base: Fix a bug where enabling read when an error comes in

If you enable read on a gensio when an error has already come in and the
gensio is going through error shutdown, the enable can be missed and the
read callback reporting the error won't happen.

Expand the states where a read enable will cause a deferred op.

Signed-off-by: Corey Minyard <[email protected]>

Author: cminyard

ISSUES

@@ -9,14 +9,6 @@ this in test_mdns.py if there are multiple callbacks to mdns_cb.
 There is not an easy workaround, it is probably better to catch
 exceptions in all callbacks.
 
-Windows has some strange behavior that I can't track down dealing with
-socket closing.  It looks like in some cases that a close won't be
-reported in winsock_func() if the source and destination sockets are
-both in the same process.  That seems to be why
-test_ssl_tcp_accept_connect.py fails occassionally, I (think I) have
-verified that one side of the connection gets closed but the close is
-not reported on the other side.
-
 On Windows, a new pty spews out a bunch of terminal control stuff and
 messes things up for testing and for things like rsync.  Some way to
 disable this is needed.  This breaks test_pty_basic.py.  See

lib/gensio_base.c

@@ -1461,7 +1461,9 @@ basen_set_read_callback_enable(struct basen_data *ndata, bool enabled)
 	/* Nothing to do, let the read/open handling wake things up. */
 	ndata->deferred_read = true;
     } else if (enabled && (read_pending || ndata->ll_err) &&
-	       ndata->state == BASEN_OPEN) {
+	       (ndata->state == BASEN_OPEN ||
+		ndata->state == BASEN_IO_ERR_CLOSE ||
+		ndata->state == BASEN_IN_LL_IO_ERR_CLOSE)) {
 	ndata->deferred_read = true;
 	basen_sched_deferred_op(ndata);
     } else {