From: Azat Khuzhin <azat@libevent.org>
Date: Sun, 3 Feb 2019 15:47:14 +0000 (+0300)
Subject: test/ssl/bufferevent_wm: explicitly break the loop once client/server received enough
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=ae9b285d2d7c9b898049072c157d50769d8014ea;p=libevent

test/ssl/bufferevent_wm: explicitly break the loop once client/server received enough

There can be tricky cases (that can be reproduced by reducing
SO_RCVBUF/SO_SNDBUF to 6144, on linux, and be aware, since linux doubles
this const), when there is still write event pending, although we read
enough.

This should be fixed in a more sophisticated way, but to backport the
patch, let's simply break the loop manually.

The ssl/bufferevent_wm originally failed on solaris.
---

diff --git a/test/regress_ssl.c b/test/regress_ssl.c
index b486b803..68c28114 100644
--- a/test/regress_ssl.c
+++ b/test/regress_ssl.c
@@ -806,6 +806,7 @@ struct wm_context
 	size_t limit;
 	size_t get;
 	struct bufferevent *bev;
+	struct wm_context *neighbour;
 };
 static void
 wm_transfer(struct bufferevent *bev, void *arg)
@@ -821,6 +822,9 @@ wm_transfer(struct bufferevent *bev, void *arg)
 			ctx->server ? "server" : "client", bev));
 		bufferevent_setcb(bev, NULL, NULL, NULL, NULL);
 		bufferevent_disable(bev, EV_READ);
+		if (ctx->neighbour->get >= ctx->neighbour->limit) {
+			event_base_loopbreak(bufferevent_get_base(bev));
+		}
 	} else {
 		ctx->get += drain;
 		evbuffer_drain(in, drain);
@@ -953,6 +957,9 @@ regress_bufferevent_openssl_wm(void *arg)
 	tt_assert(bev);
 	client.bev = bev;
 
+	server.neighbour = &client;
+	client.neighbour = &server;
+
 	bufferevent_setwatermark(bev, EV_READ, 0, client.wm_high);
 	bufferevent_setcb(bev, wm_transfer, NULL, wm_eventcb, &client);