Hacker News Reader

Comment by herodotus | Hacker News Reader

When I worked at Apple I filed a Radar (bug-report) asking for the mail client to check that, if the text of an <a> tag was a url, that the text matched the href field. What followed, on the Radar, was a lengthy debate about this. If I recall correctly, the people who opposed basically argued that, if this feature was implemented by the mail client, spammers would simply find another way to inject false links. We (those who wanted the feature) lost. But I still think that, any web app that shows "http://whatever" in a link field should ensure that the href field is "http://whatever".

View on HN

Sounds reasonable only if the text is a URI handler format; http:// file:// ftp:// steam:// etc.

But then what about almost-uri text. www.yourbank.com without the https://. Or lookalikes "https:\\" or... what about proxies? does https://l33th4x.com?proxy=www.bank.com count if the text is www.bank.com?

Filtering crap like this sounds reasonable but very quickly becomes an exercise in what I call "Giving a mouse a cookie." Now you have a huge complex chunk of code to parse and filter URLs/URIs and every look-alike you can think of; Did you remember that automatic deserialization that kicked in when your values were sent to a callback?

2 days of work later, your new build has fancy-pants a tag filtering that contains and unknown number of bugs and phishers just register and use new domains that look kinda legit and follow your new text/link rules. www.security-wellsfargo-audit.com/login looks legit to the mark, your mail client allowed it so it must be OK.

This sounds an awful lot like how software development in general works...

Isn't all software just some version of "Giving a mouse a cookie."?

uh... ok... I'll bite... why do you call it "giving a mouse a cookie"

See https://en.wikipedia.org/wiki/If_You_Give_a_Mouse_a_Cookie

"If you give a mouse a cookie, he'll ask for a glass of milk"

"If you give a mouse a glass of milk, he'll ask for a straw"

And the story keeps going like that, with more and more requests coming in. It's almost like never-ending scope creep, except the book has an infinite loop in it.

There's also an animated series on Amazon now... my youngest loves it.

Let me google that for you: https://en.wikipedia.org/wiki/If_You_Give_a_Mouse_a_Cookie

That filtering code should already be there in web app.

I'm agreeing with the "no" votes here. It's extra work and testing complexity for a one off case that's trivially avoided by spammers.

And where do you draw the line? Should it flag a link with text "htp://ebay.com" that goes somewhere else? "ebay" with a href somewhere else?

There's no technical workaround to educating users.

I'm partial to a simpler and complete solution: just always force display of href text on hyperlinks, ignoring the markup that's between <a> and </a>. Nothing good comes from displaying the text/image instead of the actual URL; at best it's used (usually overused) for extra aesthetic touch that's not otherwise useful, at worst it's used by advertisers and scammers to lie to people about the link's destination (tracking links and phishing).

I guarantee you that I can craft URLs users will click even if the whole URL is exposed. This "solution" makes developers feel better but provides essentially zero additional security.

Can you give an example? And will it never help?

I think it’s silly to argue against this. It’s like saying “computer security is hard so why bother at all.”

It’s a continued arms race, where you continue to make things harder and harder. This strategy is working the rate at which people are hacked on platforms like iOS is a fraction of what it used to be like for general computing. There will always be security holes, but you plug them as you find them just as you create mitigation’s against classes of problems to the best of your ability. Why make it easy for the attacker?

Never is a silly standard to measure against.

I'm clearly not saying don't do it at all, I'm saying that this approach won't succeed at anything other than making developers feel like they're Doing Something™. Actual spam filtering and 2fa are examples of real security, showing users the URL is an example of security theater.

The corollary is the nonsensical "security is hard so let's force non-technical users to do it".

Sandboxing to prevent malware installation and password managers to prevent phishing are excellent technical workarounds for this.

For the latter, as one of my sibling posts points out the user will go "Huh, stupid password manager didn't fill out my password - I'll have to do it manually". Often the password manager even helps them do this in the name of user convenience.

Only Brick wall UX works. That's what WebAuthn does here. Don't offer the user a way to "continue anyway", don't ask them confusing questions, just a brick wall and no way forward.

The user will probably be emotional. Scammers work hard to make your users afraid, or horny, or confused, and so they really, really want to give their bank credentials to https://honest-this-is-your-bank.not-a-scam.example/tmp/back... and nothing you can tell them is going to make them stop wanting to do that.

