3 ms·

How does Google create one of the most carefully guarded TLS connections? Should other sites model their implementation?

tptacek 13y ago
By controlling both sides of the connection, and by investing in people like Adam Langley. And yes, other sites should. Not "control both sides of the connection", which is unrealistic, but in modeling their server configurations on Google's so they can take maximal advantage of Chrome's TLS features.
mortehu 13y ago
When Google.com's certificate was faked, it was discovered because Chrome restricts what CAs are allowed to sign Google's certificates, if I recall correctly.
- tptacek 13y ago
  Google does that for a number of other non-Google sites, too.
  - aray 13y ago
    Pretty sure its just google sites, because otherwise you might get false positives as other sites change servers/ips/certificates/etc
    - tptacek 13y ago
      Be less sure. Google provides certificate pinning as a service to other sites, who request it specifically.
    - dfc 13y ago
      https://src.chromium.org/viewvc/chrome/trunk/src/net/http/transport_security_state_static.json?view=markup https://src.chromium.org/viewvc/chrome/trunk/src/net/http/tr...
      - _delirium 13y ago
        Interesting. Kind of an eclectic small list: Google, Twitter, Tor, CryptoCat.
methehack 13y ago
I think it uses perfect forward secrecy, unlike almost any other SSL connection: http://blogs.computerworld.com/encryption/22366/can-nsa-see-through-encrypted-web-pages-maybe-so http://blogs.computerworld.com/encryption/22366/can-nsa-see-... Prefect forward secrecy is a lot more secure since if google's private key were compromised any traffic -- including traffic captured in the past -- would still be secure (baring some further compromise). Each connection has 2 possibly transient negotiated public/private key pairs made just for that connection. In theory, google could store all these pairs as well and they could be compromised, but that adds up to a lot more ifs. As near as I can tell, the extra computation required to do perfect forward secrecy is a large part of why its not more frequently implemented.

tptacek 13y ago

By controlling both sides of the connection, and by investing in people like Adam Langley. And yes, other sites should. Not "control both sides of the connection", which is unrealistic, but in modeling their server configurations on Google's so they can take maximal advantage of Chrome's TLS features.

mortehu 13y ago

When Google.com's certificate was faked, it was discovered because Chrome restricts what CAs are allowed to sign Google's certificates, if I recall correctly.

tptacek 13y ago
Google does that for a number of other non-Google sites, too.
- aray 13y ago
  Pretty sure its just google sites, because otherwise you might get false positives as other sites change servers/ips/certificates/etc
  - tptacek 13y ago
    Be less sure. Google provides certificate pinning as a service to other sites, who request it specifically.
  - dfc 13y ago
    https://src.chromium.org/viewvc/chrome/trunk/src/net/http/transport_security_state_static.json?view=markup https://src.chromium.org/viewvc/chrome/trunk/src/net/http/tr...
    - _delirium 13y ago
      Interesting. Kind of an eclectic small list: Google, Twitter, Tor, CryptoCat.

methehack 13y ago

I think it uses perfect forward secrecy, unlike almost any other SSL connection: http://blogs.computerworld.com/encryption/22366/can-nsa-see-through-encrypted-web-pages-maybe-so http://blogs.computerworld.com/encryption/22366/can-nsa-see-... Prefect forward secrecy is a lot more secure since if google's private key were compromised any traffic -- including traffic captured in the past -- would still be secure (baring some further compromise). Each connection has 2 possibly transient negotiated public/private key pairs made just for that connection. In theory, google could store all these pairs as well and they could be compromised, but that adds up to a lot more ifs. As near as I can tell, the extra computation required to do perfect forward secrecy is a large part of why its not more frequently implemented.