Skip to content

Latest commit

 

History

History
 
 

authz-service

AuthZ Service

The AuthZ Service (short for "Authorization") provides the API to determine if specific Automate actions are allowed on particular resources based on a user's team membership.

Contents of this file:

AuthZ Approach

Terminology

Authentication (or AuthN)

Authentication establishes a user's identity. AuthN is not covered in this document; see Authentication.

Authorization (or AuthZ)

Authorization defines an authenticated user's actions within the system. This document is the comprehensive resource for authorization within Chef Automate.

OPA

AuthZ Service uses the Open Policy Agent (OPA) library for authorization decisions. OPA uses general policies written in the declarative language Rego for decisions on individual requests.

Policy

Policies specify the rights to access and act on resources within Chef Automate. A single policy defines the rights of a set of users and/or teams of users to access and act on a set of resources. Policies and team membership are defined by an admin user during Chef Automate setup. A policy consists of a (Subjects, Action, Resource) tuple.

Query (or AuthZ Query)

Queries are used to determine that a user has the correct authorization to proceed with an operation. A query is an auto-generated message (IsAuthorizedReq) sent from the gateway to the Authz service to check user authorizations against the policy store. A query consists of a (Subjects, Action, Resource) tuple.

Note that the tuple for a Policy is different than a tuple for a query only in that the former may contain wildcards, as explained in detail later. More concretely, a policy states "who can do what to which resources", while a query asks "can this person do this to that resource?"

(While a query can just as easily be talked about as a "request", we are using "query" in this document to specifically mean the message from the gateway to AuthZ, and "request" to mean the HTTP request coming in to the gateway.)

Request (or HTTP Request)

This is a message sent from the Automate front-end to the gateway/back-end that contains an HTTP endpoint, an HTTP method, and possibly parameters. Parameters may be in the endpoint URL itself or in the body of the request, depending on the HTTP method. Example:

HTTP endpoint: /auth/users
HTTP method: GET or PUT
Parameter: userID=123

AuthZ with OPA

Every HTTP request to the gateway (from among all those defined in proto files in the gateway project) is, after being authenticated, sent to the AuthZ service to be authorized. Authorization determines if that particular operation is permitted for the current user. (Note that there is no authorization enforced at the individual service level, only at the gateway.) The AuthZ Service makes that determination with the OPA engine, which returns whether or not the request is granted based on its knowledge of Automate policies. These policies are all user-created, with the exception of the set of default policies that are included out of the box. Depending on your needs, you can delete some of the default policies, which is necessary to have tighter permissions. Note that the admin default policies, however, are non-deletable to prevent locking yourself out of the system completely!

The figure shows the sequence of steps involved for each HTTP request received by the gateway [1]. The happy path through the system is:

a. A request enters the gateway. (Not shown, the gateway authenticates the user sending the request.)

b. The gateway creates a query and sends it to the AuthZ service to check for authorization [2]. The query contains the subjects, action, and resource.

c. The AuthZ service uses the OPA engine [3] to attempt to match the inputs against a policy.

d. If at least one such policy is found to match, it grants access to the given resource and action [5].

e. The AuthZ service can then send a response to the gateway [6] affirming this, so the gateway then allows the original request to proceed [7] and [8].

If a policy was not matched against the inputs, the response from AuthZ to the query is deny[4] so the gateway aborts the original HTTP request and returns a 403 HTTP status.

Automate 2 Authorization

FIGURE 1

Warning It is important to note that in order to achieve a secure system, the default behavior is to deny a request. This could occur if:

  1. The API endpoint is annotated with resource and action, but there is no policy in place to allow that combination of resource and action [4].
  2. The API endpoint is not annotated with resource and action details [10]. This case should never happen in practice.

Elements of the AuthZ Query

The gateway sends a query to the AuthZ service (the IsAuthorizedReq) to determine authorization. The AuthZ query has three required elements: Subjects, Action, and Resource. As an example:

Subjects: [ user:ldap:12345, team:ldap:audit, team:ldap:proj ]
Action: read
Resource: compliance:nodes:*

Query Subjects

A Subject is a user, client, or team that is requesting permission to perform some operation.