The brick wall doesn't care about the user's emotional state and will stubbornly resist. Maybe the user will eventually realise it was a scam, maybe they won't, the brick wall doesn't know or care either way.

Brick Wall UX even helps the software engineer. When a manager asks if you can't just add a banner that says "Hi, this is our new web site, please continue to use your old credentials" and thus undo every second of training about phishing your users ever got the answer with a Brick Wall UX is that literally can't work. No matter how much they beg and cajole and swear it's just a temporary workaround, it will not work at all, so they're just going to have to go tell the Big Boss that no matter how much was spent on new-brand-name.example the login system will have to remain forever on login3.long-forgotten-brand.example because that's what some idiot picked five years ago then it was set up and too bad.

Can you elaborate on what "brick wall UX" is like to use as a user, or how one implements it? Are there known examples you could point me at? (It's a term I haven't heard before, and haven't noticed anything when searching.)

I don't think it's a known term; it seems something GP created. As for how it works, the concept is simple: if the user wants to do the wrong or potentially insecure thing, just don't let them, period. That's the brick wall.

I'm of two minds about this personally. On the one hand, I appreciate the argument that the only thing that can prevent businesses from doing something bad, stupid or abusive is if it's legally, physically or by design impossible. On the other hand, as a pro user, I do appreciate the ability to override software when it mistakenly tries to prevent me from doing something.

Unless the developer has enough foresight to account for, and sufficiently handle, every single instance where that brick wall would stop a legitimate false positive, I stand vehemently against them.

There are enough examples in the past that clearly demonstrate that developers are not benevolent or competent enough to have complete and final control over the software their users run. Sometimes this control even results in the exact opposite of what the developers originally intended, as was the case with firefox addons just a couple of days ago.

Ultimate control over software should always reside in the hands of the user.

> as a pro user, I do appreciate the ability to override software when it mistakenly tries to prevent me from doing something.

This is really the core problem in security; the world is designed for people like this by people like this, without any serious thought for the implications for the overwhelming majority of users. Do not include "I know what I'm doing" escape hatches, and security will magically get better for the many at the expense of convenience for the few.

Sure. That's why I'm not too big a fan of security. The flip side of your observation is this: the most secure form of computing is a rock. You can ensure users can't be pwnd and can't selfpwn themselves by making the device as useless as possible.

Let's give every user a tablet that has two buttons. You press one, you get a new cat picture. Press the other to "like it". That's all the user needs. All data exchange is end-to-end encrypted from the cat picture provider to tablet's input&video drivers - can't risk the spooks^Wcompetition knowing what they're looking at. They don't need to do banking - like everywhere else, they just sign a three-party contract with the tablet provider and the bank. This way, the Bad Guys can't steal users' money! Oh, the users also want to watch pictures of squirrels? There's a separate tablet for that pulling from separate provider; it's insecure to let these mix on one device!

Seriously, this is how the world would look like if security got its wish. There is a point past which security is essentially enslavement, and that's true both in physical security and computer security.

I gave the best example around today already: WebAuthn. Its ancestor U2F has the same behaviour.

The credentials in these protocols depend explicitly on the verified FQDN of the server (and thus you can only use this with HTTPS). When scammer.example asks for your credentials there literally isn't a way to give it credentials for realbank.example. No matter how sure you are that you're a very smart person and definitely need to give scammer.example access to empty your bank accounts, no way to do this is available. Maybe next week you'll still be angry you couldn't do this, maybe you'll realise it was a scam, don't care.

This is why Google reports zero successful phishing for their own systems. U2F is mandatory there. Their employees aren't magical, some will fall for scammer.example and they will be really frustrated that they can't use their Google login like it says, and some will scream at their help desk team about how stupid this is, how it's totally broken, and even after they demand that the help desk person be fired and they change their password six times and write a ten page rant on their blog they still can't give their employee credentials to the scammer and Google remains safe.

Except they're not, because a dumb enough user isn't going to think about their password manager, and they'll enter their password anyway.

They won’t even know what it is.

I also think mail clients should do that, or at least offer it as an option. Most mail readers have an option to disable loading remote content, and this would be another little way to make HTML email less dangerous.

I think (or at least hope) that most people are in the habit of hovering over links in email before clicking them. And I really hope that mail readers never start implementing Javascript. As for web apps, that's the Wild West, and a small fix like this isn't going to tame it.

Maybe, quite simple have the option to parse <a> tags down to the format "Text - URL". If I remember correctly, some sites from way back had this sort of format occasionally.

