Use of Timed Efficient Stream Loss-Tolerant Authentication (TESLA) in the Asynchronous Layered Coding (ALC) and NACK-Oriented Reliable Multicast (NORM) Protocols
Internet Engineering Task Force (IETF) V. Roca
Request for Comments: 5776 A. Francillon
Category: Experimental S. Faurite
ISSN: 2070-1721 INRIA
April 2010 Use of Timed Efficient Stream Loss-Tolerant Authentication (TESLA) in
the Asynchronous Layered Coding (ALC) and
NACK-Oriented Reliable Multicast (NORM) Protocols
Abstract
This document details the Timed Efficient Stream Loss-Tolerant
Authentication (TESLA) packet source authentication and packet
integrity verification protocol and its integration within the
Asynchronous Layered Coding (ALC) and NACK-Oriented Reliable
Multicast (NORM) content delivery protocols. This document only
considers the authentication/integrity verification of the packets
generated by the session's sender. The authentication and integrity
verification of the packets sent by receivers, if any, is out of the
scope of this document.
Status of This Memo
This document is not an Internet Standards Track specification; it is
published for examination, experimental implementation, and
evaluation.
This document defines an Experimental Protocol for the Internet
community. This document is a product of the Internet Engineering
Task Force (IETF). It represents the consensus of the IETF
community. It has received public review and has been approved for
publication by the Internet Engineering Steering Group (IESG). Not
all documents approved by the IESG are a candidate for any level of
Internet Standard; see Section 2 of RFC 5741.
Information about the current status of this document, any errata,
and how to provide feedback on it may be obtained at
http://www.rfc-editor.org/info/rfc5776.
Copyright Notice
Copyright (c) 2010 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(http://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect
to this document. Code Components extracted from this document must
include Simplified BSD License text as described in Section 4.e of
the Trust Legal Provisions and are provided without warranty as
described in the Simplified BSD License.
This document may contain material from IETF Documents or IETF
Contributions published or made publicly available before November
10, 2008. The person(s) controlling the copyright in some of this
material may not have granted the IETF Trust the right to allow
modifications of such material outside the IETF Standards Process.
Without obtaining an adequate license from the person(s) controlling
the copyright in such materials, this document may not be modified
outside the IETF Standards Process, and derivative works of it may
not be created outside the IETF Standards Process, except to format
it for publication as an RFC or to translate it into languages other
than English.
1. Introduction
Many applications using multicast and broadcast communications
require that each receiver be able to authenticate the source of any
packet it receives as well as the integrity of these packets. This
is the case with ALC [RFC5775] and NORM [RFC5740], two Content
Delivery Protocols (CDPs) designed to transfer objects (e.g., files)
reliably between a session's sender and several receivers. The NORM
protocol is based on bidirectional transmissions. Each receiver
acknowledges data received or, in case of packet erasures, asks for
retransmissions. On the opposite, the ALC protocol is based on
purely unidirectional transmissions. Reliability is achieved by
means of the cyclic transmission of the content within a carousel
and/or by the use of proactive Forward Error Correction (FEC) codes.
Both protocols have in common the fact that they operate at the
application level, on top of an erasure channel (e.g., the Internet)
where packets can be lost (erased) during the transmission.
The goal of this document is to counter attacks where an attacker
impersonates the ALC or NORM session's sender and injects forged
packets to the receivers, thereby corrupting the objects
reconstructed by the receivers.
Preventing this attack is much more complex in the case of group
communications than it is with unicast communications. Indeed, with
unicast communications, a simple solution exists: the sender and the
receiver share a secret key to compute a Message Authentication Code
(MAC) of all messages exchanged. This is no longer feasible in the
case of multicast and broadcast communications since sharing a group
key between the sender and all receivers implies that any group
member can impersonate the sender and send forged messages to other
receivers.
The usual solution to provide the source authentication and message
integrity services in the case of multicast and broadcast
communications consists of relying on asymmetric cryptography and
using digital signatures. Yet, this solution is limited by high
computational costs and high transmission overheads. The Timed
Efficient Stream Loss-tolerant Authentication (TESLA) protocol is an
alternative solution that provides the two required services, while
being compatible with high-rate transmissions over lossy channels.
This document explains how to integrate the TESLA source
authentication and packet integrity protocol to the ALC and NORM CDP.
Any application built on top of ALC and NORM will directly benefit
from the services offered by TESLA at the transport layer. In
particular, this is the case of File Delivery over Unidirectional
Transport (FLUTE).
