Ieeepro techno solutions 2014 ieee dotnet project - decentralized access control with anonymous authentication of data stored in clouds

Ieeepro techno solutions 2014 ieee dotnet project - decentralized access control with anonymous authentication of data stored in clouds

• 1.
  Decentralized Access Controlwith Anonymous Authentication of Data Stored in Clouds Sushmita Ruj‡, Milos Stojmenovic†, Amiya Nayak∗ ‡CSE, Indian Institute of Technology, IIT, Indore, India, – sush@iiti.ac.in †Singidunum University, Belgrade, Serbia – mstojmenovic@singidunum.ac.rs ∗SEECS, University of Ottawa, Canada – anayak@site.uottawa.ca Abstract—We propose a new decentralized access control scheme for secure data storage in clouds, that supports anonymous authenti- cation. In the proposed scheme, the cloud verifies the authenticity of the ser without knowing the user’s identity before storing data. Our scheme also has the added feature of access control in which only valid users are able to decrypt the stored information. The scheme prevents replay attacks and supports creation, modification, and reading data stored in the cloud. We also address user revocation. Moreover, our authentication and access control scheme is decen- tralized and robust, unlike other access control schemes designed for clouds which are centralized. The communication, computation, and storage overheads are comparable to centralized approaches. Keywords: Access control, Authentication, Attribute-based signatures, Attribute-based encryption, Cloud storage. I. INTRODUCTION Research in cloud computing is receiving a lot of attention from both academic and industrial worlds. In cloud computing, users can outsource their computation and storage to servers (also called clouds) using Internet. This frees users from the hassles of maintaining resources on-site. Clouds can provide several types of services like applications (e.g., Google Apps, Microsoft online), infrastructures (e.g., Amazon’s EC2, Eucalyptus, Nimbus), and platforms to help developers write applications (e.g., Amazon’s S3, Windows Azure). Much of the data stored in clouds is highly sensitive, for example, medical records and social networks. Security and privacy are thus very important issues in cloud computing. In one hand, the user should authenticate itself before initiating any transaction, and on the other hand, it must be ensured that the cloud does not tamper with the data that is outsourced. User privacy is also required so that the cloud or other users do not know the identity of the user. The cloud can hold the user accountable for the data it outsources, and likewise, the cloud is itself accountable for the services it provides. The validity of the user who stores the data is also verified. Apart from the technical solutions to ensure security and privacy, there is also a need for law enforcement. Recently, Wang et al. [2] addressed secure and dependable cloud storage. Cloud servers prone to Byzantine failure, where a storage server can fail in arbitrary ways [2]. The cloud is also prone to data modification and server colluding attacks. In server colluding attack, the adversary can compromise storage servers, so that it can modify data files as long as they are internally consistent. To provide secure data storage, the data needs to be encrypted. However, the data is often modified and this dynamic property needs to be taken into account while designing efficient secure storage techniques. Efficient search on encrypted data is also an important concern in clouds. The clouds should not know the query but should be able to return the records that satisfy the query. This is achieved by means of searchable encryption [3], [4]. The keywords are sent to the cloud encrypted, and the cloud returns the result without knowing the actual keyword for the search. The problem here is that the data records should have keywords associated with them to enable the search. The correct records are returned only when searched with the exact keywords. Security and privacy protection in clouds are being explored by many researchers. Wang et al. [2] addressed storage security using Reed-Solomon erasure-correcting codes. Authentication of users using public key cryptographic techniques has been studied in [5]. Many homomorphic encryption techniques have been suggested [6], [7] to ensure that the cloud is not able to read the data while performing computations on them. Using homomorphic encryption, the cloud receives ciphertext of the data and performs computations on the ciphertext and returns the encoded value of the result. The user is able to decode the result, but the cloud does not know what data it has operated on. In such circumstances, it must be possible for the user to verify that the cloud returns correct results. Accountability of clouds is a very challenging task and involves technical issues and law enforcement. Neither clouds nor users should deny any operations performed or requested. It is impor- tant to have log of the transactions performed; however, it is an important concern to decide how much information to keep in the log. Accountability has been addressed in TrustCloud [8]. Secure provenance has been studied in [9]. Considering the following situation: A Law student, Alice, wants to send a series of reports about some malpractices by authorities of University X to all the professors of University X, Research chairs of universities in the country, and students belonging to Law department in all universities in the province. She wants to remain anonymous while publishing all evidence of malpractice. She stores the information in the cloud. Access control is important in such case, so that only authorized users can access the data. It is also important to verify that the information comes from a reliable source. The problems of access control, authentication, and privacy protection should be solved IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS VOL:25 NO:2 YEAR 2014
• 2.
