Thanks for submitting the article! Let me know if anyone has any questions. If you are wondering how Apple obfuscate its software, the answer is simple: they built some extensions for LLVM that applies code transformation directly to LLVM IR. LLVM IR is an intermediate language to represent the code that will be copied in binary.

Great write-up! I was one of the Apple engineers who implemented some of these obfuscations. Fun to see an outside perspective on your work.

Thanks a lot!!! I can't believe I meet you on HN, great work so far

Thanks for taking the time to write about your reverse engineering experience. I'm more of a programmer than a reverse engineer, but I do like to dabble in RE and analysis and dealing with obfuscation and techniques like virtualization is always quite a challenge. There's probably a lot of learning material out there, but it feels difficult to digest at times, so articles like these are always helpful for getting some perspective on how these things work out in practice, and getting a feel for some of the approaches you can use when you run into a roadblock. Definitely makes me want to go mess around with some binary analysis after work.

Thanks for your comment. This article was written in one go, and it could be way better than the actual state. However I do appreciate that you mean to consider my article as a nice example on how things work out in practice. Being a programmer and knowing a little bit of reverse engineering is important (well, I'm still a programmer too!). If you have any questions, please let me know.

Regarding the difficulty of learning material, I can feel you. It was the same for me when I started (I have a lot of stuff to dig into!), and it's still. Academy papers might be difficult to understand, but once you comprehend the formalisms it'll be easier. For sure, understanding the formal formulas is another challenge that I haven't resolved yet.

> We use IDA for convenience in this article, although we must be especially careful when importing the binary into other tools (we will explain why at the end of the article)

Forgive me if I missed this being explained - I was curious what the reasoning for this was and I didn't see it! Could you elaborate? :)

Ops! Forgot to write about it (otherwise it would be so long). I did not mention the tools, but I was mainly referring to Hopper Decompiler/Disassembler (definitely no no for me). Altough it seemed the natural choice for reverse engineering macOS applications and daemons, it failed disastrousely on reverse engineering fairplayd. This is where obfuscation is really good at: feeling pain. Hopper tried to disassemblate the binary but still no luck (there was an error due to some bogus instructions referred by a dead branch). I'm seeing improvements for Hopper release by release, but there were some regressions that I noticed..

I tried to import it into Ghidra and it missed some informations during the pass of stack analysis. At the end it was a mess result to read, so I ended it up with IDA (free because I'm a student). Binary ninja also needs some license, I'm trying to afford it.

So mainly just Ghidra & Hopper were successfully tripped up by fairplayd - thanks for the explanation!

Great article overall, thanks for taking the time to write it up.

Ghidra was somehow usable, I got several crashes with Hopper. One question for more expert people than me: does Hopper employ any telemetry inside its demo version? Some issues I discovered were fixed in two days and I did not report them.

Great read! As someone who has spent a good chunk of time looking at FairPlay code in the past, it was interesting to follow someone else’s chain of thought :)

Thank you so much! This comment made my day :) The concept about story-telling was to explain in detail how I thought. Despite that, I do agree with some previous comments, some parts are too much.

I'm gonna go out on a limb and suggest that this obfuscation scheme is legacy, and they keep it merely as defense in depth and because it's tested so why not.

Modern Macs can do remote attestations from a trusted boot chain all the way up to specific apps, which obviates the need for this sort of obfuscation. The memory spaces will be protected by the operating system as long as SIP is enabled, and if it's not enabled or has been disabled / the root partition has been modified, then that will be detectable by Apple remotely. Although code obfuscation is fun (I've built a virtualization based obfuscation in the past), a properly implemented remote attestation and security architecture does obsolete it. It's therefore mostly useful on Windows/Linux PCs where these schemes don't really hang together.

Thanks for your comment. This obfuscation scheme is used only to complicate any attempt to reverse engineering (disassembling, decompilation) of the two processes involved in FairPlay DRM (I'm mainly talking about CoreFP and Fairplayd). Attackers need to spend more time on retrieving the original business logic. For sure, it's not the most hyper-defense technology Apple has employed (think about Secure Enclave).

> a properly implemented remote attestation and security architecture does obsolete it [obfuscation]

I'm not sure I've understood this part. So, if Apple implements remote attestation, would it be more difficult for attackers to reverse engineering the application? I am probably missing a point, would you mind if I ask you to expand that?

RA tells Apple what software stack you're running and that it's not virtualized (genuine hardware device). Therefore, you no longer need software obfuscation because:

1. The software is now tamperproofed (server won't release content key unless the RA contains an expected hash)

2. The memory space is protected from being read from other processes, so there's no need to try and hide the processing of secrets in the code itself.

i.e. a system based on RA can be entirely transparent, open source even, and it can still work. The only secrets are the hardware keys that act as the root of trust. The software and hardware stack does itself need to be secure of course, but Apple has got pretty good at that. And btw Apple's platforms already support remote attestation:

https://developer.apple.com/documentation/devicecheck

> The software and hardware stack does itself need to be secure of course, but Apple has got pretty good at that.

lol

Thanks for the detailed explanation.

