Apple solves the problem in hardware by streaming the persistent storage data through dedicated silicon. Not only is this extremely fast, it guards the key. For the Intel versions of Mac, this was done in the T2 coprocessor.
The flipside of this is availability. Your T2 coprocessor is now permanently tied to your data. This means if the chip dies, there's no recovery unless you have a backup encrypted with a separate key (with its own confidentiality/availability tradeoff).
(And if anything else on your motherboard dies, Apple's official answer is "you're f*cked", since they refuse to do board-level repair.)
For the threat model of most users, where hardware-based targeted attacks aren't a big concern, this is a bad tradeoff.
It's sort of a weird space we're in in the modern world, where you have to assume that anything on a "device" is ephemeral and fragile, and those of us concerned with data persistence on local hardware need to work at having a path to verifiably-restorable backups.
Cloud is a great solution for most people. But not really an option for "where do I put my decades-stale collection of old home directories" or "mbox files from email in the late 90's".
> For the threat model of most users, where hardware-based targeted attacks aren't a big concern, this is a bad tradeoff.
> hardware-based targeted attacks
You mean physical-access attacks, correct?
Is it really just these kinds of attacks that a T2 chip protects against?
AFAIK if malware has super user privilege, it can access the RAM of other processes, and therefore it can access the encryption keys stored in RAM by other processes.
If those processes could have used an encryption API that does the encryption on the chip, and therefore not need to store encryption keys in RAM, they'd be protected against this kind of attack, a kind of attack that is not hardware-based.
Considering those keys are loaded into RAM for/whilst unencripting, i don't see how it matters, cause the malware should have access to the (now) unencripted data regardless.
my colleagues' mbp m2 got into recovery mode after sonoma 14.3 update. it is mdm managed and the recovery key got from the IT is not able to decrypt the disk. login accounts tied to the drive also not able to unlock the disk. IT reached out to apple support and waiting response. because of silicon no other methods like in intel, getting into the machine works. does that mean no way of getting the data back? is it possible to install a fresh mac installation on a external disk and access mbp internal disk for decrypting?
Why does your IT department MDM devices but not have backups and/or why is any substantial amount of data not synced to servers, in repos, etc?
Given how fast networks and storage and processors are these days - if something takes more than an hour or two of diagnostic time and rarely happens, you just nuke the machine and re-image.
Since the device has to be powered on, but locked, and they're using a memory dump, they're referring to the AFU state, right?
From my understanding the BFU state is much harder given that the keys are not in memory but in the TEE. Although there has been exploits for that in the past, not sure if they can extract encryption keys.
It's also interesting to consider what's possible with arbitrary code execution vulnerabilities in early boot stages - would they be able to influence the ARM TZ?
> Over-the-air updates do not change the encryption scheme of a device, because it requires a complete re-encryption of the userdata partition.
JFC that one would have been obvious to think ahead of: what to do if the key derivation becomes compromised?
In an ideal system, each file's metadata should not just specify which master key is to be used, but also which version/KDF scheme was used - that way, newly created files could benefit from stronger encryption schemes and existing files be transparently re-encrypted in the background (or in the foreground, which may be easier to implement technically, but people won't want to wait hours for their upgrade to finish).
Passwords have been doing this for decades, every password in /etc/shadow carries a prefix indicating the hash scheme and settings (salt, rounds), and if the system default changes (like Arch did last year [1]), users get their password updated at login. In user space, for example PHP has a function to check if a re-hash is needed [2].
This paper is from 2021. I don't know the advanced details, but I do know that Android's security is improving quite fast, so it's very possible (and, likely, I'd say) that many of the issues on this paper have already been fixed