  2 simultaneously. We addressthis problem in its entirety in this paper. Access control in clouds is gaining attention because it is important that only authorized users have access to valid service. A huge amount of information is being stored in the cloud, and much of this is sensitive information. Care should be taken to ensure access control of this sensitive information which can often be related to health, important documents (as in Google Docs or Dropbox) or even personal information (as in social networking). There are broadly three types of access control: User Based Access Control (UBAC), Role Based Access Control (RBAC), and Attribute Based Access Control (ABAC). In UBAC, the access control list (ACL) contains the list of users who are authorized to access data. This is not feasible in clouds where there are many users. In RBAC (introduced by [10]), users are classified based on their individual roles. Data can be accessed by users who have matching roles. The roles are defined by the system. For example, only faculty members and senior secretaries might have access to data but not the junior secretaries. ABAC is more extended in scope, in which users are given attributes, and the data has attached access policy. Only users with valid set of attributes, satisfying the access policy, can access the data. For instance, in the above example certain records might be accessible by faculty members with more than 10 years of research experience or by senior secretaries with more than 8 years experience. The pros and cons of RBAC and ABAC are discussed in [11]. There has been some work on ABAC in clouds (for example, [12], [13], [14], [15], [16]). All these work use a cryptographic primitive known as Attribute Based Encryption (ABE). The The eXtensible Access Control Markup Language (XACML) [17] has been proposed for ABAC in clouds [18]. An area where access control is widely being used is health care. Clouds are being used to store sensitive information about patients to enable access to medical professionals, hospital staff, researchers, and policy makers. It is important to control the access of data so that only authorized users can access the data. Using ABE, the records are encrypted under some access policy and stored in the cloud. Users are given sets of attributes and corresponding keys. Only when the users have matching set of attributes, can they decrypt the information stored in the cloud. Access control in health care has been studied in [12], [13]. Access control is also gaining importance in online social net- working where users (members) store their personal information, pictures, videos and share them with selected groups of users or communities they belong to. Access control in online social networking has been studied in [19]. Such data are being stored in clouds. It is very important that only the authorized users are given access to those information. A similar situation arises when data is stored in clouds, for example in Dropbox, and shared with certain groups of people. It is just not enough to store the contents securely in the cloud but it might also be necessary to ensure anonymity of the user. For example, a user would like to store some sensitive information but does not want to be recognized. The user might want to post a comment on an article, but does not want his/her identity to be disclosed. However, the user should be able to prove to the other users that he/she is a valid user who stored the information without revealing the identity. There are cryptographic protocols like ring signatures [20], mesh signatures [21], group signatures [22], which can be used in these situations. Ring signature is not a feasible option for clouds where there are a large number of users. Group signatures assume the pre-existence of a group which might not be possible in clouds. Mesh signatures do not ensure if the message is from a single user or many users colluding together. For these reasons, a new protocol known as Attribute Based Signature (ABS) has been applied. ABS was proposed by Maji et al. [23]. In ABS, users have a claim predicate associated with a message. The claim predicate helps to identify the user as an authorized one, without revealing its identity. Other users or the