General | |
---|---|
Designers | Jean-Philippe Aumasson, Luca Henzen, Willi Meier, Raphael C.-W. Phan |
Derived from | LAKE[1] |
Successors | BLAKE2 |
Certification | SHA-3 finalist |
Detail | |
Digest sizes | 224, 256, 384 or 512 bits |
Structure | HAIFA construction |
Rounds | 14 or 16 |
Speed | 8.4cpb onCore 2 for BLAKE-256; 7.8 cpb for BLAKE-512 |
BLAKE is acryptographic hash function based onDaniel J. Bernstein'sChaChastream cipher, but a permuted copy of the input block,XORed with round constants, is added before each ChaCha round. LikeSHA-2, there are two variants differing in theword size. ChaCha operates on a 4×4 array of words. BLAKE repeatedly combines an 8-word hash value with 16 message words, truncating the ChaCha result to obtain the next hash value.BLAKE-256 andBLAKE-224 use 32-bit words and produce digest sizes of 256 bits and 224 bits, respectively, whileBLAKE-512 andBLAKE-384 use 64-bit words and produce digest sizes of 512 bits and 384 bits, respectively.
TheBLAKE2 hash function, based on BLAKE, was announced in 2012. TheBLAKE3 hash function, based on BLAKE2, was announced in 2020.
BLAKE was submitted to theNIST hash function competition by Jean-Philippe Aumasson, Luca Henzen, Willi Meier, and Raphael C.-W. Phan. In 2008, there were 51 entries. BLAKE made it to the final round consisting of five candidates but lost toKeccak in 2012, which was selected for theSHA-3 algorithm.
LikeSHA-2, BLAKE comes in two variants: one that uses 32-bit words, used for computing hashes up to 256 bits long, and one that uses 64-bit words, used for computing hashes up to 512 bits long. The core block transformation combines 16 words of input with 16 working variables, but only 8 words (256 or 512 bits) are preserved between blocks.
It uses a table of 16 constant words (the leading 512 or 1024 bits of the fractional part ofπ), and a table of 10 16-element permutations:
σ[0] = 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15σ[1] = 14 10 4 8 9 15 13 6 1 12 0 2 11 7 5 3σ[2] = 11 8 12 0 5 2 15 13 10 14 3 6 7 1 9 4σ[3] = 7 9 3 1 13 12 11 14 2 6 5 10 4 0 15 8σ[4] = 9 0 5 7 2 4 10 15 14 1 11 12 6 8 3 13σ[5] = 2 12 6 10 0 11 8 3 4 13 7 5 15 14 1 9σ[6] = 12 5 1 15 14 13 4 10 0 7 6 3 9 2 8 11σ[7] = 13 11 7 14 12 1 3 9 5 0 15 4 8 6 2 10σ[8] = 6 15 14 9 11 3 0 8 12 2 13 7 1 4 10 5σ[9] = 10 2 8 4 7 6 1 5 15 11 9 14 3 12 13 0
The core operation, equivalent to ChaCha's quarter round, operates on a 4-word column or diagonala b c d
, which is combined with 2 words of messagem[]
and two constant wordsn[]
. It is performed 8 times per full round:
j ← σ[r%10][2×i]// Index computationsk ← σ[r%10][2×i+1]a ← a + b + (m[j] ⊕ n[k])// Step 1 (with input)d ← (d ⊕ a) >>> 16c ← c + d// Step 2 (no input)b ← (b ⊕ c) >>> 12a ← a + b + (m[k] ⊕ n[j])// Step 3 (with input)d ← (d ⊕ a) >>> 8c ← c + d// Step 4 (no input)b ← (b ⊕ c) >>> 7
In the above,r
is the round number (0–13), andi
varies from 0 to 7.
The differences from the ChaCha quarter-round function are:
"BLAKE reuses the permutation of the ChaCha stream cipher with rotations done in the opposite directions. Some have suspected an advanced optimization, but in fact it originates from a typo in the original BLAKE specifications", Jean-Philippe Aumasson explains in his "Crypto Dictionary".[2]
The 64-bit version (which does not exist in ChaCha) is identical, but the rotation amounts are 32, 25, 16 and 11, respectively, and the number of rounds is increased to 16.
Throughout the NIST hash function competition, entrants are permitted to "tweak" their algorithms to address issues that are discovered. Changes that have been made to BLAKE are: the number of rounds was increased from 10/14 to 14/16. This is to be more conservative about security while still being fast.
