Add an interface to try next fail-over server after connection to the active server was unexpectedly dropped.

5 files changed, 81 insertions, 45 deletions

diff --git a/src/providers/data_provider_fo.c b/src/providers/data_provider_fo.c
index 21059d394..dff0aef24 100644
--- a/src/providers/data_provider_fo.c
+++ b/src/providers/data_provider_fo.c
@@ -155,22 +155,42 @@ static int be_svc_data_destroy(void *memptr)
     return 0;
 }
 
-int be_fo_add_service(struct be_ctx *ctx, const char *service_name)
+/*
+ * Find registered be_svc_data by service name.
+ */
+static struct be_svc_data *be_fo_find_svc_data(struct be_ctx *ctx,
+                                               const char *service_name)
 {
-    struct fo_service *service;
     struct be_svc_data *svc;
-    int ret;
+
+    if (!ctx || !ctx->be_fo) {
+        return 0;
+    }
 
     DLIST_FOR_EACH(svc, ctx->be_fo->svcs) {
         if (strcmp(svc->name, service_name) == 0) {
-            DEBUG(6, ("Failover service already initialized!\n"));
-            /* we already have a service up and configured,
-             * can happen when using both id and auth provider
-             */
-            return EOK;
+            return svc;
         }
     }
 
+    return 0;
+}
+
+int be_fo_add_service(struct be_ctx *ctx, const char *service_name)
+{
+    struct fo_service *service;
+    struct be_svc_data *svc;
+    int ret;
+
+    svc = be_fo_find_svc_data(ctx, service_name);
+    if (svc) {
+        DEBUG(6, ("Failover service already initialized!\n"));
+        /* we already have a service up and configured,
+         * can happen when using both id and auth provider
+         */
+        return EOK;
+    }
+
     /* if not in the be service list, try to create new one */
 
     ret = fo_new_service(ctx->be_fo->fo_ctx, service_name, &service);
@@ -217,11 +237,7 @@ int be_fo_service_add_callback(TALLOC_CTX *memctx,
     struct be_svc_callback *callback;
     struct be_svc_data *svc;
 
-    DLIST_FOR_EACH(svc, ctx->be_fo->svcs) {
-        if (strcmp(svc->name, service_name) == 0) {
-            break;
-        }
-    }
+    svc = be_fo_find_svc_data(ctx, service_name);
     if (NULL == svc) {
         return ENOENT;
     }
@@ -249,11 +265,7 @@ int be_fo_add_srv_server(struct be_ctx *ctx, const char *service_name,
     char *domain;
     int ret;
 
-    DLIST_FOR_EACH(svc, ctx->be_fo->svcs) {
-        if (strcmp(svc->name, service_name) == 0) {
-            break;
-        }
-    }
+    svc = be_fo_find_svc_data(ctx, service_name);
     if (NULL == svc) {
         return ENOENT;
     }
@@ -281,16 +293,7 @@ int be_fo_get_server_count(struct be_ctx *ctx, const char *service_name)
 {
     struct be_svc_data *svc_data;
 
-    if (!ctx->be_fo) {
-        return 0;
-    }
-
-    DLIST_FOR_EACH(svc_data, ctx->be_fo->svcs) {
-        if (strcmp(svc_data->name, service_name) == 0) {
-            break;
-        }
-    }
-
+    svc_data = be_fo_find_svc_data(ctx, service_name);
     if (!svc_data) {
         return 0;
     }
@@ -304,11 +307,7 @@ int be_fo_add_server(struct be_ctx *ctx, const char *service_name,
     struct be_svc_data *svc;
     int ret;
 
-    DLIST_FOR_EACH(svc, ctx->be_fo->svcs) {
-        if (strcmp(svc->name, service_name) == 0) {
-            break;
-        }
-    }
+    svc = be_fo_find_svc_data(ctx, service_name);
     if (NULL == svc) {
         return ENOENT;
     }
@@ -349,19 +348,14 @@ struct tevent_req *be_resolve_server_send(TALLOC_CTX *memctx,
     state->ev = ev;
     state->ctx = ctx;
 
-    DLIST_FOR_EACH(svc, ctx->be_fo->svcs) {
-        if (strcmp(svc->name, service_name) == 0) {
-            state->svc = svc;
-            break;
-        }
-    }
-
+    svc = be_fo_find_svc_data(ctx, service_name);
     if (NULL == svc) {
         tevent_req_error(req, EINVAL);
         tevent_req_post(req, ev);
         return req;
     }
 