cloud can verify the user and the validity of the message stored. ABS can be combined with ABE to achieve authenticated access control without disclosing the identity of the user to the cloud. Existing work [38], [18], [12], [13], [14], [15], [16] on access control in cloud are centralized in nature. Except [38] and [18], all other schemes use attribute based encryption (ABE). The scheme in [38] uses a symmetric key approach and does not support authentication. The schemes [12], [13], [16] do not support authentication as well. Earlier work by Zhao et al. [15] provides privacy preserving authenticated access control in cloud. How- ever, the authors take a centralized approach where a single key distribution center (KDC) distributes secret keys and attributes to all users. Unfortunately, a single KDC is not only a single point of failure but difficult to maintain because of the large number of users that are supported in a cloud environment. We, therefore, emphasize that clouds should take a decentralized approach while distributing secret keys and attributes to users. It is also quite natural for clouds to have many KDCs in different locations in the world. Although Yang et al. [34] proposed a decentralized approach, their technique does not authenticate users, who want to remain anonymous while accessing the cloud. In an earlier work, Ruj et al. [16] proposed a distributed access control mechanism in clouds. However, the scheme did not provide user authentication. The other drawback was that a user can create and store a file and other users can only read the file. Write access was not permitted to users other than the creator. In the preliminary version of this paper [1], we extend our previous work with added features which enables to authenticate the validity of the message without revealing the identity of the user who has stored information in the cloud. In this version we also address user revocation, that was not addressed in [1]. We use attribute based signature scheme [24] to achieve authenticity and privacy. Unlike [24], our scheme is resistant to replay attacks, in which a user can replace fresh data with stale data from a previous write, even if it no longer has valid claim policy. This is an important property because a user, revoked of its attributes, might no longer be able to write to the cloud. We therefore add this extra feature in our scheme and modify [24] appropriately. Our scheme also allows writing multiple times which was not permitted in our earlier work [16].
• 3.
  3 A. Our Contributions Themain contributions of this paper are the following: 1) Distributed access control of data stored in cloud so that only authorized users with valid attributes can access them. 2) Authentication of users who store and modify their data on the cloud. 3) The identity of the user is protected from the cloud during authentication. 4) The architecture is decentralized, meaning that there can be several KDCs for key management. 5) The access control and authentication are both collusion resistant, meaning that no two users can collude and access data or authenticate themselves, if they are individually not authorized. 6) Revoked users cannot access data after they have been revoked. 7) The proposed scheme is resilient to replay attacks. A writer whose attributes and keys have been revoked cannot write back stale information. 8) The protocol supports multiple read and write on the data stored in the cloud. 9) The costs are comparable to the existing centralized ap- proaches, and the expensive operations are mostly done by the cloud. B. Organization The paper is organized as follows. Related work is presented in Section II. The mathematical background and assumptions are detailed in Section III. We present our privacy preserving access control scheme in Section IV followed by a real life example in Section V. The security is analyzed in Section VI. Computation complexity is discussed in Section VII, and comparison with other work is presented in Section VIII. We conclude in Section IX. II. RELATED WORK ABE was proposed by Sahai and Waters [26]. In ABE, a user has a set of attributes in addition to its unique ID. There are two classes of ABEs. In Key-policy ABE or KP-ABE (Goyal et al. [27]), the sender