Internet Engineering Task Force P. Savola
Internet Draft CSC/FUNET
Expiration Date: June 2003
December 2002
Security Considerations for 6to4
draft-savola-v6ops-6to4-security-01.txt
Status of this Memo
This document is an Internet-Draft and is subject to all provisions
of Section 10 of RFC2026.
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."
The list of current Internet-Drafts can be accessed at
http://www.ietf.org/ietf/1id-abstracts.txt.
To view the list Internet-Draft Shadow Directories, see
http://www.ietf.org/shadow.html.
Abstract
The IPv6 interim mechanism 6to4 [6TO4] uses automatic IPv6-over-IPv4
tunneling to interconnect IPv6 networks. The architecture includes
Relay Routers and Routers, which accept and decapsulate IPv4 traffic
from anywhere. There aren't many constraints on the embedded IPv6
packets, or where IPv4 traffic will be automatically tunneled to.
These could enable one to go around access controls, and more likely,
being able to perform proxy Denial of Service attacks using 6to4
relays or routers as reflectors. Anyone is also capable of spoofing
traffic from non-6to4 addresses, as if it was coming from a relay, to
a 6to4 node. This document discusses these issues in more detail and
tries to suggest enhancements to alleviate the problems.
Savola [Expires June 2003] [Page 1]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
Table of Contents
1. Introduction ............................................... 3
2. Different 6to4 Forwarding Scenarios ........................ 4
2.1. From 6to4 to 6to4 ...................................... 4
2.2. From Native to 6to4 .................................... 5
2.3. From 6to4 to Native .................................... 5
2.4. Other Models ........................................... 6
2.4.1. BGP Between 6to4 Routers and Relays ................ 6
2.4.2. 6to4 as an Optimization Method ..................... 6
2.4.3. 6to4 as Tunnel End-Point Addressing Mechanism ...... 6
3. Some Functionalities of 6to4 ............................... 7
3.1. Functions of Different 6to4 Network Components ......... 7
3.2. Non-functions of Different 6to4 Network Components ..... 8
4. Special Processing of 6to4 Packets ......................... 9
4.1. Encapsulating IPv6 Packets into IPv4 ................... 9
4.2. Decapsulating IPv4 Packets into IPv6 ................... 9
5. Threat Analysis ............................................ 9
5.1. Threats Related to Any 6to4 Node ....................... 10
5.2. Threats Related to 6to4 Routers ........................ 10
5.2.1. Attacks Against the 6to4 Pseudo-Interface .......... 10
5.2.1.1. Comparison to Situation without 6to4 ........... 10
5.2.2. Relay Spoofing, DoS against IPv6 Nodes ............. 11
5.2.2.1. Comparison to Situation without 6to4 ........... 11
5.3. Threats Related to 6to4 Relays ......................... 12
5.3.1. Attacks Against the 6to4 Pseudo-Interface .......... 12
5.3.2. Participating in DoS attacks against IPv4 .......... 13
5.3.2.1. Comparison to Situation without 6to4 ........... 13
5.3.3. Using Any IPv6 Node for Reflection ................. 13
5.3.3.1. Comparison to Situation without 6to4 ........... 14
5.3.4. IPv4 Local Directed Broadcast Attacks .............. 14
5.3.4.1. Comparison to Situation without 6to4 ........... 15
5.3.5. Theft of Service ................................... 15
5.4. Summary ................................................ 16
6. Solutions .................................................. 17
6.1. Generic Approach ....................................... 17
6.1.1. Encapsulating IPv6 into IPv4 ....................... 17
6.1.2. Decapsulating IPv4 into IPv6 ....................... 17
6.1.3. IPv4 and IPv6 Sanity Checks ........................ 18
6.1.3.1. IPv4 ........................................... 18
6.1.3.2. IPv6 ........................................... 18
6.1.3.3. Optional Ingress Filtering ..................... 18
6.1.3.4. Notes About the Checks ......................... 19
6.2. Simplified Approach .................................... 19
6.2.1. Encapsulating IPv6 into IPv4 ....................... 19
6.2.2. Decapsulating IPv4 into IPv6 ....................... 19
6.3. All Relays Must Be Anycast ............................. 20
7. Problems ................................................... 20
Savola [Expires June 2003] [Page 2]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
7.1. Implementation Considerations with Automatic Tunnels ... 20
7.2. Reduced Flexibility .................................... 21
7.3. Anyone Pretending to Be a Relay Router ................. 21
7.3.1. 6to4 Relay Routers Must Use 192.88.99.1 ........... 22
7.3.2. Limited Distribution of More Specific Routes ....... 23
8. Security Considerations .................................... 23
9. Acknowledgements ........................................... 24
10. References ................................................ 24
10.1. Normative References .................................. 24
10.2. Informative References ................................ 25
Author's Address ............................................... 25
A. Some Trivial Attack Scenarios Outlined ..................... 26
1. Introduction
The IPv6 interim mechanism "6to4" [6TO4] specifies automatic
IPv6-over-IPv4 tunneling to interconnect isolated IPv6 clouds without
explicit tunnels by embedding the tunnel IPv4 address in the IPv6
6to4 prefix.
One challenge with this mechanism is that all 6to4 routers must
accept and decapsulate IPv4 packets from every other 6to4 router;
there are no strict constraints on what the IPv6 packet may contain,
which implies a trust relationship.
Another, bigger challenge is that to interconnect native IPv6
networks and 6to4 clouds, relay routers are used as bridges between
these two clouds. Relay routers can be tricked by malicious parties
to send IPv4, or IPv6, traffic anywhere the attacker wants. With
source address spoofing, this could be called traffic "laundering" or
a "proxy" denial-of-service attack. To some extent, these reflected
attacks can also be launched off any nodes.
Even worse, anyone can send tunneled traffic, spoofed to come from
non-6to4 addresses to any 6to4 router, and the node does not have any
means to ensure its correctness, but to assume it came from a
legitimate Relay.