WordPress does it in the admin interface. A very clever hack. Something like:

    a:after {
        content: attr(href);
        display: inline-block;
        padding: 0 1ex;
    }

You give way too much credit to users if you think they hover over links and users can’t hover over links on mobile devices.

At least Thunderbird seems to do that: when an email has an <a> tag with text that looks like a URL but doesn't match the href, it throws the "this email is probably a scam" bar above the message.

...which then marks all these newsletters as scams, since the link usually first points to analytics site?

Sounds to me it’s working as intended.

Sure but regular users will see it and think "this message means nothing, I got an email I know was from my bank the other day and I saw the banner there, too!"

They should just collect stats on their own servers.

This might come as a surprise to you, but every company doesn't have infinite developer time to reinvent things which already exist.

Then they should be held accountable for their choice not to vet ads they send to their users. If this was enforced legally, a bunch of companies would suddenly be able to invest the "infinite time" required to register a damn click.

That's obviously not the hard part.

"The hard part" can still be outsourced to a third party without resorting to redirection chains, either by sending the click information to them on the server-side or by sending it client side using a script.

Either way, "the hard part" is generally undesirable to users because it compromises their privacy in order to manipulate them.

The real answer here is that tech companies are trying to solve this problem with tech - in an effort to cheap out on actually hiring some humans to look at the thing and verify that it's safe.

My eyeballs are not free. I hate advertising and advertisers. I have no pity for the advertising platform that cheaps out on security just because it's expensive.

> the mail client to check that, if the text of an <a> tag was a url, that the text matched the href field.

The use case that this breaks is doing click tracking on links using redirects from a unique url to the actual url (which would be the url displayed in the link text).

To avoid breaking this use case, the best remedy would be to prompt the user with a security warning upon clicking a mismatched link prompting them the verify the url in the url bar. The issue, is that doing this selectively teaches the wrong security practice to users: that they can improve their safety by looking at the link text rather than at the url bar after clicking the link.

> The use case that this breaks is doing click tracking on links

I, personally, would be quite happy for this use case to break.

> I, personally, would be quite happy for this use case to break.

Why? If you don't want to be tracked it is pretty easy to avoid. You should already only be getting/opening emails you care about. Emails you don't care about should be unsubscribed from and reported as spam. Granted that links should only be tracked in email you do care about, why do you not want those people to have the information they need to refine and improve these emails so they can better serve and inform you?

I would assume that happiness depends on how this use case is broken.

Would you be happy if the email just doesn't show up or gets shunted to spam? Even if it is a password reset email or a email verification email?

Would you be happy if the link just failed to open, forcing you to copy and paste the link text manually? Why not just do that on your own anyway? No need to have the email client block this for everyone just to suite your tastes.

This leaves us with just the behavior I mentioned above and my argument against it.

> why do you not want those people to have the information they need to refine and improve these emails

Well, it comes at the expense of 1) making things slower for me and 2) making it more difficult to discern phishing emails from legitimate ones. I also find it difficult to believe that all of this analytics is actually doing much to inform me about things I care about.

> Would you be happy if the email just doesn't show up or gets shunted to spam?

It's unfortuate, but I would understand it if it happend. I certainly hope most phishing emails would end up in my spam.

> 1) making things slower for me and

That is a valid reason, but I suspect the extra delay of 2x your ping when you click on a link and wait for it's target to load is fairly negligible for most people.

> 2) making it more difficult to discern phishing emails from legitimate ones

You shouldn't be relying on link text to discern phishing emails, that is what the client checking SPF records and the user checking the contents of the url bar are for.

> I also find it difficult to believe that all of this analytics is actually doing much to inform me about things I care about.

Why is that? I would think it is pretty obvious how A/B testing click-through rates for emails could easily help make those emails more informative and easier to use.

If the speed cost and privacy loss is not worth it to you, having the actual (non-tracked) URL available in the link text atleast gives users the option to opt out of that tracking.

> Why? If you don't want to be tracked it is pretty easy to avoid.

If you want to track clicks, why not use a subdomain instead of something.weird3rdparty.com? And why would the link text look like a URL? I don't see why

  <a href="https://possiblephishing.com/n1bfsd?j1h7d8Sda">www.ebay.com/viewOrder</a>

makes more sense than

  <a href="https://www.ebay.com/viewOrder">view your order</a>

> Even if it is a password reset email or a email verification email?

