0RTT Key Exchange with Full Forward Secrecy
 21 Citations
 2.4k Downloads
Abstract
Reducing latency overhead while maintaining critical security guarantees like forward secrecy has become a major design goal for key exchange (KE) protocols, both in academia and industry. Of particular interest in this regard are 0RTT protocols, a class of KE protocols which allow a client to send cryptographically protected payload in zero roundtrip time (0RTT) along with the very first KE protocol message, thereby minimizing latency. Prominent examples are Google’s QUIC protocol and the upcoming TLS protocol version 1.3. Intrinsically, the main challenge in a 0RTT key exchange is to achieve forward secrecy and security against replay attacks for the very first payload message sent in the protocol. According to cryptographic folklore, it is impossible to achieve forward secrecy for this message, because the session key used to protect it must depend on a nonephemeral secret of the receiver. If this secret is later leaked to an attacker, it should intuitively be possible for the attacker to compute the session key by performing the same computations as the receiver in the actual session.
In this paper we show that this belief is actually false. We construct the first 0RTT key exchange protocol which provides full forward secrecy for all transmitted payload messages and is automatically resilient to replay attacks. In our construction we leverage a puncturable key encapsulation scheme which permits each ciphertext to only be decrypted once. Fundamentally, this is achieved by evolving the secret key after each decryption operation, but without modifying the corresponding public key or relying on shared state.
Our construction can be seen as an application of the puncturable encryption idea of Green and Miers (S&P 2015). We provide a new generic and standardmodel construction of this tool that can be instantiated with any selectively secure hierarchical identitybased key encapsulation scheme.
Keywords
Replay Attack Forward Secrecy Transport Layer Security Server Configuration Asynchronous Messaging1 Introduction
Authenticated key exchange and TLS. The Transport Layer Security (TLS) protocol is the most important cryptographic security mechanism on the Internet today, with TLS 1.2 being the most recent standardized version [16] and TLS 1.3 under development [40]. As one core functionality TLS provides an (authenticated) key exchange (AKE) which allows two remote parties to establish a shared cryptographic key over an insecure channel like the Internet. The study of provable security guarantees for AKE protocols was initiated by the seminal work of Bellare and Rogaway [4]; the huge body of work on cryptographic analyses of the TLS key exchange(s) includes [5, 17, 26, 28].
The Demand for lowlatency key exchange. Classical AKE protocols like TLS incur a considerable latency overhead due to exchanging a relatively large number of protocol messages before the first actual (application) data messages can be transmitted under cryptographic protection. Latency is commonly measured in roundtrip time (RTT), indicating the number of rounds/round trips messaging has to take before the first application data can be sent. Even very efficient examples of highperformance AKE protocols like HMQV [27] need at least two messages (i.e., 1RTT) before either party can compute the session key.
0RTT key exchange. Reducing the latency overhead of key exchange protocols to zero roundtrip time (0RTT) while maintaining reasonable security guarantees has become a major design goal both in academia [23, 29, 36, 44] and industry [38, 40].^{1} In terms of practical designs, the two principal protocols are Google’s QUIC protocol [38] and the 0RTT mode drafted for the upcoming TLS version 1.3 [40]. While the latter is still in development, QUIC is already implemented in recent versions of the Google Chrome and Opera web browsers, is currently used on Google’s web servers, and has been proposed as an IETF standard (July 2015).
 1.
From prior communication (which may be a key exchange or some outofband communication), the client obtains a “mediumlived” (usually a couple of days) server configuration. This server configuration contains a Diffie–Hellman share \(g^s\) (with g being a generator of an algebraic group) for which the server knows s, and is signed under a public signing key certified for belonging to the server.
 2.
In the 0RTT key exchange, the client knowing \(g^s\) now picks a secret exponent x at random and sends the share \(g^x\) to the server. It also directly computes a preliminary, 0RTT key \(K_1\) from the Diffie–Hellman value \(g^{xs}\). In immediate application, this key can be used to send cryptographically protected (0RTT) application data along with the client’s keyexchange message.
 3.
The server responds with a freshly chosen, ephemeral Diffie–Hellman share \(g^y\) which is used by both the server and the client to compute the actual session key \(K_2\) from \(g^{xy}\). All further communication throughout the session is subsequently protected under \(K_2\).
Issues with 0RTT key exchange. As outlined, the 0RTT keyexchange design elegantly allows clients to initiate a secure connection with zero latency overhead, addressing an important practical problem. Unfortunately, all protocols that follow this format—including QUIC and TLS 1.3 as well as academic approaches [23, 44]—face at least one of the following two very undesirable drawbacks.
Forward Secrecy. Recall that forward secrecy essentially demands that transmitted data remains secret even if an attacker learns the secret key of one communication partner. From contemporary insight, this is considered a standard and crucial security goal of modern key exchange protocols, as it addresses data protection in the presence of passive key compromises or mass surveillance. Observe that a 0RTT key exchange of the form outlined above, however, cannot provide forward secrecy for the 0RTT application data transmitted from the client to the server. As such data is protected under the key \(K_1\), derived from \(g^{xs}\), an attacker which eavesdrops on the communication and later compromises the server’s secret exponent s (possibly long after the session has finished) can easily compute \(K_1\) and thus decrypt the 0RTT data sent. This drawback is clearly acknowledged in the design documents of QUIC and TLS 1.3 and one of the main reasons to upgrade to a second, forwardsecret key \(K_2\). Notably, the lack of forward secrecy for TLS 1.3 0RTT is true of both the original Diffie–Hellmanbased and the latest preshared key (PSK) variants of the protocol, albeit under different assumptions on which key is learned by the attacker [29, 39, 40, 42].
In 2005, Krawczyk stated that it was not possible to obtain forward secrecy for implicitlyauthenticated 2message protocols in a publickey authentication context, if there was no preexisting shared state [27]. Subsequent works referenced this idea prominently, but often dropped one or more of the original conditions [8, 11, 30]. Despite modeling changes and arguments to the contrary in relation to 1round protocols [13, 15], and work on forward secrecy for noninteractive keyexchange (NIKE) protocols [37], the assumption that forward secrecy is fundamentally impossible under limited rounds has perpetuated. In particular, the QUIC crypto specification accepts an “upper bound on the forward security of the connection” for 0RTT handshakes [31]. Likewise, this limitation is accepted as seemingly inherent in academic 0RTT designs [23, 44], and early discussions around the development of TLS 1.3 go so far as to claim that forward secrecy “can’t be done in 0RTT” [43].
Replay Attacks. In a replay attack, an attacker aims at making the receiver accept the same payload twice. Specifically, replay attacks in the example 0RTT protocol given can take the form of replaying the client’s Diffie–Hellman share \(g^x\) or the 0RTT data sent. Observe that, without further countermeasures, an adversary can simply replay (potentially multiple times) a recorded client message \(g^x\), making the server derive the same key \(K_1\) as in the original connection, and then replay the client’s 0RTT data which the server can correctly decrypt and would therefore process. Depending on the application logic, such replays can lead to severe security issues. For example, an authenticated request (e.g., via login credentials or cookie tokens) might allow an adversary to replay client actions like online orders or financial transactions.
One potential countermeasure, implemented in QUIC, is essentially to store all seen client values \(g^x\) (in a certain time frame encoded in an additional nonce value) in order to detect and reject repeated requests with the same value and nonce.^{2} Notably, this solution induces a substantial management overhead and arguably is acceptable only for certain server configurations. As such, the solution is not elegant, but effectively prevents the same key from being accepted twice by a server. We remark, though, that on a higher level applications may resend data under a laterderived key in common web scenarios, essentially rendering replay attacks on the application layer unavoidable in such cases [19, 41].
Lowlatency keyexchange designs proposed thus far widely accepted the aforementioned drawbacks on forward secrecy and replay protection as inherent to the 0RTT environment. This assumption paves the way for the following research question for the design of modern, lowlatency authenticated keyexchange protocols: Can a keyexchange protocol establish a cryptographic key in 0RTT while upholding strong forwardsecrecy and replay protection guarantees?
Contributions. In this work we introduce the notion of forwardsecret onepass key exchange and a generic construction of such a protocol, resolving the aforementioned open problem. Notable features of this protocol are summarized as follows.

The protocol provides full forward secrecy, even for the first message transmitted from the client to the server, and is automatically resilient to replay attacks. We provide a rigorous security analysis for which we develop a novel keyexchange model (in the style of Bellare and Rogaway [4]) that captures the peculiarities of forward secrecy and replay protection in 0RTT key exchange.

The protocol has the simplest message flow imaginable: the client encrypts a session key and sends it to the server. We do not need to distinguish between preliminary and final keys but only derive a single session key. The forward secrecy and replay security of the protocol stem from the fact that the longterm secret key of this scheme is evolved.

The construction and security proof are completely generic, based on any onetime signature scheme and any hierarchical identitybased key encapsulation scheme (HIBKEM) that needs to provide only a weak form of selectiveID security. This allows for flexible instantiation of the protocol with arbitrary cryptographic constructions of these primitives, adjusted with suitable deployment and efficiency parameters for a given application, and based on various hardness assumptions.

The construction and its security analysis are completely independent of a particular instantiation of building blocks, immediately yielding the first postquantum secure 0RTT key exchange protocol, via instantiation of the protocol with suitable latticebased building blocks, such as the HIBE from [1] and the onetime signature from [34].

More generally, by instantiating the protocol with different HIBKEM schemes, one can easily obtain different “cipher suites”, with different security and performance characteristics. Replacement of a cipher suite is easy, as it does not require a new security analysis of the protocol. In contrast, several consecutive research papers were required to establish the security of only the most important basic cipher suites of TLS [26, 28, 32].

We make the conceptual observation that the tool of forwardsecret puncturable publickey encryption can be leveraged to enable forwardsecret 0RTT AKE.

We carve out puncturable forwardsecret key encapsulation as a versatile building block and build it in a generic fashion from any HIBKEM scheme, in the standard model, and from a wide range of assumptions. In contrast, the cunning, but involved construction by Green and Miers [21] blends the attributebased encryption scheme of Ostrovsky, Sahai, and Waters [35] with forwardsecret encryption [9]. It therefore relies on specific assumptions and, using the FujisakiOkamoto transform [20] to achieve CCAsecurity, relies on the randomoracle model.