For more information on the TESLA protocol and its principles, please
refer to [RFC4082] and [Perrig04]. For more information on ALC and
NORM, please refer to [RFC5775], [RFC5651], and [RFC5740],
respectively. For more information on FLUTE, please refer to
[RMT-FLUTE].
1.1. Scope of This Document
This specification only considers the authentication and integrity
verification of the packets generated by the session's sender. This
specification does not consider the packets that may be sent by
receivers, for instance, NORM's feedback packets. [RMT-SIMPLE-AUTH]
describes several techniques that can be used to that purpose. Since
this is usually a low-rate flow (unlike the downstream flow), using
computing intensive techniques like digital signatures, possibly
combined with a Group MAC scheme, is often acceptable. Finally,
Section 5 explains how to use several authentication schemes in a
given session thanks to the "ASID" (Authentication Scheme IDentifier)
field.
This specification relies on several external mechanisms, for
instance:
o to communicate securely the public key or a certificate for the
session's sender (Section 2.2.2);
o to communicate securely and confidentially the group key, K_g,
used by the Group MAC feature, when applicable (Section 3.3.3).
In some situations, this group key will have to be periodically
refreshed;
o to perform secure time synchronization in indirect mode
(Section 2.3.2) or in direct mode (Section 2.3.1) to carry the
request/response messages with ALC, which is purely
unidirectional;
These mechanisms are required in order to bootstrap TESLA at a sender
and at a receiver and must be deployed in parallel to TESLA.
Besides, the randomness of the Primary Key of the key chain
(Section 3.1.2) is vital to the security of TESLA. Therefore, the
sender needs an appropriate mechanism to generate this random key.
Several technical details of TESLA, like the most appropriate way to
alternate between the transmission of a key disclosure and a
commitment to a new key chain, or the transmission of a key
disclosure and the last key of the previous key chain, or the
disclosure of a key and the compact flavor that does not disclose any
key, are specific to the target use case (Section 3.1.2). For
instance, it depends on the number of packets sent per time interval,
on the desired robustness and the acceptable transmission overhead,
which can only be optimized after taking into account the use-case
specificities.
1.2. Conventions Used in This Document
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].
1.3. Terminology and Notations
The following notations and definitions are used throughout this
document.
1.3.1. Notations and Definitions Related to Cryptographic Functions
Notations and definitions related to cryptographic functions
[RFC4082][RFC4383]:
o PRF is the Pseudo Random Function;
o MAC is the Message Authentication Code;
o HMAC is the keyed-Hash Message Authentication Code;
o F is the one-way function used to create the key chain
(Section 3.1.2.1);
o F' is the one-way function used to derive the HMAC keys
(Section 3.1.2.1);
o n_p is the length, in bits, of the F function's output. This is
therefore the length of the keys in the key chain;
o n_f is the length, in bits, of the F' function's output. This is
therefore the length of the HMAC keys;
o n_m is the length, in bits, of the truncated output of the MAC
[RFC2104]. Only the n_m most significant bits of the MAC output
are kept;
o N is the length of a key chain. There are N+1 keys in a key
chain: K_0, K_1, ..., K_N. When several chains are used, all the
chains MUST have the same length and keys are numbered
consecutively, following the time interval numbering;
o n_c is the number of keys in a key chain. Therefore, n_c = N+1;
o n_tx_lastkey is the number of additional intervals during which
the last key of the old key chain SHOULD be sent, after switching
to a new key chain and after waiting for the disclosure delay d.
These extra transmissions take place after the interval during
which the last key is normally disclosed. The n_tx_lastkey value
is either 0 (no extra disclosure) or larger. This parameter is
sender specific and is not communicated to the receiver;
o n_tx_newkcc is the number of intervals during which the commitment
to a new key chain SHOULD be sent, before switching to the new key
chain. The n_tx_newkcc value is either 0 (no commitment sent
within authentication tags) or larger. This parameter is sender
specific and is not communicated to the receiver;
o K_g is a shared group key, communicated to all group members,
confidentially, during the TESLA bootstrapping (Section 2.2);
o n_w is the length, in bits, of the truncated output of the MAC of
the optional group authentication scheme: only the n_w most
significant bits of the MAC output are kept. n_w is typically
small, a multiple of 32 bits (e.g., 32 bits).
