CF is amassing a lot of power. With 30% of all Internet Traffic going through CF and them decrypting all HTTPS traffic at the Edge...and able to change any or all of it transparently. This extract from CF's blog reads like a Government/Thread Actor's dream come true....
@thc Is there an HTML obfuscating solution against this?

@mafe @thc Yeah it's called #encryption.

It also completely defeats the point of a CDN/caching-proxy like Cloudflare.

There are a few issues at play here, but mostly the redundancy limiting ability for DDoS could be handled by #p2p content-addressed mirrors, the problem is that on the #clearnet that has major #privacy issues.

So instead of doing the smart thing and deprecating the clearnet for application-level traffic, people just went "fuck it" and took the shit (but easy/simple) option.

@lispi314 @mafe @thc

IMHO: Using a CDN just means that someone is bad at building small websites.

@wakame @mafe @thc Counterpoint: Audiovisual content gets heavy.

Even if you selfhost your webcomic on the most barebones of plain HTML sites, if you get the hug of death your site is going down/unreachable.

If it were on content-addressed #P2P network, that needn't be the case.

@thc I remember ISPs doing this kind of MITM attack. I did not realize that was possible given https.

Mostly I’m puzzled that it hasn’t gotten more attention.

@jgordon

Cloudfare doesn't have to break SSL; they're in the unique position of having direct assess to the plaintext HTML passing through them.

Some of what they're proud here is blatantly evil.

@thc

@jgordon @thc #Privacy conscious circles have been ringing the alarm for a while, but everyone keeps ignoring them or deriding them as paranoid for just seeing the writing on the walls and what is /right now/ being done with some modicum of subtlety (which somehow enables it to just fly under the radar? why?).

@thc

It surprises people that CF can see (and modify) the plain HTML even under https, but if you think about it, any CDN has to do that to be able to cache content.

Same as with a VPN, using a CDN requires some amount of trust. Unfortunately, it only requires trust from one side (the backend service) and not the other (the end user), which doesn't have any say in the outcome.

@javierg That's not entirely true. CDN can provide an extremely valuable service by only serving static things.

@thc

@riley @thc

Do you mean to serve the HTML and any response from the front-end directly to the user and only let the CDN cache images, CSS and scripts? Yes, you can do that, even in the current offering of CDNs.

It does reduce the ability of the CDN to spy you, at the cost of a more complex setup.

Still the choice of the backend service provider, not the end user.

@thc @tychotithonus But I thought they said they didn’t want power.
@PeoriaBummer @thc @tychotithonus They lied, as corporations usually do.
@lispi314 @thc @tychotithonus I don’t think they’d do something like that. That doesn’t sound very much like them at all.

@PeoriaBummer @lispi314 @thc @tychotithonus they still gotta answer subpoenas and their employees aren't immune to the three letter agencies.

Same goes for every us based company, but most don't serve that much of web traffic.

@thc if they modify the continent and re-sign the HTTPS, will the certificate signature be for them rather than the source site?

@guigsy Your question is misguided.

Despite the original architecture, most present-day SSL certificates used for HTTPS don't certify an entity but an endpoint.

@thc

@guigsy @thc they hold a certificate which says that they *are* the source site.
@womble you need to use one of their edge certs for them to intercept though. Unless I’m missing something? If you don’t use one you give up proxying and direct server requests but I think that’s pretty much it.
@guigsy Yes, and a huge number of sites use CF for SSL termination like that

@guigsy When people use CDNs, they ususally don't encrypt their shit and trust a third party, the CDN itself, to do TLS encryption. Even if servers handle TLS, it's a separate connection anyway

Users never directly sent requests to the actual servers they want content from. They send requests to CDN servers, which request the content from, which then requests it from the "end" servers. CDNs are MITM-by-design.

Client <=Encrypted or plaintext=> CDN <=Separate connection=> Service servers

@thc

@thc I'm gonna say it again: #CloudFlare / @cloudflare is a #RogueISP and it's whole business model should not only not exist but be as illegal as #Racketeering is in the #MeatSpace!

@thc The image in the parent post is a screenshot of the first paragraph of the “Parsing and modifying HTML on the fly” section of this Cloudflare blog post.

I don’t like seeing people take my side for reasons I disagree with, so:

On-the-fly HTML rewrites are standard features for any website hosting provider, esp. classic PHP-enabled web hosts. The “HTTPS-compromising intermediary” argument” doesn’t hold water if you treat a CDN as a hosting provider.

There are much better reasons to oppose CloudFlare: their “hate credits”, scope creep, and undermining of browser diversity (by sending uncommon TLS fingerprints through CAPTCHA hell) are better reasons, especially given their market share.

Incident report on memory leak caused by Cloudflare parser bug

Last Friday, Tavis Ormandy from Google’s Project Zero contacted Cloudflare to report a security problem with our edge servers. He was seeing corrupted web pages being returned by some HTTP requests run through Cloudflare.

The Cloudflare Blog
Agreed, @Seirdy, although as @thc mentioned, 30% of the interwebs traffic (BTW , commonly cited figure last year was 20%) being visible gives clownflare a lot of power to abuse. The line between a backup key for a friend and a master key that can unlock every door in the neighborhood is not so fine IMHO.
@cnx @thc Agreed, their market share makes their anti-browser-diversity TLS fingerprinting-based CAPTCHA hell extremely problematic.
@thc they're also blocking VPNs in many cases
😐
@thc source on "30% of all Internet Traffic going through CF"? I'm always curious how they stack up against others like Akamai.
@thc using cloudflare means that your data will be scraped by the NSA since its an american company. so while it has a lot of nice things to offer it is a bit of a deal with the devil..but then again thats using any american internet infrastructure

@thc

> safely rewrite http:// to https://

What for? Properly manitained websites should use https:// by default, all the time one handle it on the webs servers instead of trusting some third party… It's not Jed to do HTTPS

Al the rest is bad and instrusive (injecting google spyware, amp bullshit¹ which another google's spayware, modifying HTML…). Only idiots with 0 respect for their users (e.g marketing people) would want such intrusive crap

1. Useless extra JS won't make pages "load faster"

@thc not true from a technical pov and very misleading. CF isn’t decrypting https in order to do this. They’re handling the initial http to https encryption so of course they have access to the http.
Excessive power, maybe, but not implemented the manner implied here
@BenAveling CF decrypts all HTTPS at the edge and then forwards it via http to the origin server. That’s the most common setup. This is still true if only HTTPS is used (even with HSTS is policy enforcement).
@thc well yes. But that’s not a bug, it’s a feature. So to speak