The 6to4 specification outlined quite a few security considerations,
but it has been shown that in practice some of these have been
difficult to get implemented due to their abstract nature.
This draft analyses the 6to4 security issues in more detail and
outlines some enhancements and caveats.
Savola [Expires June 2003] [Page 3]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
Sections 2-4 are more or less introductory in nature, rehashing how
6to4 should be used today based on the 6to4 specification, so that it
is easier to understand how security could be affected. Section 5
provides a threat analysis for implementations that already implement
most of the security checks. Section 6 introduces some filtering
rules for 6to4 implementations, and section 7 discusses some
additional problems which still need some consideration. Appendix A
outlines a few possible trivial attack scenarios in the case that
very little or no security has been implemented.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
document are to be interpreted as described in [RFC2119].
2. Different 6to4 Forwarding Scenarios
It should be noted that when communicating between 6to4 and native
domains, the relays that will be used in the two directions are very
likely different; routing is highly asymmetric. Because of this, it
is not feasible to limit relays you accept traffic from with e.g.
access lists.
The first three subsections introduce the most common form of 6to4
operation. Other models are presented in the fourth subsection.
2.1. From 6to4 to 6to4
6to4 domains always exchange 6to4 traffic directly via IPv4
tunneling; the endpoint address V4ADDR is derived from 6to4 prefix
2002:V4ADDR.
.--------. _----_ .--------.
| 6to4 | _( IPv4 )_ | 6to4 |
| Router | <====> ( Internet ) <===> | Router |
'--------' (_ _) '--------'
^ '----' ^
| Direct tunneling over IPv4 |
V V
.--------. .-------.
| 6to4 | | 6to4 |
| Client | | Client |
'--------' '--------'
It is required that every 6to4 router considers every other 6to4
router it wants to talk to to be "on-link" (with IPv4 as the link-
layer). If this is restricted by increasing the prefix length from
2002::/16, some traffic will be sent to the 6to4 Relay Router, which
would forward it to other 6to4 Routers. However, if the original
Savola [Expires June 2003] [Page 4]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
destination does not have equally long prefix, the traffic it tries
to send back will be tunneled directly, and will be dropped.
Therefore, the restricted scenario with a longer prefix-length is not
globally workable and will not be considered here.
2.2. From Native to 6to4
When native domains send traffic to 6to4 address 2002:V4ADDR, it will
be routed to the topologically nearest, advertising 6to4 Relay
Router. Relay router will tunnel the traffic over IPv4 to the
corresponding IPv4 address V4ADDR. (Note that IPv4 address 9.0.0.1
here is just an example of a global IPv4 address.)
2002::/16 Closest to 'Native Client'
.--------. _----_ .------------. .--------.
| Native | _( IPv6 )_ | 6to4 Relay | Tunneled | 6to4 |
| Client | -> ( Internet ) --> | Router | =========> | Router |
'--------' (_ _) '------------' 9.0.0.1 '--------'
'----' dst=2002:0900:0000::1 |
V
.-------.
| 6to4 |
| Client |
'--------'
2.3. From 6to4 to Native
6to4 domains send traffic to native domains by tunneling it over IPv4
to their configured 6to4 Relay Router, or the closest found using
6to4 IPv4 Anycast [6TO4ANY]. The relay will decapsulate the packet
and forward it to native IPv6 Internet, the same way as any other
IPv6 packet.
Configured/found by IPv4 Anycast
.--------. _----_ .------------. .--------.
| Native | _( IPv6 )_ | 6to4 Relay | Tunneled | 6to4 |
| Client | <- ( Internet ) <-- | Router | <========= | Router |
'--------' (_ _) '------------' 192.88.99.1'--------'
'----' (or configured) ^
dst=3ffe:ffff::1 |
.-------.
| 6to4 |
| Client |
'--------'
Savola [Expires June 2003] [Page 5]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
2.4. Other Models
These are more or less special cases of 6to4 operation; in later
chapters, unless otherwise stated, only the most generic form will be
considered.
2.4.1. BGP Between 6to4 Routers and Relays
[6TO4, 5.2.2.2] presents a model where, instead of static
configuration, BGP4+ is used between 6to4 Relay Routers and 6to4
Routers.
If the 6to4 router established a BGP session between all the possible
6to4 relays, the traffic from non-6to4 sites would always go through
"home relay", and configuring "trusted relay" would not be a problem;
an alternative would be to advertise the more specific 6to4 routes
between 6to4 Relays, as described later in this memo.
This model is not known to be used at the time of writing; this is
probably caused by the fact that parties that need 6to4 are those
that are not able to run BGP; and setting up these sessions would be
much more work for relay operators.
2.4.2. 6to4 as an Optimization Method
Some seem to use 6to4 as an IPv6 connectivity "optimization method";
that is, they have also non-6to4 addresses on their nodes and border
routers, but also employ 6to4 to reach 6to4 sites.
This is typically done to be able to reach 6to4 destinations by
direct tunneling and not having to use relays at all.
Some also publish both 6to4 and non-6to4 addresses in DNS to affect
inbound connections; if the source host's default address selection
[ADDRSEL] works properly, 6to4 sources will use 6to4 addresses to
reach the site and non-6to4 nodes use non-6to4 addresses. If this
can be assumed, the security threats are significantly simplified.
2.4.3. 6to4 as Tunnel End-Point Addressing Mechanism
6to4 addresses can also be used only as an IPv6-in-IPv4 tunnel
endpoint addressing and routing mechanism.
An example of this is interconnecting 10 branch offices where nodes
use non-6to4 addresses. Only the offices' border routers need to be
aware of 6to4, and use 6to4 addresses solely for addressing the
tunnels between different branch offices. This assumes that all
outgoing traffic from the main organization (but not between branch
Savola [Expires June 2003] [Page 6]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
offices) uses one or more non-6to4 connections.
This is similar to the optimization model above, and can be used to
make the addressing and routing easier.
