Whamcloud - gitweb
LU-3469 osp: osp_sync() to flush pending changes synchronously
[fs/lustre-release.git] / lustre / lfsck / lfsck_layout.c
index c808808..98e61d2 100644 (file)
@@ -349,19 +349,40 @@ again:
 static int lfsck_layout_verify_header(struct lov_mds_md_v1 *lmm)
 {
        __u32 magic;
-       __u32 patten;
+       __u32 pattern;
 
        magic = le32_to_cpu(lmm->lmm_magic);
        /* If magic crashed, keep it there. Sometime later, during OST-object
         * orphan handling, if some OST-object(s) back-point to it, it can be
         * verified and repaired. */
-       if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
-               return -EINVAL;
+       if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3) {
+               struct ost_id   oi;
+               int             rc;
+
+               lmm_oi_cpu_to_le(&oi, &lmm->lmm_oi);
+               if ((magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
+                       rc = -EOPNOTSUPP;
+               else
+                       rc = -EINVAL;
+
+               CDEBUG(D_LFSCK, "%s LOV EA magic %u on "DOSTID"\n",
+                      rc == -EINVAL ? "Unknown" : "Unsupported",
+                      magic, POSTID(&oi));
 
-       patten = le32_to_cpu(lmm->lmm_pattern);
+               return rc;
+       }
+
+       pattern = le32_to_cpu(lmm->lmm_pattern);
        /* XXX: currently, we only support LOV_PATTERN_RAID0. */
-       if (patten != LOV_PATTERN_RAID0)
+       if (lov_pattern(pattern) != LOV_PATTERN_RAID0) {
+               struct ost_id oi;
+
+               lmm_oi_cpu_to_le(&oi, &lmm->lmm_oi);
+               CDEBUG(D_LFSCK, "Unsupported LOV EA pattern %u on "DOSTID"\n",
+                      pattern, POSTID(&oi));
+
                return -EOPNOTSUPP;
+       }
 
        return 0;
 }
@@ -652,15 +673,6 @@ out:
        }
 }
 
-static inline bool is_dummy_lov_ost_data(struct lov_ost_data_v1 *obj)
-{
-       if (fid_is_zero(&obj->l_ost_oi.oi_fid) &&
-           obj->l_ost_gen == 0 && obj->l_ost_idx == 0)
-               return true;
-
-       return false;
-}
-
 static void lfsck_layout_le_to_cpu(struct lfsck_layout *des,
                                   const struct lfsck_layout *src)
 {
@@ -1153,6 +1165,17 @@ out:
        return rc;
 }
 
+static void lfsck_layout_record_failure(const struct lu_env *env,
+                                                struct lfsck_instance *lfsck,
+                                                struct lfsck_layout *lo)
+{
+       lo->ll_objs_failed_phase1++;
+       if (unlikely(lo->ll_pos_first_inconsistent == 0))
+               lo->ll_pos_first_inconsistent =
+                       lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
+                                                       lfsck->li_di_oit);
+}
+
 static int lfsck_layout_master_async_interpret(const struct lu_env *env,
                                               struct ptlrpc_request *req,
                                               void *args, int rc)
@@ -1662,6 +1685,45 @@ static int lfsck_layout_trans_stop(const struct lu_env *env,
 }
 
