Re: [IPFIX] New Version Notification for draft-trammell-ipfix-tcpcontrolbits-revision-00.txt
Brian Trammell <[email protected]>
| Newsgroups | gmane.ietf.ipfix |
|---|---|
| Message-ID | <[email protected]> |
On 10 Sep 2013, at 10:32 , Lothar Braun <[email protected]> wrote: > Hi, > > I have some clarification questions/comments inline (I'm commenting on version 02) > > >> >> >> IPFIX Working Group B. Trammell >> Internet-Draft ETH Zurich >> Intended status: Informational P. Aitken >> Expires: March 13, 2014 Cisco Systems, Inc >> September 09, 2013 >> >> >> Revision of the tcpControlBits IPFIX Information Element >> draft-trammell-ipfix-tcpcontrolbits-revision-02.txt >> >> Abstract >> >> This document revises the tcpControlBits IPFIX Information Element >> defined in [RFC5102] to reflect changes to the TCP Flags header field >> since [RFC0793]. >> >> Status of This Memo >> >> This Internet-Draft is submitted in full conformance with the >> provisions of BCP 78 and BCP 79. >> >> Internet-Drafts are working documents of the Internet Engineering >> Task Force (IETF). Note that other groups may also distribute >> working documents as Internet-Drafts. The list of current Internet- >> Drafts is at http://datatracker.ietf.org/drafts/current/. >> >> Internet-Drafts are draft documents valid for a maximum of six months >> and may be updated, replaced, or obsoleted by other documents at any >> time. It is inappropriate to use Internet-Drafts as reference >> material or to cite them other than as "work in progress." >> >> This Internet-Draft will expire on March 13, 2014. >> >> Copyright Notice >> >> Copyright (c) 2013 IETF Trust and the persons identified as the >> document authors. All rights reserved. >> >> This document is subject to BCP 78 and the IETF Trust's Legal >> Provisions Relating to IETF Documents >> (http://trustee.ietf.org/license-info) in effect on the date of >> publication of this document. Please review these documents >> carefully, as they describe your rights and restrictions with respect >> to this document. Code Components extracted from this document must >> include Simplified BSD License text as described in Section 4.e of >> the Trust Legal Provisions and are provided without warranty as >> described in the Simplified BSD License. >> >> >> >> >> Trammell & Aitken Expires March 13, 2014 [Page 1] >> Internet-Draft IPFIX tcpControlBits September 2013 >> >> >> 1. Introduction >> >> Octets 12 and 13 of the TCP header encode the data offset (header >> length) in four bits, as well as 12 bits of flags. The least >> significant 6 bits of these were defined in [RFC0793] as URG, ACK, >> PSH, RST, SYN, and FIN for TCP control. Subsequently, [RFC3168] >> defined the CWR and ECE flags for Explicit Congestion Notification >> (ECN) negotiation and signaling; [RFC3540] additionally defined the >> NS flag for the ECN Nonce Sum. >> >> As defined in the IANA IPFIX Information Element Registry >> [IANA-IPFIX], taken from [RFC5102], the tcpControlBits Information >> Element for IPFIX [I-D.ietf-ipfix-protocol-rfc5101bis] only covers >> the original six bits from [RFC0793]. To allow IPFIX to be used to >> measure the use of ECN, and to bring the IPFIX Information Element >> definition in line with the current definition of the TCP Flags >> header field, it is necessary to revise this definition. >> >> The revised definition of the Information Element in Section 2 was >> developed and approved through the IE-DOCTORS process >> [I-D.ietf-ipfix-ie-doctors] in August 2013. Section 5.1 of >> [I-D.ietf-ipfix-ie-doctors] states "This process should not in any >> way be construed as allowing the IE-DOCTORS to overrule IETF >> consensus. Specifically, Information Elements in the IANA IE >> registry which were added with IETF consensus require IETF consensus >> for revision or deprecation". Since the tcpControlBits Information >> Element was defined in [RFC5102], an IETF Proposed Standard, any >> revision of this Information Element definition requires IETF >> Consensus. The publication of this document fulfills that >> requirement. >> >> The following section defines the revised tcpControlBits Information >> Element as in Section 9.1 of [I-D.ietf-ipfix-ie-doctors]. >> >> 2. The tcpControlBits Information Element >> >> ElementId: 6 >> Data Type: unsigned16 >> Data Type Semantics: flags >> Description: TCP control bits observed for the packets of this >> Flow. This information is encoded as a bit field; for each TCP >> control bit, there is a bit in this set. The bit is set to 1 if >> any observed packet of this Flow has the corresponding TCP control >> bit set to 1. The bit is cleared to 0 otherwise. >> >> The values of each bit are shown below, per the definition of the >> bits in the TCP header [RFC0793]: >> >> >> >> >> Trammell & Aitken Expires March 13, 2014 [Page 2] >> Internet-Draft IPFIX tcpControlBits September 2013 >> >> >> MSb LSb >> 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 >> +---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+ >> | | | N | C | E | U | A | P | R | S | F | >> | Zero | Future | S | W | C | R | C | S | S | Y | I | >> | (Data Offset) | Use | | R | E | G | K | H | T | N | N | >> +---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+---+ >> >> bit flag >> value name description >> ------+-----+------------------------------------- >> 0x8000 Zero (see tcpHeaderLength) >> 0x4000 Zero (see tcpHeaderLength) >> 0x2000 Zero (see tcpHeaderLength) >> 0x1000 Zero (see tcpHeaderLength) >> 0x0800 Future Use >> 0x0400 Future Use >> 