_namespace_scan_local_lpf(env, com);
-
-       com->lc_new_checked = 0;
-       com->lc_new_scanned = 0;
-       com->lc_time_last_checkpoint = cfs_time_current();
-       com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
-                               cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
-
        di = iops->init(env, obj, 0, BYPASS_CAPA);
        if (IS_ERR(di))
                RETURN(PTR_ERR(di));
 
-       fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
+       if (first)
+               fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
+       else
+               fid_zero(&fid);
        rc = iops->get(env, di, (const struct dt_key *)&fid);
        if (rc < 0)
                GOTO(fini, rc);
 
-       /* Skip the start one, which either has been processed or non-exist. */
-       rc = iops->next(env, di);
-       if (rc != 0)
-               GOTO(put, rc);
+       if (first) {
+               /* The start one either has been processed or does not exist,
+                * skip it. */
+               rc = iops->next(env, di);
+               if (rc != 0)
+                       GOTO(put, rc);
+       }
 
        do {
-               if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY3) &&
-                   cfs_fail_val > 0) {
-                       struct l_wait_info lwi;
-
-                       lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
-                                         NULL, NULL);
-                       l_wait_event(thread->t_ctl_waitq,
-                                    !thread_is_running(thread),
-                                    &lwi);
-
-                       if (unlikely(!thread_is_running(thread)))
-                               GOTO(put, rc = 0);
-               }
+               if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
+                   unlikely(!thread_is_running(thread)))
+                       GOTO(put, rc = 0);
 
                key = iops->key(env, di);
+               if (IS_ERR(key)) {
+                       rc = PTR_ERR(key);
+                       if (rc == -ENOENT)
+                               GOTO(put, rc = 1);
+
+                       goto checkpoint;
+               }
+
                fid_be_to_cpu(&fid, (const struct lu_fid *)key);
                if (!fid_is_sane(&fid)) {
                        rc = 0;
@@ -5932,7 +5896,8 @@ checkpoint:
                down_write(&com->lc_sem);
                com->lc_new_checked++;
                com->lc_new_scanned++;
-               ns->ln_fid_latest_scanned_phase2 = fid;
+               if (rc >= 0 && fid_is_sane(&fid))
+                       ns->ln_fid_latest_scanned_phase2 = fid;
                if (rc > 0)
                        ns->ln_objs_repaired_phase2++;
                else if (rc < 0)
@@ -5952,7 +5917,7 @@ checkpoint:
                        ns->ln_time_last_checkpoint = cfs_time_current_sec();
                        ns->ln_objs_checked_phase2 += com->lc_new_checked;
                        com->lc_new_checked = 0;
-                       rc = lfsck_namespace_store(env, com);
+                       rc = lfsck_namespace_store(env, com, false);
                        up_write(&com->lc_sem);
                        if (rc != 0)
                                GOTO(put, rc);
@@ -5978,12 +5943,56 @@ put:
 fini:
        iops->fini(env, di);
 
-       CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop: rc = %d\n",
-              lfsck_lfsck2name(lfsck), rc);
-
        return rc;
 }
 
+static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
+                                               struct lfsck_component *com)
+{
+       struct lfsck_instance   *lfsck  = com->lc_lfsck;
+       struct lfsck_namespace  *ns     = com->lc_file_ram;
+       int                      rc;
+       int                      i;
+       ENTRY;
+
+       while (!list_empty(&lfsck->li_list_lmv)) {
+               struct lfsck_lmv_unit *llu;
+
+               spin_lock(&lfsck->li_lock);
+               llu = list_entry(lfsck->li_list_lmv.next,
+                                struct lfsck_lmv_unit, llu_link);
+               list_del_init(&llu->llu_link);
+               spin_unlock(&lfsck->li_lock);
+
+               rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
+               if (rc <= 0)
+                       RETURN(rc);
+       }
+
+       CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
+              lfsck_lfsck2name(lfsck));
+
+       lfsck_namespace_scan_local_lpf(env, com);
+
+       com->lc_new_checked = 0;
+       com->lc_new_scanned = 0;
+       com->lc_time_last_checkpoint = cfs_time_current();
+       com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
+                               cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
+
+       i = lfsck_sub_trace_file_fid2idx(&ns->ln_fid_latest_scanned_phase2);
+       rc = lfsck_namespace_double_scan_one_trace_file(env, com,
+                               com->lc_sub_trace_objs[i].lsto_obj, true);
+       while (rc > 0 && ++i < LFSCK_STF_COUNT)
+               rc = lfsck_namespace_double_scan_one_trace_file(env, com,
+                               com->lc_sub_trace_objs[i].lsto_obj, false);
+
+       CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop at the No. %d "
+              "trace file: rc = %d\n", lfsck_lfsck2name(lfsck), i, rc);
+
+       RETURN(rc);
+}
+
 static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env,
                                               struct lfsck_component *com,
                                               struct lfsck_position *pos)
@@ -6034,7 +6043,7 @@ static int lfsck_namespace_double_scan_result(const struct lu_env *env,
                ns->ln_status = LS_FAILED;
        }
 
