Re: expired SIGs in an active zone

Paul Vixie <paul@vix.com> Fri, 08 February 2002 17:48 UTC

Received: from psg.com (exim@psg.com [147.28.0.62]) by ietf.org (8.9.1a/8.9.1a) with ESMTP id MAA17096 for <dnsext-archive@lists.ietf.org>; Fri, 8 Feb 2002 12:48:06 -0500 (EST)
Received: from lserv by psg.com with local (Exim 3.33 #1) id 16ZEvw-000IFO-00 for namedroppers-data@psg.com; Fri, 08 Feb 2002 09:35:36 -0800
Received: from as.vix.com ([204.152.187.70]) by psg.com with esmtp (Exim 3.33 #1) id 16ZEvv-000IEw-00 for namedroppers@ops.ietf.org; Fri, 08 Feb 2002 09:35:35 -0800
Received: from as.vix.com (localhost [127.0.0.1]) by as.vix.com (Postfix) with ESMTP id AACC328EB3 for <namedroppers@ops.ietf.org>; Fri, 8 Feb 2002 09:35:28 -0800 (PST) (envelope-from vixie@as.vix.com)
From: Paul Vixie <paul@vix.com>
To: DNSEXT WG Mailing list <namedroppers@ops.ietf.org>
Subject: Re: expired SIGs in an active zone
In-Reply-To: Message from "Scott Rose" <scottr@antd.nist.gov> of "Fri, 08 Feb 2002 11:00:38 EST." <003201c1b0b9$c08d2b40$b9370681@antd.nist.gov>
X-Mailer: mh-e 6.0; nmh 1.0.4; Emacs 21.1
Date: Fri, 08 Feb 2002 09:35:28 -0800
Message-Id: <20020208173528.AACC328EB3@as.vix.com>
Sender: owner-namedroppers@ops.ietf.org
Precedence: bulk

> Def:  An expired SIG is a SIG record who's "expiration" section is in the
> past. not who's TTL has expired.

fine, so far.  but after that we get into several kinds of disagreement:

> -When dealing with expired SIG records in an authoritative zone, the server
> SHOULD drop them from the zone when the SIG expires.  This entails the
> server periodically checking SIG expiration times.

this is just impossible.  authority servers do not auto-edit zones.  if you
really mean this then you'll have to structure it as a dynamic update and
increment the serial number and show the change in IXFRs and so forth.  but
i for one would never, ever, (ever) turn this functionality on in my servers.

> - Servers should avoid placing expired SIGs in DNS responses.  NOTE:
> <- this isn't a SHOULD NOT since it might prove more difficult for
> servers to implement than resolvers to ignore expired SIGs Some other
> warning/on line signing might occur, but that is not a protocol issue,
> but an operational one.

if you really want this behaviour then you'll have to ratchet down the
TTL as the expiry gets closer in time.  otherwise you'll have cached SIGs
"floating around" whose signatures are expired but not their DNS TTL's,
and you won't have solved whatever problem you're worried about.  but if
you plan to ratchet down your TTL's in authority responses, then every
time you do it you'll have changed the zone's identity, and you'll have
to autoincrement the serial number and show this delta in IXFR's.

i went through all of this logic for a different purpose (see attached)
and was told by the committee that my draft had completely proved that the
implementation complexity was just unimaginably higher than anybody had
thought.

> Open question:  Should deleting expired SIG records require an updating of
> the zone serial number?  (I think "no" personally, but there are open
> issues).
> 
> So this is another debate for those of us that have tired of the Opt-In
> debate (and give the Verisign guys a chance to catch their breaths ;-)

i've got to say that last time i participated in a secret-handshake meeting
in a smoke filled back room that had the result of making a recommendation
to the namedroppers committee, we did a better job explaining our motives
in public than you've done here.  what are the motives?  2535 requires the
clients and servers to have securely synchronized clocks.  that means a
client, of either direct authoritative data or recursively cached data,
will know whether the signature is expired or not.  why are you protecting
the client from something it HAS to be able to deal with ANYWAY?

btw, here's another copy of DEFUPD for those whose nightmares have stopped:

--------





   DNSIND Working Group                                    Paul Vixie (ISC)
   INTERNET-DRAFT                                                  May 1996
   <draft-ietf-dnsind-defupd-00.txt>

   Amends: RFC 1035, [UPDATE]


       Deferred Dynamic Updates in the Domain Name System (DNS DEFUPD)


   Status of this Memo

      This document is an Internet-Draft.  Internet-Drafts are working
      documents of the Internet Engineering Task Force (IETF), its areas,
      and its working groups.  Note that other groups may also distribute
      working documents as Internet-Drafts.

      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.''