We formalize 0RTT key exchange security with forward secrecy. This is a nontrivial extension of previous models (particularly [24]) in that it needs to take evolving state, (semi)synchronized time, and accordingly conditioned forward secrecy into account in the security experiment.
We consider the established concepts as valuable towards the understanding of forwardsecret 0RTT key exchange, its foundations, and its connection to, in particular, asynchronous messaging.
Highlevel protocol description. The basic outline of our protocol is the simplest one can imagine. We use a publickey key encapsulation mechanism (KEM)^{3} to transport a random session key from the client to the server. That is, the server is in possession of a longterm key pair \(( pk , sk )\) for the KEM, and the client uses \( pk \) to encapsulate a key. This immediately yields a 0RTT protocol, because we can send encrypted payload data along with the encapsulated key. However, of course, it does not yet provide forward secrecy or security against replay attacks.
The key idea to achieve these additional properties is not to modify the protocol, but to modify the way the server stores and processes its secret key. More precisely, we construct and use a special puncturable forwardsecure KEM (PFSKEM). Consider a server with longterm secret key \( sk \). When receiving an encapsulated session key in ciphertext \(c_1\), the server can use this scheme to proceed as follows.
 1.
It decrypts \(c_1\) using \( sk \).
 2.
The server then derives a new secret key \( sk _{\setminus c_1}\) from \( sk \), which is “punctured at position \(c_1\)”. This means that \( sk _{\setminus c_1}\) can be used to decrypt all ciphertexts except for \(c_1\).
 3.
Finally, the server deletes \( sk \).
This process is executed repeatedly for all ciphertexts received by the server. That is, when the server receives a second ciphertext \(c_2\) from the same or a different client, it again “punctures” \( sk _{\setminus c_1 }\) to obtain a new secret key \( sk _{\setminus c_1,c_2 }\), which can be used to decrypt all ciphertexts except for \(c_1\) and \(c_2\). Note that this yields forward secrecy, because an attacker that obtains \( sk _{\setminus c_1,c_2 }\) will not be able to use this key to decrypt \(c_1\) or \(c_2\), and thus will not be able to learn the session key of previous sessions.
The drawback of using this approach naïvely is that the size of secret keys grows linearly with the number of sessions, which is of course impractical. For efficiency reasons, we therefore add an additional time component to the protocol, which requires only loosely synchronized clocks between client and server. Within each time slot, the size of the secret key grows linearly with the number of sessions. However, at the end of the time slot, the server is able to “purge” the key, which reduces its size back to a factor logarithmic in the number of time intervals. We stress that the loose time synchronization is included in our protocol’s design only for efficiency reasons, but is not needed to achieve the desired security goals.
A particularly beneficial aspect of this approach is that the server’s public key \( pk \) remains static over its entire lifetime (which would typically be 1–2 years in practice, but longer lifetimes are easily possible), because there is no QUIClike server configuration that needs to be frequently updated at clientside. Thus, this yields a protocol without the need to frequently replace the server configuration \(g^s\) at the client.
The maximal size of punctured secret keys, and thus the storage requirement of the protocol, depends on the size of time slots. Longer time slots (several hours or possibly even a few days, depending on the number of connections during this time) require more storage, but only loosely synchronized clocks. Short time slots (a few minutes) require less storage, but more precisely synchronized clocks. These parameters can be chosen depending on the individual characteristics of a server and the services that it provides.
Related work. The idea of forwardsecret encryption based on hierarchical identitybased encryption is due to Canetti, Halevi, and Katz [9]. Pointcheval and Sanders [37] studied forward secrecy for noninteractive keyexchange protocols based on multilinear maps. Both approaches however only provide coarsegrained forward secrecy with respect to time periods, whereas we aim at a finegrained, immediate notion of forward secrecy in the setting of key exchange.
With a similar goal in mind, the previously mentioned work of Green and Miers [21] achieves forward secrecy in the context of asynchronous messaging.^{4} Their construction blends the attributebased encryption scheme of Ostrovsky, Sahai, and Waters [35] with the scheme of Canetti, Halevi, and Katz [9] or, alternatively, with the scheme of Boneh, Boyen, and Goh [7]. This makes their scheme relatively complex and bound to specific algebraic settings and complexity assumptions. Moreover, their scheme achieves only CPA security, and requires the random oracle model [3] and the FujisakiOkamoto transform [20] to achieve CCA security. In contrast, we describe a simple, natural and directly CCAsecure construction based on any hierarchical identitybased KEM (HIBKEM), which can be instantiated from any HIBKEM that only needs to provide weak selectiveID security.
The security of the QUIC protocol was formally analyzed by Fischlin and Günther [18] as well as Lychev et al. [33]. Krawczyk and Wee [29] described the OPTLS protocol as a foundation for TLS 1.3, including a 0RTT mode. For TLS 1.3, Cremers et al. [14] conducted a toolsupported analysis of TLS 1.3 including a draft 0RTT handshake mode, and Fischlin and Günther [19] analyzed the provable security of both Diffie–Hellman and PSKbased 0RTT handshake drafts. Foundational definitions and generic constructions of 0RTT key exchange from other cryptographic primitives were given by Hale et al. [23]. All these works consider security models and constructions without forward secrecy of the first message. In a related, but different direction, CohnGordon et al. [12] consider postcompromise security for keyexchange protocols that use key ratcheting, where the session key is frequently updated during the lifetime of a single session.
Outline of the paper. Section 2 introduces the necessary building blocks for our construction as well as puncturable forwardsecret key encapsulation (PFSKEM), before we provide a generic PFSKEM construction from HIBE. We formalize forwardsecret onepass key exchange protocols (FSOPKE) in Sect. 3, together with a corresponding security model. In Sect. 4 we provide a generic construction of FSOPKE with server authentication from PFSKEM and prove its security in the FSOPKE model. In Sect. 5 we analyze the size of keys and messages for different deployment parameters.
2 Generic Construction of Puncturable Encryption
2.1 Building Blocks
Let us begin with recapping the definition and security of onetime signature schemes, as well as hierarchical identitybased key encapsulation schemes.
Definition 1
(OneTime Signatures). A onetime signature scheme \(\mathsf {OTSIG}\) consists of three probabilistic polynomialtime algorithms (\(\mathsf {OTSIG.KGen}\), \(\mathsf {OTSIG.Sign}\), \(\mathsf {OTSIG.Vfy}\)).

\(\mathsf {OTSIG.KGen}(1^\lambda )\) takes as input a security parameter \(\lambda \) and outputs a public key \( pk _{OT}\) and a secret key \( sk _{OT}\).

\(\mathsf {OTSIG.Sign}( sk _{OT},m)\) takes as input a secret key and a message \(m\in \{0,1\}^n\). Output is a signature \(\sigma \).

\(\mathsf {OTSIG.Vfy}( pk _{OT}, m, \sigma )\) input is a public key, a message \(m\in \{0,1\}^n\) and a signature \(\sigma \). If \(\sigma \) is a valid signature for m under \( pk _{OT}\), then the algorithm outputs 1, else 0.
Consider the following security experiment \(G_{\mathcal {A},\mathsf {OTSIG}}^{\mathsf {sEUF{\text { }}1{\text { }}\mathsf {CMA}}}(\lambda )\) played between a challenger \(\mathcal {C}\) and an adversary \(\mathcal {A}\).
 1.
The challenger \(\mathcal {C}\) computes \(( pk _{OT}, sk _{OT})\mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\mathsf {OTSIG.KGen}(1^\lambda )\) and runs \(\mathcal {A}\) with input \( pk _{OT}\).
 2.
\(\mathcal {A}\) may query one arbitrary message m to the challenger. \(\mathcal {C}\) replies with \(\sigma \mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\mathsf {OTSIG.Sign}( sk _{OT},m)\).
 3.\(\mathcal {A}\) eventually outputs a message \(m^*\) and a signature \(\sigma ^*\). We denote the event that \(\mathsf {OTSIG.Vfy}( pk _{OT}, m^*, \sigma ^*)=1\) and \((m^*,\sigma ^*)\ne (m,\sigma )\) by$$\begin{aligned} G_{\mathcal {A},\mathsf {OTSIG}}^{\mathsf {sEUF{\text { }}1{\text { }}\mathsf {CMA}}}(\lambda ) = 1. \end{aligned}$$
Definition 2
In our generic construction we use a hierarchical identitybased key encapsulation scheme (\(\mathsf {HIBKEM}\)) [6]. \(\mathsf {HIBKEM}\) schemes enable a user to encapsulate a symmetric key with the recipients identity. An identity at depth t in the hierarchical tree is represented by a vector \(\mathsf {ID}_{\vert t}=(I_1,\cdots , I_t)\). Ancestors of the identity \(\mathsf {ID}_{\vert t}\) are identities represented by vectors \(\mathsf {ID}_{\vert s}=(J_1,\cdots ,J_s)\) with \(1\le s<t\) and \(I_i=J_i\) for \(1\le i \le s\).
Definition 3
(HIBKEM [6]). A hierarchical identitybased key encapsulation scheme \(\mathsf {HIBKEM}\) consists of four probabilistic polynomialtime algorithms \((\mathsf {HIBKEM.KGen}, \mathsf {HIBKEM.Del}, \mathsf {HIBKEM.Enc}, \mathsf {HIBKEM.Dec})\).

\(\mathsf {HIBKEM.KGen}(1^\lambda )\) takes as input a security parameter \(\lambda \) and outputs an a public key \( pk \) and an initial secret key (or master key) msk, which we refer as the private key at depth 0. We assume that \( pk \) implicitly defines the identity space \(\mathcal {ID}\) and the key space \(\mathcal {K}\).

\(\mathsf {HIBKEM.Del}(\mathsf {ID}_{\vert t}, sk _{\mathsf {ID}'\vert s})\) takes as input an identity \(\mathsf {ID}_{\vert t}\) at depth t and the private key of an ancestor identity \(\mathsf {ID}'_{\vert s}\) at depth \(s<t\) or the master key msk. Output is a secret key \( sk _{\mathsf {ID}\vert t}\).

\(\mathsf {HIBKEM.Enc}( pk , \mathsf {ID})\) takes a input the public key \( pk \) and the target \(\mathsf {ID}\). The algorithm outputs a ciphertext \(\mathsf {CT}\) and a symmetric key \(\mathsf {K}\).

\(\mathsf {HIBKEM.Dec}( sk _{\mathsf {ID}}, \mathsf {CT})\) takes as input a secret key \(sk_{\mathsf {ID}}\) and a ciphertext \(\mathsf {CT}\). Output is a symmetric key \(\mathsf {K}\) or \(\bot \) if decryption fails.
Consider the following selectiveID CPA security experiment \(G_{\mathcal {A},\mathsf {HIBKEM}}^{\mathsf {IND{\text { }}sID{\text { }}CPA}}(\lambda )\) played between a challenger \(\mathcal {C}\) and an adversary \(\mathcal {A}\).
 1.
\(\mathcal {A}\) outputs the target identity \(\mathsf {ID}^*\) on which it wants to be challenged.
 2.
The challenger generates the system parameters and computes \(( pk , msk)\mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\mathsf {HIBKEM.KGen}(1^\lambda )\). \(\mathcal {C}\) generates \((\mathsf {K}_0,\mathsf {CT}^*)\mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\mathsf {HIBKEM.Enc}( pk , \mathsf {ID}^*)\) and \(K_1\mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\mathcal {K}\). Then the challenger sends \((K_b, \mathsf {CT}^*, pk \)) to \(\mathcal {A}\) where \(b\mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\{0,1\}\).
 3.
\(\mathcal {A}\) may query an \(\mathsf {HIBKEM.Del}\) oracle. The \(\mathsf {HIBKEM.Del}\) oracle outputs the secret key of a requested identity \(\mathsf {ID}\). The only restriction is, that the attacker \(\mathcal {A}\) is not allowed to ask the \(\mathsf {HIBKEM.Del}\) oracle for the secret key of \(\mathsf {ID}^*\) or any ancestor identity of \(\mathsf {ID}^*\).
 4.Finally, \(\mathcal {A}\) eventually outputs a guess \(b'\). We denote the event that \(b=b'\) by$$\begin{aligned} G_{\mathcal {A},\mathsf {HIBKEM}}^{\mathsf {IND{\text { }}sID{\text { }}CPA}}(\lambda ) = 1 \end{aligned}$$
Definition 4
2.2 Puncturable ForwardSecret Key Encapsulation
We now formally introduce the definition of a puncturable forwardsecret key encapsulation (PFSKEM) scheme as well as its corresponding correctness definition and security notion.
Definition 5
(PFSKEM). A puncturable forwardsecret key encapsulation scheme \(\mathsf {PFSKEM}\) consists of five probabilistic polynomialtime algorithms (\(\mathsf {PFSKEM.KGen}\), \(\mathsf {PFSKEM.Enc}\), \(\mathsf {PFSKEM.PnctCxt}\), \(\mathsf {PFSKEM.Dec}\), \(\mathsf {PFSKEM.PnctInt}\)).