For an HTTP request initiated by a chef server or chef client, the Subjects array will have just a single element with the ID of the token that was used, in the form token:SOME_ID.

For an HTTP request initiated by a browser, this is an array containing the user ID of the requestor plus the IDs of any teams that user belongs to. Teams may be locally maintained (within Automate) or may be those managed by external identity management systems (LDAP or SAML). The Subjects array will contain one element of the form user:PROVIDER:SOME_ID and zero or more elements of the form team:PROVIDER:SOME_ID.

Query Action

An Action is a verb being attempted by the requestor. This could be a CRUD verb (e.g. "create") or a more specific action name. Unlike the Subjects, which is inferred from the user, the Action is inferred from the HTTP endpoint in the HTTP request that the user has initiated.

Specifically, here is where it comes from:

Each public API method is specified by a definition in a proto file. Included in that specification is the action associated with the endpoint. Below the HTTP endpoint is /auth/teams, the HTTP method is GET, and the Action is read.

  rpc GetTeams (teams.request.GetTeamsReq) returns (teams.response.Teams) {
    option (google.api.http).get = "/auth/teams";
    option (chef.automate.api.policy).resource = "auth:teams";
    option (chef.automate.api.policy).action = "read";
  };

As an example, in this query...

subjects: [user:local:[email protected]]
action: read
resource: cfgmgmt:nodes

... the Action reveals what the user wants permission to do with the Resource "cfgmgmt:nodes".

Query Resource

A Resource describes the data that the requestor is attempting to act upon. This could be, for example, all Compliance profiles or a particular Infrastructure Automation node.

Specifically, here is where it comes from:

Just like the Action, the Resource comes from the endpoint definition in the proto file. In the example above, the Resource is auth:teams. Unlike the other elements, a Resource may be concrete (e.g. /auth/teams in the previous example) or abstract (e.g. /auth/users/{email}) as, for example:

   rpc GetUser (users.request.Email) returns (users.response.User) {
    option (google.api.http).get = "/auth/users/{email}";
    option (chef.automate.api.policy).resource = "auth:users:{email}";
    option (chef.automate.api.policy).action = "read";
  };

When an actor (browser or chef client/server) sends an HTTP request, it will, of course, provide a concrete path, (e.g. /auth/users/[email protected]). The gateway matches that to either a concrete or abstract endpoint. If it is an abstract one, as in this example, then the parameters recognized in the match are applied to the resource, turning that abstract resource (auth:users:{email}) into a concrete one (auth:users:[email protected]), which can then be evaluated for permissions.

Elements of the AuthZ Policy

AuthZ queries are comprised of a (Subjects, Action, Resource) tuple, as just described. Policy creation requests are superficially the same -- a (Subjects, Action, Resource) tuple -- but the values of each of those differs between policies and queries, as does the origination.

That is, an AuthZ query is generated internally by the gateway as actors interact with Automate, for any action that goes to the back end. Creating a policy, on the other hand, is a specific action performed by a system administrator, who explicitly provides a (Subjects, Action, Resource) tuple.

Policy Subjects

The Subjects array contains one or more user IDs, team IDs, and/or token IDs that the policy applies to. As on the query side, an individual Subject can be a user, team, or non-human actor. It can also include wildcards as described below.

Policy Action

An Action is either a specific verb or a wildcard * that applies the policy to any verb. A specific verb is typically a CRUD verb (e.g. "create") but could be a more specific action name.

Policy Resource

A Resource is the data the policy is protecting. It can also include wildcards as described below.

Mapping Requests to Endpoints to Policies

The (Subjects, Action, Resource) tuple of an AuthZ query is evaluated by the AuthZ service as shown earlier (steps [3] to [5] in Figure 1 above). These elements correspond to eponymous elements in Automate policies. Thus, each Automate policy has Subjects, Action, and Resource. The set of all policies -- specified by an admin user during Chef Automate setup -- collectively define what is allowed or denied for every Automate user. The (Subjects, Action, Resource) tuple of an individual query is matched against the set of policies, yielding one, none, or multiple policies. As long as there is at least one matching policy, the request is allowed. If there are no matching policies, the request is denied.