> The software and hardware stack does itself need to be secure of course

Oh this is what I'm missing. It's a huge assumption that I wish that can be true!

Is it not true? Not everything needs to be secure, just the right parts. The test is whether you can jailbreak the devices and not be detected by their RA scheme. As far as I know, there's no (public) way to do this.

BTW, this tech isn't new. In practice if you are vertically integrating, it's possible to make things secure enough. Games consoles have been doing this for years. Even in the Xbox 360 era, the use of local exploits was detectable the moment you connected to Xbox Live, and AFAIK Xbox One remains completely unmoddable/unjailbroken even after a decade into its lifespan.

There's a tech talk here by a member of the Xbox team who talk about how they secured it against physical attack:

https://www.youtube.com/watch?v=U7VwtOrwceo

But bear in mind, RA was never the weak point even of the 360.

Making remote attestation secure is a well studied problem in the industry. It's been done several times. You have to be a competent tech firm producing your own hardware/software combos, and you need a competent security team, but there are several companies that meet that criteria and Apple is definitely one of them.

So from an attackers perspective nowadays it sounds like focusing on encryption weaknesses and/or hardware issues (e.g. glitching) or firmware vulnerabilities is the right place to look?

I don't know much about Apple's hardware but at least consoles are secured against both those things. They have anti-glitch circuitry. The boot ROM doesn't even do comparisons against computed hashes, it just extends PCRs with them so it's glitch-proof by design even if the core stability monitoring fails. The Xbox One doesn't even expose most of the keys to software at any point. The keys flow from the hardware parts of the security complex to the RAM decryption/hashing engine via dedicated wires on the SoC.

Also, the entire stack is renewable. Unless you find a bug in the boot ROM they will just patch it and months of work will be toast within days. The boot ROMs are (a) encrypted and (b) very heavily reviewed and pen tested. Again, don't know about Apple but all these modern security architectures are more or less the same. The underlying theory is universal and sound, it just boils down to varying levels of cost / effort / backwards compatibility / generality.

So I'd say there are no right places to look anymore. There's always the potential for bugs in the tiny parts of the systems that act as the roots of trust, but these are small pieces of code and it's possible with enough break/fix cycles and review to make them perfect.

All the above rests on a few assumptions:

• Attackers of limited motivation. Xbox guys set a budget of $600 for hacking a specific console. If you're willing to spend more than that on a physical attack then they accept defeat (i.e. FIB workstations are out of scope).

• Platform vendors with tight control over hardware. PCs are insecure against physical attacks by design due to general disagreement and lack of consensus over whether it really matters / what the threat model is. So there are RA schemes but they're hardly used and mostly sold to enterprises wanting to defend against malware.

• Goal is to defend the whole stack. PC platforms can do RA of isolated worlds, this is how SGX works, and it's in theory secure against physical attack (encrypted memory) but SGX enclaves are very limited in what they can do. In theory you could build a secure path to the GPU, but in practice to do that requires a billion NDAs and only works with some GPUs etc and there's no encrypted path for input devices. On iDevices, consoles and other places with vertical integration that's solvable.

God, all this sounds like a nightmare. I can't wait for laws that prohibit platforms/software from refusing service/content to users on the basis of the level of control they have over devices used to interact with it.

Dark times.

Users love this stuff. It lets them buy cheap consoles that are sold below cost and subsidised by game royalties. Heavy gamers subsidise light gamers, and both can effectively "pay off" the true cost of the hardware over time as they buy titles. So it's a bit like zero-interest credit.

Also it eliminates cheating in multiplayer games, and users love that too.

And finally it stops gamers who play by the rules and buy games from feeling like mugs when their mates are playing for free, because there's no piracy.

You think users are going to vote to end all that? They already voted with their feet and embraced consoles on a massive scale. Both console and mobile gaming dwarfs PC gaming.

Consider privacy. One might say "Users love this stuff. They get complex and effective services for free, all in exchange for contributing their data towards ads. Purchasers and advertisers subsidize light users who just consume the content."

And yet, we got laws like GDPR on the ideological basis that personal data is above the concept of "market" and about the individual, period. Your business model be damned.

The same thing should happen here. Both the complete control over all parts/SoCs of a device, and the right to the lack of negative consequences for choosing to exercise that control (such as being second-class citizens on the platform that runs on that device in terms of content/service availability) are paramount to a digital free society, and should be regulated as such, putting them above the concept of "market", just as privacy was.

