BUG/MINOR: tcp: avoid closing fd when socket failed in tcp_bind_listener
authorWilliam Dauchy <w.dauchy@criteo.com>
Wed, 12 Feb 2020 09:09:14 +0000 (10:09 +0100)
committerWilly Tarreau <w@1wt.eu>
Wed, 12 Feb 2020 14:32:44 +0000 (15:32 +0100)
we were trying to close file descriptor even when `socket` call was
failing.
this should fix github issue #499

this should be backported to all versions >= v1.8

Fixes: 153659f1ae69a1 ("MINOR: tcp: When binding socket, attempt to
reuse one from the old proc.")
Signed-off-by: William Dauchy <w.dauchy@criteo.com>
(cherry picked from commit c0e23aef051dd3916ef3eedf934837942cb7db71)
Signed-off-by: Willy Tarreau <w@1wt.eu>

src/proto_tcp.c

index 64cdc0b..ec0d981 100644 (file)
@@ -841,8 +841,8 @@ int tcp_bind_listener(struct listener *listener, char *errmsg, int errlen)
                        if (getsockopt(fd, IPPROTO_TCP, TCP_MAXSEG, &default_tcp_maxseg,
                            &ready_len) == -1)
                                ha_warning("Failed to get the default value of TCP_MAXSEG\n");
+                       close(fd);
                }
-               close(fd);
        }
        if (default_tcp6_maxseg == -1) {
                default_tcp6_maxseg = -2;