Support X25519Kyber768Draft00 hybrid post-quantum KEM #43

bwesterb · 2023-04-14T11:00:48Z

See https://datatracker.ietf.org/doc/draft-westerbaan-cfrg-hpke-xyber768d00/

I'm not that experienced in Rust, so I believe various things can be done a bit more ergonomically.

As a Kyber implementation, I use the one by Argyle Software. I carefully reviewed the basic implementation (not avx2!) and found some issues. I fixed those in the fork that's actually used. That fork also adds deterministic key generation and an API to convert private->public key.

I considered using liboqs-rust's implementation of Rust instead, but it unfortunately doesn't support the deterministic key generation required.

Test vectors were generated with our Go implementation in Circl.

To do:

Check test vectors

bwesterb · 2023-04-28T17:11:13Z

Hi @rozbb, when would you have time to have a look at this?

rozbb · 2023-05-01T06:06:10Z

Hi Bas. Thanks for this! Sorry, I've been extremely swamped at work. I intend to take a look at this more carefully this week, but on first glance, I'm not sure this is something that should be in the upstream repo. Lots of parts of this, including the primitives and the spec itself, are quite new. I think it might be best to keep this as a separate repo for now, both to deter people from using it prematurely, and because I don't think I have the bandwidth to maintain and update it with the libs and specs.

bwesterb · 2023-05-01T11:26:53Z

Thanks Michael, looking forward to your review.

Lots of parts of this, including the primitives and the spec itself, are quite new.

True. The final version of Kyber, and its integration with HPKE, will be different. Hence the "draft00" in the name. However, it's likely that this version is going to see wide (preliminary) deployment in, among other things, post-quantum ECH. (Just like this version of Kyber already landed in BoringSSL.)

I don't think I have the bandwidth to maintain and update it with the libs and specs.

I will create PRs to keep it up to date with the libs/spec. I understand if you'd only want to add it once it's completely settled.

rozbb · 2023-05-06T06:36:57Z

(sorry, ran checks while looking at this)

First off, these are great edits, and it fits in really well with the existing library! I'm glad the structure was such that a non-DH KEM was more or less pluggable into the framework.

That said, I realized as I was going through the code that I don't think I'll be able to meaningfully look at the PRs and iterations as they come in. It might be best for your own productivity to iterate the feature in your fork, and send a huge lump of a PR when nearing standardization. Does that make sense? Sorry, I wish I had more bandwidth lately to keep up to date with the cool PQ stuff, but I don't :(

See https://datatracker.ietf.org/doc/draft-westerbaan-cfrg-hpke-xyber768d00/

bwesterb · 2023-05-08T09:46:42Z

Thanks for having a look!

and send a huge lump of a PR when nearing standardization.

I think the final PR will be almost identical to the current one, barring the following changes:

codepoint/name for the final hybrid;
bump of Kyber crate for the final version;
updated test vectors;
use ExtractAndExpand(x25519_ss || kyber_ss, kyber_ct) instead of x25519_ss || kyber_ss as shared secret for IND-CCA2 robustness, as the final version of Kyber will not hash the ciphertext in anymore.

bwesterb · 2023-05-19T14:07:55Z

IANA assigned a codepoint for this version of X25519+Kyber768Draft00.

fix: Serde impls for kyber

rozbb · 2023-07-11T04:37:46Z

Sorry for the long delay on this. I think merging into a branch would be appropriate. I'm gonna take a closer look this weekend and hopefully merge. Thank you again for the contribution. It would be very nice to be the first HPKE instance with PQ support once it drops!

rozbb

Left some notes in there, and pushed some changes. Lmk if you agree with them.

One nagging thing I can't solve. In cargo doc --all-features I can't make the xyber KEM get exported to be alongside the other KEMs in the kem module. Do you know how to fix this by any chance?