 /**
+ * Get the system default stripe size.
+ *
+ * \param[in] env      pointer to the thread context
+ * \param[in] lfsck    pointer to the lfsck instance
+ * \param[out] size    pointer to the default stripe size
+ *
+ * \retval             0 for success
+ * \retval             negative error number on failure
+ */
+static int lfsck_layout_get_def_stripesize(const struct lu_env *env,
+                                          struct lfsck_instance *lfsck,
+                                          __u32 *size)
+{
+       struct lov_user_md      *lum = &lfsck_env_info(env)->lti_lum;
+       struct dt_object        *root;
+       int                      rc;
+
+       root = dt_locate(env, lfsck->li_next, &lfsck->li_local_root_fid);
+       if (IS_ERR(root))
+               return PTR_ERR(root);
+
+       /* Get the default stripe size via xattr_get on the backend root. */
+       rc = dt_xattr_get(env, root, lfsck_buf_get(env, lum, sizeof(*lum)),
+                         XATTR_NAME_LOV, BYPASS_CAPA);
+       if (rc > 0) {
+               /* The lum->lmm_stripe_size is LE mode. The *size also
+                * should be LE mode. So it is unnecessary to convert. */
+               *size = lum->lmm_stripe_size;
+               rc = 0;
+       } else if (unlikely(rc == 0)) {
+               rc = -EINVAL;
+       }
+
+       lfsck_object_put(env, root);
+
+       return rc;
+}
+
+/**
  * \retval      +1: repaired
  * \retval       0: did nothing
  * \retval     -ve: on error
@@ -1674,13 +1736,36 @@ static int lfsck_layout_refill_lovea(const struct lu_env *env,
                                     struct lov_ost_data_v1 *slot,
                                     int fl, __u32 ost_idx)
 {
-       struct ost_id   *oi     = &lfsck_env_info(env)->lti_oi;
-       int              rc;
+       struct ost_id           *oi     = &lfsck_env_info(env)->lti_oi;
+       struct lov_mds_md_v1    *lmm    = buf->lb_buf;
+       int                      rc;
 
        fid_to_ostid(cfid, oi);
        ostid_cpu_to_le(oi, &slot->l_ost_oi);
        slot->l_ost_gen = cpu_to_le32(0);
        slot->l_ost_idx = cpu_to_le32(ost_idx);
+
+       if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE) {
+               struct lov_ost_data_v1 *objs;
+               int                     i;
+               __u16                   count;
+
+               count = le16_to_cpu(lmm->lmm_stripe_count);
+               if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1)
+                       objs = &lmm->lmm_objects[0];
+               else
+                       objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
+               for (i = 0; i < count; i++, objs++) {
+                       if (objs != slot && lovea_slot_is_dummy(objs))
+                               break;
+               }
+
+               /* If the @slot is the last dummy slot to be refilled,
+                * then drop LOV_PATTERN_F_HOLE from lmm::lmm_pattern. */
+               if (i == count)
+                       lmm->lmm_pattern &= ~cpu_to_le32(LOV_PATTERN_F_HOLE);
+       }
+
        rc = dt_xattr_set(env, parent, buf, XATTR_NAME_LOV, fl, handle,
                          BYPASS_CAPA);
        if (rc == 0)
@@ -1695,59 +1780,67 @@ static int lfsck_layout_refill_lovea(const struct lu_env *env,
  * \retval     -ve: on error
  */
 static int lfsck_layout_extend_lovea(const struct lu_env *env,
+                                    struct lfsck_instance *lfsck,
                                     struct thandle *handle,
                                     struct dt_object *parent,
                                     struct lu_fid *cfid,
                                     struct lu_buf *buf, int fl,
-                                    __u32 ost_idx, __u32 ea_off)
+                                    __u32 ost_idx, __u32 ea_off, bool reset)
 {
        struct lov_mds_md_v1    *lmm    = buf->lb_buf;
        struct lov_ost_data_v1  *objs;
        int                      rc;
+       __u16                    count;
        ENTRY;
 
-       if (fl == LU_XATTR_CREATE) {
-               LASSERT(buf->lb_len == lov_mds_md_size(ea_off + 1,
-                                                      LOV_MAGIC_V1));
+       if (fl == LU_XATTR_CREATE || reset) {
+               __u32 pattern = LOV_PATTERN_RAID0;
+
+               count = ea_off + 1;
+               LASSERT(buf->lb_len == lov_mds_md_size(count, LOV_MAGIC_V1));
+
+               if (ea_off != 0 || reset)
+                       pattern |= LOV_PATTERN_F_HOLE;
 