has an access policy to encrypt data. A writer whose attributes and keys have been revoked cannot write back stale information. The receiver receives attributes and secret keys from the attribute authority and is able to decrypt information if it has matching attributes. In Ciphertext-policy, CP-ABE ([28], [29]), the receiver has the access policy in the form of a tree, with attributes as leaves and monotonic access structure with AND, OR and other threshold gates. All the approaches take a centralized approach and allow only one KDC, which is a single point of failure. Chase [30] proposed a multi-authority ABE, in which there are several KDC authorities (coordinated by a trusted authority) which distribute attributes and secret keys to users. Multi-authority ABE protocol was studied in [31], [32], which required no trusted authority which requires every user to have attributes from at all the KDCs. Recently, Lewko and Waters [35] proposed a fully decentralized ABE where users could have zero or more attributes from each authority and did not require a trusted server. In all these cases, decryption at user’s end is computation intensive. So, this technique might be inefficient when users access using their mobile devices. To get over this problem, Green et al. [33] proposed to outsource the decryption task to a proxy server, so that the user can compute with minimum resources (for example, hand held devices). How- ever, the presence of one proxy and one key distribution center makes it less robust than decentralized approaches. Both these approaches had no way to authenticate users, anonymously. Yang et al. [34] presented a modification of [33], authenticate users, who want to remain anonymous while accessing the cloud. To ensure anonymous user authentication Attribute Based Sig- natures were introduced by Maji et al. [23]. This was also a centralized approach. A recent scheme by the same authors [24] takes a decentralized approach and provides authentication with- out disclosing the identity of the users. However, as mentioned earlier in the previous section it is prone to replay attack. III. BACKGROUND In this section, we present our cloud storage model, adversary model and the assumptions we have made in the paper. Table I presents the notations used throughout the paper. We also describe mathematical background used in our proposed solution. TABLE I NOTATIONS Symbols Meanings Uu u-th User/Owner Aj j-th KDC A Set of KDCs Lj Set of attributes that KDC Aj possesses lj = |Lj| Number of attributes that KDC Aj possesses I[j, u] Set of attributes that Aj gives to user Uu for encryption/decryption Iu Set of attributes that user Uu possesses J[j, u] Set of attributes that Aj gives to user Uu for claim attributes Ju Set of attributes that user Uu possesses as claim attributes AT[j] KDC which has attribute j P K[j]/SK[j] Public key/secret key of KDC Aj for encryption/decryption ski,u Secret key given by Aj corresponding to attribute i given to user Uu T P K/P SK Trustee public key/secret key AP K[j]/ASK[j] Public key/secret key of KDC Aj for verifying claim X Boolean access structure Y Claim policy τ Time instant R Access matrix of dimension m × h M Matrix of dimension l × t corresponding to the claim predicate MSG Message |MSG| Size of message MSG C Ciphertext H, H Hash functions, example SHA-1 A. Assumptions We make the following assumptions in our work. 1) The cloud is honest-but-curious, which means that the cloud administrators can be interested in viewing user’s
• 4.
  4 content, but cannotmodify it. This is a valid assumption that has been made in [12], [13]. Honest-but-curious model of adversaries do not tamper with data so that they can keep the system functioning normally and remain undetected. 2) Users can have either read or write or both accesses to a file stored in the cloud. 