Hacker News Reader

I disagree. It may seem good on paper, but it gives you too much of a false sense of security. Security measures like this often seem to work, but they are papering over a deeper problem. Usually this is being done because user input is not being handled carefully, and if so, the assumption that blocking some keywords "defangs" potential exploits is usually easy to prove false. Consider the case of eBay and JSFuck[1].

I dislike the mentality that leads to this; WAFs, lazy pentesting and compliance checkboxes have created a substantial body of absolute bullshit security theater and I have absolutely zero doubt that this has convinced companies it's "safe" to put insanely poorly written software partially out on the open internet and that they've "done their due diligence" so to speak. And then I get a letter in the mail apologizing that my most confidential information has yet again been leaked by some company I barely or didn't have any real choice to give my data to. I'm sure they all care deeply about my data and that's why it was stolen using decades-old Java serialization library vulnerabilities.

[1]: https://blog.checkpoint.com/research/ebay-platform-exposed-t...

A lot of critical responses here are saying "this distracts from getting people to do the right thing". I'm open to that, but what is the plan for forcing organizations that are doing the wrong thing to do the right thing? We're talking about businesses that mismanage the sensitive data of millions of people. "They should be doing things right" doesn't seem like an adequate response to this situation.

Correct, we should be legally enforcing that they do the right thing, with legislation that has actual fangs, for companies like Equifax. It should be a potential bankruptcy event when you leak most of America's social security numbers through sheer incompetence.

The problem is that WAF-style security-theater is being enshrined as the industry standard instead, which means that we're just going to get more of these problems instead of less. In other words, a half-measure like this doesn't just distract from doing the right thing, it's actually literally useless for any real security, and instead it's more likely to allow serious security issues to go unnoticed for a much longer period of time.

I agree with you about security theater. The only reason I'm favorable to this approach is because it seems to provide accountable harm reduction in this case. It would stop many typical attacks while being very easy for any random government bureaucrat to test compliance with.

> legislation that has actual fangs, for companies like Equifax. It should be a potential bankruptcy event when you leak most of America's social security numbers

Immutable identifiers should never have this much weight. When your encryption key is compromised, you roll it. When your SSN is leaked, you're doing damage control for the rest of your life. This problem did not begin with Equifax.

It's a 7-digit number that encoded most people's birth region in it until two decades ago and gets asked for by all sorts of randos having anything to do with finance. The problem here is that something so readily shared and easily compromised and impossible to change has this much weight in our identification protocols.

You are not wrong, but I'd argue that mutable ID methods would be worse.

Ok, context and perspective will matter a lot here. If you are wanting to hide from authority then mutable id sounds good. If you are wanting to "know the history" of a person then mutable is bad.

In our social contract the "inability" to change id is baked into the way be behave, and the consequences for bad behavior.

Equally there are lots of good reasons to be able to get a quick history of say a prospective employee, loan recipient, tennant, supplier, customer, and so on.

If I can apply for a new Social Security number every month/year then that number does become useless as a form of id.

But it would then just need to be replaced with something else. Too much of what we do is predicated on our historical behavior. Having a 12-month limit on identity would break, well, just about any kind of contract.

And in most contract situations uou most definitely want to know "who" you are contracting with.

I truly feel that is neither here nor there. As long as we don't fix what's wrong with Equifax, we can trust absolutely nothing to stay private with them or similar entities for any substantial amount of time. Therefore, we actually need to fix this problem anyways. Having just one of these fixed will not solve much.

I am receiving approximately one parcel per month informing me of a new data breach.

> I'm open to that, but what is the plan for forcing organizations that are doing the wrong thing to do the right thing?

i'd say figure out how bad what they did is. Not hashing and building sql strings like that: go to jail for a long time.

Gross carelessness with private data: go to jail for a very long time.

About the same time as if you purposely published the data, or purposely dropped a table, because thats about what you did. If you dont have the competence, you have no business writing such a system. If you dont know how to calculate a bridge that wont collapse, or dont know how to follow the plans and build it properly, you have no business being there.

If an organization has such a password policy, that can be interpreted as the person in charge of setting this policy thinks their organization doesn't have enough people with the competence and organizational power to prevent SQL injection vulnerabilities. Which would reflect poorly on any institution, but especially a university (which should be a bastion of people with competence and organizational power).

As for common password advise, my take on your argument would be that we should all be using these keywords in our passwords to quickly surface these bugs, lest they be hidden and only used by attackers.

It can be interpreted as you suggest, but does not necessarily come from the person in charge. It might mean that auditors forced the installation of a WAF because of $REGULATIONS (irrespective of the actual code quality) and refused to allow weakening the rules for password fields.

EDIT: as they don't actually check for all the banned strings, the "auditor + mandatory misconfigured WAF" hypothesis expressed above is not valid in this case. But let it stay as an otherwise-plausible explanation, or maybe something that was valid in the past.

