[PATCH] fscrypt: don't use hardware offload Crypto API drivers

[Eric Biggers]

On Fri, Jun 13, 2025 at 11:01:03AM +0200, Maxime MERE wrote:
> Hello,
> 
> On 6/11/25 22:58, Eric Biggers wrote:
> > To protect users from these buggy and seemingly unhelpful drivers that I
> > have no way of testing, let's make fscrypt not use them.  Unfortunately
> > there is no direct support for doing so in the Crypto API, but we can
> > achieve something very close to it by disallowing algorithms that have
> > ASYNC, ALLOCATES_MEMORY, or KERN_DRIVER_ONLY set.
> 
> I agree that software drivers are more efficient and less prone to bugs than
> hardware drivers. However, I would like to highlight the fact that certain
> ST products (the STM32MP2x series) have features that allow the loading of a
> secret key via an internal bus from a Secure OS to the CRYP peripheral
> (usable by the kernel). This enables cryptographic operations to be
> delegated to the non-secure side (the kernel) without exposing the key.
> 
> If fscrypt no longer supports hardware drivers, then this type of
> functionality could not be used, which I find unfortunate because it is
> something that might interest users.

What?  fscrypt doesn't support that anyway, and there isn't any path forward to
supporting it in a way that would actually improve security.  (Considering how
fscrypt's key derivation etc. works.)

fscrypt does support hardware wrapped *inline encryption* keys, which is
actually designed properly and does work.

Honestly, the responses to this thread so far have made it even more clear that
this patch is the right decision.

- Eric