Am Mittwoch, 19. Juli 2017, 10:20:07 CEST schrieb Horia Geantă:
Hi Horia,
> Hi,
>
> AFAICS, IEEE 1619-2007 standard mentions only XTS-AES-128 and
> XTS-AES-256, meaning that the keys should be either 256 or 512 bits.
>
> Further, NIST SP800-38E mentions that an implementation may restrict
> support to only one of XTS-AES-{128,256}, but does not explicitly allow
> other cipher suites.
>
> I don't see this enforcement in crypto/xts.c (XTS SW implementation),
> though I noticed that XTS mode is used for other ciphers besides AES.
>
> More, tcrypt has xts(aes) speed tests checking also for XTS-AES-192.
> For HW crypto engines adhering strictly to IEEE P1619-2007 this is a
> problem:
> [...]
> tcrypt: test 5 (384 bit key, 16 byte blocks):
> caam_jr 8020000.jr: key size mismatch
> tcrypt: setkey() failed flags=200000
> [...]
>
> Is the kernel more permissive than the standard?
The standard(s) only define XTS for AES 128/256. It does not (according to my
knowledge) prohibit the use of other key sizes (or even other ciphers, such as
xts(serpent)).
Thus, I would say that the kernel implements the standard and allows a more
flexible use than the standard specifies.
Unfortunately, the kernel crypto API only allows a min and max keysize
specification in the cipher definition structure.
Once solution could be to introduce another flag. When that flag is set,
xts_check_key() could deny any key other than 256 or 512 bits. But maybe that
solution is more of a hack than a real solution.
Or tcrypt could be changed to not use XTS with 192/384 bit keys.
Ciao
Stephan
