notfed / puree Goto Github PK

Designed a network protocol without plaintext markup myself and used... uh... the only possible way to organize such protocol :) My protocol has no configuration freedom, though.

If you support slots, is there a reason to not use them? This way you can generate proper key and change password and key independently. I would use simple array of slots as the first header.

header1

slot
slot
...
slot

slot

salt
mac
key
header2 address

header2

type
data

Leave it to the second header to specify its type and any other chained headers.

If you want to provide 3 bits of public information, you can encode them in random data. Provide a short salt then take 3 bits from hash(salt,password). If all combinations are equally probable, it doesn't expose information about password. You will need to generate the salt 8 times to see which gives necessary value. But those three metaparameters don't look very useful. CipherHasParams - isn't this flag provided by the header type? DontStretchPassword - sure this can be tried unconditionally and doesn't need to be encoded anywhere, you also duplicate it as parameter character and to prevent ambiguity it looks like you need to specify the parameter character unconditionally. UseAntiForensicRegion - you didn't specify the design rationale which seems to be ensured destruction by partial overwrite, then it doesn't hurt to use it unconditionally.

Hi - very nice paper. There is another effort done by the dedis lab at EPFL, Switzerland: https://github.com/dedis/purb. It has some work on how to efficiently store different encryption schemes.