看了一圈觉得还是WikiPedia的比较通俗靠谱:
The large memory requirements of scrypt come from a large vector of pseudorandom bit strings that are generated as part of the algorithm. Once the vector is generated, the elements of it are accessed in a pseudo-random order and combined to produce the derived key. A straightforward implementation would need to keep the entire vector in RAM so that it can be accessed as needed.
Because the elements of the vector are generated algorithmically, each element could be generated on the fly as needed, only storing one element in memory at a time and therefore cutting the memory requirements significantly. However, the generation of each element is intended to be computationally expensive, and the elements are expected to be accessed many times throughout the execution of the function. Thus there is a significant trade-off in speed in order to get rid of the large memory requirements.
This sort of time–memory trade-off often exists in computer algorithms: speed can be increased at the cost of using more memory, or memory requirements decreased at the cost of performing more operations and taking longer. The idea behind scrypt is to deliberately make this trade-off costly in either direction. Thus an attacker could use an implementation that doesn’t require many resources (and can therefore be massively parallelized with limited expense) but runs very slowly, or use an implementation that runs more quickly but has very large memory requirements and is therefore more expensive to parallelize.
The algorithm includes the following parameters:
Function scrypt
Inputs:
Passphrase: Bytes string of characters to be hashed
Salt: Bytes random salt
CostFactor (N): Integer CPU/memory cost parameter
BlockSizeFactor (r): Integer blocksize parameter (8 is commonly used)
ParallelizationFactor (p): Integer Parallelization parameter. (1..232-1 * hLen/MFlen)
DesiredKeyLen: Integer Desired key length in bytes
Output:
DerivedKey: Bytes array of bytes, DesiredKeyLen long
Step 1. Generate expensive salt
blockSize ← 128*BlockSizeFactor //Length (in bytes) of the SMix mixing function output (e.g. 128*8 = 1024 bytes)
Use PBKDF2 to generate initial 128*BlockSizeFactor*p bytes of data (e.g. 128*8*3 = 3072 bytes)
Treat the result as an array of p elements, each entry being blocksize bytes (e.g. 3 elements, each 1024 bytes)
[B0...Bp−1] ← PBKDF2HMAC-SHA256(Passphrase, Salt, 1, blockSize*ParallelizationFactor)
Mix each block in B 2CostFactor times using ROMix function (each block can be mixed in parallel)
for i ← 0 to p-1 do
Bi ← ROMix(Bi, 2CostFactor)
All the elements of B is our new "expensive" salt
expensiveSalt ← B0∥B1∥B2∥ ... ∥Bp-1 //where ∥ is concatenation
Step 2. Use PBKDF2 to generate the desired number of bytes, but using the expensive salt we just generated
return PBKDF2HMAC-SHA256(Passphrase, expensiveSalt, 1, DesiredKeyLen);
Function ROMix(Block, Iterations)
Create Iterations copies of X
X ← Block
for i ← 0 to Iterations−1 do
Vi ← X
X ← BlockMix(X)
for i ← 0 to Iterations−1 do
//Convert first 8-bytes of the last 64-byte block of X to a UInt64, assuming little endian (Intel) format
j ← Integerify(X) mod N
X ← BlockMix(X xor Vj)
return X
Where Integerify is a bijective function from {0, 1}k to {0,...,2k− 1}.
Function BlockMix(B):
The block B is r 128-byte chunks (which is equivalent of 2r 64-byte chunks)
r ← Length(B) / 128;
Treat B as an array of 2r 64-byte chucks
[B0...B2r-1] ← B
X ← B2r−1
for i ← 0 to 2r−1 do
X ← Salsa20/8(X xor Bi) //Salsa20/8 hashes from 64-bytes to 64-bytes
Yi ← X
return ← Y0∥Y2∥...∥Y2r−2 ∥ Y1∥Y3∥...∥Y2r−1
Where Salsa20/8 is the 8-round version of Salsa20.