3 ms·

Isn't the idea that you cannot compromise the TPM remotely? At least Apples SecureEnclave require physical access and the fingerprint.

jsnell 5y ago
You cannot compromise the TPM(e.g. read the keys stored in it, decrement increment-only counters), but most of the point of a TPM is that programs can call an interface and have the TPM perform operations using those keys. If someone compromises the machine enough to be able to run arbitrary code, they can issue those TPM operations remotely just the same as the legit software. What you describe is somewhat related, in that one can use a combination of a TPM + some kind of a biometric sensor to build a system like TouchID or Windows Hello, and that combination would not be remotely operable. But if e.g. your mTLS client cert is stored in the TPM, you certainly would not expect to swipe a fingerprint reader on every connection that the browser establishes to the mTLS domains.
- littlecranky67 5y ago
  Thats why most of those TPMs have a requirement of physical interaction, i.e. you need to press a key on a Yubikey, or put a fingerprint down on an iPhone before they start this operation.
heavyset_go 5y ago
I think the idea is that you cannot impersonate a specific TPM remotely. Get root on a system, though, and you can do whatever you want with the hardware you now have access to, including using its TPM for your own purposes.
- littlecranky67 5y ago
  But thats exactly how it doesn't work - you are referring to Software-level TPMs. I.e. the SecureEnclave in iPhones work in a way that you can always put keys INTO the TPM chip, but cannot read it out back later, but rather have the SecureEnclave perform the crypto on chip (requiring some form of authentication, like a fingerprint read or a button press on Yubikeys). Even a full-root access should not compromise that.

3 ms·

Isn't the idea that you cannot compromise the TPM remotely? At least Apples SecureEnclave require physical access and the fingerprint.

by littlecranky67 5y ago

jsnell 5y ago
You cannot compromise the TPM(e.g. read the keys stored in it, decrement increment-only counters), but most of the point of a TPM is that programs can call an interface and have the TPM perform operations using those keys. If someone compromises the machine enough to be able to run arbitrary code, they can issue those TPM operations remotely just the same as the legit software. What you describe is somewhat related, in that one can use a combination of a TPM + some kind of a biometric sensor to build a system like TouchID or Windows Hello, and that combination would not be remotely operable. But if e.g. your mTLS client cert is stored in the TPM, you certainly would not expect to swipe a fingerprint reader on every connection that the browser establishes to the mTLS domains.
- littlecranky67 5y ago
  Thats why most of those TPMs have a requirement of physical interaction, i.e. you need to press a key on a Yubikey, or put a fingerprint down on an iPhone before they start this operation.
heavyset_go 5y ago
I think the idea is that you cannot impersonate a specific TPM remotely. Get root on a system, though, and you can do whatever you want with the hardware you now have access to, including using its TPM for your own purposes.
- littlecranky67 5y ago
  But thats exactly how it doesn't work - you are referring to Software-level TPMs. I.e. the SecureEnclave in iPhones work in a way that you can always put keys INTO the TPM chip, but cannot read it out back later, but rather have the SecureEnclave perform the crypto on chip (requiring some form of authentication, like a fingerprint read or a button press on Yubikeys). Even a full-root access should not compromise that.