The WAF hypothesis still holds if you assume the password change page is governed by the same WAF. The WAF would reject any new password it finds offensive, and the "rule" about not using certain SQL keywords just matches people's fuzzy understanding of why these passwords are rejected. It was probably even true, a couple WAF rule updates ago.

Counterpoint - however pathetic it may be, it's better that they publicly owned up to their lack of confidence in this way. The question is, what do we do about it once we see something like that?

> The question is, what do we do about it once we see something like that?

Run away and stay as far away from their products and services as possible.

It's a university

Ok don't go there to study CS :)

> especially a university (which should be a bastion of people with competence and organizational power)

Generally the opposite. University staff positions pay pretty poorly compared to what the most competent people can earn elsewhere.

There are some bright staff people at universities who are there for other reasons besides the pay, but the competence of an average university staff software developer is not great.

This is also why over the past two decades universities have trended to SaaS subscriptions rather than building software in-house

Yeah, there's very much a "cobbler's children have the worst shoes" aspect of this that some people don't get.

> especially a university (which should be a bastion of people with competence and organizational power).

Those people are not the ones running the IT systems. Even the software engineering and information systems departments have no say. IME university IT are understaffed, outsource a lot (often without a choice), and often have to follow decrees from higher management that don't seem to have been thought through.

Your password should not go anywhere near a database.

It should be salted and hashed a few hundred thousand times and that compared to the salted, hashed version stored on file.

If you can't even manage that, you have no business writing software that can store credentials. And I mean that. Software security starts with acknowledging that data is toxic and will bankrupt you if you refuse to respect it.

Many people in positions of great responsibility don't do what they should. What do we do about that?

I wouldn't be telling anyone to implement crappy password policies as a workaround.

I'd tell them to do it properly or not at all, and remind them that in many jurisdictions, knowingly implementing poor data controls earns you some actionable liability. PII is no joke.

This isn't controversial when you're telling people you can't do your own gas work without certification, or electrics without experience. It's okay to demand competence.

What you're recommending is the status quo. The status quo has led to me receiving regular letters informing of massive security breaches where all my PII is disclosed and I have no recourse of any kind, other than possibly spending my life fighting a giant company in a class action lawsuit when I have no time or skill set to do so.

What I'm recommending is just part of the bare minimum security requirement for an authentication system. SQL parameterisation, transport encryption, input validation (and more) are all as important.

Stopping SQL keywords is a distraction dressed up like security. It's harmful.

> Stopping SQL keywords is a distraction dressed up like security. It's harmful.

We are talking about how to mitigate harm from people who are already doing the wrong thing. Saying that this distracts from doing the right thing misses the point. They are already doing the wrong thing. You saying they should not do the wrong thing does not stop them from doing the wrong thing. Once again, your proposal is the failing status quo.

I'm really trying to meet you halfway here but I can't imagine a scenario where I had such low confidence in a third party application that I'd wrap it in input-filtering cotton wool and feel like that was safe enough.

A project that both stores plaintext passwords and fails to use parametrisation (something that's been standard practice for over two decades) is untrustable. It's an untenable liability.

Maybe I'm wrong. Could you explain when you think this would be acceptable?

It's not that I think it's acceptable. I'm thinking about how to reduce the harm of people doing unacceptable things that we can't stop them from doing. I would be very transparent about this. I would say this is harm reduction for people who are doing the wrong thing and is a waste of everyone else's time, but we have to do it anyway because software engineering has failed to govern itself as a profession and the government has failed to hold it accountable for its disasters.

The key thing is that it's both easy to test and would stop many attacks. Anyone can check whether the password field will take these forbidden keywords and patterns.

No question it's a pathetic thing to have to resort to, but pathetic is where things are at right now.

Does it really reduce harm? Bypassing something like this should be, like, one checkbox in metasploit. It's like requiring all walls to be wallpapered to make sure no-one spots any cracks in the walls.

If they're storing raw data without parameters, you can't have any confidence they do that with every other SQL query. You'd have to block every SQL token from every input.

Imagining we're happy to sacrifice these words to the gods of security theatre, what layer are you suggesting this goes? Browser level just means hackers can make raw requests. HTTPDS could block it, like an overzealous WAF, but that still needs implementing. Service providers shouldn't be able to see this stuff because of transport layer security.

There's no sensible way to make this make sense.

I do agree that a lot of entities are not adequately prosecuted for their incompetent data handling but fixing that seems far more realistic than banning the word "drop", everywhere.

https://en.wikipedia.org/wiki/Swiss_cheese_model

Most software you use is untrustable. You get a binary and you pray the vendor isn't an idiot. Even if you watch all the SQL commands it does in normal operations, you can't be sure an attacker cannot send some operation you don't know about and do a regular SQL command.

It's okay to demand competence, but I wouldn't expect competence to materialize overnight.

