tap-as-a-service/specs/mitaka/tap-as-a-service.rst

20 KiB

Tap-as-a-Service for Neutron

Launchpad blueprint:

https://blueprints.launchpad.net/neutron/+spec/tap-as-a-service

This spec explains an extension for the port mirroring functionality. Port mirroring involves sending a copy of packets ingressing and/or egressing one port (where ingress means entering a VM and egress means leaving a VM) to another port, (usually different from the packet's original destination). A port could be attached to a VM or a networking resource like router.

While the blueprint describes the functionality of mirroring Neutron ports as an extension to the port object, the spec proposes to offer port mirroring as a service, which will enable more advanced use-cases (e.g. intrusion detection) to be deployed.

The proposed port mirroring capability shall be introduced in Neutron as a service called "Tap-as-a-Service".

Problem description

Neutron currently does not support the functionality of port mirroring for tenant networks. This feature will greatly benefit tenants and admins, who want to debug their virtual networks and gain visibility into their VMs by monitoring and analyzing the network traffic associated with them (e.g. IDS).

This spec focuses on mirroring traffic from one Neutron port to another; future versions may address mirroring from a Neutron port to an arbitrary interface (not managed by Neutron) on a compute host or the network controller.

Different usage scenarios for the service are listed below:

  1. Tapping/mirroring network traffic ingressing and/or egressing a particular Neutron port.
  2. Tapping/mirroring all network traffic on a tenant network.
  3. Tenant or admin will be able to do tap/traffic mirroring based on a policy rule and set destination as a Neutron port, which can be linked to a virtual machine as normal Nova operations or to a physical machine via l2-gateway functionality.
  4. Admin will be able to do packet level network debugging for the virtual network.
  5. Provide a way for real time analytics based on different criteria, like tenants, ports, traffic types (policy) etc.

Note that some of the above use-cases are not covered by this proposal, at least for the first step.

Proposed change

The proposal is to introduce a new Neutron service plugin, called "Tap-as-a-Service", which provides tapping (port-mirroring) capability for Neutron networks; tenant or provider networks. This service will be modeled similar to other Neutron services such as the firewall, load-balancer, L3-router etc.

The proposed service will allow the tenants to create a tap service instance to which they can add Neutron ports that need to be mirrored by creating tap flows. The tap service itself will be a Neutron port, which will be the destination port for the mirrored traffic.

The destination Tap-as-a-Service Neutron port should be created beforehand on a network owned by the tenant who is requesting the service. The ports to be mirrored that are added to the service must be owned by the same tenant who created the tap service instance. Even on a shared network, a tenant will only be allowed to mirror the traffic from ports that they own on the shared network and not traffic from ports that they do not own on the shared network.

The ports owned by the tenant that are mirrored can be on networks other than the network on which tap service port is created. This allows the tenant to mirror traffic from any port it owns on a network on to the same Tap-as-a-Service Neutron port.

The tenant can launch a VM specifying the tap destination port for the VM interface (--nic port-id=tap_port_uuid), thus receiving mirrored traffic for further processing (dependent on use case) on that VM.

The following would be the work flow for using this service from a tenant's point of view

  1. Create a Neutron port which will be used as the destination port. This can be a part of ordinary VM launch.
  2. Create a tap service instance, specifying the Neutron port.
  3. If you haven't yet, launch a monitoring or traffic analysis VM and connect it to the destination port for the tap service instance.
  4. Associate Neutron ports with a tap service instance if/when they need to be monitored.
  5. Disassociate Neutron ports from a tap service instance if/when they no longer need to be monitored.
  6. Destroy a tap-service instance when it is no longer needed.
  7. Delete the destination port when it is no longer neeeded.

Please note that the normal work flow of launching a VM is not affected while using TaaS.

Alternatives

As an alternative to introducing port mirroring functionality under Neutron services, it could be added as an extension to the existing Neutron v2 APIs.

Data model impact

Tap-as-a-Service introduces the following data models into Neutron as database schemas.