-       rc = lfsck_namespace_store(env, com);
+       rc = lfsck_namespace_store(env, com, false);
        up_write(&com->lc_sem);
 
        return rc;
@@ -6045,6 +6054,13 @@ lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
                                                  struct ptlrpc_request *req,
                                                  void *args, int rc)
 {
+       if (rc == 0) {
+               struct lfsck_async_interpret_args *laia = args;
+               struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
+
+               ltd->ltd_synced_failures = 1;
+       }
+
        return 0;
 }
 
@@ -6081,9 +6097,13 @@ static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
        struct lfsck_tgt_descs            *ltds  = &lfsck->li_mdt_descs;
        struct lfsck_tgt_desc             *ltd;
        struct ptlrpc_request_set         *set;
+       __u32                              idx;
        int                                rc    = 0;
        ENTRY;
 
+       if (!lad->lad_incomplete)
+               RETURN_EXIT;
+
        set = ptlrpc_prep_set();
        if (set == NULL)
                GOTO(out, rc = -ENOMEM);
@@ -6092,25 +6112,12 @@ static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
        memset(laia, 0, sizeof(*laia));
        lad->lad_touch_gen++;
 
-       spin_lock(&ltds->ltd_lock);
-       while (!list_empty(&lad->lad_mdt_list)) {
-               ltd = list_entry(lad->lad_mdt_list.next,
-                                struct lfsck_tgt_desc,
-                                ltd_namespace_list);
-               if (ltd->ltd_namespace_gen == lad->lad_touch_gen)
-                       break;
+       down_read(&ltds->ltd_rw_sem);
+       cfs_foreach_bit(lad->lad_bitmap, idx) {
+               ltd = LTD_TGT(ltds, idx);
+               LASSERT(ltd != NULL);
 
-               ltd->ltd_namespace_gen = lad->lad_touch_gen;
-               list_move_tail(&ltd->ltd_namespace_list,
-                              &lad->lad_mdt_list);
-               if (!lad->lad_incomplete ||
-                   !cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
-                       ltd->ltd_namespace_failed = 0;
-                       continue;
-               }
-
-               ltd->ltd_namespace_failed = 1;
-               spin_unlock(&ltds->ltd_lock);
+               laia->laia_ltd = ltd;
                rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
                        lfsck_namespace_assistant_sync_failures_interpret,
                        laia, LFSCK_NOTIFY);
@@ -6118,10 +6125,8 @@ static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
                        CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
                               "to sync failure with MDT %x: rc = %d\n",
                               lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
-
-               spin_lock(&ltds->ltd_lock);
        }
-       spin_unlock(&ltds->ltd_lock);
+       up_read(&ltds->ltd_rw_sem);
 
        rc = ptlrpc_set_wait(set);
        ptlrpc_set_destroy(set);
@@ -6165,7 +6170,7 @@ int lfsck_verify_linkea(const struct lu_env *env, struct dt_device *dev,
                        struct dt_object *obj, const struct lu_name *cname,
                        const struct lu_fid *pfid)
 {
-       struct linkea_data       ldata  = { 0 };
+       struct linkea_data       ldata  = { NULL };
        struct lu_buf            linkea_buf;
        struct thandle          *th;
        int                      rc;
@@ -6244,7 +6249,7 @@ int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
                          char *name, struct lu_fid *pfid)
 {
        struct lu_name           *cname = &lfsck_env_info(env)->lti_name;
-       struct linkea_data        ldata = { 0 };
+       struct linkea_data        ldata = { NULL };
        int                       rc;
 
        rc = lfsck_links_read(env, obj, &ldata);
@@ -6264,82 +6269,6 @@ int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
 }
 