Why would you want to share password reset info with a third party? I get the "easy tracking" argument for newsletters (but I'm pretty sure most are not GDPR compliant), but for password reset emails? why?

> Would you be happy if the link just failed to open

Why would it? You can put a working link in the email directly - you've proven that by putting a different link in the email than the one you pretend you're linking to.

> If you want to track clicks, why not use a subdomain instead of something.weird3rdparty.com?

I don't see what relevance a subdomain has here. As the original issue was described, it was blocking any links where the url doesn't match the link text (if the link text is a url). This means that even tracking that is done on the same domain (e.g. example.com/emailTracking/{{unique_string}} redirects to example.com/viewOrder/412) would be blocked.

If you start only blocking those links if the domains don't match, things become more complicated to implement and test, especially once you consider subdomains, etc. The question remains, what is the point of this and is it worth it? Are we really promoting good security practices or just adding security theater cruft?

> Why would you want to share password reset info with a third party?

Who said anything about a third party? As you mention above, this blocks people who are doing their own tracking on the same domain. I don't have any numbers, but I assume that ESPs like MailGun and SendGrid are probably the most common third parties used to track email link clicks (since they will automatically substitute link urls with trackable rediracts if you enable it.) In this case, that third party already has access to the entire content and metadata of the email and giving them link click data is a relatively small addition.

> Why would it? You can put a working link in the email directly - you've proven that by putting a different link in the email than the one you pretend you're linking to.

Email links already work one way, perhaps if Apple implemented this feature eventually every email sender out there would switch to links that would work in Apple mail again. In the mean time, Apple has to deal with confused and dissatisfied users wondering why links in their emails don't work.

How would you implement link url / text mismatch blocking? Have you thought through the consequences for your users and their understanding of security or satisfaction with your product?

> Are we really promoting good security practices or just adding security theater cruft?

We're taking away easy attacks, just like blocking known spammers from delivering emails to your MTA. Yes, it doesn't solve spam completely, but it reduces the amount. Add other things like SPF, DKIM etc and you can identify even more malicious emails and warn the user. And certainly, the OP's description mentioned matching urls (though the text would have to match the href, so "ebay.com" would likely be fine for "https://ebay.com/something"), but again: what's the harm in linking to the actual URL? Why the need to hide it, IF you already make the link text look like a URL? I'm sure there are also reasons why a site operator wants to avoid SSL, but I do like the fact that many browsers do warn users when (potentially) sensitive data gets transmitted via plain text.

> Who said anything about a third party?

The super majority of click-tracking runs via third parties. And yes, every additional detail you share with them (PII has gotten a lot of attention lately, and I'm beginning to come around and love GDPR) will put your users more at risk - now it's not just your database that risks exposing your users when a leak/breach happens, it's also your email provider's.

> How would you implement link url / text mismatch blocking? Have you thought through the consequences for your users and their understanding of security or satisfaction with your product?

I'm sure that it's not trivial, but few things are, so rejecting it for that reason doesn't sound like a good idea to me. "Hey, let's not do SSL, it's not that simple to build and might inconvenience a user that has their clock set 100 years into the future"

> We're taking away easy attacks

What attacks are blocked by this that are not also blocked by using SPF and warning users when SPF is not present or doesn't match?

> what's the harm in linking to the actual URL? Why the need to hide it, IF you already make the link text look like a URL?

This is practically irrelevant. We are discussing the costs/benefits of the implementation of a client side feature, not the ideal form that all emails sent should adhere to (which is a far broader and more complicated topic.)

However, I do have several practical reasons why the link text and url matching can have negative consequences in practice:

Third party tracking issues:

1) Practical difficulty: Most link click tracking is done via third parties (usually the sender's ESP). No click tracking tool I've seen offered by an ESP provides any way to use the link url as the link text (since the link url is usually processed and rewritten by the ESP after the email is composed and sent.)

2) Confused users: The average user has no idea what an ESP is and could be needlessly confused / frightened when shown links so some random domain even when the operator of that domain is trusted by the sender.

First party tracking issues:

3) Removal of user choice: If you only give the track-able url in both link target and text, you FORCE your users to be tracked, rather than giving them the option to select the link text and paste it directly into their browser.

4) User experience: The url that is shown in the link text can be much more informative as to what it does (e.g. shows you your order) rather than an uninformative, generic click tracking link.

> The super majority of click-tracking runs via third parties.

