req_fsm: Use status 408 for reset streams

The 503 synth and 500 minimal response status codes are too misleading
in this context, where the failure is attributed to the client. Among
existing 4XX status codes, this is the closest if we stretch the timeout
definition to "didn't complete rapidly enough before the client went
away".

bin/varnishd/cache/cache_req_fsm.c

@@ -289,8 +289,13 @@ cnt_vclfail(struct worker *wrk, struct req *req)
 
 	Req_Rollback(ctx);
 
-	req->err_code = 503;
-	req->err_reason = "VCL failed";
+	if (req->req_reset) {
+		req->err_code = 408;
+		req->err_reason = "Client disconnected";
+	} else {
+		req->err_code = 503;
+		req->err_reason = "VCL failed";
+	}
 	req->req_step = R_STP_SYNTH;
 	req->doclose = SC_VCL_FAILURE;
 	req->filter_list = NULL;
@@ -306,6 +311,7 @@ cnt_synth(struct worker *wrk, struct req *req)
 {
 	struct vsb *synth_body;
 	ssize_t sz, szl;
+	uint16_t status;
 	uint8_t *ptr;
 	const char *body;
 
@@ -340,7 +346,8 @@ cnt_synth(struct worker *wrk, struct req *req)
 		}
 		VSB_destroy(&synth_body);
 		(void)VRB_Ignore(req);
-		(void)req->transport->minimal_response(req, 500);
+		status = req->req_reset ? 408 : 500;
+		(void)req->transport->minimal_response(req, status);
 		req->doclose = SC_VCL_FAILURE; // XXX: Not necessary any more ?
 		VSLb_ts_req(req, "Resp", W_TIM_real(wrk));
 		http_Teardown(req->resp);

bin/varnishtest/tests/t02025.vtc

@@ -46,7 +46,7 @@ varnish v1 -expect req_reset == 1
 # is interpreted as before a second elapsed. Session VXIDs showing up
 # numerous times become increasingly more suspicious. The format can of
 # course be extended to add anything else useful for data mining.
-shell -expect "1000 ${localhost}" {
+shell -expect "1000 ${localhost} 408" {
 	varnishncsa -n ${v1_name} -d \
-		-q 'Timestamp:Reset[2] < 1.0' -F '%{VSL:Begin[2]}x %h'
+		-q 'Timestamp:Reset[2] < 1.0' -F '%{VSL:Begin[2]}x %h %s'
 }