Re: [ippm] Benjamin Kaduk's Discuss on draft-ietf-ippm-stamp-07: (with DISCUSS and COMMENT)

[Greg Mirsky <[e-mail]>]

Hi Benjamin,
thank you for your thorough review and detailed comments. Please find
answers, notes, and the proposed updates below in-line tagged GIM>>.
I much appreciate your feedback, suggestions to address your concerns.

Regards,
Greg

On Wed, Sep 4, 2019 at 5:50 PM Benjamin Kaduk via Datatracker <
[e-mail]> wrote:

> Benjamin Kaduk has entered the following ballot position for
> draft-ietf-ippm-stamp-07: Discuss
>
> When responding, please keep the subject line intact and reply to all
> email addresses included in the To and CC lines. (Feel free to cut this
> introductory paragraph, however.)
>
>
> Please refer to https://www.ietf.org/iesg/statement/discuss-criteria.html
> for more information about IESG DISCUSS and COMMENT positions.
>
>
> The document, along with other ballot positions, can be found here:
> https://datatracker.ietf.org/doc/draft-ietf-ippm-stamp/
>
>
>
> ----------------------------------------------------------------------
> DISCUSS:
> ----------------------------------------------------------------------
>
> We don't ever clearly state that the protocol allows for packet sizes
> other than the listed 44- and 112-octet variants, that content larger
> than that is to be treated as padding unless directed otherwise by
> configuration, that the reflected packet must be the same size as the
> incoming packet, and how a Session-Reflector should set any such padding
> that it needs to add in order to produce a same-sized packet.
>
GIM>> We had discussed this and the current working version of the draft in
Section 4.2 refers to the STAMP Optional Extensions
<https://datatracker.ietf.org/doc/draft-ietf-ippm-stamp-option-tlv/> draft:
   STAMP supports symmetrical test packets.  The base STAMP Session-
   Sender packet has a minimum size of 44 octets in unauthenticated
   mode, see Figure 2, and 112 octets in the authenticated mode, see
   Figure 4.  The variable length of a test packet in STAMP is supported
   by using Extra Padding TLV defined in
   [I-D.ietf-ippm-stamp-option-tlv].
As discussed in Section 4.6 Interoperability with TWAMP Light, TWAMP Light
Session-Reflector will treat STAMP optional extensions as Padding and, if
configured to symmetrical packet size mode, will respond with Padding as
per RFC 6038. This draft defines the use of only base STAMP packets and the
discussion of all extensions is in the draft-ietf-ippm-stamp-option-tlv.

>
> This document hardcodes the truncated HMAC-SHA-256 algorithm.  Per BCP
> 201, what is the procedure for cryptographic algorithm agility?
>
GIM>> Support of other cryptographic algorithms is important but the WG
agreed that in this specification only the use of HMAC-SHA-256 is defined.
Future specifications may define the use of other, more advanced
cryptographic algorithms, possibly providing an update to the STAMP YANG
data model <https://datatracker.ietf.org/doc/draft-ietf-ippm-stamp-yang/>.

>
> Please also consider the discussion in BCP 107 about key lifecycles and
> key management, including whether it is appropriate to use a
> key-derivation function to produce short-term (e.g., per flow) keys from
> a long-lived key (e.g., one fixed in static configuration).
>
GIM>> In the course of the discussion, we've clarified in the section
Integrity Protection in STAMP that:
   HMAC uses its own key, and the definition of the
   mechanism to distribute the HMAC key is outside the scope of this
   specification.  One example is to use an orchestrator to configure
   HMAC key based on STAMP YANG data model [I-D.ietf-ippm-stamp-yang].

>
> What is the input plaintext to the HMAC computation?  In the case of
> future extensions, does the HMAC field remain at its current fixed
> offset in the packet or move to always be the last 16 octets?  Is any
> additional padding/TLV content protected by the HMAC?
>
> What error does the error estimate ... estimate?
> Clock skew between sender and receiver?
>
GIM>> The Error Estimate field has been originally defined in RFC 4656
One-Way Active Measurement Protocol. One flag (S) indicates whether the
originator of the timestamp has clock synchronized to UTC (GPS, NTP or
PTP). Other fields can be used to express the error estimate of the
timestamping process.

