Whamcloud - gitweb
LU-14930 mdt: abort_recov_mdt shouldn't abort client recovery
[fs/lustre-release.git] / lustre / ldlm / ldlm_lib.c
index 3626bda..64214e2 100644 (file)
@@ -27,7 +27,6 @@
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
  */
 
 /**
@@ -176,9 +175,10 @@ int client_import_add_nids_to_conn(struct obd_import *imp, lnet_nid_t *nids,
        list_for_each_entry(conn, &imp->imp_conn_list, oic_item) {
                if (class_check_uuid(&conn->oic_uuid, nids[0])) {
                        *uuid = conn->oic_uuid;
+                       spin_unlock(&imp->imp_lock);
                        rc = class_add_nids_to_uuid(&conn->oic_uuid, nids,
                                                    nid_count);
-                       break;
+                       RETURN(rc);
                }
        }
        spin_unlock(&imp->imp_lock);
@@ -983,6 +983,7 @@ static int rev_import_flags_update(struct obd_import *revimp,
 
        revimp->imp_msghdr_flags |= MSGHDR_CKSUM_INCOMPAT18;
 
+       revimp->imp_connect_data = *data;
        rc = sptlrpc_import_sec_adapt(revimp, req->rq_svc_ctx, &req->rq_flvr);
        if (rc) {
                CERROR("%s: cannot get reverse import %s security: rc = %d\n",
@@ -1097,6 +1098,7 @@ int target_handle_connect(struct ptlrpc_request *req)
        struct obd_connect_data *data, *tmpdata;
        int size, tmpsize;
        lnet_nid_t *client_nid = NULL;
+       struct ptlrpc_connection *pcon = NULL;
 
        ENTRY;
 
@@ -1202,13 +1204,12 @@ int target_handle_connect(struct ptlrpc_request *req)
         * processing, so we needs to allow lw_client to be connected at
         * anytime, instead of only the initial connection
         */
-       lw_client = (data->ocd_connect_flags & OBD_CONNECT_LIGHTWEIGHT) != 0;
+       lw_client = OCD_HAS_FLAG(data, LIGHTWEIGHT);
 
        if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_INITIAL) {
                initial_conn = true;
-               mds_conn = (data->ocd_connect_flags & OBD_CONNECT_MDS) != 0;
-               mds_mds_conn = (data->ocd_connect_flags &
-                               OBD_CONNECT_MDS_MDS) != 0;
+               mds_conn = OCD_HAS_FLAG(data, MDS);
+               mds_mds_conn = OCD_HAS_FLAG(data, MDS_MDS);
 
                /*
                 * OBD_CONNECT_MNE_SWAB is removed at 2.14
@@ -1266,26 +1267,27 @@ int target_handle_connect(struct ptlrpc_request *req)
                export = NULL;
                rc = -EALREADY;
        } else if ((mds_conn || (lw_client && initial_conn) ||
-                  data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) &&
-                  export->exp_connection != NULL) {
+                  OCD_HAS_FLAG(data, MDS_MDS)) && export->exp_connection) {
                spin_unlock(&export->exp_lock);
                if (req->rq_peer.nid != export->exp_connection->c_peer.nid) {
                        /* MDS or LWP reconnected after failover. */
                        LCONSOLE_WARN("%s: Received %s connection from %s, removing former export from %s\n",
                                      target->obd_name,
-                                     mds_conn ? "MDS" : "LWP",
+                                     lw_client ? "LWP" : "MDS",
                                      libcfs_nid2str(req->rq_peer.nid),
                                      libcfs_nid2str(export->exp_connection->c_peer.nid));
                } else {
-                       /* New MDS connection from the same NID. */
-                       LCONSOLE_WARN("%s: Received new %s connection from %s, removing former export from same NID\n",
+                       /* New connection from the same NID. */
+                       LCONSOLE_WARN("%s: Received new %s connection from %s, %s former export from same NID\n",
                                      target->obd_name,
-                                     mds_conn ? "MDS" : "LWP",
-                                     libcfs_nid2str(req->rq_peer.nid));
+                                     lw_client ? "LWP" : "MDS",
+                                     libcfs_nid2str(req->rq_peer.nid),
+                                     OCD_HAS_FLAG(data, MDS_MDS) ?
+                                     "keep" : "remove");
                }
 