\(\mathsf {PFSKEM.KGen}(1^\lambda )\) takes as input a security parameter \(\lambda \) and outputs a public key \( PK \) and an initial secret key \( SK \).

\(\mathsf {PFSKEM.Enc}( PK , \tau )\) takes as input a public key and a time period \(\tau \). Output is a ciphertext \(\mathsf {CT}\) and a symmetric key \(\mathsf {K}\).

\(\mathsf {PFSKEM.PnctCxt}( SK ,\tau ,\mathsf {CT})\) input is the current secret key \( SK \), a time period \(\tau \) and additionally a ciphertext \(\mathsf {CT}\). The algorithm outputs a new secret key \( SK '\).

\(\mathsf {PFSKEM.Dec}( SK , \tau , \mathsf {CT})\) takes as input a secret key \( SK \), time period \(\tau \) and a ciphertext \(\mathsf {CT}\). Output is a symmetric key \(\mathsf {K}\) or \(\bot \) if decapsulation fails.

\(\mathsf {PFSKEM.PnctInt}( SK ,\tau )\) takes as input a secret key \( SK \) and a time interval \(\tau \). Output is a secret key \( SK '\) for the next time interval \(\tau +1\).
Definition 6
(Correctness of PFSKEM). For all \(\lambda , n \in \mathbb {N}\), any \(( PK , SK ) \mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\mathsf {PFSKEM.KGen}(1^\lambda )\), any time period \(\tau ^*\), any \((\mathsf {K},\mathsf {CT}^*)\mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\mathsf {PFSKEM.Enc}( PK , \tau ^*)\), and any (arbitrary interleaved) sequence \(i = 0, \dots , n1\) of invocations of \( SK ' \mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\mathsf {PFSKEM.PnctCxt}( SK , \tau , \mathsf {CT})\) for any \((\tau , \mathsf {CT}) \ne (\tau ^*, \mathsf {CT}^*)\) or \( SK ' \mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\mathsf {PFSKEM.PnctInt}( SK , \tau )\) for any \(\tau \ne \tau ^*\) it holds that \(\mathsf {PFSKEM.Dec}( SK ', \tau ^*, \mathsf {CT}^*) = \mathsf {K}\).
Beyond the regular correctness definition above, we further define an extended variant of correctness which demands that decapsulation under a previously punctured out timeinterval and ciphertext yields an error symbol \(\bot \).
Definition 7
(Extended Correctness of PFSKEM). For all \(\lambda , n \in \mathbb {N}\), any \(( PK , SK ) \mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\mathsf {PFSKEM.KGen}(1^\lambda )\), any time period \(\tau ^*\), any \((\mathsf {K},\mathsf {CT}^*)\mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\mathsf {PFSKEM.Enc}( PK ,\tau ^*)\), and any (arbitrary interleaved) sequence \(i = 0, \dots , n1\) of invocations of \( SK ' \mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\mathsf {PFSKEM.PnctCxt}( SK , \tau _i, \mathsf {CT}_i)\) for any \((\tau _i, \mathsf {CT}_i)\) or \( SK ' \mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\mathsf {PFSKEM.PnctInt}( SK ',\tau _i')\) for any \(\tau _i'\) it holds that if \((\tau _i, \mathsf {CT}_i) = (\tau ^*, \mathsf {CT}^*)\) or \(\tau _i' = \tau ^*\) for some \(i \in \{0,\dots ,n1\}\), then \(\mathsf {PFSKEM.Dec}( SK ',\tau ^*,\mathsf {CT}^*) = \bot \).
The security of a \(\mathsf {PFSKEM}\) scheme is defined by the following selectivetime CCA security experiment \(G_{\mathcal {A},\mathsf {PFSKEM}}^{\mathsf {IND{\text { }}sT{\text { }}CCA}}(\lambda )\) played between a challenger \(\mathcal {C}\) and an attacker \(\mathcal {A}\).
 1.
In the beginning, \(\mathcal {A}\) outputs the target time \(\tau ^*\).
 2.
The challenger \(\mathcal {C}\) generates a fresh key pair \(( PK , SK ) \mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\mathsf {PFSKEM.KGen}(1^\lambda )\). It computes \((\mathsf {CT}^*, \mathsf {K}_0^*) \mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\mathsf {PFSKEM.Enc}( PK , \tau ^*)\) and selects \(\mathsf {K}_1^* \mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\mathcal {K}\). Additionally, it chooses a bit \(b\mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\{0,1\}\) and then sends \(( PK ,\mathsf {CT}^*,\mathsf {K}_b^*)\) to \(\mathcal {A}\).
 3.\(\mathcal {A}\) can now ask a polynomial number of the following queries:

\(\mathtt {PFSKEM.Dec}(\tau , \mathsf {CT})\): The challenger computes \(\mathsf {K}\mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\mathsf {PFSKEM.Dec}( SK , \tau , \mathsf {CT})\) and returns \(\mathsf {K}\) to \(\mathcal {A}\).

\(\mathtt {PFSKEM.PnctCxt}(\tau , \mathsf {CT})\): The challenger runs \( SK \mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\mathsf {PFSKEM.PnctCxt}( SK ,\tau ,\mathsf {CT})\) and returns symbol \(\top \).

\(\mathtt {PFSKEM.PnctInt}(\tau )\): The challenger runs \( SK \mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\mathsf {PFSKEM.PnctInt}( SK ,\tau )\) and returns symbol \(\top \).

\(\mathtt {PFSKEM.Corrupt}()\): The challenger aborts and outputs a random bit if \(\mathcal {A}\) has not queried \(\mathtt {PFSKEM.PnctCxt}(\tau ^*, \mathsf {CT}^*)\) or \(\mathtt {PFSKEM.PnctInt}(\tau ^*)\) before. Otherwise, the challenger returns the current secret key \( SK \) to \(\mathcal {A}\).

 4.\(\mathcal {A}\) eventually outputs a guess \(b'\). We denote the event that \(b=b'\) by$$\begin{aligned} G_{\mathcal {A},\mathsf {PFSKEM}}^{\mathsf {IND{\text { }}sT{\text { }}CCA}}(\lambda ) = 1 \ . \end{aligned}$$
Definition 8
2.3 A Generic PFSKEM Construction from HIBKEM
We have now set up the necessary building blocks for our generic PFSKEM construction. In this construction, we deploy a HIBKEM scheme over a binary hierarchy tree comprising time intervals in the upper part and identifiers within these intervals in the lower part. The latter identifiers are carefully crafted to be public keys of a onetime signature scheme, conveniently enabling our construction to achieve CCA security.
We start with a short description of the binary tree, where the root node has the label \(\epsilon \). The left child of a node under label n is labeled with \(n_0\) and the right child with \(n_1\). In a \(\mathsf {HIBKEM}\) scheme every identity \(\mathsf {ID}_i\) is represented by a node \(n_i\) of the hierarchy tree T and with \( sk _i\) we denote the secret key corresponding to node \(n_i\). The root node has the corresponding master secret key msk of the \(\mathsf {HIBKEM}\) scheme. To identify specific nodes in the tree we need the following functions.

\(\mathsf {Parent}(T, n)\). On input of a description of a tree T and a node n, this function outputs the label of the direct ancestor of n or \(\bot \) if it does not exists.

\(\mathsf {Sibling}(T, n)\). On input of a description of a tree T and a label of a node n this function outputs the other child \(n' \ne n\) of the node \(\mathsf {Parent}(T, n)\) or \(\bot \) if it does not exists.
On input of a description of a tree T, a set of secret keys and nodes \( SK = \lbrace ( sk _1,n_1),\ldots ,( sk _u,n_u)\rbrace \) and a node n, the following algorithm computes a new set of secret keys and nodes \( SK '\). The secret keys in \( SK '\) can neither be used to derive the secret key of n nor of its descendants.
 \(\mathsf {PunctureTree}(T, SK , n)\). Create an empty set \( SK '{:=}\{\}\). Then, for all tuples \(( sk _i,n_i)\) in \( SK \):

If \(n_i\) is neither an ancestor nor a descendant of n in the tree and if \(n_i\ne n\), then set \( SK '{:=} SK '\cup {( sk _i, n_i)}\) and \( SK {:=} SK \setminus \lbrace ( sk _i,n_i)\rbrace \).
If there is a remaining node \(n'\) with its secret key \( sk '\) in \( SK \) and if \(n'\) is an ancestor of n, then set \( ntmp {:=}n\), and while \(\mathsf {Parent}( ntmp )\ne \bot \):
if \(n'\) is an ancestor of \(\mathsf {Sibling}(ntmp)\) then \( SK '{:=} SK '\cup \lbrace \mathsf {HIBKEM.Del}(\mathsf {Sibling}( ntmp ), sk '), \mathsf {Sibling}(ntmp))\rbrace \)

\( ntmp {:=}\mathsf {Parent}( ntmp )\).
Output is the set of secret keys and nodes \( SK '\).