1.3.2. Notations and Definitions Related to Time
Notations and definitions related to time:
o i is the time interval index. Interval numbering starts at 0 and
increases consecutively. Since the interval index is stored as a
32-bit unsigned integer, wrapping to 0 might take place in long
sessions.
o t_s is the sender local time value at some absolute time (in NTP
timestamp format);
o t_r is the receiver local time value at the same absolute time (in
NTP timestamp format);
o T_0 is the start time corresponding to the beginning of the
session, i.e., the beginning of time interval 0 (in NTP timestamp
format);
o T_int is the interval duration (in milliseconds);
o d is the key disclosure delay (in number of intervals);
o D_t is the upper bound of the lag of the receiver's clock with
respect to the clock of the sender;
o S_sr is an estimated bound of the clock drift between the sender
and a receiver throughout the duration of the session;
o D^O_t is the upper bound of the lag of the sender's clock with
respect to the time reference in indirect time synchronization
mode;
o D^R_t is the upper bound of the lag of the receiver's clock with
respect to the time reference in indirect time synchronization
mode;
o D_err is an upper bound of the time error between all the time
references, in indirect time synchronization mode;
o NTP timestamp format consists in a 64-bit unsigned fixed-point
number, in seconds relative to 0h on 1 January 1900. The integer
part is in the first 32 bits, and the fraction part in the last 32
bits [RFC1305].
2. Using TESLA with ALC and NORM: General Operations
2.1. ALC and NORM Specificities That Impact TESLA
The ALC and NORM protocols have features and requirements that
largely impact the way TESLA can be used.
In the case of ALC:
o ALC is massively scalable: nothing in the protocol specification
limits the number of receivers that join a session. Therefore, an
ALC session potentially includes a huge number (e.g., millions or
more) of receivers;
o ALC can work on top of purely unidirectional transport channels:
this is one of the assets of ALC, and examples of unidirectional
channels include satellite (even if a back channel might exist in
some use cases) and broadcasting networks like Digital Video
Broadcasting - Handhelds / Satellite services to Handhelds (DVB-
H/SH);
o ALC defines an on-demand content delivery model [RFC5775] where
receivers can arrive at any time, at their own discretion,
download the content and leave the session. Other models (e.g.,
push or streaming) are also defined;
o ALC sessions are potentially very long: a session can last several
days or months during which the content is continuously
transmitted within a carousel. The content can be either static
(e.g., a software update) or dynamic (e.g., a web site).
Depending on the use case, some of the above features may not apply.
For instance, ALC can also be used over a bidirectional channel or
with a limited number of receivers.
In the case of NORM:
o NORM has been designed for medium-size sessions: indeed, NORM
relies on feedback messages and the sender may collapse if the
feedback message rate is too high;
o NORM requires a bidirectional transport channel: the back channel
is not necessarily a high-data rate channel since the control
traffic sent over it by a single receiver is an order of magnitude
lower than the downstream traffic. Networks with an asymmetric
connectivity (e.g., a high-rate satellite downlink and a low-rate
return channel) are appropriate.
2.2. Bootstrapping TESLA
In order to initialize the TESLA component at a receiver, the sender
MUST communicate some key information in a secure way, so that the
receiver can check the source of the information and its integrity.
Two general methods are possible:
o by using an out-of-band mechanism, or
o by using an in-band mechanism.
The current specification does not recommend any mechanism to
bootstrap TESLA. Choosing between an in-band and out-of-band scheme
is left to the implementer, depending on the target use case.
However, it is RECOMMENDED that TESLA implementations support the use
of the in-band mechanism for interoperability purposes.
2.2.1. Bootstrapping TESLA with an Out-Of-Band Mechanism
For instance, [RFC4442] describes the use of the MIKEY (Multimedia
Internet Keying) protocol to bootstrap TESLA. As a side effect,
MIKEY also provides a loose time synchronization feature from which
TESLA can benefit. Other solutions, for instance, based on an
extended session description, are possible, on the condition that
these solutions provide the required security level.
2.2.2. Bootstrapping TESLA with an In-Band Mechanism
This specification describes an in-band mechanism. In some use
cases, it might be desired that bootstrapping take place without
requiring the use of an additional external mechanism. For instance,
each device may feature a clock with a known time-drift that is
negligible in front of the time accuracy required by TESLA, and each
device may embed the public key of the sender. It is also possible
that the use case does not feature a bidirectional channel that
prevents the use of out-of-band protocols like MIKEY. For these two
examples, the exchange of a bootstrap information message (described
in Section 3.4.1) and the knowledge of a few additional parameters
(listed below) are sufficient to bootstrap TESLA at a receiver.
