Latest revision as of 17:14, 15 May 2017

HOW TO WRITE INSECURE CODE

Introduction

In the interest of ensuring that there will be a future for hackers, criminals, and others who want to destroy the digital future, this paper captures tips from the masters on how to create insecure code. With a little creative use of these tips, you can also ensure your own financial future. Be careful, you don't want to make your code look hopelessly insecure, or your insecurity may be uncovered and fixed.

The idea for this article comes from Roedy Green's How to write unmaintainable code. You may find the one page version more readable. Actually, making your code unmaintainable is a great first step towards making it insecure and there are some great ideas in this article, particularly the section on camouflage. Also many thanks to Steven Christey from MITRE who contributed a bunch of particularly insecure items.

Special note for the slow to pick up on irony set. This essay is a joke! Developers and architects are often bored with lectures about how to write secure code. Perhaps this is another way to get the point across.

General Principles

Avoid the tools

To ensure an application is forever insecure, you have to think about how security vulnerabilities are identified and remediated. Many software teams believe that automated tools can solve their security problems. So if you want to ensure vulnerabilities, simply make them difficult for automated tools to find. This is a lot easier than it sounds. All you have to do is make sure your vulnerabilities don't match anything in the tool's database of signatures. Your code can be as complex as you want, so it's pretty easy to avoid getting found. In fact, most inadvertent vulnerabilities can't be found by tools anyway.

Always use default deny

Apply the principle of "Default Deny" when building your application. Deny that your code can ever be broken, deny vulnerabilities until there's a proven exploit, deny to your customers that there was ever anything wrong, and above all - deny responsibility for flaws. Blame the dirty cache buffers.

Be a shark

Always be on the move. Leave security problems to operations staff.

Complexity

Distribute security mechanisms

Security checks should be designed so that they are as distributed as possible throughout the codebase. Try not to follow a consistent pattern and don't make it easy to find all the places where the mechanism is used. This will virtually ensure that security is implemented inconsistently.

Spread the wealth

Another great way to avoid being found is to make sure your security holes aren't located in one place in your code. It's very difficult for analysts to keep all of your code in their head, so by spreading out the holes you prevent anyone from finding or understanding them.

Use dynamic code

The single best way to make it difficult for a security analyst (or security tool for that matter) to follow through an application and uncover a flaw is to use dynamic code. It can be almost impossible to trace the flow through code that is loaded at runtime. Features like reflection and classloading are beautiful features for hiding vulnerabilities. Enable as many "plugin" points as possible.

Secure Languages

Type safety

Means anything you enter at the keyboard is secure.

Secure languages

Pick only programming languages that are completely safe and don't require any security knowledge or special programming to secure.

Mix languages

Different languages have different security rules, so the more languages you include the more difficult it will be to learn them all. It's hard enough for development teams to even understand the security ramifications of one language, much less three or four. You can use the transitions between languages to hide vulnerabilities too.

Trust Relationships

Rely on security checks done elsewhere

It's redundant to do security checks twice, so if someone else says that they've done a check, there's no point in doing it again. When possible, it's probably best to just assume that others are doing security right, and not waste time doing it yourself. Web services and other service interfaces are generally pretty secure, so don't bother checking what you send or what they return.

Trust insiders

Malicious input only comes from the Internet, and you can trust that all the data in your databases is perfectly validated, encoded, and sanitized for your purposes.

Code wants to be free!

Drop your source code into repositories that are accessible by all within the company. This also prevents having to email those hard-coded shared secrets around.

Logging

Use your logs for debugging

Nobody will be able to trace attacks on your code if you fill up the logs with debugging nonsense. Extra points for making your log messages undecipherable by anyone except you. Even more points if the log messages look as though they might be security relevant.

Don't use a logging framework

The best approach to logging is just to write to stdout, or maybe to a file. Logging frameworks make the logs too easy to search and manage to be sure that nobody will ever review them.

Encryption

Build your own encryption scheme

The standard encryption mechanisms are way too complicated to use. It's much easier to make up an encryption algorithm that scrambles up secret stuff. You don't need to bother with a key or anything, just mix it all up and nobody will figure it out.