Illustrating the described algorithm, we provide an example in Fig. 2, with a tree where the nodes are labeled as described earlier. \( SK \) consists of the tuple \(\{(msk,\epsilon )\}\), where msk is the initial secret key of a \(\mathsf {HIBKEM}\). We would like to puncture the secret key SK for the input \(n_{01}\). In order to do so, we must delete all keys in \( SK \) that can be used to derive the secret keys for the nodes with label “01” or with the prefix “01”. For this, we run the algorithm \(\mathsf {PunctureTree}\) with input \((T, SK , 01)\). In Fig. 2 the gray nodes denote the labels for which we have to derive the secret keys within the new \(\mathsf {PFSKEM}\) secret key \(SK'\). The secret keys in \(SK'\) can only be used to generate secret keys for identities which are not ancestors or descendants of the punctured node “01”.
To obtain a symmetric key at time \(\tau \), one can use the encapsulation algorithm of the \(\mathsf {HIBKEM}\) scheme with input \(( PK ,\tau \vert \vert pk _{OT})\). Correspondingly, the secret key \( SK \) of the \(\mathsf {PFSKEM}\) scheme can be punctured via the previously defined algorithm \(\mathsf {PunctureTree}(T, SK , n)\) by deleting the secret key for the identity \(\mathsf {ID}=\tau \vert \vert pk _{OT}\) in the \(\mathsf {HIBKEM}\) scheme including all secret keys of ancestors of \(\mathsf {ID}\). Particularly, this can be accomplished by using the previously defined algorithm \(\mathsf {PunctureTree}(T, SK , n)\). Decapsulation uses the secret key of the identity \(\mathsf {ID}=\tau \vert \vert pk _{OT}\) with a ciphertext \(\mathsf {CT}\) and outputs the symmetric key or \(\bot \) if the key is already deleted or the signature of the ciphertext is not valid.
As we establish next, our \(\mathsf {PFSKEM}\) construction is selectivetime CCAsecure (according to Definition 8) if the underlying \(\mathsf {HIBKEM}\) scheme is \(\mathsf {IND{\text { }}sID{\text { }}CPA}\)secure and the \(\mathsf {OTSIG}\) scheme is \(\mathsf {sEUF{\text { }}1{\text { }}\mathsf {CMA}}\)secure (cf. Definitions 4 and 2).
Theorem 1
Proof
An attacker \(\mathcal {A}\) on the \(\mathsf {PFSKEM}\) scheme outputs a target time period \(\tau ^*\) and can make the queries described in the security experiment for \(\mathsf {PFSKEM}\) schemes.

\(\mathtt {PFSKEM.Dec}(\tau , \mathsf {CT}_{\mathsf {PFSKEM}}=(\mathsf {CT}_\mathsf {HIBKEM}, \sigma , pk _{OT}))\) with \(\tau \ne \tau ^*\): \(\mathcal {B}_{\mathsf {HIBKEM}}\) can query the \(\mathsf {HIBKEM}\) challenger for the secret key of identity \(\tau \vert \vert pk_{OT}\), because \(\tau \vert \vert pk_{OT}\) is not an ancestor identity of \(\tau ^*\vert \vert pk_{OT}^*\). With the secret key it is possible to decapsulate the key for \(\mathsf {CT}^*_\mathsf {HIBKEM}\).

\(\mathtt {PFSKEM.Dec}(\tau , \mathsf {CT}_{\mathsf {PFSKEM}}=(\mathsf {CT}_\mathsf {HIBKEM}, \sigma , pk _{OT}))\): \( \mathcal {B}_{\mathsf {HIBKEM}}\) can query the \(\mathsf {HIBKEM}\) challenger for the secret key of identity \(\tau ^*\vert \vert pk_{OT}\).

\(\mathtt {PFSKEM.Corrupt}\): If adversary \(\mathcal {A}\) did not call \(\mathtt {PFSKEM.PnctCxt}(\tau ^*, \mathsf {CT}^*)\) or \(\mathtt {PFSKEM.PnctInt}(\tau ^*)\) before, then \(\mathcal {B}_{\mathsf {HIBKEM}}\) aborts and outputs a random bit, else \(\mathcal {B}_{\mathsf {HIBKEM}}\) can query the \(\mathsf {HIBKEM}\) challenger for all secret keys of the requested identities and send them to \(\mathcal {A}\).
3 ForwardSecret OnePass Key Exchange Protocols
3.1 Syntax
Protocols in a 0RTT–like setting, where only one message is transmitted between two key exchange protocol partners, have been the object of previous design interest. In particular, a similar scenario was considered by Halevi and Krawczyk under the notion of onepass key exchange [24]. Aiming for efficiency and optimal key management, we extend their setting by allowing shared state between several executions of the protocol and introduce a discretized notion of time.
Definition 9
(FSOPKE). A forwardsecret onepass key exchange (FSOPKE) protocol supporting \(\tau _{max}\) time periods and providing mutual or unilateral (serveronly) authentication consists of the following four probabilistic algorithms.

\(\mathsf {FSOPKE}.\mathsf {KGen}(1^\lambda , r, \tau _{max}) \rightarrow ( pk , sk )\) . On input the security parameter \(1^\lambda \), a role \(r\in \{\mathsf {client}, \mathsf {server}\}\), and the maximum number of time periods \(\tau _{max}\in \mathbb {N}\), this algorithm outputs a public/secret key pair \(( pk , sk )\) for the specified role.

\(\mathsf {FSOPKE}.\mathsf {RunC}( sk , pk ) \rightarrow ( sk ', k, m)\) . On input a secret key \( sk \) and a public key \( pk \), this algorithm outputs a (potentially modified) secret key \( sk '\), a session key \(k \in \{0,1\}^* \cup \{\bot \}\), and a message \(m \in \{0,1\}^* \cup \{\bot \}\).

\(\mathsf {FSOPKE}.\mathsf {RunS}( sk , pk , m) \rightarrow ( sk ', k)\) . On input of a secret key \( sk \), a public key \( pk \), and a message \(m \in \{0,1\}^*\), this algorithm outputs a (potentially modified) secret key \( sk '\) and a session key \(k \in \{0,1\}^* \cup \{\bot \}\). For a unilateral authenticating protocol, \( pk = \bot \) indicates that the client is not authenticated.

\(\mathsf {FSOPKE}.\mathsf {TimeStep}( sk , r) \rightarrow sk '\) . On input a secret key \( sk \) and an according role \(r\in \{\mathsf {client}, \mathsf {server}\}\), this algorithm outputs a (potentially modified) secret key \( sk '\).

for all \(( pk _i, sk _i) \leftarrow \mathsf {FSOPKE}.\mathsf {KGen}(1^\lambda , \mathsf {client}, \tau _{max})\),

for all \(( pk _j, sk _j) \leftarrow \mathsf {FSOPKE}.\mathsf {KGen}(1^\lambda , \mathsf {server}, \tau _{max})\),
 for any \(n \in \mathbb {N}\) with \(n < \tau _{max}\) and all

\( sk _i' \leftarrow \mathsf {FSOPKE}.\mathsf {TimeStep}^n( sk _i, \mathsf {client})\)

\( sk _j' \leftarrow \mathsf {FSOPKE}.\mathsf {TimeStep}^n( sk _j, \mathsf {server})\)
(where \(\mathsf {FSOPKE}.\mathsf {TimeStep}^n\) indicates n iterative applications of \(\mathsf {FSOPKE}.\mathsf {TimeStep}\)),


for all \(( sk _i'', k_i, m) \leftarrow \mathsf {FSOPKE}.\mathsf {RunC}( sk _i', pk _j)\),
 and for all

\(( sk _j'', k_j) \leftarrow \mathsf {FSOPKE}.\mathsf {RunS}( sk _j', pk _i,m)\) (for mutual authentication)

resp. \(( sk _j'', k_j) \leftarrow \mathsf {FSOPKE}.\mathsf {RunS}( sk _j',\bot ,m)\) (for unilateral authentication),

A forwardsecret onepass key exchange protocol is used by a client and a server party as follows. First of all, both parties generate public/secret key pairs \(( pk , sk ) \leftarrow \mathsf {FSOPKE}.\mathsf {KGen}(1^\lambda ,r,\tau _{max})\) for their according role \(r= \mathsf {client}\) resp. \(r= \mathsf {server}\). To proceed in time (stepwise), they can invoke \(\mathsf {FSOPKE}.\mathsf {TimeStep}\) on their respective secret keys (up to \(\tau _{max}\,  1\) times). Two parties holding secret keys in the same time frame then communicate by the client running \(\mathsf {FSOPKE}.\mathsf {RunC}\) on its secret key and the public key of its intended partner, obtaining the joint session key and a message; transmitting the latter to the server. The server then invokes \(\mathsf {FSOPKE}.\mathsf {RunS}\) on its secret key, the (intended) client’s public key (or \(\bot \) in case of unilateral authentication), and the obtained message, which outputs, by correctness, the same joint session key.
Note that this (0RTT) session key is the only session key derived. Unlike in QUIC and TLS 1.3, we demand that this key immediately enjoys full forward secrecy and replay protection, making an upgrade to another key unnecessary. This demand is realized via the forthcoming security model in Sect. 3.2.
3.2 Security Model
We denote by \(\mathcal {I}= \mathcal {C}\,\mathbin {{\dot{\cup }}}\,\mathcal {S}\) the set of identities modeling both clients (\(\mathcal {C}\)) and servers (\(\mathcal {S}\)) in the system, each identity \(u \in \mathcal {I}\) being associated with a public/secret key pair \(( pk _u, sk _u)\). Here, the publickey part \( pk _u\) is generated once and fixed, whereas \( sk _u\) can be modified by (the sessions of) the according party over time. Each identity u moreover holds the local, current time in a variable denoted by \(\tau _u \in \mathbb {N}\), initialized to \(\tau _u \leftarrow 1\).

\(\mathsf {role}\in \{\mathsf {client}, \mathsf {server}\}\) indicates the role of the session. We demand that \(\mathsf {role}= \mathsf {client}\) resp. \(\mathsf {role}= \mathsf {server}\) if and only if \(u \in \mathcal {C}\) resp. \(u \in \mathcal {S}\).

\(\mathsf {id}\in \mathcal {I}\) indicates the owner of the session (e.g., u for a session \(\pi _u^i\)).

\(\mathsf {pid}\in \mathcal {I}\cup \{\bot \}\) indicates the intended communication partner, and is set exactly once. Setting \(\mathsf {pid}= \bot \) is possible if \(\mathsf {role}= \mathsf {server}\) to indicate the client is not authenticated. Initially, \(\mathsf {pid}= \bot \) can also be set (if \(\mathsf {role}= \mathsf {server}\)) to indicate that the client’s identity is to be learned within the protocol (i.e., postspecified).

\(\mathsf {trans}\in \{0,1\}^* \cup \{\bot \}\) records the (single) sent, resp. received, message.

\(\mathsf {time}\in \mathbb {N}\) records the time interval used when processing the sent, resp. received, message.

\(\mathsf {key}\in \{0,1\}^* \cup \{\bot \}\) is the session key derived in the session.

\(\mathsf {keystate}\in \{\mathsf {fresh}, \mathsf {revealed}\}\) indicates whether the session key has been revealed. Initially \(\mathsf {keystate}= \mathsf {fresh}\).
We write, e.g., \(\pi _u^i.\mathsf {key}\) when referring to state variables of a specific session.
Definition 10

\(\pi _u^i.\mathsf {trans}= \pi _v^j.\mathsf {trans}\), i.e., they share the same transcript,

\(\pi _u^i.\mathsf {time}= \pi _v^j.\mathsf {time}\), i.e., they run in the same time interval,

\(\pi _u^i.\mathsf {role}= \mathsf {client}\wedge \pi _v^j.\mathsf {role}= \mathsf {server}\), i.e., they run in opposite roles,

\(\pi _u^i.\mathsf {pid}= \pi _v^i.\mathsf {id}\), i.e., the server session is owned by the client’s intended partner, and

\(\pi _u^i.\mathsf {id}= \pi _v^i.\mathsf {pid}\vee \pi _v^j.\mathsf {pid}= \bot \), i.e., the client session is owned by the server’s intended partner or the server considers its partner to be unauthenticated.
We assume the adversary \(\mathcal {A}\) controls the network, is responsible for transporting messages, and hence allowed to arbitrary modify, drop, or reorder messages. It interacts with the key exchange protocol and sessions via the following queries.

