r9409: fix a problem that volker noticed with web page timeouts causing smbd

[samba.git] / source4 / web_server / web_server.c

diff --git a/source4/web_server/web_server.c b/source4/web_server/web_server.c
index 4962185526707c4fdf838a0484e2de8e3e291e78..e85c4f20c79cc448d06e6fe06086e1f224b23d5c 100644 (file)
--- a/source4/web_server/web_server.c
+++ b/source4/web_server/web_server.c
@@ -52,7 +52,11 @@ static void websrv_timeout(struct event_context *event_context,
                           struct timeval t, void *private)
 {
        struct websrv_context *web = talloc_get_type(private, struct websrv_context);
-       stream_terminate_connection(web->conn, "websrv_timeout: timed out");
+       struct stream_connection *conn = web->conn;
+       web->conn = NULL;
+       /* TODO: send a message to any running esp context on this connection
+          to stop running */
+       stream_terminate_connection(conn, "websrv_timeout: timed out"); 
 }
 
 /*
@@ -108,7 +112,17 @@ static void websrv_recv(struct stream_connection *conn, uint16_t flags)
                        web->input.partial.data[web->input.content_length] = 0;
                }
                EVENT_FD_NOT_READABLE(web->conn->event.fde);
+
+               /* the reference/unlink code here is quite subtle. It
+                is needed because the rendering of the web-pages, and
+                in particular the esp/ejs backend, is semi-async.  So
+                we could well end up in the connection timeout code
+                while inside http_process_input(), but we must not
+                destroy the stack variables being used by that
+                rendering process when we handle the timeout. */
+               talloc_reference(web->task, web);
                http_process_input(web);
+               talloc_unlink(web->task, web);
        }
        return;