Whamcloud - gitweb
LU-14119 lfsck: replace dt_lookup() with dt_lookup_dir()
[fs/lustre-release.git] / lustre / lfsck / lfsck_namespace.c
index 5a09bdc..ff660c4 100644 (file)
@@ -20,7 +20,7 @@
  * GPL HEADER END
  */
 /*
- * Copyright (c) 2013, 2016, Intel Corporation.
+ * Copyright (c) 2013, 2017, Intel Corporation.
  */
 /*
  * lustre/lfsck/lfsck_namespace.c
 
 #define DEBUG_SUBSYSTEM S_LFSCK
 
-#include <lustre/lustre_idl.h>
 #include <lu_object.h>
 #include <dt_object.h>
 #include <md_object.h>
 #include <lustre_fid.h>
 #include <lustre_lib.h>
 #include <lustre_net.h>
-#include <lustre/lustre_user.h>
 
 #include "lfsck_internal.h"
 
 #define LFSCK_NAMESPACE_MAGIC_V1       0xA0629D03
 #define LFSCK_NAMESPACE_MAGIC_V2       0xA0621A0B
+#define LFSCK_NAMESPACE_MAGIC_V3       0xA06249FF
 
 /* 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
+#define LFSCK_NAMESPACE_MAGIC          LFSCK_NAMESPACE_MAGIC_V3
 
 enum lfsck_nameentry_check {
        LFSCK_NAMEENTRY_DEAD            = 1, /* The object has been unlinked. */
@@ -88,7 +87,7 @@ static void lfsck_namespace_assistant_req_fini(const struct lu_env *env,
                                               struct lfsck_assistant_req *lar)
 {
        struct lfsck_namespace_req *lnr =
-                       container_of0(lar, struct lfsck_namespace_req, lnr_lar);
+               container_of(lar, struct lfsck_namespace_req, lnr_lar);
 
        if (lnr->lnr_lmv != NULL)
                lfsck_lmv_put(env, lnr->lnr_lmv);
@@ -104,8 +103,8 @@ static void lfsck_namespace_le_to_cpu(struct lfsck_namespace *dst,
        dst->ln_status = le32_to_cpu(src->ln_status);
        dst->ln_flags = le32_to_cpu(src->ln_flags);
        dst->ln_success_count = le32_to_cpu(src->ln_success_count);
-       dst->ln_run_time_phase1 = le32_to_cpu(src->ln_run_time_phase1);
-       dst->ln_run_time_phase2 = le32_to_cpu(src->ln_run_time_phase2);
+       dst->ln_run_time_phase1 = le64_to_cpu(src->ln_run_time_phase1);
+       dst->ln_run_time_phase2 = le64_to_cpu(src->ln_run_time_phase2);
        dst->ln_time_last_complete = le64_to_cpu(src->ln_time_last_complete);
        dst->ln_time_latest_start = le64_to_cpu(src->ln_time_latest_start);
        dst->ln_time_last_checkpoint =
@@ -164,9 +163,11 @@ static void lfsck_namespace_le_to_cpu(struct lfsck_namespace *dst,
        dst->ln_local_lpf_skipped = le64_to_cpu(src->ln_local_lpf_skipped);
        dst->ln_local_lpf_failed = le64_to_cpu(src->ln_local_lpf_failed);
        dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size);
-       dst->ln_time_latest_reset = le32_to_cpu(src->ln_time_latest_reset);
+       dst->ln_time_latest_reset = le64_to_cpu(src->ln_time_latest_reset);
        dst->ln_linkea_overflow_cleared =
                                le64_to_cpu(src->ln_linkea_overflow_cleared);
+       dst->ln_agent_entries_repaired =
+                               le64_to_cpu(src->ln_agent_entries_repaired);
 }
 
 static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
@@ -176,8 +177,8 @@ static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
        dst->ln_status = cpu_to_le32(src->ln_status);
        dst->ln_flags = cpu_to_le32(src->ln_flags);
        dst->ln_success_count = cpu_to_le32(src->ln_success_count);
-       dst->ln_run_time_phase1 = cpu_to_le32(src->ln_run_time_phase1);
-       dst->ln_run_time_phase2 = cpu_to_le32(src->ln_run_time_phase2);
+       dst->ln_run_time_phase1 = cpu_to_le64(src->ln_run_time_phase1);
+       dst->ln_run_time_phase2 = cpu_to_le64(src->ln_run_time_phase2);
        dst->ln_time_last_complete = cpu_to_le64(src->ln_time_last_complete);
        dst->ln_time_latest_start = cpu_to_le64(src->ln_time_latest_start);
        dst->ln_time_last_checkpoint =
@@ -236,9 +237,11 @@ static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
        dst->ln_local_lpf_skipped = cpu_to_le64(src->ln_local_lpf_skipped);
        dst->ln_local_lpf_failed = cpu_to_le64(src->ln_local_lpf_failed);
        dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size);
-       dst->ln_time_latest_reset = cpu_to_le32(src->ln_time_latest_reset);
+       dst->ln_time_latest_reset = cpu_to_le64(src->ln_time_latest_reset);
        dst->ln_linkea_overflow_cleared =
                                cpu_to_le64(src->ln_linkea_overflow_cleared);
+       dst->ln_agent_entries_repaired =
+                               cpu_to_le64(src->ln_agent_entries_repaired);
 }
 
 static void lfsck_namespace_record_failure(const struct lu_env *env,
@@ -309,7 +312,7 @@ static int lfsck_namespace_load_bitmap(const struct lu_env *env,
        }
 
        if (ns->ln_bitmap_size == 0) {
-               lad->lad_incomplete = 0;
+               clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
                CFS_RESET_BITMAP(bitmap);
 
                RETURN(0);
@@ -323,9 +326,9 @@ static int lfsck_namespace_load_bitmap(const struct lu_env *env,
                RETURN(rc >= 0 ? -EINVAL : rc);
 
        if (cfs_bitmap_check_empty(bitmap))
-               lad->lad_incomplete = 0;
+               clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
        else
-               lad->lad_incomplete = 1;
+               set_bit(LAD_INCOMPLETE, &lad->lad_flags);
 
        RETURN(0);
 }
@@ -449,7 +452,7 @@ static int lfsck_namespace_init(const struct lu_env *env,
        memset(ns, 0, sizeof(*ns));
        ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
        ns->ln_status = LS_INIT;
-       ns->ln_time_latest_reset = cfs_time_current_sec();
+       ns->ln_time_latest_reset = ktime_get_real_seconds();
        down_write(&com->lc_sem);
        rc = lfsck_namespace_store(env, com);
        if (rc == 0)
@@ -559,7 +562,7 @@ int lfsck_namespace_trace_update(const struct lu_env *env,
 
        if (new != 0) {
                rc = dt_insert(env, obj, (const struct dt_rec *)&new,
-                              (const struct dt_key *)key, th, 1);
+                              (const struct dt_key *)key, th);
                if (rc != 0)
                        GOTO(log, rc);
        }
@@ -593,8 +596,7 @@ int lfsck_namespace_check_exist(const struct lu_env *env,
        if (unlikely(lfsck_is_dead_obj(obj)))
                RETURN(LFSCK_NAMEENTRY_DEAD);
 
-       rc = dt_lookup(env, dir, (struct dt_rec *)fid,
-                      (const struct dt_key *)name);
+       rc = dt_lookup_dir(env, dir, name, fid);
        if (rc == -ENOENT)
                RETURN(LFSCK_NAMEENTRY_REMOVED);
 
@@ -757,19 +759,41 @@ again:
        return rc;
 }
 
+static inline bool linkea_reclen_is_valid(const struct linkea_data *ldata)
+{
+       if (ldata->ld_reclen <= 0)
+               return false;
+
+       if ((char *)ldata->ld_lee + ldata->ld_reclen >
+           (char *)ldata->ld_leh + ldata->ld_leh->leh_len)
+               return false;
+
+       return true;
+}
+
+static inline bool linkea_entry_is_valid(const struct linkea_data *ldata,
+                                        const struct lu_name *cname,
+                                        const struct lu_fid *pfid)
+{
+       if (!linkea_reclen_is_valid(ldata))
+               return false;
+
+       if (cname->ln_namelen <= 0 || cname->ln_namelen > NAME_MAX)
+               return false;
+
+       if (!fid_is_sane(pfid))
+               return false;
+
+       return true;
+}
+
 static int lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
                                               struct lu_name *cname,
                                               struct lu_fid *pfid,
                                               char *buf, const int buflen)
 {
        linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
-       if (unlikely(ldata->ld_reclen <= 0 ||
-                    ldata->ld_reclen + sizeof(struct link_ea_header) >
-                       ldata->ld_leh->leh_len ||
-                    cname->ln_namelen <= 0 ||
-                    cname->ln_namelen > NAME_MAX ||
-                    cname->ln_namelen >= buflen ||
-                    !fid_is_sane(pfid)))
+       if (unlikely(!linkea_entry_is_valid(ldata, cname, pfid)))
                return -EINVAL;
 
        /* To guarantee the 'name' is terminated with '0'. */
@@ -787,9 +811,7 @@ static void lfsck_linkea_del_buf(struct linkea_data *ldata,
 
        /* If current record is corrupted, all the subsequent
         * records will be dropped. */
-       if (unlikely(ldata->ld_reclen <= 0 ||
-                    ldata->ld_reclen + sizeof(struct link_ea_header) >
-                       ldata->ld_leh->leh_len)) {
+       if (unlikely(!linkea_reclen_is_valid(ldata))) {
                void *ptr = ldata->ld_lee;
 
                ldata->ld_leh->leh_len = sizeof(struct link_ea_header);
@@ -827,7 +849,10 @@ static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
        while (ldata->ld_lee != NULL) {
                ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
                                   ldata->ld_lee->lee_reclen[1];
-               if (unlikely(ldata->ld_reclen == oldlen &&
+               if (unlikely(!linkea_reclen_is_valid(ldata))) {
+                       lfsck_linkea_del_buf(ldata, NULL);
+                       LASSERT(ldata->ld_lee == NULL);
+               } else if (unlikely(ldata->ld_reclen == oldlen &&
                             memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
                        repeated++;
                        if (!remove)
@@ -920,8 +945,7 @@ again:
        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,
-                              (const struct dt_key *)info->lti_key);
+               rc = dt_lookup_dir(env, parent, info->lti_key, &tfid);
                if (rc != 0 && rc != -ENOENT)
                        GOTO(log, rc);
 
