Make timeout property for PasswordResetTokenGenerator

[Alexander Voloshchenko]

During project development our team needs to create several types of tokens. One of them will be used in case of account reset password. The second one is for account activation. Django itself has a good class for token generation called PasswordResetTokenGenerator. And now for account activation, we are using our own class called ActivationTokenGenerator, a subclass of PasswordResetTokenGenerator with overridden _make_hash_value method. And it works, but there is one problem. And this problem is called "timeout". For now, every token created with PasswordResetTokenGenerator will have timeout from settings.PASSWORD_RESET_TIMEOUT variable and can be changed only by changing this variable value. But what if we need different timeouts for different tokens? And I don't think we want changing timeout for activation token using a variable which is screaming about password reset (PASSWORD_RESET_TIMEOUT), we would like to use smth called ACTIVATION_TOKEN_TIMEOUT
So there is a solution: why not create a timeout property for PasswordResetTokenGenerator class? Almost in the same way as it was done with _secret and algorithm fields.So our development team come up with an idea to create a PR which will add this functionality to the Django project. But before this we decided to search similar solutions in django PRs. And we found them! Ticket 30423 https://code.djangoproject.com/ticket/30423 sounds good enough, but it was closed with wontfix label.So the question is: why not to add this fine feature to the PasswordResetTokenGenerator ? And if people find this useful - why not to merge one o the existing PRs?

[Danilov Maxim]

Hi Alexander.

 

You can simply override ‘check_token’ to avoide this harcoded settings.PASSWORD_RESET_TIMEOUT on the 57 line.

 

And for yours task it can be with super call  and after add additional check.

 

It is only some additional lines:

 

MyСlassFromPasswordResetTokenGenerator(…):

 

    MY_OWN_TIMEOUT = your_timeout

 

    def check_token(self, *args, **kwargs):

        return super().check_token(*args, **kwargs) or self._my_check_token_function_with_other_timeout(*args, **kwargs))

 

_my_check_token_function_with_other_timeout – should check if super returns false not from last ‘if’ in PasswordResetTokenGenerator.check_token

 

For us it works without any problem.

 

Mit freundlichen Grüßen,

DI Mag. Maxim Danilov

 

+43(681)207 447 76

ma...@wpsoft.at

--
You received this message because you are subscribed to the Google Groups "Django developers (Contributions to Django itself)" group.
To unsubscribe from this group and stop receiving emails from it, send an email to django-develop...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/django-developers/4f810acc-f2b4-4931-a84e-a59332a825f1n%40googlegroups.com.

[Alexander Voloshchenko]

Hi Max!

Thanks for your answer

But honestly, I don't really understand how You can check if super returns false not from the last 'if'.

Now the function has 5 places where it can return False, and yes, we don't like the last case specifically. But to be honest I don't see a way to check the first 4 without copying them from the framework code to our project (which sounds a bit wrong for us and absolutely not what we would like to see in a project)

Maybe I misunderstood You and maybe You have time to expand the concept because now I can hardly imagine it without copying code from django into the project codebase.

[Carlton Gibson]

Hi Alexander. 

I think the point from ticket #30423 is that we don't want to complicate PasswordResetTokenGenerator in order to make it into a more general purpose token generator. 

Much better is for you to take inspiration from it to implement what you need in your project. (It's the underlying signing code that you really want to lean on Django for.) 

Kind Regards,

Carlton

On Mon, 22 Aug 2022 at 17:26, Alexander Voloshchenko <voloshchen...@gmail.com> wrote:

During project development our team needs to create several types of tokens. One of them will be used in case of account reset password. The second one is for account activation. Django itself has a good class for token generation called PasswordResetTokenGenerator. And now for account activation, we are using our own class called ActivationTokenGenerator, a subclass of PasswordResetTokenGenerator with overridden _make_hash_value method. And it works, but there is one problem. And this problem is called "timeout". For now, every token created with PasswordResetTokenGenerator will have timeout from settings.PASSWORD_RESET_TIMEOUT variable and can be changed only by changing this variable value. But what if we need different timeouts for different tokens? And I don't think we want changing timeout for activation token using a variable which is screaming about password reset (PASSWORD_RESET_TIMEOUT), we would like to use smth called ACTIVATION_TOKEN_TIMEOUT
So there is a solution: why not create a timeout property for PasswordResetTokenGenerator class? Almost in the same way as it was done with _secret and algorithm fields.So our development team come up with an idea to create a PR which will add this functionality to the Django project. But before this we decided to search similar solutions in django PRs. And we found them! Ticket 30423 https://code.djangoproject.com/ticket/30423 sounds good enough, but it was closed with wontfix label.So the question is: why not to add this fine feature to the PasswordResetTokenGenerator ? And if people find this useful - why not to merge one o the existing PRs?

--

[Roger Gammans]

Hi

If it's intended as a reference implementation, then I would expect PasswordResetTokenGenerator  to use Signer or TimestampSigner internally , but I was surprised to discover that it didn't use either.

Isn't those entry points the better API to advise for use rather than the direct use of hmac based algorithms?

[Alexander Voloshchenko]

Hi Carlton!

I can understand what you're saying, and I don't think the TimestampSigner solution is bad

But again, it's a little unclear why in the end, in line 132 of https://github.com/django/django/blob/7e6b537f5b92be152779fc492bb908d27fe7c52a/django/contrib/auth/tokens.py#L132 we assign an object of class PasswordResetTokenGenerator to a variable named default_token_generator, so should it be the default way to create tokens? why then default_token_generator should generate tokens with a timeout limit depending on the variable PASSWORD_RESET_TIMEOUT?

and if somehow this class was meant to be a way to create tokens, why not put timeout as separate property?

[Carlton Gibson]

Yes, likely. 

My point was (merely) that we **don't** want to add additional extension points to PasswordResetTokenGenerator.

--

You received this message because you are subscribed to the Google Groups "Django developers (Contributions to Django itself)" group.
To unsubscribe from this group and stop receiving emails from it, send an email to django-develop...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/django-developers/6458aad7056857164d976b1c7ce46bcd0249d292.camel%40gammascience.co.uk.

[Robert Musił]

Hey Carlton,

I'm reading through the replies and I feel like I'm missing point completely.

First some background - `PasswordResetTokenGenerator` is heavily referenced over the web (or more accurately - the `default_token_genrator`) when it comes to account activation. I personally believe the solution is pretty genius for this particular use case, simplifies the flow without relying on the persistence layer. The implementation is also well-contained within the class so makes using it super simple.

When I use `default_token_generator` I'm however limited by two things - the fields that make up the hash (user.pk, user.password, login_timestamp and current timestamp) - which I can EASILY overwrite with `_make_hash_value` and `PASSWORD_RESET_TIMEOUT_DAYS` which is NOT easily overwritten, due to the method `check_token` having too many responsibilities. 

So the use case is definitely there, but the argument here is "we don't want to", but the reasoning behind that is unclear - is it not worth to maintain this piece of code because it's deprecated? This would be flagged then, no? Or do you feel like it's already a concrete implementation given its name ("PasswordResetTokenGenerator") - would the solution then be abstracting the parts of it to a base class, would that be acceptable?

[Carlton Gibson]

Hey Robert. 

> ... the argument here is "we don't want to", but the reasoning behind that is unclear ... 

For me, the reasoning would be that we get several security reports against Django pretty much every week, and the extent to which we make security sensitive areas **generic** for the sake of only a subset of users increases the attack surface, which makes these harder to contain and deal with. (**Every** hook adds a maintenance cost, no matter how small. The question is always whether it's worth it.) 

The diff on django/contrib/auth/tokens.py is small: https://github.com/django/django/pull/11302/files

I'm not sure we want to make PasswordResetTokenGenerator usable as a generic token generator. (Use the signing tools to create your own no?) 

Perhaps others will say that yes this is worth it (and then we come to a consensus, or not) — but my original reply was attempting to expand the reason for the original decision to turn it down.

C.

To view this discussion on the web visit https://groups.google.com/d/msgid/django-developers/3f42f83a-2090-4627-bbf1-1dd86df34913n%40googlegroups.com.