Some parameters cannot be communicated in-band. In particular:
o the sender or group controller MUST either communicate the public
key of the sender or a certificate (which also means that a PKI
has been set up) to all receivers, so that each receiver be able
to verify the signature of the bootstrap message and direct time
synchronization response messages (when applicable).
o when time synchronization is performed with NTP/SNTP (Simple
Network Time Protocol), the sender or group controller MUST
communicate the list of valid NTP/SNTP servers to all the session
members (sender included), so that they are all able to
synchronize themselves on the same NTP/SNTP servers.
o when the Group MAC feature is used, the sender or group controller
MUST communicate the K_g group key to all the session members
(sender included). This group key may be periodically refreshed.
The way these parameters are communicated is out of the scope of this
document.
2.3. Setting Up a Secure Time Synchronization
The security offered by TESLA heavily relies on time. Therefore, the
session's sender and each receiver need to be time synchronized in a
secure way. To that purpose, two general methods exist:
o direct time synchronization, and
o indirect time synchronization.
It is also possible that a given session includes receivers that use
the direct time synchronization mode while others use the indirect
time synchronization mode.
2.3.1. Direct Time Synchronization
When direct time synchronization is used, each receiver asks the
sender for a time synchronization. To that purpose, a receiver sends
a direct time synchronization request (Section 4.2.2.1). The sender
then directly answers each request with a direct time synchronization
response (Section 3.4.2), signing this reply. Upon receiving this
response, a receiver first verifies the signature, and then
calculates an upper bound of the lag of his clock with respect to the
clock of the sender, D_t. The details on how to calculate D_t are
given in Section 2.4.1.
This synchronization method is both simple and secure. Yet, there
are two potential issues:
o a bidirectional channel must exist between the sender and each
receiver, and
o the sender may collapse if the incoming request rate is too high.
Relying on direct time synchronization is not expected to be an issue
with NORM since (1) bidirectional communications already take place,
and (2) NORM scalability is anyway limited. Yet, it can be required
that a mechanism, that is out of the scope of this document, be used
to spread the transmission of direct time synchronization request
messages over time if there is a risk that the sender may collapse.
But direct time synchronization is potentially incompatible with ALC
since (1) there might not be a back channel, and (2) there are
potentially a huge number of receivers and therefore a risk that the
sender will collapse.
2.3.2. Indirect Time Synchronization
When indirect time synchronization is used, the sender and each
receiver must synchronize securely via an external time reference.
Several possibilities exist:
o sender and receivers can synchronize through an NTPv3 (Network
Time Protocol version 3) [RFC1305] hierarchy of servers. The
authentication mechanism of NTPv3 MUST be used in order to
authenticate each NTP message individually. It prevents, for
instance, an attacker from impersonating an NTP server;
o they can synchronize through an NTPv4 (Network Time Protocol
version 4) [NTP-NTPv4] hierarchy of servers. The Autokey security
protocol of NTPv4 MUST be used in order to authenticate each NTP
message individually;
o they can synchronize through an SNTPv4 (Simple Network Time
Protocol version 4) [RFC4330] hierarchy of servers. The
authentication features of SNTPv4 must then be used. Note that
TESLA only needs a loose (but secure) time synchronization, which
is in line with the time synchronization service offered by SNTP;
o they can synchronize through a GPS or Galileo (or similar) device
that also provides a high precision time reference. Spoofing
attacks on the GPS system have recently been reported. Depending
on the use case, the security achieved will or will not be
acceptable;
o they can synchronize thanks to a dedicated hardware, embedded on
each sender and receiver, that provides a clock with a time-drift
that is negligible in front of the TESLA time accuracy
requirements. This feature enables a device to synchronize its
embedded clock with the official time reference from time to time
(in an extreme case once, at manufacturing time), and then to
remain autonomous for a duration that depends on the known maximum
clock drift.
A bidirectional channel is required by the NTP/SNTP schemes. On the
opposite, with the GPS/Galileo and high precision clock schemes, no
such assumption is made. In situations where ALC is used on purely
unidirectional transport channels (Section 2.1), using the NTP/SNTP
schemes is not possible. Another aspect is the scalability
requirement of ALC, and to a lesser extent of NORM. From this point
of view, the above mechanisms usually do not raise any problem,
unlike the direct time synchronization schemes. Therefore, using
indirect time synchronization can be a good choice. It should be
noted that the NTP/SNTP schemes assume that each client trusts the
sender and accepts aligning its NTP/SNTP configuration to that of the
sender. If this assumption does not hold, the sender SHOULD offer an
alternative solution.
