I am trying to implement the HMAC algorithm in Mathematica. The algorithm is a relatively simple way of signing messages with a key.
Here is some pseudocode from Wikipedia:
My attempt:
hash[string_String, method_String : "SHA256"] := FromCharacterCode @
Interpreter["HexInteger"] @ StringPartition[IntegerString[Hash[string, method], 16], 2];
HMAC[key_String, message_String, method_String : "SHA256", blockSize_Integer : 64] :=
Module[
{char54, char92, key2, ipad, opad},
{char54, char92} = FromCharacterCode /@ {54, 92};
key2 = Switch[StringLength @ key,
blockSize, key,
l_ /; l > blockSize, hash[key, method],
_, StringPadRight[key, blockSize, FromCharacterCode @ 0]];
ipad = FromCharacterCode[BitXor @@ Map[ToCharacterCode, {StringRepeat[char54, blockSize], key2}]];
opad = FromCharacterCode[BitXor @@ Map[ToCharacterCode, {StringRepeat[char92, blockSize], key2}]];
hash @ StringJoin[opad, hash @ StringJoin[ipad, message]]];
A Java analog for testing:
Needs["JLink`"];
LoadJavaClass["javax.crypto.Mac"];
javaHMAC[key_, message_] := Module[
{mac, keySpec, bytes},
JavaBlock[
mac = Mac`getInstance["HmacSHA256"];
keySpec = JavaNew["javax.crypto.spec.SecretKeySpec", ToCharacterCode[key], "HmacSHA256"];
mac @ init[keySpec];
bytes = mac @ doFinal[ToCharacterCode[message]] + 256;
FromCharacterCode @ bytes]];
In my particular case I am using this to communicate with Amazon Web Services. At the bottom of this page they provide a test vector which involves repeated HMAC computations. For these inputs (using outputs as the key for the next input)
key = "AWS4" <> "wJalrXUtnFEMI/K7MDENG+bPxRfiCYEXAMPLEKEY";
messages = {"20120215", "us-east-1", "iam", "aws4_request"};
we should obtain
"kSecret = '41575334774a616c725855746e46454d492f4b374d44454e472b62507852666943594558414d504c454b4559'
kDate = '969fbb94feb542b71ede6f87fe4d5fa29c789342b0f407474670f0c2489e0a0d'
kRegion = '69daa0209cd9c5ff5c8ced464a696fd4252e981430b10e3d3fd8e2f197d7a70c'
kService = 'f72cfd46f26bc4643f06a11eabb6c0ba18780c19a8da0c31ace671265e3c87fa'
kSigning = 'f4780e2d9f65fa895f9c67b32ce1baf0b0d8a43505a000a1a9e090d414db404d'";
StringSplit[StringSplit[%, "\n"], "= "][[All, 2]];
correct = StringReplace[%, "'" -> ""];
My function passes this test. Swapping it out for the Java method will show that the same results are obtained.
FoldList[HMAC, key, messages];
StringJoin /@ IntegerString[ToCharacterCode @ %, 16, 2];
MapThread[SequenceAlignment, {%, correct}];
% // Column
{41575334774a616c725855746e46454d492f4b374d44454e472b62507852666943594558414d504c454b4559}
{969fbb94feb542b71ede6f87fe4d5fa29c789342b0f407474670f0c2489e0a0d}
{69daa0209cd9c5ff5c8ced464a696fd4252e981430b10e3d3fd8e2f197d7a70c}
{f72cfd46f26bc4643f06a11eabb6c0ba18780c19a8da0c31ace671265e3c87fa}
{f4780e2d9f65fa895f9c67b32ce1baf0b0d8a43505a000a1a9e090d414db404d}
But on a slightly different set of messages, from the bottom of this page, my function fails on one of the steps.
messages2 = {"20110909", "us-east-1", "iam", "aws4_request"};
FoldList[HMAC, key, messages2];
hex2 = StringJoin /@ IntegerString[ToCharacterCode @ %, 16, 2];
FoldList[javaHMAC, key, messages2];
javaHex2 = StringJoin /@ IntegerString[ToCharacterCode @ %, 16, 2];
MapThread[SequenceAlignment, {hex2, javaHex2}] // Column
{41575334774a616c725855746e46454d492f4b374d44454e472b62507852666943594558414d504c454b4559}
{a83ed188be5f4b074d7f66349f5077fbcdf797bf3471fb9d5f32730f936d41a5}
{957b5875f33834a85374b750011dc2d6f0e1d6896eeb891d36a73c711961ad6e}
{{,0},116249d060bff83faa1a627e85a4c6f83ce50d89c334765878dcf76e28bfc6,{,e}}
{{a4b,98f1},d,{,889f},e,{1,c},4,{b2f,f44},21,{2,adc},5,{8145,22b},a,{,b0ce1},f,{6,},8,{1,2e6},9,{d7,29},c,{4e4,262},e,{17a88,d},1,{31,5e},5a9,{f85,4c9},0,{33,ef},d,{b,},1,{d858,},e,{6,},3,{f52,b0e},7,{51,}}
In this case the Java output matches the answer provided by AWS.
I cannot get this implementation to work either.
Answer
Try something like
hash[string_String, method_String: "SHA256"] :=
FromCharacterCode@IntegerDigits[Hash[string, method], 256, 32]
The original method you used will drop leading zeros. You can even see that in your example data set:
{{,0},116249d060bff83faa1a627e85a4c6f83ce50d89c334765878dcf76e28bfc6,{,e}}
ETA: I should have mentioned that passing in the algorithm but then hard-coding the length of the expected hash to 32 bytes is extremely sloppy of me. Do not do that in production code.
Comments
Post a Comment