Do what we've successfully done for hundreds of years: Fine and potentially prosecute them. Stop pretending software engineers are special snowflakes, and apply the same standards we use for bridges, roads, etc. None of those are perfect, but they do manage to do things like stop using known-bad materials, which is broadly equivalent to allowing script injection via password fields.

>Your password should not go anywhere near a database.

>

> It should be salted and hashed a few hundred thousand times and that compared to the salted, hashed version stored on file.

>

> If you can't even manage that, you have no business writing software that can store credentials

That's misunderstanding the nature of the vulnerability. It's not about where the password is stored, but where it is entered. Before it can be salted & hashed, there's software that has to decide where the password input starts and ends. If it gets that wrong, that's how you get the vulnerability.

It's not even clear that there is a vulnerability problem here, though there is obviously a usability/UI problem. It could very well be that having those words in your password doesn't compromise security anywhere, but some systems might reject the password before even attempting to authenticate with it.

It might be a method to avoid trigger an overenthusiastic WAF, but these are all SQL tokens and keywords. It seems far more likely that they've blocked them in a desperate effort to catch injection attacks.

There might not be an actual attack vector, they might block these words in all inputs. It just smells like incompetence.

WAFs commonly block SQL keywords to prevent SQL injection, so it still holds that the WAF could do it.

> but where it is entered.

Hmm. This strongly reminds me of gpt promt injection. I guess why they're both called injection attacks

> That's misunderstanding the nature of the vulnerability.

No, they're 100% correct. If the password is properly stored, there's no possibility of injection, because what gets sent to the database is something like a hex string, or just a bytestring, depending on reprsentation.

> It's not about where the password is stored, but where it is entered. Before it can be salted & hashed, there's software that has to decide where the password input starts and ends. If it gets that wrong, that's how you get the vulnerability.

???

Where it is entered looks like this:

  <input type="password" name="password">

… the actual text of the password makes no appearances, ever.

Whatever code you're using to generate HTML/DOM nodes, or SQL queries, should be parameterized and automatically escaping all inputs. If it isn't, that's the security issue, and trying to kludge around it with idiotic restrictions won't work. (Other commenters have already alluded to the problems with the OP's attempt.)

But even then, a password should never hit either of those interfaces: there's no reason to render it into the DOM, or into SQL.

(There are some other comments about this might be to help users work around failures in a WAF. That could be, but it's orthogonal.)

I find this to be a common defense in depth trap. A lot of engineering effort get threw at the wrong layer, when the problem can be much more efficiently solved on another layer.

It renders the whole organization working in fear -- When you have to worry about the system inserting password in plaintext into a database table, there are also a million other terrible things that can go wrong in this system, like what if your DBA copy-paste a SQL from stackoverflow? There's just endless work.

If your org has incompetence engineers, then maybe just don't let them implement their own authentication system. Use popular open source frameworks and/or buy a commercial product.

What is the attack vector this protects against though?

If the authentication flow is doing anything other than salting/hashing the password and then throwing away the original plaintext password, the entire system really shouldn't be used at all.

It (partially) protects against the attack vector in which the system was badly coded and simply does not do what everyone with basic competency insists all systems must do.

I strongly suspect that this is the most common attack vector in cybersecurity.

But why would you ever send a plaintext password into a sql query?

Because you are lazy, irresponsible, and/or incompetent.

Or anywhere else except a hash function for that matter?

Since being clever is often a footgun, I'll admit to one idea I had years ago; I never really gave it the smell test, and it only existed for a personal project. But if one person thought of it, others may have as well.

I was learning to write sqlite C functions for fun, and thought since storing plain passwords has been such an issue, why not offload the responsibility from the programmer and let the db handle it fully -- consume plaintext, salt, hash, etc, and could transparently upgrade algorithms when necessary.

Luckily I've learned enough to recognize that my skillset is not of the level necessary to fully evaluate the risks. Today the idea makes me feel uncomfortable.

one general problem with this (and also audit logging) is making sure that such changes can't be rolled back. I think the consensus the last time I looked at it was to do it in a separate connection or batched at the end of the main transaction. but audit logging is just such a tremendous pain in general and everyone's got their own specific requirements.

I think in general you would not want to do this in the database anyway (definitely not unless it's sharded and not really even then). You are taking that hashing load off your application servers and moving it inside the DB after all. When your DB is out of CPU, it's out, they're tough to scale. And in SQLite you are holding the DB write lock for longer, which serializes every other thread that needs to write. That's also true of holding connections/resources in traditional DBs. It's notionally fine for toy problems and you can probably do "clever" shit like sharding uuids across multiple files by hash to scale a bit further. But increasingly, even as someone who likes the idea of a "richer" interface to RDBMS, and doesn't mind writing SQL functions/triggers/views where it makes sense, I think the database really just not is the place for unnecessary gizmos on performance grounds either. Don't put hashing in your DB.

An unhashed, unsalted password should literally never touch a database.

Which is to say this kind of sanitization on passwords is meaningless if the barest of security standards are in place.

https://xkcd.com/463/