From: Richard Levitte <levitte@openssl.org>
Date: Thu, 28 Sep 2006 12:23:15 +0000 (+0000)
Subject: Fixes for the following claims:
X-Git-Tag: OpenSSL_0_9_8e~77
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=7e2bf831006fa4e0a2c1f1f929234a75b9410efa;p=openssl

Fixes for the following claims:

  1) Certificate Message with no certs

  OpenSSL implementation sends the Certificate message during SSL
  handshake, however as per the specification, these have been omitted.

  -- RFC 2712 --
     CertificateRequest, and the ServerKeyExchange shown in Figure 1
     will be omitted since authentication and the establishment of a
     master secret will be done using the client's Kerberos credentials
     for the TLS server.  The client's certificate will be omitted for
     the same reason.
  -- RFC 2712 --

  3) Pre-master secret Protocol version

  The pre-master secret generated by OpenSSL does not have the correct
  client version.

  RFC 2712 says, if the Kerberos option is selected, the pre-master
  secret structure is the same as that used in the RSA case.

  TLS specification defines pre-master secret as:
         struct {
             ProtocolVersion client_version;
             opaque random[46];
         } PreMasterSecret;

  where client_version is the latest protocol version supported by the
  client

  The pre-master secret generated by OpenSSL does not have the correct
  client version. The implementation does not update the first 2 bytes
  of random secret for Kerberos Cipher suites. At the server-end, the
  client version from the pre-master secret is not validated.

PR: 1336
---

diff --git a/ssl/s3_clnt.c b/ssl/s3_clnt.c
index 26788858d7..278be82294 100644
--- a/ssl/s3_clnt.c
+++ b/ssl/s3_clnt.c
@@ -1796,8 +1796,10 @@ int ssl3_send_client_key_exchange(SSL *s)
 				n+=2;
 				}
  
-			if (RAND_bytes(tmp_buf,sizeof tmp_buf) <= 0)
-			    goto err;
+			    tmp_buf[0]=s->client_version>>8;
+			    tmp_buf[1]=s->client_version&0xff;
+			    if (RAND_bytes(&(tmp_buf[2]),sizeof tmp_buf-2) <= 0)
+				goto err;
 
 			/*  20010420 VRS.  Tried it this way; failed.
 			**	EVP_EncryptInit_ex(&ciph_ctx,enc, NULL,NULL);
diff --git a/ssl/s3_srvr.c b/ssl/s3_srvr.c
index 098eea13ce..9f84e3e4dc 100644
--- a/ssl/s3_srvr.c
+++ b/ssl/s3_srvr.c
@@ -300,8 +300,9 @@ int ssl3_accept(SSL *s)
 
 		case SSL3_ST_SW_CERT_A:
 		case SSL3_ST_SW_CERT_B:
-			/* Check if it is anon DH or anon ECDH */
-			if (!(s->s3->tmp.new_cipher->algorithms & SSL_aNULL))
+			/* Check if it is anon DH or anon ECDH or KRB5 */
+			if (!(s->s3->tmp.new_cipher->algorithms & SSL_aNULL)
+				&& !(s->s3->tmp.new_cipher->algorithms & SSL_aKRB5))
 				{
 				ret=ssl3_send_server_certificate(s);
 				if (ret <= 0) goto end;
@@ -1995,6 +1996,25 @@ int ssl3_get_client_key_exchange(SSL *s)
 				SSL_R_DATA_LENGTH_TOO_LONG);
 			goto err;
 			}
+		if (!((p[0] == (s->client_version>>8)) && (p[1] == (s->client_version & 0xff))))
+		    {
+		    /* The premaster secret must contain the same version number as the
+		     * ClientHello to detect version rollback attacks (strangely, the
+		     * protocol does not offer such protection for DH ciphersuites).
+		     * However, buggy clients exist that send random bytes instead of
+		     * the protocol version.
+		     * If SSL_OP_TLS_ROLLBACK_BUG is set, tolerate such clients. 
+		     * (Perhaps we should have a separate BUG value for the Kerberos cipher)
+		     */
+		    if (!((s->options & SSL_OP_TLS_ROLLBACK_BUG) &&
+			   (p[0] == (s->version>>8)) && (p[1] == (s->version & 0xff))))
+		        {
+			SSLerr(SSL_F_SSL3_GET_CLIENT_KEY_EXCHANGE,
+			       SSL_AD_DECODE_ERROR);
+			goto err;
+			}
+		    }
+
 		EVP_CIPHER_CTX_cleanup(&ciph_ctx);
 
                 s->session->master_key_length=