Whamcloud - gitweb
LU-1399 config: check lustre_cfg_new() return
[fs/lustre-release.git] / lustre / mdt / mdt_handler.c
index 9abfa46..3d5de6f 100644 (file)
@@ -69,6 +69,7 @@
 #include <lustre_param.h>
 #include <lustre_quota.h>
 #include <lustre_lfsck.h>
+#include <lustre_nodemap.h>
 
 mdl_mode_t mdt_mdl_lock_modes[] = {
         [LCK_MINMODE] = MDL_MINMODE,
@@ -388,12 +389,70 @@ static void mdt_pack_size2body(struct mdt_thread_info *info,
        b->mbo_blocks = ma->ma_som->msd_blocks;
 }
 
+#ifdef CONFIG_FS_POSIX_ACL
+/*
+ * Pack ACL data into the reply. UIDs/GIDs are mapped and filtered by nodemap.
+ *
+ * \param      info    thread info object
+ * \param      repbody reply to pack ACLs into
+ * \param      o       mdt object of file to examine
+ * \param      nodemap nodemap of client to reply to
+ * \retval     0       success
+ * \retval     -errno  error getting or parsing ACL from disk
+ */
+int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
+                     struct mdt_object *o, struct lu_nodemap *nodemap)
+{
+       const struct lu_env     *env = info->mti_env;
+       struct md_object        *next = mdt_object_child(o);
+       struct lu_buf           *buf = &info->mti_buf;
+       int rc;
+
+       buf->lb_buf = req_capsule_server_get(info->mti_pill, &RMF_ACL);
+       buf->lb_len = req_capsule_get_size(info->mti_pill, &RMF_ACL,
+                                          RCL_SERVER);
+       if (buf->lb_len == 0)
+               return 0;
+
+       rc = mo_xattr_get(env, next, buf, XATTR_NAME_ACL_ACCESS);
+       if (rc < 0) {
+               if (rc == -ENODATA) {
+                       repbody->mbo_aclsize = 0;
+                       repbody->mbo_valid |= OBD_MD_FLACL;
+                       rc = 0;
+               } else if (rc == -EOPNOTSUPP) {
+                       rc = 0;
+               } else {
+                       CERROR("%s: unable to read "DFID" ACL: rc = %d\n",
+                              mdt_obd_name(info->mti_mdt),
+                              PFID(mdt_object_fid(o)), rc);
+               }
+       } else {
+               rc = nodemap_map_acl(nodemap, buf->lb_buf,
+                                    rc, NODEMAP_FS_TO_CLIENT);
+               /* if all ACLs mapped out, rc is still >= 0 */
+               if (rc < 0) {
+                       CERROR("%s: nodemap_map_acl unable to parse "DFID
+                              " ACL: rc = %d\n", mdt_obd_name(info->mti_mdt),
+                              PFID(mdt_object_fid(o)), rc);
+               } else {
+                       repbody->mbo_aclsize = rc;
+                       repbody->mbo_valid |= OBD_MD_FLACL;
+                       rc = 0;
+               }
+       }
+       return rc;
+}
+#endif
+
 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
                         const struct lu_attr *attr, const struct lu_fid *fid)
 {
-        struct md_attr *ma = &info->mti_attr;
+       struct md_attr          *ma = &info->mti_attr;
+       struct obd_export       *exp = info->mti_exp;
+       struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
 
-        LASSERT(ma->ma_valid & MA_INODE);
+       LASSERT(ma->ma_valid & MA_INODE);
 