The details on how to calculate an upper bound of the lag of a
receiver's clock with respect to the clock of the sender, D_t, are
given in Section 2.4.2.
2.4. Determining the Delay Bounds
Let us assume that a secure time synchronization has been set up.
This section explains how to define the various timing parameters
that are used during the authentication of received packets.
2.4.1. Delay Bound Calculation in Direct Time Synchronization Mode
In direct time synchronization mode, synchronization between a
receiver and the sender follows the following protocol [RFC4082]:
o The receiver sends a direct time synchronization request message
to the sender, that includes t_r, the receiver local time at the
moment of sending (Section 4.2.2.1).
o Upon receipt of this message, the sender records its local time,
t_s, and sends to the receiver a direct time synchronization
response that includes t_r (taken from the request) and t_s,
signing this reply (Section 3.4.2).
o Upon receiving this response, the receiver first verifies that he
actually sent a request with t_r and then checks the signature.
Then he calculates D_t = t_s - t_r + S_sr, where S_sr is an
estimated bound of the clock drift between the sender and the
receiver throughout the duration of the session. This document
does not specify how S_sr is estimated.
After this initial synchronization, at any point throughout the
session, the receiver knows that: T_s < T_r + D_t, where T_s is the
current time at the sender and T_r is the current time at the
receiver.
2.4.2. Delay Bound Calculation in Indirect Time Synchronization Mode
In indirect time synchronization, the sender and the receivers must
synchronize indirectly using one or several time references.
2.4.2.1. Single Time Reference
Let us assume that there is a single time reference.
1. The sender calculates D^O_t, the upper bound of the lag of the
sender's clock with respect to the time reference. This D^O_t
value is then communicated to the receivers (Section 3.2.1).
2. Similarly, a receiver R calculates D^R_t, the upper bound of the
lag of the receiver's clock with respect to the time reference.
3. Then, for receiver R, the overall upper bound of the lag of the
receiver's clock with respect to the clock of the sender, D_t, is
the sum: D_t = D^O_t + D^R_t.
The D^O_t and D^R_t calculation depends on the time synchronization
mechanism used (Section 2.3.2). In some cases, the synchronization
scheme specifications provide these values. In other cases, these
parameters can be calculated by means of a scheme similar to the one
specified in Section 2.4.1, for instance, when synchronization is
achieved via a group controller [RFC4082].
2.4.2.2. Multiple Time References
Let us now assume that there are several time references (e.g.,
several NTP/SNTP servers). The sender and receivers first
synchronize with the various time references, independently. It
results in D^O_t and D^R_t. Let D_err be an upper bound of the time
error between all of the time references. Then, the overall value of
D_t within receiver R is set to the sum: D_t = D^O_t + D^R_t + D_err.
In some cases, the D_t value is part of the time synchronization
scheme specifications. For instance, NTPv3 [RFC1305] defines
algorithms that are "capable of accuracies in the order of a
millisecond, even after extended periods when synchronization to
primary reference sources has been lost". In practice, depending on
the NTP server stratum, the accuracy might be a little bit worse. In
that case, D_t = security_factor * (1ms + 1ms), where the
security_factor is meant to compensate several sources of inaccuracy
in NTP. The choice of the security_factor value is left to the
implementer, depending on the target use case.
2.5. Cryptographic Parameter Values
The F (resp. F') function output length is given by the n_p (resp.
n_f) parameter. The n_p and n_f values depend on the PRF function
chosen, as specified below:
+------------------------+---------------------+
| PRF name | n_p and n_f |
+------------------------+---------------------+
| HMAC-SHA-1 | 160 bits (20 bytes) |
| HMAC-SHA-224 | 224 bits (28 bytes) |
| HMAC-SHA-256 (default) | 256 bits (32 bytes) |
| HMAC-SHA-384 | 384 bits (48 bytes) |
| HMAC-SHA-512 | 512 bits (64 bytes) |
+------------------------+---------------------+
The computing of regular MAC (resp. Group MAC) makes use of the n_m
(resp. n_w) parameter, i.e., the length of the truncated output of
the function. The n_m and n_w values depend on the MAC function
chosen, as specified below: