Comment by peterwwillis | Hacker News Reader

Comment by peterwwillis | Hacker News Reader

Parent Full thread

It's been the golden age for script kiddies ever since fuzzing and injection became the most effective blind attacks. Who needs a database password when you have %27 ?

It used to be you had to actually break into a system to exfiltrate all its data. Now you just make an HTTP query. Owning a big system was really important because bandwidth and server space was expensive. Now you rent some VPS space with bitcoin you made from spam or DDoS-for-hire using someone else's botnet that had been sitting around with a default password, and use it to distribute pirated media like it's text files. Mass-scan for SQL injections, inject some malware you found on a forum, and amass a botnet to play with. What a time to be a script kiddie.

I'm not sure about the modern age, but to me the golden age of blackhats was pre-2003, when nobody was really watching their networks or systems and advanced techniques were everywhere with zero defenses. Metasploit and the age of shitty webapps and packaged malware ushered in the dumbing-down of blackhats as a general concept.

(Get off my lawn!)

I'm not sure if security didn't actually improve dramatically. Web frameworks seem to have SQL injection and cross-site scripting more or less under control these days. Cloud setups and containerisation should also help (servers are more easily rebuild and therefore more likely to be fully patched, infrastructure-as-code is easier to review, most servers/containers are only running one service etc.)

Patching is not easier now, and wasn't difficult before. There's more steps now, it's not easier to review.

Look at it from the blackhat perspective. I don't care what the hell loops you're going through on the backend. If there's a 0day, i'm going to use it and it's going to work, because you don't even know about it yet, much less have a patch built into a binary pushed to mirrors that your Docker image building box needs to update to before you can rebuild your image and push it to your servers and do a maintenance window to switch to the new services.

It doesn't matter if you are running in a VPS in a container in a virtual machine in an emulator, because I can still use an SQL call to dump everyone's passwords, or get voting records and account details for one of your political parties. If I execute code you're still going to get owned because you don't have security patches in your kernels and you don't use signed binaries (none of which are new technologies, btw).

I would even go so far as to say you would never in a million years find exploited code in one of your systems because you somehow believe container apps are immune to basic vulnerabilities. Containers are just fancy chroots.

I disagree with your assessment for two reasons:

1. Breaking up my app (eg, across containers) increases exploitation difficulty, because it's harder to chain weaknesses between parts of my stack. A weakness in the API layer doesn't let you run against the DB directly -- you either need to find a weakness in the interface between containers or use a container jailbreak exploit. You also have fewer system libraries to use, since the container only has those needed by either the API or DB, not both.

2. The frequency of rebuilding means that a lot of system components get patched faster. Sure, you can burn a 0-day on my app, and there's relatively little I can do about it, but all of my builds are <1week old, so you're using new exploits, not easy-to-get legacy malware.

The point of security is always economics. So containers (and more generally, frequent rebuilds and virtualization) provide security by raising the complexity and freshness of exploits required.

That's a non-trivial gain.

Abstraction has never been a big barrier to chained vulnerabilities. And hackers do not rely on a dynamic linker or shared libraries to deliver their payloads. (Script kiddies do, because they're using some paint-by-numbers malware package) Any actual hardening of a system will more than secure it from "the excess of unnecessary binaries" that some people fear.

So you're running brand new code on your production systems on a regular basis? Fantastic! I'll find a new 0day in the latest releases since those are the ones that have had the least review. If you're using an old stable build, the 0days keep working until one gets leaked, and then you just find another one.

How do you figure containers raise complexity of an exploit? The exploit is of the app or the stack, but the container isn't part of the app stack. A container is literally just a way in which you run your app. If I can execute code, the game is over, in all container systems that don't prevent traditional exploits. The only thing containers make easier is namespaces, which is a white hat mitigation. Like locks on doors, they only keep honest people out.

By the way, exploits don't really need to be fresh. Many exploits getting patched these days are years old, and some were patched years ago and a recent update re-instituted the vuln. You can resist 99% of attacks using 10 year old technology, but nobody implements it. They just say stuff like "I put it in a container!" and your boss is happy so everyone goes on with life.

Containers make things a bit harder from a persistence view. There was a great CCC talk on pwning architectures which are just based on aws lambda this year [1], worth a look.

1: https://app.media.ccc.de/v/33c3-7865-gone_in_60_milliseconds

I completely agree with you and I find the ignorance in our industry mind blowing. The industry has pushed for sloppy dev practices (Lean/Agile) and has time after time claimed that all of the "extra" stuff like strict access controls and discipline "aren't needed features."

Mongo DB security didn't come until version 3. VERSION 3. It was assumed you could just firewall it off and life is good.

Everyone uses Lean/Agile now so calling effectively the entire industry sloppy is a bit stupid on your part. There is nothing in the project management process that forbids you from including security as a high priority or making sure the software meets certain quality criteria.

Also where did you get the idea that security didn't come to MongoDB 3 ? You're talking nonsense.

> Everyone uses Lean/Agile now

HN bubble generalization.

Also, was the stupid part necessary?

> If there's a 0day, i'm going to use it

Sure, but with the number of unpatched systems out there most black hats aren't paying (in time/effort or straight cash) for 0-days except for exceptional circumstances. 0-days are great for targeted attacks, state actors, and hack-for-hire/contractors, but for the everyday dude who wants a botnet? Why bother...

Look at Mirai - that wasn't anything close to a 0-day and it had a huuuuge impact. Anyone running a similar botnet could be extorting hosts and providers to this day if it hadn't been used in such a public way.

I've run java apps on boxes that are several years out of date OS patching wise and have many bugs, explain to be how you'd exploit those?

You'd need to:

* Get through the CDN to hit origin * Get through the patched internet facing rproxies * Exploit the JVM to get code execution, while your target area is something like spring * Get a code exec on that box as the appserver user _THEN_ use your j33t exploit..

