3 ms·
I don't understand this comment. HTTPS Everywhere forces large sites to use MITMs? No, it forces them to implement good security, which MITMs are incompatible w
by StavrosK 12y ago
I don't understand this comment. HTTPS Everywhere forces large sites to use MITMs? No, it forces them to implement good security, which MITMs are incompatible with. Companies just go "yeah we know that we need to be secure, but we can't be bothered doing that because we like our CDN too much, so we'll just put everything at risk with horrible hacks".
- Animats 12y ago"I don't understand this comment." A content-delivery-network using HTTPS is a MITM. Decryption occurs at the CDN, and everything is in the clear inside the CDN's operation. That's a good place for interception, lawful and otherwise, because so much can be captured there. It's a scaling problem. If you only use HTTPS for crucial items such as logins and credit cards, the load is small enough that you don't need a content delivery network for those pages. A small number of high-security machines can handle the important stuff. The non-secure pages can go through a CDN, which can cache them. With HTTPS Everywhere, if you're big enough to need a CDN, and use the same CDN for both secure and non-secure pages, you've exposed everyone's credentials inside the CDN. How much do you trust your CDN? Are you feeling lucky? What we really need is less SSL and more page signing. There's a W3C proposal for attaching the secure hash of a page to its URL, and validating that in the browser. That guarantees the page you asked for is the one you get, while allowing caching. The caching operation can't change anything without the browser rejecting it. This catches tampering at the router level, the cache level, and the ISP (we're looking at you, Comcast) level. So you can have the performance gains of distributed caching without the risk of tampering.
- toyg 12y ago> There's a W3C proposal for attaching the secure hash of a page to its URL My google-fu seems weak, any chance you could link it? It sounds like a great idea in principle (although you'll need buy-in from browser AND server vendors).
- Animats 12y agohttp://www.w3.org/TR/SRI/ http://www.w3.org/TR/SRI/ It's called "subresource integrity". Mozilla and Google have reps working on the spec. Example: <script src="http://code.jquery.com/jquery-1.10.2.min.js" integrity="ni:///sha-256;C6CB9UYIS9UJeqinPHWTHVqh_E1uhG5Twh-Y5qFQmYg?ct=application/javascript"> Anybody in the transmission path can cache that, but they can't change it. This improves caching performance in general, because you don't need timed expiration. Load JQuery-1.10.2 once, and never load it again unless it changes.
- richbradshaw 12y agoWell, they can change it if it and the resource are not served over HTTPS by changing both the hash and the response for the resource. This doesn't add much unless at least the parent page is using HTTPS, else the hash can simply be stripped anyway. If browsers had an HSTS type precache list for sites that must use hashes, if they aren't using HTTPS then the resource and the hash can be changed.
- otoburb 12y ago>What we really need is less SSL and more page signing. This may more realistically address MITM tampering attacks, but doesn't address the second concern the EFF and others are trying to mitigate, namely passive MITM eavesdropping[1]. CDNs are definitely a prime interception point if HTTPS can't be scaled more securely. To address the (hopefully short-term) scaling issues, I wonder if users will be asked to pick between a fully encrypted but slightly slower experience (anti-tamper + anti-eavesdrop), or a 'mostly secure' (anti-tamper only) and faster experience. [1] https://www.eff.org/https-everywhere/faq https://www.eff.org/https-everywhere/faq