        b->mbo_atime      = attr->la_atime;
        b->mbo_mtime      = attr->la_mtime;
@@ -401,8 +460,12 @@ void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
        b->mbo_mode       = attr->la_mode;
        b->mbo_size       = attr->la_size;
        b->mbo_blocks     = attr->la_blocks;
-       b->mbo_uid      = attr->la_uid;
-       b->mbo_gid      = attr->la_gid;
+       b->mbo_uid        = nodemap_map_id(nodemap, NODEMAP_UID,
+                                          NODEMAP_FS_TO_CLIENT,
+                                          attr->la_uid);
+       b->mbo_gid        = nodemap_map_id(nodemap, NODEMAP_GID,
+                                          NODEMAP_FS_TO_CLIENT,
+                                          attr->la_gid);
        b->mbo_flags      = attr->la_flags;
        b->mbo_nlink      = attr->la_nlink;
        b->mbo_rdev       = attr->la_rdev;
@@ -779,25 +842,27 @@ out:
 static int mdt_getattr_internal(struct mdt_thread_info *info,
                                 struct mdt_object *o, int ma_need)
 {
-        struct md_object        *next = mdt_object_child(o);
-        const struct mdt_body   *reqbody = info->mti_body;
-        struct ptlrpc_request   *req = mdt_info_req(info);
-        struct md_attr          *ma = &info->mti_attr;
-        struct lu_attr          *la = &ma->ma_attr;
-        struct req_capsule      *pill = info->mti_pill;
-        const struct lu_env     *env = info->mti_env;
-        struct mdt_body         *repbody;
-        struct lu_buf           *buffer = &info->mti_buf;
-        int                     rc;
-       int                     is_root;
-        ENTRY;
+       struct md_object        *next = mdt_object_child(o);
+       const struct mdt_body   *reqbody = info->mti_body;
+       struct ptlrpc_request   *req = mdt_info_req(info);
+       struct md_attr          *ma = &info->mti_attr;
+       struct lu_attr          *la = &ma->ma_attr;
+       struct req_capsule      *pill = info->mti_pill;
+       const struct lu_env     *env = info->mti_env;
+       struct mdt_body         *repbody;
+       struct lu_buf           *buffer = &info->mti_buf;
+       struct obd_export       *exp = info->mti_exp;
+       struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
+       int                      rc;
+       int                      is_root;
+       ENTRY;
 
-        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
-                RETURN(err_serious(-ENOMEM));
+       if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))
+               RETURN(err_serious(-ENOMEM));
 
-        repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
+       repbody = req_capsule_server_get(pill, &RMF_MDT_BODY);
 
-        ma->ma_valid = 0;
+       ma->ma_valid = 0;
 
        if (mdt_object_remote(o)) {
                /* This object is located on remote node.*/
@@ -997,33 +1062,8 @@ static int mdt_getattr_internal(struct mdt_thread_info *info,
        }
 #ifdef CONFIG_FS_POSIX_ACL
        else if ((exp_connect_flags(req->rq_export) & OBD_CONNECT_ACL) &&
-                (reqbody->mbo_valid & OBD_MD_FLACL)) {
-                buffer->lb_buf = req_capsule_server_get(pill, &RMF_ACL);
-                buffer->lb_len = req_capsule_get_size(pill,
-                                                      &RMF_ACL, RCL_SERVER);
-                if (buffer->lb_len > 0) {
-                        rc = mo_xattr_get(env, next, buffer,
-                                          XATTR_NAME_ACL_ACCESS);
-                        if (rc < 0) {
-                                if (rc == -ENODATA) {
-                                       repbody->mbo_aclsize = 0;
-                                       repbody->mbo_valid |= OBD_MD_FLACL;
-                                        rc = 0;
-                                } else if (rc == -EOPNOTSUPP) {
-                                        rc = 0;
-                               } else {
-                                       CERROR("%s: unable to read "DFID
-                                              " ACL: rc = %d\n",
-                                              mdt_obd_name(info->mti_mdt),
-                                              PFID(mdt_object_fid(o)), rc);
-                               }
-                        } else {
-                               repbody->mbo_aclsize = rc;
-                               repbody->mbo_valid |= OBD_MD_FLACL;
-                               rc = 0;
-                       }
-               }
-       }
+                (reqbody->mbo_valid & OBD_MD_FLACL))
+               rc = mdt_pack_acl2body(info, repbody, o, nodemap);
 #endif
 
        if (reqbody->mbo_valid & OBD_MD_FLMDSCAPA &&
@@ -1195,7 +1235,7 @@ static int mdt_swap_layouts(struct tgt_session_info *tsi)
        info = tsi2mdt_info(tsi);
 
        if (info->mti_dlm_req != NULL)
-               ldlm_request_cancel(req, info->mti_dlm_req, 0);
+               ldlm_request_cancel(req, info->mti_dlm_req, 0, LATF_SKIP);
 
        if (req_capsule_get_size(info->mti_pill, &RMF_CAPA1, RCL_CLIENT))
                mdt_set_capainfo(info, 0, &info->mti_body->mbo_fid1,
@@ -1269,43 +1309,32 @@ out:
 }
 
 static int mdt_raw_lookup(struct mdt_thread_info *info,
-                          struct mdt_object *parent,
-                          const struct lu_name *lname,
-                          struct ldlm_reply *ldlm_rep)
+                         struct mdt_object *parent,
+                         const struct lu_name *lname,
+                         struct ldlm_reply *ldlm_rep)
 {
-        struct md_object *next = mdt_object_child(info->mti_object);
-        const struct mdt_body *reqbody = info->mti_body;
-        struct lu_fid *child_fid = &info->mti_tmp_fid1;
-        struct mdt_body *repbody;
-        int rc;
-        ENTRY;
-
-       if (reqbody->mbo_valid != OBD_MD_FLID)
-                RETURN(0);
+       struct lu_fid   *child_fid = &info->mti_tmp_fid1;
+       int              rc;
+       ENTRY;
 
