[Gen-art] Re: Gen-ART Review of draft-ietf-v6ops-icmpv6-filtering-recs-01.txt

Thanks for the careful review.

Responses in line:

Spencer Dawkins wrote:
> Dear Elwyn and Janos,
>
> I am the assigned Gen-ART reviewer for this draft. For
> background on Gen-ART, please see the FAQ
> at<http://www.alvestrand.no/ietf/gen/art/gen-art-FAQ.html>. (or ask 
> Elwyn, who is also a Gen-ART reviewer).
>
> Summary - This document is almost ready for publication as an 
> Informational RFC.
>
> I did find a small number of questions, tagged as 
> "Spencer/More-than-Editorial", which appear below, that I'd like you 
> to consider along with any other Last Call comments you may  receive.
>
> There are also editorial notes tagged as "Spencer/Editorial".
>
> Thanks,
>
> Spencer Dawkins
>
> Throughout the document - weasel-wording about "site local" appears in 
> a number of places. Do we have a sense whether this is still 
> necessary? If not, would be nice to delete; if still necessary, might 
> be nice to collect the guidance on site-locals into one place, to 
> ensure that it's consistent.
Hmm!  I don't see this.  Site local addresses (and the term 'site 
local') are only mentioned in s4.2.5 in the context of a particular 
specification.  All the other instances of 'site' are talking about site 
boundaries and characteristics of sites rather than site local addresses 
and I don't think they can be collected up.
>
> Throughout the document - there are many recommendations that point 
> out IPv6 resistence to scanning attacks as part of the recommendation. 
> I'm wondering if some of these could be removed (since there is an 
> explicit section on resistance to scanning attacks in the document).
The main discussion occurs once in s3.2.  Pointers back to s3.2 occur in 
s4.2.1, s4.3.1, sA.5.  This doesn't feel like 'many'.
>
> Abstract
>
>   In networks supporting IPv6 the Internet Control Message Protocol
>   version 6 (ICMPv6) plays a fundamental role with a large number of
>   functions, and a correspondingly large number of message types and
>   options.  A number of security risks are associated with uncontrolled
>   forwarding of ICMPv6 messages.  On the other hand, compared with IPv4
>   and the corresponding protocol ICMP, ICMPv6 is essential to the
>   functioning of IPv6 rather than a useful auxiliary.
>
> Spencer/Editorial: I struggled with the wording of the last two 
> sentences,
> just because it's so central to the point of the document. I'd suggest
> something like "While a number of security risks are associated with
> uncontrolled forwarding of ICMPv6 messages, ICMPv6 is essential to the
> functioning of IPv6. This indicates that current filtering strategies
> designed for the corresponding protocol ICMP in IPv4 networks should be
> revisited, because these strategies are accommodating a useful auxiliary
> protocol that may not be required for correct functioning." But this
> paragraph doesn't seem as clear as the corresponding text in Section 1.
How about:
"ICMPv6 is essential to the functioning of IPv6 but there are a number 
of security risks associated with uncontrolled forwarding of ICMPv6 
messages. Filtering strategies designed for the corresponding protocol 
ICMP in IPv4 networks are not directly applicable, because these 
strategies are intended to accommodate a useful auxiliary protocol that 
may not be required for correct functioning."
>
>
> 1.  Introduction
>
>   ICMPv6 has a large number of functions defined in a number of sub-
>   protocols, and there are a correspondingly large number of messages
>   and options within these messages.  The functions currently defined
>   are:
>
> Spencer/Editorial: The following text is really dense. Is it possible to
> provide any hiearchy/abstraction, even if it's only a list of the 15 
> or so
> functions, without message detail, followed by the full version of the 
> list
> appearing below? If it's possible to combine functions (for example,
> grouping all of the discovery functions into one category, followed by 
> all
> end-to-end signaling functions, followed by MIPv6 ...), that would be
> better. A table would be better. Other proposals would be better.
I'll take this one under advisement.  I agree it is dense (it has been 
pruned once already).  My feeling is that some grouping might be the 
best answer.  The ordering is mostly there already. Thus:

