wiki:S3/S3AAA

Version 77 (modified by Dominic König, 10 years ago) ( diff )

--

DeveloperGuidelinesS3Framework | S3AAA

S3 Authentication, Authorization and Accounting

Authentication is the act of establishing or confirming someone's identity.
Authorization is the concept of allowing access to resources only to those permitted to use them.
Accounting refers to the tracking of user actions - an audit trail.

Overview

AAA functions for S3 are implemented in the modules/s3/s3aaa.py module. This module extends the web2py Auth class as AuthS3 (Authentication), and defines additional classes for role management, access control and audit.

ComponentLocationFunction
AuthS3modules/s3/s3aaa.pyAuthentication, Login
S3Permissionmodules/s3/s3aaa.pyAuthorization of Access, ACLs
S3Auditmodules/s3/s3aaa.pyData access logging, audit trail
S3RoleManagermodules/s3/s3aaa.pyRESTful method to manage roles and ACLs
Admin controllerscontrollers/admin.pyUser Management, role management

Authentication

The AuthS3 class extends web2py's Auth class, which is documented here:

NB Whilst the Authentication functions of the Auth class are only slightly modified, the Authorisation & Audit functions are not used at all, so ignore those when working with Sahana.

  • more coming...

Current user

  • coming soon...

Interactive Login

  • coming soon...

HTTP Basic Authentication

It is possible to access privileged resources by providing the username/password within each request, rather than the usual method of having the login stored within the session.

This is done by using the login_bare() method of AuthS3.

  • more coming...

Roles

Roles are defined in the auth_group table, which is defined by the AuthS3 module (in modules/s3/s3aaa.py).
Each role has an ID, a unique name and an optional description.

Access permissions are granted to roles, while a user gets permissions by assigning roles to them. Role assignment is stored in the auth_membership table, which is defined by the AuthS3 class (in modules/s3/s3aaa.py).

At the start of every request, the IDs of all roles of the currently logged-in user are stored as list in session.s3.roles (in models/00_utils.py).

In cases where the user logs in during the request (e.g. by HTTP Basic Auth), a refresh of this list is also triggered by the login_bare() method of AuthS3.

Roles can be managed in the S3RoleManager graphical interface (Administration menu => User Management => Roles).

The following roles are pre-defined in S3 and cannot be changed:

IDNameDescription
1Administratorsystem administrator
2Authenticatedall authenticated users
3Anonymousunauthenticated users
4Editordata editor

The first registered user gets the Administrator role assigned.
Users with the Administrator role always have all permissions, and may access all pages in Eden. The Administrator may also manage Users, Roles and ACLs.

Users with the Editor role may access all data with all methods, except they can not manage Users, Roles or ACLs.

Every authenticated user gets automatically the Authenticated role assigned. This role assignment cannot be revoked.

Simple Authorization

Simple authorization means, that

  • anonymous users have read-only permission
  • authenticated users have full access
  • the Administrator may additionally access the user management

This is the fallback model wherever there are no ACLs defined.

ACLs

Access Control Lists (ACLs) are bit arrays with each bit representing a permission to access data with a particular method:

BitValuePermission
auth.permission.CREATE0x01may create new records
auth.permission.READ0x02may read or list records
auth.permission.UPDATE0x04may update existing records
auth.permission.DELETE0x08may delete records

ACLs are combinations of these bits (by bitwise OR), e.g. an ACL with the value 0x06 defines permissions to read and update records, while no permission to add or to delete any records.

ACLs are stored per role and request destination in the s3_permission table, which is defined by the S3Permission class (in modules/s3/s3aaa.py).

ACLs can be managed by the S3RoleManager graphical interface (Administration menu => User Management => Roles).

For every destination (controller/function/table) two ACLs can be defined to apply depending on whether a user owns the record or not:

  • one ACL for users owning a record (Owner ACL = oacl)
  • one ACL for any other user not owning the record (User ACL = uacl).

If a user owns a record, then the most permissive of the User ACL and the Owner ACL gets applied, otherwise only the User ACL applies.

