2010/11/22 Daniel P. Berrange <berrange(a)redhat.com>:
On Sat, Nov 20, 2010 at 06:10:21PM +0100, Matthias Bolte wrote:
> gnulib wraps Windows' SOCKET handle based send() and recv() functions
> into file descriptor based ones that are use in libvirt. By default
> GnuTLS uses the SOCKET handle based send() and recv() on Windows. This
> makes gnutls_handshake() fail internally with a WSAENOTSOCK error because
> libvirt passes a file descriptor; GnuTLS needs the SOCKET handle.
This doesn't entirely make any sense to me. GNUTLS also uses GNULIB,
including all its socket wrappers for send/recv. If the push/pull
function is NULL, gnulib does this
if (session->internals._gnutls_push_func == NULL)
{
i = send (GNUTLS_POINTER_TO_INT (fd), &ptr[n - left], left, 0);
And this 'send' impl maps to GNULIBs replacement in GNUTLS tree, which
is identical to libvirt's 'send' impl
So AFAICT, the syscall behaviour will be identical both with & without
your proposed patch.
The only potential difference I can see is that when push/pull are
NULL, GNUTLS calls into WSAGetLastError (); to set the errno
which is somewhat redundant as GNULIB has already called that and
used it to set errno.
Daniel
You're right GNUTLS uses GNULIB. I missed that fact. But why does my
patch make a difference then? Without it the TLS transport doesn't
work and with the patch it works. I'll have to investigate.
Matthias