0x0200 Future Use >> 0x0100 NS ECN Nonce Sum >> 0x0080 CWR Congestion Window Reduced >> 0x0040 ECE ECN Echo >> 0x0020 URG Urgent Pointer field significant >> 0x0010 ACK Acknowledgment field significant >> 0x0008 PSH Push Function >> 0x0004 RST Reset the connection >> 0x0002 SYN Synchronize sequence numbers >> 0x0001 FIN No more data from sender >> >> >> As the most significant four bits of octets 12 and 13 of the TCP >> header [RFC0793] are used to encode the TCP data offset (header >> length), the corresponding bits in this IE must be exported as >> zero and must be ignored by the collector; use the tcpHeaderLength >> Information Element to encode this value. > > Are there good reasons for the two "must" in this sentence? Well, we'd really like to enable the following code on the first packet (assuming you have an alignment-friendly architecture): tcp_control_bits = ntohl((uint16_t *)(tcp_hdr_base + 12)) & 0x0FFF; and the following on subsequent packets: tcp_control_bits |= ntohl((uint16_t *)(tcp_hdr_base + 12)) & 0x0FFF; > (And shouldn't this be a MUST if this really should be zero?). This section doesn't use 2119 language because it's meant to be inserted into the registry, where, IIRC, 2119 isn't explicitly in effect. > We could, at some point in the future, decide to use those four bits for something else. I'm not sure how you'd get such a proposal past the IE doctors :). The other header-field-like IEs in the information model are clearly taken from a portion of the packet, so an IE that used the top four bits for something else would break that precedent. > This happened for example with the CWR and ECE bits of the previous definition that had a requirement to be zero. Later on, it turned out that those bits had to be used for encoding something that the original definition hadn't thought about. Right, but in that case, the bits were reserved for future use in TCP, not in IPFIX. The mistake we made in the 5101 revision of this IE is referencing 793 only, and not the updates to it. (This is why there's a difference between Zero and Future Use in this IE definition). > I understand that these four bits are not likely to be used in the TCP header length to encode any useful information for flow data, but we might decide to use those four bits for encoding other information, e.g. some more information about the TCP connection state (or whatever else we might think about). I'm a big fan of "each IE means one thing", even at the expense of efficiency, especially as these IEs are beginning to see use outside IPFIX. But it's entirely possible if unlikely that at some point in the future we'll reuse these four bits for something else, at which point the solution is a matter for the people writing the redefinition, and will probably look a lot like what we did for the ECN flag bits. > What I would like to have would be some text that sets the bits to reserved and says something like: > > "Bits 0x1000 through 0x.8000 are reserved for future use. A collector must not assume to receive information about the TCP header length; the field tcpHeaderLength Information Element should be used to encode this value." > > What do you think? Again, I'm not sure that I'd want to explicitly encourage such reuse. > >> Each of the three future use bits (0x800, 0x400, and 0x200) should >> be exported as one if the corresponding bit is observed in the TCP >> headers of the packets of this Flow, as they may be subsequent to >> a future update of [RFC0793]. > > What does "as one" in "exported AS ONE if" mean? Does this mean that these bits must be exported as seen in the packet headers? > >> If exported as a single octet with reduced length encoding, > > Nit: RFC 5101 calls it reduced size encoding (at least most of the time ;)) Yeah, it turns out I'm the only one who calls it reduced-length. Will fix, thanks. >> this >> Information Element covers the low-order octet of this field (i.e, >> bits 0x80 to 0x01), omitting the ECN Nonce Sum and the three >> Future Use bits. A collector receiving this Information Element >> with reduced length encoding must not assume anything about the >> content of these four bits. > > I just wanted to raise this point (without objecting). This would clearly be an exception to what I would expect from reduced size encoding in RFC 5101bis: > > The reduction in size can be to any number of > octets smaller than the original type if the data value still fits, > i.e., so that only leading zeroes are dropped. > > Whenever I see reduced size encoding, I assume that there is no further information in the field and that all leading bits are zero because of this text. But I would assume this is not a hard requirement from RFC 5101(bis), and it is probably ok if the data type has an explicit definition that states that the first bits are not zero. Well, it could mean two things: the top bits are zero because the Nonce Sum bit was never set on any packet in the flow (which, given studies on the usage of ECN, and anecdotes about ECN Nonce deployment, I would say is the case 99.99...% of the time), or because it was not observed (since nobody uses ECN nonce, nobody measures it, either). Perhaps we should state that unsigned16 encoding should ONLY be used by EPs connected to MPs which actually know how to measure at least one of the bits, so that 0s here won't be interpreted as actual assertions of no flag present. Cheers, Brian _______________________________________________ IPFIX mailing list [email protected] https://www.ietf.org/mailman/listinfo/ipfix