summaryrefslogtreecommitdiff
path: root/doc/development/permissions.md
blob: 3abadc98501fbec2627437b7549866e1bb80fcb6 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
---
stage: Manage
group: Authentication and Authorization
info: To determine the technical writer assigned to the Stage/Group associated with this page, see https://about.gitlab.com/handbook/product/ux/technical-writing/#assignments
---

# Permission development guidelines

There are multiple types of permissions across GitLab, and when implementing
anything that deals with permissions, all of them should be considered.

## Instance

### User types

Each user can be one of the following types:

- Regular.
- External - access to groups and projects only if direct member.
- [Internal users](internal_users.md) - system created.
- [Auditor](https://gitlab.com/gitlab-org/gitlab/-/blob/master/ee/app/policies/ee/base_policy.rb#L9):
  - No access to projects or groups settings menu.
  - No access to Admin Area.
  - Read-only access to everything else.
- [Administrator](https://gitlab.com/gitlab-org/gitlab/-/blob/master/app/policies/base_policy.rb#L6) - read-write access.

See the [permissions page](../user/permissions.md) for details on how each user type is used.

## Groups and Projects

### General permissions

Groups and projects can have the following visibility levels:

- public (`20`) - an entity is visible to everyone
- internal (`10`) - an entity is visible to authenticated users
- private (`0`) - an entity is visible only to the approved members of the entity

By default, subgroups can **not** have higher visibility levels.
For example, if you create a new private group, it cannot include a public subgroup.

The visibility level of a group can be changed only if all subgroups and
sub-projects have the same or lower visibility level. For example, a group can be set
to internal only if all subgroups and projects are internal or private.

WARNING:
If you migrate an existing group to a lower visibility level, that action does not migrate subgroups
in the same way. This is a [known issue](https://gitlab.com/gitlab-org/gitlab/-/issues/22406).

Visibility levels can be found in the `Gitlab::VisibilityLevel` module.

### Feature specific permissions

Additionally, the following project features can have different visibility levels:

- Issues
- Repository
  - Merge request
  - Forks
  - Pipelines
- Analytics
- Requirements
- Security and Compliance
- Wiki
- Snippets
- Pages
- Operations
- Metrics Dashboard

These features can be set to "Everyone with Access" or "Only Project Members".
They make sense only for public or internal projects because private projects
can be accessed only by project members by default.

### Members

Users can be members of multiple groups and projects. The following access
levels are available (defined in the
[`Gitlab::Access`](https://gitlab.com/gitlab-org/gitlab/-/blob/master/lib/gitlab/access.rb)
module):

- No access (`0`)
- [Minimal access](../user/permissions.md#users-with-minimal-access) (`5`)
- Guest (`10`)
- Reporter (`20`)
- Developer (`30`)
- Maintainer (`40`)
- Owner (`50`)

If a user is the member of both a project and the project parent groups, the
highest permission is the applied access level for the project.

If a user is the member of a project, but not the parent groups, they
can still view the groups and their entities (like epics).

Project membership (where the group membership is already taken into account)
is stored in the `project_authorizations` table.

NOTE:
In [GitLab 14.9](https://gitlab.com/gitlab-org/gitlab/-/issues/351211) and later, projects in personal namespaces have a maximum role of Owner.
Because of a [known issue](https://gitlab.com/gitlab-org/gitlab/-/issues/219299) in GitLab 14.8 and earlier, projects in personal namespaces have a maximum role of Maintainer.

### Confidential issues

[Confidential issues](../user/project/issues/confidential_issues.md) can be accessed
only by project members who are at least
reporters (they can't be accessed by guests). Additionally they can be accessed
by their authors and assignees.

### Licensed features

Some features can be accessed only if the user has the correct license plan.

## Permission dependencies

Feature policies can be quite complex and consist of multiple rules.
Quite often, one permission can be based on another.

Designing good permissions means reusing existing permissions as much as possible
and making access to features granular.

In the case of a complex resource, it should be broken into smaller pieces of information
and each piece should be granted a different permission.

A good example in this case is the _Merge Request widget_ and the _Security reports_.
Depending on the visibility level of the _Pipelines_, the _Security reports_ are either visible
in the widget or not. So, the _Merge Request widget_, the _Pipelines_, and the _Security reports_,
have separate permissions. Moreover, the permissions for the _Merge Request widget_
and the _Pipelines_ are dependencies of the _Security reports_.