  1. tap_service
Attribute Name Type Access (CRUD) Default Value Validation/ Conversion Description
id UUID R, all generated N/A UUID of the tap service instance.
project_id String CR, all Requester N/A ID of the project creating the service
name String CRU, all Empty N/A Name for the service instance.
description String CRU, all Empty N/A Description of the service instance.
port_id UUID CR, all N/A UUID of a valid Neutron port An existing Neutron port to which traffic will be mirrored
status String R, all N/A N/A The operation status of the resource (ACTIVE, PENDING_foo, ERROR, ...)
  1. tap_flow
Attribute Name Type Access (CRUD) Default Value Validation/ Conversion Description
id UUID R, all generated N/A UUID of the tap flow instance.
name String CRU, all Empty N/A Name for the tap flow instance.
description String CRU, all Empty N/A Description of the tap flow instance.
tap_service_id UUID CR, all N/A Valid tap service UUID UUID of the tap service instance.
source_port UUID CR, all N/A UUID of a valid Neutron port UUID of the Neutron port that needed to be mirrored
direction ENUM (IN, OUT, BOTH) CR, all BOTH Whether to mirror the traffic leaving or arriving at the source port IN: Network -> VM OUT: VM -> Network
status String R, all N/A N/A The operation status of the resource (ACTIVE, PENDING_foo, ERROR, ...)

REST API impact

Tap-as-a-Service shall be offered over the RESTFull API interface under the following namespace:

http://wiki.openstack.org/Neutron/TaaS/API_1.0

The resource attribute map for TaaS is provided below:

direction_enum = ['IN', 'OUT', 'BOTH']

RESOURCE_ATTRIBUTE_MAP = {
    'tap_service': {
        'id': {'allow_post': False, 'allow_put': False,
               'validate': {'type:uuid': None}, 'is_visible': True,
               'primary_key': True},
        'project_id': {'allow_post': True, 'allow_put': False,
                       'validate': {'type:string': None},
                       'required_by_policy': True, 'is_visible': True},
        'name': {'allow_post': True, 'allow_put': True,
                 'validate': {'type:string': None},
                 'is_visible': True, 'default': ''},
        'description': {'allow_post': True, 'allow_put': True,
                        'validate': {'type:string': None},
                        'is_visible': True, 'default': ''},
        'port_id': {'allow_post': True, 'allow_put': False,
                             'validate': {'type:uuid': None},
                             'is_visible': True},
        'status': {'allow_post': False, 'allow_put': False,
                   'is_visible': True},
    },
    'tap_flow': {
        'id': {'allow_post': False, 'allow_put': False,
               'validate': {'type:uuid': None}, 'is_visible': True,
               'primary_key': True},
        'name': {'allow_post': True, 'allow_put': True,
                 'validate': {'type:string': None},
                 'is_visible': True, 'default': ''},
        'description': {'allow_post': True, 'allow_put': True,
                        'validate': {'type:string': None},
                        'is_visible': True, 'default': ''},
        'tap_service_id': {'allow_post': True, 'allow_put': False,
                      'validate': {'type:uuid': None},
                      'required_by_policy': True, 'is_visible': True},
        'source_port': {'allow_post': True, 'allow_put': False,
                      'validate': {'type:uuid': None},
                      'required_by_policy': True, 'is_visible': True},
        'direction': {'allow_post': True, 'allow_put': False,
                             'validate': {'type:string': direction_enum},
                             'is_visible': True},
        'status': {'allow_post': False, 'allow_put': False,
                   'is_visible': True},
    }
}

Security impact

A TaaS instance comprises a collection of source Neutron ports (whose ingress and/or egress traffic are being mirrored) and a destination Neutron port (where the mirrored traffic is received). Security Groups will be handled differently for these two classes of ports, as described below:

Destination Side:

Ingress Security Group filters, including the filter that prevents MAC-address spoofing, will be disabled for the destination Neutron port. This will ensure that all of the mirrored packets received at this port are able to reach the monitoring VM attached to it.

Source Side:

Ideally it would be nice to mirror all packets entering and/or leaving the virtual NICs associated with the VMs that are being monitored. This means capturing ingress traffic after it passes the inbound Security Group filters and capturing egress traffic before it passes the outbound Security Group filters.

However, due to the manner in which Security Groups are currently implemented in OpenStack (i.e. north of the Open vSwitch ports, using Linux IP Tables) this is not possible because port mirroring support resides inside Open vSwitch. Therefore, in the first version of TaaS, Security Groups will be ignored for the source Neutron ports; this effectively translates into capturing ingress traffic before it passes the inbound Security Group filters and capturing egress traffic after it passes the outbound Security Group filters. In other words, port mirroring will be implemented for all packets entering and/or leaving the Open vSwitch ports associated with the respective virtual NICs of the VMs that are being monitored.

There is a separate effort that has been initiated to implement Security Groups within OpenvSwitch. A later version of TaaS may make use of this feature, if and when it is available, so that we can realize the ideal behavior described above. It should be noted that such an enhancement should not require a change to the TaaS data model.