3. Some Functionalities of 6to4
We list some, relatively obvious features of different 6to4
components to better undestand what's the required behaviour.
3.1. Functions of Different 6to4 Network Components
o Non-6to4 (Native) Node
If native IPv6 nodes want to communicate with 6to4 nodes,
they send the traffic along normally. The traffic will
reach the topologically closest, advertising 6to4 Relay
Router, and will be tunneled to the destination 6to4 Router,
which will pass it to the 6to4 node via normal forwarding
process.
o 6to4 Host
A host, usually autoconfigured and has an address from a
6to4 prefix, but doesn't have a 6to4 pseudo-interface. It
doesn't need to know anything about 6to4, and it acts like a
normal IPv6 Host in every manner. Note that 6to4 Hosts can
also be 6to4 Routers in some scenarios, but then 6to4 Router
functionalities, below, apply.
o 6to4 Router
Acts at the border of a 6to4 domain. It does not have a
native, global IPv6 address. More specifically:
- provide "native-like" IPv6 connectivity to local clients
and routers
- if packets are sent to foreign 6to4 addresses, tunnel
them to the destination 6to4 router using IPv4
- if packets are sent to locally configured 6to4
addresses, forward them normally
- if packets are sent to non-6to4 addresses, tunnel them
to the configured/closest-by-anycast 6to4 Relay Router,
which will pass them on
- if packets are received from 6to4 addresses, decapsulate
the IPv4 packets received from 6to4 routers
Savola [Expires June 2003] [Page 7]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
- if packets are received from non-6to4 addresses,
decapsulate the IPv4 packets received from 6to4 Relay
Router closest to the source (note: it is not easily
distinguishable that the packet was really received from
a Relay router, not from a spoofing third party.)
o 6to4 Relay Router
Acts as a relay between all 6to4 domains and native IPv6;
more specifically:
- advertises the reachability of the 2002::/16 prefix to
native IPv6 routing, thus receiving traffic to all 6to4
addresses from closest native IPv6 nodes
- (if implements RFC3068) advertise the reachability of
IPv4 '6to4 Relay anycast prefix' (192.88.99.0/24) to
IPv4 routing, thus receiving some tunneled traffic to
native IPv6 nodes from 6to4 Routers
- if packets are received from 6to4 addresses through
tunneling, decapsulate them and forwards them on using
normal IPv6 routing
- if packets are received through normal IPv6 routing from
native addresses, and are destined for 2002::/16, tunnel
them to the corresponding 6to4 Router
3.2. Non-functions of Different 6to4 Network Components
What should not happen; this forms a basis for the security checks.
The lists are not exhaustive.
o 6to4 Router or Relay
- use private, broadcast or reserved IPv4 addresses in tunnels,
or the matching 6to4 prefixes
- receive traffic from 6to4 Routers where the IPv4 tunnel
source address does not match the 6to4 prefix
- receive traffic where the destination IPv6 address is not a
global address; in particular, e.g. link-local addresses,
mapped addresses and such should not be used
o 6to4 Router
- send traffic to other 6to4 domains through 6to4 Relay Router
or via some third party 6to4 Router
- receive traffic from other 6to4 domains via a 6to4 Relay
Router
- receive traffic to other than to your own 6to4 prefix(es)
o 6to4 Relay Router
Savola [Expires June 2003] [Page 8]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
- receive traffic from 6to4 to 6to4
4. Special Processing of 6to4 Packets
One could summarize the special processing of 6to4 as follows:
o Relay Router
1. incoming from native, tunneled to 6to4
2. tunneled from 6to4, going to native
o Router
1. tunneled from relay, source is native
2. tunneled to relay, destination is native
3. tunneled directly, destination is 6to4
4.1. Encapsulating IPv6 Packets into IPv4
IPv6 packets are encapsulated into IPv4 in three scenarios:
1. 6to4 Router sends packets to other 6to4 Routers (2002::/16
destination)
2. 6to4 Router sends packets to its configured/nearest-by-anycast
6to4 Relay Router (non-2002::/16 destination)
3. 6to4 Relay Router sends packets from native IPv6 sources to
6to4 Routers (2002::/16 destination)
4.2. Decapsulating IPv4 Packets into IPv6
IPv6 packets are decapsulated from IPv4 in three scenarios:
1. 6to4 Router receives packets from other 6to4 Routers (2002::/16
source)
2. 6to4 Router receives packets from a node, supposedly 6to4 Relay
Router closest to the source (non-2002::/16 source)
3. 6to4 Relay Router receives packets from 6to4 Routers, to be
sent to native IPv6 destinations (2002::/16 source)
5. Threat Analysis
The 6to4 threat analysis presented here focuses on 6to4
implementations which have implemented most if not all security
checks listed in [6TO4]; in particular, checks that always match
2002:V4ADDR and V4ADDR must be implemented. Many implementations are
known to be problematic at least in some cases.
The appendix lists some additional trivial threats which are
applicable if no or only little security is implemented.
Savola [Expires June 2003] [Page 9]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
The IPv4 address blocks 8.0.0.0/24 and 9.0.0.0/24 are only used for
demonstrative purposes, representing global IPv4 addresses.
5.1. Threats Related to Any 6to4 Node
Any 6to4 node can be made to participate in a DoS attack listed in
5.2.2 below, being used as "dst". The threat will be discussed
there.
5.2. Threats Related to 6to4 Routers
Note that in this memo, a loose interpretation of "6to4 router" is
used; it is also used to refer to those 6to4 nodes which have the
6to4 pseudo-interface.
There are two main classes of threats; attacks against the 6to4
pseudo-interface and attacks relying on being able to abuse the fact
that it is difficult for 6to4 routers to tell whether packets from
non-6to4 nodes come from legitimate relays.
5.2.1. Attacks Against the 6to4 Pseudo-Interface
Unless the 6to4 pseudo-interface has been sufficiently protected,
it's possible to remotely attack against the pseudo-interface with
tunneled link-local packets, just as if it were in a local network.