@@ -936,8 +960,7 @@ again:
 
        /* Re-check whether the name conflict with othrs after taken
         * the ldlm lock. */
-       rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
-                      (const struct dt_key *)info->lti_key);
+       rc = dt_lookup_dir(env, parent, info->lti_key, &tfid);
        if (rc == 0) {
                if (!lu_fid_eq(cfid, &tfid)) {
                        exist = false;
@@ -1007,7 +1030,7 @@ again:
        }
 
        memset(la, 0, sizeof(*la));
-       la->la_ctime = cfs_time_current_sec();
+       la->la_ctime = ktime_get_real_seconds();
        la->la_valid = LA_CTIME;
        rc = dt_declare_attr_set(env, orphan, la, th);
        if (rc != 0)
@@ -1032,7 +1055,7 @@ again:
                        rec->rec_type = S_IFDIR;
                        rec->rec_fid = pfid;
                        rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
-                                      (const struct dt_key *)dotdot, th, 1);
+                                      (const struct dt_key *)dotdot, th);
                        if (rc != 0)
                                GOTO(unlock, rc);
                }
@@ -1051,7 +1074,7 @@ again:
                rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
                rec->rec_fid = cfid;
                rc = dt_insert(env, parent, (const struct dt_rec *)rec,
-                              (const struct dt_key *)cname->ln_name, th, 1);
+                              (const struct dt_key *)cname->ln_name, th);
                if (rc == 0 && S_ISDIR(rec->rec_type)) {
                        dt_write_lock(env, parent, 0);
                        rc = dt_ref_add(env, parent, th);
@@ -1087,6 +1110,99 @@ log:
        return rc;
 }
 
+static int lfsck_lmv_set(const struct lu_env *env,
+                        struct lfsck_instance *lfsck,
+                        struct dt_object *obj,
+                        struct lmv_mds_md_v1 *lmv)
+{
+       struct dt_device *dev = lfsck->li_next;
+       struct thandle *th = NULL;
+       struct lu_buf buf = { lmv, sizeof(*lmv) };
+       int rc;
+
+       ENTRY;
+
+       th = dt_trans_create(env, dev);
+       if (IS_ERR(th))
+               RETURN(PTR_ERR(th));
+
+       rc = dt_declare_xattr_set(env, obj, &buf, XATTR_NAME_LMV, 0, th);
+       if (rc)
+               GOTO(stop, rc);
+
+       rc = dt_trans_start_local(env, dev, th);
+       if (rc != 0)
+               GOTO(stop, rc);
+
+       rc = dt_xattr_set(env, obj, &buf, XATTR_NAME_LMV, 0, th);
+       if (rc)
+               GOTO(stop, rc);
+
+       EXIT;
+stop:
+       dt_trans_stop(env, dev, th);
+
+       return rc;
+}
+
+static int lfsck_lmv_delete(const struct lu_env *env,
+                           struct lfsck_instance *lfsck,
+                           struct dt_object *obj)
+{
+       struct dt_device *dev = lfsck->li_next;
+       struct thandle *th = NULL;
+       int rc;
+
+       ENTRY;
+
+       th = dt_trans_create(env, dev);
+       if (IS_ERR(th))
+               RETURN(PTR_ERR(th));
+
+       rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LMV, th);
+       if (rc)
+               GOTO(stop, rc);
+
+       rc = dt_trans_start_local(env, dev, th);
+       if (rc != 0)
+               GOTO(stop, rc);
+
+       rc = dt_xattr_del(env, obj, XATTR_NAME_LMV, th);
+       if (rc)
+               GOTO(stop, rc);
+
+       EXIT;
+stop:
+       dt_trans_stop(env, dev, th);
+
+       return rc;
+}
+
+static inline int lfsck_object_is_shard(const struct lu_env *env,
+                                       struct lfsck_instance *lfsck,
+                                       struct dt_object *obj,
+                                       const struct lu_name *lname)
+{
+       struct lfsck_thread_info *info = lfsck_env_info(env);
+       struct lmv_mds_md_v1 *lmv = &info->lti_lmv;
+       int rc;
+
+       rc = lfsck_shard_name_to_index(env, lname->ln_name, lname->ln_namelen,
+                                      lfsck_object_type(obj),
+                                      lfsck_dto2fid(obj));
+       if (rc < 0)
+               return 0;
+
+       rc = lfsck_read_stripe_lmv(env, lfsck, obj, lmv);
+       if (rc == -ENODATA)
+               return 0;
+
+       if (!rc && lmv->lmv_magic == LMV_MAGIC_STRIPE)
+               return 1;
+
+       return rc;
+}
+
 /**
  * Add the specified name entry back to namespace.
  *
@@ -1097,13 +1213,17 @@ log:
  * it is quite possible that the name entry is lost. Then the LFSCK
  * should add the name entry back to the namespace.
  *
+ * If \a child is shard, which means \a parent is a striped directory,
+ * if \a parent has LMV, we need to delete it before insertion because
+ * now parent's striping is broken and can't be parsed correctly.
+ *
  * \param[in] env      pointer to the thread context
  * \param[in] com      pointer to the lfsck component
  * \param[in] parent   pointer to the directory under which the name entry
  *                     will be inserted into
  * \param[in] child    pointer to the object referenced by the name entry
  *                     that to be inserted into the parent
- * \param[in] name     the name for the child in the parent directory
+ * \param[in] lname    the name for the child in the parent directory
  *
  * \retval             positive number for repaired cases
  * \retval             0 if nothing to be repaired
@@ -1113,19 +1233,26 @@ static int lfsck_namespace_insert_normal(const struct lu_env *env,
                                         struct lfsck_component *com,
                                         struct dt_object *parent,
                                         struct dt_object *child,
-                                        const char *name)
+                                        const struct lu_name *lname)
 {
-       struct lfsck_thread_info        *info   = lfsck_env_info(env);
-       struct lu_attr                  *la     = &info->lti_la;
-       struct dt_insert_rec            *rec    = &info->lti_dt_rec;
-       struct lfsck_instance           *lfsck  = com->lc_lfsck;
+       struct lfsck_thread_info *info = lfsck_env_info(env);
+       struct lu_attr *la = &info->lti_la;
+       struct dt_insert_rec *rec = &info->lti_dt_rec;
+       struct lfsck_instance *lfsck = com->lc_lfsck;
        /* The child and its name may be on different MDTs. */
-       const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
-       const struct lu_fid             *cfid   = lfsck_dto2fid(child);
-       struct dt_device                *dev    = lfsck->li_next;
-       struct thandle                  *th     = NULL;
-       struct lfsck_lock_handle        *llh    = &info->lti_llh;
-       int                              rc     = 0;
+       const struct lu_fid *pfid = lfsck_dto2fid(parent);
+       const struct lu_fid *cfid = lfsck_dto2fid(child);
+       struct dt_device *dev = lfsck->li_next;
+       struct thandle *th = NULL;
+       struct lfsck_lock_handle *llh = &info->lti_llh;
+       struct lmv_mds_md_v1 *lmv = &info->lti_lmv;
+       struct lu_buf buf = { lmv, sizeof(*lmv) };
+       /* whether parent's LMV is deleted before insertion */
+       bool parent_lmv_deleted = false;
+       /* whether parent's LMV is missing */
+       bool parent_lmv_lost = false;
+       int rc = 0;
+
        ENTRY;
 
        /* @parent/@child may be based on lfsck->li_bottom,
@@ -1135,9 +1262,6 @@ static int lfsck_namespace_insert_normal(const struct lu_env *env,
        if (IS_ERR(parent))
                GOTO(log, rc = PTR_ERR(parent));
 
-       if (unlikely(!dt_try_as_dir(env, parent)))
-               GOTO(log, rc = -ENOTDIR);
-
        child = lfsck_object_locate(dev, child);
        if (IS_ERR(child))
                GOTO(log, rc = PTR_ERR(child));
@@ -1145,11 +1269,57 @@ static int lfsck_namespace_insert_normal(const struct lu_env *env,
        if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
                GOTO(log, rc = 1);
 
-       rc = lfsck_lock(env, lfsck, parent, name, llh,
-                       MDS_INODELOCK_UPDATE, LCK_PW);
-       if (rc != 0)
+       rc = lfsck_lock(env, lfsck, parent, lname->ln_name, llh,
+                       MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
+                       MDS_INODELOCK_XATTR, LCK_EX);
+       if (rc)
                GOTO(log, rc);
 