      To learn the current status of any Internet-Draft, please check the
      ``1id-abstracts.txt'' listing contained in the Internet-Drafts Shadow
      Directories on ftp.is.co.za (Africa), nic.nordu.net (Europe),
      munnari.oz.au (Pacific Rim), ds.internic.net (US East Coast), or
      ftp.isi.edu (US West Coast).


   Abstract

      Not all applications that perform dynamic updates using the protocol
      specified in [UPDATE] want their updates applied immediately.  A case
      in point is [DHCP], wherein the DHCP lease time should control the
      lifetime of associated DNS data even if the DHCP client or server is
      not available at the time the DHCP lease expires.

      The essence of this proposal is that DNS Dynamic Updates should be
      deferrable for some time delay period, after which they will be
      executed normally.  Furthermore, RRs added or updated by a deferred
      update can have expiration times specified for them, as enforced by
      the automatic Dynamic Updates.  Automatic serial number changes (as
      in [UPDATE]), dynamic zone slave notification (see [NOTIFY]) and
      incremental zone transfer (see [IXFR]) will jointly see to it that
      the zone changes are propagated with expedience.



   Expires November 1996                                          [Page 1]

   INTERNET-DRAFT                 DNS DEFUPD                       May 1996


   1 - New Assigned Numbers

      Opcode:   DEFUPD (6?)
      RRtype:   TUU (34?)
      RRtype:   TUE (35?)


   2 - Message Format

   The format and encoding of a DEFUPD is identical to that of UPDATE as
   defined in [UPDATE 2], except that the Opcode is DEFUPD rather than
   UPDATE, and there are two new RR types that can be used in the
   Additional Data section.

   2.1 - Additional Data Section: TUU RR

   In addition to the optional uses described in [UPDATE 2.6], a DEFUPD
   request's Additional Data section can include a TUU (Time Until Update)
   RR as follows:

      Owner:      same as ZNAME (see [UPDATE 2.3])
      Class:      same as ZCLASS (see [UPDATE 2.3])
      Type:       TUU (new RRtype for this protocol)
      TTL:        deferral time, relative, in seconds
      RDLENGTH:   0
      RDATA:      empty

   Of particular note is the TTL, which contains the relative time delay,
   in seconds, starting from the time this DEFUPD is received by the
   primary master, before operations contained in the Update Section (see
   [UPDATE 2.5]) will actually be performed.

















   Expires November 1996                                          [Page 2]

   INTERNET-DRAFT                 DNS DEFUPD                       May 1996


   2.2 - Additional Data Section: TUE RR

   In addition to the optional uses described in [UPDATE 2.6], a DEFUPD
   request's Additional Data section can include a TUU (Time Until Expiry)
   RR as follows:

      Owner:      same as ZNAME (see [UPDATE 2.3])
      Class:      same as ZCLASS (see [UPDATE 2.3])
      Type:       TUE (new RRtype for this protocol)
      TTL:        expiry time, relative, in seconds
      RDLENGTH:   0
      RDATA:      empty

   Of particular note is the TTL, which contains the expiration time delay,
   in seconds, starting from the time this DEFUPD is received by the
   primary master, of all RRs added or updated by operations in the Update
   Section (see [UPDATE 2.5]).

   3 - Server Behavior

   A server, upon receiving a DEFUPD request, will first scan the request's
   Additional Data section in search of TUU or TUE RRs.  If no RRs of
   either type TUU or TUE are found, then this request will be processed as
   a normal UPDATE with no special behaviour.  If any TUU or TUE RRs are
   found, then processing continues as follows.

   3.1 - Verify TUU RR

   If any TUU RRs are found in the Additional Data section, this update
   will be processed with Deferral as explained below.  If more than one
   TUU RR is found, signal FORMERR to requestor.  The TUU RR's owner name
   and class are compared to ZNAME and ZCLASS; if a mismatch occurs, signal
   FORMERR to requestor.  The TUU RR's RDLENGTH/RDATA is ignored by
   implementations of this specification, but future specifications may
   make use of this field.

   3.2 - Verify TUE RR

   If any TUE RRs are found in the Additional Data section, this update
   will be processed with Expiry as explained below.  If more than one TUE
   RR is found, signal FORMERR to requestor.  The TUE RR's owner name and
   class are compared to ZNAME and ZCLASS; if a mismatch occurs, signal
   FORMERR to requestor.  The TUE RR's RDLENGTH/RDATA is ignored by
   implementations of this specification, but future specifications may
   make use of this field.



   Expires November 1996                                          [Page 3]

   INTERNET-DRAFT                 DNS DEFUPD                       May 1996


   3.3 - Deferral

   If a TUU RR was specified in the Additional Data section, this update
   will be processed with Deferral.  Deferral means that the update will
   not be applied synchronously to the requestor's transaction cycle, but
   instead will be applied asynchronously at some potentially later time.
   The delay period is measured in seconds and expressed in the TUU's TTL.

