[
https://issues.apache.org/jira/browse/SENTRY-551?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14230524#comment-14230524
]
Gregory Chanan commented on SENTRY-551:
---------------------------------------
A couple of questions:
bq. the relationship between group and roles will be stored in the Database and
must through the Sentry Service
What about the user -> group mapping, that is still determined by the client
before communicating with the Sentry Service based on the group mapping defined
in sentry-site.xml?
bq. If the requestUser dosen’t belong to the asked groups, it has no access get
the roles information the groups belong to.
There are two things I don't understand here:
1) Why is the group mapping consulted twice (once by the client, to be able to
pass to the sentry service) and once by the sentry service for the "no acces"
check. There should be a single source of truth for each invocation -- either
the groups are the client groups or the groups are the service groups. Mixing
them is just confusing.
2) What are you trying to prevent by having this "no access" check? Presumably
the sentry service should have its own authentication/authorization mechanism,
I'm not sure how this check fits in.
> Extend the ProviderBackend interface method getRoles when Solr/Sentry
> integation with DB Provider
> -------------------------------------------------------------------------------------------------
>
> Key: SENTRY-551
> URL: https://issues.apache.org/jira/browse/SENTRY-551
> Project: Sentry
> Issue Type: Bug
> Reporter: guoquanshen
> Assignee: guoquanshen
> Attachments: SENTRY-551.001.patch
>
>
> In document-level security, the class QueryDocAuthorizationComponent is used
> to get roles the request user belong to and
> adds a FilterQuery with the roles to restrict certain documents.Right now the
> procedure of QueryDocAuthorizationComponent used to get roles the request
> user belong to as following:
> {code}Roles= QueryDocAuthorizationComponent.getRoles(requestUser)
> ->ProviderBackend.getRoles(GroupMappingService.getGroups(requestUser)){code}
> The interface ProviderBackend as following:
> {code}public interface ProviderBackend {
> public ImmutableSet<String> getRoles(Set<String> groups, ActiveRoleSet
> roleSet);
> ...
> }{code}
> The interface must be extended as following:
> {code}public interface ProviderBackend {
> public ImmutableSet<String> getRoles(String requestor, Set<String> groups,
> ActiveRoleSet roleSet);
> ...
> }{code}
> That's the reason:
> ProviderBackend.getRoles(groups)
> The roles which the group belong to has been gotten through the
> providerBackend. Currently the Solr only uses a simple policy file as the
> backend, the format policy is as following:
> [groups]
> ops_group = ops_role
> [roles]
> ops_role = ...
> This policy is parsed through the class SimpleFileProviderBackend, it
> implements the interface ProviderBackend. This class is used to determine the
> relationship between the roles and groups. You can see that when user call
> SimpleFileProviderBackend.getRole(ops_group), it returns ops_role.
> However, when security search integration with DB store, the relationship
> between group and roles will be stored in the Database and must through the
> Sentry Service. There is a SimpleDBProviderBackend implemented the interface
> ProviderBackend can communication with Sentry Service
> to get roles information. The Sentry Service exposes the thrift API like
> this: getRoles(requestUser, groups). Why the Sentry Service exposed API has a
> extra parameter requestUser, it is a protection mechanism. If the requestUser
> dosen’t belong to the asked groups, it has no access get the roles
> information the groups belong to.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
