base64: fix for unsigned chars (e.g. ARM).

[ccan] / ccan / io / io.c

diff --git a/ccan/io/io.c b/ccan/io/io.c
index c5c9bee0daf0329f433eec930058bfd20905f96c..12df06d82cb6474fed825cd381f7f225fc6213b1 100644 (file)
--- a/ccan/io/io.c
+++ b/ccan/io/io.c
@@ -97,7 +97,6 @@ struct io_conn *io_new_conn_(const tal_t *ctx, int fd,
        conn->fd.fd = fd;
        conn->finish = NULL;
        conn->finish_arg = NULL;
-       list_node_init(&conn->always);
 
        if (!add_conn(conn))
                return tal_free(conn);
@@ -120,6 +119,16 @@ struct io_conn *io_new_conn_(const tal_t *ctx, int fd,
        return conn;
 }
 
+bool io_conn_exclusive(struct io_conn *conn, bool exclusive)
+{
+       return backend_set_exclusive(&conn->plan[IO_IN], exclusive);
+}
+
+bool io_conn_out_exclusive(struct io_conn *conn, bool exclusive)
+{
+       return backend_set_exclusive(&conn->plan[IO_OUT], exclusive);
+}
+
 void io_set_finish_(struct io_conn *conn,
                    void (*finish)(struct io_conn *, void *),
                    void *arg)
@@ -145,7 +154,9 @@ static struct io_plan *set_always(struct io_conn *conn,
        struct io_plan *plan = &conn->plan[dir];
 
        plan->status = IO_ALWAYS;
-       backend_new_always(conn);
+       /* Only happens on OOM, and only with non-default tal_backend. */
+       if (!backend_new_always(plan))
+               return NULL;
        return io_set_plan(conn, dir, NULL, next, arg);
 }
 
@@ -414,27 +425,14 @@ void io_ready(struct io_conn *conn, int pollflags)
                        || conn->plan[IO_IN].status == IO_POLLING_STARTED);
 }
 
-void io_do_always(struct io_conn *conn)
+void io_do_always(struct io_plan *plan)
 {
-       /* There's a corner case where the in next_plan wakes up the
-        * out, placing it in IO_ALWAYS and we end up processing it immediately,
-        * only to leave it in the always list.
-        *
-        * Yet we can't just process one, in case they are both supposed
-        * to be done, so grab state beforehand.
-        */
-       bool always_out = (conn->plan[IO_OUT].status == IO_ALWAYS);
+       struct io_conn *conn;
 
-       if (conn->plan[IO_IN].status == IO_ALWAYS)
-               if (!next_plan(conn, &conn->plan[IO_IN]))
-                       return;
+       assert(plan->status == IO_ALWAYS);
+       conn = container_of(plan, struct io_conn, plan[plan->dir]);
 
-       if (always_out) {
-               /* You can't *unalways* a conn (except by freeing, in which
-                * case next_plan() returned false */
-               assert(conn->plan[IO_OUT].status == IO_ALWAYS);
-               next_plan(conn, &conn->plan[IO_OUT]);
-       }
+       next_plan(conn, plan);
 }
 
 void io_do_wakeup(struct io_conn *conn, enum io_direction dir)
@@ -531,6 +529,18 @@ bool io_plan_out_started(const struct io_conn *conn)
        return conn->plan[IO_OUT].status == IO_POLLING_STARTED;
 }
 
+/* Despite being a TCP expert, I missed the full extent of this
+ * problem.  The legendary ZmnSCPxj implemented it (with the URL
+ * pointing to the explanation), and I imitate that here. */
+struct io_plan *io_sock_shutdown(struct io_conn *conn)
+{
+       if (shutdown(io_conn_fd(conn), SHUT_WR) != 0)
+               return io_close(conn);
+
+       /* And leave unset .*/
+       return &conn->plan[IO_IN];
+}
+       
 bool io_flush_sync(struct io_conn *conn)
 {
        struct io_plan *plan = &conn->plan[IO_OUT];