homepage

Python SSL doesn't support DH ciphers in in all version tested.
This is a serious security issue because it's not possible to use as a server or client Perfect Forward Secrecy [1] security provided by DHE and ECDH ciphers .
In order to enable DH ciphers the SSL implementation the in the file Modules/_ssl.c, it must issue a DH_generate_parameters() if a cipher is DH.
For example PHP handling of DH ciphers, look php-5.3.8/ext/openssl/openssl.c : 
#if !defined(NO_DH)
 case OPENSSL_KEYTYPE_DH:
 {
 DH *dhpar = DH_generate_parameters(req->priv_key_bits, 2, NULL, NULL);
 int codes = 0;
 if (dhpar) {
 DH_set_method(dhpar, DH_get_default_method());
 if (DH_check(dhpar, &codes) && codes == 0 && DH_generate_key(dhpar)) {
 if (EVP_PKEY_assign_DH(req->priv_key, dhpar)) {
 return_val = req->priv_key;
 }
 } else {
 DH_free(dhpar);
 }
 }
 }
 break;
#endif
 default:
An important security fix, to support and enable by default DH ciphers has to be done.
[1] http://en.wikipedia.org/wiki/Perfect_forward_secrecy 

Other example for DH and ECC from:
https://github.com/bumptech/stud/blob/master/stud.c
#ifndef OPENSSL_NO_DH
static int init_dh(SSL_CTX *ctx, const char *cert) {
 DH *dh;
 BIO *bio;
 assert(cert);
 bio = BIO_new_file(cert, "r");
 if (!bio) {
 ERR_print_errors_fp(stderr);
 return -1;
 }
 dh = PEM_read_bio_DHparams(bio, NULL, NULL, NULL);
 BIO_free(bio);
 if (!dh) {
 ERR("{core} Note: no DH parameters found in %s\n", cert);
 return -1;
 }
 LOG("{core} Using DH parameters from %s\n", cert);
 SSL_CTX_set_tmp_dh(ctx, dh);
 LOG("{core} DH initialized with %d bit key\n", 8*DH_size(dh));
 DH_free(dh);
#ifdef NID_X9_62_prime256v1
 EC_KEY *ecdh = NULL;
 ecdh = EC_KEY_new_by_curve_name(NID_X9_62_prime256v1);
 SSL_CTX_set_tmp_ecdh(ctx,ecdh);
 EC_KEY_free(ecdh);
 LOG("{core} ECDH Initialized with NIST P-256\n");
#endif
 return 0;
}
#endif /* OPENSSL_NO_DH */

The ssl module doesn't directly handle keys, it just gives a PEM file to OpenSSL's ssl functions. So I don't understand what should be done precisely here, or even if something has to be done at all.

Please look at how PHP implement the feature.
It doesn't use any PEM or any Key File, but just initiatlize the DH parameters.
Stud instead, ask the user to generate "offline" the DH parameters and save it into the PEM file.
I think that the PHP approach it's better than the STUD one:
It does not require any file or key to generate DH parameters.
This is the way to have supported ciphers such as DHE-RSA-AES256-SHA (
http://www.openssl.org/docs/apps/ciphers.html ) that now cannot be used because the Python SSL binding doesn't initialize the DH parameters.

Well the OpenSSL docs say "DH_generate_parameters() may run for several hours before finding a suitable prime", which sounds like a good reason not to do it every time your program is run.
Anyway, SSL_CTX_set_tmp_dh() should allow us to set DH parameters on a SSL context, PEM_read_DHparams() to read them from a PEM file, and OpenSSL's source tree has a couple of PEM files with "strong" DH parameters for various key sizes.

Wow, i saw your patch for ECC SSL ciphers on http://bugs.python.org/issue13627 .
Do you think we can use the same method/concept as ssl.OP_SINGLE_ECDH_USE but ssl.OP_SINGLE_DH_USE for DH?

> Wow, i saw your patch for ECC SSL ciphers on http://bugs.python.org/issue13627 .
> 
> Do you think we can use the same method/concept as
> ssl.OP_SINGLE_ECDH_USE but ssl.OP_SINGLE_DH_USE for DH?
Of course.

In the meantime i added two other tickets on security and performance improvements of Python SSL support, to make it really complete and comparable to Apache/Dovecot/PHP in terms of configuration and capability: 
Python SSL stack doesn't support ordering of Ciphers
http://bugs.python.org/issue13635
Python SSL stack doesn't support Compression configuration
http://bugs.python.org/issue13634 

Here is a patch adding the load_dh_params method on SSL contexts, and the OP_SINGLE_DH_USE option flag.

Per the Red Hat problems in issue13627 I just tried this patch on Fedora 16. Everything built just fine. However, the patch doesn't apply cleanly to tip an longer:
[meadori@motherbrain cpython]$ patch -p1 < ../patches/dh.patch 
patching file Doc/library/ssl.rst
Hunk #2 succeeded at 715 (offset 27 lines).
patching file Lib/ssl.py
Hunk #1 succeeded at 68 with fuzz 2.
patching file Lib/test/dh512.pem
patching file Lib/test/ssl_servers.py
Hunk #1 succeeded at 180 (offset 1 line).
Hunk #2 succeeded at 194 (offset 1 line).
patching file Lib/test/test_ssl.py
Hunk #2 succeeded at 101 with fuzz 2.
Hunk #3 succeeded at 541 (offset 3 lines).
Hunk #4 FAILED at 1200.
Hunk #5 succeeded at 1858 with fuzz 2 (offset 29 lines).
1 out of 5 hunks FAILED -- saving rejects to file Lib/test/test_ssl.py.rej
patching file Modules/_ssl.c
Hunk #1 succeeded at 1922 (offset 20 lines).
Hunk #2 succeeded at 2082 (offset 22 lines).
Hunk #3 succeeded at 2539 with fuzz 2 (offset 24 lines).
After fixing the unit test hunk everything builds and the SSL unit tests pass.

New changeset 33dea851f918 by Antoine Pitrou in branch 'default':
Issue #13626: Add support for SSL Diffie-Hellman key exchange, through the
http://hg.python.org/cpython/rev/33dea851f918 

Thank you Meador. I've committed an updated patch.