3) All communications between users/clouds are secured by Secure Shell Protocol, SSH. B. Formats of access policies Access policies can be in any of the following formats: 1) Boolean functions of attributes, 2) Linear Secret Sharing Scheme (LSSS) matrix, or 3) Monotone span programs. Any access struc- ture can be converted into a Boolean function [35]. An example of a Boolean function is ((a1 ∧a2 ∧a3)∨(a4 ∧a5))∧(a6 ∨a7)), where a1, a2, . . . , a7 are attributes. Let Y : {0, 1}n → {0, 1} be a monotone Boolean function [24]. A monotone span program for Y over a field F is an l × t matrix M with entries in F, along with a labeling function a : [l] → [n] that associates each row of M with an input variable of Y, such that, for every (x1, x2 . . . , xn) ∈ {0, 1}n , the following condition is satisfied: Y(x1, x2, . . . , xn) = 1 ⇔ ∃v ∈ F1×l : vM = [1, 0, 0, . . ., 0] and (∀i : xa(i) = 0 ⇒ vi = 0) In other words, Y(x1, x2, . . . , xn) = 1 if and only if the rows of M indexed by {i|xa(i) = 1} span the vector [1, 0, 0, . . ., 0]. Span programs can be constructed from Boolean functions in a similar way as shown later in Section V. C. Mathematical background We will use bilinear pairings on elliptic curves. Let G be a cyclic group of prime order q generated by g. Let GT be a group of order q. We can define the map e : G × G → GT . The map satisfies the following properties: 1) e(aP, bQ) = e(P, Q)ab for all P, Q ∈ G and a, b ∈ Zq, Zq = {0, 1, 2, . . ., q − 1}. 2) Non-degenerate: e(g, g) = 1. Bilinear pairing on elliptic curves groups are used. We do not discuss the pairing functions which mainly use Weil and Tate pairings [36] and computed using Miller’s algorithm. The choice of curve is an important consideration because it determines the complexity of pairing operations. D. Attribute based encryption Attribute based encryption with multiple authorities as pro- posed by Lewko and Waters [35] proceeds as follows [16]: 1) System Initialization: Select a prime q, generator g of G0, groups G0 and GT of order q, a map e : G0 × G0 → GT , and a hash function H : {0, 1}∗ → G0 which maps the identities of users to G0. The hash function used here is SHA-1. Each KDC Aj ∈ A has a set of attributes Lj. The attributes disjoint (Li Lj = φ for i = j). Each KDC also chooses two random exponents αi, yi ∈ Zq. The secret key of KDC Aj is SK[j] = {αi, yi, i ∈ Lj}. (1) The public key of KDC Aj is published: PK[j] = {e(g, g)αi , gyi , i ∈ Lj}. (2) 2) Key generation and distribution by KDCs: User Uu receives a set of attributes I[j, u] from KDC Aj, and corresponding secret key ski,u for each i ∈ I[j, u] ski,u = gαi H(u)yi , (3) where αi, yi ∈ SK[j]. Note that all keys are delivered to the user securely using the user’s public key, such that only that user can decrypt it using its secret key. 3) Encryption by sender: The encryption function is ABE.Encrypt(MSG, X). Sender decides about the access tree X. LSSS matrix R can be derived as described in III-B. Sender encrypts message MSG as follows: 1) Choose a random seed s ∈ Zq and a random vector v ∈ Zh q , with s as its first entry; h is the number of leaves in the access tree (equal to the number of rows in the corresponding matrix R). 2) Calculate λx = Rx · v, where Rx is a row of R. 3) Choose a random vector w ∈ Zh q with 0 as the first entry. 4) Calculate ωx = Rx · w. 5) For each row Rx of R, choose a random ρx ∈ Zq. 6) The following parameters are calculated: C0 = MSGe(g, g)s C1,x = e(g, g)λx e(g, g)απ(x)ρx , ∀x C2,x = gρx ∀x C3,x = gyπ(x)ρx gωx ∀x, (4) where π(x) is mapping from Rx to the attribute i that is located at the corresponding leaf of the access tree. 7) The ciphertext C is sent by the sender (it also includes the access tree via R matrix): C = R, π, C0, {C1,x, C2,x, C3,x, ∀x} . (5) 4) Decryption by receiver: The decryption function is ABE.Decrypt(C, {ski,u}), where C is given by equation (13). Receiver Uu takes as input ciphertext C, secret keys {ski,u}, group G0, and outputs message msg. It obtains the access matrix R and mapping π from C. It then executes the following steps: 1) Uu calculates the set of attributes {π(x) : x ∈ X} Iu that are common to itself and the access matrix. X is the set of rows of R. 2) For each of these attributes, it checks if there is a subset X of rows of R, such that the vector (1, 0 . . ., 0) is their linear combination. If not, decryption is impossible. If yes, it calculates constants cx ∈ Zq, such that x∈X cxRx = (1, 0, . . . , 0). 3) Decryption proceeds as follows: a) For each x ∈ X , dec(x) = C1,xe(H(u),C3,x) e(skπ(x),u,C2,x) b) Uu computes MSG = C0/Πx∈X dec(x). E. Attribute based signature scheme Attribute based signature scheme [24] has the following steps.
• 5.