Threats to Neighbor Discovery are listed in [SEND].
The potential effects of an attack can be mitigated if the interface
is insulated from the other interfaces (e.g. a separate neighbor
cache). In practise, this is not the case.
The attack can be eliminated by restricting the use of 6to4 pseudo-
interface: if any packet with scope smaller than global is received,
it must be silently discarded. (Site-local might be allowed in some
restricted scenarios.) This may conflict with future uses of [6TO4,
3.1].
5.2.1.1. Comparison to Situation without 6to4
Even though rather simply fixable, this attack is not new as such;
the same is possible using automatic tunneling [MECH] or configured
tunneling (if one is able to spoof source IPv4 address to that of the
tunnel end-point).
However, as automatic tunneling is being deprecated, the worst
problem comes from 6to4; any open decapsulation is bad.
Savola [Expires June 2003] [Page 10]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
5.2.2. Relay Spoofing, DoS against IPv6 Nodes
6to4 Routers receive packets from non-6to4 source addresses through
Relay Routers, as described in section 2.2 and 4.2 point 2.
In the general case, the 6to4 router cannot distinguish Relay routers
from malicious nodes sending IPv4-encapsulated IPv6 traffic directly
to the 6to4 router.
This makes two kinds of attacks possible:
o unidirectional source address spoofing
o Denial-of-Service attack reflection against native IPv6 nodes
In both scenarios, the attacker sends IPv4-encapsulated IPv6 packets
to the 6to4 router with contents like:
src = 3ffe:1122:3344::1 (forged)
dst = 2002:0900:0002::1
src_v4 = 8.0.0.1
dst_v4 = 9.0.0.2 (matching dst)
Now the 6to4 router receives these packets from 8.0.0.1, decapsulates
them, discarding the IPv4 header containing the source address
8.0.0.1 and processes them as normal ("dst" has been guessed or
obtained using one of a number of techniques).
In the first scenario, it is assumed that "src" is somehow specially
trusted (at least to some extent) more than some other packets. This
kind of weak trust based on IP addresses is commonplace. This
enables unidirectional (as the replies will be lost) source address
spoofing, but may be enough for e.g. exploiting some remote
vulnerabilities in connectionless protocol server applications.
In the second scenario, if "dst" exists, the replies (e.g. TCP SYN
ACK, TCP RST, ICMP Echo Reply, input sent to UDP echo service, etc.)
are sent to the victim "src", above. All the traces from the
original attacker src_v4 have been discarded. These return packets
will go through a relay.
These attacks are similar to ones shown in [RHHASEC].
5.2.2.1. Comparison to Situation without 6to4
The unidirectional spoofing attack exists without 6to4 too, but it
requires the attacker is able to spoof IPv6 source addresses. With
6to4, one is able to launch this attack without any spoofing at all.
A restriction is that the source address cannot be spoofed to belong
Savola [Expires June 2003] [Page 11]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
to the destination site as only non-6to4 addresses can be spoofed
(assuming correct implementations). Blindly trusting source address
of packets coming from the Internet without other precautions is very
bad practise, though -- and this attack would typically be useful
only for spoofing destination site -- which is not possible, and can
be protected against with egress filtering.
The Denial-of-Service attack is also not really new; the only twists
come from the fact that traces of an attack are more easily lost and
that attacker does not really have to even spoof his IPv4 address to
be able to reasonably discreetly launch the attack.
However, it can be argued that this DoS attack is not all that
serious because:
o 6to4 as an enabling mechanism does not provide any possibility
for packet multiplication, attacks are generally 1:1
o victim receives packets as replies from "dst": unless echo
service (e.g. UDP port 7) is used, the attacker has reasonably
little control on the content of packets; for example, pure "SYN"
TCP packets often used for flooding cannot be sent.
o attack packets pass through choke point(s), namely (one or more)
6to4 relays; in addition to physical limitations, these could
implement some form 6to4-site-specific traffic limiting
o one has to know a valid destination address (however, this is
easily guessable or deducible with e.g. an ICMP echo request with
a limited Hop Count).
The attack can be launched from hosts whose connection is ingress-
filtered, too. However, often this is not a real factor, as usually
the attackers are just zombies and real attackers may not even care
if the real source is discovered.
5.3. Threats Related to 6to4 Relays
6to4 Relays are also subject to attacks, but usually in a different
role than 6to4 Routers; usually Relays' function is anonymization of
the attack and losing trails, not reflection -- as properly
implemented relays should be resistant to reflection.
5.3.1. Attacks Against the 6to4 Pseudo-Interface
The threats are the same as against 6to4 routers.
Savola [Expires June 2003] [Page 12]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
5.3.2. Participating in DoS attacks against IPv4
An attack specific to 6to4 Relays is similar to 6to4 Routers, but
against IPv4; the attacker sends IPv6-native packets with an IPv4
address he wants to bomb as the 6to4 destination address, like:
src = 3ffe:1122:3344::1 (spoofed or real attacker)
dst = 2002:0900:0002::1 (victim)
Now the 6to4 relay receives these packets, and encapsulates in IPv4
packets that are sent towards 9.0.0.2; the destination may not have a
faintest idea of what IPv6 is, but is bombed with packets coming from
the relay's IPv4 address, including IPv6 packets as the payload.
5.3.2.1. Comparison to Situation without 6to4
Slightly different arguments apply; below are reasons why this can be
considered not too serious an attack:
o 6to4 as an enabling mechanism does not provide any possibility
for packet multiplication, attacks are generally 1:1
o victim receives packets as sent by the source; if the victim is
an IPv4-only node, it just sees "protocol 41" packets which are
not typically dangerous and easily filtered
o 6to4 relay's source IPv4 address is used in packets, and tracing
is easier
o source IPv6 address (spoofed or real) is in the packets which may
make manual tracing easier
o attack packets pass through choke point(s), namely (one or more)
6to4 relays; in addition to physical limitations, these could
implement some form 6to4-site-specific traffic limiting
And as counter-arguments, some more serious aspects of it are:
o victim receives packets as sent by the source; if the victim is
6to4 node, IPv6 packet can include almost anything; however if
IPv6 source addess is not spoofed, this attack is nothing new.
o the relays can be chosen by the attacker, so if there are a large
number of relays, he can pick ones that are known best suited for
the attack (e.g. bad security policy, ones using 192.88.99.1 as
source for more difficult tracing, etc.)