\(\mathsf {NewSession}(u, role, pid, m)\) . Initializes a new session of identity \(u \in \mathcal {I}\), taking role \(\mathsf {role}\in \{\mathsf {client}, \mathsf {server}\}\) and intended communication partner \(\mathsf {pid}\in \mathcal {I}\cup \{\bot \}\) (where \(\mathsf {pid}= \bot \) for a server session indicates an unauthenticated client partner). If \(\mathsf {role}\ne \mathsf {server}\), we require that \(m=\bot \).
If \(\mathsf {role}= \mathsf {client}\), invoke \(( sk _u, k, m) \leftarrow \mathsf {FSOPKE}.\mathsf {RunC}( sk _u, pk _{\mathsf {pid}})\), else invoke \(( sk _u, k) \leftarrow \mathsf {FSOPKE}.\mathsf {RunS}( sk _u, pk _{\mathsf {pid}},m)\), where \( pk _{\bot } = \bot \).
Register a new session \(\pi _u^i\) with \(\mathsf {role}= role\), \(\mathsf {id}= u\), \(\mathsf {pid}= pid\), \(\mathsf {trans}= m\), \(\mathsf {time}= \tau _u\), and \(\mathsf {key}= k\).
If \(\mathsf {role}= \mathsf {client}\), return m. If \(\mathsf {role}= \mathsf {server}\), return \(\bot \) if \(k = \bot \), and \(\top \) otherwise.

\(\mathsf {Reveal}(\pi _u^i)\) . Reveals the session key of a specific session, if derived.
If \(\pi _u^i.\mathsf {key}\ne \bot \), set \(\pi _u^i.\mathsf {keystate}\leftarrow \mathsf {revealed}\) and return \(\mathsf {key}\), else return \(\bot \).

\(\mathsf {Corrupt}(u)\) . Corrupts the longterm state of an identity \(u \in \mathcal {I}\). This query can be asked at most once per identity u and, from this point on, no further queries to (sessions of) u are allowed.
Let \(\mathsf {Corrupt}(u)\) be the \(\varsigma \)th query issued by \(\mathcal {A}\); we set \(\varsigma ^{corr}_u \leftarrow \varsigma \), where \(\varsigma ^{corr}_u = \infty \) for uncorrupted identities. Likewise, we record the identity’s current time \(\tau _u\) at corruption and set \( \tau ^{corr}_u \leftarrow \tau _u\).
Return \( sk _u\).

\({\mathsf {Tick}}(u)\) . Forward the state of some identity \(u \in \mathcal {I}\) by one time step by invoking \( sk _u \leftarrow \mathsf {FSOPKE}.\mathsf {TimeStep}( sk _u)\). Record the new time as \(\tau _u \leftarrow \tau _u + 1\).

\(\mathsf {Test}(\pi _u^i)\) . Allows the adversary to challenge a derived session key and is asked exactly once. This oracle is given a secret bit \(b_{\mathsf {test}}\in \{0,1\}\) chosen at random in the security game.
If \(\pi _u^i.\mathsf {key}= \bot \), return \(\bot \).
Set \(\tau ^{t}\leftarrow \pi ^{t}.\mathsf {time}\). If \(b_{\mathsf {test}}= 0\), return \(\pi _u^i.\mathsf {key}\), else return a random key chosen according to the probability distribution specified by the protocol.
Definition 11

The challenger generates keys and state for all parties \(u \in \mathcal {I}\) as \(( pk _u, sk _u) \leftarrow \mathsf {FSOPKE}.\mathsf {KGen}(1^\lambda )\) and chooses a random bit \(b_{\mathsf {test}}\mathop {\leftarrow }\limits ^{{\scriptscriptstyle \$}}\{0,1\}\).

The adversary \(\mathcal {A}\) receives \((u, pk _u)\) for all \(u \in \mathcal {I}\) and has access to the queries \(\mathsf {NewSession}\), \(\mathsf {Reveal}\), \(\mathsf {Corrupt}\), \({\mathsf {Tick}}\), and \(\mathsf {Test}\). Record for the \(\mathsf {Test}\) query, being the \(\varsigma ^{test}\)th query, the tested session \(\pi ^{t}\).

Eventually, \(\mathcal {A}\) stops and outputs a guess \(b \in \{0,1\}\).
 1.
\(\pi ^{t}.\mathsf {keystate}= \mathsf {fresh}\), i.e., \(\mathcal {A}\) has not issued a \(\mathsf {Reveal}\) query to the test session.
 2.
\(\pi _v^j.\mathsf {keystate}= \mathsf {fresh}\) for any session \(\pi _v^j\) such that \(\pi _v^j\) and \(\pi ^{t}\) are partners, i.e., \(\mathcal {A}\) has not issued a \(\mathsf {Reveal}\) query to a session partnered with the test session.
 3.
\(\varsigma ^{corr}_u > \varsigma ^{test}\) for \(u = \pi ^{t}.\mathsf {id}\), i.e., the owner of the test session has not been corrupted before the \(\mathsf {Test}\) query was issued.
 4.if \(\pi ^{t}.\mathsf {role}= \mathsf {client}\) and \(\varsigma ^{corr}_v \ne \infty \), for \(v = \pi ^{t}.\mathsf {pid}\), then one of the following must hold:

There exists a session \(\pi _v^j\) partnered with \(\pi ^{t}\), i.e., a session of the intended server partner processed the client session’s message in the intended time interval.

\(\tau ^{t}< \tau ^{corr}_v\), i.e., the intended partner was corrupted in a time interval after that of the tested session.

 5.
if \(\pi ^{t}.\mathsf {role}= \mathsf {server}\) and \(\pi ^{t}.\mathsf {pid}\ne \bot \), then \(\varsigma ^{corr}_v > \varsigma ^{test}\) for \(v = \pi ^{t}.\mathsf {pid}\), i.e., the intended client partner of a tested server session has not been corrupted before the \(\mathsf {Test}\) query was issued.
 6.
if \(\pi ^{t}.\mathsf {role}= \mathsf {server}\) and \(\pi ^{t}.\mathsf {pid}= \bot \), then there exists a session \(\pi _v^j\) partnered with \(\pi ^{t}\), i.e., when testing a server session without authenticated partner, there must exist an honest communication partner to the tested server session \(\pi ^{t}\).
Remark 1
Notably, our security model requires both forward secrecy and replay protection from a FSOPKE protocol. Furthermore, it captures unilateral authentication (of the server) and mutual authentication simultaneously.
As expected, we restrict \(\mathsf {Reveal}\) on both partner sessions involved in the test session (conditions 1 and 2). However, our notion of partnering in Definition 10 lends more power to an adversary than is typically provided. Partnering is defined not only with respect to the session transcripts, partner IDs, and roles, but also with respect to time. Consequently, if the two sessions are not operating within the same time interval, \(\mathsf {Reveal}\) queries are, in fact, permitted on the intended partner session to the test session – even if all other aspects of partnering are fulfilled (condition 2).
To ensure replay protection, the adversary is allowed to test and reveal matching sessions of the same role; we only forbid testing and revealing two matching sessions of opposite roles (via the partnering condition). This explicitly allows for replaying of a client’s message to two server sessions (i.e., spawning two server sessions on input of the same client message m) and revealing one server session while testing the other session. Hence, our model requires that secure protocols prevent replays.
For forward secrecy, corruption of the tested identity is allowed after the \(\mathsf {Test}\) query was issued (condition 3). This applies to both clients (if the client identity exists) and servers.
Server corruption under a tested client session in the 0RTT setting necessitates special considerations (condition 4). First we consider the scenario that the intended partner server session processes messages in the same time interval as the test query, i.e. \(\tau ^{t}\). In this case a tested client’s message must have been processed by the intended partner server session before the server is corrupted^{5} to exclude the following trivial attack: observe that an adversary spawning a new client session (with some \(\mathsf {pid}= v\), outputting a message m) which it subsequently tests, may obtain the secret key \( sk _v\) of the (server) identity v through a \(\mathsf {Corrupt}(v)\) query such that, by correctness of the FSOPKE protocol, it can process message m and derive the correct session key. In this manner, an adversary would always be trivially able to win the key secrecy game. Hence, condition 4 (first item) encodes the strongest possible forward secrecy guarantees in such a scenario: whenever a client’s message has been processed by the server, the corresponding session key becomes forwardsecret w.r.t. compromises of the server’s longterm secret.
Alternatively, we consider the scenario where the intended partner server session processes messages in a time interval after that used in the tested session, i.e. \(\tau ^{t}< \tau ^{corr}_v\). If the server session’s time interval is ahead of that of the tested client session then different session keys are computed. Yet this implies that there are no immediate forward secrecy guarantees should the client’s clock be ahead of the server’s time interval, since the server’s clock can be moved forward after corruption of the server. Thus, condition 4 (second item) gives an additional forward secrecy guarantee: the tested session key is forwardsecret w.r.t. compromises of the server’s longterm secret for any future time interval.
As with corruption of the test session identity (condition 3), if a server session is tested such that a partnered client identity is defined, corruption of the partnered client is restricted until after the test query has been made (condition 5). We do guarantee security if the client is corrupted immediately after it has issued the test session message, but before the server has processed it, due to potential authentication by the client. Should the message be signed, for example, such corruption would allow an adversary to tamper with the message. Thus, for compromises of the client’s longterm secret, we demand forward secrecy immediately after the server establishes the session key.
For the case of unilateral authentication, we must naturally restrict \(\mathsf {Test}\) queries on the server side to cases where an honest partnered client exists (condition 6), as otherwise the adversary can take the role of the client and hence trivially learns the key.
Finally, all security guarantees are required to be provided independent of the time stepping mechanism, making the latter a functional property of a FSOPKE scheme which does not affect the scheme’s security. For example, a scheme could liberally allow session key establishment even if the states of both of the involved sessions are off by a number of time steps. While this is beyond the requirements for a correct scheme, key secrecy still requires that such session keys are secure.
In our model, we do not consider randomness or sessionstate reveal queries [10, 30], but note that it could be augmented with such queries.
4 Constructions
For the construction of a forwardsecret 0RTT key exchange protocol we now first focus on the more common case where only the server authenticates. Our construction builds on puncturable forwardsecure key encapsulation and leverages some synchronization of time between parties in the system. Later, we discuss how to adapt this construction to scenarios where relying on time synchronization is not an option.
4.1 Construction Based on Synchronized Time
We construct a forwardsecret onepass key exchange protocol in a generic way from any puncturable forwardsecure key encapsulation scheme. For our construction, we assume that clients and servers hold some roughly synchronized time, but stress that we are concerned with time intervals rather than exact time and, hence, synchronization for example on the same day is sufficient for our scheme. Aiming at unilateral (serveronly) authentication, clients do not hold longterm key material (i.e., we have \( pk = \bot \) for clients) and only (mis)use their secret key to store the current time interval.
Definition 12

\(\mathsf {FSOPKE}.\mathsf {KGen}(1^\lambda , r, \tau _{max}) \rightarrow ( pk , sk )\) .

