On Tue, Jan 21, 2020 at 2:50 AM Seena Fallah <seenafallah@gmail.com> wrote:

OPA can be used in companies that uses many services like k8s, Ceph,... and want to have one central point for authorizing users so they can maintenance their access for each user on each service for example and etc. It’s just a use case and so it’s really good to have it. I think this is the biggest use case for having OPA in products that gets an option to centralize authorizing for all types of services.

Performance for this model is issue like having keystone with Ceph. So I think it’s based on users that active this integrations at all.

The model for writing policies to radosgw isn’t really good I think because of the reason above if this accrued there is always two copies of policies and it doesn’t sounds good for maintaining.

If bucket policy disable, s3 clients like boto3 and etc will not work for setting polices but I think when someone is enabling OPA for authorizing it will also have an API for his/her OPA server to set/del policies and they can call these APIs to set/del policies.
And for extensions like PublicAccessBlock, it will disable because OPA is just authorizing requests and Ceph doesn’t authorize any request when OPA integration is enabled so OPA should handle any incoming policies were made by S3 policies. So it doesn’t make conflicts and if OPA integration is enabled it won’t work as we return 405 on each set/del policies requests and if OPA is disabled users can use this policies.

On Tue, Jan 21, 2020 at 2:05 AM Casey Bodley <cbodley@redhat.com> wrote:
I am a big fan of the IAM policy documents, both because of the
flexibility and expressiveness they provide, and because they're in a
format that all of our s3 clients understand.

I'm not familiar enough with OPA to know what extra capabilities it
offers that IAM policy cannot, but I have serious concerns about the
performance and scalability of a model where radosgw has to send
blocking RPCs to OPA in order to authorize each and every request.

On the other hand, consider a model where a Policy Agent exercises its
control over authorization by writing IAM documents to radosgw, which we
use to cheaply authorize requests out of our metadata cache. I would
imagine that this model could cover a lot of interesting use cases,
without breaking support for existing s3 applications that rely on
bucket policy - as the proposal to reject PutBucketPolicy requests would.

Is this something that OPA could feasibly do?

For use cases that aren't supported by the existing policy grammar,
we're open to maintaining extensions to these documents. We already
implement a number of s3 extensions [1][2] that are easily accessible
via python/boto and the aws cli.

But a model where radosgw outsources authorization entirely is a hard
sell, because it conflicts with feature development going forward. One
example would be support for PublicAccessBlock [3], where radosgw needs
full visibility into policy to detect cases where public access would be
granted.

[1]
https://docs.ceph.com/docs/master/radosgw/s3/python/#using-s3-api-extensions

[2]
https://github.com/ceph/ceph/blob/master/examples/boto3/service-2.sdk-extras.json

