Re: [nvo3] BFD over VXLAN: Trapping BFD Control packet at VTEP

[Jeffrey Haas <[e-mail]>]

I also agree with Joel.

-- Jeff


> On Oct 31, 2019, at 11:59 AM, Joel M. Halpern <[e-mail]> wrote:
> 
> Explicitly restricting the discard behavior to the management VNI takes care of my concern.
> 
> Thank you,
> Joel
> 
> On 10/31/2019 11:48 AM, Greg Mirsky wrote:
>> Hi Jeff,
>> thank you for the detailed clarification of your questions. Please find my follow-up notes in-lined tagged GIM2>>.
>> Regards,
>> Greg
>> On Wed, Oct 30, 2019 at 2:14 PM Jeffrey Haas <[e-mail] <mailto:[e-mail]>> wrote:
>>    Greg,
>>    On Wed, Oct 30, 2019 at 01:58:30PM -0700, Greg Mirsky wrote:
>>     > On Wed, Oct 30, 2019 at 1:27 PM Jeffrey Haas <[e-mail]
>>    <mailto:[e-mail]>> wrote:
>>     >
>>     > > Greg,
>>     > >
>>     > > From the updated text:
>>     > >
>>     > > "At the same time, a service layer BFD session may be used
>>    between the
>>     > > tenants of VTEPs IP1 and IP2 to provide end-to-end fault
>>    management. In
>>     > > such case, for VTEPs BFD Control packets of that session are
>>     > > indistinguishable from data packets.  If end-to-end defect
>>    detection is
>>     > > realized as the set of concatenated OAM domains, e.g., VM1-1 -
>>    IP1 --
>>     > > IP2 - VM2-1, then the BFD session over VXLAN between VTEPs SHOULD
>>     > > follow the procedures described in Section 6.8.17 [RFC5880]."
>>     > >
>>     > > In the case that two VMs are running BFD to each other as a user
>>     > > application
>>     > > rather than as part of the virtualized environment, it's
>>    unlikely that
>>     > > they'd be treated as concatenated domains.  To do so, the
>>    tenant VMs would
>>     > > have to have a sense that they are indeed virtual.
>>     > >
>>     > > Is your intent in this text that BFD implementations on the
>>    server should
>>     > > detect BFD sessions between servers and change them to a
>>    concatenated
>>     > > session?
>>     > >
>>     > GIM>> No, we do not suggest that the concatenation of BFD sessions be
>>     > automagical. That may be controlled via the management plane though.
>>    Then my suggestion is we may not want this text.
>>    It's fine to say "if tenants want to run BFD to each other, and that is
>>    standard BFD (RFC 5881) from the perspective of those tenants" if that's
>>    your intent.  Leave automagic out of the spec. :-)
>> GIM2>> I'd take the passage referring to the concatenated path out. That will leave it as:
>>    At the same time, a service layer BFD session may be used between the
>>    tenants of VTEPs IP1 and IP2 to provide end-to-end fault management.
>>    In such case, for VTEPs BFD Control packets of that session are
>>    indistinguishable from data packets.
>>     > > Section 5 comment:
>>     > >
>>     > > :   The UDP destination port and the TTL of the inner IP packet
>>    MUST be
>>     > > :   validated to determine if the received packet can be
>>    processed by
>>     > > :   BFD.  BFD Control packets with unknown MAC address MUST NOT be
>>     > > :   forwarded to VMs.
>>     > >
>>     > > I'd suggest pushing the second sentence into the prior section
>>    since it
>>     > > deals with MAC addresses rather than the UDP procedures.
>>     > >
>>     > GIM>> Could you please clarify your suggestion - move to Section
>>    4 or to
>>     > the preceding paragraph? I think it is the latter but wanted to
>>    make sure.
>>    Full section 5 from your draft-8 candidate:
>>    : 5.  Reception of BFD Packet from VXLAN Tunnel
>>    :
>>    :    Once a packet is received, the VTEP MUST validate the packet.     If the
>>    :    Destination MAC of the inner Ethernet frame matches one of the MAC
>>    :    addresses associated with the VTEP the packet MUST be processed
>>    :    further.  If the Destination MAC of the inner Ethernet frame
>>    doesn't
>>    :    match any of VTEP's MAC addresses, then the processing of the
>>    :    received VXLAN packet MUST follow the procedures described in
>>    :    Section 4.1 [RFC7348].
>>    It's not clear what that procedure is, with respect to BFD.  Section 4.1
>>    basically says is that when a mapping is discovered, deliver it to
>>    that VM
>>    with headers removed.
>>    Section 4.1 really doesn't discuss dropping behavior.
>>    :
>>    :    The UDP destination port and the TTL of the inner IP packet MUST be
>>    :    validated to determine if the received packet can be processed by
>>    :    BFD.
>>    This is fine.
>>    :    BFD Control packets with unknown MAC address MUST NOT be
>>    :    forwarded to VMs.
>>    This appears to be clarifying the missing point in the prior
>>    paragraph.  If
>>    that's the case, why is this sentence not part of the prior paragraph?
>> GIM>> So I thought. Moving the sentence to the first paragraph highlighted the contradiction others had pointed earlier:
>> On the one hand:
>>    If the Destination MAC of the inner Ethernet frame doesn't
>>    match any of VTEP's MAC addresses, then the processing of the
>>    received VXLAN packet MUST follow the procedures described in
>>    Section 4.1 [RFC7348].
>> To which we add:
>>    BFD Control packets with unknown MAC address
>>    MUST NOT be forwarded to VMs.
>> But the unknown MACs are treated as BUM according to the last paragraph in Section 4.2 of RFC 7348:
>>    Note that multicast frames and "unknown MAC destination" frames are
>>    also sent using the multicast tree, similar to the broadcast frames.
>> In light of that, can this draft require that BFD packets with unknown MAC be dropped and not flooded over the corresponding to the VNI domain? I think that in addition to moving the sentence up the statement must be updated:
>> OLD TEXT:
>>    BFD Control packets with unknown MAC address
>>    MUST NOT be forwarded to VMs.
>> NEW TEXT:
>>    If the BFD session is using the Management VNI (Section 6),
>>    BFD Control packets with unknown MAC address
>>    MUST NOT be forwarded to VMs.
>>  Comments? Suggestions?
>>    -- Jeff