Contradiction between specs and code to detect a V2 handshake
This is a contradiction on how to differentiate V1 and V2 handshake. It's confusing if someone wants to implement his own version of Tor (JTor, silvertunnel,...).
tor-spec (and 130-v2-conn-protocol):
In "certificates up-front" (a.k.a "the v1 handshake"),
[...] The initiator's ClientHello MUST NOT include any
ciphersuites other than:
TLS_DHE_RSA_WITH_AES_256_CBC_SHA
TLS_DHE_RSA_WITH_AES_128_CBC_SHA
SSL_DHE_RSA_WITH_3DES_EDE_CBC_SHA
SSL_DHE_DSS_WITH_3DES_EDE_CBC_SHA
The actual implementation:
/* Now we need to see if there are any ciphers whose presence means we're
* dealing with an updated Tor. */
for (i = 0; i < sk_SSL_CIPHER_num(session->ciphers); ++i) {
SSL_CIPHER *cipher = sk_SSL_CIPHER_value(session->ciphers, i);
const char *ciphername = SSL_CIPHER_get_name(cipher);
if (strcmp(ciphername, TLS1_TXT_DHE_RSA_WITH_AES_128_SHA) &&
strcmp(ciphername, TLS1_TXT_DHE_RSA_WITH_AES_256_SHA) &&
strcmp(ciphername, SSL3_TXT_EDH_RSA_DES_192_CBC3_SHA) &&
strcmp(ciphername, "(NONE)")) {
log_debug(LD_NET, "Got a non-version-1 cipher called '%s'", ciphername);
// return 1;
goto dump_list;
}
}
So, in practice, the use of the SSL_DHE_DSS_WITH_3DES_EDE_CBC_SHA (SSL3_TXT_EDH_DSS_DES_192_CBC3_SHA) cipher suite is considered as a sign of the use of the V2 handshake. This contradicts all specifications.
(According to 124-tls-certificates and tor.git history, two of these cipher suites {AES_256 and DSS} have never been used in Tor v0/1, they seem to be include for better censorship resistance).
Trac:
Username: cced