>
> I think we need to require some level of cryptographic protection
> whenever control information is included in a Session-Sender's test
> packet.  That is, that a Session-Reflector MUST NOT act on control
> information received in unauthenticated packets.  (That said, this
> document itself does not describe a way to include control information,
> so perhaps the note about "optional control information communicated in
> the Session-Sender's test packet" in Section 4 is misplaced.
>
GIM>> Thank you for catching this. Clearly, it must be removed:
OLD TEXT:
   STAMP Session-Reflector receives Session-Sender's packet and acts
   according to the configuration and optional control information
   communicated in the Session-Sender's test packet.
NEW TEXT:
   STAMP Session-Reflector receives Session-
   Sender's packet and acts according to the configuration.

In Section 4.2.1:
>
>    o  Timestamp and Receiver Timestamp fields are each eight octets
>       long.  The format of these fields, NTP or PTPv2, indicated by the
>       Z flag of the Error Estimate field as described in Section 4.1.
>
> I think you need to explicitly say that "Timestamp" is echoed from the
> received packet and "Receiver Timestamp" is determined locally as close
> to (reciept? transmission?) as possible.
>
GIM>> You've helped find a typo that makes the name of the field confusing.
The field is tagged correctly in Figure 5 - Receive Timestamp. In fact, the
Receive Timestamp is also the local to the Session-Reflector. It is the
time value the Reflector received the STAMP test packet. The value in the
Timestamp field is taken at the transmission of the reflected packet. The
Sender Timestamp field is a copy of the Timestamp field in the
Session-Sender's test packet. I propose the update as follows:
OLD TEXT:
   o  Timestamp and Receiver Timestamp fields are each eight octets
      long.  The format of these fields, NTP or PTPv2, indicated by the
      Z flag of the Error Estimate field as described in Section 4.1.
NEW TEXT:
   o  Timestamp and Receive Timestamp fields are each eight octets long.
      The format of these fields, NTP or PTPv2, indicated by the Z flag
      of the Error Estimate field as described in Section 4.2.  Receive
      Timestamp is the time the test packet was received by the Session-
      Reflector.  Timestamp - the time taken by the Session-Reflector at
      the start of transmitting the test packet.

>
> I think we need greater clarity on whether the normative statements in
> Section 4.4 apply only to STAMP peers that are aware they are
> interacting with TWAMP Light, or apply to all STAMP peers (see Comment
> for further discussion on why the current text seems internally
> inconsistent).
>
>
> In Section 4.1.1:
>
>    o  Timestamp is eight octets long field.  STAMP node MUST support
>       Network Time Protocol (NTP) version 4 64-bit timestamp format
>       [RFC5905], the format used in [RFC5357].  STAMP node MAY support
>       IEEE 1588v2 Precision Time Protocol truncated 64-bit timestamp
>       format [IEEE.1588.2008], the format used in [RFC8186].
>
> I think a note that which one is in use will be configured by the
> configuration/management function is in order.  Except that the Z bit
> below confuses things terribly...
>
>       The STAMP Session-Sender and Session-Reflector MAY use, not use,
>       or set value of the Z field in accordance with the timestamp
>       format in use.  This optional field is to enhance operations, but
>       local configuration or defaults could be used in its place.
>
> ... since, as noted by the secdir reviewer, this line just confuses
> everything.  Either keep the "must be zero" semantics of 4656 or the
> "MUST match reality" semantics of 8186, but this middle case is actively
> harmful.
>
GIM>> As result of the discussion, this text is changed to:
NEW TEXT:
      The STAMP Session-Sender and Session-Reflector MUST use the NTP 64
      bit format of a timestamp (Z field value of 0).  as the default.
      A configuration/management function MAY configure STAMP Session-
      Sender and Session-Reflector to using the PTPv2 truncated format
      of a timestamp (Z field value of 1).
Hope it is clearer now.