Hash values of an empty string:
BLAKE-224("") =7dc5313b1c04512a174bd6503b89607aecbee0903d40a8a569c94eedBLAKE-256("") =716f6e863f744b9ac22c97ec7b76ea5f5908bc5b2f67c61510bfc4751384ea7aBLAKE-384("") =c6cbd89c926ab525c242e6621f2f5fa73aa4afe3d9e24aed727faaadd6af38b620bdb623dd2b4788b1c8086984af8706BLAKE-512("") =a8cfbbd73726062df0c6864dda65defe58ef0cc52a5625090fa17601e1eecd1b628e94f396ae402a00acc9eab77b4d4c2e852aaaa25a636d80af3fc7913ef5b8
Changing a single bit causes each bit in the output to change with 50% probability, demonstrating anavalanche effect:
BLAKE-512("The quick brown fox jumps over the lazy dog") =1f7e26f63b6ad25a0896fd978fd050a1766391d2fd0471a77afb975e5034b7ad2d9ccf8dfb47abbbe656e1b82fbc634ba42ce186e8dc5e1ce09a885d41f43451BLAKE-512("The quick brown fox jumps over the lazy dof") =a701c2a1f9baabd8b1db6b75aee096900276f0b86dc15d247ecc03937b370324a16a4ffc0c3a85cd63229cfa15c15f4ba6d46ae2e849ed6335e9ff43b764198a
(In this example 266 matching bits out of 512 is about 52% due to the random nature of the avalanche.)
General | |
---|---|
Designers | Jean-Philippe Aumasson, Samuel Neves,Zooko Wilcox-O'Hearn, Christian Winnerlein |
Derived from | BLAKE |
Detail | |
Digest sizes | up to 64 bytes (BLAKE2b); up to 32 bytes (BLAKE2s); arbitrary (BLAKE2X) |
Rounds | 10 or 12 |
Speed | 3.5cpb onCore i5 (Ivy Bridge) for BLAKE2b[3] |
BLAKE2 is a cryptographic hash function based on BLAKE, created by Jean-Philippe Aumasson, Samuel Neves,Zooko Wilcox-O'Hearn, and Christian Winnerlein. The design goal was to replace the widely used, but broken,MD5 andSHA-1 algorithms in applications requiring high performance in software. BLAKE2 was announced on December 21, 2012.[4] Areference implementation is available underCC0, theOpenSSL License, and theApache License 2.0.[5][6]
BLAKE2b is faster than MD5, SHA-1, SHA-2, and SHA-3, on 64-bit x86-64 and ARM architectures.[5] Its creators state that BLAKE2 provides better security than SHA-2 and similar to that of SHA-3: immunity tolength extension, indifferentiability from a random oracle, etc.[7]
BLAKE2 removes addition of constants to message words from BLAKE round function, changes two rotation constants, simplifies padding, adds parameter block that is XOR'ed with initialization vectors, and reduces the number of rounds from 16 to 12 forBLAKE2b (successor of BLAKE-512), and from 14 to 10 forBLAKE2s (successor of BLAKE-256).
BLAKE2 supports keying, salting, personalization, and hash tree modes, and can output digests from 1 up to 64 bytes for BLAKE2b, or up to 32 bytes for BLAKE2s. There are also parallel versions designed for increased performance onmulti-core processors;BLAKE2bp (4-way parallel) andBLAKE2sp (8-way parallel).
BLAKE2X is a family ofextendable-output functions (XOFs). Whereas BLAKE2 is limited to 64-byte digests, BLAKE2X allows for digests of up to 256 GiB. BLAKE2X is itself not an instance of a hash function, and must be based on an actual BLAKE2 instance. An example of a BLAKE2X instance could beBLAKE2Xb16MiB, which would be a BLAKE2X version based on BLAKE2b producing 16,777,216-byte digests (or exactly 16MiB, hence the name of such an instance).[8]
BLAKE2b and BLAKE2s are specified in RFC 7693. Optional features using the parameter block (salting, personalized hashes, tree hashing, et cetera), are not specified, and thus neither is support for BLAKE2bp, BLAKE2sp, or BLAKE2X.[9]
BLAKE2b uses an initialization vector that is the same as the IV used by SHA-512. These values aretransparently obtained by taking the first 64 bits of the fractional parts of the positive square roots of the first eight prime numbers.
