Using Rack HEAD causes CookieStore security warnings

[dlee]

Rack::Session::Cookie accepts a :secret option in its initializer. The initializer stores that secret in an instance variable and uses it in the #set_session method to HMAC sign the data.

Rack HEAD added security warnings for when the Rack::Session::Cookie middleware is initialized without a secret.

This is causing the warning to show up because of the way Rails uses Rack::Session::Cookie:

Rails uses a subclass of Rack::Session::Cookie called ActionDispatch::Session::CookieStore that overwrites #set_session and #set_cookie to do its own signed cookie implementation using ActionDispatch::Cookies::SignedCookieJar (using Rack env "action_dispatch.secret_token" instead of the @secret instance variable). Since Rails doesn't need the @secret instance variable, it initializes ActionDispatch::Session::CookieStore without providing a :secret option.

To avoid the warning, we should just not use Rack::Session::Cookie. We're overriding a lot of the functionality for Rack::Session::Cookie anyway, to the point that its most important functionality (cookie signing) is not being used at all.

[steveklabnik]

Would you be interested in working up a pull request to fix this? You seem to have a pretty good handle on the problem.

[dlee]

@steveklabnik Is not using Rack::Session::Cookie the correct direction?

There's the other option of using Rack::Session::Cookie properly, at the cost of losing the ability to use ActionDispatch::Cookies::SignedCookieJar for the signing.

[steveklabnik]

I am not 100% sure what the right thing is to to, to be honest. I'm still wrapping my head around session code in Rails.

[dlee]

Can we get someone in Rails core who's familiar with Rack integration to chime in? What's our philosophy behind integration with Rack? Do we try to use as many standard Rack middleware as possible, or is that not a priority?

[steveklabnik]

That kind of question is better served by a post to the rails-core list. The chances of a reply in Issues is pretty low.

[senny]

@dlee I also suggest you start a discussion on the mailing list. The bug tracker is primarily used to track bugs and pull requests with attached code (for new features). For discussions the mailing list is better suited.

@steveklabnik should we close this ticket in favor of a post on the core-list?

[dlee]

@senny I can see how the discussion about "our philosophy behind integration with Rack" is better suited for a mailing list, but I still consider the main issue here to be a bug.

[homakov]

I would love to join. Rack stuff is tiny and can be rewritten
I actually was about to write encrypted session jar. Not signed, just encrypted with secret so you could not read it. WDYT about it?

[steveklabnik]

@tenderlove and @NZKoz , what is the behavior that's intended here?

[NZKoz]

yeah, we need to avoid using the rack session subclass as this warning isn't relevant for us. It doesn't seem to give us much behaviour that we don't then turn around and re-implement.

A pull request for that would be a good start.

@homakov we intend to ship an encrypted cookie store with 4.0, and also start using the key derivation code rather than the raw secret, perhaps we can look into that once we've removed the rack subclassing?

[homakov]

@NZKoz good idea! anyone is on it or is it just future plan? I see more pros of 'encrypted' than in 'signed'. It will allow to develop database-less(stateless) applications - developer is sure that data from session is not readable/writable, each user is just carrying his own DB.
One thing - performance. Doesn't it look slower to decrypt every session everytime.. gotta check

[heartpunk]

Looks like the discussion here (that I wasn't yet aware of) makes #8792 not the appropriate fix for #8789. If I am up to the task of refactoring to what it seems the consensus here is (moving away from subclassing rack's session stuff), would that be a desirable pull request?

[guilleiguaran]

@tehgeekmeister yes, a pull request for this would be great

[Sega100500]

Rails bug?

rack/rack#485 (comment)

[NZKoz]

Yes, we're subclassing where we don't need to. The rack guys have (quite reasonably) added a warning and our code is triggering it. You can safely disregard it.

As mentioned by @steveklabnik and @tehgeekmeister the preferred fix should be to stop subclassing inappropriately, and we'd like a pull request for that.

[heartpunk]

Just got my dev environment set up and tests runnable. Getting started on the changes now!

[heartpunk]

It looks like we're going to have to clone a lot of functionality from rack to do this. Seems like it might be better to maintain the subclassing. I'll happily do the work if everyone disagrees, but as the person doing it, it's starting to feel like a bad idea.

For now, I'll just keep going and get a pull request up so there's something explicit for people to look at, and see what I mean.

[colszowka]

Maybe it would be nice at this point to team up with the rack team to figure out how it'd be possible to move all this custom functionality you're mentioning into rack itself and have Rails have nothing to do with the details of session and cookie implementation? Is there functionality in the whole rails cookie implementation that would not benefit rack apps in general?

[heartpunk]

There's some that I suspect other frameworks don't want to have. Some of it is around making autoloading work nicely (

rails/actionpack/lib/action_dispatch/middleware/session/abstract_store.rb

Line 58 in cb3181e

$1.constantize

for example). Other stuff seems to be duplication, for example, the whole thing that's caused this is that rack now can handle cookie signing, which I assume it couldn't earlier, since rails has an implementation of that too.