Hard-code your keys

Hard-coding is the best way to retain control. This way those pesky operations folks won't be able to monkey with (they say "rotate") the keys, and they'll be locked into the source code where only smart people can understand them. This will also make it easier to move from environment to environment as the hard-coded key will be the same everywhere. This in turn means your code is secure everywhere.

Smart coders don't read manuals

Just keep adding bytes until the algorithm stops throwing exceptions. Or just save yourself time and cut-and-paste from a popular coding site. This looks like a good start: byte[] keyBytes = new byte[] { 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f, 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17 };

Pack up your sensitive payload

If you need to pass things that are sensitive, like passwords or PII, just pack it up in an encrypted blob. Then just email the key to the developer who is coding the other side of the transaction, and punch out early. This will also get you past Data Loss Prevention (DLP) tools which is a huge bonus.

Encryption is hard

A great way to improve the security posture of your site is to use client-side JavaScript to encrypt passwords and other sensitive data that are submitted. This can be used with or without HTTPS (TLS/SSL). If you use it without HTTPS, you can save some money by skipping the annual purchase of a signed third party certificate.

HTTPS makes your application bullet proof

If you use HTTPS encryption between your web application endpoint and the client, there is absolutely no way for anyone to steal any of the data that is transmitted.

Authentication

Build your own authentication scheme

Authentication is simple, so just use your common sense and implement. Don't worry about esoteric stuff you hear about like session prediction, hashing credentials, brute force attacks, and so on. Only NSA eggheads could possibly ever figure that stuff out. And if users want to choose weak passwords, that's their own fault.

Just use a hash

Salt is for pigs. Its way too much work to salt and hash passwords. Besides, only the application can get to the database and hashes are one way trap doors, so why would you need it?

Sessions are inherently secure

Sessions timeout after 20 minutes, so don't worry about them. You can put the SESSIONID in the URL, log files, or wherever else you feel like. What could an attacker do with a SESSIONID anyway? It's just a bunch of random letters and numbers.

Single Sign-On is easy

If you need to submit a username and password (or other secret knocks) from one site to another, pack it up in an encrypted blob as above. This way if anyone finds the blob, they'll have a heck of a time decrypting it, which is what they'll need to do to be able to encrypt it again and create the blob themselves.

A one-way hash cannot be decrypted. Therefore, this is a way to take the latter approach and make it unbreakable.

Authorization

Forget about it

See Authentication. Only ivory tower knuckleheads think there's a difference between authentication and authorization.

Just let your authorization scheme evolve

It's really hard to figure out all the access control rules for your application, so use a just in time development methodology. This way you can just code up new rules when you figure them out. If you end up with hundreds or thousands of lines of authorization code scattered throughout your application, you're on the right track.

Privilege makes code just work

Using the highest privilege levels makes your product easier to install and run.

Optimize the developer... always

Each developer, independent of one another, must decide where authorization decision and enforcement is made.

Trust the client

RIA clients and fat clients that use remote services can make decisions about what the end-user can or can't see.

Volunteer to authorize access to other systems

When a service you are calling, for example, has inappropriate access controls just be nice and live with it. After all it will be your fault when a direct object reference permits any user of your system to see any data on their service. Should that happen, it will be a chance for you to show your dedication when you're up at 3 AM on a Saturday morning sorting out a breach.

Policy

Forget about it

Policy can't be known until the user requests arrive. Put authorization logic where you see fit. Sometimes you need to roll it into the UI. Sometimes it's in the data layer. Sometimes it's both.

See Authentication. Only ivory tower knuckleheads think that security requirements can be known at design time and expressed in a runtime-readable/declarative format.

The tool knows how to do it

Let your automated tools dictate your security requirements. Tailoring is for suckers.

Input Validation

Validation is for suckers

Your application is already protected by a firewall, so you don't have to worry about attacks in user input. The security experts who keep nagging you are just looking to keep themselves in a job and know nothing about programming.

Let developers validate their way

Don't bother with a standard way of doing validation, you're just cramping developer style. To create truly insecure code, you should try to validate as many different ways as possible, and in as many different places as possible.

Trust the client

If the client doesn't, by design, produce wacky encoding, for example, then wacky encoding isn't a threat.

Documentation

If you can build it and it appears to work then why describe it?

The most successful applications do not waste time with requirements, security or otherwise. Optimize the development by keeping the developers from having to read.

Security is just another option

Assume that your sysadmins will RTFM and change the default settings you specified in a footnote on page 124.

Don't document how security works

There is no point in writing down all the details of a security design. If someone wants to figure out if it works, they should check the code. After all, the code may change and then the documentation would be useless.

Freedom to innovate

Standards are really just guidelines for you to add your own custom extensions.

Print is dead

You already know everything about security, what else is there to learn? Books are for lamers, mailing lists and blogs are for media whores and FUD-tossing blowhards.

Coding

Most APIs are safe

Don't waste time poring through documentation for API functions. It's generally pretty safe to assume that APIs do proper validation, exception handling, logging, and thread safety.

Don't use security patterns

Make sure there's no standard way of implementing validation, logging, error handling, etc... on your project. It's best when developers are left free to express themselves and channel their inner muse in their code. Avoid establishing any security coding guidelines, that'll just inhibit creativity.

Make sure the build process has lots of steps

You want to maximize the number of steps in the build process that have to occur in the right order to make a successful build. It's best if only one person knows how to actually set up all the config files and build the distribution. If you do have steps written down, you should have lots of notes distributed across a bunch of files and howto's in lots of locations.

Testing

Testing is overrated

Testing is expensive and time-consuming. Just code and ship immediately. The market wants your newest release yesterday. If anything is broken, your loyal users will let you know.

Test with a browser

All you need to test a web application is a browser. That way, you ensure that your code will work perfectly for all of your legitimate users. And what do you care if it doesn't work for hackers, attackers, and criminals? Using a fancy security testing tool like WebScarab is a waste of your valuable time.

Test with only one browser

After all, they all work the same way and follow standards and never differ in terms of security.

Build test code into every module

You should build test code throughout the application. That way you'll have it there in production in case anything goes wrong and you need to run it.

Security warnings are all false alarms

Everyone knows that security warnings are all false alarms, so you can safely just ignore them. Why should you waste your valuable time chasing down possible problems when it's not your money on the line. If a tool doesn't produce perfect results, then just forget it.

Libraries

Use lots of precompiled libraries

Libraries are great because you don't have to worry about whether the code has any security issues. You can trust your business to just about any old library that you download off the Internet and include in your application. Don't worry about checking the source code, it's probably not available and it's too much trouble to recompile. And it would take way too long to check all that source code anyway.

Don't worry about using the latest version of components

Just because a library has known vulnerabilities doesn't mean it's not safe to use. If there's no proof that *my* application is vulnerable then you're just wasting my time.

Culture

"I got a security approval"

So what if it was several years ago, it's basically the same code anyway.

"Security has to *prove* it before I do anything"

So what if you exploited my application, you cheated by using hacking

"Hello, this is an internal application, so security doesn't matter"

We have firewalls and stuff

DevOps

Automation is enough security

Forget about integrity, if we have a problem with a node, all we need to do is kill it and stand another up

Automated Scanning

It just creates more garbage nobody is going to read anyways

Access Control on DevOps systems

IAM roles and how keys are managed don't matter much, we can just check them right into GitHub and it'll be lost in a world of anonymity. Hell, we can even put all our code there so we can just pull it from anywhere.

Static Analysis

See automated scanning

Dynamic Analysis

See automated scanning

Dependency checking

No one would ever think to put a payload in a library, and hey, the signature on libraries are probably always there.

Auditing is a waste of money

There is no reason to check actual configurations written when we are using playbooks and configuration automation the tell us a node started out with a good configuration. Why would it ever change? Also see Automation is enough security

We'll never lose control of our nodes

Let's just write that root key down on paper and share it with the team. Nobody would ever leak it and create a botnet on our dollar.

CORS

The cloud is all about sharing! Let's CORS * everything!

Logging Services

We're using other peoples infrastructure anyways, why bother with logs or alerts? It's just a waste of time