txscrypt 3.0.0

How do I store a password?

Easy.

from txscrypt import computeKey
d = computeKey(password)
d.addCallback(storeSomewhere)

computeKey is a function. You give it the plaintext password, in bytes. If your user is giving you a password in Unicode, encode it first. You get a deferred that will fire, at some point, with a magical string of bytes. Store it.

Okay. How do I check a password?

from txscrypt import checkPassword
d = checkPassword(stored, provided)

In this snippet, stored is the thing you got from computeKey. provided is the password as provided by the user. Give it the same treatment you gave the password before you passed it to computeKey. For example, if it’s Unicode, encode it.

You get a deferred. At some point in the future, it will fire with either True if the password matched or False if it didn’t.

Why is the magical string base64-encoded?

You’re not supposed to care about what’s in it.

It’s not entirely base64 encoded, but it does contain some base64 encoded parts. If it weren’t, it’d have a bunch of NUL bytes and other gnarly stuff (i.e. not printable ASCII) in it, and that makes a lot of storage stuff balk.

If you must know, it looks like this:

COMMENT:PARAMS:ENCODED_KEY:ENCODED_SALT

Where COMMENT is currently always txscrypt, PARAMS is a JSON object containing the parameters used to invoke the scrypt hash function (N, r, p, buflen), ENCODED_KEY is the base64 encoded key, ENCODED_SALT is the base64 encoded salt.

Earlier versions of txscrypt used the raw bytes produced by scrypt. Some third party tools bit off those strings after the first NUL byte. Unluckily, this was immediately after the word “scrypt”, which were the first bytes of that string.

But what about salts?

txscrypt takes care of this for you.

(It computes a salt of sufficient length using your OS’ cryptographically secure random number generator. Currently, that length is 256 bits.)

But what about timing attacks?

txscrypt takes care of this for you.

(It relies on salts being of sufficient length. There might be side channels related to multiple executions of scrypt on the same machine; but to the author’s knowledge this has never been demonstrated as an attack.)

But what about starving the thread pool?

txscrypt takes care of this for you.

(It creates a new thread pool just for running scrypt in. This means that scrypt doesn’t compete against, say, DNS resolution, or things you pass to deferToThread.)

But what about shutting down the thread pool?

txscrypt takes care of this for you.

(It tells the reactor to stop the thread pool at the start of its own shutdown procedure. That does mean that you have to have the reactor running for any computations to work.)

When should I create my own Wrapper object?

If you want to change:

• the salt length

• the thread pool

• any of the default scrypt hash parameters (N, r, p, buflen)

So, basically, never. If the number of iterations is insecure, file a bug with txscrypt, so it can be amended.

Changelog