5.3.3. Using Any IPv6 Node for Reflection
Any IPv6 node will respond to IPv6 packets destined to the node with
source address belonging to 2002::/16.
Savola [Expires June 2003] [Page 13]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
This attack is applicable if:
o the relay chosen by the attacker does not check that IPv4 source
and 2002::/16 source address match, or
o the attacker's IPv6 connection is not ingress-filtered (and it
was really a non-6to4 source)
The IPv6 source address being forged, the node will participate as an
unwilling intermediary to an attack through a 6to4 relay against any
IPv4 node (or 6to4 node), like:
src = 2002:0900:0002::1 (forged, target of the attack)
dst = 3ffe:1122:3344::1 (intermediary node)
This is not new: similar attack with any other spoofed source address
is possible if ingress filtering is not enabled. The only difference
here is that when attacking IPv4 nodes, the relay's IPv4 source
address is seen as the immediate source of the attack. Closer
inspection (after packet reflection) reveals the IPv6 datagram with
(possibly spoofed) 2002::/16 destination address.
5.3.3.1. Comparison to Situation without 6to4
This attack is a reflected variation of the attack above; the
implications are similar to those in section 5.2.2.1:
o A non-compliant 6to4 implementation or IPv6 source address
spoofing is required
o 6to4 as an enabling mechanism does not provide any possibility
for packet multiplication, attacks are generally 1:1
o victim receives packets as replies from "dst": unless echo
service (e.g. UDP port 7) is used, the attacker has reasonably
little control on the content of packets; for example, pure "SYN"
TCP packets often used for flooding cannot be sent.
o attack packets pass through choke point(s), namely (one or more)
6to4 relays; in addition to physical limitations, these could
implement some form 6to4-site-specific traffic limiting
5.3.4. IPv4 Local Directed Broadcast Attacks
This threat is applicable if the relay does not check whether the
IPv4 address it tries to send encapsulated IPv6 packets to is a local
broadcast address. This threat is mentioned in [6TO4]. The packet
could be sent as follows:
Savola [Expires June 2003] [Page 14]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
src = 3ffe:ffff:5678::aaaa (may be forged)
dst = 2002:0900:00ff::bbbb
9.0.0.255 is assumed the the relay's broadcast address. After
receiving the packet natively, if the relay doesn't check the
destination address for subnet broadcast, it would send the
encapsulated IP-IP packet to 9.0.0.255. This would be received by
all nodes in the subnet, and the responses would be directed to the
relay.
5.3.4.1. Comparison to Situation without 6to4
This is a special form of "directed broadcast" attack which cannot be
protected against except by the mentioned check. However, there is a
significant difference: the reply packets are sent back to the relay.
Therefore only the non-compliant device can suffer from this; the
rest of the Internet cannot be affected.
5.3.5. Theft of Service
The administrators of 6to4 Relay Routers often want to use some
policy to limit the use of relay.
The policy control is usually done by applying some restrictions in
where the routing information for 2002::/16 and/or 192.88.99.0/24 (if
[6TO4ANY] is used) will spread.
Some users may be able to use the service regardless of these
controls using:
o configuring the address of the relay using its IPv4 address
instead of 192.88.99.1
o using Routing Header to route IPv6 packets to reach some 6to4
Relay (some other routing tricks like neighbors setting static
routes are also possible)
The former can be protected against using configured access lists in
the relay; this is only feasible if the number of IP networks the
relay is supposed to serve is relatively low. Another possible way
to mitigate this is to filter out arriving tunneled packets with
protocol 41 (IPv6) which do not have the the 192.88.99.1 as the
destination address.
The latter has similar issues: the IPv6 source address could be
checked so the packets to the relay only come from the valid IPv6
addresses which are able to reach the relay anyway.
Savola [Expires June 2003] [Page 15]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
Of these, except in really restricted scenarios, only the first may
be of some interest, and the mitigating the problem by access list is
rather straightforward.
As this threat does not have implications on other than the
organization providing Relay, it is not further analysed.
5.4. Summary
Note: TBD.