  5 1) System Initialization:Select a prime q, and groups G1 and G2, which are of order q. We define the mapping ˆe : G1 × G1 → G2. Let g1, g2 be generators of G1 and hj be generators of G2, for j ∈ [tmax], for arbitrary tmax. Let H be a hash function. Let A0 = ha0 0 , where a0 ∈ Z∗ q is chosen at random. (T Sig, T V er) mean T Sig is the private key with which a message is signed and T V er is the public key used for verification. The secret key for the trustee is T SK = (a0, T Sig) and public key is T PK = (G1, G2, H, g1, A0, h0, h1, . . . , htmax , g2, T V er). 2) User registration: For a user with identity Uu the KDC draws at random Kbase ∈ G. Let K0 = K 1/a0 base . The following token γ is output γ = (u, Kbase, K0, ρ), (6) where ρ is signature on u||Kbase using the signing key T Sig. 3) KDC setup: Choose a, b ∈ Z∗ q randomly and compute: Aij = ha j , Bij = hb j, for Ai ∈ A, j ∈ [tmax]. The private key of i-th KDC is ASK[i] = (a, b) and public key APK[i] = (Aij, Bij|j ∈ [tmax]). 4) Attribute generation: The token verification algorithm ver- ifies the signature contained in γ using the signature verification key T V er in T PK. This algorithm extracts Kbase from γ using (a, b) from ASK[i] and computes Kx = K 1/(a+bx) base , x ∈ J[i, u]. The key Kx can be checked for consistency using algorithm ABS.KeyCheck(T PK, APK[i], γ, Kx), which checks ˆe(Kx, AijBx ij) = ˆe(Kbase, hj), for all x ∈ J[i, u] and j ∈ [tmax]. 5) Sign: The algorithm ABS.Sign(T PK, {APK[i] : i ∈ AT [u]}, γ, {Kx : x ∈ Ju}, MSG, Y), has input the public key of the trustee, the secret key of the signer, the message to be signed and the policy claim Y. The policy claim is first converted into the span program M ∈ Zl×t q , with rows labeled with attributes. Mx denotes row x of M. Let π denote the mapping from rows to the attributes. So, π (x) is the mapping from Mx to attribute x. A vector v is computed which satisfies the assignment {x : x ∈ J[i, u]}. Compute μ = H(MSG||Y). Choose r0 ∈ Z∗ q and ri ∈ Zq, i ∈ Ju, and compute: Y = Kr0 base,Si = (Kvi i )r0 .(g2gμ 1 )ri (∀i ∈ Ju) (7) W = Kr0 0 ,Pj = Πi∈AT [u](AijB π (i) ij )Mij ri (∀j ∈ [t]) (8) The signature is calculated as σ = (Y, W, S1, S2, . . . , St, P1, P2, . . . , Pt). (9) 6) Verify: Algorithm ABS.V erify(T PK, σ = (Y, W, S1, S2, . . . , St, P1, P2, . . . , Pt), MSG, Y), converts Y to the corresponding monotone program M ∈ Zl×t q , with rows labeled with attributes. Compute μ = H(MSG||Y). If Y = 1, ABS.V erify = 0 meaning false. Otherwise, the following constraints are checked. ˆe(W, A0) ? = ˆe(Y, h0) (10) Πi∈lˆe(Si, Ai jB π (i) i j )Mij ) ? = ˆe(Y, h1)ˆe(g2gμ 1 , P1), j = 1 ˆe(g2gμ 1 , Pj), j > 1, (11) where i = AT [i]. Fig. 1. Our secure cloud storage model IV. PROPOSED PRIVACY PRESERVING AUTHENTICATED ACCESS CONTROL SCHEME In this section we propose our privacy preserving authenticated access control scheme. According to our scheme a user can create a file and store it securely in the cloud. This scheme consists of use of the two protocols ABE and ABS, as discussed in Section III-D and III-E respectively. We will first discuss our scheme in details and then provide a concrete example to demonstrate how it works. We refer to the Fig. 1. There are three users, a creator, a reader and writer. Creator Alice receives a token γ from the trustee, who is assumed to be honest. A trustee can be someone like the federal government who manages social insurance numbers etc. On presenting her id (like health/social insurance number), the trustee gives her a token γ. There are multiple KDCs (here 2), which can be scattered. For example, these can be servers in different parts of the world. A creator on presenting the token to one or more KDCs receives keys for encryption/decryption and signing. In the Fig. 1, SKs are secret keys given for decryption, Kx are keys for signing. The message MSG is encrypted under the access policy X. The access policy decides who can access the data stored in the cloud. The creator decides on a claim policy Y, to prove her authenticity and signs the message under this claim. The ciphertext C with signature is c, and is sent to the cloud. The cloud verifies the signature and stores the ciphertext C. When a reader wants to read, the cloud sends C. If the user has attributes matching with access policy, it can decrypt and get back original message. Write proceeds in the same way as file creation. By designating the verification process to the cloud, it relieves the individual users from time consuming verifications. When a reader wants to read some data stored in the cloud, it tries to decrypt it using the secret keys it receives from the KDCs. If it has enough attributes matching with the access policy, then it decrypts the information stored in the cloud.