Consider a simple (demo) set of policies of the form (subject, action, resource):

1: (team:local:admins, read, auth:teams)
2: (user:local:user1, update, compliance:node:*)

The first policy states that an actor that is a member of the local admins team is granted read access to the auth:teams resource. As noted earlier, the auth:teams resource is tied to the GetTeams RPC and /auth/teams HTTP endpoint. Thus, this HTTP request coming in for /auth/teams from an admins member --

subjects: [user:local:123, team:local:admins, team:local:other]
action: read
resource: auth:teams

-- is allowed to proceed because it matches policy #1 above. (See [5] in Figure 1.)

Figure 2 shows all of this again, slowed way down by a factor of -- well, a huge factor -- illustrating how the pieces interconnect.

AuthZ Mapping

FIGURE 2

As another example--this time with wildcards--assume that a request comes in for the /compliance/special/5 endpoint. Further assume that that endpoint is annotated with the Action of update and Resource of compliance:node:* (conveniently in sync with policy 2 given above). In this instance, the requesting user is user2 so the generated AuthZ tuple is

subjects: [user:local:user2, team:local:something]
action: update
resource: compliance:node:5

As the only policy (#2) regarding this particular Action and Resource is for user1, the AuthZ service does not authorize the call so the gateway then denies the original HTTP request ([6] in Figure 1).

If, on the other hand, the request came in for user1, the result would be allowed because the wildcard in policy #2 matches any value at the end of that resource string, so compliance:node:5 or compliance:node:00000000-0000-4000-8000-000000000000 (or many others) would be allowed.

The next section provides further details about wildcards.

Wildcards

AuthZ uses wildcards in policy definitions so that you can create a policy that covers a large range of possible values. For example, if an Action may be create, read, update, or delete, rather than having to create four nearly identical policies each with a different Action, you can simply specify an Action of * that will match any value in an AuthZ query.

Wild though they be, wildcards do have rules, however. This section is focused on wildcards but by necessity has to introduce general formats as well, but formats will be discussed more in the following section.

An Action consists of lowercase alphabetic characters, and possibly underscores... or just a wildcard. Figure 3 illustrates this.

Policy Action

FIGURE 3 -- POLICY ACTION

A Subject may contain a wildcard in one of four possible locations, specifying:

  • any users (or similarly any teams) of a certain provider, using e.g. user:ldap:*
  • any teams (or similarly any users) at all, using, e.g. team:*
  • any API tokens, using token:*
  • any requestor at all, using *

Those are reflected in the railroad diagram in Figure 4. (It is not critical to absorb the entire diagram. It is here more for completeness for those who might want or need to know at some point.) The easiest way to understand what the diagram conveys for our purposes here is to trace a path through the diagram that "spells out" each of the above four expressions. Start at the left terminus and look for a path that gets you all the way to the right terminus.

Policy Subject

FIGURE 4 -- POLICY SUBJECT

A Resource is somewhat complex as well. What you can immediately glean from Figure 5 -- given your mastery of Figures 3 and 4 -- is that a Resource may consist of ... a bunch of stuff... or a single wildcard. But look closer. And squint. It also shows that a Resource might also consist of ... a bunch of different stuff... followed by a colon then a single asterisk. So a resource may either be just a wildcard, or the last term in a resource may be a wildcard. You can not mix a wildcard with text within a single term, as in stuff:pre*. A few examples of valid resources with wildcards are * or compliance:* or compliance:nodes:*.

Policy Resource

FIGURE 5 -- POLICY RESOURCE

Policy Semantics

Arguably the key to understanding how AuthZ does permissioning is understanding semantics of the policy.

It is important to keep in mind that an AuthZ query needs to only match some policy to be granted access. To put that another way, the permissions for a given query are determined by the union of all matching policies. For example, if a user is part of team A and team B, but only team A is granted permission by a policy to access a given resource, the query is allowed. (As long as some policy covers a subject/action/resource, that is sufficient to grant access.)

Let's again consider the triumvirate of a policy definition.

Action Semantics

Policy Action

FIGURE 6 -- ACTION SEMANTICS

An Action, as you have seen, is very straightforward. It is a verb (typically a CRUD operation) or a wildcard. Nothing to add about it here, other than to include it for completeness.