 /**
- * Remove the name entry from the parent directory.
- *
- * No need to care about the object referenced by the name entry,
- * either the name entry is invalid or redundant, or the referenced
- * object has been processed has been or will be handled by others.
- *
- * \param[in] env      pointer to the thread context
- * \param[in] lfsck    pointer to the lfsck instance
- * \param[in] parent   pointer to the lost+found object
- * \param[in] name     the name for the name entry to be removed
- * \param[in] type     the type for the name entry to be removed
- *
- * \retval             0 for success
- * \retval             negative error number on failure
- */
-int lfsck_remove_name_entry(const struct lu_env *env,
-                           struct lfsck_instance *lfsck,
-                           struct dt_object *parent,
-                           const char *name, __u32 type)
-{
-       struct dt_device        *dev    = lfsck->li_next;
-       struct thandle          *th;
-       struct lustre_handle     lh     = { 0 };
-       int                      rc;
-       ENTRY;
-
-       rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
-                             MDS_INODELOCK_UPDATE, LCK_EX);
-       if (rc != 0)
-               RETURN(rc);
-
-       th = dt_trans_create(env, dev);
-       if (IS_ERR(th))
-               GOTO(unlock, rc = PTR_ERR(th));
-
-       rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
-       if (rc != 0)
-               GOTO(stop, rc);
-
-       if (S_ISDIR(type)) {
-               rc = dt_declare_ref_del(env, parent, th);
-               if (rc != 0)
-                       GOTO(stop, rc);
-       }
-
-       rc = dt_trans_start(env, dev, th);
-       if (rc != 0)
-               GOTO(stop, rc);
-
-       rc = dt_delete(env, parent, (const struct dt_key *)name, th,
-                      BYPASS_CAPA);
-       if (rc != 0)
-               GOTO(stop, rc);
-
-       if (S_ISDIR(type)) {
-               dt_write_lock(env, parent, 0);
-               rc = dt_ref_del(env, parent, th);
-               dt_write_unlock(env, parent);
-       }
-
-       GOTO(stop, rc);
-
-stop:
-       dt_trans_stop(env, dev, th);
-
-unlock:
-       lfsck_ibits_unlock(&lh, LCK_EX);
-
-       CDEBUG(D_LFSCK, "%s: remove name entry "DFID"/%s "
-              "with type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
-              PFID(lfsck_dto2fid(parent)), name, type, rc);
-
-       return rc;
-}
-
-/**
  * Update the object's name entry with the given FID.
  *
  * \param[in] env      pointer to the thread context
@@ -6434,6 +6363,7 @@ int lfsck_namespace_setup(const struct lu_env *env,
        struct lfsck_namespace  *ns;
        struct dt_object        *root = NULL;
        struct dt_object        *obj;
+       int                      i;
        int                      rc;
        ENTRY;
 
@@ -6452,7 +6382,7 @@ int lfsck_namespace_setup(const struct lu_env *env,
        com->lc_ops = &lfsck_namespace_ops;
        com->lc_data = lfsck_assistant_data_init(
                        &lfsck_namespace_assistant_ops,
-                       "lfsck_namespace");
+                       LFSCK_NAMESPACE);
        if (com->lc_data == NULL)
                GOTO(out, rc = -ENOMEM);
 
@@ -6465,6 +6395,9 @@ int lfsck_namespace_setup(const struct lu_env *env,
        if (com->lc_file_disk == NULL)
                GOTO(out, rc = -ENOMEM);
 
+       for (i = 0; i < LFSCK_STF_COUNT; i++)
+               mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
+
        root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
        if (IS_ERR(root))
                GOTO(out, rc = PTR_ERR(root));
@@ -6472,23 +6405,20 @@ int lfsck_namespace_setup(const struct lu_env *env,
        if (unlikely(!dt_try_as_dir(env, root)))
                GOTO(out, rc = -ENOTDIR);
 
-       obj = local_index_find_or_create(env, lfsck->li_los, root,
-                                        lfsck_namespace_name,
-                                        S_IFREG | S_IRUGO | S_IWUSR,
-                                        &dt_lfsck_features);
+       obj = local_file_find_or_create(env, lfsck->li_los, root,
+                                       LFSCK_NAMESPACE,
+                                       S_IFREG | S_IRUGO | S_IWUSR);
        if (IS_ERR(obj))
                GOTO(out, rc = PTR_ERR(obj));
 
        com->lc_obj = obj;
-       rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_features);
-       if (rc != 0)
-               GOTO(out, rc);
-
        rc = lfsck_namespace_load(env, com);
-       if (rc > 0)
-               rc = lfsck_namespace_reset(env, com, true);
-       else if (rc == -ENODATA)
+       if (rc == -ENODATA)
                rc = lfsck_namespace_init(env, com);
+       else if (rc < 0)
+               rc = lfsck_namespace_reset(env, com, true);
+       else
+               rc = lfsck_namespace_load_sub_trace_files(env, com, false);
        if (rc != 0)
                GOTO(out, rc);