Re: [TLS] DTLS Key Separation PR

[Eric Rescorla <[e-mail]>]

On Wed, Oct 9, 2019 at 7:01 PM Martin Thomson <[e-mail]> wrote:

> tl;dr keep the space.
>
> I had a little trouble reproducing the 12 from RFC 8446, so I
> double-checked.
>
> ....
>
> Working from the base for SHA-256:
>
> The last block of SHA-256 is rounded up to 448 bits (56 bytes), less one
> to allow for padding.  Therefore we have 55 bytes to use without having to
> run two blocks through SHA-256.
>
> HMAC-Hash = H(K XOR opad || H(K XOR ipad || text))
>
> Here `K XOR ipad` is the 32 bytes output size of SHA-256, so we are down
> to 23 bytes for text before it adds a block.
>

In HMAC the key consumes an entire input block, so that you are guaranteed
a compression function cycle before the data is added:

1) append zeros to the end of K to create a B byte string
        (e.g., if K is of length 20 bytes and B=64, then K will be
         appended with 44 zero bytes 0x00)
    (2) XOR (bitwise exclusive-OR) the B byte string computed in step
        (1) with ipad

https://tools.ietf.org/rfcmarkup?doc=2104#section-2

This shifts everything to the right 32 bytes, which, because context
is 32 bytes, gets us back to 12, I think, though I haven't re-worked
through it.

Here's Ilari's original analysis, if that's helpful:
https://mailarchive.ietf.org/arch/msg/tls/n_TFape7L4HHoKLxGo8CkimZziI

-Ekr






> HKDF-Expand =  HMAC-Hash(PRK, info | 0x01)
>
> This takes one more.  Down to 22.
>
> HKDF-Expand-Label passes info in the form of:
>
>        struct {
>            uint16 length = Length;
>            opaque label<7..255> = "tls13 " + Label;
>            opaque context<0..255> = Context;
>        } HkdfLabel;
>
> which has a minimal overhead of 2 + 1 + len("tls13 ") + 1 = 10.  So we get
> 12.
>
> "c ap traffic" is 12 bytes long, so yeah it *looks* like we're stuck if we
> care about not adding too many extra hash iterations.
>
> ....
>
> But if you look at the key schedule, we always provide a context for those
> cases we use "c ap traffic".  Those will always spill over into the next
> iteration as Context is 32 bytes.  So for those cases, we have a whole 32
> bytes extra to play with.  The only cases with an empty Context are
> "derived" and "res binder"|"ext binder".  Those max out at 10, so we seem
> to have two whole bytes of wiggle room.
>
> You can safely add the space.
>
> On Wed, Oct 2, 2019, at 08:40, Eric Rescorla wrote:
> > Hi folks,
> >
> > As discussed in Montreal, I've prepared a PR to give us DTLS/TLS key
> separation.
> >
> > See:
> > https://github.com/tlswg/dtls13-spec/pull/99
> >
> > Sadly. we didn't have enough space for "dtls13 " so I went for "dtls13"
> >
> > -Ekr
> >
> > _______________________________________________
> > TLS mailing list
> > [e-mail]
> > https://www.ietf.org/mailman/listinfo/tls
> >
>
> _______________________________________________
> TLS mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/tls
>