Subject Semantics

Whoa! Subjects sure look different all of a sudden, though, compared to what you saw earlier.

Policy Subject

FIGURE 7 -- SUBJECT SEMANTICS

Figure 7 is divided into two. Mostly we have been talking about individual subjects. The top part of the figure, though, is specifically describing the plural Subjects property. Which is to say, Subjects is either an array of Subject elements, or a single wildcard.

The lower half of Figure 7 expounds upon a single Subject, with arguably a bit higher-level view than you saw in Figure 4. The provider in the figure is one of the literals from Figure 4: local, ldap, or saml. The value is an ID value appropriate to the provider. You can also see here just which terms may specify a wildcard instead of a value. Doing that means that the containing policy will allow all values for that term. So you could, for example, create a policy to grant access to all users (user:*). Alternately, if you want to restrict a policy to just local users, use user:local:* instead. Likewise, a policy that would grant access to all chef servers and clients, as well as any user using API tokens, would be token:*.

Resource Semantics

Taking a high-level perspective on resources also reveals some interesting patterns.

Policy Resource

FIGURE 8 -- RESOURCE SEMANTICS

Resources are arguably the key to understanding permissioning because they involve not just wildcards but also the added dimension of hierarchy. Figure 8 reveals that there are ten types of valid resources. Here are concrete examples of them, all quite deliberately rooted from the same "branch" hierarchy here for discussion purposes. In a nutshell, a resource can have any number of terms, separated by colons, with a value or a wildcard at the end.

[ 1] cfgmgmt:nodes:<some node id here>:runs:<some run id here>
[ 2] cfgmgmt:nodes:<some node id here>:runs:*
[ 3] cfgmgmt:nodes:<some node id here>:runs
[ 4] cfgmgmt:nodes:<some node id here>:*
[ 5] cfgmgmt:nodes:<some node id here>
[ 6] cfgmgmt:nodes:*
[ 7] cfgmgmt:nodes
[ 8] cfgmgmt:*
[ 9] cfgmgmt
[10] *

Rule 1: A wildcard matches any value for the given term.

You have already been exposed to this case, but just to be explicit, say there is a Resource of cfgmgmt:nodes:* on a policy.

Now let's say a query came in checking permissions for node 23, so the Resource in the query is cfgmgmt:nodes:23. The AuthZ service would grant permission because the policy wildcard matches any value. That is represented concisely here in the first row. The additional rows show that this applies for any term of a Resource.

Query                       Policy                 Allow or Deny?
-------                     ------                 --------------
cfgmgmt:nodes:23            cfgmgmt:nodes:*        Allow
cfgmgmt:nodes               cfgmgmt:*              Allow
cfgmgmt                     *                      Allow
compliance:nodes            cfgmgmt:*              Deny
compliance                  *                      Allow

Rule 2: A wildcard is inclusive of everything deeper in the hierarchy.

A policy with a Resource of cfgmgmt:nodes:23:* would match anything deeper in the same branch:

Query                       Policy                 Allow or Deny?
-------                     ------                 --------------
cfgmgmt:nodes:23:runs       cfgmgmt:nodes:23:*     Allow
cfgmgmt:nodes:23:runs:199   cfgmgmt:nodes:23:*     Allow
cfgmgmt:nodes:5:runs:199    cfgmgmt:nodes:23:*     Deny

Rule 3: A wildcard is exclusive of its container.

This case is notably not included under rule 2:

Query                       Policy                 Allow or Deny?
-------                     ------                 --------------
cfgmgmt:nodes:23            cfgmgmt:nodes:23:*     Deny

While that might seem odd at first glance, let's consider a variation, just going up a level to look at a different set of semantic objects:

Query                       Policy                 Allow or Deny?
-------                     ------                 --------------
cfgmgmt:nodes:23            cfgmgmt:nodes:*        Allow
cfgmgmt:nodes               cfgmgmt:nodes:*        Deny

Here we have a policy that says permission is granted for any value (node) under cfgmgmt:nodes. So a query coming in to access node 23 is granted, clearly. But the second query, to access the container of the nodes itself, is denied. A Resource of cfgmgmt:nodes might, for example, be attached to a ListNodes method or a CreateNode method. And it is conceivable that you might want to grant access to particular nodes, but not necessarily to see (list) all the nodes or to create a node.