                if (req->rq_peer.nid == export->exp_connection->c_peer.nid &&
-                   data->ocd_connect_flags & OBD_CONNECT_MDS_MDS) {
+                   OCD_HAS_FLAG(data, MDS_MDS)) {
                        /*
                         * Because exports between MDTs will always be
                         * kept, let's do not fail such export if they
@@ -1343,7 +1345,7 @@ no_export:
        } else if (lustre_msg_get_conn_cnt(req->rq_reqmsg) == 1 &&
                   rc != EALREADY) {
                if (!strstr(cluuid.uuid, "mdt"))
-                       LCONSOLE_WARN("%s: Rejecting reconnect from the known client %s (at %s) because it is indicating it is a new client",
+                       LCONSOLE_WARN("%s: Rejecting reconnect from the known client %s (at %s) because it is indicating it is a new client\n",
                                      target->obd_name, cluuid.uuid,
                                      libcfs_nid2str(req->rq_peer.nid));
                GOTO(out, rc = -EALREADY);
@@ -1476,7 +1478,16 @@ dont_check_exports:
         */
        ptlrpc_request_change_export(req, export);
 
+       pcon = ptlrpc_connection_get(req->rq_peer, req->rq_self, &cluuid);
+       if (pcon == NULL)
+               GOTO(out, rc = -ENOTCONN);
+
        spin_lock(&export->exp_lock);
+
+       if (export->exp_disconnected) {
+               spin_unlock(&export->exp_lock);
+               GOTO(out, rc = -ENODEV);
+       }
        if (export->exp_conn_cnt >= lustre_msg_get_conn_cnt(req->rq_reqmsg)) {
                spin_unlock(&export->exp_lock);
                CDEBUG(D_RPCTRACE,
@@ -1489,24 +1500,23 @@ dont_check_exports:
        }
        LASSERT(lustre_msg_get_conn_cnt(req->rq_reqmsg) > 0);
        export->exp_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
-       spin_unlock(&export->exp_lock);
-
-       if (export->exp_connection != NULL) {
-               /* Check to see if connection came from another NID. */
-               if (export->exp_connection->c_peer.nid != req->rq_peer.nid)
-                       obd_nid_del(export->exp_obd, export);
 
+       /* Check to see if connection came from another NID. */
+       if (export->exp_connection != NULL &&
+           export->exp_connection->c_peer.nid != req->rq_peer.nid) {
+               obd_nid_del(export->exp_obd, export);
                ptlrpc_connection_put(export->exp_connection);
+               export->exp_connection = NULL;
        }
 
-       export->exp_connection = ptlrpc_connection_get(req->rq_peer,
-                                                      req->rq_self,
-                                                      &cluuid);
-       if (!export->exp_connection)
-               GOTO(out, rc = -ENOTCONN);
-
+       if (export->exp_connection == NULL) {
+               export->exp_connection = pcon;
+               pcon = NULL;
+       }
        obd_nid_add(export->exp_obd, export);
 
+       spin_unlock(&export->exp_lock);
+
        lustre_msg_set_handle(req->rq_repmsg, &conn);
 
        rc = rev_import_reconnect(export, req);
@@ -1582,6 +1592,8 @@ out:
                spin_unlock(&target->obd_dev_lock);
                class_decref(target, "find", current);
        }
+       if (pcon)
+               ptlrpc_connection_put(pcon);
        req->rq_status = rc;
        RETURN(rc);
 }
@@ -1737,6 +1749,8 @@ static void target_finish_recovery(struct lu_target *lut)
                              atomic_read(&obd->obd_connected_clients),
                              obd->obd_stale_clients,
                              obd->obd_stale_clients == 1 ? "was" : "were");
+               if (obd->obd_stale_clients && do_dump_on_eviction(obd))
+                       libcfs_debug_dumplog();
        }
 
        ldlm_reprocess_recovery_done(obd->obd_namespace);
@@ -1829,6 +1843,7 @@ void target_cleanup_recovery(struct obd_device *obd)
                return;
        }
        obd->obd_recovering = obd->obd_abort_recovery = 0;
+       obd->obd_abort_recov_mdt = 0;
        spin_unlock(&obd->obd_dev_lock);
 
        spin_lock(&obd->obd_recovery_task_lock);
@@ -2715,6 +2730,7 @@ static int target_recovery_thread(void *arg)
 
        thread->t_env = env;
        thread->t_id = -1; /* force filter_iobuf_get/put to use local buffers */
+       thread->t_task = current;
        env->le_ctx.lc_thread = thread;
        tgt_io_thread_init(thread); /* init thread_big_cache for IO requests */
 
@@ -2777,6 +2793,7 @@ static int target_recovery_thread(void *arg)
         */
        spin_lock(&obd->obd_dev_lock);
        obd->obd_recovering = obd->obd_abort_recovery = 0;
+       obd->obd_abort_recov_mdt = 0;
        spin_unlock(&obd->obd_dev_lock);
        spin_lock(&obd->obd_recovery_task_lock);
        target_cancel_recovery_timer(obd);