• 6.
  6 A. Data storagein clouds A user Uu first registers itself with one or more trustees. For simplicity we assume there is one trustee. The trustee gives it a token γ = (u, Kbase, K0, ρ), where ρ is the signature on u||Kbase signed with the trustees private key T Sig (By Equation 6). The KDCs are given keys PK[i], SK[i] for encryption/decryption and ASK[i], APK[i] for signing/verifying. The user on presenting this token obtains attributes and secret keys from one or more KDCs. A key for an attribute x belonging to KDC Ai is calculated as Kx = K 1/(a+bx) base , where (a, b) ∈ ASK[i]. The user also receives secret keys skx,u for encrypting messages. The user then creates an access policy X which is a monotone Boolean function. The message is then encrypted under the access policy as C = ABE.Encrypt(MSG, X) (12) The user also constructs a claim policy Y to enable the cloud to authenticate the user. The creator does not send the message MSG as is, but uses the time stamp τ and creates H(C)||τ. This is done to prevent replay attacks. If the time stamp is not sent, then the user can write previous stale message back to the cloud with a valid signature, even when its claim policy and attributes have been revoked. The original work by Maji et al. [24] suffers from replay attacks. In their scheme, a writer can send its message and correct signature even when it no longer has access rights. In our scheme a writer whose rights have been revoked cannot create a new signature with new time stamp and thus cannot write back stale information. It then signs the message and calculates the message signature as σ = ABS.Sign(Public key of trustee, Public key of KDCs, token, signing key, message, access claim) The following information is then sent in the cloud. c = (C, τ, σ, Y). (13) The cloud on receiving the information verifies the access claim using the algorithm ABS.verify. The creator checks the value of V = ABS.V erify(T PK, σ, c, Y). If V = 0, then authentication has failed and the message is discarded. Else, the message (C, τ) is stored in the cloud. B. Reading from the cloud When a user requests data from the cloud, the cloud sends the ciphertext C using SSH protocol. Decryption proceeds using algorithm ABE.Decrypt(C, {ski,u}) and the message MSG is calculated as given in Section III-D4. C. Writing to the cloud To write to an already existing file, the user must send its message with the claim policy as done during file creation. The cloud verifies the claim policy, and only if the user is authentic, is allowed to write on the file. Fig. 2. Example of claim policy D. User revocation We have just discussed how to prevent replay attacks. We will now discuss how to handle user revocation. It should be ensured that users must not have the ability to access data, even if they possess matching set of attributes. For this reason, the owners should change the stored data and send updated information to other users. The set of attributes Iu possessed by the revoked user Uu is noted and all users change their stored data that have attributes i ∈ Iu. In [13], revocation involved changing the public and secret keys of the minimal set of attributes which are required to decrypt the data. We do not consider this approach because here different data are encrypted by the same set of attributes, so such a minimal set of attributes is different for different users. Therefore, this does not apply to our model. Once the attributes Iu are identified, all data that possess the attributes are collected. For each such data record, the following steps are then carried out: 1) A new value of s, snew ∈ Zq is selected. 2) The first entry of vector vnew is changed to new snew. 3) λx = Rxvnew is calculated, for each row x corresponding to leaf attributes in Iu. 4) C1,x is recalculated for x. 5) New value of C1,x is securely transmitted to the cloud. 6) New C0 = Me(g, g)snew is calculated and stored in the cloud 7) New value of C1,x is not stored with the data, but is transmitted to users, who wish to decrypt the data. We note here that the new value of C1,x is not stored in the cloud but transmitted to the non-revoked users who have attribute corresponding to x. This prevents a revoked user to decrypt the new value of C0 and get back the message. V. REAL LIFE EXAMPLE We now revisit the problem we stated in the introduction. We will use a relaxed setting. Suppose Alice is a Law student and wants to send a series of reports about malpractices by authorities of University X to all the professors of University X, Research chairs of universities X, Y, Z and students belonging to Law department in university X. She wants to remain anonymous, while publishing all evidence. All information is stored in the cloud. It is important that users should not be able to know her
• 7.