   Compiling hpke v0.10.0 (/Users/micro/projects/rust-hpke)
warning: ambiguous glob re-exports
  --> src/kem.rs:19:9
   |
14 | pub use dhkem::*;
   |         -------- but the name `EncappedKey` in the type namespace is also re-exported here
...
19 | pub use xyber768d00::*;
   |         ^^^^^^^^^^^^^^ the name `EncappedKey` in the type namespace is first re-exported here
   |
   = note: `#[warn(ambiguous_glob_reexports)]` on by default

Thank you again for this! I agree with the spec as far as I understand it. It's extremely simple, which I think is a feature.

rozbb · 2023-07-17T04:00:28Z

Cargo.toml

+#   https://github.com/Argyle-Software/kyber/issues/73
+#   https://github.com/Argyle-Software/kyber/issues/75
+#   https://github.com/Argyle-Software/kyber/issues/77
+git = "https://github.com/bwesterb/argyle-kyber"


Note for future: we should probably not rely on this code or forks of it. I agree with your commits @bwesterb but I think that library needs quite a bit of work to get to a good state.

There are some good people working on a better Rust implementation of Kyber. Let's promptly switch when it's out.

I did very carefully check every bit of the code in that fork, and I feel confident in deploying it. (Not so much upstream though.)

rozbb · 2023-07-17T05:41:54Z

src/kem/xyber768d00.rs

+
+    const KEM_ID: u16 = 0x30;
+
+    fn derive_keypair(ikm: &[u8]) -> (Self::PrivateKey, Self::PublicKey) {


A note for your spec. DeriveKeypair currently says

For a given KEM, the ikm parameter given to DeriveKeyPair() SHOULD have length at least Nsk, and SHOULD have at least Nsk bytes of entropy.

This probably isn't true anymore. In this case, the recommendation should probably be 96 bytes

The HPKE spec stipulates that

For a given KEM, the ikm parameter given to DeriveKeyPair() SHOULD have length at least Nsk, and SHOULD have at least Nsk bytes of entropy.

For a crypto system where a secret key is some random blob directly reinterpreted, such as with ECC, this makes sense. In general, it doesn't. Kyber768, for instance, stretches a single 32 byte seed into two secret values e_1 and e_2 that have roughly 3119 bits of entropy. That pulls the maximum security down to 256 bits, but that's completely fine as Kyber768 targets a AES-192 level of security anyway.

Is the goal a property like: if one of the underlying primitives is broken, the other holds up? So if say X25519 is broken, and you only have 128 bits of entropy in the Kyber setup, then the security level is below the desired 192, no?

This is offtopic, ofc. Just interested in the hybrid model

Officially Kyber768 targets NIST PQC level 3, which is defined as being as hard to attack as AES-192. The security target of this hybrid is just "128 bits" though, which is the security level of X25519. We picked Kyber768 instead of Kyber512, as the Kyber team themselves are more comfortable with Kyber768 as the default in case of cryptanalytic improvements.

Is the goal a property like: if one of the underlying primitives is broken, the other holds up?

Yes, more precisely, it's designed to be IND-CCA2 robust. That is: if one is broken, the combined hybrid (as used in HPKE) is still IND-CCA2 (classically — not necessarily PQ of course if Kyber is broken). See the security considerations section of the I-D.

So if say X25519 is broken, and you only have 128 bits of entropy in the Kyber setup, then the security level is below the desired 192, no?

There is 256 bits of entropy in the Kyber setup.

Correct me if I'm wrong, I know "entropy math" is famously hand-wavy. But if you have a 32 byte seed that you stretch into 32 + 64 bytes (X25519 and Kyber respectively) and say the entire X25519 is broken, then there should only be 2/3 of the entropy left, ie 170 bits. Similarly if Kyber is fully broken then there's only 1/3 = 85 bits left. Is that wildly wrong?

Let me simplify. I generated a secret 128-bit key X, and computed Y = SHAKE128(X, 256). The first 128 bits of Y are 64ba9be5763117f981a6a47af2352481. With your reasoning, there is no entropy left in X. In a sense there is some merit to it: it's likely there is only one X matching that Y. But, computationally, it's infeasible to find such X: it's the preimage resistance of SHAKE128. Also, finding the second part of Y (without trying to find X first) is computationally hard. The generic security property is the flat sponge claim.

The same happens here: if Kyber is so broken that we can recover its seed, then we're not done yet. To go from that seed, to the original seed, or the seed used for X25519, you'd need to break SHA2.

In fact, a Kyber public key contains a value called rho (used to generate the matrix A), that's derived from the private seed in a similar way. In fact, the last 32 bytes of the private seed are stretched to rho and sigma. The latter is used to generate the final private values s_1 and s_2.