Open image in new window Generate a public/secret key pair \(( PK , SK ) \leftarrow \mathsf {PFSKEM.KGen}(1^\lambda )\). Set \( pk \leftarrow ( PK , \tau _{max})\), \(\tau \leftarrow 1\), and \( sk \leftarrow ( SK , \tau , \tau _{max})\), and output \(( pk , sk )\).

Open image in new window Set \( pk \leftarrow \bot \), \(\tau \leftarrow 1\), and \( sk \leftarrow (\tau )\), and output \(( pk , sk )\).

\(\mathsf {FSOPKE}.\mathsf {RunC}( sk , pk ) \rightarrow ( sk ', k, m)\) . Parse \( sk = (\tau )\) and \( pk = ( PK , \tau _{max})\). If \(\tau > \tau _{max}\), then abort and output \(( sk ,\bot ,\bot )\).
Otherwise, compute \((\mathsf {CT}, \mathsf {K}) \leftarrow \mathsf {PFSKEM.Enc}( PK , \tau )\), set \(k \leftarrow \mathsf {K}\) and \(m \leftarrow \mathsf {CT}\), and output \(( sk , k, m)\).

\(\mathsf {FSOPKE}.\mathsf {RunS}( sk , pk = \bot , m) \rightarrow ( sk ', k)\) . Parse \( sk = ( SK ,\tau ,\tau _{max})\). If \( SK = \bot \) or \(\tau > \tau _{max}\), then abort and output \(( sk , \bot )\).
Compute \(\mathsf {K}\leftarrow \mathsf {PFSKEM.Dec}( SK , \tau , m)\). If \(\mathsf {K}= \bot \), then abort and output \(( sk , \bot )\).
Otherwise, issue \( SK ' \leftarrow \mathsf {PFSKEM.PnctCxt}( SK ,\tau ,m)\). Let \( sk \leftarrow ( SK ', \tau , \tau _{max})\), set \(k \leftarrow \mathsf {K}\), and output \(( sk , k)\).

\(\mathsf {FSOPKE}.\mathsf {TimeStep}( sk , r) \rightarrow sk '\) .

Open image in new window Parse \( sk = ( SK ,\tau ,\tau _{max})\). If \(\tau \ge \tau _{max}\), then set \( sk \leftarrow (\bot , \tau + 1, \tau _{max})\), and output \( sk \). Otherwise, let \( SK ' \leftarrow \mathsf {PFSKEM.PnctInt}( SK , \tau )\), set \( sk \leftarrow ( SK ', \tau + 1, \tau _{max})\), and output \( sk \).

Open image in new window Parse \( sk = (\tau )\), set \( sk \leftarrow (\tau + 1)\), and output \( sk \).
Correctness follows from the correctness of the underlying \(\mathsf {PFSKEM}\) scheme; the details are omitted here due to space limitations.
Security Analysis. We now investigate the security of our construction and show that it is a secure forwardsecret onepass key exchange protocol with unilateral authentication.
Theorem 2
Proof
Let \(\mathcal {A}\) be an adversary against the security of \(\mathsf {FSOPKE_U}\). We proceed in a sequence of games, bounding the introduced difference in \(\mathcal {A}\)’s advantage for each step. By \(\mathsf {Adv}_{i}\) we denote \(\mathcal {A}\)’s advantage in one of the ith game.
Game 0. This is the original security experiment, with adversarial advantage \(\mathsf {Adv}_{0}=\mathsf {Adv}_{\mathcal {A},\mathsf {FSOPKE_U}}^{\mathsf {FSOPKE{\text { }}sec}}(\lambda )\).

either \(\pi ^{t}_c = \pi ^{t}\), i.e., \(\pi ^{t}_c\) is the tested session, or

\(\pi ^{t}_c\) is partnered with the tested (server) session \(\pi ^{t}\).
For the second case, observe that if a server is tested, by condition 6 of the \(\mathsf {FSOPKE{\text { }}sec}\) security game in Definition 11, there must exist such a partnered client session \(\pi ^{t}_c\) with \(\pi ^{t}_c.\mathsf {pid}= \pi ^{t}.\mathsf {id}\) in order for \(\mathcal {A}\) to win.
Game 4. In this game hop, we replace the key \(\mathsf {k^*}\) derived in the tested session \(\pi ^{t}\) by one chosen uniformly at random from the output space of \(\mathsf {PFSKEM.Dec}\). We show that any adversary that distinguishes the change from Game 3 to Game 4 with nonnegligible advantage can be turned into an algorithm \(\mathcal {B}\) which wins in \(G_{\mathcal {A},\mathsf {PFSKEM}}^{\mathsf {IND{\text { }}sT{\text { }}CCA}}\) with the same advantage.
In this reduction, \(\mathcal {B}\) first outputs the time interval \(\tau ^*\) guessed in Game 2 as the time interval it wants to be challenged on in \(G_{\mathcal {A},\mathsf {PFSKEM}}^{\mathsf {IND{\text { }}sT{\text { }}CCA}}\). It then obtains a challenge public key \( PK ^*\), which it associates with the server identity \(s^*\) within the \( pk ^*= ( PK ^*, \tau _{max})\) guessed in Game 1. For all other identities \(u \in \mathcal {I}\setminus \{s^*\}\), algorithm \(\mathcal {B}\) generates appropriate public/secret key pairs on its own following \(\mathsf {FSOPKE}.\mathsf {KGen}\). In particular, it generates \(\mathsf {PFSKEM}\) keys for all other server identities \(s \in \mathcal {S}\setminus \{s^*\}\). Furthermore, \(\mathcal {B}\) obtains a challenge ciphertext \(\mathsf {CT}^*\) and key \(\mathsf {K}^*\), with \(\mathsf {K}^*\) either being the real key encapsulated in \(\mathsf {CT}^*\) or and independently chosen random one.

\(\mathsf {NewSession}(u, role, pid, m)\) . We distinguish the following cases:

For all client sessions \(\pi _u^i\) (\(u \in \mathcal {C}\)) except for the client session \(\pi ^{t}_c\) guessed in Game 3, \(\mathcal {B}\) simulates \(\mathsf {NewSession}\) queries as specified in the security game.

For the guessed client session \(\pi ^{t}_c\), \(\mathcal {B}\) does not invoke \(\mathsf {PFSKEM.Enc}\) but uses its challenge key \(\mathsf {K}^*\) as the session key k and the challenge ciphertext \(\mathsf {CT}^*\) as the output message m. Observe that, through Games 1–3, we ensure that \(\pi ^{t}_c\) uses time interval \(\tau ^*\) and public key \( pk ^*\) (and hence the challenge \(\mathsf {PFSKEM}\) public key \( PK ^*\)) of server \(s^*\).

For all server sessions \(\pi _s^i\) not owned by the server identity \(s^*\) guessed in Game 1 (i.e., \(s \in \mathcal {S}\setminus \{s^*\}\)), \(\mathcal {B}\) simulates \(\mathsf {NewSession}\) queries as specified, using the according (selfgenerated) secret key \( sk _s\).

For all server sessions \(\pi _{s^*}^i\) owned by \(s^*\) and not partnered with the guessed client session \(\pi ^{t}_c\), \(\mathcal {B}\) uses its oracles \(\mathtt {PFSKEM.Dec}\) and \(\mathtt {PFSKEM.PnctCxt}\) from the selective ID, selective time CCA game to simulate the operations for the \(\mathsf {NewSession}\) query. Note that, as \(\pi _{s^*}^i\) is not partnered with \(\pi ^{t}_c\) (though having opposite roles and \(\pi ^{t}_c.\mathsf {pid}= s^*\)), we have \((\pi ^{t}_c.\mathsf {time},\pi ^{t}_c.\mathsf {trans}) = (\tau ^*,\mathsf {CT}^*) \ne (\pi _{s^*}^i.\mathsf {time},\pi _{s^*}^i.\mathsf {trans})\) and are hence allowed to call the \(\mathtt {PFSKEM.Dec}\) oracle on this input.

For the first server session \(\pi ^{t}_{s^*}\) owned by \(s^*\) which is partnered with the guessed client session \(\pi ^{t}_c\), \(\mathcal {B}\) sets the session key to be the challenge key \(k \leftarrow \mathsf {K}^*\) and invokes \(\mathtt {PFSKEM.PnctCxt}(\tau ^*, \mathsf {CT}^*)\).
Note that partnering in particular implies \(\pi ^{t}_{s^*}\) holds the same time as \(\pi ^{t}_c\) and obtains the message of \(\pi ^{t}_c\), i.e., \(\pi ^{t}_c.\mathsf {time}= \tau ^*= \pi ^{t}_{s^*}.\mathsf {time}\) and \(\pi ^{t}_c.\mathsf {trans}= m = \pi ^{t}_{s^*}.\mathsf {trans}\). Furthermore, \(\mathtt {PFSKEM.PnctCxt}\) was not invoked before on \((\tau ^*, \mathsf {CT}^*)\). Hence, by correctness, \(\pi ^{t}_{s^*}\) establishes the same session key \(\mathsf {K}^*\) as \(\pi ^{t}_c\).

For any further server session \(\pi _{s^*}^i\) partnered with \(\pi ^{t}_c\), \(\mathcal {B}\) sets \(k \leftarrow \bot \). By Definition 7, we know that any such session would obtain \(\bot \leftarrow \mathsf {PFSKEM.Dec}( SK ,\tau ^*,\mathsf {CT}^*)\), as \(\mathtt {PFSKEM.PnctCxt}\) has been called before on \((\tau ^*,\mathsf {CT}^*)\).

\(\mathsf {Reveal}(\pi _u^i)\) . First, observe that any winning adversary \(\mathcal {A}\) cannot call \(\mathsf {Reveal}\) on the sessions \(\pi ^{t}_c\) and \(\pi ^{t}_{s^*}\) by conditions 1 and 2 of the security model, as one of them is the tested session and the other, if it exists, is partnered with the tested session.
For all other sessions, \(\mathcal {B}\) holds the correct key from simulation of the \(\mathsf {NewSession}\) queries above, and can therefore respond to according \(\mathsf {Reveal}\) queries as specified.