Can you find me some email link tracking services that are not also involved in sending that email in the first place?

You do expose additional information about the user's clicks and IP address. If this is something your company is concerned with, you probably need to implement first-party click tracking both for email AND your website. You should probably also run your own ESP.

If you are trying to promote the use of first-party click tracking (or just discourage third-party click tracking), it would be far better to block all links that go to domains that don't match the sender's domain (or alternately that don't match the sender's domain's SPF record.)

> I'm sure that it's not trivial, but few things are, so rejecting it for that reason doesn't sound like a good idea to me.

I am not rejecting it because it's not trivial. I am rejecting it because thinking through how that implementation would realistically work makes me thing that it would accomplish almost nothing and possibly even negatively impact users' understanding of security.

You seem to think that is not true, so I am asking how this feature can be implemented in way that has a positive impact on security.

My biggest concern is that we shouldn't do anything to teach people that sometimes they CAN trust the link text rather than needing to check the actual URL they end up at. As far as I can think, all the attacks that this stops are better stopped by checking SPF and strongly warning users when it is not present or does not match.

> My biggest concern is that we shouldn't do anything to teach people that sometimes they CAN trust the link text rather than needing to check the actual URL they end up at.

I don't believe that "copy link text to avoid tracking" is a relevant part here, so I still don't understand why you'd want to give the impression of a text-link with a different URL. I see lots of malicious reasons, but I don't see valid ones where there is a strong case that this is a necessity. Why should we teach users that "don't trust your lying eyes, just click on whatever" is ever a good idea? Why shouldn't we teach them to not touch something that is trying to deceive them? If we teach them to ignore these things, we're making it easier for scammers.

Sure, blocking links to domains that don't match the sender's may be something as well, but I do see lots of cases where that's totally normal, i.e. me sending you an email saying "hey, I read your blog entry, and this site here does what you want". Mind you, that's just a link, it's not a link that is trying to confuse you about it's true target.

> You do expose additional information about the user's clicks and IP address. If this is something your company is concerned with

If any company isn't concerned with that, they either don't do business in Europe or they should talk to a lawyer. ;)

> If you are trying to promote the use of first-party click tracking (or just discourage third-party click tracking)

Neither is my intent, I just don't see valid reasons to pretend you're linking to one URL when you're linking to another when there's an easy alternative: put words into the linktext, not URLs.

It's like FB's idea to pressure users into giving them their passwords for their email account. Terrible idea, no valid business case ("it's easy and we can really check that they are the owner of that email account" isn't valid), but lots of reasons for malicious actors, so somebody telling you "give me your email password" is a warning sign for everybody. Trying to confuse users about what URL you're linking to is as well.

> why do you not want those people to have the information they need to refine and improve these emails so they can better serve and inform you?

Email A/B testing is always about marketing and trying to manipulate users, in my experience. If the content is the same, but a different button gets more clicks, it's hard for me to believe users are finding a benefit in that. I think it's far more likely that the content isn't all that compelling, but the UI tweaks managed to tickle some part of the reader's subconscious in the right way for them to click.

In other words, emails with quality content don't have marketing teams running them.

This is pretty much like all targeted advertising. The benefit to the company is very clear, but I don't think I'd be missing much if email analytics went away.

Just as a personal opinion that I'm sure plenty of people disagree with, I dislike it when products are hyper-customized based on a lot of analytics. For instance, Netflix has a ton of data on me, but all of their tests and micro optimizations make me use the app less. I would much prefer their product if they threw all of that info in the trash, rolled the UI back 5 years, and showed me more generalized start ratings again.

Yeah but that's annoying and almost every one who isn't "techy" for lack of a better term won't even bother reading it and just press ok.

I mean can you honestly tell you read the Google privacy notice or do you just click the down arrow till the ok box appears?

> Yeah but that's annoying

Yes it is annoying. My point is if you want to be annoying, you should be annoying the user when they visit ANY external links as trusting the link text isn't a practice that should taught/encouraged.

> almost every one who isn't "techy" for lack of a better term won't even bother reading it and just press ok.

That depends on how short and well worded the alert is. However, there is certainly an attention budget that can be used up with pointless / low value alerts (which is why I would assume that Apple did not go this route).

Email marketers would just stop having text in <a> tags that dont look like URLs to get around it. They already do that anyways for the most part.

Phishing attempts will mimic real emails, so they will do the same.

Reply on news.ycombinator.com