Update with new default boringssl no-aes cipher suites. Re-enable tests.

webrtc/base/opensslstreamadapter.cc

Index: webrtc/base/opensslstreamadapter.cc
diff --git a/webrtc/base/opensslstreamadapter.cc b/webrtc/base/opensslstreamadapter.cc
index 270faa0809f42013c01aeb35854d7c98b3b9cc1f..f445dd4140bb95e43575e7df3a649c467035bea8 100644
--- a/webrtc/base/opensslstreamadapter.cc
+++ b/webrtc/base/opensslstreamadapter.cc
@@ -160,12 +160,27 @@ static int kDefaultSslCipher12 =
 static int kDefaultSslEcCipher12 =
     static_cast<uint16_t>(TLS1_CK_ECDHE_ECDSA_WITH_AES_128_GCM_SHA256);
 // Fallback cipher for DTLS 1.2 if hardware-accelerated AES-GCM is unavailable.
-// TODO(davidben): Switch to the standardized CHACHA20_POLY1305 variant when
-// available.
+
+#ifdef TLS1_CK_ECDHE_RSA_WITH_CHACHA20_POLY1305_SHA256

[davidben_webrtc, 2016/01/06 04:04:44]

Do you actually need this? Surely by now DEPS rolls should be such that you can
assume the new one?

[torbjorng (webrtc), 2016/01/13 13:16:47]

OK. I added this to for the benefit of third parties I thought may pick up an
installed version of boringssl.  Reverted back now.

[davidben_webrtc, 2016/01/13 20:42:53]

Pre-installed versions of BoringSSL are explicitly not supported. The README
should be pretty clear on this. It should only be used pinned at particular
versions because we don't have any hard promises of ABI/API stability. Just a
soft "if every consumer has to change everything all the time, that's kind of
annoying" API stability.

(Although that soft "stability" is, of course, contingent on consumers not going
out of their way to make changing BoringSSL versions hard, as your test seems to
be doing. ;-) )

[torbjorng (webrtc), 2016/01/14 14:27:09]

:-)

I cannot take credit for those tests, unfortunately/fortunately.

+    // This was added in boringssl 13414b3a048f17074253d23d03c0ab97c49b54e2,
+    // changing the fallback.  For now, support old boringssl versions.
+static int kDefaultSslCipher12NoAesGcm =
+    static_cast<uint16_t>(TLS1_CK_ECDHE_RSA_WITH_CHACHA20_POLY1305_SHA256);

[davidben_webrtc, 2016/01/06 04:04:44]

So, this is fine, but it seems poor that you'd need to be so over-fit to
BoringSSL's defaults. Why even have the test at all? (There's some chance we'll
have a last-minute change to this too since the IETF's been mumbling about
switching the PRF from SHA256 to SHA384 or something.)

[torbjorng (webrtc), 2016/01/13 13:16:47]

We might want to clean this up at some point, as well as other parts of the
tests.

[davidben_webrtc, 2016/01/13 20:42:53]

Yeah, I would definitely ask that you change this, though I'm fine with this CL
as-is and then you unravel that check as a follow-up. Otherwise we'll probably
play this game again and again in the future.

[torbjorng (webrtc), 2016/01/14 14:27:09]

I suppose somebody added these tests as a test of sanity, and while it has the
now observed disadvantage, sanity checks have their obvious advantages. A better
test would be "did we get a reasonably strong cipher suite?" as default.

+#else
 static int kDefaultSslCipher12NoAesGcm =
     static_cast<uint16_t>(TLS1_CK_ECDHE_RSA_CHACHA20_POLY1305_OLD);
+#endif
+
+#ifdef TLS1_CK_ECDHE_ECDSA_WITH_CHACHA20_POLY1305_SHA256
+    // This was added in boringssl 13414b3a048f17074253d23d03c0ab97c49b54e2,
+    // changing the fallback.  For now, support old boringssl versions.
+static int kDefaultSslEcCipher12NoAesGcm =
+    static_cast<uint16_t>(TLS1_CK_ECDHE_ECDSA_WITH_CHACHA20_POLY1305_SHA256);
+#else
 static int kDefaultSslEcCipher12NoAesGcm =
     static_cast<uint16_t>(TLS1_CK_ECDHE_ECDSA_CHACHA20_POLY1305_OLD);
+#endif
+
 #else  // !OPENSSL_IS_BORINGSSL
 // OpenSSL sorts differently than BoringSSL, so the default cipher doesn't
 // change between TLS 1.0 and TLS 1.2 with the current setup.
@@ -173,7 +188,7 @@ static int kDefaultSslCipher12 =
     static_cast<uint16_t>(TLS1_CK_ECDHE_RSA_WITH_AES_256_CBC_SHA);
 static int kDefaultSslEcCipher12 =
     static_cast<uint16_t>(TLS1_CK_ECDHE_ECDSA_WITH_AES_256_CBC_SHA);
-#endif
+#endif  // OPENSSL_IS_BORINGSSL
 
 #if defined(_MSC_VER)
 #pragma warning(pop)