-        LASSERT(!info->mti_cross_ref);
+       LASSERT(!info->mti_cross_ref);
 
-        /* Only got the fid of this obj by name */
-        fid_zero(child_fid);
-        rc = mdo_lookup(info->mti_env, next, lname, child_fid,
-                        &info->mti_spec);
-#if 0
-        /* XXX is raw_lookup possible as intent operation? */
-        if (rc != 0) {
-                if (rc == -ENOENT)
-                        mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
-                RETURN(rc);
-        } else
-                mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
+       /* Only got the fid of this obj by name */
+       fid_zero(child_fid);
+       rc = mdo_lookup(info->mti_env, mdt_object_child(info->mti_object),
+                       lname, child_fid, &info->mti_spec);
+       if (rc == 0) {
+               struct mdt_body *repbody;
 
-        repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
-#endif
-        if (rc == 0) {
-                repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
+               repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
                repbody->mbo_fid1 = *child_fid;
                repbody->mbo_valid = OBD_MD_FLID;
-        }
-        RETURN(1);
+               mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS);
+       } else if (rc == -ENOENT) {
+               mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
+       }
+
+       RETURN(rc);
 }
 
 /*
@@ -1323,7 +1352,6 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info,
         struct mdt_body        *reqbody   = NULL;
         struct mdt_object      *parent    = info->mti_object;
         struct mdt_object      *child;
-        struct md_object       *next      = mdt_object_child(parent);
         struct lu_fid          *child_fid = &info->mti_tmp_fid1;
         struct lu_name         *lname     = NULL;
         struct mdt_lock_handle *lhp       = NULL;
@@ -1338,7 +1366,8 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info,
        LASSERT(ergo(is_resent,
                     lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT));
 
-       LASSERT(parent != NULL);
+       if (parent == NULL)
+               RETURN(-ENOENT);
 
        if (info->mti_cross_ref) {
                /* Only getattr on the child. Parent is on another node. */
@@ -1351,7 +1380,7 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info,
 
                rc = mdt_check_resent_lock(info, child, lhc);
                if (rc < 0) {
-                       RETURN(-EPROTO);
+                       RETURN(rc);
                } else if (rc > 0) {
                        mdt_lock_handle_init(lhc);
                        mdt_lock_reg_init(lhc, LCK_PR);
@@ -1427,10 +1456,15 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info,
        }
 
        if (lu_name_is_valid(lname)) {
-               rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
-               if (rc != 0) {
-                       if (rc > 0)
-                               rc = 0;
+               /* Always allow to lookup ".." */
+               if (unlikely(lname->ln_namelen == 2 &&
+                            lname->ln_name[0] == '.' &&
+                            lname->ln_name[1] == '.'))
+                       info->mti_spec.sp_permitted = 1;
+
+               if (info->mti_body->mbo_valid == OBD_MD_FLID) {
+                       rc = mdt_raw_lookup(info, parent, lname, ldlm_rep);
+
                        RETURN(rc);
                }
 
@@ -1447,8 +1481,8 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info,
 
                 /* step 2: lookup child's fid by name */
                 fid_zero(child_fid);
-                rc = mdo_lookup(info->mti_env, next, lname, child_fid,
-                                &info->mti_spec);
+               rc = mdo_lookup(info->mti_env, mdt_object_child(parent), lname,
+                               child_fid, &info->mti_spec);
                if (rc == -ENOENT)
                        mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_NEG);
 
@@ -1973,6 +2007,7 @@ static int mdt_quotactl(struct tgt_session_info *tsi)
        int                      id, rc;
        struct mdt_device       *mdt = mdt_exp2dev(exp);
        struct lu_device        *qmt = mdt->mdt_qmt_dev;
+       struct lu_nodemap       *nodemap = exp->exp_target_data.ted_nodemap;
        ENTRY;
 
        oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
@@ -1993,9 +2028,11 @@ static int mdt_quotactl(struct tgt_session_info *tsi)
                /* deprecated, not used any more */
                RETURN(-EOPNOTSUPP);
                /* master quotactl */
-       case Q_GETINFO:
        case Q_SETINFO:
        case Q_SETQUOTA:
+               if (!nodemap_can_setquota(nodemap))
+                       RETURN(-EPERM);
+       case Q_GETINFO:
        case Q_GETQUOTA:
                if (qmt == NULL)
                        RETURN(-EOPNOTSUPP);
@@ -2034,6 +2071,13 @@ static int mdt_quotactl(struct tgt_session_info *tsi)
                }
        }
 