Ah I think I understand now. Thanks for the link too

rozbb · 2023-07-17T05:50:38Z

Cargo.toml

@@ -39,7 +41,8 @@ rand_core = { version = "0.6", default-features = false }
 p256 = { version = "0.13", default-features = false, features = ["arithmetic", "ecdh"], optional = true}
 p384 = { version = "0.13", default-features = false, features = ["arithmetic", "ecdh"], optional = true}
 sha2 = { version = "0.10", default-features = false }
-serde = { version = "1.0", default-features = false, optional = true }
+serde = { version = "1.0", default-features = false, optional = true, features = ["derive"] }
+serde-big-array = { version = "0.5", optional = true }


I don't love this dep. I don't know who maintains it and it uses a good amount of unsafe. Flagging for later.

For context, it's from: bwesterb#2 @OtaK

Completely agree with @rozbb. I don't like this dependency either but until serde adopts const generics (tracking issue here: serde-rs/serde#1937) we'll have to rely on a third party implementation of it.

For context, I also saw that serde_with supports const generics and we could totally switch to this implementation -which seems safer- but for the needs of something that is under a feature flag, it is clearly bloated vs serde-big-array so I prioritized code size for this.

I worked around it entirely. We're already using GenericArray, so I think it makes sense to piggyback off of their serde impls. See newest commit.

bwesterb · 2023-07-17T12:25:04Z

src/kem/xyber768d00.rs

+
+    /// Deterministically derives a keypair from the given input keying material and ciphersuite
+    /// ID. The keying material should have as many bits of entropy as it extracts for its seeds,
+    /// i.e., 768 bits.


This "768 bits" doesn't make sense to me. See this comment.

bwesterb · 2023-07-17T12:29:04Z

Left some notes in there, and pushed some changes. Lmk if you agree with them.

They're a clear improvement, thanks.

There is an issue with the comments on the seed size: I left a comment.

One nagging thing I can't solve. In cargo doc --all-features I can't make the xyber KEM get exported to be alongside the other KEMs in the kem module. Do you know how to fix this by any chance?

Sorry, I don't.

Thank you again for this! I agree with the spec as far as I understand it. It's extremely simple, which I think is a feature.

Thank you! That was a design goal. :)

…GenericArray

rozbb · 2023-07-18T00:39:48Z

OK I'm pretty happy with it now. Lmk what you think about the switch to GenericArray. Only thing left to do is figure out why the build is broken.

bwesterb · 2023-07-18T12:06:31Z

OK I'm pretty happy with it now. Lmk what you think about the switch to GenericArray.

lgtm

See https://datatracker.ietf.org/doc/draft-westerbaan-cfrg-hpke-xyber768d00/ --------- Co-authored-by: Mathieu Amiot <[email protected]> Co-authored-by: Michael Rosenberg <[email protected]>

bwesterb mentioned this pull request Apr 14, 2023

Adding a DeriveKeyPair functionality open-quantum-safe/liboqs#1206

Open

bwesterb marked this pull request as draft April 14, 2023 11:19

bwesterb force-pushed the xyber768d00 branch from 01abe7b to ea7e235 Compare April 14, 2023 14:02

bwesterb marked this pull request as ready for review April 14, 2023 14:02

bwesterb changed the title ~~[WIP] Support X25519Kyber768Draft00 hybrid post-quantum KEM~~ Support X25519Kyber768Draft00 hybrid post-quantum KEM Apr 14, 2023

Support X25519Kyber768Draft00 hybrid post-quantum KEM

71efb5c

See https://datatracker.ietf.org/doc/draft-westerbaan-cfrg-hpke-xyber768d00/

bwesterb force-pushed the xyber768d00 branch from ea7e235 to 71efb5c Compare May 8, 2023 09:36