Rule 4: A literal (non-wildcard) is non-hierarchical.

The absence of a wildcard in a policy means that requests must be exact matches. There is no implied hierarchical permissions. A few examples:

Query                       Policy                 Allow or Deny?
-------                     ------                 --------------
cfgmgmt:nodes               cfgmgmt:nodes          Allow
cfgmgmt:nodes:23            cfgmgmt:nodes          Deny
cfgmgmt:nodes:23            cfgmgmt:nodes:23       Allow
cfgmgmt:nodes:23:runs:99    cfgmgmt:nodes:23       Deny

Rule 5: Overlapping policies are harmless.

You can safely add policies with resources that overlap by wildcard hierarchy inclusiveness. (Note the same set of policies is used for each query.)

Query                       Policy                   Allow or Deny?
-------                     ------                   --------------
                          / cfgmgmt:nodes:*         \
cfgmgmt:nodes:23            cfgmgmt:*                Allow
                          \ cfgmgmt:nodes:23:runs:* /

                          / cfgmgmt:nodes:*         \
cfgmgmt:nodes:42            cfgmgmt:*                Allow
                          \ cfgmgmt:nodes:23:runs:* /

                          / cfgmgmt:nodes:*         \
cfgmgmt:nodes:23:runs:11    cfgmgmt:*                Allow
                          \ cfgmgmt:nodes:23:runs:* /

                          / cfgmgmt:nodes:*         \
cfgmgmt:nodes:42:runs:11    cfgmgmt:*                Allow
                          \ cfgmgmt:nodes:23:runs:* /

                          / cfgmgmt:nodes:*         \
cfgmgmt:special             cfgmgmt:*                Allow
                          \ cfgmgmt:nodes:23:runs:* /

Managing Policies

So You Want to Create Policies

In order to create policies you need to know:

  1. What provider your organization uses for users and teams. This could be local, meaning you will manage users and teams within Chef Automate, or it could be ldap or saml, depending on your needs.
  2. Teams and/or user IDs to be permissioned.
  3. Actions and resources associated with public API methods.

A cross-reference chart from API methods to actions and resources is provided here.

Default Policies

Out of the box, Chef Automate provides a set of default policies that provide access to all non-administrative tasks to all users, and access to all administrative tasks to all users in the local admins group (team:local:admins). This allows you to explore the user interface immediately, unencumbered with policy roadblocks limiting what you can see. The set of default policies is here.

That set of default policies may even be fine for certain organizations. For others, it will be necessary to add and remove policies to lock the system down so that known users are authorized to access specific resources.

Policy API

Documentation of the complete Policy API are available here. Link to the swagger docs, or should the content be inline here??

Managing policies is done via the command line. These setup steps will make the process as productive as possible for you.

  1. Install the jq JSON formatter to be able to get pretty-printed output (brew install jq).
  2. Install the jo JSON generator to be able to type JSON concisely (brew install jo).
  3. In habitat studio, generate a non-expiring token (generate_supertoken).
  4. Copy that token into the TOK value in the following snippet and then execute to define these shell variables:
export TARGET_HOST=https://a2-dev.test/api/v0
export TOK="RyABKRdmg49UvOs7o73Q_Igcyng="
export TOKEN_ARG="api-token: $TOK"

With those preliminaries, the following concise syntax gives you access to the Policy API:

List Policies

This endpoint lists all the policies currently in the policy store.

curl -sSkH "$TOKEN_ARG" $TARGET_HOST/auth/policies | jq .

Introspect Permissions for Current User

Introspection lets you see which endpoints you have permission to access. (If you used the supertoken generator above, you should have permissions for everything.)

curl -sSkH "$TOKEN_ARG" $TARGET_HOST/auth/introspect | jq .

Create Policy

jo saves typing a lot of quotation marks required by the JSON format. This command executes an HTTP POST request.

curl -sSkH "$TOKEN_ARG" $TARGET_HOST/auth/policies --data "$(jo -p subjects=$(jo -a user:local:[email protected]) resource=auth:teams:* action=read version=v0)" | jq .