+       if (oqctl->qc_type == USRQUOTA)
+               id = nodemap_map_id(nodemap, NODEMAP_UID,
+                                   NODEMAP_CLIENT_TO_FS, id);
+       else if (oqctl->qc_type == GRPQUOTA)
+               id = nodemap_map_id(nodemap, NODEMAP_UID,
+                                   NODEMAP_CLIENT_TO_FS, id);
+
        repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL);
        if (repoqc == NULL)
                RETURN(err_serious(-EFAULT));
@@ -2329,8 +2373,14 @@ int mdt_check_resent_lock(struct mdt_thread_info *info,
 
                lock = ldlm_handle2lock(&lhc->mlh_reg_lh);
                LASSERT(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT);
-               LASSERTF(lock != NULL, "Invalid lock handle "LPX64"\n",
-                        lhc->mlh_reg_lh.cookie);
+               if (lock == NULL) {
+                       /* Lock is pinned by ldlm_handle_enqueue0() as it is
+                        * a resend case, however, it could be already destroyed
+                        * due to client eviction or a raced cancel RPC. */
+                       LDLM_DEBUG_NOLOCK("Invalid lock handle "LPX64"\n",
+                                         lhc->mlh_reg_lh.cookie);
+                       RETURN(-ESTALE);
+               }
 
                if (!fid_res_name_eq(mdt_object_fid(mo),
                                     &lock->l_resource->lr_name)) {
@@ -2819,6 +2869,7 @@ void mdt_thread_info_init(struct ptlrpc_request *req,
 
         info->mti_spec.no_create = 0;
        info->mti_spec.sp_rm_entry = 0;
+       info->mti_spec.sp_permitted = 0;
 
        info->mti_spec.u.sp_ea.eadata = NULL;
        info->mti_spec.u.sp_ea.eadatalen = 0;
@@ -2885,7 +2936,8 @@ static int mdt_tgt_connect(struct tgt_session_info *tsi)
        if (OBD_FAIL_CHECK(OBD_FAIL_TGT_DELAY_CONDITIONAL) &&
            cfs_fail_val ==
            tsi2mdt_info(tsi)->mti_mdt->mdt_seq_site.ss_node_id)
-               schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE, HZ * 3);
+               schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE,
+                       msecs_to_jiffies(3 * MSEC_PER_SEC));
 
        rc = tgt_connect(tsi);
        if (rc != 0)
@@ -3817,10 +3869,9 @@ static void mdt_stack_pre_fini(const struct lu_env *env,
        lustre_cfg_bufs_reset(bufs, mdt_obd_name(m));
        lustre_cfg_bufs_set_string(bufs, 1, NULL);
        lcfg = lustre_cfg_new(LCFG_PRE_CLEANUP, bufs);
-       if (!lcfg) {
-               CERROR("%s: cannot alloc lcfg\n", mdt_obd_name(m));
-               return;
-       }
+       if (lcfg == NULL)
+               RETURN_EXIT;
+
        top->ld_ops->ldo_process_config(env, top, lcfg);
        lustre_cfg_free(lcfg);
        EXIT;
@@ -3853,10 +3904,9 @@ static void mdt_stack_fini(const struct lu_env *env,
                strcat(flags, "A");
        lustre_cfg_bufs_set_string(bufs, 1, flags);
        lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
-       if (!lcfg) {
-               CERROR("Cannot alloc lcfg!\n");
-               return;
-       }
+       if (lcfg == NULL)
+               RETURN_EXIT;
+
        LASSERT(top);
        top->ld_ops->ldo_process_config(env, top, lcfg);
        lustre_cfg_free(lcfg);
@@ -3980,7 +4030,7 @@ static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
        lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
 
        lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
-       if (!lcfg)
+       if (lcfg == NULL)
                GOTO(free_bufs, rc = -ENOMEM);
 
        rc = class_attach(lcfg);
@@ -4002,6 +4052,8 @@ static int mdt_stack_init(const struct lu_env *env, struct mdt_device *mdt,
        lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
 
        lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
+       if (lcfg == NULL)
+               GOTO(class_detach, rc = -ENOMEM);
 
        rc = class_setup(obd, lcfg);
        if (rc)
@@ -4108,7 +4160,7 @@ static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
        lustre_cfg_bufs_set_string(bufs, 3, lprof->lp_dt);
 
        lcfg = lustre_cfg_new(LCFG_ATTACH, bufs);
-       if (!lcfg)
+       if (lcfg == NULL)
                GOTO(cleanup_mem, rc = -ENOMEM);
 
        rc = class_attach(lcfg);
@@ -4133,6 +4185,8 @@ static int mdt_quota_init(const struct lu_env *env, struct mdt_device *mdt,
                                   mdt->mdt_bottom->dd_lu_dev.ld_obd->obd_name);
 
        lcfg = lustre_cfg_new(LCFG_SETUP, bufs);
+       if (lcfg == NULL)
+               GOTO(class_detach, rc = -ENOMEM);
 
        rc = class_setup(obd, lcfg);
        if (rc)
@@ -4988,90 +5042,6 @@ static int mdt_connect_internal(struct obd_export *exp,
        return 0;
 }
 
-/* mds_connect copy */
-static int mdt_obd_connect(const struct lu_env *env,
-                           struct obd_export **exp, struct obd_device *obd,
-                           struct obd_uuid *cluuid,
-                           struct obd_connect_data *data,
-                           void *localdata)
-{
-        struct obd_export      *lexp;
-        struct lustre_handle    conn = { 0 };
-        struct mdt_device      *mdt;
-        int                     rc;
-        ENTRY;
-
-        LASSERT(env != NULL);
-        if (!exp || !obd || !cluuid)
-                RETURN(-EINVAL);
-
-       mdt = mdt_dev(obd->obd_lu_dev);
-
-       /*
-        * first, check whether the stack is ready to handle requests
-        * XXX: probably not very appropriate method is used now
-        *      at some point we should find a better one
-        */
-       if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
-           !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
-               rc = obd_get_info(env, mdt->mdt_child_exp,
-                                 sizeof(KEY_OSP_CONNECTED),
-                                 KEY_OSP_CONNECTED, NULL, NULL, NULL);
-               if (rc)
-                       RETURN(-EAGAIN);
-               set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
-       }
-
-        rc = class_connect(&conn, obd, cluuid);
-        if (rc)
-                RETURN(rc);
-
-        lexp = class_conn2export(&conn);
-        LASSERT(lexp != NULL);
-
-        rc = mdt_connect_internal(lexp, mdt, data);
-        if (rc == 0) {
-                struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
-
-                LASSERT(lcd);
-               memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
-               rc = tgt_client_new(env, lexp);
-                if (rc == 0)
-                        mdt_export_stats_init(obd, lexp, localdata);
-
-               /* For phase I, sync for cross-ref operation. */
-               lexp->exp_keep_sync = 1;
-        }
-
-        if (rc != 0) {
-                class_disconnect(lexp);
-                *exp = NULL;
-        } else {
-                *exp = lexp;
-        }
-
-        RETURN(rc);
-}
-
-static int mdt_obd_reconnect(const struct lu_env *env,
-                             struct obd_export *exp, struct obd_device *obd,
-                             struct obd_uuid *cluuid,
-                             struct obd_connect_data *data,
-                             void *localdata)
-{
-        int                     rc;
-        ENTRY;
-
-        if (exp == NULL || obd == NULL || cluuid == NULL)
-                RETURN(-EINVAL);
-
-        rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
-        if (rc == 0)
-                mdt_export_stats_init(obd, exp, localdata);
-
-        RETURN(rc);
-}
-
 static int mdt_ctxt_add_dirty_flag(struct lu_env *env,
                                   struct mdt_thread_info *info,
                                   struct mdt_file_data *mfd)
@@ -5187,6 +5157,7 @@ static int mdt_obd_disconnect(struct obd_export *exp)
         LASSERT(exp);
         class_export_get(exp);
 
+       nodemap_del_member(exp);
        rc = server_disconnect_export(exp);
        if (rc != 0)
                CDEBUG(D_IOCTL, "server disconnect error: rc = %d\n", rc);
@@ -5196,6 +5167,102 @@ static int mdt_obd_disconnect(struct obd_export *exp)
        RETURN(rc);
 }
 
