Whamcloud - gitweb
LU-1095 debug: conditionally print recovery status
[fs/lustre-release.git] / lustre / ldlm / ldlm_lib.c
index 2fa220b..54780a7 100644 (file)
@@ -266,7 +266,9 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg)
 
         /* In a more perfect world, we would hang a ptlrpc_client off of
          * obd_type and just use the values from there. */
-        if (!strcmp(name, LUSTRE_OSC_NAME)) {
+       if (!strcmp(name, LUSTRE_OSC_NAME) ||
+           (!strcmp(name, LUSTRE_OSP_NAME) &&
+            !is_osp_on_ost(lustre_cfg_buf(lcfg, 0)))) {
                 rq_portal = OST_REQUEST_PORTAL;
                 rp_portal = OSC_REPLY_PORTAL;
                 connect_op = OST_CONNECT;
@@ -274,7 +276,9 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg)
                 cli->cl_sp_to = LUSTRE_SP_OST;
                 ns_type = LDLM_NS_TYPE_OSC;
 
-        } else if (!strcmp(name, LUSTRE_MDC_NAME)) {
+       } else if (!strcmp(name, LUSTRE_MDC_NAME) ||
+                  (!strcmp(name, LUSTRE_OSP_NAME) &&
+                   is_osp_on_ost(lustre_cfg_buf(lcfg, 0)))) {
                 rq_portal = MDS_REQUEST_PORTAL;
                 rp_portal = MDC_REPLY_PORTAL;
                 connect_op = MDS_CONNECT;
@@ -291,6 +295,14 @@ int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg)
                 cli->cl_flvr_mgc.sf_rpc = SPTLRPC_FLVR_INVALID;
                 ns_type = LDLM_NS_TYPE_MGC;
 
+       } else if (!strcmp(name, LUSTRE_OSP_NAME)) {
+               rq_portal = OST_REQUEST_PORTAL;
+               rp_portal = OSC_REPLY_PORTAL;
+               connect_op = OST_CONNECT;
+               cli->cl_sp_me = LUSTRE_SP_CLI;
+               cli->cl_sp_to = LUSTRE_SP_OST;
+               ns_type = LDLM_NS_TYPE_OSC;
+
         } else {
                 CERROR("unknown client OBD type \"%s\", can't setup\n",
                        name);
@@ -744,7 +756,7 @@ int target_handle_connect(struct ptlrpc_request *req)
         int rc = 0;
         char *target_start;
         int target_len;
-        int mds_conn = 0;
+       bool     mds_conn = false, lw_client = false;
         struct obd_connect_data *data, *tmpdata;
         int size, tmpsize;
         lnet_nid_t *client_nid = NULL;
@@ -870,7 +882,10 @@ int target_handle_connect(struct ptlrpc_request *req)
 
         if ((lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_INITIAL) &&
             (data->ocd_connect_flags & OBD_CONNECT_MDS))
-                mds_conn = 1;
+               mds_conn = true;
+
+       if ((data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) != 0)
+               lw_client = true;
 
         /* lctl gets a backstage, all-access pass. */
         if (obd_uuid_equals(&cluuid, &target->obd_uuid))
@@ -881,7 +896,11 @@ int target_handle_connect(struct ptlrpc_request *req)
                 goto no_export;
 
         /* we've found an export in the hash */
+
+       cfs_spin_lock(&export->exp_lock);
+
         if (export->exp_connecting) { /* bug 9635, et. al. */
+               cfs_spin_unlock(&export->exp_lock);
                 LCONSOLE_WARN("%s: Export %p already connecting from %s\n",
                               export->exp_obd->obd_name, export,
                               libcfs_nid2str(req->rq_peer.nid));
@@ -889,6 +908,7 @@ int target_handle_connect(struct ptlrpc_request *req)
                 export = NULL;
                 rc = -EALREADY;
         } else if (mds_conn && export->exp_connection) {
+               cfs_spin_unlock(&export->exp_lock);
                 if (req->rq_peer.nid != export->exp_connection->c_peer.nid)
                         /* mds reconnected after failover */
                         LCONSOLE_WARN("%s: Received MDS connection from "
@@ -908,6 +928,7 @@ int target_handle_connect(struct ptlrpc_request *req)
                    req->rq_peer.nid != export->exp_connection->c_peer.nid &&
                    (lustre_msg_get_op_flags(req->rq_reqmsg) &
                     MSG_CONNECT_INITIAL)) {
+               cfs_spin_unlock(&export->exp_lock);
                 /* in mds failover we have static uuid but nid can be
                  * changed*/
                 LCONSOLE_WARN("%s: Client %s seen on new nid %s when "
@@ -920,13 +941,12 @@ int target_handle_connect(struct ptlrpc_request *req)
                 class_export_put(export);
                 export = NULL;
         } else {
-                cfs_spin_lock(&export->exp_lock);
-                export->exp_connecting = 1;
-                cfs_spin_unlock(&export->exp_lock);
-                LASSERT(export->exp_obd == target);
+               export->exp_connecting = 1;
+               cfs_spin_unlock(&export->exp_lock);
+               LASSERT(export->exp_obd == target);
 