   3.3.1 - Store Deferred Update

   Subject to per-server, per-zone, and per-RRset quotas, this UPDATE
   message is stored, persistently, on the name server.  If per-RRset
   quotas are implemented, it is recommended that a DEFUPD ``count
   against'' all RRsets mentioned in the Update Section.  If an
   implementation defined quota is exceeded by this deferred update, or if
   persistent storage is unavailable, signal SERVFAIL to requestor (leaving
   the zone in its former state).  Note that even a deferred update whose
   TUU's TTL is zero (0), specifying immediate application, should be
   subject to quotas if the name server implements quotas.

   3.3.2 - Send Early Response

   Signal NOERROR to requestor.

   3.3.3 - Apply Deferred Update

   When a period of time equal to or greater than the TUU's TTL (measured
   in seconds) has elapsed since a DEFUPD was first received at the primary
   master, the DEFUPD message is applied to the zone as an UPDATE would be,
   except that no error can be signalled to the requestor.  Thus, all
   errors not found and reported at the time the DEFUPD request was
   received are silent, affecting only the continued processing of the
   deferred update.  Note that all sections are processed, including those
   processed before the deferred update were stored.  Thus, prerequisites
   must hold before and after the deferral period.













   Expires November 1996                                          [Page 4]

   INTERNET-DRAFT                 DNS DEFUPD                       May 1996


   3.4 - Expiry Processing

   When a DEFUPD is applied, either during the requestor's transaction
   cycle or following the optional Deferral period, the inclusion of a TUE
   RR in the Additional Data section will cause this update to be processed
   with Expiry.

   Expiry as expressed in the TUE's TTL is the time, in seconds, before all
   RRs added or modified by the Update Section will be automatically
   deleted by the primary master server.  This time is relative to the time
   the DEFUPD message is processed, which might be after the delay period
   specified by a TUU RR.

   3.4.1 - Initial TTL Limits

   The TTL of all added or updated RRs in the Update Section will be
   maximized silently to one half of the Expiry time.  This will cause
   downstream caching name servers to purge RRsets containing this RR at
   least once before expiry.

   3.4.2 - TTL Half Life

   Each time an RR's expiry reaches half of its previous value, that RR's
   TTL will be reduced to half of its previous value, until the TTL reaches
   a value less than or equal to sixty (60), i.e., one minute of real time,
   at which time the TTL will not be automatically reduced further by the
   primary master.  It should be noted that RRs held in a server's memory
   need not be swept for half life processing, as long as the TTL changes
   appear when the RR next becomes externally visible, and as long as the
   ``zone has changed'' processing (see below) is done at the time of the
   half life expiration.

   Whenever the TTL is automatically reduced by this process, the zone will
   be considered ``changed'' for the purpose of automatic SOA SERIAL
   increment (see [UPDATE 3.6]) and real time zone slave notification (see
   [NOTIFY]).












   Expires November 1996                                          [Page 5]

   INTERNET-DRAFT                 DNS DEFUPD                       May 1996


   3.4.3 - Automatic Deletion

   When the time since an RR was added or updated by a DEFUPD with Expiry
   exceeds the TUE's TTL as specified in that update, all RRs added or
   updated by that DEFUPD's Update Section will be automatically deleted by
   the primary master.  This deletion can be deferred until the deleted RRs
   would next become visible, so long as the ``zone has changed''
   processing (see below) is done at the time of expiration (i.e., when the
   automatic deletion is first deferred.)

   Whenever automatic deletion occurs, the zone will be considered
   ``changed'' for the purpose of automatic SOA SERIAL increment (see
   [UPDATE 3.6]) and real time zone slave notification (see [NOTIFY]).

   3.5 - Requirements for Persistence

   Stored deferred updates should persist across name server restarts.

   3.5.1 - Restarts and Deferral

   In the event of a name server restart, all deferred updates whose TUU
   has expired must take effect before any network requests are processed
   using data from the affected zone, and before any Expiry processing
   takes place on RRs in the affected zone.

   3.5.2 - Restarts and Expiry

   In the event of a name server restart, all expiries must be checked as
   of the current time before any network responses are generated using
   data from the affected zone.

   If an RR's expiry time has been reached while the name server was not
   running, that RR will be deleted.  Otherwise, the RR's TTL will be set
   to one half of the time remaining before expiration, and half life
   processing as specified in Section 3.4.2 will be restarted.