+/* mds_connect copy */
+static int mdt_obd_connect(const struct lu_env *env,
+                          struct obd_export **exp, struct obd_device *obd,
+                          struct obd_uuid *cluuid,
+                          struct obd_connect_data *data,
+                          void *localdata)
+{
+       struct obd_export       *lexp;
+       struct lustre_handle    conn = { 0 };
+       struct mdt_device       *mdt;
+       int                      rc;
+       lnet_nid_t              *client_nid = localdata;
+       ENTRY;
+
+       LASSERT(env != NULL);
+       if (!exp || !obd || !cluuid)
+               RETURN(-EINVAL);
+
+       mdt = mdt_dev(obd->obd_lu_dev);
+
+       /*
+        * first, check whether the stack is ready to handle requests
+        * XXX: probably not very appropriate method is used now
+        *      at some point we should find a better one
+        */
+       if (!test_bit(MDT_FL_SYNCED, &mdt->mdt_state) && data != NULL &&
+           !(data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT)) {
+               rc = obd_get_info(env, mdt->mdt_child_exp,
+                                 sizeof(KEY_OSP_CONNECTED),
+                                 KEY_OSP_CONNECTED, NULL, NULL, NULL);
+               if (rc)
+                       RETURN(-EAGAIN);
+               set_bit(MDT_FL_SYNCED, &mdt->mdt_state);
+       }
+
+       rc = class_connect(&conn, obd, cluuid);
+       if (rc)
+               RETURN(rc);
+
+       lexp = class_conn2export(&conn);
+       LASSERT(lexp != NULL);
+
+       rc = mdt_connect_internal(lexp, mdt, data);
+       if (rc == 0) {
+               struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd;
+
+               LASSERT(lcd);
+               memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
+               rc = tgt_client_new(env, lexp);
+               if (rc == 0) {
+                       rc = nodemap_add_member(*client_nid, lexp);
+                       if (rc != 0 && rc != -EEXIST)
+                               goto out;
+
+                       mdt_export_stats_init(obd, lexp, localdata);
+               }
+
+               /* For phase I, sync for cross-ref operation. */
+               spin_lock(&lexp->exp_lock);
+               lexp->exp_keep_sync = 1;
+               spin_unlock(&lexp->exp_lock);
+       }
+out:
+       if (rc != 0) {
+               class_disconnect(lexp);
+               *exp = NULL;
+       } else {
+               *exp = lexp;
+       }
+
+       RETURN(rc);
+}
+
+static int mdt_obd_reconnect(const struct lu_env *env,
+                            struct obd_export *exp, struct obd_device *obd,
+                            struct obd_uuid *cluuid,
+                            struct obd_connect_data *data,
+                            void *localdata)
+{
+       lnet_nid_t             *client_nid = localdata;
+       int                     rc;
+       ENTRY;
+
+       if (exp == NULL || obd == NULL || cluuid == NULL)
+               RETURN(-EINVAL);
+
+       rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data);
+       if (rc == 0) {
+               rc = nodemap_add_member(*client_nid, exp);
+               if (rc == 0 || rc == -EEXIST)
+                       mdt_export_stats_init(obd, exp, localdata);
+       }
+
+       RETURN(rc);
+}
+
 /* FIXME: Can we avoid using these two interfaces? */
 static int mdt_init_export(struct obd_export *exp)
 {
@@ -5257,25 +5324,6 @@ static int mdt_destroy_export(struct obd_export *exp)
        RETURN(0);
 }
 
-/** The maximum depth that fid2path() will search.
- * This is limited only because we want to store the fids for
- * historical path lookup purposes.
- */
-#define MAX_PATH_DEPTH 100
-
-/** mdt_path() lookup structure. */
-struct path_lookup_info {
-       __u64                   pli_recno;      /**< history point */
-       __u64                   pli_currec;     /**< current record */
-       struct lu_fid           pli_fid;
-       struct lu_fid           pli_fids[MAX_PATH_DEPTH]; /**< path, in fids */
-       struct mdt_object       *pli_mdt_obj;
-       char                    *pli_path;      /**< full path */
-       int                     pli_pathlen;
-       int                     pli_linkno;     /**< which hardlink to follow */
-       int                     pli_fidcount;   /**< number of \a pli_fids */
-};
-
 int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
                   struct linkea_data *ldata)
 {
@@ -5307,8 +5355,21 @@ int mdt_links_read(struct mdt_thread_info *info, struct mdt_object *mdt_obj,
        return linkea_init(ldata);
 }
 