>
> (I also support Barry and Magnus' Discusses.)
>
GIM>> It took some time to address them.

>
>
> ----------------------------------------------------------------------
> COMMENT:
> ----------------------------------------------------------------------
>
> Section 1
>
> I'll note several grammar nits, inline, though perhaps some of them will
> not apply after the rewrite in response to the secdir review:
>
>    Development and deployment of Two-Way Active Measurement Protocol
>
> "the Two-Way"
>
GIM>> Applied, thank you.

>
>    (TWAMP) [RFC5357] and its extensions, e.g., [RFC6038] that defined
>    features such as Reflect Octets and Symmetrical Size for TWAMP
>
> comma after TWAMP
>
GIM>> Done.

>
>    provided invaluable experience.  Several independent implementations
>    exist, have been deployed and provide important operational
>    performance measurements.  At the same time, there has been
>    noticeable interest in using a more straightforward mechanism for
>    active performance monitoring that can provide deterministic behavior
>    and inherit separation of control (vendor-specific configuration or
>
> "inherit" from what?
>
GIM>> Right, should have been "inherent". Now in the working version.

>
>    orchestration) and test functions.  One of such is Performance
>
> "One such mechanism is"
>
GIM>> This passage updated to:
   Recent work on IP Edge to Customer Equipment using TWAMP Light from
   Broadband Forum [BBF.TR-390] demonstrated that interoperability among
   implementations of TWAMP Light is challenged because the composition
   and operation of TWAMP Light were not sufficiently specified in
   [RFC5357].

>
>    Measurement from IP Edge to Customer Equipment using TWAMP Light from
>    Broadband Forum [BBF.TR-390] used as the reference TWAMP Light that,
>
> I'm not sure what the intent here is, but maybe ", which is used as the
> reference TWAMP Light".
>
GIM>> Replaced by the sentence I've copied above.

>
>    according to [RFC8545], includes sub-set of TWAMP-Test functions in
>
> I'd also suggest starting a new sentence for "According to [RFC8545]"
> (and adding the then-needed "this" and "a" for "this includes a").
>
GIM>> Re-worded as follows:
   According to [RFC8545], TWAMP Light includes sub-set of
   TWAMP-Test functions to provide comprehensive solution requires
   support by other applications that provide, for example, control and
   security.


>
>    combination with other applications that provide, for example,
>    control and security.  This document defines an active performance
>    measurement test protocol, Simple Two-way Active Measurement Protocol
>    (STAMP), that enables measurement of both one-way and round-trip
>    performance metrics like delay, delay variation, and packet loss.
>
> I agree with the secdir reviewer that the relationship between STAMP and
> TWAMP Light could be much more clear.
>
GIM>> The new paragraph at the closing of the Introduction section:
   This document defines an active performance measurement test
   protocol, Simple Two-way Active Measurement Protocol (STAMP), that
   enables measurement of both one-way and round-trip performance
   metrics like delay, delay variation, and packet loss.  Some TWAMP
   extensions, e.g., [RFC7750] are supported by the extensions to STAMP
   base specification in [I-D.ietf-ippm-stamp-option-tlv].

>
> Section 2.1
>
>    MBZ May be Zero
>
> I commonly see this expand to "Must be zero"; requiring the sender to
> not set any bits seems more likely to preserve the ability to use the
> field for future extensibility, since a recipient that sees a nonzero
> bit knows it was consciously set (i.e., with intent to use the
> extension) rather than inadvertently set by someone expecting it to be
> ignored.
> (Also, if the bits are covered under the HMAC, then the recipient can't
> actually ignore them, since they have to be used to verify the HMAC.)
>
GIM>> Changed MBZ full form to the Must-be-zero. Named padding fields in
unauthenticated mode - Reserved. Would that be acceptable?