+       rc = lfsck_object_is_shard(env, lfsck, child, lname);
+       if (rc < 0)
+               GOTO(unlock, rc);
+
+       if (rc == 1) {
+               rc = lfsck_read_stripe_lmv(env, lfsck, parent, lmv);
+               if (!rc) {
+                       /*
+                        * To add a shard, we need to convert parent to a
+                        * plain directory by deleting its LMV, and after
+                        * insertion set it back.
+                        */
+                       rc = lfsck_lmv_delete(env, lfsck, parent);
+                       if (rc)
+                               GOTO(unlock, rc);
+                       parent_lmv_deleted = true;
+                       lmv->lmv_layout_version++;
+                       lfsck_lmv_header_cpu_to_le(lmv, lmv);
+               } else if (rc == -ENODATA) {
+                       struct lu_seq_range *range = &info->lti_range;
+                       struct seq_server_site *ss = lfsck_dev_site(lfsck);
+
+                       rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
+                       if (rc)
+                               GOTO(unlock, rc);
+
+                       fld_range_set_mdt(range);
+                       rc = fld_server_lookup(env, ss->ss_server_fld,
+                                      fid_seq(lfsck_dto2fid(parent)), range);
+                       if (rc)
+                               GOTO(unlock, rc);
+
+                       parent_lmv_lost = true;
+                       lmv->lmv_magic = LMV_MAGIC;
+                       lmv->lmv_master_mdt_index = range->lsr_index;
+                       lmv->lmv_layout_version++;
+                       lfsck_lmv_header_cpu_to_le(lmv, lmv);
+               } else {
+                       GOTO(unlock, rc);
+               }
+       }
+
+       if (unlikely(!dt_try_as_dir(env, parent)))
+               GOTO(unlock, rc = -ENOTDIR);
+
        th = dt_trans_create(env, dev);
        if (IS_ERR(th))
                GOTO(unlock, rc = PTR_ERR(th));
@@ -1157,7 +1327,7 @@ static int lfsck_namespace_insert_normal(const struct lu_env *env,
        rec->rec_type = lfsck_object_type(child) & S_IFMT;
        rec->rec_fid = cfid;
        rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
-                              (const struct dt_key *)name, th);
+                              (const struct dt_key *)lname->ln_name, th);
        if (rc != 0)
                GOTO(stop, rc);
 
@@ -1167,8 +1337,14 @@ static int lfsck_namespace_insert_normal(const struct lu_env *env,
                        GOTO(stop, rc);
        }
 
-       memset(la, 0, sizeof(*la));
-       la->la_ctime = cfs_time_current_sec();
+       if (parent_lmv_lost) {
+               rc = dt_declare_xattr_set(env, parent, &buf, XATTR_NAME_LMV,
+                                         0, th);
+               if (rc)
+                       GOTO(stop, rc);
+       }
+
+       la->la_ctime = ktime_get_real_seconds();
        la->la_valid = LA_CTIME;
        rc = dt_declare_attr_set(env, parent, la, th);
        if (rc != 0)
@@ -1183,7 +1359,7 @@ static int lfsck_namespace_insert_normal(const struct lu_env *env,
                GOTO(stop, rc);
 
        rc = dt_insert(env, parent, (const struct dt_rec *)rec,
-                      (const struct dt_key *)name, th, 1);
+                      (const struct dt_key *)lname->ln_name, th);
        if (rc != 0)
                GOTO(stop, rc);
 
@@ -1195,7 +1371,12 @@ static int lfsck_namespace_insert_normal(const struct lu_env *env,
                        GOTO(stop, rc);
        }
 
-       la->la_ctime = cfs_time_current_sec();
+       if (parent_lmv_lost) {
+               rc = dt_xattr_set(env, parent, &buf, XATTR_NAME_LMV, 0, th);
+               if (rc)
+                       GOTO(stop, rc);
+       }
+
        rc = dt_attr_set(env, parent, la, th);
        if (rc != 0)
                GOTO(stop, rc);
@@ -1208,12 +1389,15 @@ stop:
        dt_trans_stop(env, dev, th);
 
 unlock:
+       if (parent_lmv_deleted)
+               lfsck_lmv_set(env, lfsck, parent, lmv);
+
        lfsck_unlock(llh);
 
 log:
        CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
               "the name %s and type %o to the parent "DFID": rc = %d\n",
-              lfsck_lfsck2name(lfsck), PFID(cfid), name,
+              lfsck_lfsck2name(lfsck), PFID(cfid), lname->ln_name,
               lfsck_object_type(child) & S_IFMT, PFID(pfid), rc);
 
        if (rc != 0) {
@@ -1287,9 +1471,7 @@ static int lfsck_namespace_create_orphan_dir(const struct lu_env *env,
                        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);
+               rc = dt_lookup_dir(env, lfsck->li_lpf_root_obj, name, &tfid);
                if (rc != 0)
                        GOTO(log, rc = (rc == -ENOENT ? -ENXIO : rc));
 
@@ -1316,8 +1498,7 @@ again:
        do {
                namelen = snprintf(name, 31, DFID"-P-%d",
                                   PFID(cfid), idx++);
-               rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
-                              (const struct dt_key *)name);
+               rc = dt_lookup_dir(env, parent, name, &tfid);
                if (rc != 0 && rc != -ENOENT)
                        GOTO(log, rc);
        } while (rc == 0);
@@ -1329,8 +1510,7 @@ again:
 
        /* Re-check whether the name conflict with othrs after taken
         * the ldlm lock. */
-       rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
-                      (const struct dt_key *)name);
+       rc = dt_lookup_dir(env, parent, name, &tfid);
        if (unlikely(rc == 0)) {
                lfsck_unlock(llh);
                goto again;
@@ -1396,8 +1576,8 @@ again:
                lmv->lmv_master_mdt_index = lfsck_dev_idx(lfsck);
                lfsck_lmv_header_cpu_to_le(lmv2, lmv);
                lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
-               rc = dt_declare_xattr_set(env, orphan, &lmv_buf,
-                                         XATTR_NAME_LMV, 0, th);
+               rc = dt_declare_xattr_set(env, orphan, &lmv_buf, XATTR_NAME_LMV,
+                                         0, th);
                if (rc != 0)
                        GOTO(stop, rc);
        }
@@ -1433,13 +1613,13 @@ again:
 
        rec->rec_fid = cfid;
        rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
-                      (const struct dt_key *)dot, th, 1);
+                      (const struct dt_key *)dot, th);
        if (rc != 0)
                GOTO(unlock2, rc);
 
        rec->rec_fid = lfsck_dto2fid(parent);
        rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
-                      (const struct dt_key *)dotdot, th, 1);
+                      (const struct dt_key *)dotdot, th);
        if (rc != 0)
                GOTO(unlock2, rc);
 
@@ -1457,7 +1637,7 @@ again:
 
        rec->rec_fid = cfid;
        rc = dt_insert(env, parent, (const struct dt_rec *)rec,
-                      (const struct dt_key *)name, th, 1);
+                      (const struct dt_key *)name, th);
        if (rc == 0) {
                dt_write_lock(env, parent, 0);
                rc = dt_ref_add(env, parent, th);
@@ -1588,16 +1768,22 @@ again:
                lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
                                                    true);
 
-       if (buflen < ldata_new.ld_leh->leh_len) {
+       /*
+        * linkea may change because it doesn't take lock in the first read, if
+        * it becomes larger, restart from beginning.
+        */
+       if ((ldata_new.ld_leh->leh_reccount > 0 ||
+            unlikely(ldata_new.ld_leh->leh_overflow_time)) &&
+           buflen < ldata_new.ld_leh->leh_len) {
                dt_write_unlock(env, obj);
                dt_trans_stop(env, dev, th);
                lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
                               ldata_new.ld_leh->leh_len);
+               buflen = linkea_buf.lb_len;
                goto again;
        }
 
-       if (ldata_new.ld_leh->leh_reccount > 0 ||
-           unlikely(ldata->ld_leh->leh_overflow_time))
+       if (buflen)
                rc = lfsck_links_write(env, obj, &ldata_new, th);
        else
                rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
@@ -1796,8 +1982,7 @@ static int lfsck_namespace_replace_cond(const struct lu_env *env,
                goto replace;
        }
 
-       rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
-                      (const struct dt_key *)name);
+       rc = dt_lookup_dir(env, parent, name, &tfid);
        if (rc == -ENOENT) {
                exist = false;
                goto replace;
@@ -1897,7 +2082,7 @@ replace:
                GOTO(stop, rc);
 
        rc = dt_insert(env, parent, (const struct dt_rec *)rec,
-                      (const struct dt_key *)name, th, 1);
+                      (const struct dt_key *)name, th);
 
        GOTO(stop, rc = (rc == 0 ? 1 : rc));
 
@@ -2024,7 +2209,7 @@ int lfsck_namespace_repair_dirent(const struct lu_env *env,
        struct lfsck_thread_info        *info   = lfsck_env_info(env);
        struct dt_insert_rec            *rec    = &info->lti_dt_rec;
        const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
-       const struct lu_fid             *cfid   = lfsck_dto2fid(child);
+       struct lu_fid                   cfid    = {0};
        struct lu_fid                    tfid;
        struct lfsck_instance           *lfsck  = com->lc_lfsck;
        struct dt_device                *dev    = lfsck->li_next;
@@ -2034,6 +2219,8 @@ int lfsck_namespace_repair_dirent(const struct lu_env *env,
        int                              rc     = 0;
        ENTRY;
 
+       if (child)
+               cfid = *lfsck_dto2fid(child);
        parent = lfsck_object_locate(dev, parent);
        if (IS_ERR(parent))
                GOTO(log, rc = PTR_ERR(parent));
@@ -2060,7 +2247,8 @@ int lfsck_namespace_repair_dirent(const struct lu_env *env,
 
        if (update) {
                rec->rec_type = lfsck_object_type(child) & S_IFMT;
-               rec->rec_fid = cfid;
+               LASSERT(!fid_is_zero(&cfid));
+               rec->rec_fid = &cfid;
                rc = dt_declare_insert(env, parent,
                                       (const struct dt_rec *)rec,
                                       (const struct dt_key *)name2, th);
@@ -2080,8 +2268,7 @@ int lfsck_namespace_repair_dirent(const struct lu_env *env,
 
 
        dt_write_lock(env, parent, 0);
-       rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
-                      (const struct dt_key *)name);
+       rc = dt_lookup_dir(env, dt_object_child(parent), name, &tfid);
        /* Someone has removed the bad name entry by race. */
        if (rc == -ENOENT)
                GOTO(unlock2, rc = 0);
@@ -2091,7 +2278,7 @@ 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)
@@ -2104,7 +2291,7 @@ int lfsck_namespace_repair_dirent(const struct lu_env *env,
        if (update) {
                rc = dt_insert(env, parent,
                               (const struct dt_rec *)rec,
-                              (const struct dt_key *)name2, th, 1);
+                              (const struct dt_key *)name2, th);
                if (rc != 0)
                        GOTO(unlock2, rc);
        }
@@ -2126,8 +2313,8 @@ stop:
        /* We are not sure whether the child will become orphan or not.
         * Record it in the LFSCK trace file for further checking in
         * the second-stage scanning. */
-       if (!update && !dec && rc == 0)
-               lfsck_namespace_trace_update(env, com, cfid,
+       if (!update && !dec && child && rc == 0)
+               lfsck_namespace_trace_update(env, com, &cfid,
                                             LNTF_CHECK_LINKEA, true);
 
 unlock1:
@@ -2140,7 +2327,7 @@ log:
               "entry for: parent "DFID", child "DFID", name %s, type "
               "in name entry %o, type claimed by child %o. repair it "
               "by %s with new name2 %s: rc = %d\n",
-              lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
+              lfsck_lfsck2name(lfsck), PFID(pfid), PFID(&cfid),
               name, type, update ? lfsck_object_type(child) : 0,
               update ? "updating" : "removing", name2, rc);
 