### Permission dependencies of Secure features

Secure features have complex permissions since these features are integrated
into different features like Merge Requests and CI flow.

 Here is a list of some permission dependencies.

| Activity level | Resource | Locations |Permission dependency|
|----------------|----------|-----------|-----|
| View | License information | Dependency list, License Compliance | Can view repository |
| View | Dependency information | Dependency list, License Compliance | Can view repository |
| View | Vulnerabilities information | Dependency list | Can view security findings |
| View | Black/Whitelisted licenses for the project | License Compliance, merge request  | Can view repository |
| View | Security findings | merge request, CI job page, Pipeline security tab | Can read the project and CI jobs |
| View | Vulnerability feedback | merge request | Can read security findings |
| View | Dependency List page | Project | Can access Dependency information |
| View | License Compliance page | Project | Can access License information|

## Where should permissions be checked?

We should typically apply defense-in-depth (implementing multiple checks at
various layers) starting with low-level layers, such as finders and services,
followed by high-level layers, such as GraphQL, public REST API, and controllers.

See [Guidelines for reusing abstractions](reusing_abstractions.md).

Protecting the same resources at many points means that if one layer of defense is compromised
or missing, customer data is still protected by the additional layers.

See the permissions section in the [Secure Coding Guidelines](secure_coding_guidelines.md#permissions).

### Considerations

Services or finders are appropriate locations because:

- Multiple endpoints share services or finders so downstream logic is more likely to be re-used.
- Sometimes authorization logic must be incorporated in DB queries to filter records.
- Permission checks at the display layer should be avoided except to provide better UX
  and not as a security check. For example, showing and hiding non-data elements like buttons.

The downsides to defense-in-depth are:

- `DeclarativePolicy` rules are relatively performant, but conditions may perform database calls.
- Higher maintenance costs.

### Exceptions

Developers can choose to do authorization in only a single area after weighing
the risks and drawbacks for their specific case.

Prefer domain logic (services or finders) as the source of truth when making exceptions.

Logic, like backend worker logic, might not need authorization based on the current user.
If the service or finder's constructor does not expect `current_user`, then it typically won't
check permissions.

### Tips

If a class accepts `current_user`, then it may be responsible for authorization.

### Example: Adding a new API endpoint

By default, we authorize at the endpoint. Checking an existing ability may make sense; if not, then we probably need to add one.

As an aside, most endpoints can be cleanly categorized as a CRUD (create, read, update, destroy) action on a resource. The services and abilities follow suit, which is why many are named like `Projects::CreateService` or `:read_project`.

Say, for example, we extract the whole endpoint into a service. The `can?` check will now be in the service. Say the service reuses an existing finder, which we are modifying for our purposes. Should we make the finder check an ability?

- If the finder doesn't accept `current_user`, and therefore doesn't check permissions, then probably no.
- If the finder accepts `current_user`, and doesn't check permissions, then it would be a good idea to double check other usages of the finder, and we might consider adding authorization.
- If the finder accepts `current_user`, and already checks permissions, then either we need to add our case, or the existing checks are appropriate.

### Refactoring permissions

#### Finding existing permissions checks

As mentioned [above](#where-should-permissions-be-checked), permissions are
often checked in multiple locations for a single endpoint or web request. As a
result, finding the list of authorization checks that are run for a given endpoint
can be challenging.

To assist with this, you can locally set `GITLAB_DEBUG_POLICIES=true`.

This outputs information about which abilities are checked in the requests
made in any specs that you run. The output also includes the line of code where the
authorization check was made. Caller information is especially helpful in cases
where there is metaprogramming used because those cases are difficult to find by
grepping for ability name strings.

Example:

```shell
# example spec run

GITLAB_DEBUG_POLICIES=true bundle exec rspec spec/controllers/groups_controller_spec.rb:162

# permissions debug output when spec is run; if multiple policy checks are run they will all be in the debug output.

POLICY CHECK DEBUG -> policy: GlobalPolicy, ability: create_group, called_from: ["/gitlab/app/controllers/application_controller.rb:245:in `can?'", "/gitlab/app/controllers/groups_controller.rb:255:in `authorize_create_group!'"]
```

This flag is meant to help learn more about authorization checks while
refactoring and should not remain enabled for any specs on the default branch.