I'm in favor of discussing with rack people as well. Seems like a good course of action. I'll dig into this more tomorrow and build up a list of duplications/points of tight coupling, since that will be useful whichever way we decide to go.

[NZKoz]

The true goal is to avoid that error to be honest, it's caused because our session implementation just delegates to cookies.signed or cookies.encrypted depending on how it's configured.

Here's our set_session:

https://github.com/rails/rails/blob/master/actionpack/lib/action_dispatch/middleware/session/cookie_store.rb#L70-L72

Here's rack's

https://github.com/rack/rack/blob/master/lib/rack/session/cookie.rb#L131-L145

The main issue here's is we're calling super's initialize which is triggering an error even though we don't use secret.

If changing the messy subclassing isn't an option, then simply override initialize and pass the secret correctly, though the value will never be used, it'll avoid the warnings.

[mjtko]

While @spastorino committed a workaround in cb3181e to suppress the message for now, I think the better fix is to subclass from Rack::Session::Abstract::ID instead of Rack::Session::Cookie, as we are providing our own cookie implementation in rails.

Here's my approach to do that: https://github.com/mjtko/rails/compare/fix;cookie-store-inheritance

If somebody on the core team could comment and shout if they think this looks reasonable, I'll open it as a PR.

/cc @tenderlove and @NZKoz

[tengla]

Quick and dirty solution, if seeing this error message irritates you.

Take a look at the error message, where the whining file is referenced,

Go to initializer under module Compatibility.

module Compatibility
  def initialize(app, options = {})
    options[:key] ||= '_session_id'
    # add secret if not supplied by options.
    options[:secret] ||= SecureRandom.hex(16)
    super
  end
  ...

edit:
This is also what @spastorino suggests in cb3181e.

[heartpunk]

I'm in support of @athlite's suggestion, myself. It's close to the PR I had originally submitted, though I think a little cleaner.

[guilleiguaran]

@athlite @mjtko talked yesterday with @spastorino. the quick workaround is only for 3.1.x and 3.2.x, in master we shouldn't use it.

[mjtko]

@guilleiguaran Indeed! Hence my reversion of that commit and an alternative solution posed in my branch. 😄

[heartpunk]

Alright, I'm 👍 for @mjtko's approach, then. The approach I was working on is much uglier in comparison.

[mjtko]

Right, I'm going to raise my branch as a PR for further discussion.

[guilleiguaran]

@mjtko open PR please :)

[mjtko]

As above, PR now open as #8824. 😁

[guilleiguaran]

Fixed with the PR

[ndbroadbent]

Will this fix also be released for 3.2.x, or should we use gem 'rack', '1.4.1' to get rid of the warning?

[guilleiguaran]

@ndbroadbent the warning should disappear in the next maintenance release of 3.2.x

[sozgul]

I am having some issues with Rack, just bounced through 3-4 issues and landed here but god its so fucking hard to actually understand all the content here I am grateful more capable people are working on it. Since I cant really contribute much understanding myself I decided to thank you guys instead.

thank you

[spastorino]

@sozgul thanks for your words. I'm interested in knowing what issues are you having so we can help you

[steveklabnik]

@sozgul agreed, what @spastorino said. Please open a new issue and we can help you with your problem directly.

[sozgul]

@spastorino, @steveklabnik, I really appreciate how fast you guys are to offer help. Just to clear the air, turns out I was the problem, Rack was having a problem with me not the other way around. My profile would make it obvious that I am not half as seasoned as a developer compared to you guy, but since you guys are so helpful I will contact you if I run into a real problem about rails. I do not have that much of a web presence yet, that will hopefully change in the future, if you guys also think I can help you with anything, please reach out.

[steveklabnik]

Great. :)

[mhenrixon]

I am still having this incredibly annoying message that are destroying my spec output. I am on rails 3.2.13 and I thought this should have been back ported am I wrong? I can't use rack 1.4.1 since rails requires 1.4.5. Suggestions?

[drnic]

Also getting this issue with latest 3.2.12 rails

[spastorino]

@drnic really? I thought it was fixed. Same happens in 3.2.13? @guilleiguaran do you remember when was this exactly fixed?

[drnic]

@spastorino I'm seeing this:

=> Booting Puma
=> Rails 3.2.13 application starting in development on http://0.0.0.0:3000
=> Call with -d to detach
=> Ctrl-C to shutdown server
Using github credentials for localhost
        SECURITY WARNING: No secret option provided to Rack::Session::Cookie.
        This poses a security threat. It is strongly recommended that you
        provide a secret to prevent exploits that may be possible from crafted
        cookies. This will not be supported in future versions of Rack, and
        future versions will even invalidate your existing user cookies.

        Called from: /Users/drnic/.rvm/gems/ruby-1.9.3-p392/gems/actionpack-3.2.13/lib/action_dispatch/middleware/stack.rb:43:in `new'.

[mjtko]

@spastorino - I seem to recall that you fixed the warning for 3.2.13 with this commit: d9a0480

[spastorino]

@drnic what middlewares are you using?

[grepit]

same problem here