o correct and fully-implemented base security features are a pre-
requisite for reasonably safe operation
o being able to spoof IPv4 or IPv6 packets enables one to launch
similar or more powerful attacks even currently
o some 6to4 attacks provide an additional layer of indirection,
which may or may not be useful
o 6to4 as an enabling mechanism does not provide any possibility
for packet multiplication which would affect global Internet,
attacks are generally 1:1
o typically the reflected packets have restricted content, limiting
the usability in an attack
o attacks typically have either 6to4 relay router's address or some
other information which could be used in manual tracing
o attack packets pass through choke point(s), namely (one or more)
6to4 relays; in addition to physical limitations, these could
implement some form 6to4-site-specific traffic limiting
o attacks could in theory be traceable using an extension of
[ITRACE] or [REVITRACE], but as those haven't been specified,
much less used, the point seems rather academic yet
When considering motives for DoS attacks and how to protect against
them (and considering the cost, and whether the protection actually
buys you anything), the following should not be forgotten:
o IPv4 and IPv6 ingress filtering are not likely to be commonplace
for a long time; until it is, you cannot really depend on it
o the real attacker (launching a DoS or DDoS) may not really even
care whether some zombie nodes get found out
o techniques to trace DoS attacks are still in infancy (or not even
there) yet; due to time anything takes to get deployed, it is not
clear whether tracing mechanisms even for basic DoS attack
mechanisms would get reasonably widely deployed before it was
time to (more or less) retire 6to4
o DoS attacks are something that, in practise, operational people
have to be able to deal with anyway
Savola [Expires June 2003] [Page 16]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
6. Solutions
6.1. Generic Approach
6.1.1. Encapsulating IPv6 into IPv4
src and dst MUST pass ipv6-sanity checks, else drop (defined below)
if src=2002
src MUST match src_v4
[ the scenario: 4.1. case 1. or 2. ]
if dst=2002
dst_v4 SHOULD NOT be assigned to the router (avoid misconfigurations)
[ the scenario: 4.1. case 1. ]
fi
elif dst=2002
dst_v4 MAY have to match one of ipv4 equiv. of 6to4 prefixes masked by a
user-specified prefix length (restricting who can use the relay)
[ the scenario: 4.1. case 3. ]
else
drop
[ the scenario: we somehow got a native-native ipv6 packet ]
fi
accept
6.1.2. Decapsulating IPv4 into IPv6
src_v4 and dst_v4 MUST pass ipv4-sanity checks, else drop (defined below)
src and dst MUST pass ipv6-sanity checks, else drop (defined below)
if dst=2002
dst MUST match dst_v4
src_v4 may be restricted to be 6to4_ipv4_anycast [not useful for a long time yet]
[ the scenario: 4.2. case 1. or 2. ]
if src=2002
src MUST match src_v4
dst_v4 SHOULD be assigned to the router (see notes below)
[ the scenario: 4.2. case 1. ]
fi
elif src=2002
src MUST match src_v4
dst_v4 SHOULD be assigned to the router (see notes below)
src_v4 MAY have to match one of ipv4 equiv. of 6to4 prefixes masked by a
user-specified prefix length (restricting who can use the relay)
[ the scenario: 4.2. case 3. ]
else
drop
[ the scenario: we somehow got a native-native ipv6 packet ]
fi
accept
Savola [Expires June 2003] [Page 17]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
6.1.3. IPv4 and IPv6 Sanity Checks
6.1.3.1. IPv4
IPv4 address MUST be a global unicast address, as required by the
6to4 specification. The disallowed addresses include those defined
in [RFC1812], and others widely used and known not to be global.
These are:
o 0.0.0.0/8 (the system has no address assigned yet)
o 10.0.0.0/8 (private)
o 127.0.0.0/8 (loopback)
o 172.16.0.0/12 (private)
o 192.168.0.0/16 (private)
o 169.254.0.0/16 (IANA Assigned DHCP link-local)
o 224.0.0.0/4 (multicast)
o 255.0.0.0/8 (broadcast)
In addition it MUST be checked that the address is not any of the
system's broadcast addresses. This is especially important if the
implementation is made so that it can receive and process
encapsulated IPv4 packets arriving at its broadcast addresses, or try
to send encapsulated IPv4 packets to one of its broadcast addresses.
6.1.3.2. IPv6
IPv6 address MUST NOT be:
o 0::/16 (compatible, mapped addresses, loopback, unspecified,
...)
o fe80::/10 (link-local)
o fec0::/10 (site-local)
o ff02::/16 (link-local multicast)
Other multicast could also be considered for filtering.
In addition, it MUST be checked that equivalent 2002:V4ADDR checks,
where V4ADDR is any of the above IPv4 addresses, will not be passed.
6.1.3.3. Optional Ingress Filtering
In addition, the implementation may perform some form of ingress
filtering (e.g. Unicast Reverse Path Forwarding checks). For
example, if the 6to4 Router has multiple interfaces, of which some
are "internal", receiving either IPv4 or IPv6 packets with source
address belonging to any of these internal networks from the Internet
might be disallowed.
Savola [Expires June 2003] [Page 18]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
If these checks are implemented, it is RECOMMENDED that they default
to disabled.
6.1.3.4. Notes About the Checks
The rule 'dst_v4 SHOULD be assigned to the router' is not needed if
the implementation is made in such a way that it only accepts and
processes encapsulated IPv4 packets arriving on unicast IPv4
addresses, and that if destination address is known to be a local
broadcast address, not try to encapsulate and send packets to it.
Some checks, especially the IPv4/IPv6 Sanity Checks, could be at
least partially implementable with system-level access lists, if one
would like to avoid placing too many restrictions in the 6to4
implementation itself. This depends on how many hooks for ACL's are
in place. In practice it seems like this could not be done
effectively enough unless the access list mechanism is able to parse
the encapsulated packets within IP-IP.
6.2. Simplified Approach
This makes some assumptions about the implementation as pointed above
to simplify the above rules.
6.2.1. Encapsulating IPv6 into IPv4
src and dst MUST pass ipv6-sanity checks, else drop
if src=2002
src MUST match src_v4
elif dst=2002
(accept)
else
drop
fi
accept
6.2.2. Decapsulating IPv4 into IPv6
src_v4 and dst_v4 MUST pass ipv4-sanity checks, else drop
src and dst MUST pass ipv6-sanity checks, else drop
if dst=2002
dst MUST match dst_v4
if src=2002
src MUST match src_v4
fi
elif src=2002
src MUST match src_v4
else
Savola [Expires June 2003] [Page 19]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
drop
fi
accept
6.3. All Relays Must Be Anycast
In this model, all Relay Routers must always use 6to4 IPv4 Anycast
address as the source address.
In order for this to be implementable, every single 6to4 Relay Router
must use only anycast. This may be too significant a deployment
restriction.
Also, in certain circumstances if Unicast RPF is used, the use of
IPv4 anycast as a source address globally will fail the checks and
result in dropped packets.
The checks would be:
if src_v4 is 6to4 ipv4 anycast address
accept [coming from relay]
elif src_v4 is non-global ipv4 unicast address
drop
elif src = 2002
src MUST match src_v4 [coming from other 6to4 routers]
accept
else
drop
fi
7. Problems
7.1. Implementation Considerations with Automatic Tunnels
There is a problem with multiple transition mechanisms if strict
security checks are implemented. This may vary a bit from
implementation to implementation.