Delete Policy

Use "List Policies" above to find the unique identifier for the policy you wish to delete, and supply it here with an HTTP DELETE request.

curl -sSkH "$TOKEN_ARG" $TARGET_HOST/auth/policies/f4fd7963-21db-4b5b-a8df-aa688056ed9d --request DELETE | jq .

Introspection (How to Query Permissions)

In many cases it is useful to know a priori what calls on a given view one can make before deciding what to render. When UI code makes any call to the backend, e.g. "/cfgmgmt/stats/run_counts", the system checks whether the current user has permission to access that endpoint. If so, the call proceeds; otherwise, a 403 error is returned. While that provides a robust security model, it does not by itself provide a robust user experience.

Introspection, in the context of Chef Automate, is the process of determining permissions for a user. That is, we take the user id and the team membership for the user, and check this against all policies in the system, and return a list of those endpoints that the user is allowed to access on at least one HTTP method. Endpoints where none of the methods are accessible are elided.

Introspection is intended for use by the Automate UI so that it may present a better user experience. When rendering a page, the UI code should first make a call to the introspection endpoint. From the list returned, it should then choose appropriate elements to render so that they will all perform permitted operations (as opposed to blindly putting controls for everything and the user then encounters 403/unauthorized responses for the disallowed operations). Note that policies may change at any time so there is no guarantee against 403 errors.

The introspection endpoint, /auth/introspect, supports two methods:

  1. GET - Does an "IntrospectAll", returning permissions for all non-parameterized endpoints (i.e. concrete endpoints like "/auth/teams" but not abstract endpoints like "/auth/users/{email}").
  2. POST - Does an "Introspect", returning data for the single endpoint path specified in the payload (an example payload might be "path=/auth/users/[email protected]" corresponding to "/auth/users/{email}").

Keep in mind that the POST here refers to the /auth/introspect endpoint -- not to the target endpoint that is passed as an argument. So though the target endpoint might use an HTTP GET method (such as with /auth/users/{email}) you still introspect it with the POST method on the introspection endpoint.

Data from either call returns the same structure:

{
  endpoints: {
    <endpoint_path>: {
      "get": <boolean>,
      "put": <boolean>,
      "post": <boolean>,
      "delete": <boolean>,
      "patch": <boolean>
    },
    <endpoint_path>: {
      "get": <boolean>,
      "put": <boolean>,
      "post": <boolean>,
      "delete": <boolean>,
      "patch": <boolean>
    },
    . . .
  }
}

Here is an example, looking at just one protobuf file and showing how the non-parameterized methods in the proto file map to the resultant list.

IntrospectAll

FIGURE 9

That list returned by IntrospectAll skips all the parameterized endpoints because it would be impossible to know a priori the exhaustive set. That is, if there is an endpoint /compliance/nodes/{node_id}, the UI will fill in that {node_id} at the time of rendering for however many nodes are being displayed and make those calls to the back end. But if we want to get a list of permissions in advance, the backend does not have a list of all possible node IDs and, even if it did, it would be prohibitive to check all of them.

At the time of page rendering the UI code knows what set of IDs it is interested in rendering. So the second rpc method Introspect may be used to query a concrete (filled-in) version of that single endpoint. So it could, for example, it could introspect on /compliance/nodes/23 to see if permission on that node is granted and, if so, render it. Then similarly for the remaining ones to be considered for display.

Introspection supports single (e.g. /abc/def/{email}) or multiple (e.g. /abc/def/{email}/runs/{id}) place holders within a path. In the degenerate case, this also includes no place holders, so you could introspect any single endpoint, parameterized or not.

Thus one may introspect for concrete forms of any of these and more:

  • /auth/users
  • /auth/teams/{id}/users
  • /auth/users/{email}
  • /cfgmgmt/nodes/{node_id}/runs/{run_id}

Here is what happens with a parameterized endpoint (Figure 10). The system checks all abstract, parameterized endpoints against the user's input, matching against the place holders. For those that match, the parameter from the URL is swapped into the same place holder in the resource. Then, just like standard introspection (Figure 2), that resource/action pair is sent off to the policy store to determine if any policies permit the operation.

