On 13/07/2015 21:14, Jason Gunthorpe wrote:
> On Mon, Jun 22, 2015 at 03:42:37PM +0300, Haggai Eran wrote:
>> + switch (ib_event->event) {
>> + case IB_CM_REQ_RECEIVED:
>> + req->device = req_param->listen_id->device;
>> + req->port = req_param->port;
>> + req->local_gid = &req_param->primary_path->sgid;
>> + req->service_id = req_param->primary_path->service_id;
>> + req->pkey = be16_to_cpu(req_param->primary_path->pkey);
>
> I feel pretty strongly that we should be using the pkey from the work
> completion, not the pkey in the message.
>
> The reason, if someone is using pkey like vlan, and expecting a
> container to never receive packets outside the assigned pkey, then we
> need to check each and every packet for the correct pkey before
> associating it with that container.
The way I see it is that you have one RDMA CM agent in the system, and
the header parameters address it. This agent allows addressing several
namespaces, and they are de-muxed according to the parameters in the
payload.
So a container never receives a packet outside of its assigned pkeys,
but the pkey you look at (as well as the GID, and possibly the IP
address) all come from the payload.
> When doing the namespace patches you should probably also look at
> other CM GMPs than just the REQ and how the paths are setup and
> consider what to do with the pkey. I'd probably suggest that the pkey
> should be forced throughout the entire process to ensure it always
> matches the ip device - at least for containers that is the right
> thing.. I probably wouldn't turn it on for the root namespace though..
Once a connection has been established, following GMPs use a unique ID
to address this connection, so no more de-muxing is needed. What is
really missing here I guess is a mechanism that would enforce containers
to only use certain pkeys - perhaps with something like an RDMA cgroup.
It could force containers to only use approved pkeys not only with RDMA
CM, but through uverbs, and other user-space interfaces.
Haggai
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