Consider three mechanisms using automatic tunneling: 6to4, ISATAP
[ISATAP] and Automatic Tunneling using Compatible Addresses [MECH].
All of these use IP-IP (protocol 41) [IPIP] IPv4 encapsulation with,
more or less, a pseudo-interface.
When a router, which has any two of these enabled, receives an IPv4
encapsulated IPv6 packet:
Savola [Expires June 2003] [Page 20]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
src_v4 = 10.0.0.1
dst_v4 = 20.20.20.20
src = 3ffe:ffff::1
dst = 2002:1010:1010::2
what can it do? How should it decide which transitionary mechanism
this belongs to; there is no "transitionary mechanism number" in IPv6
or IPv4 header to signify this. (This can also be viewed as a
flexibility benefit.)
Without any kind of security checks (in any of implemented methods)
these often just "work" as the mechanisms aren't differentiated but
handled in "one big lump".
Configured tunneling [MECH] does not suffer from this as it is point-
to-point, and based on src/dst pairs of both IPv4 and IPv6 addresses
it can be deduced which logical tunnel interface is in the question.
Solutions for this include not using more than one automatic
tunneling mechanisms in the same system or binding different
mechanisms to different IPv4 addresses.
7.2. Reduced Flexibility
There is a worry about too strict rules limiting the (future)
flexibility of 6to4. If later, for some reason, one would want to
introduce new revolutionary ways to use 6to4, strict checking in all
relevant nodes might prevent it, as new updated version would have to
be deployed everywhere before the new method could be used.
On the other hand, one could argue that 6to4 has always been intended
as an intermediate mechanism, and that future flexibility should not
be critical. However, it is difficult to predict how long the
intermediate period will be.
7.3. Anyone Pretending to Be a Relay Router
6to4 Routers receive traffic from non-6to4 ("native") sources via
6to4 Relays. 6to4 Routers have no way of matching IPv4 source
address of the relay with non-6to4 IPv6 address of the source. In
consequence, anyone can spoof any non-6to4 IPv6 address he wants by
sending traffic, encapsulated, directly to 6to4 Routers.
Of course, as the source IPv4 address may be logged, many may spoof
their IPv4 source address, but the ability to do so is not be
required: it is unlikely that source IPv4 (but rather, the spoofed
IPv6 address) will be logged anywhere -- this would be equivalent to
logging the MAC-address of IP packets.
Savola [Expires June 2003] [Page 21]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
Unfortunately, this problem is very difficult to solve properly.
There have been two rough ideas:
o Every 6to4 Relay MUST configure and use "192.88.99.1" as the
source address of packets that are encapsulated towards 6to4
Routers.
o Every 6to4 Relay MUST participate in an eBGP multi-hop peering
mesh (which can be hierarchical): there more specific routes will
be advertised.
It should be noted that if IPv6 operators do not implement ingress
filtering for IPv6, so that spoofing IPv6 is not more difficult than
spoofing IPv4, these problems have only little impact on the overall
security of 6to4 nodes.
These two are analyzed in order.
7.3.1. 6to4 Relay Routers Must Use 192.88.99.1
This is a natural extension of [6TO4ANY], but it should be noted that
the routers do not necessarily have to _advertise_ 192.88.99.1 or
provide service using that address, just source packets from it. (If
not advertised, the problems with unicast RPF-like mechanisms become
even more difficult.)
6to4 Routers would verify, as pointed above, that encapsulated
traffic coming from non-6to4 sources has "192.88.99.1" as source IPv4
address.
This has the (minor) advantage that the attacker must be able to
spoof to use the source address "192.88.99.1" to be able to inject
spoofed IPv6 traffic. This seems like an advantage (but not
necessarily major enough to warrant it) to the current status, but
one still cannot protect from those who can spoof "192.88.99.1", of
course.
The disadvantage is that it becomes difficult to tell relays apart:
debugging will be very difficult as everyone is using the same
address. Another, much smaller, disadvantage is that sites employing
unicast RPF-checks may lose some packets under certain circumstances
(to be elaborated if this is considered significant).
If that would help any, the Relays could also pick any valid address
from 192.88.99.0/24 -- there would be space to (somehow) distinguish
254 Relay routers.
Savola [Expires June 2003] [Page 22]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
7.3.2. Limited Distribution of More Specific Routes
If 6to4 prefixes more specific than 2002::/16 could be advertised,
the traffic model between native<->6to4 and 6to4<-> could be changed
so that only one Relay would always be used, most often the one
closest to the 6to4 Router.
This approach was rejected due, as IPv6 routing table was not to be
polluted by 6to4 prefixes derived of IPv4 prefixes.
However, the problem can be avoided with some effort: creating a
separate, possibly hierarchical based on IPv6 connections, peering
mesh for 6to4 Relay routers. This could be done by forming eBGP
[BGP] multi-hop peerings between Relays, and advertising more
specific routes (e.g. the same superblocks of IPv4 addresses one
expects to service) to all the other Routers.
In that way, the global routing table would not be impacted at all.
This seems to have at least three potential issues:
o Every Relay should be part of this (if one wants to have some
extra safety that the addresses haven't been spoofed)
o The route from a native source takes the path to the first Relay,
and there (possibly through other Relays) to the last Relay to
tunnel the packet to the 6to4 Router; this adds at least some
latency
o The mechanism of redistributing IPv4 6to4 client addresses to
other relays as 6to4 prefixes need work
Of these, only the last requires more discussion. It could be argued
that this advertising should either be manually configured once (ie.
relatively static prefixes, or generated from IPv4 route-objects in
RADB etc.) or generated automatically (e.g. when a 6to4 Router first
sends a "triggering" packet through the Relay). Further analysis on
this is TBD if necessary.