@@ -2232,7 +2419,7 @@ static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
        dt_delete(env, obj, (const struct dt_key *)dotdot, th);
 
        rc = dt_insert(env, obj, (const struct dt_rec *)rec,
-                      (const struct dt_key *)dotdot, th, 1);
+                      (const struct dt_key *)dotdot, th);
        if (rc != 0)
                GOTO(unlock, rc);
 
@@ -2412,7 +2599,7 @@ lfsck_namespace_dsd_single(const struct lu_env *env,
 
 lost_parent:
                lmv = &info->lti_lmv;
-               rc = lfsck_read_stripe_lmv(env, child, lmv);
+               rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
                if (rc != 0 && rc != -ENODATA)
                        GOTO(out, rc);
 
@@ -2441,7 +2628,7 @@ lost_parent:
                if (rc >= 0) {
                        /* Add the missing name entry to the parent. */
                        rc = lfsck_namespace_insert_normal(env, com, parent,
-                                                       child, cname->ln_name);
+                                                          child, cname);
                        if (unlikely(rc == -EEXIST)) {
                                /* Unfortunately, someone reused the name
                                 * under the parent by race. So we have
@@ -2479,8 +2666,7 @@ lost_parent:
                GOTO(out, rc);
        }
 
-       rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
-                      (const struct dt_key *)cname->ln_name);
+       rc = dt_lookup_dir(env, parent, cname->ln_name, &tfid);
        if (rc == -ENOENT) {
                /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
                 * has ever tried to verify some remote MDT-object that resides
@@ -2499,7 +2685,8 @@ lost_parent:
                }
 
                lfsck_ibits_unlock(lh, LCK_EX);
-               rc = lfsck_namespace_check_name(env, parent, child, cname);
+               rc = lfsck_namespace_check_name(env, lfsck, parent, child,
+                                               cname);
                if (rc == -ENOENT)
                        goto lost_parent;
 
@@ -2523,7 +2710,7 @@ lost_parent:
 
                /* Add the missing name entry back to the namespace. */
                rc = lfsck_namespace_insert_normal(env, com, parent, child,
-                                                  cname->ln_name);
+                                                  cname);
                if (unlikely(rc == -ESTALE))
                        /* It may happen when the remote object has been
                         * removed, but the local MDT is not aware of that. */
@@ -2688,7 +2875,8 @@ 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 && !fid_is_zero(pfid) &&
+                   !lu_fid_eq(pfid, &tfid) && once) {
                        linkea_next_entry(ldata);
                        continue;
                }
@@ -2718,8 +2906,7 @@ again:
                        continue;
                }
 
-               rc = dt_lookup(env, parent, (struct dt_rec *)&tfid,
-                              (const struct dt_key *)cname->ln_name);
+               rc = dt_lookup_dir(env, parent, cname->ln_name, &tfid);
                *pfid2 = *lfsck_dto2fid(parent);
                if (rc == -ENOENT) {
                        lfsck_object_put(env, parent);
@@ -2739,7 +2926,7 @@ again:
                         * directory contains the specified child, but such
                         * parent does not match the dotdot name entry, then
                         * trust the linkEA. */
-                       if (!lu_fid_eq(pfid, pfid2)) {
+                       if (!fid_is_zero(pfid) && !lu_fid_eq(pfid, pfid2)) {
                                *type = LNIT_UNMATCHED_PAIRS;
                                rc = lfsck_namespace_repair_unmatched_pairs(env,
                                                com, child, pfid2, cname);
@@ -3049,11 +3236,8 @@ static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
 
        if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
            !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
-               CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the"
-                      "the namespace LFSCK, then the LFSCK cannot guarantee"
-                      "all the name entries have been verified in first-stage"
-                      "scanning. So have to skip orphan related handling for"
-                      "the directory object "DFID" with remote name entry\n",
+               CDEBUG(D_LFSCK,
+                      "%s: some MDT(s) maybe NOT take part in the the namespace LFSCK, then the LFSCK cannot guarantee all the name entries have been verified in first-stage scanning. So have to skip orphan related handling for the directory object "DFID" with remote name entry\n",
                       lfsck_lfsck2name(lfsck), PFID(cfid));
 
                RETURN(0);
@@ -3081,8 +3265,7 @@ lock:
                GOTO(out, rc = 0);
        }
 
-       rc = dt_lookup(env, child, (struct dt_rec *)pfid,
-                      (const struct dt_key *)dotdot);
+       rc = dt_lookup_dir(env, child, dotdot, pfid);
        if (rc != 0) {
                if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
                        dt_read_unlock(env, child);
@@ -3222,11 +3405,6 @@ out:
        return rc;
 }
 
-#define lfsck_time_before(a, b)                \
-       (typecheck(__u32, a) &&         \
-        typecheck(__u32, b) &&         \
-        ((int)(a) - (int)(b) < 0))
-
 static inline bool
 lfsck_namespace_linkea_stale_overflow(struct linkea_data *ldata,
                                      struct lfsck_namespace *ns)
@@ -3235,8 +3413,7 @@ lfsck_namespace_linkea_stale_overflow(struct linkea_data *ldata,
         * local time based, so need NOT to care about clock drift
         * among the servers. */
        return ldata->ld_leh->leh_overflow_time &&
-              lfsck_time_before(ldata->ld_leh->leh_overflow_time,
-                                ns->ln_time_latest_reset);
+              ldata->ld_leh->leh_overflow_time < ns->ln_time_latest_reset;
 }
 
 /**
@@ -3334,6 +3511,146 @@ log:
 }
 
 /**
+ * Verify the object's agent entry.
+ *
+ * If the object claims to have agent entry but the linkEA does not contain
+ * remote parent, then remove the agent entry. Otherwise, if the object has
+ * no agent entry but its linkEA contains remote parent, then will generate
+ * agent entry for it.
+ *
+ * \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
+ *
+ * \retval             positive number for repaired cases
+ * \retval             0 if nothing to be repaired
+ * \retval             negative error number on failure
+ */
+static int lfsck_namespace_check_agent_entry(const struct lu_env *env,
+                                            struct lfsck_component *com,
+                                            struct dt_object *obj)
+{
+       struct linkea_data ldata = { NULL };
+       struct lfsck_thread_info *info = lfsck_env_info(env);
+       struct lfsck_namespace *ns = com->lc_file_ram;
+       struct lfsck_instance *lfsck = com->lc_lfsck;
+       struct lu_fid *pfid = &info->lti_fid2;
+       struct lu_name *cname = &info->lti_name;
+       struct lu_seq_range *range = &info->lti_range;
+       struct seq_server_site *ss = lfsck_dev_site(lfsck);
+       __u32 idx = lfsck_dev_idx(lfsck);
+       int rc;
+       bool remote = false;
+       ENTRY;
+
+       if (!(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT))
+               RETURN(0);
+
+       rc = lfsck_links_read_with_rec(env, obj, &ldata);
+       if (rc == -ENOENT || rc == -ENODATA)
+               RETURN(0);
+
+       if (rc && rc != -EINVAL)
+               GOTO(out, rc);
+
+       /* We check the agent entry again after verifying the linkEA
+        * successfully. So invalid linkEA should be dryrun mode. */
+       if (rc == -EINVAL || unlikely(!ldata.ld_leh->leh_reccount))
+               RETURN(0);
+
+       linkea_first_entry(&ldata);
+       while (ldata.ld_lee != NULL && !remote) {
+               linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
+                                   cname, pfid);
+               if (!linkea_entry_is_valid(&ldata, cname, pfid))
+                       GOTO(out, rc = 0);
+
+               fld_range_set_mdt(range);
+               rc = fld_server_lookup(env, ss->ss_server_fld,
+                                      fid_seq(pfid), range);
+               if (rc)
+                       GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
+
+               if (range->lsr_index != idx)
+                       remote = true;
+               else
+                       linkea_next_entry(&ldata);
+       }
+
+       if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
+           (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
+               struct dt_device *dev = lfsck_obj2dev(obj);
+               struct linkea_data ldata2 = { NULL };
+               struct lustre_handle lh = { 0 };
+               struct lu_buf linkea_buf;
+               struct thandle *handle;
+
+               if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
+                       GOTO(out, rc = 1);
+
+               rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
+                                     MDS_INODELOCK_UPDATE |
+                                     MDS_INODELOCK_XATTR, LCK_EX);
+               if (rc)
+                       GOTO(out, rc);
+
+               handle = dt_trans_create(env, dev);
+               if (IS_ERR(handle))
+                       GOTO(unlock, rc = PTR_ERR(handle));
+
+               lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
+                              ldata.ld_leh->leh_len);
+               rc = dt_declare_xattr_set(env, obj, &linkea_buf,
+                               XATTR_NAME_LINK, LU_XATTR_REPLACE, handle);
+               if (rc)
+                       GOTO(stop, rc);
+
+               rc = dt_trans_start_local(env, dev, handle);
+               if (rc)
+                       GOTO(stop, rc);
+
+               dt_write_lock(env, obj, 0);
+               rc = lfsck_links_read2_with_rec(env, obj, &ldata2);
+               if (rc) {
+                       if (rc == -ENOENT || rc == -ENODATA)
+                               rc = 0;
+                       GOTO(unlock2, rc);
+               }
+
+               /* If someone changed linkEA by race, then the agent
+                * entry will be updated by lower layer automatically. */
+               if (ldata.ld_leh->leh_len != ldata2.ld_leh->leh_len ||
+                   memcmp(ldata.ld_buf->lb_buf, ldata2.ld_buf->lb_buf,
+                          ldata.ld_leh->leh_len) != 0)
+                       GOTO(unlock2, rc = 0);
+
+               rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK,
+                                 LU_XATTR_REPLACE, handle);
+               if (!rc)
+                       rc = 1;
+
+               GOTO(unlock2, rc);
+
+unlock2:
+               dt_write_unlock(env, obj);
+stop:
+               dt_trans_stop(env, dev, handle);
+unlock:
+               lfsck_ibits_unlock(&lh, LCK_EX);
+       }
+
+       GOTO(out, rc);
+
+out:
+       if (rc > 0)
+               ns->ln_agent_entries_repaired++;
+       if (rc)
+               CDEBUG(D_LFSCK, "%s: repair agent entry for "DFID": rc = %d\n",
+                      lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
+       return rc;
+}
+
+/**
  * Double scan the MDT-object for namespace LFSCK.
  *
  * If the MDT-object contains invalid or repeated linkEA entries, then drop
@@ -3380,6 +3697,8 @@ static int lfsck_namespace_double_scan_one(const struct lu_env *env,
        if (S_ISDIR(lfsck_object_type(child))) {
                dt_read_unlock(env, child);
                rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
+               if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
+                       rc = lfsck_namespace_check_agent_entry(env, com, child);
 
