6
votes

I have a working application that establishes an SSL connection to a server. The server uses a self-signed certificate and the client loads a certificate authority chain to tell it that the server is OK to trust. I did that with code like this on the client:

SSL_METHOD* method = TLSv1_client_method();
_ctx = SSL_CTX_new(method);
if ( SSL_CTX_load_verify_locations(_ctx, "ca-all.crt", NULL) != 1 )
{
    return false;
}
_ssl = SSL_new(_ctx);
int val = SSL_set_fd(_ssl, _socket->GetFD());
if ( val != SSL_SUCCESS )
{
    int err = SSL_get_error(_ssl, val);
    return false;
}
val = SSL_connect(_ssl);

And on the server:

  if ( SSL_CTX_use_certificate_chain_file( g_ctx, "ca-chain1.crt" ) <= 0 ) {
    return 1;
  }
  ppem_file = getenv( "PEM_FILE" );
  if ( ppem_file == NULL ) {
    ppem_file = pem_file;
  }
  if ( SSL_CTX_use_certificate_file( g_ctx, ppem_file,
                                     SSL_FILETYPE_PEM ) <= 0 ) {
    return 1;
  }
  if ( SSL_CTX_use_PrivateKey_file( g_ctx, ppem_file,
                                    SSL_FILETYPE_PEM ) <= 0 ) {
    return 2;
  }

I'm trying to modify this code so that the server also verifies the client's peer certificate (self-signed, using same issuer as the server) and having a bit of trouble. I haven't found good "conceptual overview" documentation anywhere, and that seems to be a typical hurdle with the OpenSSL libraries.

On the client I added this after the SSL_CTX_load_verify_locations() call:

if ( SSL_CTX_use_certificate_file(_ctx, "generic_client.pem", SSL_FILETYPE_PEM ) != 1 )
{
    return false;
}

On the server I added this after the SSL_CTX_use_PrivateKey_file() call:

  STACK_OF(X509_NAME) *list;
  list = SSL_load_client_CA_file( "ca_chain2.crt" );
  if( list == NULL ) {
    return 4;
  }
  SSL_CTX_set_client_CA_list( g_ctx, list );
  SSL_CTX_set_verify( g_ctx, SSL_VERIFY_PEER, NULL );

The connection fails because the certificate doesn't validate. The client seems to load the certificate fine and if I comment out the SSL_CTX_set_verify line, the client connects without trouble (because its certificate is never verified).

It seems that the server doesn't think that the client's certificate authority chain is good. What am I missing here?

From the commandline I can run: openssl verify -CAfile ca-chain2.crt generic_client.pem And it passes, so I have the right certificate data available, I must just be using it wrong somehow.

2

2 Answers

7
votes

On the server, you must also call SSL_CTX_load_verify_locations(). This function tells the server what certificates to use for certificate verification; the SSL_CTX_set_client_CA_list() function sets the list of allowed CAs that are sent to the client in the handshake. Both are required.

(You also need a SSL_CTX_use_PrivateKey_file() call on the client, after the use_certificate_file call, but I guess you're doing that and just left it out).

0
votes

SSL_CTX_set_client_CA_list sets the CA list. A CA certificate, by definition, is distinct from a user certificate (e.g. it has the CA bit set). So I recommend you create a proper CA (whose CA certificate is self-signed), and use that to sign both the client and the server certificate. I assume that OpenSSL isn't expecting that the client will actually use the CA certificate for communication also.