Introspect

FIGURE 10

Most, but not all, endpoints in the public API put their parameters directly in the URL, similar to the above example. But a given endpoint might accept its data in the body of the request rather than in the URL (using the HTTP POST method, for example).

Introspection provides support for this as well. You need to provide an additional parameter in the introspection request, called parameters. This should be a list containing the name and value of each parameter of the target endpoint. For this example, there is only one necessary parameter (entity_uuid).

{
  "parameters": [
    "entity_uuid=zz123"
  ],
  "path": "/ingest/events/chef/run"
}

The list of parameters provided to the introspection call should include only those parameters required by the abstract resource. In the above example the resource is ingest:nodes:{entity_uuid}:runs. That is, even if the actual target endpoint requires more parameters, those should not be included when introspecting the endpoint.

Introspection on these POST endpoints works almost the same as shown for GET in Figure 10. But rather than extract parameters from the URL, they come from the request body.

Spinning up the GRPC Service Locally

This section describes how to bring up services in a standalone fashion. It is the quickest for iterative development but farthest from the way services are used in the "real world". This is NOT the recommended way to exercise services -- see the next section on running in the Habitat Studio -- but this involves fewer moving parts so it is quite useful if this is your first time here or for troubleshooting. If, for example, you want to use the interactive debugger, it is easier outside Habitat Studio (not sure if it is even possible inside). On the other hand, if you are already familiar with Habitat Studio, then using the studio might be simpler as you do not need your own GO setup and configuration.

Start the service

In this directory, launch the service in the background:

$ make start &

Connect to the service directly

From the root of the service directory, confirm the service is responding properly with this sanity check:

$ grpcurl -insecure -cert ../../dev/certs/authz-service.crt --key ../../dev/certs/authz-service.key localhost:9094 chef.automate.domain.authz.Authorization/GetVersion

<Version Response>

Next, test creating a policy using the Policy API:

$ echo '{"subjects": ["teams:local:admins"], "resource": "compliance:nodes:foo", "action": "read" }' | grpcurl -v -insecure -cert ../../dev/certs/authz-service.crt -key ../../dev/certs/authz-service.key -d @ localhost:9094 chef.automate.domain.authz.Authorization/CreatePolicy

Then, when you send an IsAuthorized query with the same parameters as the policy, you should see that the query is authorized:

$ echo '{"subjects": ["teams:local:admins"], "resource": "compliance:nodes:foo", "action": "read" }' | grpcurl -v -insecure -cert ../../dev/certs/authz-service.crt -key ../../dev/certs/authz-service.key -d @ localhost:9094 chef.automate.domain.authz.Authorization/IsAuthorized

<Successful Authorization Response>

Test the IsAuthorized call with an unauthorized group name by using any other string in the subjects, action, or resource field:

$ echo '{"subjects": ["teams:local:unauthorized_team"], "resource": "compliance:nodes:foo", "action": "read" }' | grpcurl -v -insecure -cert ../../dev/certs/authz-service.crt -key ../../dev/certs/authz-service.key -d @ localhost:9094 chef.automate.domain.authz.Authorization/IsAuthorized

<Failed Authorization Response>

You can also try adding policies with wildcard attributes. The policy created with the command below allows any user to do any action on any Compliance resource:

$ echo '{"subjects": ["*"], "resource": "compliance:*", "action": "*" }' | grpcurl -v -insecure -cert ../../dev/certs/authz-service.crt -key ../../dev/certs/authz-service.key -d @ localhost:9094 chef.automate.domain.authz.Authorization/CreatePolicy

NOTE: grpcurl is configured to omit empty responses, which is why the returned unauthorized response is an empty object.

Connect to the service through Automate Gateway

If not already running, launch the gateway service in the background:

$ cd ../automate-gateway
$ make run &

The GetVersion method is exposed on the gateway, so may be invoked:

$ curl --insecure -H "Authorization: Bearer dev" https://localhost:2000/authz/version
{"name":"authz-service","version":"0.0.1","sha":"TODO","built":"TODO"}

The IsAuthorized method, however, is not exposed on the gateway; it is intended strictly for use by other gateway services.

Spinning up the GRPC Service in Habitat Studio Locally

This section describes how to bring up services in Habitat Studio. It is a bit closer to "real world" conditions but involves a bit more overhead. Even so, it provides commands to make iterative development nearly as quick as running services in isolation as discussed above, and is the recommended way to exercise services locally.