                RETURN(rc);
        }
@@ -3495,7 +3814,7 @@ lost_parent:
 
                                /* Add the missing name entry to the parent. */
                                rc = lfsck_namespace_insert_normal(env, com,
-                                               parent, child, cname->ln_name);
+                                                       parent, child, cname);
                                if (unlikely(rc == -EEXIST))
                                        /* Unfortunately, someone reused the
                                         * name under the parent by race. So we
@@ -3536,8 +3855,7 @@ lost_parent:
                        continue;
                }
 
-               rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
-                              (const struct dt_key *)cname->ln_name);
+               rc = dt_lookup_dir(env, parent, cname->ln_name, cfid);
                if (rc != 0 && rc != -ENOENT) {
                        lfsck_object_put(env, parent);
 
@@ -3615,7 +3933,8 @@ lost_parent:
                        GOTO(out, rc = 0);
                }
 
-               rc = lfsck_namespace_check_name(env, parent, child, cname);
+               rc = lfsck_namespace_check_name(env, lfsck, parent, child,
+                                               cname);
                if (rc == -ENOENT)
                        goto lost_parent;
 
@@ -3641,7 +3960,7 @@ lost_parent:
 
                /* Add the missing name entry back to the namespace. */
                rc = lfsck_namespace_insert_normal(env, com, parent, child,
-                                                  cname->ln_name);
+                                                  cname);
                if (unlikely(rc == -ESTALE))
                        /* It may happen when the remote object has been
                         * removed, but the local MDT is not aware of that. */
@@ -3739,6 +4058,9 @@ out:
                        rc = 1;
        }
 
+       if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
+               rc = lfsck_namespace_check_agent_entry(env, com, child);
+
        return rc;
 }
 
@@ -3746,78 +4068,85 @@ static void lfsck_namespace_dump_statistics(struct seq_file *m,
                                            struct lfsck_namespace *ns,
                                            __u64 checked_phase1,
                                            __u64 checked_phase2,
-                                           __u32 time_phase1,
-                                           __u32 time_phase2)
+                                           time64_t time_phase1,
+                                           time64_t time_phase2, bool dryrun)
 {
+       const char *postfix = dryrun ? "inconsistent" : "repaired";
+
        seq_printf(m, "checked_phase1: %llu\n"
                   "checked_phase2: %llu\n"
-                  "updated_phase1: %llu\n"
-                  "updated_phase2: %llu\n"
+                  "%s_phase1: %llu\n"
+                  "%s_phase2: %llu\n"
                   "failed_phase1: %llu\n"
                   "failed_phase2: %llu\n"
                   "directories: %llu\n"
-                  "dirent_repaired: %llu\n"
-                  "linkea_repaired: %llu\n"
-                  "nlinks_repaired: %llu\n"
+                  "dirent_%s: %llu\n"
+                  "linkea_%s: %llu\n"
+                  "nlinks_%s: %llu\n"
                   "multiple_linked_checked: %llu\n"
-                  "multiple_linked_repaired: %llu\n"
+                  "multiple_linked_%s: %llu\n"
                   "unknown_inconsistency: %llu\n"
-                  "unmatched_pairs_repaired: %llu\n"
-                  "dangling_repaired: %llu\n"
-                  "multiple_referenced_repaired: %llu\n"
-                  "bad_file_type_repaired: %llu\n"
-                  "lost_dirent_repaired: %llu\n"
+                  "unmatched_pairs_%s: %llu\n"
+                  "dangling_%s: %llu\n"
+                  "multiple_referenced_%s: %llu\n"
+                  "bad_file_type_%s: %llu\n"
+                  "lost_dirent_%s: %llu\n"
                   "local_lost_found_scanned: %llu\n"
                   "local_lost_found_moved: %llu\n"
                   "local_lost_found_skipped: %llu\n"
                   "local_lost_found_failed: %llu\n"
                   "striped_dirs_scanned: %llu\n"
-                  "striped_dirs_repaired: %llu\n"
+                  "striped_dirs_%s: %llu\n"
                   "striped_dirs_failed: %llu\n"
                   "striped_dirs_disabled: %llu\n"
                   "striped_dirs_skipped: %llu\n"
                   "striped_shards_scanned: %llu\n"
-                  "striped_shards_repaired: %llu\n"
+                  "striped_shards_%s: %llu\n"
                   "striped_shards_failed: %llu\n"
                   "striped_shards_skipped: %llu\n"
-                  "name_hash_repaired: %llu\n"
-                  "linkea_overflow_cleared: %llu\n"
+                  "name_hash_%s: %llu\n"
+                  "linkea_overflow_%s: %llu\n"
+                  "agent_entries_%s: %llu\n"
                   "success_count: %u\n"
-                  "run_time_phase1: %u seconds\n"
-                  "run_time_phase2: %u seconds\n",
+                  "run_time_phase1: %lld seconds\n"
+                  "run_time_phase2: %lld seconds\n",
                   checked_phase1,
                   checked_phase2,
+                  dryrun ? "inconsistent" : "updated",
                   ns->ln_items_repaired,
+                  dryrun ? "inconsistent" : "updated",
                   ns->ln_objs_repaired_phase2,
                   ns->ln_items_failed,
                   ns->ln_objs_failed_phase2,
                   ns->ln_dirs_checked,
-                  ns->ln_dirent_repaired,
-                  ns->ln_linkea_repaired,
-                  ns->ln_objs_nlink_repaired,
+                  postfix, ns->ln_dirent_repaired,
+                  postfix, ns->ln_linkea_repaired,
+                  postfix, ns->ln_objs_nlink_repaired,
                   ns->ln_mul_linked_checked,
-                  ns->ln_mul_linked_repaired,
+                  postfix, ns->ln_mul_linked_repaired,
                   ns->ln_unknown_inconsistency,
-                  ns->ln_unmatched_pairs_repaired,
-                  ns->ln_dangling_repaired,
-                  ns->ln_mul_ref_repaired,
-                  ns->ln_bad_type_repaired,
-                  ns->ln_lost_dirent_repaired,
+                  postfix, ns->ln_unmatched_pairs_repaired,
+                  postfix, ns->ln_dangling_repaired,
+                  postfix, ns->ln_mul_ref_repaired,
+                  postfix, ns->ln_bad_type_repaired,
+                  postfix, ns->ln_lost_dirent_repaired,
                   ns->ln_local_lpf_scanned,
                   ns->ln_local_lpf_moved,
                   ns->ln_local_lpf_skipped,
                   ns->ln_local_lpf_failed,
                   ns->ln_striped_dirs_scanned,
-                  ns->ln_striped_dirs_repaired,
+                  postfix, ns->ln_striped_dirs_repaired,
                   ns->ln_striped_dirs_failed,
                   ns->ln_striped_dirs_disabled,
                   ns->ln_striped_dirs_skipped,
                   ns->ln_striped_shards_scanned,
-                  ns->ln_striped_shards_repaired,
+                  postfix, ns->ln_striped_shards_repaired,
                   ns->ln_striped_shards_failed,
                   ns->ln_striped_shards_skipped,
-                  ns->ln_name_hash_repaired,
+                  postfix, ns->ln_name_hash_repaired,
+                  dryrun ? "inconsistent" : "cleared",
                   ns->ln_linkea_overflow_cleared,
+                  postfix, ns->ln_agent_entries_repaired,
                   ns->ln_success_count,
                   time_phase1,
                   time_phase2);
@@ -3897,7 +4226,7 @@ static int lfsck_namespace_reset(const struct lu_env *env,
                memset(ns, 0, sizeof(*ns));
        } else {
                __u32 count = ns->ln_success_count;
-               __u64 last_time = ns->ln_time_last_complete;
+               time64_t last_time = ns->ln_time_last_complete;
 
                memset(ns, 0, sizeof(*ns));
                ns->ln_success_count = count;
@@ -3905,7 +4234,7 @@ static int lfsck_namespace_reset(const struct lu_env *env,
        }
        ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
        ns->ln_status = LS_INIT;
-       ns->ln_time_latest_reset = cfs_time_current_sec();
+       ns->ln_time_latest_reset = ktime_get_real_seconds();
 
        rc = lfsck_load_one_trace_file(env, com, root, &com->lc_obj,
                                       &dt_lfsck_namespace_features,
@@ -3918,7 +4247,7 @@ static int lfsck_namespace_reset(const struct lu_env *env,
        if (rc != 0)
                GOTO(out, rc);
 
