PasswordHash Tutorial


Passlib supports a large number of hash algorithms, all of which can be imported from the passlib.hash module. While the exact options and behavior will vary between each algorithm, all of the hashes provided by Passlib use the same interface, defined by the passlib.ifc.PasswordHash abstract class.

The PasswordHash class provides a generic interface for interacting individually with the various hashing algorithms. It offers methods and attributes for a number of use-cases, which fall into three general categories:

  • Creating & verifying hashes
  • Examining the configuration of a hasher, and customizing the defaults.
  • Assorting supplementary methods.

See also

  • passlib.ifc – API reference of all the methods and attributes of the PasswordHash class.
  • passlib.context.CryptContext – For working with multiple hash formats at once (such a user account table with multiple existing hash formats).

Hashing & Verifying

While all the hashers in passlib.hash offer a range of methods and attributes, the main activities applications will need to perform is hashing and verifying passwords. This can be done with the PasswordHash.hash() and PasswordHash.verify() methods.


Changed in 1.7:

Prior releases used PasswordHash.encrypt() for hashing, which has now been renamed to PasswordHash.hash(). A compatibility alias is present in 1.7, but will be removed in Passlib 2.0.


First, import the desired hash. The following example uses the pbkdf2_sha256 class (which derives from PasswordHash):

>>> # import the desired hasher
>>> from passlib.hash import pbkdf2_sha256

Use PasswordHash.hash() to hash a password. This call takes care of unicode encoding, picking default rounds values, and generating a random salt:

>>> hash = pbkdf2_sha256.hash("password")
>>> hash

Note that since each call generates a new salt, the contents of the resulting hash will differ between calls (despite using the same password as input):

>>> hash2 = pbkdf2_sha256.hash("password")
>>> hash2


Subsequently, you can call PasswordHash.verify() to check user input against an existing hash:

>>> pbkdf2_sha256.verify("password", hash)

>>> pbkdf2_sha256.verify("joshua", hash)

Unicode & non-ASCII Characters

Sidenote regarding unicode passwords & non-ASCII characters:

For the majority of hash algorithms and use-cases, passwords should be provided as either unicode (or utf-8-encoded bytes).

One exception is legacy hashes that were generated using a different character encoding. In this case, passwords should be encoded using the correct encoding before they are passed to verify(); otherwise users may not be able to log in successfully.

For proper internationalization, applications should also take care to ensure unicode inputs are normalized to a single representation before hashing. The passlib.utils.saslprep() function can be used for this purpose.

Customizing the Configuration

The using() Method

Each hasher contains a number of informational attributes. many of which can be customized to change the properties of the hashes generated by PasswordHash.hash(). When you want to change the defaults, you don’t have to modify the hasher class directly, or pass in the options to each call to PasswordHash.hash().

Instead, all the hashes offer a PasswordHash.using() method. This is a powerful method which accepts most hash informational attributes, as well as some other hash-specific configuration keywords; and returns a subclass of the original hasher (or a object with an identical interface). The returned object inherits the defaults settings from it’s parent, but integrates any values you choose to override.


Changed in 1.7:

Prior releases required you to pass custom settings to each PasswordHash.encrypt() call. That usage pattern is deprecated, and will be removed in Passlib 2.0; code should be switched to use PasswordHash.using(), as shown below.

Usage Example

As an example, if the hasher you select supports a variable number of iterations (such as pbkdf2_sha256), you can specify a custom value using the rounds keyword.

Here, the default class uses 29000 rounds:

>>> from passlib.hash import pbkdf2_sha256

>>> pbkdf2_sha256.default_rounds

>>> pbkdf2_sha256.hash("password")

But if we call PasswordHash.using(), we can override this value:

>>> custom_pbkdf2 = pbkdf2_sha256.using(rounds=123456)
>>> custom_pbkdf2.default_rounds

>>> custom_pbkdf2.hash("password")

Other Keywords

While hashes frequently have additional keywords supported by using, the basic set of settings you can customize can be found by inspecting the PasswordHash.setting_kwds attribute:

>>> pbkdf2_sha256.settings_kwds
("salt", "salt_size", "rounds")

For instance, the following generates pbkdf2 hashes with a 32-byte salt instead of the default 16:

>>> pbkdf2_sha256.using(salt_size=8).hash("password")

This method is also used internally by the CryptContext class it order to create a custom hasher configured based on the CryptContext policy it was provided.