Start the service in Habitat Studio

First, enter the studio. Make sure you are in the root of the a2 repository!

$ cd your-code-dir
$ cd a2
$ rm -r results  # usually a good idea to clean this dir out first!
$ hab studio enter

... Lots of useful details reported about using the studio here...

[1][default:/src:0]#

Generally it is useful to always start with this export command to make subsequent tasks faster, as the studio preamble describes:

[1][default:/src:0]# export GO_FAST=true

Next, invoke the start_* command to launch the service of interest:

[2][default:/src:0]# start_authz_service

Note that in the studio, these start_* commands automatically run services in the background; you do not need to force that with a trailing ampersand.

Connect to the service directly in Habitat Studio

Confirm the service is responding properly with this sanity check. Note that the paths to the certificates differ from the same invocation outside the studio, simply because you are in a different directory. Here you are in the a2 root, whereas earlier you were in the service directory.

[3][default:/src:0]# grpcurl -v -insecure -cert dev/certs/authz-service.crt -key dev/certs/authz-service.key localhost:9094 chef.automate.domain.authz.Authorization/GetVersion

<Version Response>

Next, populate the service with a policy.

[3][default:/src:0]# echo '{"subjects": ["teams:local:admins"], "resource": "config-mgmt:nodes:foo", "action": "read" }' | grpcurl -v -insecure -cert dev/certs/authz-service.crt -key dev/certs/authz-service.key -d @ localhost:9094 chef.automate.domain.authz.Authorization/CreatePolicy

<Created Policy Response>

Test the IsAuthorized call with the policy you just created.

[4][default:/src:0]# echo '{"subjects": ["teams:local:admins"], "resource": "config-mgmt:nodes:foo", "action": "read" }' | grpcurl -v -insecure -cert dev/certs/authz-service.crt -key dev/certs/authz-service.key -d @ localhost:9094 chef.automate.domain.authz.Authorization/IsAuthorized

<Successful Authorization Response>

Connect to the service through Automate Gateway in Habitat Studio

Just above you launched the target service of interest. Now you need to add the gateway service.

[5][default:/src:0]# start_gateway_service

The GetVersion method is exposed on the gateway, so may be invoked. However, Habitat Studio is like a clean room; very little in the way of utilities are pre-installed. Thus, to use curl you need to install it first. Then you can run the same curl invocation you saw earlier.

[6][default:/src:1]# hab pkg install -b core/curl

[7][default:/src:1]# curl --insecure -H "Authorization: Bearer dev" https://localhost:2000/authz/version
{"name":"authz-service","version":"0.0.1","sha":"TODO","built":"TODO"}

Keep in mind that the Automate gateway exposes services through conventional HTTP protocols. Thus, you can run the very same curl command outside Habitat Studio and it will still work!

As mentioned in the previous section, the IsAuthorized method is not exposed on the gateway; it is intended strictly for use by other gateway services. Thus, you cannot access it with the HTTP protocol.

API

func GetVersion

func GetVersion() VersionInfo, Error

Returns details about the version of the Authorization service. The Error is always returned as nil.

func IsAuthorized

func IsAuthorized(IsAuthorizedReq) IsAuthorizedResp, Error

Determines if an API request is authorized. If so, IsAuthorizedResp.Authorized is true and the Error is nil. If not, the Error is non-nil and the IsAuthorizedResp is nil.

type IsAuthorizedReq

type IsAuthorizedReq struct {
  Subjects  string // the user, and the teams they belong to, whose access is being determined
  Resource string // the resource for which access is wanted
  Action   string // the action on the resource
}

type IsAuthorizedResp

type IsAuthorizedResp struct {
  Authorized bool // set to true to indicate access is allowed
}

type VersionInfo

type VersionInfo struct {
  Name    string // service name
  Version string // service version
  Sha     string // checksum of the git HEAD used to build the binary
  Built   string // creation date of the binary
}

Other Resources

To be sure, there are a lot of moving parts to understand to successfully navigate the world of Automate2 services. Besides the above, be sure to review the A2 Development Environment.