-       lad->lad_incomplete = 0;
+       clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
        CFS_RESET_BITMAP(lad->lad_bitmap);
 
        rc = lfsck_namespace_store(env, com);
@@ -3958,14 +4287,19 @@ static void lfsck_namespace_close_dir(const struct lu_env *env,
        struct lfsck_instance           *lfsck  = com->lc_lfsck;
        struct lfsck_lmv                *llmv   = lfsck->li_lmv;
        struct lfsck_namespace_req      *lnr;
-       __u32                            size   =
-                               sizeof(*lnr) + LFSCK_TMPBUF_LEN;
-       bool                             wakeup = false;
+       struct lu_attr *la = &lfsck_env_info(env)->lti_la2;
+       __u32 size = sizeof(*lnr) + LFSCK_TMPBUF_LEN;
+       int rc;
+       bool wakeup = false;
        ENTRY;
 
        if (llmv == NULL)
                RETURN_EXIT;
 
+       rc = dt_attr_get(env, lfsck->li_obj_dir, la);
+       if (rc)
+               RETURN_EXIT;
+
        OBD_ALLOC(lnr, size);
        if (lnr == NULL) {
                ns->ln_striped_dirs_skipped++;
@@ -3974,7 +4308,7 @@ static void lfsck_namespace_close_dir(const struct lu_env *env,
        }
 
        lso = lfsck_assistant_object_init(env, lfsck_dto2fid(lfsck->li_obj_dir),
-                       NULL, lfsck->li_pos_current.lp_oit_cookie, true);
+                       la, lfsck->li_pos_current.lp_oit_cookie, true);
        if (IS_ERR(lso)) {
                OBD_FREE(lnr, size);
                ns->ln_striped_dirs_skipped++;
@@ -3990,9 +4324,12 @@ static void lfsck_namespace_close_dir(const struct lu_env *env,
        lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
        lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
        lnr->lnr_size = size;
+       lnr->lnr_type = lso->lso_attr.la_mode;
 
        spin_lock(&lad->lad_lock);
-       if (lad->lad_assistant_status < 0) {
+       if (lad->lad_assistant_status < 0 ||
+           unlikely(!thread_is_running(&lfsck->li_thread) ||
+                    !thread_is_running(&lad->lad_thread))) {
                spin_unlock(&lad->lad_lock);
                lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
                ns->ln_striped_dirs_skipped++;
@@ -4059,9 +4396,9 @@ static int lfsck_namespace_checkpoint(const struct lu_env *env,
                ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
        } else {
                ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
-               ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
-                               HALF_SEC - lfsck->li_time_last_checkpoint);
-               ns->ln_time_last_checkpoint = cfs_time_current_sec();
+               ns->ln_run_time_phase1 += ktime_get_seconds() -
+                                         lfsck->li_time_last_checkpoint;
+               ns->ln_time_last_checkpoint = ktime_get_real_seconds();
                ns->ln_items_checked += com->lc_new_checked;
                com->lc_new_checked = 0;
        }
@@ -4103,7 +4440,7 @@ static int lfsck_namespace_prep(const struct lu_env *env,
        }
 
        down_write(&com->lc_sem);
-       ns->ln_time_latest_start = cfs_time_current_sec();
+       ns->ln_time_latest_start = ktime_get_real_seconds();
        spin_lock(&lfsck->li_lock);
 
        if (ns->ln_flags & LF_SCANNED_ONCE) {
@@ -4184,19 +4521,32 @@ static int lfsck_namespace_exec_oit(const struct lu_env *env,
                                    struct lfsck_component *com,
                                    struct dt_object *obj)
 {
-       struct lfsck_thread_info *info  = lfsck_env_info(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_fid            *pfid  = &info->lti_fid2;
-       struct lu_name           *cname = &info->lti_name;
-       struct lu_seq_range      *range = &info->lti_range;
-       struct seq_server_site   *ss    = lfsck_dev_site(lfsck);
-       struct linkea_data        ldata = { NULL };
-       __u32                     idx   = lfsck_dev_idx(lfsck);
-       int                       rc;
+       struct lfsck_thread_info *info = lfsck_env_info(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_fid *pfid = &info->lti_fid2;
+       struct lu_name *cname = &info->lti_name;
+       struct lu_seq_range *range = &info->lti_range;
+       struct seq_server_site *ss = lfsck_dev_site(lfsck);
+       struct linkea_data ldata = { NULL };
+       __u32 idx = lfsck_dev_idx(lfsck);
+       struct lu_attr la = { .la_valid = 0 };
+       bool remote = false;
+       int rc;
        ENTRY;
 
+       rc = dt_attr_get(env, obj, &la);
+       if (unlikely(rc || (la.la_valid & LA_FLAGS &&
+                           la.la_flags & LUSTRE_ORPHAN_FL))) {
+               CDEBUG(D_INFO,
+                      "%s: skip orphan "DFID", %llx/%x: rc = %d\n",
+                      lfsck_lfsck2name(lfsck), PFID(fid),
+                      la.la_valid, la.la_flags, rc);
+
+               return rc;
+       }
+
        rc = lfsck_links_read(env, obj, &ldata);
        if (rc == -ENOENT)
                GOTO(out, rc = 0);
@@ -4221,14 +4571,47 @@ static int lfsck_namespace_exec_oit(const struct lu_env *env,
                GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
        }
 
+       if (rc && rc != -ENODATA)
+               GOTO(out, rc);
+
        if (rc == -ENODATA || unlikely(!ldata.ld_leh->leh_reccount)) {
                rc = lfsck_namespace_check_for_double_scan(env, com, obj);
 
                GOTO(out, rc);
        }
 
-       if (rc != 0)
-               GOTO(out, rc);
+       linkea_first_entry(&ldata);
+       while (ldata.ld_lee != NULL) {
+               linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
+                                   cname, pfid);
+               if (!fid_is_sane(pfid)) {
+                       rc = lfsck_namespace_trace_update(env, com, fid,
+                                                 LNTF_CHECK_PARENT, true);
+               } else if (!linkea_entry_is_valid(&ldata, cname, pfid)) {
+                       GOTO(out, rc);
+               } else {
+                       fld_range_set_mdt(range);
+                       rc = fld_server_lookup(env, ss->ss_server_fld,
+                                              fid_seq(pfid), range);
+                       if ((rc == -ENOENT) ||
+                           (!rc && range->lsr_index != idx)) {
+                               remote = true;
+                               break;
+                       }
+               }
+               if (rc)
+                       GOTO(out, rc);
+
+               linkea_next_entry(&ldata);
+       }
+
+       if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
+           (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
+               rc = lfsck_namespace_trace_update(env, com, fid,
+                                                 LNTF_CHECK_AGENT_ENTRY, true);
+               if (rc)
+                       GOTO(out, rc);
+       }
 
        /* Record multiple-linked object. */
        if (ldata.ld_leh->leh_reccount > 1) {
@@ -4238,23 +4621,11 @@ static int lfsck_namespace_exec_oit(const struct lu_env *env,
                GOTO(out, rc);
        }
 
-       linkea_first_entry(&ldata);
-       linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
-       if (!fid_is_sane(pfid)) {
+       if (remote)
                rc = lfsck_namespace_trace_update(env, com, fid,
-                                                 LNTF_CHECK_PARENT, true);
-       } else {
-               fld_range_set_mdt(range);
-               rc = fld_local_lookup(env, ss->ss_server_fld,
-                                     fid_seq(pfid), range);
-               if ((rc == -ENOENT) ||
-                   (rc == 0 && range->lsr_index != idx))
-                       rc = lfsck_namespace_trace_update(env, com, fid,
-                                               LNTF_CHECK_LINKEA, true);
-               else
-                       rc = lfsck_namespace_check_for_double_scan(env, com,
-                                                                  obj);
-       }
+                                                 LNTF_CHECK_LINKEA, true);
+       else
+               rc = lfsck_namespace_check_for_double_scan(env, com, obj);
 
        GOTO(out, rc);
 
@@ -4280,17 +4651,15 @@ static int lfsck_namespace_exec_dir(const struct lu_env *env,
        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,
-                    lad->lad_prefetched < bk->lb_async_windows ||
-                    !thread_is_running(mthread) ||
-                    thread_is_stopped(athread),
-                    &lwi);
+       wait_event_idle(mthread->t_ctl_waitq,
+                       lad->lad_prefetched < bk->lb_async_windows ||
+                       !thread_is_running(mthread) ||
+                       !thread_is_running(athread));
 