>
> Section 3
>
>    be achieved through various means.  Command Line Interface, OSS/BSS
>    (operations support system/business support system as a combination
>    of two systems used to support a range of telecommunication services)
>    using SNMP or controllers in Software-Defined Networking using
>    Netconf/YANG are but a few examples.
>
> nit: if "using SNMP or controllers[...]" is supposed to be separate from
> "OSS/BSS", then some additional punctuation/conjunction is needed.
>
GIM>> Also re-worded as:
   The configuration and management of the STAMP Session-
   Sender, Session-Reflector, and management of the STAMP sessions are
   outside the scope of this document and can be achieved through
   various means.  A few examples are:  Command Line Interface,
   telecommunication services' OSS/BSS systems, SNMP, and Netconf/YANG-
   based SDN controllers.

>
> Section 4
>
>    number.  A STAMP implementation of Session-Sender MUST be able to use
>    UDP port numbers from User, a.k.a.  Registered, Ports and Dynamic,
>    a.k.a.  Private or Ephemeral, Ports ranges defined in [RFC6335].
>
> Able to use as source, destination, or both?  (We just talked about
> destination but not source in the previous sentence.)
>
GIM>> The text is now in Section 4.1. Will clarify that it applies to the
destination port:
   A STAMP implementation of Session-Sender MUST be able to use as the
   destination UDP port numbers from User, a.k.a.  Registered, Ports and
   Dynamic, a.k.a.  Private or Ephemeral, Ports ranges defined in
   [RFC6335].

>
> Section 4.1
>
>    Because STAMP supports symmetrical test packets, STAMP Session-Sender
>    packet has a minimum size of 44 octets in unauthenticated mode, see
>    Figure 2, and 112 octets in the authenticated mode, see Figure 4.
>
> nit: I don't see how merely "support"ing (as opposed to "require"ing or
> "use"ing) symmetrical packets implies these minimum packet sizes.  (That
> is, I find the word "because" unjustified absent some statement that
> requires the Session-Reflector packets to be that size and a requirement
> for the symmetry is present.)
>
GIM>> The use of the symmetrical test packets is the default behavior:
NEW TEXT:
   A STAMP Session-Reflector supports symmetrical size of test packets
   [RFC6038] as the default behavior.  Because of that, the base STAMP
   Session-Sender packet has a minimum size of 44 octets in
   unauthenticated mode, see Figure 2, and 112 octets in the
   authenticated mode, see Figure 4.  The variable length of a test
   packet in STAMP is supported by using Extra Padding TLV defined in
   [I-D.ietf-ippm-stamp-option-tlv].

>
> Section 4.2
>
>       That implies that the STAMP Session-Reflector MUST keep a state
>       for each accepted STAMP-test session, uniquely identifying STAMP-
>       test packets to one such session instance, and enabling adding a
>       sequence number in the test reply that is individually incremented
>       on a per-session basis.
>
> How does it "accept a STAMP-test session"?
>
GIM>> Would s/accepted/configured/ work?

>
> Section 4.2.1
>
>       *  in the stateful mode the Session-Reflector counts the received
>          STAMP test packets in each test session and uses that counter
>          to set the value of the Sequence Number field.
>
> Should we say anything about whether the initial sequence number (having
> received one packet from the Session-Sender) is zero or one?
>
GIM>> In the description of the format of the Session-Sender
unauthenticated test packet stated:
   o  Sequence Number is four octets long field.  For each new session
      its value starts at zero and is incremented with each transmitted
      packet.
Will add similar note for the Session-Reflector:
OLD TEXT:
      *  in the stateful mode the Session-Reflector counts the received
         STAMP test packets in each test session and uses that counter
         to set the value of the Sequence Number field.
NEW TEXT:
      *  in the stateful mode, the Session-Reflector counts the
         transmitted STAMP test packets.  It starts with zero and is
         incremented by one for each subsequent packet for each test
         session.  The Session-Reflector uses that counter to set the
         value of the Sequence Number field.