Record Ownership

Tables can implement a record ownership by adding two meta fields:

Field nameTypeDescription
owned_by_userinteger (reference auth_user)ID of the user who owns this record
(defaults to the ID of the user who created the record)
owned_by_roleinteger (reference auth_group)ID of the group (role) who own the record

These meta fields are contained in s3_authorstamp() and hence also in s3_meta_fields().

A user is considered the owner of a record if they are either the individual owner of the record (user ID == owned_by_user), or they are a member of the owner role (owned_by_role in user roles).

If a record has no owner, i.e. if both owned_by_user and owned_by_role are None, then all authenticated users are considered the owner of this record.

In tables which do not define either of these meta-fields, ownership rules are not applied (uacl only).

NOTE: you can have both an individual record owner and an owner role for the same record at the same time, where the individual owner doesn't need to have the owner role.

Ownership vs. Access Permission

NOTE: Ownership for a record and access permissions on this record are independent from each other!

The fact alone that a user owns a record doesn't give him any permissions on that record. He still needs to have a role assigned for which an ACL definition exists which gives him permissions on that table.

But that also means that the role which determines the user's ownership of a record, and the role which determines the user's effective access permissions for that record, do not have to be the same (in fact, the ownership-determining role doesn't need to have any permissions on the table at all): the effective permissions would still be applied as per the most permissive role the user is assigned to.

Example:

Have these things:

  1. A role OrgX Staff, which is routinely assigned to staff members of organisation X
  2. A role Boss, which is assigned to all organisation admins (independent of the organisation!)
  3. A role Clerk, which is assigned to all helpdesk officers (independent of the organisation!)
  4. A record Y in the table aaa_bbbbb, which has its owned_by_role field set to OrgX Staff
  5. An ACL for the aaa_bbbbb table, which sets (uacl=CREATE, oacl=ALL) for the Boss role
  6. An ACL for the aaa_bbbbb table, which sets (uacl=NONE, oacl=READ) for the Clerk role

With this configuration, a user who has the OrgX Staff role, would own record Y. However, the fact that he owns the record doesn't give him any permission to access it.

(Note that there is intentionally no ACL defined on aaa_bbbbb for role OrgX Staff!)

If the user would have both, the OrgX Staff and the Boss roles, then he would own the record Y (as per owned_by_role) and also be permitted to read, update and delete this record (as per ACL for Boss), and additionally, he could add new records to aaa_bbbbb.

If instead the user would have the OrgX Staff and the Clerk roles, then he would also own the record Y (as per owned_by_role), but just be permitted to read that record (as per ACL for Clerk).

If the user would only be Boss, then he could only create new records in aaa_bbbbb, but could not access record Y (since that would require ownership of that record).

If the user would only be Clerk, then he could not see record Y at all. And he could not either create new records in aaa_bbbbb.

Restrictions

ACLs can be defined for controllers, and for particular functions inside controllers (Controller ACLs).
ACLs can additionally be defined for individual database tables (Table ACLs).

System-wide Policy

The system-wide permission model is configured as security.policy in deployment settings models/000_config.py. This defaults to simple authorization (security.policy = 1).

To configure the system-wide policy to use ACLs, set security.policy to:

deployment_settings.security.policy = 3 # Apply Controller ACLs

or:

deployment_settings.security.policy = 4 # Apply both Controller and Function ACLs

or:

deployment_settings.security.policy = 5 # Apply Controller, Function and Table ACLs

Controller Restriction

To use controller ACLs, it must be specified for which controllers ACLs are to be used.

This can be done by setting the respective controller to restricted=True in deployment_settings.modules (models/000_config.py):

    dvi = Storage(
            name_nice = T("Disaster Victim Identification"),
            description = T("Disaster Victim Identification"),

            restricted = True, # Apply controller ACLs for the dvi module

            module_type = 10,
            resources = Storage(
                dvi_recreq = {"importer" : True},
            )
        ),

If restricted is False or undefined for a controller, then simple authorization is used for controller access.

NB: We could assume that if there is no ACL defined for a controller then this controller is unrestricted. However, this would require a second DB lookup per run to check whether the controller is restricted at all in case no specific ACL for the current user is found, and in order to keep this efficient, we introduced this deployment setting (which is also less confusing behavior).

