4 ms·

Thanks for your comment. This obfuscation scheme is used only to complicate any attempt to reverse engineering (disassembling, decompilation) of the two process

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?

baby_souffle 3y ago
> 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...
mike_hearn 3y ago
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 https://developer.apple.com/documentation/devicecheck
- nicolodev 3y ago
  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!
  - mike_hearn 3y ago
    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 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.
    - drvdevd 3y ago
      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?
      - mike_hearn 3y ago
        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.

baby_souffle 3y ago

> 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...

mike_hearn 3y ago

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

nicolodev 3y ago
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!
- mike_hearn 3y ago
  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 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.
  - drvdevd 3y ago
    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?
    - mike_hearn 3y ago
      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.