-                rc = target_handle_reconnect(&conn, export, &cluuid);
-        }
+               rc = target_handle_reconnect(&conn, export, &cluuid);
+       }
 
         /* If we found an export, we already unlocked. */
         if (!export) {
@@ -978,8 +998,8 @@ no_export:
               export ? (long)export->exp_last_request_time : 0);
 
         /* If this is the first time a client connects, reset the recovery
-         * timer */
-        if (rc == 0 && target->obd_recovering)
+        * timer. Discard lightweight connections which might be local */
+       if (!lw_client && rc == 0 && target->obd_recovering)
                 check_and_start_recovery_timer(target, req, export == NULL);
 
         /* We want to handle EALREADY but *not* -EALREADY from
@@ -997,7 +1017,8 @@ no_export:
         client_nid = &req->rq_peer.nid;
 
         if (export == NULL) {
-                if (target->obd_recovering) {
+               /* allow lightweight connections during recovery */
+               if (target->obd_recovering && !lw_client) {
                         cfs_time_t t;
                        int        c; /* connected */
                        int        i; /* in progress */
@@ -1128,7 +1149,7 @@ dont_check_exports:
                              &export->exp_nid_hash);
         }
 
-        if (target->obd_recovering && !export->exp_in_recovery) {
+       if (target->obd_recovering && !export->exp_in_recovery && !lw_client) {
                 int has_transno;
                 __u64 transno = data->ocd_transno;
 
@@ -1168,7 +1189,7 @@ dont_check_exports:
         }
 
         /* Tell the client we're in recovery, when client is involved in it. */
-        if (target->obd_recovering)
+       if (target->obd_recovering && !lw_client)
                 lustre_msg_add_op_flags(req->rq_repmsg, MSG_CONNECT_RECOVERING);
 
         tmp = req_capsule_client_get(&req->rq_pill, &RMF_CONN);
@@ -1355,17 +1376,20 @@ static void target_exp_dequeue_req_replay(struct ptlrpc_request *req)
 #ifdef __KERNEL__
 static void target_finish_recovery(struct obd_device *obd)
 {
-        time_t elapsed_time = max_t(time_t, 1, cfs_time_current_sec() -
-                                    obd->obd_recovery_start);
         ENTRY;
 
-        LCONSOLE_INFO("%s: Recovery over after %d:%.02d, of %d clients "
-                      "%d recovered and %d %s evicted.\n", obd->obd_name,
-                      (int)elapsed_time / 60, (int)elapsed_time % 60,
-                      obd->obd_max_recoverable_clients,
-                      cfs_atomic_read(&obd->obd_connected_clients),
-                      obd->obd_stale_clients,
-                      obd->obd_stale_clients == 1 ? "was" : "were");
+       /* only log a recovery message when recovery has occurred */
+       if (obd->obd_recovery_start) {
+               time_t elapsed_time = max_t(time_t, 1, cfs_time_current_sec() -
+                                       obd->obd_recovery_start);
+               LCONSOLE_INFO("%s: Recovery over after %d:%.02d, of %d clients "
+                       "%d recovered and %d %s evicted.\n", obd->obd_name,
+                       (int)elapsed_time / 60, (int)elapsed_time % 60,
+                       obd->obd_max_recoverable_clients,
+                       cfs_atomic_read(&obd->obd_connected_clients),
+                       obd->obd_stale_clients,
+                       obd->obd_stale_clients == 1 ? "was" : "were");
+       }
 
         ldlm_reprocess_all_ns(obd->obd_namespace);
         cfs_spin_lock(&obd->obd_recovery_task_lock);
@@ -2484,119 +2508,6 @@ void target_send_reply(struct ptlrpc_request *req, int rc, int fail_id)
        EXIT;
 }
 
