FAQ: Add troubleshooting procedure and example.

[ovs.git] / OPENFLOW-1.1+

                OpenFlow 1.1+ support in Open vSwitch
                =====================================

Open vSwitch support for OpenFlow 1.1, 1.2, and 1.3 is a work in
progress.  This file describes the work still to be done.

The Plan
--------

OpenFlow version support is not a build-time option.  A single build
of Open vSwitch must be able to handle all supported versions of
OpenFlow.  Ideally, even at runtime it should be able to support all
protocol versions at the same time on different OpenFlow bridges (and
perhaps even on the same bridge).

At the same time, it would be a shame to litter the core of the OVS
code with lots of ugly code concerned with the details of various
OpenFlow protocol versions.

The primary approach to compatibility is to abstract most of the
details of the differences from the core code, by adding a protocol
layer that translates between OF1.x and a slightly higher-level
abstract representation.  The core of this approach is the many struct
ofputil_* structures in lib/ofp-util.h.

As a consequence of this approach, OVS cannot use OpenFlow protocol
definitions that closely resemble those in the OpenFlow specification,
because openflow.h in different versions of the OpenFlow specification
defines the same identifier with different values.  Instead,
openflow-common.h contains definitions that are common to all the
specifications and separate protocol version-specific headers contain
protocol-specific definitions renamed so as not to conflict,
e.g. OFPAT10_ENQUEUE and OFPAT11_ENQUEUE for the OpenFlow 1.0 and 1.1
values for OFPAT_ENQUEUE.  Generally, in cases of conflict, the
protocol layer will define a more abstract OFPUTIL_* or struct
ofputil_*.

Here are the current approaches in a few tricky areas:

    * Port numbering.  OpenFlow 1.0 has 16-bit port numbers and later
      OpenFlow versions have 32-bit port numbers.  For now, OVS
      support for later protocol versions requires all port numbers to
      fall into the 16-bit range, translating the reserved OFPP_* port
      numbers.

    * Actions.  OpenFlow 1.0 and later versions have very different
      ideas of actions.  OVS reconciles by translating all the
      versions' actions (and instructions) to and from a common
      internal representation.

OpenFlow 1.1
------------

The list of remaining work items for OpenFlow 1.1 is below.  It is
probably incomplete.

    * Implement Write-Actions instruction.

    * The new in_phy_port field in OFPT_PACKET_IN needs some kind of
      implementation.  It has a sensible interpretation for tunnels
      but in general the physical port is not in the datapath for OVS
      so the value is not necessarily meaningful.  We might have to
      just fix it as the same as in_port.

    * On OF1.1+ flow_mods, updates by MODIFY are now much better
      specified.  Check that OVS implements the new behavior, fix it
      if not.

    * On OF1.1+ flow_mods, DELETE now ignores buffer_id.

    * OFPST_PORT and OFPST_QUEUE stats.  These differ little from
      OF1.0 to OF1.1 (just the size of the port number field) but do
      require abstraction (Done?).

    * OFPT_TABLE_MOD stats.  This is new in OF1.1, so we need to
      implement it.  It should be implemented so that the default OVS
      behavior does not change.

    * Document how OVS does packet buffering.

    * MPLS.  Simon Horman maintains a patch series that adds this
      feature.  It needs review and possible revision before it is
      merged.

    * SCTP.  Joe Stringer maintains a patch series that adds this
      feature.  It has received review comments that need to be
      addressed before it is merged.

    * Match and set double-tagged VLANs (QinQ).  This requires kernel
      work for reasonable performance.

    * VLANs tagged with 88a8 Ethertype.  This requires kernel work for
      reasonable performance.

    * Groups.

OpenFlow 1.2
------------

OpenFlow 1.2 support requires OpenFlow 1.1 as a prerequisite, plus the
following additional work.  (This is based on the change log at the
end of the OF1.2 spec.  I didn’t compare the specs carefully yet.)

    * Use new OpenFlow extensible error infrastructure, on OF1.2+
      only, instead of the OVS-specific extension used until now.

    * OFPT_FLOW_MOD:

        * MODIFY and MODIFY_STRICT commands now never insert new flows
          in the table.  We will still need variations that do,
          though, both to support older OpenFlow protocols and to get
          sensible behavior for the internal implementation of the
          NXAST_LEARN action.

        * New flag OFPFF_RESET_COUNTS.

        * New cookie field behavior.

        * Add ability to delete flow in all tables.

        * Update DESIGN to describe OF1.2 behavior also.

    * Implement OFPT_ROLE_REQUEST.  Patch submitted by Jarno
      Rajahalme, currently under revision.

    * Add ability to turn off packet buffering with OFPCML_NO_BUFFER.

OpenFlow 1.3
------------

OpenFlow 1.3 support requires OpenFlow 1.2 as a prerequisite, plus the
following additional work.  (This is based on the change log at the
end of the OF1.3 spec, reusing most of the section titles directly.  I
didn’t compare the specs carefully yet.)

    * Add support for multipart requests.

    * Add OFPMP_TABLE_FEATURES statistics.

    * More flexible table miss support.

    * IPv6 extension header handling support.  Fully implementing this
      requires kernel support.  This likely will take some careful and
      probably time-consuming design work.  The actual coding, once
      that is all done, is probably 2 or 3 days work.

    * Per-flow meters.  Similar to IPv6 extension headers in kernel
      and design requirements.  Might be politically difficult to add
      directly to the kernel module, since its functionality overlaps
      with tc.  Ideally, therefore, we could implement these somehow
      with tc, but I haven’t investigated whether that makes sense.

    * Per-connection event filtering.  OF1.3 adopted Open vSwitch’s
      existing design for this feature so implementation should be
      easy.

    * Auxiliary connections.  These are optional, so a minimal
      implementation would not need them.  An implementation in
      generic code might be a week’s worth of work.  The value of an
      implementation in generic code is questionable, though, since
      much of the benefit of axuiliary connections is supposed to be
      to take advantage of hardware support.  (We could make the
      kernel module somehow send packets across the auxiliary
      connections directly, for some kind of “hardware” support, if we
      judged it useful enough.)

    * MPLS BoS matching.  (Included in Simon's MPLS series?)

    * Provider Backbone Bridge tagging.  I don’t plan to implement
      this (but we’d accept an implementation).

    * Rework tag order.  I’m not sure whether we need to do anything
      for this.

    * Duration for stats.

    * On-demand flow counters.  I think this might be a real
      optimization in some cases for the software switch.

How to contribute
-----------------

If you plan to contribute code for a feature, please let everyone know
on ovs-dev before you start work.  This will help avoid duplicating
work.

Please consider the following:

    * Testing.  Please test your code.

    * Unit tests.  Please consider writing some.  The tests directory
      has many examples that you can use as a starting point.

    * ovs-ofctl.  If you add a feature that is useful for some
      ovs-ofctl command then you should add support for it there.

    * Documentation.  If you add a user-visible feature, then you
      should document it in the appropriate manpage and mention it in
      NEWS as well.

    * Coding style (see the CodingStyle file at the top of the source
      tree).

    * The patch submission guidelines (see SubmittingPatches).  I
      recommend using “git send-email”, which automatically follows a
      lot of those guidelines.

Bug Reporting
-------------

Please report problems to bugs@openvswitch.org.