+/**
+ * Given an MDT object, try to look up the full path to the object.
+ * Part of the MDT layer implementation of lfs fid2path.
+ *
+ * \param[in]     info  Per-thread common data shared by MDT level handlers.
+ * \param[in]     obj   Object to do path lookup of
+ * \param[in,out] fp    User-provided struct to store path information
+ *
+ * \retval 0 Lookup successful, path information stored in fp
+ * \retval -EAGAIN Lookup failed, usually because object is being moved
+ * \retval negative errno if there was a problem
+ */
 static int mdt_path_current(struct mdt_thread_info *info,
-                           struct path_lookup_info *pli)
+                           struct mdt_object *obj,
+                           struct getinfo_fid2path *fp)
 {
        struct mdt_device       *mdt = info->mti_mdt;
        struct mdt_object       *mdt_obj;
@@ -5321,6 +5382,7 @@ static int mdt_path_current(struct mdt_thread_info *info,
        int                     reclen;
        struct linkea_data      ldata = { 0 };
        int                     rc = 0;
+       bool                    first = true;
        ENTRY;
 
        /* temp buffer for path element, the buffer will be finally freed
@@ -5330,16 +5392,14 @@ static int mdt_path_current(struct mdt_thread_info *info,
                RETURN(-ENOMEM);
 
        ldata.ld_buf = buf;
-       ptr = pli->pli_path + pli->pli_pathlen - 1;
+       ptr = fp->gf_path + fp->gf_pathlen - 1;
        *ptr = 0;
        --ptr;
-       pli->pli_fidcount = 0;
-       pli->pli_fids[0] = *(struct lu_fid *)mdt_object_fid(pli->pli_mdt_obj);
-       *tmpfid = pli->pli_fids[0];
+       *tmpfid = fp->gf_fid = *mdt_object_fid(obj);
+
        /* root FID only exists on MDT0, and fid2path should also ends at MDT0,
         * so checking root_fid can only happen on MDT0. */
-       while (!lu_fid_eq(&mdt->mdt_md_root_fid,
-                         &pli->pli_fids[pli->pli_fidcount])) {
+       while (!lu_fid_eq(&mdt->mdt_md_root_fid, &fp->gf_fid)) {
                struct lu_buf           lmv_buf;
 
                mdt_obj = mdt_object_find(info->mti_env, mdt, tmpfid);
@@ -5367,18 +5427,17 @@ static int mdt_path_current(struct mdt_thread_info *info,
                linkea_entry_unpack(lee, &reclen, tmpname, tmpfid);
                /* If set, use link #linkno for path lookup, otherwise use
                   link #0.  Only do this for the final path element. */
-               if (pli->pli_fidcount == 0 &&
-                   pli->pli_linkno < leh->leh_reccount) {
+               if (first && fp->gf_linkno < leh->leh_reccount) {
                        int count;
-                       for (count = 0; count < pli->pli_linkno; count++) {
+                       for (count = 0; count < fp->gf_linkno; count++) {
                                lee = (struct link_ea_entry *)
                                     ((char *)lee + reclen);
                                linkea_entry_unpack(lee, &reclen, tmpname,
                                                    tmpfid);
                        }
-                       if (pli->pli_linkno < leh->leh_reccount - 1)
+                       if (fp->gf_linkno < leh->leh_reccount - 1)
                                /* indicate to user there are more links */
-                               pli->pli_linkno++;
+                               fp->gf_linkno++;
                }
 
                lmv_buf.lb_buf = info->mti_xattr_buf;
@@ -5392,7 +5451,7 @@ static int mdt_path_current(struct mdt_thread_info *info,
 
                        /* For slave stripes, get its master */
                        if (le32_to_cpu(lmm->lmv_magic) == LMV_MAGIC_STRIPE) {
-                               pli->pli_fids[pli->pli_fidcount] = *tmpfid;
+                               fp->gf_fid = *tmpfid;
                                continue;
                        }
                } else if (rc < 0 && rc != -ENODATA) {
@@ -5403,72 +5462,81 @@ static int mdt_path_current(struct mdt_thread_info *info,
 
                /* Pack the name in the end of the buffer */
                ptr -= tmpname->ln_namelen;
-               if (ptr - 1 <= pli->pli_path)
+               if (ptr - 1 <= fp->gf_path)
                        GOTO(out, rc = -EOVERFLOW);
                strncpy(ptr, tmpname->ln_name, tmpname->ln_namelen);
                *(--ptr) = '/';
 