-       if (unlikely(!thread_is_running(mthread)) ||
-                    thread_is_stopped(athread))
+       if (unlikely(!thread_is_running(mthread) ||
+                    !thread_is_running(athread)))
                return 0;
 
        if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
@@ -4305,7 +4674,9 @@ static int lfsck_namespace_exec_dir(const struct lu_env *env,
        }
 
        spin_lock(&lad->lad_lock);
-       if (lad->lad_assistant_status < 0) {
+       if (lad->lad_assistant_status < 0 ||
+           unlikely(!thread_is_running(mthread) ||
+                    !thread_is_running(athread))) {
                spin_unlock(&lad->lad_lock);
                lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
                return lad->lad_assistant_status;
@@ -4367,9 +4738,9 @@ static int lfsck_namespace_post(const struct lu_env *env,
        spin_unlock(&lfsck->li_lock);
 
        if (!init) {
-               ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
-                               HALF_SEC - lfsck->li_time_last_checkpoint);
-               ns->ln_time_last_checkpoint = cfs_time_current_sec();
+               ns->ln_run_time_phase1 += ktime_get_seconds() -
+                                         lfsck->li_time_last_checkpoint;
+               ns->ln_time_last_checkpoint = ktime_get_real_seconds();
                ns->ln_items_checked += com->lc_new_checked;
                com->lc_new_checked = 0;
        }
@@ -4420,22 +4791,21 @@ lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
 
        if (ns->ln_status == LS_SCANNING_PHASE1) {
                struct lfsck_position pos;
-               cfs_duration_t duration = cfs_time_current() -
-                                         lfsck->li_time_last_checkpoint;
-               __u64 checked = ns->ln_items_checked + com->lc_new_checked;
-               __u64 speed = checked;
-               __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);
+               time64_t duration = ktime_get_seconds() -
+                                   lfsck->li_time_last_checkpoint;
+               u64 checked = ns->ln_items_checked + com->lc_new_checked;
+               u64 speed = checked;
+               u64 new_checked = com->lc_new_checked;
+               time64_t rtime = ns->ln_run_time_phase1 + duration;
 
                if (duration != 0)
-                       do_div(new_checked, duration);
+                       new_checked = div64_s64(new_checked, duration);
 
                if (rtime != 0)
-                       do_div(speed, rtime);
+                       speed = div64_s64(speed, rtime);
 
-               lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0);
+               lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0,
+                                               bk->lb_param & LPF_DRYRUN);
                seq_printf(m, "average_speed_phase1: %llu items/sec\n"
                           "average_speed_phase2: N/A\n"
                           "average_speed_total: %llu items/sec\n"
@@ -4478,38 +4848,37 @@ lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
 
                lfsck_pos_dump(m, &pos, "current_position");
        } else if (ns->ln_status == LS_SCANNING_PHASE2) {
-               cfs_duration_t duration = cfs_time_current() -
-                                         com->lc_time_last_checkpoint;
+               time64_t duration = ktime_get_seconds() -
+                                   com->lc_time_last_checkpoint;
                __u64 checked = ns->ln_objs_checked_phase2 +
                                com->lc_new_checked;
                __u64 speed1 = ns->ln_items_checked;
                __u64 speed2 = checked;
                __u64 speed0 = speed1 + speed2;
-               __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);
-               __u32 time0 = ns->ln_run_time_phase1 + rtime;
+               __u64 new_checked = com->lc_new_checked;
+               time64_t rtime = ns->ln_run_time_phase2 + duration;
+               time64_t time0 = ns->ln_run_time_phase1 + rtime;
 
                if (duration != 0)
-                       do_div(new_checked, duration);
+                       new_checked = div64_s64(new_checked, duration);
 
                if (ns->ln_run_time_phase1 != 0)
-                       do_div(speed1, ns->ln_run_time_phase1);
+                       speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
                else if (ns->ln_items_checked != 0)
                        time0++;
 
                if (rtime != 0)
-                       do_div(speed2, rtime);
+                       speed2 = div64_s64(speed2, rtime);
                else if (checked != 0)
                        time0++;
 
                if (time0 != 0)
-                       do_div(speed0, time0);
+                       speed0 = div64_s64(speed0, time0);
 
                lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
                                                checked,
-                                               ns->ln_run_time_phase1, rtime);
+                                               ns->ln_run_time_phase1, rtime,
+                                               bk->lb_param & LPF_DRYRUN);
                seq_printf(m, "average_speed_phase1: %llu items/sec\n"
                           "average_speed_phase2: %llu objs/sec\n"
                           "average_speed_total: %llu items/sec\n"
@@ -4525,25 +4894,26 @@ lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
                __u64 speed1 = ns->ln_items_checked;
                __u64 speed2 = ns->ln_objs_checked_phase2;
                __u64 speed0 = speed1 + speed2;
-               __u32 time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
+               time64_t time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
 
                if (ns->ln_run_time_phase1 != 0)
-                       do_div(speed1, ns->ln_run_time_phase1);
+                       speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
                else if (ns->ln_items_checked != 0)
                        time0++;
 
                if (ns->ln_run_time_phase2 != 0)
-                       do_div(speed2, ns->ln_run_time_phase2);
+                       speed2 = div64_s64(speed2, ns->ln_run_time_phase2);
                else if (ns->ln_objs_checked_phase2 != 0)
                        time0++;
 
                if (time0 != 0)
-                       do_div(speed0, time0);
+                       speed0 = div64_s64(speed0, time0);
 
                lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
                                                ns->ln_objs_checked_phase2,
                                                ns->ln_run_time_phase1,
-                                               ns->ln_run_time_phase2);
+                                               ns->ln_run_time_phase2,
+                                               bk->lb_param & LPF_DRYRUN);
                seq_printf(m, "average_speed_phase1: %llu items/sec\n"
                           "average_speed_phase2: %llu objs/sec\n"
                           "average_speed_total: %llu items/sec\n"
@@ -4844,7 +5214,7 @@ static int lfsck_namespace_query(const struct lu_env *env,
        return rc;
 }
 
-static struct lfsck_operations lfsck_namespace_ops = {
+static const struct lfsck_operations lfsck_namespace_ops = {
        .lfsck_reset            = lfsck_namespace_reset,
        .lfsck_fail             = lfsck_namespace_fail,
        .lfsck_close_dir        = lfsck_namespace_close_dir,
@@ -4894,26 +5264,28 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env,
                                    struct dt_object *child,
                                    struct lfsck_namespace_req *lnr)
 {
-       struct lfsck_thread_info        *info   = lfsck_env_info(env);
-       struct lu_attr                  *la     = &info->lti_la;
-       struct dt_allocation_hint       *hint   = &info->lti_hint;
-       struct dt_object_format         *dof    = &info->lti_dof;
-       struct dt_insert_rec            *rec    = &info->lti_dt_rec;
-       struct lmv_mds_md_v1            *lmv2   = &info->lti_lmv2;
-       const struct lu_name            *cname;
-       const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
-       const struct lu_fid             *cfid   = lfsck_dto2fid(child);
-       struct linkea_data               ldata  = { NULL };
-       struct lfsck_lock_handle        *llh    = &info->lti_llh;
-       struct lu_buf                    linkea_buf;
-       struct lu_buf                    lmv_buf;
-       struct lfsck_instance           *lfsck  = com->lc_lfsck;
-       struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
-       struct dt_device                *dev    = lfsck->li_next;
-       struct thandle                  *th     = NULL;
-       int                              rc     = 0;
-       __u16                            type   = lnr->lnr_type;
-       bool                             create;
+       struct lfsck_thread_info *info = lfsck_env_info(env);
+       struct lu_attr *la = &info->lti_la;
+       struct dt_allocation_hint *hint = &info->lti_hint;
+       struct dt_object_format *dof = &info->lti_dof;
+       struct dt_insert_rec *rec = &info->lti_dt_rec;
+       struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
+       const struct lu_name *cname;
+       const struct lu_fid *pfid = lfsck_dto2fid(parent);
+       const struct lu_fid *cfid = lfsck_dto2fid(child);
+       struct linkea_data ldata = { NULL };
+       struct lfsck_lock_handle *llh = &info->lti_llh;
+       struct lustre_handle rlh = { 0 };
+       struct lustre_handle clh = { 0 };
+       struct lu_buf linkea_buf;
+       struct lu_buf lmv_buf;
+       struct lfsck_instance *lfsck = com->lc_lfsck;
+       struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
+       struct dt_device *dev = lfsck->li_next;
+       struct thandle *th = NULL;
+       int rc = 0;
+       __u16 type = lnr->lnr_type;
+       bool create;
        ENTRY;
 
        cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
@@ -4945,7 +5317,7 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env,
                GOTO(log, rc);
 
        rc = lfsck_lock(env, lfsck, parent, lnr->lnr_name, llh,
-                       MDS_INODELOCK_UPDATE, LCK_PR);
+                       MDS_INODELOCK_UPDATE, LCK_PW);
        if (rc != 0)
                GOTO(log, rc);
 
@@ -4953,17 +5325,40 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env,
        if (rc != 0)
                GOTO(log, rc);
 
+       if (dt_object_remote(child)) {
+               rc = lfsck_remote_lookup_lock(env, lfsck, parent, child, &rlh,
+                                             LCK_EX);
+               if (rc != 0)
+                       GOTO(log, rc);
+       }
+
+       rc = lfsck_ibits_lock(env, lfsck, child, &clh,
+                             MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
+                             MDS_INODELOCK_XATTR, LCK_EX);
+       if (rc != 0)
+               GOTO(unlock_remote_lookup, rc);
+
        /* Set the ctime as zero, then others can know it is created for
         * repairing dangling name entry by LFSCK. And if the LFSCK made
         * wrong decision and the real MDT-object has been found later,
         * then the LFSCK has chance to fix the incosistency properly. */
        memset(la, 0, sizeof(*la));
-       la->la_mode = (type & S_IFMT) | 0600;
-       la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
-                       LA_ATIME | LA_MTIME | LA_CTIME;
-
-       child->do_ops->do_ah_init(env, hint, parent, child,
-                                la->la_mode & S_IFMT);
+       if (S_ISDIR(type))
+               la->la_mode = (type & S_IFMT) | 0700;
+       else
+               la->la_mode = (type & S_IFMT) | 0600;
+       la->la_valid = LA_TYPE | LA_MODE | LA_CTIME;
+
+       /*
+        * if it's directory, skip do_ah_init() to create a plain directory
+        * because it may have shards already, which will be inserted back
+        * later, besides, it may be remote, and creating stripe directory
+        * remotely is not supported.
+        */
+       if (S_ISREG(type))
+               child->do_ops->do_ah_init(env, hint,  parent, child, type);
+       else if (S_ISDIR(type))
+               child->do_ops->do_ah_init(env, hint,  NULL, child, type);
 
        memset(dof, 0, sizeof(*dof));
        dof->dof_type = dt_mode_to_dft(type);
@@ -4973,7 +5368,7 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env,
 
        th = dt_trans_create(env, dev);
        if (IS_ERR(th))
-               GOTO(log, rc = PTR_ERR(th));
+               GOTO(unlock_child, rc = PTR_ERR(th));
 