IV0 = 0x6a09e667f3bcc908// Frac(sqrt(2))IV1 = 0xbb67ae8584caa73b// Frac(sqrt(3))IV2 = 0x3c6ef372fe94f82b// Frac(sqrt(5))IV3 = 0xa54ff53a5f1d36f1// Frac(sqrt(7))IV4 = 0x510e527fade682d1// Frac(sqrt(11))IV5 = 0x9b05688c2b3e6c1f// Frac(sqrt(13))IV6 = 0x1f83d9abfb41bd6b// Frac(sqrt(17))IV7 = 0x5be0cd19137e2179// Frac(sqrt(19))
Pseudocode for the BLAKE2b algorithm. The BLAKE2b algorithm uses 8-byte (UInt64) words, and 128-byte chunks.
Algorithm BLAKE2bInput: MMessage to be hashed cbMessageLen: Number, (0..2128)Length of the message in bytes KeyOptional 0..64 byte key cbKeyLen: Number, (0..64)Length of optional key in bytes cbHashLen: Number, (1..64)Desired hash length in bytesOutput: HashHash of cbHashLen bytesInitialize State vectorh withIV h0..7 ← IV0..7Mix key size (cbKeyLen) and desired hash length (cbHashLen) into h0 h0 ← h0 xor 0x0101kknnwherekk is Key Length (in bytes)nnis Desired Hash Length (in bytes)Each time we Compress we record how many bytes have been compressed cBytesCompressed ← 0 cBytesRemaining ← cbMessageLenIf there was a key supplied (i.e. cbKeyLen > 0)then pad with trailing zeros to make it 128-bytes (i.e. 16 words)and prepend it to the messageMif (cbKeyLen > 0)then M ← Pad(Key, 128) || M cBytesRemaining ← cBytesRemaining + 128end ifCompress whole 128-byte chunks of the message, except the last chunkwhile (cBytesRemaining > 128)do chunk ← get next 128 bytes of messageM cBytesCompressed ← cBytesCompressed + 128increase count of bytes that have been compressed cBytesRemaining ← cBytesRemaining - 128decrease count of bytes inM remaining to be processed h ← Compress(h, chunk, cBytesCompressed, false)false ⇒ this is not the last chunkend whileCompress the final bytes fromM chunk ← get next 128 bytes of messageMWe will get cBytesRemaining bytes (i.e. 0..128 bytes) cBytesCompressed ← cBytesCompressed+cBytesRemainingThe actual number of bytes leftover inM chunk ← Pad(chunk, 128)IfM was empty, then we will still compress a final chunk of zeros h ← Compress(h, chunk, cBytesCompressed, true)true ⇒ this is the last chunkResult ← first cbHashLen bytes of little endian state vector hEnd Algorithm BLAKE2b
TheCompress function takes a full 128-byte chunk of the input message and mixes it into the ongoing state array:
Function CompressInput: hPersistent state vector chunk128-byte (16 double word) chunk of message to compress t: Number, 0..2128Count of bytes that have been fed into the Compression IsLastBlock: BooleanIndicates if this is the final round of compressionOutput: hUpdated persistent state vectorSetup local work vector V V0..7 ← h0..7First eight items are copied from persistent state vectorh V8..15 ← IV0..7Remaining eight items are initialized from theIVMix the 128-bit countert into V12:V13 V12 ← V12xor Lo(t)Lo 64-bits of UInt128t V13 ← V13xor Hi(t)Hi 64-bits of UInt128tIf this is the last block then invert all the bits in V14if IsLastBlockthen V14 ← V14xor 0xFFFFFFFFFFFFFFFFTreat each 128-byte messagechunk as sixteen 8-byte (64-bit) wordsm m0..15 ← chunkTwelve rounds of cryptographic message mixingfor ifrom 0to 11doSelect message mixing schedule for this round.BLAKE2b uses 12 rounds, while SIGMA has only 10 entries. S0..15 ← SIGMA[imod 10]Rounds 10 and 11 use SIGMA[0] and SIGMA[1] respectively Mix(V0, V4, V8, V12, m[S0], m[S1]) Mix(V1, V5, V9, V13, m[S2], m[S3]) Mix(V2, V6, V10, V14, m[S4], m[S5]) Mix(V3, V7, V11, V15, m[S6], m[S7]) Mix(V0, V5, V10, V15, m[S8], m[S9]) Mix(V1, V6, V11, V12, m[S10], m[S11]) Mix(V2, V7, V8, V13, m[S12], m[S13]) Mix(V3, V4, V9, V14, m[S14], m[S15])end forMix the upper and lower halves of V into ongoing state vector h h0..7 ← h0..7xor V0..7 h0..7 ← h0..7xor V8..15Result ← hEnd Function Compress
TheMix function is called by theCompress function, and mixes two 8-byte words from the message into the hash state. In most implementations this function would be written inline, or as an inlined function.
