Project

General

Profile

Actions

Patch #16685

closed

Introduce the request_store gem to hold User.current and prevent data leakage in error messages

Added by Holger Just almost 10 years ago. Updated over 9 years ago.

Status:
Closed
Priority:
Normal
Assignee:
-
Category:
Accounts / authentication
Target version:
Start date:
Due date:
% Done:

100%

Estimated time:

Description

Recently, there were several issues where the User.current was not properly initialized and thus the setting from the request before it was used, e.g. #16511 with r13041 which can lead to data disclosure. This issue and all similar ones can be fully circumvented by ensuring that the current user is reset before the request even reaches our rails code.

This patch which was extracted from Planio proposes the introduction of the request_store gem. It adds a middleware to provide a true request-local data store based on Thread.current. This ensures that the current user always needs to be set explicitly and can't be taken from previous requests, even if the logic to setup the user somehow fails or is circumvented due to early error handlers. The fail-safe default is always User.anonymous which should be sufficient even in the face of an auth error.

This patch is against the current master branch on Github.


Files


Related issues

Related to Redmine - Feature #31911: Update request_store gem to 1.4ClosedJean-Philippe Lang

Actions
Actions #1

Updated by Toshi MARUYAMA almost 10 years ago

  • Target version set to 2.4.6
Actions #2

Updated by Toshi MARUYAMA almost 10 years ago

  • Description updated (diff)
Actions #3

Updated by Toshi MARUYAMA almost 10 years ago

Test fails on drone.io.

https://drone.io/github.com/marutosi/redmine/211
https://drone.io/github.com/marutosi/redmine/212

  1) Failure:
test_api_should_trigger_basic_http_auth_with_basic_authorization_header(Redmine::ApiTest::AuthenticationTest) 
[/xxxxxxxx/redmine/test/integration/api_test/authentication_test.rb:34]:
Expected response to be a <401>, but was <500>

Tests on CI Server pass and I cannot reproduce on my CentOS 6.5.
http://www.redmine.org/builds/index.html

Actions #4

Updated by Toshi MARUYAMA almost 10 years ago

On Ubuntu 12.04.4 LTS,
ruby 1.9.3p545 (2014-02-24 revision 45159) [x86_64-linux]

  1) Failure:
test_api_should_trigger_basic_http_auth_with_basic_authorization_header(Redmine::ApiTest::AuthenticationTest)
[test/integration/api_test/authentication_test.rb:34]:
Expected response to be a <401>, but was <500>
Mocha::ExpectationError (unexpected invocation: #<AnyInstance:UsersController>.authenticate_with_http_basic()
unsatisfied expectations:
- expected exactly once, not yet invoked: #<AnyInstance:ApplicationController>.authenticate_with_http_basic(any_parameters)
):
  app/controllers/application_controller.rb:125:in `find_current_user'
  app/controllers/application_controller.rb:100:in `user_setup'
  test/integration/api_test/authentication_test.rb:33:in `test_api_should_trigger_basic_http_auth_with_basic_authorization_header'
Actions #5

Updated by Etienne Massip almost 10 years ago

Is it really useful to add a new runtime dependency? Wouldn't a complete test be enough instead? And if jot, could that change be targeted to next major release?

Actions #6

Updated by Toshi MARUYAMA almost 10 years ago

Etienne Massip wrote:

Is it really useful to add a new runtime dependency?

mocha is in "test" group.

Wouldn't a complete test be enough instead?

It depends web servers, so we cannot add new tests.
https://github.com/steveklabnik/request_store#the-problem

And if not, could that change be targeted to next major release?

This issue is security fix depends on #16511 with r13041,
so I think it should be targeted to minor release.

Actions #7

Updated by Etienne Massip almost 10 years ago

I meant request store, not mocha. This should be handled by RoR, not by a specific dependency.

Actions #8

Updated by Toshi MARUYAMA almost 10 years ago

We usually upgrade Rails version for security fix in minor release.
And r13110 does not change application behaviors.

Actions #9

Updated by Etienne Massip almost 10 years ago

Sure but this is not a security fix, only a potential fix (which from my very personal pov may be a bit of overkill since it seems that all possibilities of having a wrong user in thread are now prevented).

Actions #10

Updated by Toshi MARUYAMA almost 10 years ago

  • Target version changed from 2.4.6 to 2.6.0
  • % Done changed from 0 to 100

OK. I have reverted 2.4-stable and 2.5-stable revisions.

Actions #11

Updated by Etienne Massip almost 10 years ago

Thanks, it was just my personal opinion; it may be safer than adding a new gem which might itself contain some security issue we're not aware of!

Actions #12

Updated by Holger Just almost 10 years ago

Etienne: It's way too easy to introduce data leaks when dealing with before filters, either in the core or in Plugins, so this adds an additional safety net.

Also, the gem is actually about 10 lines of code, so not that much of a dependency burden. You could add it into the core if you would be willing to support this code (which I think is pointless here). If you read the code on github, you'll notice that there isn't much room for bugs. Also, the library is well tested.

In my opinion, it adds a rather easy-to-use safty net under some easy to break code paths.

Actions #13

Updated by Toshi MARUYAMA almost 10 years ago

request_store 1.0.6 breaks many tests.
http://www.redmine.org/builds/logs/build_trunk_postgresql_ruby-2.0.0-p0_1969.html

 1) Failure: test_api_should_accept_switch_user_header_for_admin_user(Redmine::ApiTest::AuthenticationTest) 
...
Actions #14

Updated by Holger Just almost 10 years ago

This is probably caused by this commit. In 1.0.6, the thread store is cleared after a request has finished (and not before a request started as before). That means, that after the middleware was passed on the way up again, User.current is not valid anymore.

The solution for this would be to fix the integration tests to not rely on an "old" User.current. In fact, not having this old value still set after a request is a good thing and solves many potential leakage issues.

Actions #15

Updated by Jean-Philippe Lang over 9 years ago

  • Status changed from New to Closed
Actions #16

Updated by Go MAEDA over 4 years ago

Actions

Also available in: Atom PDF