>
> Section 4.2.2
>
>                                                               Also,
>    STAMP Session-Reflector test packet format in authenticated mode
>    includes a key (HMAC) ([RFC2104]) hash at the end of the PDU.  The
>    detailed use of the HMAC field is in Section 4.3.
>
> nit: "keyed"
>
GIM>> Done, thank you

>
> Section 4.3
>
> I think we should have a statement about HMAC key (non-)reuse across
> separate measurement sessions.
>
> I agree with the secdir reviewer that the confidentiality protection
> seems like something that would be done at a "lower" level, not a
> "higher" level.
>
GIM>> Resulting from our discussion with SecDir, the following text is now
in the Integrity Protection in STAMP section:
 HMAC uses its own key; key management and the
   mechanisms to distribute the HMAC key is outside the scope of this
   specification.  One example is to use an orchestrator to configure
   HMAC key based on STAMP YANG data model [I-D.ietf-ippm-stamp-yang].
Would you suggest additional text or an update?

>
> Section 4.4
>
>    In the former case, the Session-Sender MAY not be aware that its
>
> It's unclear that this "MAY" is normative as opposed to descriptive.
>
GIM>> Yes, it should be in descriptive form s/MAY/might/

>
>    Session-Reflector does not support STAMP.  For example, a TWAMP Light
>    Session-Reflector may not support the use of UDP port 862 as defined
>    in [RFC8545].  Thus STAMP Session-Sender MAY use port numbers as
>    defined in Section 4.  If any of STAMP extensions are used, the TWAMP
>    Light Session-Reflector will view them as Packet Padding field.  The
>    Session-Sender SHOULD use the default format for its timestamps -
>    NTP.  And it MAY use PTPv2 timestamp format.
>
> Given the above note about not knowing that the peer is TWAMP Light vs.
> STAMP, it seems that this SHOULD/MAY apply to all STAMP implementations,
> not just ones that are interacting with TWAMP Light.  Which in turn might
> suggest that the normative statements are best made in a different
> section.
> (Also (nit), where do we say that NTP is the default format?)
>
GIM>> We've clarified the default format for timestamp when addressing
other review comments. Now the draft states in Section 4.2.1:
      The STAMP Session-Sender and Session-Reflector MUST use the NTP 64
      bit format of a timestamp (Z field value of 0).  as the default.
And, as I've mentioned in response to the question above, the draft
clarifies for PTPv2 format:
      A configuration/management function MAY configure STAMP Session-
      Sender and Session-Reflector to using the PTPv2 truncated format
      of a timestamp (Z field value of 1).
I hope it is not seen as duplication and the message is consistent.


>
>    In the latter scenario, if a TWAMP Light Session-Sender does not
>    support the use of UDP port 862, the test management system MUST set
>    STAMP Session-Reflector to use UDP port number as defined in
>    Section 4.  If the TWAMP Light Session-Sender includes Packet Padding
>    field in its transmitted packet, the STAMP Session-Reflector will
>    return the reflected packet of the symmetrical size if the size of
>    the received test packet is larger than the size of the STAMP base
>    packet.  The Session-Reflector MUST be set to use the default format
>    for its timestamps, NTP.
>
> On the other hand, if we take the same approach here, and assume that
> the Session-Reflector may not know that the Session-Sender is TWAMP
> Light vs. STAMP, then this MUST would seem to always apply, and thus
> prevent the Session-Reflector from ever using the PTPv2 timestamp
> format, in which case the text related to its doing so is "dead code"
> and should be removed to avoid confusion.
>
GIM>> When we say in the draft that a Session-Sender or Session-Reflector
"know" something, we imly that that is known to an operator, the one who
configures, manages the test session. If both entities support STAMP, then
the test session may be instantiated using Netconf/YANG and use PTPv2
format. If only one entity is STAMP-based, then operator may assume that
the remote node only supprots STAMP and set its system to use NTP format.
Do you see that reasonable?

>
> Section 8.2
>
> RFC 2104 needs to be a normative reference.  The truncation of the HMAC
> is simple enough that we probably don't need to consider RFC 4868
> normative just for it, though.
>
GIM>> Agreed and moved to the Normative list though it causes Downref:
 ** Downref: Normative reference to an Informational RFC: RFC 2104