+    state->svc = svc;
     state->attempts = 0;
 
     subreq = fo_resolve_service_send(state, ev,
@@ -474,16 +468,22 @@ int be_resolve_server_recv(struct tevent_req *req, struct fo_server **srv)
     return EOK;
 }
 
+void be_fo_try_next_server(struct be_ctx *ctx, const char *service_name)
+{
+    struct be_svc_data *svc;
+
+    svc = be_fo_find_svc_data(ctx, service_name);
+    if (svc) {
+        fo_try_next_server(svc->fo_service);
+    }
+}
+
 int be_fo_run_callbacks_at_next_request(struct be_ctx *ctx,
                                         const char *service_name)
 {
     struct be_svc_data *svc;
 
-    DLIST_FOR_EACH(svc, ctx->be_fo->svcs) {
-        if (strcmp(svc->name, service_name) == 0) {
-            break;
-        }
-    }
+    svc = be_fo_find_svc_data(ctx, service_name);
     if (NULL == svc) {
         return ENOENT;
     }
diff --git a/src/providers/dp_backend.h b/src/providers/dp_backend.h
index 69872b6c4..a31e458f7 100644
--- a/src/providers/dp_backend.h
+++ b/src/providers/dp_backend.h
@@ -177,6 +177,12 @@ struct tevent_req *be_resolve_server_send(TALLOC_CTX *memctx,
                                           struct be_ctx *ctx,
                                           const char *service_name);
 int be_resolve_server_recv(struct tevent_req *req, struct fo_server **srv);
+/*
+ * Instruct fail-over to try next server on the next connect attempt.
+ * Should be used after connection to service was unexpectedly dropped
+ * but there is no authoritative information on whether active server is down.
+ */
+void be_fo_try_next_server(struct be_ctx *ctx, const char *service_name);
 
 int be_fo_run_callbacks_at_next_request(struct be_ctx *ctx,
                                         const char *service_name);
diff --git a/src/providers/fail_over.c b/src/providers/fail_over.c
index f364a8c5f..3dab38a34 100644
--- a/src/providers/fail_over.c
+++ b/src/providers/fail_over.c
@@ -1275,6 +1275,27 @@ fo_set_port_status(struct fo_server *server, enum port_status status)
     }
 }
 
+void fo_try_next_server(struct fo_service *service)
+{
+    struct fo_server *server;
+
+    if (!service) {
+        DEBUG(1, ("Bug: No service supplied\n"));
+        return;
+    }
+
+    server = service->active_server;
+    if (!server) {
+        return;
+    }
+
+    service->active_server = 0;
+
+    if (server->port_status == PORT_WORKING) {
+        server->port_status = PORT_NEUTRAL;
+    }
+}
+
 void *
 fo_get_server_user_data(struct fo_server *server)
 {
diff --git a/src/providers/fail_over.h b/src/providers/fail_over.h
index 24b2c1f5f..f1a071da9 100644
--- a/src/providers/fail_over.h
+++ b/src/providers/fail_over.h
@@ -155,6 +155,12 @@ void fo_set_server_status(struct fo_server *server,
 void fo_set_port_status(struct fo_server *server,
                         enum port_status status);
 
+/*
+ * Instruct fail-over to try next server on the next connect attempt.
+ * Should be used after connection to service was unexpectedly dropped
+ * but there is no authoritative information on whether active server is down.
+ */
+void fo_try_next_server(struct fo_service *service);
 
 void *fo_get_server_user_data(struct fo_server *server);
 
diff --git a/src/providers/ldap/sdap_id_op.c b/src/providers/ldap/sdap_id_op.c
index de498bd77..1e20c75e1 100644
--- a/src/providers/ldap/sdap_id_op.c
+++ b/src/providers/ldap/sdap_id_op.c
@@ -703,6 +703,9 @@ int sdap_id_op_done(struct sdap_id_op *op, int retval, int *dp_err_out)
             && op->conn_data == op->conn_cache->cached_connection) {
         /* do not reuse failed connection */
         op->conn_cache->cached_connection = NULL;
+
+        DEBUG(5, ("communication error on cached connection, moving to next server\n"));
+        be_fo_try_next_server(op->conn_cache->be, op->conn_cache->service->name);
     }
 
     int dp_err;