                memset(lmm, 0, buf->lb_len);
                lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
-               /* XXX: currently, we only support LOV_PATTERN_RAID0. */
-               lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
+               lmm->lmm_pattern = cpu_to_le32(pattern);
                fid_to_lmm_oi(lfsck_dto2fid(parent), &lmm->lmm_oi);
                lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
-               /* XXX: We cannot know the stripe size,
-                *      then use the default value (1 MB). */
-               lmm->lmm_stripe_size =
-                       cpu_to_le32(LOV_DESC_STRIPE_SIZE_DEFAULT);
-               objs = &(lmm->lmm_objects[ea_off]);
+
+               rc = lfsck_layout_get_def_stripesize(env, lfsck,
+                                                    &lmm->lmm_stripe_size);
+               if (rc != 0)
+                       RETURN(rc);
+
+               objs = &lmm->lmm_objects[ea_off];
        } else {
-               __u16   count = le16_to_cpu(lmm->lmm_stripe_count);
-               int     gap   = ea_off - count;
                __u32   magic = le32_to_cpu(lmm->lmm_magic);
+               int     gap;
 
-               /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3
-                * which has been verified in lfsck_layout_verify_header()
-                * already. If some new magic introduced in the future,
-                * then layout LFSCK needs to be updated also. */
-               if (magic == LOV_MAGIC_V1) {
-                       objs = &(lmm->lmm_objects[count]);
-               } else {
-                       LASSERT(magic == LOV_MAGIC_V3);
+               count = le16_to_cpu(lmm->lmm_stripe_count);
+               if (magic == LOV_MAGIC_V1)
+                       objs = &lmm->lmm_objects[count];
+               else
                        objs = &((struct lov_mds_md_v3 *)lmm)->
                                                        lmm_objects[count];
-               }
 
-               if (gap > 0)
+               gap = ea_off - count;
+               if (gap >= 0)
+                       count = ea_off + 1;
+               LASSERT(buf->lb_len == lov_mds_md_size(count, magic));
+
+               if (gap > 0) {
                        memset(objs, 0, gap * sizeof(*objs));
+                       lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_HOLE);
+               }
+
                lmm->lmm_layout_gen =
                            cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
                objs += gap;
-
-               LASSERT(buf->lb_len == lov_mds_md_size(ea_off + 1, magic));
        }
 
-       lmm->lmm_stripe_count = cpu_to_le16(ea_off + 1);
+       lmm->lmm_stripe_count = cpu_to_le16(count);
        rc = lfsck_layout_refill_lovea(env, handle, parent, cfid, buf, objs,
                                       fl, ost_idx);
 
@@ -1917,7 +2010,7 @@ static int lfsck_layout_recreate_parent(const struct lu_env *env,
        memset(la, 0, sizeof(*la));
        la->la_uid = rec->lor_uid;
        la->la_gid = rec->lor_gid;
-       la->la_mode = S_IFREG | S_IRUSR | S_IWUSR;
+       la->la_mode = S_IFREG | S_IRUSR;
        la->la_valid = LA_MODE | LA_UID | LA_GID;
 
        memset(dof, 0, sizeof(*dof));
@@ -1993,9 +2086,9 @@ static int lfsck_layout_recreate_parent(const struct lu_env *env,
        rc = dt_create(env, pobj, la, NULL, dof, th);
        if (rc == 0)
                /* 3b. Add layout EA for the MDT-object. */
-               rc = lfsck_layout_extend_lovea(env, th, pobj, cfid, ea_buf,
-                                              LU_XATTR_CREATE, ltd->ltd_index,
-                                              ea_off);
+               rc = lfsck_layout_extend_lovea(env, lfsck, th, pobj, cfid,
+                                              ea_buf, LU_XATTR_CREATE,
+                                              ltd->ltd_index, ea_off, false);
        dt_write_unlock(env, pobj);
        if (rc < 0)
                GOTO(stop, rc);
@@ -2394,14 +2487,31 @@ again:
                LASSERT(buf->lb_len >= rc);
 
                buf->lb_len = rc;
-               rc = lfsck_layout_extend_lovea(env, handle, parent, cfid, buf,
-                                              fl, ost_idx, ea_off);
+               rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
+                                              buf, fl, ost_idx, ea_off, false);
 