\(\mathsf {Corrupt}(u)\) . For the server identity \(s^*\) involved in the tested session \(\pi ^{t}\), \(\mathcal {B}\) invokes its \(\mathtt {PFSKEM.Corrupt}\) oracle to obtain the \(\mathsf {PFSKEM}\) secret key \( SK ^*\), which it returns within \( sk ^*= ( SK ^*,\tau _{s^*},\tau _{max})\). Observe, that if \(\mathcal {A}\) calls \(\mathsf {Corrupt}(s^*)\) without losing, we are ensured that \(\mathcal {B}\) has called \(\mathtt {PFSKEM.PnctCxt}(\tau ^*,\mathsf {CT}^*)\) and/or \(\mathtt {PFSKEM.PnctInt}(\tau ^*)\) before \(\mathsf {Corrupt}(s^*)\), and hence also does not lose in the selectivetime CCA security game:

If \(\pi ^{t}= \pi ^{t}_{s^*}\) is a server session (owned by \(s^*\)), condition 3 of the security model ensures that \(s^*\) can only be corrupted after \(\pi ^{t}\) has accepted. In the process of \(\pi ^{t}\) accepting (with \(\pi ^{t}.\mathsf {time}= \tau ^*\) and \(\pi ^{t}.\mathsf {trans}= \mathsf {CT}^*\)), \(\mathcal {B}\) must have invoked \(\mathtt {PFSKEM.PnctCxt}(\tau ^*,\mathsf {CT}^*)\), and therefore before corruption of \(s^*\).

If \(\pi ^{t}= \pi ^{t}_c\) is a client session, condition 4 of the security model ensures that either there exists a partnered server session (\(\pi ^{t}_{s^*}\)) that processed \(\mathsf {CT}^*\) in the time interval \(\tau ^*\) or that \(s^*\) gets corrupted in a time interval \(\tau ^{corr}_{s^*} > \pi ^{t}.\mathsf {time}= \tau ^*\). Hence, \(\mathcal {B}\) must have invoked \(\mathtt {PFSKEM.PnctCxt}(\tau ^*,\mathsf {CT}^*)\) or \(\mathtt {PFSKEM.PnctInt}(\tau ^*)\), respectively, before corruption of \(s^*\).^{6}

For any other (client or server) identity \(u \ne s^*\), \(\mathcal {B}\) maintains the corresponding secret key \( sk _u\) and can therefore respond to according \(\mathsf {Corrupt}\) queries as specified.

\({\mathsf {Tick}}(u)\) . Algorithm \(\mathcal {B}\) conducts the time stepping procedures as specified, using its oracle \(\mathtt {PFSKEM.PnctInt}\) on the (unknown) secret key \( SK ^*\) corresponding to the \(\mathsf {PFSKEM}\) challenge public key \( PK ^*\).

\(\mathsf {Test}(\pi ^{t})\) . Observe that the tested session \(\pi ^{t}\) must be either the client session \(\pi ^{t}_c\) guessed in Game 3 or the (first) server session \(\pi ^{t}_{s^*}\) owned by \(s^*\) partnered with \(\pi ^{t}_c\). Algorithm \(\mathcal {B}\), in both cases, simply outputs \(\pi ^{t}.\mathsf {key}= \mathsf {K}^*\) as the response of the \(\mathsf {Test}\) query.
When \(\mathcal {A}\) stops and outputs a guess \(b \in \{0,1\}\), \(\mathcal {B}\) stops as well and outputs b as its own guess.
Observe that algorithm \(\mathcal {B}\) correctly answers all queries of \(\mathcal {A}\) and, in the case that \(\mathsf {K}^*\) is the real key encapsulated in \(\mathsf {CT}^*\), perfectly simulates Game 3, while it perfectly simulates Game 4 if \(\mathsf {K}^*\) is chosen independently at random. Algorithm \(\mathcal {B}\) moreover obeys all restrictions in the selective ID, selective time CCA security game of Definition 8 if \(\mathcal {A}\) adheres to the conditions in the \(\mathsf {FSOPKE}\) security game.
4.2 Variant Without Synchronized Time
For those environments where more relaxed requirements for time synchronization are preferable, we outline a variant of our forwardsecret 0RTT key exchange construction above that does not rely on synchronized time. For this variant, we essentially combine the \(\mathsf {FSOPKE_U}\) construction from Definition 12, restricted to a single time interval, with the concept of server configurations used in recent key exchange protocol designs, namely Google’s QUIC protocol [31] and TLS 1.3 with Diffie–Hellmanbased 0RTT mode [39]. A server configuration here essentially is a publicly accessible string that contains a semistatic public key, signed with the longterm signing key of the corresponding party. Utilizing this string, a forwardsecret 0RTT key exchange protocol variant without time synchronization then works as follows.
For each time interval (e.g., a set number of days or weeks), servers generate a \(\mathsf {PFSKEM}\) key pair (i.e., with \(\tau _{max}= 1\)), which they sign and publish within a server configuration. Clients can then retrieve and use the currently offered public key for the server to establish connections within this time interval.
We stress that, while introducing a slightly higher communication overhead, this variant offers the same security properties as the timesynchronized one. In particular recall that, due to puncturing, compromising the semistatic secret key for some time interval does not endanger the forward secrecy of priorly established connections within the same time interval. Indeed, the choice of how often to publish new server configurations (i.e., how long the conceptual time intervals are) is a purely functional one, based on the performance tradeoff between storage and computation overhead for \(\mathsf {PFSKEM}\) keys covering a shorter or longer interval (and hence more or fewer connections).
5 Analysis
We analyze our protocol for security levels \(\lambda \in \{80, 128, 256\}\). We instantiate our scheme based on the DDHbased HIBE scheme from [6] and the discrete logbased onetime signature scheme from [22, Sect. 5.4]. We consider groups with asymmetric bilinear map \( e : \mathbb {G}_1 \times \mathbb {G}_2 \rightarrow \mathbb {G}_T \) where groups are of order p such that \(p = 2^{2 \lambda }\) for the given security parameter \(\lambda \). Thus, an element of \({\mathbb {Z}}_p\) can be represented by \(2\lambda \) bits. We assume a setting based on BarretoNaehrig curves [2], where elements of \(\mathbb {G}_1\) can be represented by \(2\lambda \) bits, while elements of \(\mathbb {G}_2\) have size \(4 \lambda \) bits. In this setting, we can instantiate our PFSKEM (and thus our FSOPKE) as follows.

A ciphertext consist of three elements of \(\mathbb {G}_1\) (from the HIBE of [6]) plus three \(\mathbb {G}_1\)elements for \( pk _{OT}\), plus two \({\mathbb {Z}}_p\)elements for \(\sigma \). Thus, ciphertexts have size \(6 \times \mathbb {G}_1 + 2 \times {\mathbb {Z}}_p = 16 \lambda \) bits.

A public key contains \(2 \lambda +35\) elements of \(\mathbb {G}_2\), which amounts to \(8\lambda ^2 +140\lambda \) bits.