If an attacker is capable of doing that, then absolutely nothing you can do is capable of stopping them anyway.

Patching is only important for your internet facing stuff, depending on your environment.

I WANT to patch such systems, we all agree that it's poor practice to do so; sometimes life is unfair though and in the grand scheme of things I don't see any significant risk in this scenario..

I worked at a company with that kind of app. It got owned three different ways. A user reported it after they saw it listed in a forum somewhere.

It was mainly app vulns, not platform, but there were certainly things that could have avoided it. The system was just not hardened at all, and the devs were sloppy.

Luckily they had basic network security best practices and so it was confined to frontend, but blah blah SQL blah blah MITM, still not a good situation to be in.

Security is not that much improved to be true.

Docker is just a fancy chroot, a motivated attacker can escape from it, it remains far less secure than a freebsd jail or a solaris zone.

Except maybe some somewhat safer compiled languages (rust, newer versions of C++), and a few interesting tools (like AFL), there was not much improvements in security mechanism in late years.

SELinux is still so annoying that it's commonly the first thing disabled in many install.

A large amount of sysadmins and developers still do a chmod -R 777, not fully understanding what there are doing.

And in the area of micro services, the traditional 3/4 tiers architecture can easily be replaced by a clusterfuck of containers deployed in a flat network where it's impossible to monitor what is speaking with what. And it's much easier to make a mistake like putting a database on the Internet with no authentication whatsoever...

What is more, because of containers, infrastructures tends to be far more diverse than ever: python, node, php, ruby, go, java, each with various version, from various distribution, with various libraries. Each of these environments must be properly monitored, but when you deal with thousands of containers, hundreds of technological stacks, with 3 or 4 versions for each layer, it becomes a nightmare.

In languages that encourage bundling, Java with war webapps for example, I've rarely seen developers properly monitor security issues of what they ship. Too often I've seen only opportunistic updates. And in some extreme cases I've seen products shipping version of libraries 10 or even 15 years old. Containers being a way to generalize bundling, I don't see it being an improvement in that regard. Quite the contrary in fact.

However, to be completely honest, new tools like chef/puppet/containers and new methodologies like Infrastructure as Code somewhat helped to be less frighten about updates as it reduces the risk/impact of something breaking in the process, no more 'don't touch it, it will break comething'. Even containers can help in that regard, it's easier to do AB testing, avoid downtime or put servers in maintenance with them.

Fundamentally, some things are better, some things are worth, and in the end the security level has not really changed.

And honestly, it's a truly frightening view. Software/Infrastructure security is becoming more and more important as sensitive operations are increasingly done through websites (ex: banking operation), private data is increasingly stored in the "cloud" and an increasing number of devices are now connected.

Security flows are still as numerous as before, but their impact, once exploited, tends to be far greater. If it continues like that, something really bad will happen, exactly what is difficult to predict (massive data leak? major industrial disaster caused by a hack? an election being manipulated?), but it will happen.

The OS and even the lower level libs are rarely the first target anymore unless there are some nice pre-packed pwns on msploit or whereever; sure -- there are exploitable bugs in such things but even getting to these hosts are hard now we're in the age of CDN's and ELBs and so on.. Also NX, ASLR, nonexec heap, tools like selinux etc etc have made pwning these as 0day a much harder thing.

So; the noobs have gone upstack. Why bother trying to break out of an ASLR'd JVM behind a CDN with selinux and etc just to get a shell on a box if your goal is to read a .properties so you can dump the db out? Instead you can probably pull the data one rq at a time with some bad form sanitation or whatever? The latter also means you're targetting code the actual target has written which is more likely to have easy to find issues as no one but the target is looking over it than say, openssl which has rather a few eyes on it..

There's a flipside to config mgt from security tho.. Tools like chef/puppet, while enabling you to maintain a _basic_ level of state actually tend to be great attack surfaces and so overall reduce your sec. How many ops teams out there would notice if you pwned one of their lappys and replaced some puppet handler code before they push it into CI and it gets run across their entire estate? Over-Using config mgt (like using it to orcestrate app deploys) with even mid skilled ops folks generally makes an attackers job much easier than say, rolling out immutable containers..

Most places I've seen without much thought have ended up with superpowerful CI boxes that by owning you get absolutely EVERYTHING and they'll happily give out keys to all the devs while not letting them near prod... >_<

If you have a shell account on a CI and the jenkins taskrunner user has access to ssh in anywhere you want to be, then you're just a commit away from having those rights too...

Just my $0.02

(edit: actually, if you just have commit rights to any ./script that is running on that jenkins, you can do everything that jenkins can do without even needing a shell...

Your jenkins builds the artefacts for your app and ALSO does the deploys? Bet you can get into prod with two lines of nc and push)...

> How many ops teams out there would notice if you pwned oneof their lappys

Why are their lappys on the internet? Aren't we discussing online vulnerabilities?

After all, if you pwn an admin-credentialled computer, you can do all sorts of damage in any but the most locked-down shop. Puppet has little to do with it per se - it's the privileges an ops laptop will have that matters.

Why is a laptop on the internet? I mean, it's much easier to pwn an employee laptop and then use that to get everywhere then it is owning breaking into an infra over the internet...

Even if you can't spearphish 'em; several beers at a meetup or three followed by a subtle rpi install when you're somewhere useful usually has better results.

So I've.. Been told.

(edit: The point was though that in almost all targets, all you need is commit rights to a single repo to own everything; so you don't even need a sysadmin just a dev... Hell, PM's and BA's sometimes even have commit rights...)

Googling %27 didn't find much - can you elaborate?

URL-encoded single quote character (for SQL injections).

Reply on news.ycombinator.com