                GOTO(unlock_parent, rc);
        }
 
        lmm = buf->lb_buf;
        rc1 = lfsck_layout_verify_header(lmm);
+
+       /* If the LOV EA crashed, the rebuild it. */
+       if (rc1 == -EINVAL) {
+               if (bk->lb_param & LPF_DRYRUN)
+                       GOTO(unlock_parent, rc = 1);
+
+               LASSERT(buf->lb_len >= rc);
+
+               buf->lb_len = rc;
+               memset(lmm, 0, buf->lb_len);
+               rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
+                                              buf, fl, ost_idx, ea_off, true);
+
+               GOTO(unlock_parent, rc);
+       }
+
+       /* For other unknown magic/pattern, keep the current LOV EA. */
        if (rc1 != 0)
                GOTO(unlock_parent, rc = rc1);
 
@@ -2411,7 +2521,7 @@ again:
         * be updated also. */
        magic = le32_to_cpu(lmm->lmm_magic);
        if (magic == LOV_MAGIC_V1) {
-               objs = &(lmm->lmm_objects[0]);
+               objs = &lmm->lmm_objects[0];
        } else {
                LASSERT(magic == LOV_MAGIC_V3);
                objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
@@ -2433,8 +2543,9 @@ again:
                        goto again;
 
                buf->lb_len = rc;
-               rc = lfsck_layout_extend_lovea(env, handle, parent, cfid, buf,
-                                              fl, ost_idx, ea_off);
+               rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
+                                              buf, fl, ost_idx, ea_off, false);
+
                GOTO(unlock_parent, rc);
        }
 
@@ -2444,7 +2555,7 @@ again:
        for (i = 0; i < count; i++, objs++) {
                /* The MDT-object was created via lfsck_layout_recover_create()
                 * by others before, and we fill the dummy layout EA. */
-               if (is_dummy_lov_ost_data(objs)) {
+               if (lovea_slot_is_dummy(objs)) {
                        if (i != ea_off)
                                continue;
 
@@ -2496,7 +2607,7 @@ again:
                dt_trans_stop(env, dt, handle);
        lfsck_layout_unlock(&lh);
        if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1)
-               objs = &(lmm->lmm_objects[ea_off]);
+               objs = &lmm->lmm_objects[ea_off];
        else
                objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
        rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
@@ -2688,16 +2799,24 @@ put:
        return rc > 0 ? 0 : rc;
 }
 
-/* For the MDT-object with dangling reference, we need to re-create
- * the missed OST-object with the known FID/owner information. */
-static int lfsck_layout_recreate_ostobj(const struct lu_env *env,
+/* For the MDT-object with dangling reference, we need to repare the
+ * inconsistency according to the LFSCK sponsor's requirement:
+ *
+ * 1) Keep the inconsistency there and report the inconsistency case,
+ *    then give the chance to the application to find related issues,
+ *    and the users can make the decision about how to handle it with
+ *    more human knownledge. (by default)
+ *
+ * 2) Re-create the missed OST-object with the FID/owner information. */
+static int lfsck_layout_repair_dangling(const struct lu_env *env,
                                        struct lfsck_component *com,
                                        struct lfsck_layout_req *llr,
-                                       struct lu_attr *la)
+                                       const struct lu_attr *pla)
 {
        struct lfsck_thread_info        *info   = lfsck_env_info(env);
        struct filter_fid               *pfid   = &info->lti_new_pfid;
        struct dt_allocation_hint       *hint   = &info->lti_hint;
+       struct lu_attr                  *cla    = &info->lti_la2;
        struct dt_object                *parent = llr->llr_parent->llo_obj;
        struct dt_object                *child  = llr->llr_child;
        struct dt_device                *dev    = lfsck_obj2dt_dev(child);
@@ -2706,12 +2825,30 @@ static int lfsck_layout_recreate_ostobj(const struct lu_env *env,
        struct lu_buf                   *buf;
        struct lustre_handle             lh     = { 0 };
        int                              rc;
+       bool                             create;
        ENTRY;
 