-               /* Store the parent fid for historic lookup */
-               if (++pli->pli_fidcount >= MAX_PATH_DEPTH)
-                       GOTO(out, rc = -EOVERFLOW);
-               pli->pli_fids[pli->pli_fidcount] = *tmpfid;
+               /* keep the last resolved fid to the client, so the
+                * client will build the left path on another MDT for
+                * remote object */
+               fp->gf_fid = *tmpfid;
+
+               first = false;
        }
 
 remote_out:
        ptr++; /* skip leading / */
-       memmove(pli->pli_path, ptr, pli->pli_path + pli->pli_pathlen - ptr);
+       memmove(fp->gf_path, ptr, fp->gf_path + fp->gf_pathlen - ptr);
 
-       EXIT;
 out:
-       return rc;
+       RETURN(rc);
 }
 
-/* Returns the full path to this fid, as of changelog record recno. */
+/**
+ * Given an MDT object, use mdt_path_current to get the path.
+ * Essentially a wrapper to retry mdt_path_current a set number of times
+ * if -EAGAIN is returned (usually because an object is being moved).
+ *
+ * Part of the MDT layer implementation of lfs fid2path.
+ *
+ * \param[in]     info  Per-thread common data shared by mdt level handlers.
+ * \param[in]     obj   Object to do path lookup of
+ * \param[in,out] fp    User-provided struct for arguments and to store path
+ *                     information
+ *
+ * \retval 0 Lookup successful, path information stored in fp
+ * \retval negative errno if there was a problem
+ */
 static int mdt_path(struct mdt_thread_info *info, struct mdt_object *obj,
-                   char *path, int pathlen, __u64 *recno, int *linkno,
-                   struct lu_fid *fid)
+                   struct getinfo_fid2path *fp)
 {
        struct mdt_device       *mdt = info->mti_mdt;
-       struct path_lookup_info *pli;
        int                     tries = 3;
        int                     rc = -EAGAIN;
        ENTRY;
 
-       if (pathlen < 3)
+       if (fp->gf_pathlen < 3)
                RETURN(-EOVERFLOW);
 
        if (lu_fid_eq(&mdt->mdt_md_root_fid, mdt_object_fid(obj))) {
-               path[0] = '\0';
+               fp->gf_path[0] = '\0';
                RETURN(0);
        }
 
-       OBD_ALLOC_PTR(pli);
-       if (pli == NULL)
-               RETURN(-ENOMEM);
-
-       pli->pli_mdt_obj = obj;
-       pli->pli_recno = *recno;
-       pli->pli_path = path;
-       pli->pli_pathlen = pathlen;
-       pli->pli_linkno = *linkno;
-
        /* Retry multiple times in case file is being moved */
        while (tries-- && rc == -EAGAIN)
-               rc = mdt_path_current(info, pli);
-
-       /* return the last resolved fids to the client, so the client will
-        * build the left path on another MDT for remote object */
-       *fid = pli->pli_fids[pli->pli_fidcount];
-
-       *recno = pli->pli_currec;
-       /* Return next link index to caller */
-       *linkno = pli->pli_linkno;
-
-       OBD_FREE_PTR(pli);
+               rc = mdt_path_current(info, obj, fp);
 
        RETURN(rc);
 }
 
+/**
+ * Get the full path of the provided FID, as of changelog record recno.
+ *
+ * This checks sanity and looks up object for user provided FID
+ * before calling the actual path lookup code.
+ *
+ * Part of the MDT layer implementation of lfs fid2path.
+ *
+ * \param[in]     info  Per-thread common data shared by mdt level handlers.
+ * \param[in,out] fp    User-provided struct for arguments and to store path
+ *                     information
+ *
+ * \retval 0 Lookup successful, path information and recno stored in fp
+ * \retval -ENOENT, object does not exist
+ * \retval negative errno if there was a problem
+ */
 static int mdt_fid2path(struct mdt_thread_info *info,
                        struct getinfo_fid2path *fp)
 {
@@ -5511,8 +5579,7 @@ static int mdt_fid2path(struct mdt_thread_info *info,
                RETURN(rc);
        }
 
-       rc = mdt_path(info, obj, fp->gf_path, fp->gf_pathlen, &fp->gf_recno,
-                     &fp->gf_linkno, &fp->gf_fid);
+       rc = mdt_path(info, obj, fp);
 
        CDEBUG(D_INFO, "fid "DFID", path %s recno "LPX64" linkno %u\n",
               PFID(&fp->gf_fid), fp->gf_path, fp->gf_recno, fp->gf_linkno);