See also

Context Keywords

While the PasswordHash.hash() example above works for most hashes, a small number of algorithms require you provide external data (such as a username) every time a hash is calculated.

An example of this is the oracle10 hash, where hashing requires a username:

>>> from passlib.hash import oracle10
>>> hash = oracle10.hash("secret", user="admin")

The difference between this and specifying something like a rounds setting (see Customizing the Configuration above) is that a configuration option only needs to be specified once, and is then encoded into the hash string itself... Whereas a context keyword represents something that isn’t stored in the hash string, and needs to be specified every time you call PasswordHash.hash() or PasswordHash.verify():

>>> oracle10.verify("secret", hash, user="admin")

In this example, if either the username OR password is wrong, verify() will fail:

>>> oracle10.verify("secret", hash, user="wronguser")

>>> oracle10.verify("wrongpassword", hash, user="admin")

Forgetting to include a context keywords when it’s required will cause a TypeError:

>>> hash = oracle10.hash("password")
Traceback (most recent call last):
    <traceback omitted>
TypeError: user must be unicode or bytes, not None

Whether a hash requires external parameters (such as user) can be determined from its documentation page; but also programmatically from its PasswordHash.context_kwds attribute:

>>> oracle10.context_kwds

>>> pbkdf2_sha256.context_kwds

Identifying Hashes

One of the rarer use-cases is the need to identify whether a string recognizably belongs to a given hasher class. This can be important in some cases, because attempting to call PasswordHash.verify() with another algorithm’s hash will result in a ValueError:

>>> from passlib.hash import pbkdf2_sha256, md5_crypt

>>> other_hash = md5_crypt.hash("password")

>>> pbkdf2_sha256.verify("password", other_hash)
Traceback (most recent call last):
    <traceback omitted>
ValueError: not a valid pbkdf2_sha256 hash

This can be prevented by using the identify method, which determines whether a hash belongs to a given algorithm:

>>> hash = pbkdf2_sha256.hash("password")
>>> pbkdf2_sha256.identify(hash)

>>> pbkdf2_sha256.identify(other_hash)

See also

In most cases where an application needs to distinguish between multiple hash formats, it will be more useful to switch to a CryptContext object, which automatically handles this and many similar tasks.


Document usage of PasswordHash.needs_update(), and how it ties into PasswordHash.using().

Choosing the right rounds value

For hash algorithms with a variable time-cost, Passlib’s PasswordHash.default_rounds values attempt to be secure enough for the average [1] system. But the “right” value for a given hash is dependant on the server, its cpu, its expected load, and its users. Since larger values mean increased work for an attacker...

The right rounds value for a given hash & server should be the largest possible value that doesn’t cause intolerable delay for your users.

For most public facing services, you can generally have signin take upwards of 250ms - 400ms before users start getting annoyed. For superuser accounts, it should take as much time as the admin can stand (usually ~4x more delay than a regular account).

Passlib’s default_rounds values are retuned periodically, starting with a rough estimate of what an “average” system is capable of, and then setting all hash.default_rounds values to take ~300ms on such a system. However, some older algorithms (e.g. bsdi_crypt) are weak enough that a tradeoff must be made, choosing “more secure but intolerably slow” over “fast but unacceptably insecure”.

For this reason, it is strongly recommended to not use a value much lower than Passlib’s default, and to use one of recommended hashes, as one of their chief qualifying features is the mere existence of rounds values which take a short enough amount of time, and yet are still considered secure.


Expand this section into a full document, including information from the following posts:

As well as maybe JS-interactive calculation helper.

[1]For Passlib 1.6.3, all hashes were retuned to take ~300ms on a system with a 3.0 ghz 64 bit CPU.