Function MixInputs: Va, Vb, Vc, Vdfour 8-byte word entries from the work vector V x, ytwo 8-byte word entries from padded message mOutput: Va, Vb, Vc, Vdthe modified versions of Va, Vb, Vc, Vd Va ← Va + Vb + xwith input Vd ← (Vdxor Va)rotateright 32 Vc ← Vc + Vdno input Vb ← (Vbxor Vc)rotateright 24 Va ← Va + Vb + ywith input Vd ← (Vdxor Va)rotateright 16 Vc ← Vc + Vdno input Vb ← (Vbxor Vc)rotateright 63Result ← Va, Vb, Vc, VdEnd Function Mix
Hash values of an empty string:
BLAKE2s-224("") =1fa1291e65248b37b3433475b2a0dd63d54a11ecc4e3e034e7bc1ef4BLAKE2s-256("") =69217a3079908094e11121d042354a7c1f55b6482ca1a51e1b250dfd1ed0eef9BLAKE2b-384("") =b32811423377f52d7862286ee1a72ee540524380fda1724a6f25d7978c6fd3244a6caf0498812673c5e05ef583825100BLAKE2b-512("") =786a02f742015903c6c6fd852552d272912f4740e15847618a86e217f71f5419d25e1031afee585313896444934eb04b903a685b1448b755d56f701afe9be2ce
Changing a single bit causes each bit in the output to change with 50% probability, demonstrating anavalanche effect:
BLAKE2b-512("The quick brown fox jumps over the lazy dog") =a8add4bdddfd93e4877d2746e62817b116364a1fa7bc148d95090bc7333b3673f82401cf7aa2e4cb1ecd90296e3f14cb5413f8ed77be73045b13914cdcd6a918BLAKE2b-512("The quick brown fox jumps over the lazy dof") =ab6b007747d8068c02e25a6008db8a77c218d94f3b40d2291a7dc8a62090a744c082ea27af01521a102e42f480a31e9844053f456b4b41e8aa78bbe5c12957bb
In addition to the reference implementation,[6] the following cryptography libraries provide implementations of BLAKE2:
General | |
---|---|
Designers | Jack O'Connor, Samuel Neves, Jean-Philippe Aumasson,Zooko Wilcox-O'Hearn |
First published | January 9, 2020; 5 years ago (2020-01-09) |
Derived from | Bao,BLAKE2 |
Detail | |
Digest sizes | 256 bits, arbitrarily extensible |
Structure | Merkle tree |
Rounds | 7 |
Speed | 0.49cpb on Cascade Lake-SP with AVX-512[25] |
BLAKE3 is a cryptographic hash function based on Bao and BLAKE2, created by Jack O'Connor, Jean-Philippe Aumasson, Samuel Neves, andZooko Wilcox-O'Hearn.[26] It was announced on January 9, 2020, atReal World Crypto.[27]
BLAKE3 is a single algorithm with many desirable features (parallelism,XOF,KDF,PRF andMAC), in contrast to BLAKE and BLAKE2, which are algorithm families with multiple variants. BLAKE3 has abinary tree structure, so it supports a practically unlimited degree of parallelism (both SIMD and multithreading) given long enough input. The officialRust andC implementations[28] aredual-licensed as public domain (CC0) and theApache License.[29]
BLAKE3 is designed to be as fast as possible. It is consistently a few times faster than BLAKE2. The BLAKE3 compression function is closely based on that of BLAKE2s, with the biggest difference being that the number of rounds is reduced from 10 to 7, a change based on the assumption that current cryptography is too conservative.[30] In addition to providing parallelism, the Merkle tree format also allows for verified streaming (on-the-fly verifying) and incremental updates.[28]