These same platform controls allow Apple to restrict user data collection from apps. Meanwhile: the notion of a "digital free society" isn't a thing.

I'd much prefer governments use their force of law to make those tracking practices impossible (for anything that isn't an outright criminal enterprise) than a private entity making them technically difficult.

This seems like an argument against defense in depth? I suppose encouraging security researchers to find vulnerabilities is a good thing and obfuscation hinders that.

RA doesn't protect you from kernel exploits.

In some DRM architectures, the key and data encrypted using it is never even exposed to the kernel.

That’s how many high resolution video DRM schemes work.

It lets you detect if the user is running a vulnerable kernel.

Apple's stack is pretty secure. When was the last iPhone jailbreak? I don't follow it closely as I'm not an iPhone user, but it feels like a long time ago now. And if an exploit is found, they can just revoke that kernel version. Apps can then ask users to apply the update to regain access to their streams.

It gets harder and takes longer to jailbreak the iPhone. iOS 15.0 – 15.4.1 was jailbroken - it has been 2 years - I would expect an exploit on 16 already existing and under embargo.

exploits are just kept private by bad actors nowadays.

Apple now gets the worst of both worlds, the harmless jailbreaking scene is dying but the bad actors are still in full force.

How is that the worst of both worlds? The harmless jailbreak world also empowers bad actors.

I never beleived that PR argument

Exactly!

> I'm not sure I've understood this part. So, if Apple implements remote attestation, would it be more difficult for attackers to reverse engineering the application? I am probably missing a point, would you mind if I ask you to expand that?

It's not to prevent RE. Apple can detect tamper so it's no concern if you RE it or not. Anything you learn from that RE work will almost certainly mean tampering with the system...

Dear Sir - a "a properly implemented remote attestation and security architecture" is one aspect of computing, but ironically, Apple Computer built its empire on "personal computers" not military style ID checking.

constant "papers, please" to run My Own Computer is not what I purchased hardware to do.. signed -US Citizen

It’s also likely strong evidence in court that you can’t acquire a copy of protected content without activity that is clearly circumvention.

Why doesn't Apple build a custom TEE to do DRM playback, and authenticate the display as well? That seems like a better solution.

Only an Apple engineer could answer this fully but here are some of my ideas as to why this hasn't been done yet:

-They still support playback on devices with no TEE. Kinda defeats the point of implementing it until this is the case.

- They are wary of moving more functionality into their TEE as it increases the attack surface.

- If the platform is already "attested" and locked down as it is the case today, moving the playback to the TEE provides only a little bit of extra security.

- They are banking on the ARM Realm Management Extension[0] coming to their chips. This would be more of a "catch all" solution to fuck over the owners of their machines in new and exciting ways.

[0]: https://fuse.wikichip.org/news/5699/arm-introduces-its-confi...

What prevents a virus maker from using Realm Management against virus/security scans?

There isn't enough known about the whole thing to answer this. But for what it's worth, Intel's version of this tech has a wiki paragraph dedicated to this concern.

https://en.wikipedia.org/wiki/Software_Guard_Extensions#SGX_...

Nice points! Thank you, I'll dig into ARM RME because it seems pretty interesting

I'm having doubts about what you mean about "TEE", I suppose it's term for "trusted execution environment". I'm not sure about I might answer, and it's probable the right question to ask to Apple engineering. I think Apple can use the same technology they used for Apple Pay via the Secure Chip. I'm talking mainly about exchanging keys via servers, decryption and then serve the content via a sort of "frontend". I wish to know more about TEE to give you a more detailed answer.

EDIT: thinking about secure enclave.. maybe (still an hypothesis) the chip does not have the bandwidth to perform decryption just in time. Probably it's a huge cost for Apple to apply something like that.

They didn't need to. If you have a robust secure boot chain then code running in kernel mode is sufficient (which is what Apple claim FairPlay Streaming is doing). The video is decrypted in kernel space and the drivers enforce HDCP.

Windows has a similar approach, but it's easier to get attacker's code into kernel space there, and PCs can't properly remotely attest so the whole thing doesn't really work (too many possible legit configurations).

> keeping the decryption key and the whole process secure to avoid the possibility of decrypting the contents of.ipa files to share the contents of an app (perhaps paid for) in the wrong hands

> Technically speaking, if there were no protection measures, a person could copy the app installer1 and pass it on to any other person. Result? Loss of revenue on the part of Apple and the developer who published the application since the archive copy is free.

> How can we protect the information contained within the archive? It is clear that we must somehow hide the contents of the archive; by doing so, even if they were to extract the IPA archive from the iPhone, attackers would not be able to access the contents.

I don't like how positive all of that makes DRM sound.

General purpose computing will always be an affront to rentiers.

FairPlay™ sounds as of it was taken straight from Scientology.