A punctured secret key contains \(R + S\) user secret keys of the HIBKEM, each consisting of \(3 \times \mathbb {G}_2 = 12\lambda \) bits. Here \(R =  pk _{OT} + \tau \) denotes the bitlength of “HIBKEMidentities”, and S denotes the number of sessions per time slot. Assuming a setting with \(2^{32}\) time slots (which should be sufficient for any conceivable practical application, even with very short time slots), and that a collisionresistant hash function with range \(\{0,1\}^{2\lambda }\) is used to compute a short representation of \( pk _{OT}\) inside the HIBKEM, we have \(R = 2\lambda + 32\). Thus, the size of the secret key as a function of S is \((S+2\lambda +32) \cdot 12\lambda \) bits.
For different values \(S \in \{2^{10}, 2^{16}, 2^{20}\}\) of sessions per time slot, and security parameters \(\lambda \in \{80, 128, 256\}\), we obtain the sizes of public keys and messages and the upper bounds on the size of secret keys displayed in Fig. 4.
Footnotes
 1.
Beyond the pure cryptographic protocol, round trips may also be induced by lowerlayer transport protocols. For example, the TCP protocol requires 1RTT for its own handshake before a higherlevel cryptographic key exchange can start. Here we focus on the overhead roundtrip time caused by the cryptographic components of the keyexchange protocol.
 2.
In case of Google this approach amounts to a few gigabytes of data to be held in shared state between multiple server instances.
 3.
This is essentially a publickey encryption scheme which can only be used to transport random keys, but not to transport payload messages.
 4.
Observe that asynchronous messaging and 0RTT key exchange are conceptually relatively close. In both settings, a data protection key is to be established while only unilateral communication is possible. While different, e.g., in constraints for latency and storage overhead, this in particular implies that our construction can also be employed in the setting of asynchronous messaging.
 5.
Recall that the adversary cannot spawn or interact with sessions of a party anymore after corrupting it.
 6.
Recall that \(\pi ^{t}_{s^*}\) must have accepted before \(s^*\) is corrupted, as afterwards no further queries to sessions owned by \(s^*\) are allowed.
Notes
Acknowledgments
We thank the anonymous reviewers for valuable comments. This work has been cofunded by the DFG as part of project S4 within the CRC 1119 CROSSING and by DFG grant JA 2445/12.
References
 1.Agrawal, S., Boneh, D., Boyen, X.: Efficient lattice (H)IBE in the standard model. In: Gilbert, H. (ed.) EUROCRYPT 2010. LNCS, vol. 6110, pp. 553–572. Springer, Heidelberg (2010). doi: 10.1007/9783642131905_28 CrossRefGoogle Scholar
 2.Barreto, P.S.L.M., Naehrig, M.: Pairingfriendly elliptic curves of prime order. In: Preneel, B., Tavares, S. (eds.) SAC 2005. LNCS, vol. 3897, pp. 319–331. Springer, Heidelberg (2006). doi: 10.1007/11693383_22 CrossRefGoogle Scholar
 3.Bellare, M., Rogaway, P.: Random oracles are practical: a paradigm for designing efficient protocols. In: Ashby, V. (ed.) ACM CCS 1993, , Fairfax, Virginia, USA, pp. 62–73. ACM Press, 3–5 November 1993Google Scholar
 4.Bellare, M., Rogaway, P.: Entity authentication and key distribution. In: Stinson, D.R. (ed.) CRYPTO 1993. LNCS, vol. 773, pp. 232–249. Springer, Heidelberg (1994). doi: 10.1007/3540483292_21 Google Scholar
 5.Bhargavan, K., Fournet, C., Kohlweiss, M., Pironti, A., Strub, P.Y., ZanellaBéguelin, S.: Proving the TLS handshake secure (as it is). In: Garay, J.A., Gennaro, R. (eds.) CRYPTO 2014. LNCS, vol. 8617, pp. 235–255. Springer, Heidelberg (2014). doi: 10.1007/9783662443811_14 CrossRefGoogle Scholar
 6.Blazy, O., Kiltz, E., Pan, J.: (Hierarchical) identitybased encryption from affine message authentication. In: Garay, J.A., Gennaro, R. (eds.) CRYPTO 2014. LNCS, vol. 8616, pp. 408–425. Springer, Heidelberg (2014). doi: 10.1007/9783662443712_23 CrossRefGoogle Scholar
 7.Boneh, D., Boyen, X., Goh, E.J.: Hierarchical identity based encryption with constant size ciphertext. In: Cramer, R. (ed.) EUROCRYPT 2005. LNCS, vol. 3494, pp. 440–456. Springer, Heidelberg (2005). doi: 10.1007/11426639_26 CrossRefGoogle Scholar
 8.Boyd, C., Cliff, Y., Gonzalez Nieto, J., Paterson, K.G.: Efficient oneround key exchange in the standard model. In: Mu, Y., Susilo, W., Seberry, J. (eds.) ACISP 2008. LNCS, vol. 5107, pp. 69–83. Springer, Heidelberg (2008). doi: 10.1007/9783540705000_6 CrossRefGoogle Scholar
 9.Canetti, R., Halevi, S., Katz, J.: A forwardsecure publickey encryption scheme. In: Biham, E. (ed.) EUROCRYPT 2003. LNCS, vol. 2656, pp. 255–271. Springer, Heidelberg (2003). doi: 10.1007/3540392009_16 CrossRefGoogle Scholar
 10.Canetti, R., Krawczyk, H.: Analysis of keyexchange protocols and their use for building secure channels. In: Pfitzmann, B. (ed.) EUROCRYPT 2001. LNCS, vol. 2045, pp. 453–474. Springer, Heidelberg (2001). doi: 10.1007/3540449876_28 CrossRefGoogle Scholar
 11.Chow, S.S.M., Choo, K.K.R.: Stronglysecure identitybased key agreement and anonymous extension. In: Garay, J.A., Lenstra, A.K., Mambo, M., Peralta, R. (eds.) ISC 2007. LNCS, vol. 4779, pp. 203–220. Springer, Heidelberg (2007). doi: 10.1007/9783540754961_14 CrossRefGoogle Scholar
 12.CohnGordon, K., Cremers, C., Garratt, L.: On postcompromise security. In: IEEE 29th Computer Security Foundations Symposium, CSF 2016, pp. 164–178 (2016)Google Scholar
 13.Cremers, C., Feltz, M.: Oneround strongly secure key exchange with perfect forward secrecy and deniability. Cryptology ePrint Archive, Report 2011/300 (2011). http://eprint.iacr.org/2011/300
 14.Cremers, C., Horvat, M., Scott, S., van der Merwe, T.: Automated analysis, verification of TLS 1.3: 0RTT, resumption and delayed authentication. In: IEEE Symposium on Security and Privacy, San Jose, CA, USA, pp. 470–485. IEEE Computer Society Press, 22–26 May 2016Google Scholar
 15.Cremers, C., Feltz, M.: Beyond eCK: perfect forward secrecy under actor compromise and ephemeralkey reveal. In: Foresti, S., Yung, M., Martinelli, F. (eds.) ESORICS 2012. LNCS, vol. 7459, pp. 734–751. Springer, Heidelberg (2012). doi: 10.1007/9783642331671_42 CrossRefGoogle Scholar
 16.Dierks, T., Rescorla, E.: The Transport Layer Security (TLS) Protocol Version 1.2. RFC 5246 (Proposed Standard), Updated by RFCs 5746, 5878, 6176, August 2008Google Scholar
 17.Dowling, B., Fischlin, M., Günther, F., Stebila, D.: A cryptographic analysis of the TLS 1.3 handshake protocol candidates. In: Ray, I., Li, N., Kruegel, C. (eds.) ACM CCS 2015, Denver, CO, USA, pp. 1197–1210. ACM Press, 12–16 October 2015Google Scholar
 18.Fischlin, M., Günther, F.: Multistage key exchange and the case of Google’s QUIC protocol. In: Ahn, G.J., Yung, M., Li, N. (eds.) ACM CCS 2014, Scottsdale, AZ, USA, pp. 1193–1204. ACM Press, 3–7 November 2014Google Scholar
 19.Fischlin, M., Günther, F.: Replay attacks on zero roundtrip time: the case of the TLS 1.3 handshake candidates. In: 2017 IEEE European Symposium on Security and Privacy. IEEE, April 2017Google Scholar
 20.Fujisaki, E., Okamoto, T.: Secure integration of asymmetric and symmetric encryption schemes. In: Wiener, M. (ed.) CRYPTO 1999. LNCS, vol. 1666, pp. 537–554. Springer, Heidelberg (1999). doi: 10.1007/3540484051_34 Google Scholar
 21.Green, M.D., Miers, I.: Forward secure asynchronous messaging from puncturable encryption. In: IEEE S&P 2015 [25], pp. 305–320 (2015)Google Scholar
 22.Groth, J.: Simulationsound NIZK proofs for a practical language and constant size group signatures. In: Lai, X., Chen, K. (eds.) ASIACRYPT 2006. LNCS, vol. 4284, pp. 444–459. Springer, Heidelberg (2006). doi: 10.1007/11935230_29 CrossRefGoogle Scholar
 23.Hale, B., Jager, T., Lauer, S., Schwenk, J.: Simple security definitions for and constructions of 0RTT key exchange. Cryptology ePrint Archive, Report 2015/1214 (2015). http://eprint.iacr.org/2015/1214
 24.Halevi, S., Krawczyk, H.: Onepass HMQV and asymmetric keywrapping. In: Catalano, D., Fazio, N., Gennaro, R., Nicolosi, A. (eds.) PKC 2011. LNCS, vol. 6571, pp. 317–334. Springer, Heidelberg (2011). doi: 10.1007/9783642193798_20 CrossRefGoogle Scholar
 25.IEEE Symposium on Security and Privacy, San Jose, CA, USA. IEEE Computer Society Press, 17–21 May 2015Google Scholar
 26.Jager, T., Kohlar, F., Schäge, S., Schwenk, J.: On the security of TLSDHE in the standard model. In: SafaviNaini, R., Canetti, R. (eds.) CRYPTO 2012. LNCS, vol. 7417, pp. 273–293. Springer, Heidelberg (2012). doi: 10.1007/9783642320095_17 CrossRefGoogle Scholar
 27.Krawczyk, H.: HMQV: a highperformance secure DiffieHellman protocol. In: Shoup, V. (ed.) CRYPTO 2005. LNCS, vol. 3621, pp. 546–566. Springer, Heidelberg (2005). doi: 10.1007/11535218_33 CrossRefGoogle Scholar
 28.Krawczyk, H., Paterson, K.G., Wee, H.: On the security of the TLS protocol: a systematic analysis. In: Canetti, R., Garay, J.A. (eds.) CRYPTO 2013. LNCS, vol. 8042, pp. 429–448. Springer, Heidelberg (2013). doi: 10.1007/9783642400414_24 CrossRefGoogle Scholar
 29.Krawczyk, H., Wee, H.: The OPTLS protocol and TLS 1.3. In: 2016 IEEE European Symposium on Security and Privacy, pp. 81–96. IEEE, March 2016Google Scholar
 30.LaMacchia, B., Lauter, K., Mityagin, A.: Stronger security of authenticated key exchange. In: Susilo, W., Liu, J.K., Mu, Y. (eds.) ProvSec 2007. LNCS, vol. 4784, pp. 1–16. Springer, Heidelberg (2007). doi: 10.1007/9783540756705_1 CrossRefGoogle Scholar
 31.Langley, A., Chang, W.T.: QUIC Crypto. https://docs.google.com/document/d/1g5nIXAIkN_Y7XJW5K45IblHd_L2f5LTaDUDwvZ5L6g/. Accessed May 2016, Revision 26 May 2016
 32.Li, Y., Schäge, S., Yang, Z., Kohlar, F., Schwenk, J.: On the security of the preshared key ciphersuites of TLS. In: Krawczyk, H. (ed.) PKC 2014. LNCS, vol. 8383, pp. 669–684. Springer, Heidelberg (2014). doi: 10.1007/9783642546310_38 CrossRefGoogle Scholar
 33.Lychev, R., Jero, S., Boldyreva, A., NitaRotaru, C.: How secure and quick is QUIC? Provable security and performance analyses. In: IEEE S&P 2015 [25], pp. 214–231 (2015)Google Scholar
 34.Lyubashevsky, V.: Lattice signatures without trapdoors. In: Pointcheval, D., Johansson, T. (eds.) EUROCRYPT 2012. LNCS, vol. 7237, pp. 738–755. Springer, Heidelberg (2012). doi: 10.1007/9783642290114_43 CrossRefGoogle Scholar
 35.Ostrovsky, R., Sahai, A., Waters, B.: Attributebased encryption with nonmonotonic access structures. In: Ning, P., De Capitani di Vimercati, S., Syverson, P.F. (eds.) ACM CCS 2007, Alexandria, Virginia, USA, pp. 195–203. ACM Press, 28–31 October 2007Google Scholar
 36.Petullo, W.M., Zhang, X., Solworth, J.A., Bernstein, D.J., Lange, T.: MinimaLT: minimallatency networking through better security. In: Sadeghi, A.R., Gligor, V.D., Yung, M. (eds.) ACM CCS 2013, Berlin, Germany, pp. 425–438. ACM Press, 4–8 November 2013Google Scholar
 37.Pointcheval, D., Sanders, O.: Forward secure noninteractive key exchange. In: Abdalla, M., Prisco, R. (eds.) SCN 2014. LNCS, vol. 8642, pp. 21–39. Springer, Heidelberg (2014). doi: 10.1007/9783319108797_2 Google Scholar
 38.QUIC, a multiplexed stream transport over UDP. https://www.chromium.org/quic
 39.Rescorla, E.: The Transport Layer Security (TLS) Protocol Version 1.3  draftietftlstls1312. https://tools.ietf.org/html/draftietftlstls1312. Accessed March 2016
 40.Rescorla, E.: The Transport Layer Security (TLS) Protocol Version 1.3  draftietftlstls1318. https://tools.ietf.org/html/draftietftlstls1318. Accessed October 2016
 41.Rescorla, E.: 0RTT and AntiReplay (IETF TLS working group mailing list). IETF Mail Archive, https://mailarchive.ietf.org/arch/msg/tls/gDzOxgKQADVfItfC4NyW3ylr7yc. Accessed March 2015
 42.Rescorla, E.: [TLS] Do we actually need semistatic DHEbased 0RTT? IETF Mail Archive, https://mailarchive.ietf.org/arch/msg/tls/c43zNQH9vGeHVnXhAb_D3cpIAIw. Accessed February 2016
 43.Williams, N.: [TLS] 0RTT security considerations (was OPTLS). IETF Mail Archive, https://mailarchive.ietf.org/arch/msg/tls/OZwGgVhySbVhU36BMX1elQ9x0GE. Accessed November 2014
 44.Wu, D.J., Taly, A., Shankar, A., Boneh, D.: Privacy, discovery, and authentication for the internet of things. In: Askoxylakis, I., Ioannidis, S., Katsikas, S., Meadows, C. (eds.) ESORICS 2016. LNCS, vol. 9879, pp. 301–319. Springer, Heidelberg (2016). doi: 10.1007/9783319457413_16 CrossRefGoogle Scholar