The Controller ACL can be defined for all functions in a controller, and additionally for particular functions inside a controller, where the function-specific ACLs override the general controller ACL. That means, you can define a general ACL for the pr controller, and a different one for the pr/person function.

The Controller ACLs are applied to all resources when accessed through this controller/function. If the Controller ACL does not give any permission for the current user (ACL value==auth.permissions.NONE==0x00), then the request is rejected as "Unauthorized". Controllers do not have to implement this check - this is done at a central place (in 00_utils.py).

Table Restriction

Once the user has passed that controller permission check (must have at least read permission), and tries to access to a particular table, then the controller checks for table-specific ACLs. This check is to be implemented by the particular controller using s3_has_permission() and s3_accessible_query (except controllers using S3CRUD only, which already contains it).

If there is no ACL defined for this table at all (i.e. for none of the users), then the table is considered unrestricted and only the controller ACLs apply.

If there exist ACLs for this table, but not for the current user, access is denied for the current user.

If there are specific ACLs defined for this table and the current user, then the most restrictive of the controller and table ACLs apply (i.e. you cannot allow on the table level what you forbid at the controller level, and vice versa).

Note: For consistency reasons, creating or deleting component records in a resource requires additional permission to update the main record, even though the main record is not changed by this operation, e.g. to add an address to a person record, you must also be permitted to update the person record.

Implementation of Access Control

Permission checking is always a two-step process:

  1. Check permission to access the controller/function
  2. Check permission to access the database table

The first step is done at a central point, in 00_utils.py before the models are loaded. If the ACLs, as defined for the current user, do not specify any permission for the target controller/function, then the request gets rejected before any models are loaded or the controller is entered.

The second step has to be implemented in the respective controller functions. This can be done in two ways:

  • the controller uses s3_rest_controller() with S3CRUD, or,
  • the controller uses auth.s3_has_permission() and/or auth.s3_accessible_query() to check permissions before exposing any data to the user

Checking Permissions

To check permissions to access a table (or a particular record) with a certain method, use the auth.s3_has_permission() method:

authorised = auth.s3_has_permission("read", db.my_table)
if authorised:
    # User may read in the db.my_table
authorised = auth.s3_has_permission("read", db.my_table, record_id=x)
if authorised:
    # User may read record x in db.my_table

The access method can be one of these strings:

  • "create": create new records in this table
  • "read": read in this table (or this particular record, if specified)
  • "update": update existing records in this table (or this particular record, if specified)
  • "delete": delete records from this table (or this particular record, if specified)

Query for Accessible Records

You can build a query for all records in a table which are accessible for the current user with a certain method, by using auth.s3_accessible_query:

# Define your query:
query = ...

# Get accessible-query (e.g., all readable record in db.my_table):
accessible = auth.s3.accessible_query("read", db.my_table)

# Combine both parts:
query = accessible & query

# Perform the query
rows = db(query).select(...)

Handling Insufficient Permissions

In case the user has insufficient permissions to access a table/record with the requested method, a well-defined response action must take place depending on the request format:

  • in HTML format:
    • already authenticated users should be informed about the insufficient permissions, and redirected to a (unrestricted) landing page
    • unauthenticated users should be requested to login, and forwarded to a login page
  • in all other formats:
    • authenticated clients must receive an HTTP 403 (Forbidden) error code to cancel the request properly
    • unauthenticated clients must receive an HTTP 401 (Authorization Required) error in order to trigger an authentication attempt
    • the client must not be redirected in either of the cases (important!)

All this is covered by the auth.permission.fail() method:

authorised = auth.shn_has_permission("delete", db.my_table)
if not authorised:
    auth.permission.fail()

You can alter the destinations for redirection by setting:

  • auth.permission.homepage for redirection when the user is logged-in, but has insufficient privileges (defaults to default/index).
  • auth.permission.loginpage for redirection when the user is not logged-in (defaults to default/user/login).

Example: redirect to my/index rather than to default/index in case of insufficient privileges of an authenticated user:

authorised = auth.shn_has_permission("delete", db.my_table)
if not authorised:
    auth.permission.homepage = URL(r=request, c="my", f="index")
    auth.permission.fail()

Data Access Tracking (Audit)

  • coming soon

DeveloperGuidelinesS3Framework

Note: See TracWiki for help on using the wiki.