#### Understanding logic for individual abilities

References to an ability may appear in a `DeclarativePolicy` class many times
and depend on conditions and rules which reference other abilities. As a result,
it can be challenging to know exactly which conditions apply to a particular
ability.

`DeclarativePolicy` provides a `ability_map` for each Policy class, which
pulls all Rules for an ability into an array.

Example:

```ruby
> GroupPolicy.ability_map.map.select { |k,v| k == :read_group_member }
=> {:read_group_member=>[[:enable, #<Rule can?(:read_group)>], [:prevent, #<Rule ~can_read_group_member>]]}

> GroupPolicy.ability_map.map.select { |k,v| k == :read_group }
=> {:read_group=>
  [[:enable, #<Rule public_group>],
   [:enable, #<Rule logged_in_viewable>],
   [:enable, #<Rule guest>],
   [:enable, #<Rule admin>],
   [:enable, #<Rule has_projects>],
   [:enable, #<Rule read_package_registry_deploy_token>],
   [:enable, #<Rule write_package_registry_deploy_token>],
   [:prevent, #<Rule all?(~public_group, ~admin, user_banned_from_group)>],
   [:enable, #<Rule auditor>],
   [:prevent, #<Rule needs_new_sso_session>],
   [:prevent, #<Rule all?(ip_enforcement_prevents_access, ~owner, ~auditor)>]]}
```

`DeclarativePolicy` also provides a `debug` method that can be used to
understand the logic tree for a specific object and actor. The output is similar
to the list of rules from `ability_map`. But, `DeclarativePolicy` stops
evaluating rules once one `prevent`s an ability, so it is possible that
not all conditions are called.

Example:

```ruby
policy = GroupPolicy.new(User.last,  Group.last)
policy.debug(:read_group)

- [0] enable when public_group ((@custom_guest_user1 : Group/139))
- [0] enable when logged_in_viewable ((@custom_guest_user1 : Group/139))
- [0] enable when admin ((@custom_guest_user1 : Group/139))
- [0] enable when auditor ((@custom_guest_user1 : Group/139))
- [14] prevent when all?(~public_group, ~admin, user_banned_from_group) ((@custom_guest_user1 : Group/139))
- [14] prevent when needs_new_sso_session ((@custom_guest_user1 : Group/139))
- [16] enable when guest ((@custom_guest_user1 : Group/139))
- [16] enable when has_projects ((@custom_guest_user1 : Group/139))
- [16] enable when read_package_registry_deploy_token ((@custom_guest_user1 : Group/139))
- [16] enable when write_package_registry_deploy_token ((@custom_guest_user1 : Group/139))
  [21] prevent when all?(ip_enforcement_prevents_access, ~owner, ~auditor) ((@custom_guest_user1 : Group/139))

=> #<DeclarativePolicy::Runner::State:0x000000015c665050
 @called_conditions=
  #<Set: {
   "/dp/condition/GroupPolicy/public_group/Group:139",
   "/dp/condition/GroupPolicy/logged_in_viewable/User:83,Group:139",
   "/dp/condition/BasePolicy/admin/User:83",
   "/dp/condition/BasePolicy/auditor/User:83",
   "/dp/condition/GroupPolicy/user_banned_from_group/User:83,Group:139",
   "/dp/condition/GroupPolicy/needs_new_sso_session/User:83,Group:139",
   "/dp/condition/GroupPolicy/guest/User:83,Group:139",
   "/dp/condition/GroupPolicy/has_projects/User:83,Group:139",
   "/dp/condition/GroupPolicy/read_package_registry_deploy_token/User:83,Group:139",
   "/dp/condition/GroupPolicy/write_package_registry_deploy_token/User:83,Group:139"}>,
 @enabled=false,
 @prevented=true>
```

#### Testing that individual policies are equivalent

You can use the `'equivalent project policy abilities'` shared example to ensure
that 2 project policy abilities are equivalent for all project visibility levels
and access levels.

Example:

```ruby
  context 'when refactoring read_pipeline_schedule and read_pipeline' do
    let(:old_policy) { :read_pipeline_schedule }
    let(:new_policy) { :read_pipeline }

    it_behaves_like 'equivalent policies'
  end
```