Keeping data privacy aspects in mind and preventing the data center admin from snooping on tenant's network traffic without their knowledge, the admin shall not be allowed to mirror traffic from any ports that belong to tenants. Hence creation of 'Tap_Flow' is only permitted on ports that are owned by the creating tenant.

If an admin wants to monitor tenant's traffic, the admin will have to join that tenant as a member. This will ensure that the tenant is aware that the admin might be monitoring their traffic.

Notifications impact

A set of new RPC calls for communication between the TaaS server and agents are required and will be put in place as part of the reference implementation.

IPv6 impact

None

Other end user impact

Users will be able to invoke and access the TaaS APIs through python-neutronclient.

Performance Impact

The performance impact of mirroring traffic needs to be examined and quantified. The impact of a tenant potentially mirroring all traffic from all ports could be large and needs more examination.

Some alternatives to reduce the amount of mirrored traffic are listed below.

  1. Rate limiting on the ports being mirrored.
  2. Filters to select certain flows ingressing/egressing a port to be mirrored.
  3. Having a quota on the number of TaaS Flows that can be defined by the tenant.

Other deployer impact

Configurations for the service plugin will be added later.

A new bridge (br-tap) mentioned in Implementation section.

Developer impact

This will be a new extension API, and will not affect the existing API.

Community impact

None

Follow up work

Going forward, TaaS would be incorporated with Service Insertion1 similar to other existing services like FWaaS, LBaaS, and VPNaaS.

While integrating Tap-as-a-Service with Service Insertion the key changes to the data model needed would be the removal of 'network_id' and 'port_id' from the 'Tap_Service' data model.

Some policy based filtering rules would help alleviate the potential performance issues.

We might want to ensure exclusive use of the destination port.

We might want to create the destination port automatically on tap-service creation, rather than specifying an existing port. In that case, network_id should be taken as a parameter for tap-service creation, instead of port_id.

We might want to allow the destination port be used for purposes other than just launching a VM on it, for example the port could be used as an 'external-port'2 to get the mirrored data out from the tenant virtual network on a device or network not managed by openstack.

We might want to introduce a way to tap a whole traffic for the specified network.

We need a mechanism to coordinate usage of various resources with other agent extensions. E.g. OVS flows, tunnel IDs, VLAN IDs.

Implementation

The reference implementation for TaaS will be based on Open vSwitch. In addition to the existing integration (br-int) and tunnel (br-tun) bridges, a separate tap bridge (br-tap) will be used. The tap bridge provides nice isolation for supporting more complex TaaS features (e.g. filtering mirrored packets) in the future.

The tapping operation will be realized by adding higher priority flows in br-int, which duplicate the ingress and/or egress packets associated with specific ports (belonging to the VMs being monitored) and send the copies to br-tap. Packets sent to br-tap will also be tagged with an appropriate VLAN id corresponding to the associated TaaS instance (in the initial release these VLAN ids may be reserved from highest to lowest; in later releases it should be coordinated with the Neutron service). The original packets will continue to be processed normally, so as not to affect the traffic patterns of the VMs being monitored.

Flows will be placed in br-tap to determine if the mirrored traffic should be sent to br-tun or not. If the destination port of a Tap-aaS instance happens to reside on the same host as a source port, packets from that source port will be returned to br-int; otherwise they will be forwarded to br-tun for delivery to a remote node.

Packets arriving at br-tun from br-tap will get routed to the destination ports of appropriate TaaS instances using the same GRE or VXLAN tunnel network that is used to pass regular traffic between hosts. Separate tunnel IDs will be used to isolate different TaaS instances from one another and from the normal (non-mirrored) traffic passing through the bridge. This will ensure that proper action can be taken on the receiving end of a tunnel so that mirrored traffic is sent to br-tap instead of br-int. Special flows will be used in br-tun to automatically learn about the location of the destination ports of TaaS instances.

Packets entering br-tap from br-tun will be forwarded to br-int only if the destination port of the corresponding TaaS instance resides on the same host. Finally, packets entering br-int from br-tap will be delivered to the appropriate destination port after the TaaS instance VLAN id is replaced with the VLAN id for the port.

Assignee(s)

  • Vinay Yadhav

Work Items

  • TaaS API and data model implementation.
  • TaaS OVS driver.
  • OVS agent changes for port mirroring.

Dependencies

None

Testing

  • Unit Tests to be added.
  • Functional tests in tempest to be added.
  • API Tests in Tempest to be added.

Documentation Impact

  • User Documentation needs to be updated
  • Developer Documentation needs to be updated

References


  1. Service base and insertion https://review.openstack.org/#/c/93128↩︎

  2. External port https://review.openstack.org/#/c/87825↩︎