Belt-HMAC-rng#5
Conversation
|
Benchmark on M1: |
| /// Belt-HMAC-HBELT random number generator core. | ||
| pub struct BeltHmacRngCore { | ||
| r: GenericArray<u8, U32>, | ||
| s: GenericArray<u8, U32>, | ||
| key: GenericArray<u8, U32>, | ||
| } | ||
|
|
||
| impl BeltHmacRngCore { | ||
| /// Fill the buffer with random data. | ||
| pub fn fill(&mut self, dest: &mut [u8]) { | ||
| //SAFETY: Key is always present, by default it is filled with zeros | ||
| let mut hmac = HmacHbelt::new_from_slice(&self.key).unwrap(); | ||
| // 𝑌𝑖 ← hmac[ℎ](𝐾, 𝑟 ‖ 𝑆); | ||
| hmac.update(&self.r); | ||
| hmac.update(&self.s); | ||
| let y = hmac.finalize_fixed_reset(); | ||
| dest[..BUFSIZE].copy_from_slice(&y); | ||
|
|
||
| // 𝑟 ← hmac[ℎ](𝐾, 𝑟). | ||
| hmac.update(&self.r); | ||
| hmac.finalize_into_reset(&mut self.r); | ||
| } | ||
| } | ||
|
|
||
| impl BlockRngCore for BeltHmacRngCore { | ||
| type Item = u32; | ||
| type Results = [u32; 8]; | ||
|
|
||
| fn generate(&mut self, results: &mut Self::Results) { | ||
| let mut buf = [0u8; BUFSIZE * BLOCKSIZE]; | ||
| self.fill(&mut buf); | ||
| for i in 0..BLOCKSIZE { | ||
| results[i] = | ||
| u32::from_le_bytes([buf[i * 4], buf[i * 4 + 1], buf[i * 4 + 2], buf[i * 4 + 3]]); | ||
| } | ||
| } | ||
| } | ||
|
|
||
| impl SeedableRng for BeltHmacRngCore { | ||
| type Seed = GenericArray<u8, U64>; | ||
|
|
||
| fn from_seed(seed: Self::Seed) -> Self { | ||
| let key = &seed[..BUFSIZE]; | ||
| let iv = &seed[BUFSIZE..BUFSIZE + BUFSIZE]; | ||
|
|
||
| let mut hmac = HmacHbelt::new_from_slice(key).unwrap(); | ||
|
|
||
| // 𝑟 ← hmac[ℎ](𝐾, 𝑆). | ||
| hmac.update(iv); | ||
| let r = hmac.finalize_fixed_reset(); | ||
|
|
||
| BeltHmacRngCore { | ||
| r, | ||
| s: GenericArray::<u8, U32>::clone_from_slice(iv), | ||
| key: GenericArray::<u8, U32>::clone_from_slice(key), | ||
| } | ||
| } | ||
| } |
There was a problem hiding this comment.
I do not know the plans for this CSPRNGs crate, but I can provide you with a little feedback.
Since the BeltHmacRngCore contains some secret data, you might want to have it impl ZeroizeOnDrop.
generate() appears to use 2 buffers—one named buf that is temporary and discarded in memory, and the other being results. One solution for this would be to keep a temporary buffer GenericArray<u8, U32> member of BeltHmacRngCore. It would be a little more ideal if finalize_into_reset() could work with a GenericArray<u32, U8>, but an alternate solution would be to use core::mem::transmute, union, or zerocopy if the duplicate buffer is a concern.
Having hmac as a member of BeltHmacRngCore could improve performance a little by avoiding initialization every time fill/generate is called.
Then the impl SeedableRng for BeltHmacRngCore might need to call .zeroize() on the input seed, or you could make a wrapper Seed type that implements ZeroizeOnDrop when that feature is activated.
There was a problem hiding this comment.
Since the BeltHmacRngCore contains some secret data, you might want to have it impl ZeroizeOnDrop.
Implemented, thank you.
Having hmac as a member of BeltHmacRngCore could improve performance a little by avoiding initialization every time fill/generate is called.
Sure, performance is increased from 9MB/s to 12MB/s.
Then the impl SeedableRng for BeltHmacRngCore might need to call .zeroize() on the input seed, or you could make a wrapper Seed type that implements ZeroizeOnDrop when that feature is activated.
generate() appears to use 2 buffers—one named buf that is temporary and discarded in memory, and the other being results. One solution for this would be to keep a temporary buffer GenericArray<u8, U32> member of BeltHmacRngCore. It would be a little more ideal if finalize_into_reset() could work with a GenericArray<u32, U8>, but an alternate solution would be to use core::mem::transmute, union, or zerocopy if the duplicate buffer is a concern.
Transmute will broke big endian platform. Don't wanna use union, because unsafe. Will check it later, thank you so much for advises.
2 participants
Hello!
Just implemented rng from STB (Belarus Crypto standard)
https://apmi.bsu.by/assets/files/std/brng-spec25.pdf
@newpavlov