bug#49449: 28: TLS connection never gets to "open" stage

["Mattias Engdegård" <mattiase@acm.org>]

[-- Attachment #1: Type: text/plain, Size: 434 bytes --]

8 juli 2021 kl. 14.54 skrev Lars Ingebrigtsen <larsi@gnus.org>:

> Sorry, nope.  :-/

I was thinking about 9c74f2fea6bf which introduced the condition of the wait mask, and maybe ac6e085cf6b2 where async TLS negotiation was added.

A hack like the one below would "solve" the problem by pretending that the first handshake attempt always fails with E_AGAIN, but it would be better to do it in a slightly more principled way.


[-- Attachment #2: tls-handshake-hack.diff --]
[-- Type: application/octet-stream, Size: 984 bytes --]

diff --git a/src/gnutls.c b/src/gnutls.c
index 22e7f2cbc1..47ac679c5c 100644
--- a/src/gnutls.c
+++ b/src/gnutls.c
@@ -623,13 +623,23 @@ gnutls_try_handshake (struct Lisp_Process *proc)
   if (non_blocking)
     proc->gnutls_p = true;
 
-  while ((ret = gnutls_handshake (state)) < 0)
+  if (non_blocking && proc->gnutls_handshakes_tried < 1)
     {
-      if (emacs_gnutls_handle_error (state, ret) == 0) /* fatal */
-	break;
+      /* HACK: don't succeed the first time for nonblocking connections,
+	 because the logic doesn't allow it (bug#49449).  */
+      ret = GNUTLS_E_AGAIN;
       maybe_quit ();
-      if (non_blocking && ret != GNUTLS_E_INTERRUPTED)
-	break;
+    }
+  else
+    {
+      while ((ret = gnutls_handshake (state)) < 0)
+	{
+	  if (emacs_gnutls_handle_error (state, ret) == 0) /* fatal */
+	    break;
+	  maybe_quit ();
+	  if (non_blocking && ret != GNUTLS_E_INTERRUPTED)
+	    break;
+	}
     }
 
   proc->gnutls_initstage = GNUTLS_STAGE_HANDSHAKE_TRIED;