-       CDEBUG(D_LFSCK, "Repair dangling reference for: parent "DFID
-              ", child "DFID", OST-index %u, stripe-index %u, owner %u:%u\n",
+       if (com->lc_lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ)
+               create = true;
+       else
+               create = false;
+
+       CDEBUG(D_LFSCK, "Found dangling reference for: parent "DFID
+              ", child "DFID", OST-index %u, stripe-index %u, owner %u:%u. %s",
               PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
-              llr->llr_ost_idx, llr->llr_lov_idx, la->la_uid, la->la_gid);
+              llr->llr_ost_idx, llr->llr_lov_idx, pla->la_uid, pla->la_gid,
+              create ? "Create the lost OST-object as required.\n" :
+                       "Keep the MDT-object there by default.\n");
+
+       if (!create)
+               RETURN(1);
+
+       memset(cla, 0, sizeof(*cla));
+       cla->la_uid = pla->la_uid;
+       cla->la_gid = pla->la_gid;
+       cla->la_mode = S_IFREG | 0666;
+       cla->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
+                       LA_ATIME | LA_MTIME | LA_CTIME;
 
        rc = lfsck_layout_lock(env, com, parent, &lh,
                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
@@ -2732,7 +2869,7 @@ static int lfsck_layout_recreate_ostobj(const struct lu_env *env,
        pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
        buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
 
-       rc = dt_declare_create(env, child, la, hint, NULL, handle);
+       rc = dt_declare_create(env, child, cla, hint, NULL, handle);
        if (rc != 0)
                GOTO(stop, rc);
 
@@ -2749,7 +2886,7 @@ static int lfsck_layout_recreate_ostobj(const struct lu_env *env,
        if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
                GOTO(unlock2, rc = 1);
 
-       rc = dt_create(env, child, la, hint, NULL, handle);
+       rc = dt_create(env, child, cla, hint, NULL, handle);
        if (rc != 0)
                GOTO(unlock2, rc);
 
@@ -2936,10 +3073,6 @@ static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
                GOTO(unlock2, rc = 0);
 
        lmm = buf->lb_buf;
-       rc = lfsck_layout_verify_header(lmm);
-       if (rc != 0)
-               GOTO(unlock2, rc);
-
        /* Someone change layout during the LFSCK, no need to repair then. */
        if (le16_to_cpu(lmm->lmm_layout_gen) != llr->llr_parent->llo_gen)
                GOTO(unlock2, rc = 0);
@@ -2954,7 +3087,7 @@ static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
         * be updated also. */
        magic = le32_to_cpu(lmm->lmm_magic);
        if (magic == LOV_MAGIC_V1) {
-               objs = &(lmm->lmm_objects[0]);
+               objs = &lmm->lmm_objects[0];
        } else {
                LASSERT(magic == LOV_MAGIC_V3);
                objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
@@ -3126,17 +3259,9 @@ static int lfsck_layout_check_parent(const struct lu_env *env,
                GOTO(out, rc);
 
        lmm = buf->lb_buf;
-       rc = lfsck_layout_verify_header(lmm);
-       if (rc != 0)
-               GOTO(out, rc);
-
-       /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
-        * been verified in lfsck_layout_verify_header() already. If some
-        * new magic introduced in the future, then layout LFSCK needs to
-        * be updated also. */
        magic = le32_to_cpu(lmm->lmm_magic);
        if (magic == LOV_MAGIC_V1) {
-               objs = &(lmm->lmm_objects[0]);
+               objs = &lmm->lmm_objects[0];
        } else {
                LASSERT(magic == LOV_MAGIC_V3);
                objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
@@ -3147,7 +3272,7 @@ static int lfsck_layout_check_parent(const struct lu_env *env,
                struct lu_fid           *tfid   = &info->lti_fid2;
                struct ost_id           *oi     = &info->lti_oi;
 
-               if (is_dummy_lov_ost_data(objs))
+               if (lovea_slot_is_dummy(objs))
                        continue;
 
                ostid_le_to_cpu(&objs->l_ost_oi, oi);
@@ -3256,13 +3381,7 @@ repair:
 
        switch (type) {
        case LLIT_DANGLING:
-               memset(cla, 0, sizeof(*cla));
-               cla->la_uid = pla->la_uid;
-               cla->la_gid = pla->la_gid;
-               cla->la_mode = S_IFREG | 0666;
-               cla->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
-                               LA_ATIME | LA_MTIME | LA_CTIME;
-               rc = lfsck_layout_recreate_ostobj(env, com, llr, cla);
+               rc = lfsck_layout_repair_dangling(env, com, llr, pla);
                break;
        case LLIT_UNMATCHED_PAIR:
                rc = lfsck_layout_repair_unmatched_pair(env, com, llr, pla);
@@ -3299,13 +3418,18 @@ out:
                        lo->ll_objs_skipped++;
                        rc = 0;
                } else {
-                       lo->ll_objs_failed_phase1++;
+                       lfsck_layout_record_failure(env, lfsck, lo);
                }
        } else if (rc > 0) {
                LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
                         "unknown type = %d\n", type);
 
