Obviously, XP is a problem. We're planning on prompting Chrome users on affected versions to update, both at startup and on the error page. SP3 does exist and one doesn't need to pass the Genuine Windows check to install it[2] so I think we have a good chance to getting users to update once sites stop working. Even if you're going to serve different chains based on signature_algorithms, that gets you at most a year of extra time. Maybe you should be putting effort into getting people to SP3 instead? :)
[1] https://technet.microsoft.com/en-us/library/security/2880823... [2] http://support.microsoft.com/kb/322389
I do at least hope that you will make efforts to alert more of the general public of what's coming and how they can prepare ahead of time. While at CloudFlare we follow the crypto lists closely and know there have been discussions for some time, dropping this news quietly for general public consumption on the Google blog on a Friday afternoon was... an odd choice.
And, for the record, for the last two years we've made it one-click simple for any of the millions of sites on CloudFlare to alert their users to move off out-of-date browsers:
If Microsoft's 2016/2017 deadline is reckless, what SHA-1 deprecation date would be right by your measure? Some have suggested ~2020 (5 years issuance from 2015). Would you really be happy depending on SHA-1 in 2020?
(Edit: on re-read, the 2020 suggestion sounds like a strawman - but that's actually what we were on the road towards and we may not have even managed that.)
If you are returning two different certificates for two different situations at Google.com, the least you could do is dedicate some of your vast engineering resources to making the plugins to support such a setup production ready. If Google uses it's technical sophistication to avoid the pain while the rest of the web burns then that's, at best, hypocritical.
I'll make you this deal: if you do it for Apache, we'll do it for NGINX. That's the right thing to do. As would be making SSL certs free, but that's a conversation for another day.
Apache/OpenSSL would have to assume that pre-TLSv1.2 clients support only SHA1 because there's no signature_algorithms extension, but for TLSv1.2 clients can Apache/OpenSSL choose between a RSA-SHA1 cert and a RSA-SHA256 cert?
The reason I ask is because ECDH certs seem to be even harder to come by than RSA-SHA256 certs right now.
That's certainly the plan. What did I say that suggests otherwise?
The transition that I'm talking about is the transition to SHA-256 overall. If you're in a position where your CA sold you a certificate that they shouldn't have then I feel sorry, but if you're blaming us for that then I think you're pointing in the wrong direction.
What would be the technical challenges facing offering free wildcard certs? Is it just engineering time, or is there something more fundamental? That seems like it would do much more for web security than pushing for higher minimum encryption standards on certs would...
The situation with Android, vendors, old versions, device abandonment and lack of security patches is terribly disappointing across the ecosystem. It does not have the same excuse of age that Windows XP does. I'm not sure how to remedy that, and I'm not even completely sure why it happened in the first place except for the melting pot of: carriers wanting complete images to test and to remain completely stable (big mistake: most software really needs security patches deployable globally within hours!); SoC vendors with closed-source binary blob drivers, and this being tolerated in the ecosystem (big mistake: this means when it's dead to the vendor it's dead to everyone, because ABI changes); and versions with high minimum memory requirements (which is improving recently, but I feel if you can't go back and run it on the ADP1, and it doesn't run worse than it ever did, you're still not really done with that yet). Projects like CyanogenMod at least help there.
I agree it is shocking how many are still out there, but given the choice between "HTTPS being secure" and "supporting insanely old/insecure software", choosing the latter seems like the kind of choice people will vividly remember when they come to regret it later.
I think there are really three outcomes:
1. HTTPS is widely used, it's secure, insanely old/insecure software is not supported. (Ideal outcome)
2. HTTPS is widely used, but using SHA1 certs for a little longer so that insanely old/insecure software is supported.
3. HTTPS is less widely used, but is secure with SHA2 certs, and insanely old/insecure software is still supported.
My concern (and Matthew's too I think) is that the aggressive deprecation of SHA1 will put us on a trajectory to outcome 3. We're at a unique point in history right now: there is incredible momentum behind converting sites to HTTPS, even sites that traditionally would not have used HTTPS (such as all static sites). The SHA1 deprecation might throw a wrench into this and cause site operators to reconsider switching to HTTPS. If not for this momentum, I'd agree that aggressively deprecating SHA1 would be the clearly correct course of action, but at this moment in history I'm deeply ambivalent. Disrupting the HTTPS momentum would be very sad, especially since switching to HTTPS provides an immediate defense against mass passive eavesdropping.
Nobody should really be issuing certs with SHA-1 anymore - certainly not from next year.
Of course, doing that is a lot of work. But at least your company is in a position to do the work once and automate it afterwards for all your customers.
FYI, we are deploying in the same manner as Cloudflare, with RSA/SHA1 and ECDSA/SHA256 side-by-side. We are committing our changes to the public ATS repository and hopefully those changes are useful to other projects. Unfortunately this is dependent on OpenSSL 1.0.2, so we might have to deploy beta code into production if the OpenSSL project can't beat Chrome's arbitrary deadline.
This enables SHA-2 certificates. Deployment of the patch is another problem, since it's a HotFix (which may have enterprise-QA issues) and not intended for general use, AFAIK. Still, I've been using it since WS2008 originally came out.
What that translates to is that it only gives Server 2003 SHA2 support as a client, not as a server. I.e. You can connect to sites that are using SHA2 certs, but you cannot bind a SHA2 cert to your own website in IIS 6/Server 2003.
So once SHA1 is completely deprecated, those hosting sites or legacy apps on Windows Server 2003 will not be able to upgrade to SHA2 certs.
[1] http://blogs.technet.com/b/pki/archive/2010/09/30/sha2-and-w...
It's a startling amount today. Just by virtue of making the announcement, Google knocks that number down quite a bit.
Honestly, in a world where your choices are "doesn't work" and "secure", and there is no in between "works but isn't secure", security becomes a LOT simpler.
The faustian choice will soon be between having no security on any platform and supporting ancient platforms. This could come at any time. It could already be too late.
Like it or not, you are a security company. You should act like it. This is not acting like it.
At CloudFlare, we have a plan to handle this gracefully for our customers (with both modern Chrome and old OSs) ahead of the change, but it's a non-trivial engineering challenge that many organizations won't make. I worry that, faced with the choice above, many organizations will simply opt not to support HTTPS at all. And, from my vantage point, that's a bigger risk to web security today than certificates signed with a SHA1 hash.