Whamcloud - gitweb
LU-12799 ptlrpc: return proper error code
[fs/lustre-release.git] / lustre / ptlrpc / import.c
index 2334fac..baf05c1 100644 (file)
@@ -614,6 +614,7 @@ static int import_select_connection(struct obd_import *imp)
                 imp->imp_conn_current = imp_conn;
         }
 
+       /* The below message is checked in conf-sanity.sh test_35[ab] */
         CDEBUG(D_HA, "%s: import %p using connection %s/%s\n",
                imp->imp_obd->obd_name, imp, imp_conn->oic_uuid.uuid,
                libcfs_nid2str(imp_conn->oic_conn->c_peer.nid));
@@ -638,7 +639,8 @@ static int ptlrpc_first_transno(struct obd_import *imp, __u64 *transno)
                req = list_entry(tmp, struct ptlrpc_request, rq_replay_list);
                *transno = req->rq_transno;
                if (req->rq_transno == 0) {
-                       DEBUG_REQ(D_ERROR, req, "zero transno in committed_list");
+                       DEBUG_REQ(D_ERROR, req,
+                                 "zero transno in committed_list");
                        LBUG();
                }
                return 1;
@@ -656,13 +658,22 @@ static int ptlrpc_first_transno(struct obd_import *imp, __u64 *transno)
        return 0;
 }
 
+int ptlrpc_connect_import(struct obd_import *imp)
+{
+       spin_lock(&imp->imp_lock);
+       return ptlrpc_connect_import_locked(imp);
+}
+
 /**
  * Attempt to (re)connect import \a imp. This includes all preparations,
  * initializing CONNECT RPC request and passing it to ptlrpcd for
  * actual sending.
+ *
+ * Assumes imp->imp_lock is held, and releases it.
+ *
  * Returns 0 on success or error code.
  */
-int ptlrpc_connect_import(struct obd_import *imp)
+int ptlrpc_connect_import_locked(struct obd_import *imp)
 {
        struct obd_device *obd = imp->imp_obd;
        int initial_connect = 0;
@@ -680,7 +691,8 @@ int ptlrpc_connect_import(struct obd_import *imp)
        int rc;
        ENTRY;
 
-       spin_lock(&imp->imp_lock);
+       assert_spin_locked(&imp->imp_lock);
+
        if (imp->imp_state == LUSTRE_IMP_CLOSED) {
                spin_unlock(&imp->imp_lock);
                CERROR("can't connect to a closed import\n");
@@ -1266,34 +1278,38 @@ static int ptlrpc_connect_interpret(const struct lu_env *env,
                   !imp->imp_invalid) {
 
                obd_import_event(imp->imp_obd, imp, IMP_EVENT_INVALIDATE);
+               /* The below message is checked in recovery-small.sh test_106 */
                DEBUG_REQ(D_HA, request, "%s: lwp recover",
                          imp->imp_obd->obd_name);
                imp->imp_remote_handle =
                        *lustre_msg_get_handle(request->rq_repmsg);
                import_set_state(imp, LUSTRE_IMP_RECOVER);
        } else {
-                DEBUG_REQ(D_HA, request, "%s: evicting (reconnect/recover flags"
-                          " not set: %x)", imp->imp_obd->obd_name, msg_flags);
-                imp->imp_remote_handle =
-                                *lustre_msg_get_handle(request->rq_repmsg);
+               DEBUG_REQ(D_HA, request,
+                         "%s: evicting (reconnect/recover flags not set: %x)",
+                         imp->imp_obd->obd_name, msg_flags);
+               imp->imp_remote_handle =
+                       *lustre_msg_get_handle(request->rq_repmsg);
                import_set_state(imp, LUSTRE_IMP_EVICTED);
-        }
+       }
 
-        /* Sanity checks for a reconnected import. */
-        if (!(imp->imp_replayable) != !(msg_flags & MSG_CONNECT_REPLAYABLE)) {
-                CERROR("imp_replayable flag does not match server "
-                       "after reconnect. We should LBUG right here.\n");
-        }
+       /* Sanity checks for a reconnected import. */
+       if (!(imp->imp_replayable) != !(msg_flags & MSG_CONNECT_REPLAYABLE))
+               CERROR("imp_replayable flag does not match server after reconnect. We should LBUG right here.\n");
 
-        if (lustre_msg_get_last_committed(request->rq_repmsg) > 0 &&
-            lustre_msg_get_last_committed(request->rq_repmsg) <
-            aa->pcaa_peer_committed) {
-               CERROR("%s went back in time (transno %lld"
-                      " was previously committed, server now claims %lld"
-                       ")!  See https://bugzilla.lustre.org/show_bug.cgi?"
-                       "id=9646\n",
+       if (lustre_msg_get_last_committed(request->rq_repmsg) > 0 &&
+           lustre_msg_get_last_committed(request->rq_repmsg) <
+           aa->pcaa_peer_committed) {
+               static bool printed;
+
+               /* The below message is checked in recovery-small.sh test_54 */
+               CERROR("%s: went back in time (transno %lld was previously committed, server now claims %lld)!\n",
                        obd2cli_tgt(imp->imp_obd), aa->pcaa_peer_committed,
                        lustre_msg_get_last_committed(request->rq_repmsg));
+               if (!printed) {
+                       CERROR("For further information, see http://doc.lustre.org/lustre_manual.xhtml#went_back_in_time\n");
+                       printed = true;
+               }
         }
 
 finish:
@@ -1630,7 +1646,7 @@ static struct ptlrpc_request *ptlrpc_disconnect_prep_req(struct obd_import *imp)
        req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_DISCONNECT,
                                        LUSTRE_OBD_VERSION, rq_opc);
        if (req == NULL)
-               RETURN(NULL);
+               RETURN(ERR_PTR(-ENOMEM));
 
        /* We are disconnecting, do not retry a failed DISCONNECT rpc if
         * it fails.  We can get through the above with a down server
@@ -1741,7 +1757,7 @@ static int ptlrpc_disconnect_idle_interpret(const struct lu_env *env,
        struct obd_import *imp = req->rq_import;
        int connect = 0;
 
-       DEBUG_REQ(D_HA, req, "inflight=%d, refcount=%d: rc = %d ",
+       DEBUG_REQ(D_HA, req, "inflight=%d, refcount=%d: rc = %d",
                  atomic_read(&imp->imp_inflight),
                  atomic_read(&imp->imp_refcount), rc);
 
@@ -1764,12 +1780,13 @@ static int ptlrpc_disconnect_idle_interpret(const struct lu_env *env,
                        connect = 1;
                }
        }
-       spin_unlock(&imp->imp_lock);
 
        if (connect) {
-               rc = ptlrpc_connect_import(imp);
+               rc = ptlrpc_connect_import_locked(imp);
                if (rc >= 0)
                        ptlrpc_pinger_add_import(imp);
+       } else {
+               spin_unlock(&imp->imp_lock);
        }
 
        return 0;