[3] https://github.com/ceph/ceph/pull/30033

On 1/20/20 12:21 PM, Seena Fallah wrote:
> I’m also agree with you Matt that it will free us from complexity of
> handling S3 policy or Swift ACL if we save the current state of OPA.
> But if we want use this state of OPA we should act for S3 policy and
> Swift ACL that if user is setting them it shouldn’t be allowed and
> return user that you can’t set them! Because now when OPA integration
> is enabled and user set bucket policy it returns success but actually
> it doesn’t work!
>
> What are your thoughts?
>
> On Sun, Jan 19, 2020 at 12:33 AM Seena Fallah <seenafallah@gmail.com
> <mailto:seenafallah@gmail.com>> wrote:
>
> I think the other problem caused when OPA integration is enabled
> and we set bucket policy is when user wants to get his/her bucket
> policy. Some policies are set through OPA (for example in OPA
> rules we have user A that has access to user B bucket so OPA
> return true on authorizing request and it acts like bucket policy)
> and some through bucket policy (s3 clients command). So when user
> is getting his/her bucket policy what data should we return? The
> policies that are set through bucket policy or OPA rules for that
> bucket?
>
> I fact I think OPA rules are not static and will change in time
> and so there should be a client interface for that OPA server that
> users could change their rules for their buckets (giving access to
> put, get, ... to someone else and etc.). So if the client exists
> there is no need to bucket policy and we can make it disable (by
> returning 405) when OPA integration is enabled (I repeat that
> still now in Ceph latest version when OPA integration is enabled
> bucket policies aren’t work!) because the policies that are set
> with bucket policy can be check with OPA, too.
>
> What’s your opinion?
>
> On Fri, Jan 17, 2020 at 9:40 PM Seena Fallah
> <seenafallah@gmail.com <mailto:seenafallah@gmail.com>> wrote:
>
> I think when OPA integration is enabled the source of truth
> for authorizing should be OPA (it is right now in Ceph and all
> requests are authorizing with OPA and Ceph doesn’t authorize
> any request by it self).
> When user is using bucket policy feature he/she wants to get
> access to someone else so when he/she is the bucket owner,
> he/she can perform this action and we should apply this policy
> for him/her. If we want policies just update within OPA
> server/client and S3 clients (s3cmd, aws, ...) don’t edit
> policies, we should reply to them that set/delpolicy isn’t
> allowed from here (return 405 for example; just for saying
> that the request that user send isn’t successful).
>
> Yes we can have some process and simplification before sending
> it to OPA but the s3 policy has a general structure so OPA
> server can decode it by it self.
>
> On Fri, Jan 17, 2020 at 9:16 PM Matt Benjamin
> <mbenjami@redhat.com <mailto:mbenjami@redhat.com>> wrote:
>
> The larger question, I think, is what OPA is supposed to
> do with it.
> The larger question I think it asks is whether OPA or Ceph
> owns a
> particular dimension of policy--or, perhaps, which owns
> policy for
> what portions of the namespace (at any particular point in
> time).
>
> Without any new interaction, when OPA is configured, OPA
> can make a
> direct authorization decision with all available
> information for
> Ceph/RGW, notwithstanding any S3 or Swift ACL or S3 policy
> that might
> exist--including any that might have been stored prior to
> turning on
> this proposed feature to push policy documents to OPA. This
> overriding property of the OPA integration when in use
> frees us from a
> lot of complexity regarding which system is the source of
> truth, and
> for what.
>
> I can see value in more sophisticated integration that
> mutually
> comprehends policy--but I'm having trouble with "send
> policy documents
> to OPA, maybe it will do something with them."
>
> Matt
>
> On Fri, Jan 17, 2020 at 12:01 PM Seena Fallah
> <seenafallah@gmail.com <mailto:seenafallah@gmail.com>> wrote:
> >
> > Hello Ash
> >
> > With bucket policy user A can get access to user B for
> putting object on bucket C. So if this policy sent to Ceph
> and OPA integration is enabled it will be discard because
> this policy isn’t sent to OPA server to be updated.
> > Here is a documentation for bucket policy:
> > https://docs.ceph.com/docs/master/radosgw/bucketpolicy/
> >
> > With this PR when user set bucket policy, the data of
> that policy will sent to OPA server to be applied and so
> OPA can get access to user that gets access to bucket via
> bucket policy.
> >
> > On Fri, Jan 17, 2020 at 8:24 PM Ash Narkar
> <ash@styra.com <mailto:ash@styra.com>> wrote:
> >>
> >> Hello Seena,
> >>
> >> The OPA integration is with the RGW and the intent is
> to check if an authenticated user is allowed to perform a
> particular action on a particular resource. For example,
> can Bob delete a bucket based on some attribute like his
> location. I am not familiar with the internals of Ceph's
> bucket policy command. It would be great to get some
> context here and discuss if the bucket policy can be
> authorized with OPA which is the intent of your PR I believe.
> >>
> >> Thanks
> >> Ash
> >>
> >> On Fri, Jan 17, 2020 at 6:33 AM Seena Fallah
> <seenafallah@gmail.com <mailto:seenafallah@gmail.com>> wrote:
> >>>
> >>> So when OPA integration is enabled the bucket policy
> from users will not work!
> >>> I think it’s about Ceph architecture not OPA because
> OPA is for authorizing the requests and bucket policy is
> one of the authorizing methods that OPA should support.
> >>>
> >>> On Fri, Jan 17, 2020 at 5:56 PM Matt Benjamin
> <mbenjami@redhat.com <mailto:mbenjami@redhat.com>> wrote:
> >>>>
> >>>> Hi Seena,
> >>>>
> >>>> As I wrote in a comment on your PR, my current
> intuition is that what
> >>>> you're doing here isn't consistent with the original
> intent of the OPA
> >>>> integration we currently have, nor with the OPA model
> in general.
> >>>>
> >>>> That said, I'd really like some feedback from OPA
> architects, CC'd.
> >>>>
> >>>> regards,
> >>>>
> >>>> Matt
> >>>>
> >>>> On Thu, Jan 16, 2020 at 5:04 AM Seena Fallah
> <seenafallah@gmail.com <mailto:seenafallah@gmail.com>> wrote:
> >>>> >
> >>>> > Hi all. In OPA integration from Ceph there is no
> integration for bucket policy.
> >>>> > When user is setting bucket policy to his/her
> bucket the OPA server won't get who get's access to that
> bucket so after that if the request is coming from the
> user (that gets access to that bucket via bucket policy)
> to access that bucket (PUT, GET,...), OPA will reject that
> because of no data in database.
> >>>> > I have create a pull request for this problem so if
> user creates a bucket policy for his/her bucket, the
> policy data will send to OPA server to be update on the
> database.
> >>>> > I think the main idea of having OPA is to have all
> authorization in OPA and Ceph don't authorize any request
> by it self.
> >>>> > Here is the pull request and I would be thankful to
> hear about your comments.
> >>>> > https://github.com/ceph/ceph/pull/32294
> >>>> > Thanks.
> >>>> > _______________________________________________
> >>>> > Dev mailing list -- dev@ceph.io <mailto:dev@ceph.io>
> >>>> > To unsubscribe send an email to dev-leave@ceph.io
> <mailto:dev-leave@ceph.io>
> >>>>
> >>>>
> >>>>
> >>>> --
> >>>>
> >>>> Matt Benjamin
> >>>> Red Hat, Inc.
> >>>> 315 West Huron Street, Suite 140A
> >>>> Ann Arbor, Michigan 48103
> >>>>
> >>>> http://www.redhat.com/en/technologies/storage
> >>>>
> >>>> tel. 734-821-5101
> >>>> fax. 734-769-8938
> >>>> cel. 734-216-5309
> >>>>
>
>
> --
>
> Matt Benjamin
> Red Hat, Inc.
> 315 West Huron Street, Suite 140A
> Ann Arbor, Michigan 48103
>
> http://www.redhat.com/en/technologies/storage
>
> tel. 734-821-5101
> fax. 734-769-8938
> cel. 734-216-5309
>
>
> _______________________________________________
> Dev mailing list -- dev@ceph.io
> To unsubscribe send an email to dev-leave@ceph.io
_______________________________________________
Dev mailing list -- dev@ceph.io
To unsubscribe send an email to dev-leave@ceph.io