Error messages:
>
>   o  Returning error messages to the source if a packet could not be
>      delivered.  Four different error messages, each with a number of
>      sub-types are specified in [RFC4443].
Connection checking:
>   o  Simple monitoring of connectivity through echo requests and
>      responses used by the ping and traceroute utilities.  The Echo
>      Request and Echo Response messages are specified in [RFC4443].
Discovery functions:
>   o  Finding neighbors (both routers and hosts) connected to the same
>      link and determining their IP and link layer addresses.  These
>      messages are also used to check the uniqueness of any addresses
>      that an interface proposes to use (Duplicate Address Detection -
>      DAD)).  Four messages - Neighbor Solicitation (NS), Neighbor
>      Advertisement (NA), Router Solicitation (RS) and Router
>      Advertisement (RA) - are specified in [RFC2461].
>   o  Ensuring that neighbors remain reachable using the same IP and
>      link layer addresses after initial discovery (Neighbor
>      Unreachability Discovery - NUD) and notifying neighbors of changes
>      to link layer addresses.  Uses NS and NA [RFC2461].
>   o  Finding routers and determining how to obtain IP addresses to join
>      the subnets supported by the routers.  Uses RS and RA
>      [RFC2461].[[anchor2: [RFC Editor: Please update references to
>      RFC2461 when the new version of RFC2461 is published.] --Authors]]
>   o  If stateless auto-configuration of hosts is enabled, communicating
>      prefixes and other configuration information (including the link
>      MTU and suggested hop limit default) from routers to hosts.  Uses
>      RS and RA [RFC2462]. [[anchor3: [RFC Editor: Please update
>      references to RFC2462 when the new version of RFC2462 is
>      published.] --Authors]]
>   o  Using SEcure Neighbor Discovery (SEND) to authenticate a router
>      attached to a link, the Certificate Path Solicitation and
>      Advertisement messages specified in [RFC3971] are used by hosts to
>      retrieve the trust chain between a trust anchor and the router
>      certificate from the router.
  o  Determining the Maximum Transmission Unit (MTU) along a path.  The
     Packet Too Big error message is essential to this function
     [RFC1981].
  o  Providing a means to discover the IPv6 addresses associated with
     the link layer address of an interface (the inverse of Neighbor
     Discovery, where the link layer address is discovered given an
     IPv6 address).  Two messages, Inverse Neighbor Discovery
     Solicitation and Advertisement messages are specified in
     [RFC3122].
  o  Communicating which multicast groups have listeners on a link to
     the multicast capable routers connected to the link.  Uses
     messages Multicast Listener Query, Multicast Listener Report (two
     versions) and Multicast Listener Done (version 1 only) as
     specified in Multicast Listener Discovery MLDv1 [RFC2710] and
     MLDv2[RFC3810].
  o  Discovering multicast routers attached to the local link.  Uses
     messages Multicast Router Advertisement, Multicast Router
     Solicitation and Multicast Router Termination as specified in
     Multicast Router Discovery [RFC4286].

Reconfiguration:
>
>   o  Redirecting packets to a more appropriate router on the local link
>      for the destination address or pointing out that a destination is
>      actually on the local link even if it is not obvious from the IP
>      address (where a link supports multiple subnets).  The Redirect
>      message is specified in [RFC2461].
>   o  Supporting renumbering of networks by allowing the prefixes
>      advertised by routers to be altered.  Uses NS, NA, RS and RA
>      together with the Router Renumbering message specified in
>      [RFC2894].
Mobile IPv6 Support:
>   o  Providing support for some aspects of Mobile IPv6 especially
>      dealing with the IPv6 Mobile Home Agent functionality provided in
>      routers and needed to support a Mobile node homed on the link.
>      The Home Agent Address Discovery Request and Reply; and Mobile
>      Prefix Solicitation and Advertisement messages are specified in
>      [RFC3775]
Experimental Extensions:
>
>   o  An experimental extension to ICMPv6 specifies the ICMP Node
>      Information Query and Response messages which can be used to
>      retrieve some basic information about nodes [I-D.ietf-ipngwg-icmp-
>      name-lookups].
>   o  The SEAmless IP MOBility (seamoby) working group specified a pair
>      of experimental protocols which use an ICMPv6 message specified in
>      [RFC4065] to help in locating an access router and moving the
>      attachment point of a mobile node from one access router to
>      another.
>
> 2.3.  Network Topology and Address Scopes
>
>   Local communications will use link-local addresses in many cases but
>   may also use global unicast addresses for example when configuring
>
> Spencer/Editorial: this sentence may be correct as written but isn't 
> easy to
> read clearly - suggest "may also use global unicast addresses when
> configuring global addresses, for example".
Agreed.
>
>   global addresses.  Also some ICMPv6 messages in local communications
>   may contravene the usual rules requiring compatible scopes for source
>   and destination addresses.
>
> 2.4.  Role in Establishing Communication
>
>   Many ICMPv6 messages have a role in establishing communications to
>   and from the firewall and such messages have to be accepted by
>   firewalls for local delivery.  Generally a firewall will also by
>
> Spencer/Editorial: "will also be"
Well spotted!
>
>   acting as a router so that all the messages that might be used in
>   configuring a router interface need to be accepted and generated.
>   This type of communication establishment messages should not be
>   passed through a firewall as they are normally intended for use
>   within a link.
>
> 3.  Security Considerations
>
>   Firewalls will normally be concerned to monitor ICMPv6 to control the
>
> Spencer/Editorial: "Firewalls will normally be used to monitor" - my
> firewall seems extremely unconcerned most of the time, even when dropping
> packets :-)
Maybe two nations separated by a language (again, sigh)!  This is 
concerned in the sense of involved (see Cambridge Online dictionary).  
Merriam-Webster is subtly different: concerned = 'culpably involved' - 
certainly applies to firewalls!
>
>   following security concerns:
>
> 3.1.  Denial of Service Attacks
>
>   ICMPv6 can be used to cause a Denial of Service(DoS) in a number of
>   ways, including simply sending excessive numbers of ICMPv6 packets to
>   destinations in the site and sending error messages which disrupt
>   established communications by causing sessions to be dropped.  Also
>   if spurious communication establishment messages can be passed on to
>
> Spencer/Editorial: is this "passed to on-link"? But it seems like odd
> English usage. "passed onto a specific link", perhaps?
Missing 'a':  Better would be 'can be infiltrated on to a link'.
>
>   link it might be possible to invalidate legitimate addresses or
>   disable interfaces.
>
> 3.4.  Renumbering Attacks
>
>   Spurious Renumbering messages could lead to the disruption of a site
>   and should not be allowed through a firewall in general.  Renumbering
>   messages are required to be authenticated with IPsec so that it is
>   difficult to carry out such attacks in practice.
>
> Spencer/Editorial: Is this saying, roughly, "Spurious Renumbering 
> messages can lead to the disruption of a site. Although Renumbering 
> messages are required to be authenticated with IPsec, so that it is 
> difficult to carry out such attacks in practice, they should not be 
> allowed through a firewall."?
Yes.  I'll reword this a bit.
>
> 3.5.  Problems Resulting from ICMPv6 Transparency
>
>   Because some ICMPv6 error packets need to be passed through a
>   firewall in both directions.  This means that the ICMPv6 error
>   packets can be exchanged between inside and outside without any
>   filtering.
>
> Spencer/More-than-Editorial: First "sentence" is a fragment - I'm not 
> sure
> that the paragraph is a complete thought, because I don't understand 
> how the
> second sentence is related to the first sentence (fragment), so I'm not
> proposing an editorial change. (Just because you need something to happen
> doesn't mean "there's no problem when it happens, so go ahead and pass 
> the
> traffic without filtering.") Someone smart, please provide suggested text
> :-)
Replace the first three sentences with (and fix the 'in might'):
"Because some ICMPv6 error packets need to be passed through a firewall 
in both directions, malicious users can potentially use these messages 
to communicate between inside and outside bypassing administrative 
inspection..  For example it might....."

>
>   Using this feature, malicious users can communicate between the
>   inside and outside of a firewall bypassing the administrator's
>   inspection (proxy, firewall etc.).  For example it might be possible
>   to carry out a covert conversation through the payload of ICMPv6
>   error messages or tunnel inappropriate encapsulated IP packets in
>   ICMPv6 error messages.  This problem can be alleviated by filtering
>   ICMPv6 errors using a deep packet inspection mechanism to ensure that
>   the packet carried as a payload is associated with legitimate traffic
>   to or from the protected network.
>
> Spencer/More-than-Editorial: Forgive my ignorance here, but are most 
> firewalls doing stateful deep inspection tracking for ICMP/ICMPv6 
> traffic today? If not, this seems to need more discussion about why 
> the additional state is the right way to go.
Many firewalls can do state tracking for ICMPv6.  I don't think we can 
say much beyond or different from this - there is a disclaimer elsewhere 
that states that we cannot require such state tracking or deep packet 
inspection capbility, and without it there is nothing to be done at the 
firewall.
>
> 4.  Filtering Recommendations
>
>   This section suggests some common considerations which should be
>   borne in mind when designing filtering rules and then categorizes the
>   rules for each class.  The categories are:
>   o  Messages that must not be dropped: usually because establishment
>      of communications will be prevented or severely impacted.
>
> Spencer/More-than-Editorial: why "usually"? ("Why would you pass messages
> you didn't have to pass?")
Trying to cover my backside!  The issue is pings.  Pings must be passed 
if you are doing Teredo and it is a really good idea to allow pings so 
that connectivity can be checked, but they are not absolutely essential 
in every case but it is going to be difficult to distinguish the two 
cases.  How to say this in one short sentence? Suggestions other than 
what I have?
>
>   o  Messages that should not be dropped: administrators need to have a
>      very good reason for dropping this category
>   o  Messages that may be dropped but it is not essential because they
>      would normally be dropped for other reasons (e.g., because they
>      would be using link-local addresses) or the protocol specification
>      would cause them to be rejected if they had passed through a
>      router.
>
> Spencer/Editorial: "Messages that may be dropped, although they would
> normally be dropped for other reasons anyway", and I think the 
> parenthetical
> should inclide both the e.g. and the "or the protocol specification" 
> clause,
> if I get the point being made.
No: There are two cases - this could be clarified as follows:
"Messages that may be dropped but it is not essential because they would 
normally be dropped by the router/firewall for other reasons (e.g., 
because they would be using link-local addresses) or the protocol 
specification would cause them to be rejected at the detsination if they 
had passed through a router."

Actually there is an additional comment to make here.  Firewalls 
implemented as bridges may need to do different things because they 
don't do routing activities.  I'll add words to s4.2.3 to cover this 
case.  This means that they shouldn't filter connection on-link 
establishment messages.
>
>
> 4.1.  Common Considerations
>
>   Many of the messages used for establishment of communications on the
>   local link will be sent with link-local addresses for at least one of
>   their source and destination.  Routers (and firewalls) conforming to
>   the IPv6 standards will not forward these packets; there is no need
>   to configure additional rules to prevent these packets traversing the
>   firewall/router.  Also the specifications of ICMPv6 messages intended
>   for use only on the local link specify various measures which would
>   allow receivers to detect if the message had passed through a
>   firewall/router, including:
>   o  Requiring that the hop limit in the IPv6 header is set to 255 on
>      transmission.  On reception the hop limit is required to be still
>      255 which would not be the case if the packet had passed through a
>      firewall/router.
>
> Spencer/Editorial: second sentence "Receivers verify that the hop 
> limit is
> still 255, to ensure that the packet has not passed through a
> firewall/router."?
That's probably better.
>
>
>   o  Checking that the source address is a link-local unicast address.
>   Accordingly it is not essential to configure firewall rules to drop
>   illegal packets of these types.  If they have non-link-local source
>   and destination addresses, allowing them to traverse the firewall,
>   they would be rejected because of the checks performed at the
>   destination.  However, firewall administrators may still wish to log
>   or drop such illegal packets.
>
> Spencer/Editorial: There really are "illegal packets", but I don't 
> think that's what you're talking about. "wish to drop 
> adminstratively-prohibited packets"? This term is used elsewhere in 
> the specification, too.
Maybe 'out of specification'.
>
>
>   In general, the scopes of source and destination addresses of ICMPv6
>   messages should be matched, and packets with mismatched addresses
>   should be dropped if they attempt to transit a router.  However some
>   of the address configuration messages carried locally on a link may
>   legitimately have mismatched addresses.  Node implementations need to
>   avoid over-zealous filtering of these messages delivered locally on a
>   link.
>
> Spencer/More-than-Editorial: I know the specification is targeted as 
> an Informational RFC, but this recommendation seems awefully 
> hand-wavish even for Informational ("you should drop them unless doing 
> so is over-zealous") - is there better guidance than "turn it off and 
> see if the phone rings"?
Ok. The point is that for almost all packets address scopes should 
match, but if you apply this test to NS and NA messages, things will break.
So the node implementation has to ensure that scope match checking is 
not done on these packets.  This is not a firewall config or (that I 
have seen) a user controllable host config issue.  We could remove this 
sentence or make it clearer that it is not something admins need to 
worry about other than not to panic when they see this sort of packet.
>
> 4.2.3.  Traffic that May be Dropped but will be Caught Anyway
>
> Spencer/Editorial: this phrase, which appears in several places in the 
> document, seemed odd to me every time I read it. Perhaps "Traffic that 
> will be dropped anyway for other reasons, so no special attention 
> required"? The text describing this is fine - the thing being 
> described just seems odd.
>
> 4.2.4.  Traffic for which a Dropping Policy Should be Defined
>
>   The message which the experimental Seamoby protocols are using will
>
> Spencer/Editorial: This is actually probably correct and consistent 
> but still looks like a number disagreement "the message are". Maybe 
> "The message type used by the experimental Seamoby protocols will ..."?
Agreed - it will than be consistent with 4.3.4.
>
>   be expected to have to cross site boundaries.  Administrators should
>
> Spencer/Editorial: "will be expected to cross site boundaries in 
> normal operation."?
>
OK
>   determine if they need to support these experiments and otherwise
>   messages of this type should be dropped:
>   o  Seamoby Experimental (Type 150)
>
> 4.3.1.  Traffic that Must NOT be Dropped
>
> Spencer/Editorial: "NOT" sure why the "NOT" is emphasized here, and it 
> looks like 2119 language gone bad :-)
>
> A.2.  Packet Too Big Error Message
>
>   If a network chooses to generate packets that are no larger than the
>   Guaranteed Minimum MTU (1280 octets) and the site's links to the
>   wider internet have corresponding MTUs, Packet Too Big messages
>   should not be expected at the firewall and could be dropped if they
>   arrive.
>
> Spencer/More-than-Editorial: Having had experience with wireless links 
> that had a smaller MTU than the "guaranteeed minimum" in IPv4, I'm not 
> entirely comfortable with this guidance ("if you don't care about 
> communicating with people who violate the spec, drop the packets, and 
> hope there's no GOOD reason why someone is violating the spec"). At a 
> minimum, there's a balance between making the network more secure by 
> not passing unnecessary packets and making the path more fragile - 
> maybe that's guidance that would be helpful?
Fair comment, but the situation is a little different in IPv4.  The 
minimum is actually advisory in IPv4 - there is no hard limit, whereas 
1280 is set in stone in the specs for IPv6 (OK very old specs had lower 
values, but those are not in use today).  I think we might get stick for 
contemplating smaller packets.
>
> A.8.  Redirect Messages
>
>   ICMPv6 Redirect Messages(Type 137) are used on the local link to
>   indicate that nodes are actually link-local and communications need
>   not go via a router, or to indicate a more appropriate first hop
>   router.  Although they can be used to make communications more
>   efficient, they are not essential to the establishment of
>   communications and may be a security vulnerability, particularly if a
>   link is not physically secured.  Conformant nodes are required to
>   provide configuration controls which suppress the generation of
>   Redirect messages and allow them to be ignored on reception.  Using
>   Redirect messages on a wireless link is particularly hazardous
>   because of the lack of physical security.
>
> Spencer/More-than-Editorial: should be something like "wireless link 
> with no link-level authentication/encryption is particularly 
> hazardous", I think - the problem isn't that the link is wireless, the 
> problem is that the link is open to eavesdropping and packet injection.
OK.  I'll craft some words to cover the situation.
>
> A.14.  Mobile IPv6 Messages
>
>   Mobile IPv6 [RFC3775] defines four ICMPv6 messages which are used to
>   support mobile operations: Home Agent Address Discovery Request, Home
>   Agent Address Discovery Reply, Mobile Prefix Solicitation and ICMP
>   Mobile Prefix Advertisement(Type 144, 145, 146 and 147) messages.
>   These messages are sent end-to-end between unicast addresses of a
>   mobile node and its home agent.  They must be expected to be sent
>   from outside a site.  The two Mobile prefix messages should be
>
> Spencer/Editorial: "They must traverse site-boundary firewalls in 
> order for Mobile IPv6 to function"?
I'll add " and must etc."
>
>   protected by the use of IPsec authentication.
>
I'll action these after last call ends.

/Elwyn

_______________________________________________
Gen-art mailing list
Gen-art@ietf.org
https://www1.ietf.org/mailman/listinfo/gen-art