                lo->ll_objs_repaired[type - 1]++;
+               if (bk->lb_param & LPF_DRYRUN &&
+                   unlikely(lo->ll_pos_first_inconsistent == 0))
+                       lo->ll_pos_first_inconsistent =
+                       lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
+                                                       lfsck->li_di_oit);
        }
        up_write(&com->lc_sem);
 
@@ -3334,7 +3458,7 @@ static int lfsck_layout_assistant(void *args)
        memset(lr, 0, sizeof(*lr));
        lr->lr_event = LE_START;
        lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN |
-                      LSV_ASYNC_WINDOWS;
+                      LSV_ASYNC_WINDOWS | LSV_CREATE_OSTOBJ;
        lr->lr_speed = bk->lb_speed_limit;
        lr->lr_version = bk->lb_version;
        lr->lr_param = bk->lb_param;
@@ -3435,6 +3559,9 @@ static int lfsck_layout_assistant(void *args)
                                com->lc_time_last_checkpoint +
                                cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
 
+                       /* flush all async updating before handling orphan. */
+                       dt_sync(env, lfsck->li_next);
+
                        while (llmd->llmd_in_double_scan) {
                                struct lfsck_tgt_descs  *ltds =
                                                        &lfsck->li_ost_descs;
@@ -3872,7 +3999,7 @@ static int lfsck_layout_master_check_pairs(const struct lu_env *env,
         * be updated also. */
        magic = le32_to_cpu(lmm->lmm_magic);
        if (magic == LOV_MAGIC_V1) {
-               objs = &(lmm->lmm_objects[0]);
+               objs = &lmm->lmm_objects[0];
        } else {
                LASSERT(magic == LOV_MAGIC_V3);
                objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
@@ -4062,14 +4189,7 @@ static void lfsck_layout_fail(const struct lu_env *env,
        down_write(&com->lc_sem);
        if (new_checked)
                com->lc_new_checked++;
-       lo->ll_objs_failed_phase1++;
-       if (lo->ll_pos_first_inconsistent == 0) {
-               struct lfsck_instance *lfsck = com->lc_lfsck;
-
-               lo->ll_pos_first_inconsistent =
-                       lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
-                                                       lfsck->li_di_oit);
-       }
+       lfsck_layout_record_failure(env, com->lc_lfsck, lo);
        up_write(&com->lc_sem);
 }
 
@@ -4163,6 +4283,9 @@ static int lfsck_layout_prep(const struct lu_env *env,
                int rc;
 
                rc = lfsck_layout_reset(env, com, false);
+               if (rc == 0)
+                       rc = lfsck_set_param(env, lfsck, start, true);
+
                if (rc != 0)
                        return rc;
        }
@@ -4213,13 +4336,26 @@ static int lfsck_layout_slave_prep(const struct lu_env *env,
                                   struct lfsck_start_param *lsp)
 {
        struct lfsck_layout_slave_data  *llsd   = com->lc_data;
+       struct lfsck_instance           *lfsck  = com->lc_lfsck;
+       struct lfsck_layout             *lo     = com->lc_file_ram;
        struct lfsck_start              *start  = lsp->lsp_start;
        int                              rc;
 
        rc = lfsck_layout_prep(env, com, start);
-       if (rc != 0 || !lsp->lsp_index_valid)
+       if (rc != 0)
                return rc;
 
+       if (lo->ll_flags & LF_CRASHED_LASTID &&
+           list_empty(&llsd->llsd_master_list)) {
+               LASSERT(lfsck->li_out_notify != NULL);
+
+               lfsck->li_out_notify(env, lfsck->li_out_notify_data,
+                                    LE_LASTID_REBUILDING);
+       }
+
+       if (!lsp->lsp_index_valid)
+               return 0;
+
        rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
        if (rc == 0 && start != NULL && start->ls_flags & LPF_ORPHAN) {
                LASSERT(!llsd->llsd_rbtree_valid);
@@ -4241,7 +4377,8 @@ static int lfsck_layout_master_prep(const struct lu_env *env,
        struct ptlrpc_thread            *mthread = &lfsck->li_thread;
        struct ptlrpc_thread            *athread = &llmd->llmd_thread;
        struct lfsck_thread_args        *lta;
-       long                             rc;
+       struct task_struct              *task;
+       int                              rc;
        ENTRY;
 
        rc = lfsck_layout_prep(env, com, lsp->lsp_start);
@@ -4260,10 +4397,11 @@ static int lfsck_layout_master_prep(const struct lu_env *env,
        if (IS_ERR(lta))
                RETURN(PTR_ERR(lta));
 
-       rc = PTR_ERR(kthread_run(lfsck_layout_assistant, lta, "lfsck_layout"));
-       if (IS_ERR_VALUE(rc)) {
+       task = kthread_run(lfsck_layout_assistant, lta, "lfsck_layout");
+       if (IS_ERR(task)) {
+               rc = PTR_ERR(task);
                CERROR("%s: Cannot start LFSCK layout assistant thread: "
-                      "rc = %ld\n", lfsck_lfsck2name(lfsck), rc);
+                      "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
                lfsck_thread_args_fini(lta);
        } else {
                struct l_wait_info lwi = { 0 };
@@ -4316,7 +4454,7 @@ static int lfsck_layout_scan_stripes(const struct lu_env *env,
         * be updated also. */
        magic = le32_to_cpu(lmm->lmm_magic);
        if (magic == LOV_MAGIC_V1) {
-               objs = &(lmm->lmm_objects[0]);
+               objs = &lmm->lmm_objects[0];
        } else {
                LASSERT(magic == LOV_MAGIC_V3);
                objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
@@ -4332,7 +4470,7 @@ static int lfsck_layout_scan_stripes(const struct lu_env *env,
                                        le32_to_cpu(objs->l_ost_idx);
                bool                     wakeup = false;
 
-               if (is_dummy_lov_ost_data(objs))
+               if (unlikely(lovea_slot_is_dummy(objs)))
                        continue;
 
                l_wait_event(mthread->t_ctl_waitq,
@@ -4408,7 +4546,7 @@ next:
                down_write(&com->lc_sem);
                com->lc_new_checked++;
                if (rc < 0)
-                       lo->ll_objs_failed_phase1++;
+                       lfsck_layout_record_failure(env, lfsck, lo);
                up_write(&com->lc_sem);
 
                if (cobj != NULL && !IS_ERR(cobj))
@@ -4478,6 +4616,8 @@ again:
        buf->lb_len = rc;
        lmm = buf->lb_buf;
        rc = lfsck_layout_verify_header(lmm);
+       /* If the LOV EA crashed, then it is possible to be rebuilt later
+        * when handle orphan OST-objects. */
        if (rc != 0)
                GOTO(out, rc);
 
@@ -4555,7 +4695,7 @@ out:
                down_write(&com->lc_sem);
                com->lc_new_checked++;
                if (rc < 0)
-                       lo->ll_objs_failed_phase1++;
+                       lfsck_layout_record_failure(env, lfsck, lo);
                up_write(&com->lc_sem);
        }
        buf->lb_len = buflen;
@@ -4579,6 +4719,17 @@ static int lfsck_layout_slave_exec_oit(const struct lu_env *env,
 
        LASSERT(llsd != NULL);
 
+       if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY5) &&
+           cfs_fail_val == lfsck_dev_idx(lfsck->li_bottom)) {
+               struct l_wait_info       lwi = LWI_TIMEOUT(cfs_time_seconds(1),
+                                                          NULL, NULL);
+               struct ptlrpc_thread    *thread = &lfsck->li_thread;
+
+               l_wait_event(thread->t_ctl_waitq,
+                            !thread_is_running(thread),
+                            &lwi);
+       }
+
        lfsck_rbtree_update_bitmap(env, com, fid, false);
 
        down_write(&com->lc_sem);