bwesterb mentioned this pull request May 8, 2023

Add support for X25519Kyber768Draft00 martinthomson/ohttp#49

Merged

OtaK and others added 2 commits June 19, 2023 10:49

fix: Serde impls for kyber

5408f64

Merge pull request #2 from OtaK/xyber768d00

e5ffd22

fix: Serde impls for kyber

bwesterb and others added 8 commits July 11, 2023 12:15

xyber: fix serde by adding derive feature

ad6336c

xyber: satisfy clippy

f6a2b38

Unified TestableDhKem and TestableKem

b04f25c

Remove compiler warning about re-exports

b54166f

Moved PromptedRng to test_util

cbaf776

Started a pass cleaning up and documenting xyber impl

4bf5439

More comments and cleanup to xyber; pinned serde-big-array

10abb20

Docs; added README warning

85b379f

rozbb reviewed Jul 17, 2023

View reviewed changes

bwesterb commented Jul 17, 2023

View reviewed changes

rozbb added 2 commits July 17, 2023 20:33

Removed serde-big-array; switched from fixed-size arrays in Kyber to …

b3217e6

…GenericArray

Fixed xyber DeriveKeypair entropy recommendation

1baf451

rozbb added 3 commits July 17, 2023 23:17

Fixed doc build

81b3b36

Made clippy happy

3027472

Made clippy happy (2)

6aa004f

rozbb changed the base branch from master to unstable-pq-xyber July 18, 2023 14:02

rozbb merged commit e519560 into rozbb:unstable-pq-xyber Jul 18, 2023
12 checks passed

wucke13 mentioned this pull request Jul 27, 2023

Make Serializable trait capable of serializing into user specified buffers #44

Closed

bwesterb mentioned this pull request Jul 28, 2023

Implement PKE Argyle-Software/kyber#30

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Support X25519Kyber768Draft00 hybrid post-quantum KEM #43

Support X25519Kyber768Draft00 hybrid post-quantum KEM #43

bwesterb commented Apr 14, 2023 •

edited

Loading

bwesterb commented Apr 28, 2023

rozbb commented May 1, 2023

bwesterb commented May 1, 2023 •

edited

Loading

rozbb commented May 6, 2023

bwesterb commented May 8, 2023 •

edited

Loading

bwesterb commented May 19, 2023

rozbb commented Jul 11, 2023

rozbb left a comment

rozbb Jul 17, 2023

bwesterb Jul 17, 2023

rozbb Jul 17, 2023 •

edited

Loading

bwesterb Jul 17, 2023 •

edited

Loading

rozbb Jul 17, 2023

bwesterb Jul 17, 2023 •

edited

Loading

rozbb Jul 17, 2023

bwesterb Jul 17, 2023 •

edited

Loading

bwesterb Jul 17, 2023 •

edited

Loading

rozbb Jul 18, 2023

rozbb Jul 17, 2023

bwesterb Jul 17, 2023

OtaK Jul 17, 2023

rozbb Jul 18, 2023

bwesterb Jul 17, 2023

rozbb Jul 18, 2023

bwesterb commented Jul 17, 2023

rozbb commented Jul 18, 2023

bwesterb commented Jul 18, 2023


		const KEM_ID: u16 = 0x30;

		fn derive_keypair(ikm: &[u8]) -> (Self::PrivateKey, Self::PublicKey) {

Support X25519Kyber768Draft00 hybrid post-quantum KEM #43

Support X25519Kyber768Draft00 hybrid post-quantum KEM #43

Conversation

bwesterb commented Apr 14, 2023 • edited Loading

bwesterb commented Apr 28, 2023

rozbb commented May 1, 2023

bwesterb commented May 1, 2023 • edited Loading

rozbb commented May 6, 2023

bwesterb commented May 8, 2023 • edited Loading

bwesterb commented May 19, 2023

rozbb commented Jul 11, 2023

rozbb left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

rozbb Jul 17, 2023 • edited Loading

Choose a reason for hiding this comment

bwesterb Jul 17, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

bwesterb Jul 17, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

bwesterb Jul 17, 2023 • edited Loading

Choose a reason for hiding this comment

bwesterb Jul 17, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

bwesterb commented Jul 17, 2023

rozbb commented Jul 18, 2023

bwesterb commented Jul 18, 2023

bwesterb commented Apr 14, 2023 •

edited

Loading

bwesterb commented May 1, 2023 •

edited

Loading

bwesterb commented May 8, 2023 •

edited

Loading

rozbb Jul 17, 2023 •

edited

Loading

bwesterb Jul 17, 2023 •

edited

Loading

bwesterb Jul 17, 2023 •

edited

Loading

bwesterb Jul 17, 2023 •

edited

Loading

bwesterb Jul 17, 2023 •

edited

Loading