Re: [PATCH v2] fscrypt: support trusted keys

[Jarkko Sakkinen]

On Mon, Aug 09, 2021 at 12:00:40PM +0200, Ahmad Fatoum wrote:
> Hello Jarkko,
> 
> On 09.08.21 11:44, Jarkko Sakkinen wrote:
> > On Fri, Aug 06, 2021 at 05:09:28PM +0200, Ahmad Fatoum wrote:
> >> Kernel trusted keys don't require userspace knowledge of the raw key
> >> material and instead export a sealed blob, which can be persisted to
> >> unencrypted storage. Userspace can then load this blob into the kernel,
> >> where it's unsealed and from there on usable for kernel crypto.
> >>
> >> This is incompatible with fscrypt, where userspace is supposed to supply
> >> the raw key material. For TPMs, a work around is to do key unsealing in
> >> userspace, but this may not be feasible for other trusted key backends.
> >>
> >> Make it possible to benefit from both fscrypt and trusted key sealing
> >> by extending fscrypt_add_key_arg::key_id to hold either the ID of a
> >> fscrypt-provisioning or a trusted key.
> >>
> >> A non fscrypt-provisioning key_id was so far prohibited, so additionally
> >> allowing trusted keys won't break backwards compatibility.
> >>
> >> Signed-off-by: Ahmad Fatoum <a.fatoum@xxxxxxxxxxxxxx>
> >> ---
> >> Tested with:
> >> https://github.com/google/fscryptctl/pull/23
> >> -	if (key->type != &key_type_fscrypt_provisioning)
> >> -		goto bad_key;
> >> -	payload = key->payload.data[0];
> >> +	if (key->type == &key_type_fscrypt_provisioning) {
> > 
> > Why does fscrypt have own key type, and does not extend 'encrypted' with a
> > new format [*]?
> 
> See the commit[1] adding it for more information. TL;DR:
> 
> fscrypt maintainers would've preferred keys to be associated with
> a "domain". So an encrypted key generated for fscrypt use couldn't be reused
> for e.g. dm-crypt. They are wary of fscrypt users being more exposed if their
> keys can be used with weaker ciphers via other kernel functionality that could
> be used to extract information about the raw key material.
> 
> Eric also mentioned dislike of the possibility of rooting encrypted keys to
> user keys. v2 is only restricted to v2, so we didn't discuss this further.
> 
> Restricting the key to fscrypt-only precludes this reuse.
> 
> My commit makes no attempts in changing that. It just adds a new way to pass
> raw key material into fscrypt. For more information, see the commit[1] adding
> that key type.
> 
> > [*] https://www.kernel.org/doc/html/v5.13/security/keys/trusted-encrypted.html
> 
> [1]: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=93edd392ca

OK, so why does the trusted key does not seal a fscrypt key, but instead
its key material is directly used?

> Cheers,
> Ahmad

/Jarkko