8. Security Considerations
This draft discusses security considerations.
If proper checks are not implemented, there are significant security
threats ranging from DoS proxy attacks to spoofing and attacks
against 6to4 pseudo-interface. These threats are analyzed in section
5.
Savola [Expires June 2003] [Page 23]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
As can be seen, there are mainly three classes of potential problem
sources:
o 6to4 routers not being able to identify whether relays are
legitimate
o wrong or impartially implemented 6to4 Routers
o relays performing packet laundering
The first is the toughest problem, still under research. The second
can be fixed by ensuring the correctness of implementations; this is
important. Thr third is also a difficult, but a fairly restricted
problem as relays are limited in number.
9. Acknowledgements
Some issues were first brought up in [TRANSAB] and Alain Durand
introduced one specific problem at IETF51 in August 2001 (though
there was some discussion on the list prior to that); these gave the
author the push to start looking into the details of securing 6to4.
Alexey Kuznetsov brought up the implementation problem with IPv6
martian checks. Christian Huitema formulated the rules that rely on
Relays using only anycast. Keith Moore brought up the point about
reduced flexibility. Brian Carpenter, Tony Hain and Vladislav
Yasevich are acknowledged for lengthy discussions. Alain Durand
reminded of relay spoofing problems. Brian Carpenter reminded of the
BGP-based 6to4 router model. Christian Huitema gave a push to a more
complete threat analysis.
10. References
10.1. Normative References
[6TO4] Carpenter, B. and Moore K., "Connection of IPv6 Domains
via IPv4 Clouds", RFC 3056, February 2001.
[6TO4ANY] Huitema, C., "An Anycast Prefix for 6to4 Relay Routers",
RFC 3068, June 2001.
[RFC1918] Rekhter, Y., Moskowitz, R., Karrenberg, D., de Groot, G.
and E. Lear, "Address Allocation for Private Internets",
BCP 5, RFC 1918, February 1996.
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119, March 1997.
Savola [Expires June 2003] [Page 24]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
10.2. Informative References
[ADDRSEL] Draves, R., "Default Address Selection for IPv6",
draft-ietf-ipv6-default-addr-select-09.txt
(work-in-progress), August 2002.
[BGP] Rekhter, Y., Li, T., "A Border Gateway Protocol 4",
RFC1771, March 1995.
[IPIP] Simpson, W., "IP in IP Tunneling", RFC 1853, October
1995.
[ISATAP] Templin, F. et al, "Intra-Site Automatic Tunnel
Addressing Protocol (ISATAP)", draft-ietf-ngtrans-
isatap-06.txt (work-in-progress), October 2002.
[ITRACE] Bellovin, S., Leech, M., Taylor, T., "ICMP Traceback
Messages", draft-ietf-itrace-01.txt (work in progress).
[MECH] Gilligan, R., and Nordmark, E. "Transition Mechanisms for
IPv6 Hosts and Routers", RFC 2893, August 2000.
[REVITRACE] Barros, C., "A Proposal for ICMP Traceback Messages",
http://www.research.att.com/lists/ietf-itrace/2000/09/
msg00044.html.
[RHHASEC] Savola, P., "Security of IPv6 Routing Header and Home
Address Options", draft-savola-ipv6-rh-ha-security-03.txt
(work-in-progress), December 2002.
[SEND] Nikander, P. (Ed.), "IPv6 Neighbor Discovery trust
models and threats", draft-ietf-send-psreq-00.txt
(work-in-progress), October 2002.
[TRANSAB] Hagino, J., "Possible abuse against IPv6 transition
technologies", draft-itojun-ipv6-transition-abuse-01.txt
(expired, work-in-progress), July 2000.
Author's Address
Pekka Savola
CSC/FUNET
Espoo, Finland
EMail: psavola@funet.fi
Savola [Expires June 2003] [Page 25]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
A. Some Trivial Attack Scenarios Outlined
Here, a few trivial attack scenarios are outlined -- ones that are
prevented by implementing checks listed in [6TO4] or in this memo.
When two 6to4 Routers send traffic to each others' domains, packet
sent by RA to RB is like (note: addresses from 8.0.0.0/24 are just
examples of global IPv4 addresses):
src = 2002:0800:0001::aaaa
dst = 2002:0800:0002::bbbb
src_v4 = 8.0.0.1 (added when encapsulated to IPv4)
dst_v4 = 8.0.0.2 (added when encapsulated to IPv4)
When the packet is received by IPv4 stack on RB, it will be
decapsulated so that only src and dst remain, as originally sent by
RA:
src = 2002:0800:0001::aaaa
dst = 2002:0800:0002::bbbb
As every other node is just one hop away (IPv6-wise) and the link-
layer (IPv4) addresses are lost, this may open a lot of possibilities
for misuse.
As an example, unidirectional IPv6 spoofing is made trivial because
nobody can check (without delving into IP-IP packets) whether the
encapsulated IPv6 addresses were authentic (With native IPv6, this
can be done by e.g. RPF-like mechanisms or access lists in upstream
routers).
src = 2002:1234:5678::aaaa (forged)
dst = 2002:0800:0002::bbbb
src_v4 = 8.0.0.1 (added when encapsulated to IPv4)
dst_v4 = 8.0.0.2 (added when encapsulated to IPv4)
A similar attack with "src" being native address is possible even
with the security checks, by having the sender node pretend to be a
6to4 Relay router.
More worries come in to the picture if e.g.
src = ::ffff:[some trusted IPv4 in a private network]
src/dst = ::ffff:127.0.0.1
src/dst = ::1
src/dst = ...
Savola [Expires June 2003] [Page 26]
Internet Draft draft-savola-v6ops-6to4-security-01.txt December 2002
Some implementations might have been careful enough to have designed
the stack to as to avoid the incoming (or reply) packets going to
IPv4 packet processing through special addresses (e.g. IPv4-mapped
addresses), but who can say for all ...
Savola [Expires June 2003] [Page 27]