   If any RR is deleted or if an RR's TTL is changed during startup, then
   the zone will be considered ``changed'' for the purpose of automatic SOA
   SERIAL increment (see [UPDATE 3.6]) and real time zone slave
   notification (see [NOTIFY]).








   Expires November 1996                                          [Page 6]

   INTERNET-DRAFT                 DNS DEFUPD                       May 1996


   4 - Requestor Behaviour

   A requestor will generate and transmit a DEFUPD request as specified in
   [UPDATE 4], except that TUU and TUE RRs can be included in the
   Additional Data section.

   4.1. The TUU RR, if specified, must have an owner name and class equal
   to the ZNAME and ZCLASS (see [UPDATE 2.3]).  The TTL should be set to
   the delay, measured in seconds, before this update should be processed
   by the primary master.  RDLENGTH should be set to 0, and RDATA should
   therefore be empty.

   4.2. The TUE RR, if specified, must have an owner name and class equal
   to the ZNAME and ZCLASS (see [UPDATE 2.3]).  The TTL should be set to
   the delay, measured in seconds, before all RRs added or changed by the
   Update Section will be automatically deleted by the primary master.
   This delay is measured starting from the time the update is applied,
   which could follow a deferral delay if a TUU RR was also included in
   this update.

   5 - Notes on Resource Consumption

   A TUE RR will require the primary master will initiate an automatic
   update approximately O(log2(TTL)) times over the life of an expiring RR.
   Even for massively sized zones, this is not considered an inappropriate
   load on the primary master server itself.

   The network bandwidth consumed due to the use of TUE RRs is more
   noticeable, although for massively sized zones, the bandwidth
   requirements should flatten somewhat as many RRs will be automatically
   updated during any given cycle of NOTIFY and AXFR/IXFR.

   6 - Security Considerations

   This protocol suffers the same abject and intentional lack of security
   as [UPDATE], from which it inherits its basic semantics.  In the absence
   of DNS Secure Updates, this protocol should not be used outside of an
   enterprise network, and only with great care within an enterprise
   network.









   Expires November 1996                                          [Page 7]

   INTERNET-DRAFT                 DNS DEFUPD                       May 1996


   7 - Discussion Items for DNSIND and Namedroppers

   Should the server's response to a DEFUPD include an opaque cookie called
   a ``deferred update ID'' which could be used in future DEFUPD requests
   to cancel or replace a previous deferred update?

   Should automatic updates caused by a TUE RR be required to be batched up
   and processed at some minimum interval?  For example, if we only checked
   for half life and expiration once per minute, this might drastically
   reduce the number of NOTIFY/AXFR/IXFR cycles caused by any given zone.
   We would have to recommend that all zones in a given server not be
   synchronized to the same timer, lest a server with many zones cause all
   of its zones to change and require NOTIFY/AXFR/IXFR in the same second.

   Astute readers will have noticed that this proposal is a precise
   superset of [UPDATE], and by adding the optional behaviour and
   definitions of TUU and TUE to [UPDATE], we could do away with the
   separate Opcode for DEFUPD.  This was only possible up until the time
   [UPDATE] went to proposed standard, but hopefully the intent was clear.

   8 - References

   [RFC1035]
      P. Mockapetris, ``Domain Names - Implementation and Specification,''
      RFC 1035, USC/Information Sciences Institute, November 1987.

   [IXFR]
      M. Ohta, ``Incremental Zone Transfer,'' Internet Draft, February
      1996, <draft-ietf-dnsind-ixfr-06.txt>.

   [NOTIFY]
      P. Vixie, ``A Mechanism for Prompt Notification of Zone Changes,''
      Internet Draft, March 1996, <draft-ietf-dnsind-notify-07.txt>.

   [UPDATE]
      P. Vixie (Ed), et al, ``Dynamic Updates in the Domain Name System,''
      Internet Draft, March 1996, <draft-ietf-dnsind-dynDNS-09.txt>.

   [DHCP]
      Y. Rechter, ``Interaction between DHCP and DNS,'' Internet Draft,
      February 1996, <draft-ietf-dhc-dhcp-dns-00.txt>.







   Expires November 1996                                          [Page 8]

   INTERNET-DRAFT                 DNS DEFUPD                       May 1996


   9 - Acknowledgements

   Yakov Rechter assisted in the design of this protocol.  Robert Elz
   assisted in the requirements definition.

   10 - Author's Addresses

      Paul Vixie
         Internet Software Consortium
         Star Route Box 159A
         Woodside, CA 94062
         +1 415 747 0204
         <paul@vix.com>



































   Expires November 1996                                          [Page 9]


to unsubscribe send a message to namedroppers-request@ops.ietf.org with
the word 'unsubscribe' in a single line as the message text body.
archive: <http://ops.ietf.org/lists/namedroppers/>