        /* 1a. create child. */
        rc = dt_declare_create(env, child, la, hint, dof, th);
@@ -5037,6 +5432,21 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env,
        if (rc != 0)
                GOTO(stop, rc);
 
+       /* 7a. if child is remote, delete and insert to generate local agent */
+       if (dt_object_remote(child)) {
+               rc = dt_declare_delete(env, parent,
+                                      (const struct dt_key *)lnr->lnr_name,
+                                      th);
+               if (rc)
+                       GOTO(stop, rc);
+
+               rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
+                                      (const struct dt_key *)lnr->lnr_name,
+                                      th);
+               if (rc)
+                       GOTO(stop, rc);
+       }
+
        rc = dt_trans_start_local(env, dev, th);
        if (rc != 0)
                GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
@@ -5057,14 +5467,14 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env,
                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, 1);
+                              (const struct dt_key *)dot, th);
                if (rc != 0)
                        GOTO(unlock, rc);
 
                /* 4b. insert dotdot into child dir */
                rec->rec_fid = pfid;
                rc = dt_insert(env, child, (const struct dt_rec *)rec,
-                              (const struct dt_key *)dotdot, th, 1);
+                              (const struct dt_key *)dotdot, th);
                if (rc != 0)
                        GOTO(unlock, rc);
 
@@ -5080,6 +5490,23 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env,
        /* 6b. insert linkEA for child. */
        rc = dt_xattr_set(env, child, &linkea_buf,
                          XATTR_NAME_LINK, 0, th);
+       if (rc)
+               GOTO(unlock, rc);
+
+       /* 7b. if child is remote, delete and insert to generate local agent */
+       if (dt_object_remote(child)) {
+               rc = dt_delete(env, parent,
+                              (const struct dt_key *)lnr->lnr_name, th);
+               if (rc)
+                       GOTO(unlock, rc);
+
+               rec->rec_type = type;
+               rec->rec_fid = cfid;
+               rc = dt_insert(env, parent, (const struct dt_rec *)rec,
+                              (const struct dt_key *)lnr->lnr_name, th);
+               if (rc)
+                       GOTO(unlock, rc);
+       }
 
        GOTO(unlock, rc);
 
@@ -5089,6 +5516,11 @@ unlock:
 stop:
        dt_trans_stop(env, dev, th);
 
+unlock_child:
+       lfsck_ibits_unlock(&clh, LCK_EX);
+unlock_remote_lookup:
+       if (dt_object_remote(child))
+               lfsck_ibits_unlock(&rlh, LCK_EX);
 log:
        lfsck_unlock(llh);
        CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
@@ -5121,7 +5553,7 @@ static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
        const struct lu_name       *cname;
        struct thandle             *handle   = NULL;
        struct lfsck_namespace_req *lnr      =
-                       container_of0(lar, struct lfsck_namespace_req, lnr_lar);
+               container_of(lar, struct lfsck_namespace_req, lnr_lar);
        struct dt_object           *dir      = NULL;
        struct dt_object           *obj      = NULL;
        struct lfsck_assistant_object *lso   = lar->lar_parent;
@@ -5155,11 +5587,9 @@ static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
                repaired = true;
        }
 
-       if (unlikely(fid_is_zero(&lnr->lnr_fid))) {
-               if (strcmp(lnr->lnr_name, dotdot) != 0)
-                       LBUG();
-               else
-                       rc = lfsck_namespace_trace_update(env, com, pfid,
+       if (unlikely(fid_is_zero(&lnr->lnr_fid) &&
+                    strcmp(lnr->lnr_name, dotdot) == 0)) {
+               rc = lfsck_namespace_trace_update(env, com, pfid,
                                                LNTF_CHECK_PARENT, true);
 
                GOTO(out, rc);
@@ -5749,7 +6179,7 @@ static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
 
        /* b5. insert child's FID into the LFSCK trace file. */
        rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
-                      (const struct dt_key *)key, th, 1);
+                      (const struct dt_key *)key, th);
 
        GOTO(stop, rc = (rc == 0 ? 1 : rc));
 
@@ -6146,23 +6576,22 @@ checkpoint:
                if (rc < 0 && bk->lb_param & LPF_FAILOUT)
                        GOTO(put, rc);
 
-               if (unlikely(cfs_time_beforeq(com->lc_time_next_checkpoint,
-                                             cfs_time_current())) &&
+               if (unlikely(com->lc_time_next_checkpoint <=
+                            ktime_get_seconds()) &&
                    com->lc_new_checked != 0) {
                        down_write(&com->lc_sem);
-                       ns->ln_run_time_phase2 +=
-                               cfs_duration_sec(cfs_time_current() +
-                               HALF_SEC - com->lc_time_last_checkpoint);
-                       ns->ln_time_last_checkpoint = cfs_time_current_sec();
+                       ns->ln_run_time_phase2 += ktime_get_seconds() -
+                                                 com->lc_time_last_checkpoint;
+                       ns->ln_time_last_checkpoint = ktime_get_real_seconds();
                        ns->ln_objs_checked_phase2 += com->lc_new_checked;
                        com->lc_new_checked = 0;
                        lfsck_namespace_store(env, com);
                        up_write(&com->lc_sem);
 
-                       com->lc_time_last_checkpoint = cfs_time_current();
+                       com->lc_time_last_checkpoint = ktime_get_seconds();
                        com->lc_time_next_checkpoint =
                                com->lc_time_last_checkpoint +
-                               cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
+                               LFSCK_CHECKPOINT_INTERVAL;
                }
 
                lfsck_control_speed_by_self(com);
@@ -6213,9 +6642,9 @@ static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
 
        com->lc_new_checked = 0;
        com->lc_new_scanned = 0;
-       com->lc_time_last_checkpoint = cfs_time_current();
+       com->lc_time_last_checkpoint = ktime_get_seconds();
        com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
-                               cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
+                                      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,
@@ -6260,9 +6689,9 @@ static int lfsck_namespace_double_scan_result(const struct lu_env *env,
        struct lfsck_namespace  *ns     = com->lc_file_ram;
 
        down_write(&com->lc_sem);
-       ns->ln_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
-                                 HALF_SEC - com->lc_time_last_checkpoint);
-       ns->ln_time_last_checkpoint = cfs_time_current_sec();
+       ns->ln_run_time_phase2 += ktime_get_seconds() -
+                                 com->lc_time_last_checkpoint;
+       ns->ln_time_last_checkpoint = ktime_get_real_seconds();
        ns->ln_objs_checked_phase2 += com->lc_new_checked;
        com->lc_new_checked = 0;
 
@@ -6343,7 +6772,7 @@ static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
        int                                rc    = 0;
        ENTRY;
 
-       if (!lad->lad_incomplete)
+       if (!test_bit(LAD_INCOMPLETE, &lad->lad_flags))
                RETURN_EXIT;
 
        set = ptlrpc_prep_set();
@@ -6357,7 +6786,8 @@ static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
        down_read(&ltds->ltd_rw_sem);
        cfs_foreach_bit(lad->lad_bitmap, idx) {
                ltd = lfsck_ltd2tgt(ltds, idx);
-               LASSERT(ltd != NULL);
+               if (unlikely(!ltd))
+                       continue;
 
                laia->laia_ltd = ltd;
                rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
@@ -6370,7 +6800,7 @@ static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
        }
        up_read(&ltds->ltd_rw_sem);
 
-       rc = ptlrpc_set_wait(set);
+       rc = ptlrpc_set_wait(env, set);
        ptlrpc_set_destroy(set);
 
        GOTO(out, rc);
@@ -6385,7 +6815,7 @@ out:
        EXIT;
 }
 
-struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
+const struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
        .la_handler_p1          = lfsck_namespace_assistant_handler_p1,
        .la_handler_p2          = lfsck_namespace_assistant_handler_p2,
        .la_fill_pos            = lfsck_namespace_assistant_fill_pos,
@@ -6496,6 +6926,9 @@ int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
 
        linkea_first_entry(&ldata);
        linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
+       if (!linkea_entry_is_valid(&ldata, cname, pfid))
+               return -EINVAL;
+
        /* To guarantee the 'name' is terminated with '0'. */
        memcpy(name, cname->ln_name, cname->ln_namelen);
        name[cname->ln_namelen] = 0;
@@ -6569,7 +7002,7 @@ int lfsck_update_name_entry(const struct lu_env *env,
                GOTO(stop, rc);
 
        rc = dt_insert(env, dir, (const struct dt_rec *)rec,
-                      (const struct dt_key *)name, th, 1);
+                      (const struct dt_key *)name, th);
        if (rc == 0 && S_ISDIR(type) && !exists) {
                dt_write_lock(env, dir, 0);
                rc = dt_ref_add(env, dir, th);
@@ -6648,13 +7081,16 @@ int lfsck_namespace_setup(const struct lu_env *env,
 
        com->lc_obj = obj;
        rc = lfsck_namespace_load(env, com);
-       if (rc == -ENODATA)
+       if (rc == -ENODATA) {
                rc = lfsck_namespace_init(env, com);
-       else if (rc < 0)
+       } else if (rc < 0) {
                rc = lfsck_namespace_reset(env, com, true);
-       else
+       } else {
                rc = lfsck_load_sub_trace_files(env, com,
                        &dt_lfsck_namespace_features, LFSCK_NAMESPACE, false);
+               if (rc)
+                       rc = lfsck_namespace_reset(env, com, true);
+       }
        if (rc != 0)
                GOTO(out, rc);