4 ms·
Why I’m not enabling Bitcode (2015)
- MichaelGG 10y agoThis article was criticised the first time on HN. This part is particularly fanciful: >an adversary could find a vulnerability in the LLVM backend to obtain remote code execution on Apple’s Bitcode compilation infrastructure to inject a compiler trojan that would affect every single app on the App Store that was submitted with Bitcode. That seems like a very much "just so" failure scenario. There's no reason to think this would open up a huge, recommend exploitable attack surface. And even if it did, it seems highly unlikely it'd be limited to just apps submitted with Bitcode. Same for the backdoor risk. Thinking suddenly Apple would be a target is silly. There may be good reasons against Bitcode but these security hysterics don't help the argument.
- lern_too_spel 10y agoA compiler provides a very large attack surface. Once an attacker exploits it, they can directly attack bitcode applications without any further exploits. Attacking non-bitcode applications requires exploiting certificate validation as well.
- eridius 10y agoOne would assume Apple is running the compiler backend in a heavily locked-down container or VM.
- lern_too_spel 10y agoBut that sandbox has access to all bitcode apps by necessity. That's the whole point — once the bitcode compiler has been owned, all bitcode apps are immediately vulnerable, while compromising non-bitcode apps requires significantly more additional work.
- gpderetta 10y agoWhy would it have access to all apps? It only need access to the app it is currently compiling.
- lern_too_spel 10y agoDepends on how it is implemented. If it is implemented as the GGP described, as a container or VM back end service that compiles multiple apps, that's a problem. That is precisely the type of problem that would only affect bitcode apps that the top post didn't think could exist.
- dguido 10y agoI completely agree. It's fairly simple to isolate and sandbox individual compiler processes in the cloud and fully mitigate this risk. You can be sure that whatever processes the bitcode on Apple's servers is only supposed to touch a very limited number of files, syscalls, etc. If any cloud instance deviates from that norm, it should be trivial to detect that deviation. So, yes, I would probably try and submit malformed bitcode to Apple if I wanted to be a jerk... but I think Apple has probably thought of this too given how incredibly obvious it is.
- dguido 10y agoAs another security expert, I see tremendous security opportunities presented by bitcode. Apple can start providing high-fidelity feedback to application developers about security risks in their applications through both XCode and the app submission process, and may be able to remediate exploited issues on-the-fly by transforming bitcode.
- bluejekyll 10y agoHow much could you actually transform with a high degree of certainty that you don't break the application? I just don't buy that they is much gain here, for anything. Possibly you'd be able to swap out assembly instructions for new/better ones that the processor provides, but do you really want Apple to be giving your customers transformed application binaries which you haven't tested?
- dguido 10y ago> How much could you actually transform with a high degree of certainty that you don't break the application? Compilers routinely "modify" the programs given to them via optimization passes and the like. This is normal. By having access to bitcode, you're essentially given a second chance at compiling an application. Forget to turn on control flow integrity? Want to optimize out a useless API call? You can fix that. We even do it far more aggressively by lifting binaries back to LLVM bitcode with mcsema, then outputting significantly modified binaries with new functionality. See https://github.com/trailofbits/mcsema https://github.com/trailofbits/mcsema for more info. There is breakage when we do this because we can't be sure of the correct control flow through the application, but you won't have this problem if you're not lifting back from a binary. You're also only focusing on the modification part. Don't forget that bitcode provides a far more suitable target for analysis than binary code.
- bluejekyll 10y agoI wish Apple would drop this entirely. The biggest thing it does is force you as a developer to use their tools. The actual benefits are very limited. It would be more interesting if they encouraged source uploads, they built an entire CI/CD system for testing and releasing your code. Then they could recompile for anyone opting to use that. For some background, read this on Rust and bitcode targets for Apple devices (only appleTV and appleWatch require bitcode, it's optional for everything else): https://github.com/rust-lang/rust/issues/35968 https://github.com/rust-lang/rust/issues/35968
- dguido 10y agoSomething I've learned from being a consultant is that there are an infinite number of possible build system configurations for real applications. Apple gets the best of all worlds by asking for bitcode since it abstracts any unique requirements on the developer end but provides them with something of nearly identical utility as source code.
- bluejekyll 10y agoIt's not though... its not CPU architecture independent, source code is, or can be.
- i336_ 10y agoI still wonder what the real agenda behind bitcode might be. It's hard to imagine it's only to make binary size a tiny bit smaller: it would surely have been easier and faster to have Xcode just categorize each application's assets and binaries by screen size and architecture, eg by pushing a ZIP (or whatever) file of separate binaries. This is assuming otool or a similar utility can't extract individual architectures out of a fat binary, which I imagine you can probably cleanly do as well. The app control thing does have me worried, those concerns are undeniable.