BUG/MEDIUM: mux-h1: always apply the timeout on half-closed connections

The condition in h1_refresh_timeout() seems insufficient to properly
take care of the half-closed timeout, because depending on the ordering
of operations when performing the last send() to a client, the stream
may or may not still be there and we may fail to shrink the client
timeout on our last opportunity to do so.

Here we want to make sure that the timeout is always reduced when the
last chunk was sent and the shutdown completed, regardless of the
presence of a stream or not. This is what this patch does.

This should be backported as far as 2.0, and should fix the issue
reported in #541.
This commit is contained in:
Willy Tarreau 2020-09-08 15:40:57 +02:00
parent ec00251c88
commit 4313d5ae98

View File

@ -461,11 +461,19 @@ static void h1_refresh_timeout(struct h1c *h1c)
{
if (h1c->task) {
h1c->task->expire = TICK_ETERNITY;
if ((!h1c->h1s && !conn_is_back(h1c->conn)) || b_data(&h1c->obuf)) {
if (h1c->flags & H1C_F_CS_SHUTDOWN) {
/* half-closed connections switch to clientfin/serverfin
* timeouts so that we don't hang too long on clients
* that have gone away (especially in tunnel mode).
*/
h1c->task->expire = tick_add(now_ms, h1c->shut_timeout);
task_queue(h1c->task);
TRACE_DEVEL("refreshing connection's timeout (half-closed)", H1_EV_H1C_SEND, h1c->conn);
} else if ((!h1c->h1s && !conn_is_back(h1c->conn)) || b_data(&h1c->obuf)) {
/* front connections waiting for a stream, as well as any connection with
* pending data, need a timeout.
*/
h1c->task->expire = tick_add(now_ms, ((h1c->flags & (H1C_F_CS_SHUTW_NOW|H1C_F_CS_SHUTDOWN))
h1c->task->expire = tick_add(now_ms, ((h1c->flags & H1C_F_CS_SHUTW_NOW)
? h1c->shut_timeout
: h1c->timeout));
task_queue(h1c->task);