-int target_handle_qc_callback(struct ptlrpc_request *req)
-{
-        struct obd_quotactl *oqctl;
-        struct client_obd *cli = &req->rq_export->exp_obd->u.cli;
-
-        oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
-        if (oqctl == NULL) {
-                CERROR("Can't unpack obd_quotactl\n");
-                RETURN(-EPROTO);
-        }
-
-        cli->cl_qchk_stat = oqctl->qc_stat;
-
-        return 0;
-}
-
-#ifdef HAVE_QUOTA_SUPPORT
-int target_handle_dqacq_callback(struct ptlrpc_request *req)
-{
-#ifdef __KERNEL__
-        struct obd_device *obd = req->rq_export->exp_obd;
-        struct obd_device *master_obd = NULL, *lov_obd = NULL;
-        struct obd_device_target *obt;
-        struct lustre_quota_ctxt *qctxt;
-        struct qunit_data *qdata = NULL;
-        int rc = 0;
-        ENTRY;
-
-        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DROP_QUOTA_REQ))
-                RETURN(rc);
-
-        rc = req_capsule_server_pack(&req->rq_pill);
-        if (rc) {
-                CERROR("packing reply failed!: rc = %d\n", rc);
-                RETURN(rc);
-        }
-
-        LASSERT(req->rq_export);
-
-        qdata = quota_get_qdata(req, QUOTA_REQUEST, QUOTA_EXPORT);
-        if (IS_ERR(qdata)) {
-                rc = PTR_ERR(qdata);
-                CDEBUG(D_ERROR, "Can't unpack qunit_data(rc: %d)\n", rc);
-                req->rq_status = rc;
-                GOTO(out, rc);
-        }
-
-        /* we use the observer */
-        if (obd_pin_observer(obd, &lov_obd) ||
-            obd_pin_observer(lov_obd, &master_obd)) {
-                CERROR("Can't find the observer, it is recovering\n");
-                req->rq_status = -EAGAIN;
-                GOTO(out, rc);
-        }
-
-        obt = &master_obd->u.obt;
-        qctxt = &obt->obt_qctxt;
-
-        if (!qctxt->lqc_setup || !qctxt->lqc_valid) {
-                /* quota_type has not been processed yet, return EAGAIN
-                 * until we know whether or not quotas are supposed to
-                 * be enabled */
-                CDEBUG(D_QUOTA, "quota_type not processed yet, return "
-                       "-EAGAIN\n");
-                req->rq_status = -EAGAIN;
-                GOTO(out, rc);
-        }
-
-        cfs_down_read(&obt->obt_rwsem);
-        if (qctxt->lqc_lqs_hash == NULL) {
-                cfs_up_read(&obt->obt_rwsem);
-                /* quota_type has not been processed yet, return EAGAIN
-                 * until we know whether or not quotas are supposed to
-                 * be enabled */
-                CDEBUG(D_QUOTA, "quota_ctxt is not ready yet, return "
-                       "-EAGAIN\n");
-                req->rq_status = -EAGAIN;
-                GOTO(out, rc);
-        }
-
-        LASSERT(qctxt->lqc_handler);
-        rc = qctxt->lqc_handler(master_obd, qdata,
-                                lustre_msg_get_opc(req->rq_reqmsg));
-        cfs_up_read(&obt->obt_rwsem);
-        if (rc && rc != -EDQUOT)
-                CDEBUG(rc == -EBUSY  ? D_QUOTA : D_ERROR,
-                       "dqacq/dqrel failed! (rc:%d)\n", rc);
-        req->rq_status = rc;
-
-        rc = quota_copy_qdata(req, qdata, QUOTA_REPLY, QUOTA_EXPORT);
-        if (rc < 0) {
-                CERROR("Can't pack qunit_data(rc: %d)\n", rc);
-                GOTO(out, rc);
-        }
-
-        /* Block the quota req. b=14840 */
-        OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_BLOCK_QUOTA_REQ, obd_timeout);
-        EXIT;
-
-out:
-        if (master_obd)
-                obd_unpin_observer(lov_obd);
-        if (lov_obd)
-                obd_unpin_observer(obd);
-
-        rc = ptlrpc_reply(req);
-        return rc;
-#else
-        return 0;
-#endif /* !__KERNEL__ */
-}
-#endif /* HAVE_QUOTA_SUPPORT */
-
 ldlm_mode_t lck_compat_array[] = {
         [LCK_EX] LCK_COMPAT_EX,
         [LCK_PW] LCK_COMPAT_PW,