Hacker News Reader

Comment by AnthonyMouse | Hacker News Reader

We started off with SSL/TLS being used for payments systems and logins where its absence is very important to not ignore. Then we said we should use it for everything, which is good, but now it's showing up in different contexts.

If it says the certificate for your bank is expired, you need to stop. If it says the certificate for the 10 year old public blog post that was linked by a 5 year old Reddit post as describing the solution to your problem, that should not matter, and you just want to read the non-secret contents of whatever is on that page regardless of whether the site's maintainer turned on HTTP to HTTPS redirects and then neglected to renew the certificate.

And people understand this, and people are rightfully going to devise workarounds for the second case, and it's ridiculous to expect them to not.

View on HN

> We started off with SSL/TLS being used for payments systems and logins ... If it says the certificate for your bank is expired, you need to stop. If it says the certificate for the 10 year old public blog post that was linked by a 5 year old Reddit post as describing the solution to your problem, that should not matter

Non-HTTPS pages can be tampered with to inject any content into them e.g. into a blog post page, you could inject a login form ("sign in via Google to unlock this post"), a donation payment form ("donate for more content like this!"), or malware installers ("your browser is out of date, click to update" banner).

I think pushing to protect non-tech savvy users makes sense here. I see even a lot of developers not understanding risks like the above, so it's a losing battle thinking non-techy users can be educated about it and be cautious enough.

> Non-HTTPS pages can be tampered with to inject any content into them e.g. into a blog post page, you could inject a login form ("sign in via Google to unlock this post"), a donation payment form ("donate for more content like this!"), or malware installers.

Who cares? The blog author could be malicious. The blog might have been sold 5 years ago and now hosts malicious content.

Stop trying to nanny people, it's unbecoming.

But you can atleast acknowledge that there is a difference between trusting a blog you are actively seeking out and all the entities between you and said blog, right?

Not really. The context is some random reddit post from close to a decade ago linking it. Might as well have typed a random URL in your browser.

The people getting their money and accounts stolen care. The blog owner cares about the reputational hit. People who care about the success of the web care because it makes the web more risky than people using mobile apps.

Yes and if you slit your wrists with a knife you will likely die. Do stupid things, bad things happen. Stop nannying people.

So visiting a website is doing a stupid thing? Your recommendation is essentially telling people to stop visiting websites or accept bad things will happen to them. This is not a healthy viewpoint for growing the web.

The stupid behaviors listed above are putting your google password or payment info into a random blog, or running a program it gives you because it says you need to update.

Doing that is stupid whether it's http or valid https or broken https.

Okay, but if a user went to a http version of YouTube and put in your payment info to buy a movie, as opposed to remembering it should take him to an https Google page, I would find that a plausible situation that is hard to blame the user for. Attackers being able to hijack the reputation of sites is problematic.

> Attackers being able to hijack the reputation of sites is problematic.

And the whole point of this thread is that some sites have 0 reputation to hijack.

Youtube does not have 0 reputation. The point of this thread is arguing the merits of requiring sites that don't handle sensitive informationfto use https.

Nobody mentioned YouTube until your previous post. The example of 0 reputation was a ten year old blog post found via a comment.

And "logins" were explicitly listed as important to secure, in that same comment. So that double covers YouTube for just about everybody.

> The people getting their money and accounts stolen care.

> People who care about the success of the web care because it makes the web more risky than people using mobile apps.

The main comparison here is whether a middleman injected it or the blog inserted it server-side. The level of risk is similar either way.

> The blog owner cares about the reputational hit.

If the blog hasn't been updated in ages, they probably don't.

>The level of risk is similar either way.

There is still risk, but this is a form of risk which is not neccessary and can be reduced.

>If the blog hasn't been updated in ages, they probably don't.

We are talking about blogs that don't use https because they don't sell things. Expired certificates are out of scope of this comment thread.

> There is still risk, but this is a form of risk which is not neccessary and can be reduced.

It reduces it a little bit. But if you drop the risk of a random site being malicious by 25% that's not a very important change. The user still has to be wary. That reduction is not worth anything as drastic as blocking the site.

> We are talking about blogs that don't use https because they don't sell things. Expired certificates are out of scope of this comment thread.

I got the impression we were primarily talking about broken https. It's definitely not out of scope entirely:

"If it says the certificate for your bank is expired, you need to stop. If it says the certificate for the 10 year old public blog post that was linked by a 5 year old Reddit post as describing the solution to your problem, that should not matter, and you just want to read the non-secret contents of whatever is on that page regardless of whether the site's maintainer turned on HTTP to HTTPS redirects and then neglected to renew the certificate."

> If it says the certificate for your bank is expired, you need to stop.

No I don't. At least not if it's recent. A certificate that expired in the last month is roughly equal in safety to a certificate that's valid for another month or two.

Expiration is a backup safety measure and the risk is mostly based on how long it's been since the certificate was issued.

Unless any banks are going around leaking keys right after they expire for some weird reason?

Err what? That certificate may well have been leaked, but because it expired the bank doesn’t not consider it an issue, no need to revoke it.

Certificate validity is binary. either it all is, or it isn’t. this included “not before”

We scream at the expired certificate, yet happily let CloudFlare be an official MitM. How ironic is that? :)

Not only that, banks are generally pretty diligent about that sort of thing and have enough customers and resources that if their website is misconfigured someone is going to report it immediately and they're going to fix it immediately. Which means that a certificate error on a bank site is suspicious.

Whereas a certificate error on a disused blog is pretty much what you'd expect from a disused blog.

The chance that happened is pretty low. What kind of breach gets old keys but nothing else of note?

The browser does not know what the web page is, and attacks on genuinely sensitive websites like banks wildly outnumber attacks on 10-year-old public blog posts. Browsers are doing exactly the right thing here.

The certificate error on the 10 year old blog post isn't an attack, it's an old site that still exists but isn't actively maintained, which happens all the time.

The browser cannot know that.

That's why you need the human who can know that to make the decision.

Isn’t that what’s happening? The browser shows a big warning with context and then the human makes a decision to click the “do this it’s fine” button.

The exception is cases where the site operator has opted in to HSTS or similar to indicate their decision to rather disable access than allow insecure access.

The issue is that there shouldn't be a case where it's impossible to override it, because then the problem is unfixable when the misconfiguration occurs in that case.

It's also overloading what HSTS is supposed to be for. Normally if you go to a site via HTTP, it loads without comment, which is less secure than loading a site via HTTPS that has an expired or self-signed certificate. HSTS it to say, don't do that.

Allowing the site to tell the user agent to do something according to the wishes of the site owner in a way that can't be overridden by the user is not something that user agents should ever do.

Your disagreement seems to be with the spec for HSTS, which does say it should prevent showing a site with HTTP or with an invalid cert.

https://datatracker.ietf.org/doc/html/rfc6797#section-12.1

Literally the entire point of HSTS is to allow websites to communicate to browsers that they can only be loaded in a valid TLS session. It's a defense against TLS-stripping. If you turn that on for your blog, that's on you. Serious question: what could they be "overloading" about HSTS here? What other purpose do you think it had?

Reply on news.ycombinator.com