  7 identity, but musttrust that the information is from a valid source. For this reason she also sends a claim message which states that she “Is a law student” or “Is a student counselor” or “Professor at university X”. The tree corresponding to the claim policy is shown in Figure 2. The leaves of the tree consists of attributes and the intermediary nodes consists of Boolean operators. In this example the attributes are “Student”, “Prof”, “Dept Law”, “Uni X”, “Counselor”. The above claim policy can be written as a Boolean function of attributes as ((Student AND Dept Law) OR (Prof AND Uni X)) OR (Student Counselor). Boolean functions can also be represented by access tree, with attributes at the leaves and AND(∧) and OR(∨) as the intermediate nodes and root. Boolean functions can be converted to LSSS matrix as below: Let v[x] be parents vector. If node x = AND, then the left child is (v[x]|1), and the right child is (0, . . . , 1). If x = OR, then both children also have unchanged vector v[x]. Finally, pad with 0s in front, such that all vectors are of equal length. The proof of validity of the algorithm is given in [25]. We do not present it here due to lack of space. Using this algorithm, the span program for this policy is M = ⎛ ⎜ ⎜ ⎜ ⎜ ⎝ 1 1 0 −1 1 1 0 −1 1 0 ⎞ ⎟ ⎟ ⎟ ⎟ ⎠ . An assignment v = (v1, v2, v3, v4, v5) satisfies this span program if vM = (1, 0). The cloud should verify that Alice indeed satisfies this claim. Since she is a Law student, v = (1, 1, 0, 0, 0) and is a valid assignment. As a valid user she can then store all the encrypted records under the set of access policy that she has decided. The access policy in case of Alice is ((Prof AND Uni. X) OR (Research Chair AND ((Uni X OR Uni Y) OR Uni Z)) OR ((Student AND Dept Law)AND Uni X) Later when a valid user, say Bob wants to modify any of these reports he also attaches a set of claims which the cloud verifies. For example, Bob is a research chair and might send a claim “Research chair” or “Department head” which is then verified by the cloud. It then sends the encrypted data to the Bob. Since Bob is a valid user and has matching attributes, he can decrypt and get back the information. If Bob wants to read the contents without modifying them, then there is no need to attach a claim. He will be able to decrypt only if he is a Professor in University X or a Research chair in one of the universities X, Y, Z or a student belonging to Department of Law in university X. Here it is to be noted that the attributes can belong to several KDCs. For example the Professors belonging to university X have credentials given by the university X, and a Ph. D. degree from a University P, the student counselor might be a psychologist au- thorized by the Canadian Psychological Association and assigned an employee number by a university, the research chairs can be jointly appointed by the universities X, Y , Z and the government. The students can have credentials from the university and also a department. Initially Alice goes to a trustee for example the Canadian health service and presents her a health insurance number or federal agency presents her a social insurance number. Either or both of these trustees can give her token(s) γ = (u, Kbase, K0, ρ). With the token she approaches the KDCs in the university X and department D and obtains the secret keys for decryption and for keys Kx and Ky for signing the assess policy. She can also access the public keys APK[i] of other KDCs. The entire process is carried on in the following way: A. Data Storage in clouds Let the data be denoted by MSG, X is the access policy- ((Prof AND Uni. X) OR (Research Chair AND ((Uni X OR Uni Y) OR Uni Z)) OR ((Student AND Dept Law)AND Uni X) Alice encrypts the data and obtains the ciphertext C = Enc(MSG, X). Alice also decides on a claim policy Y which is shown in Figure 2. From the matrix, v = (1, 1, 0, 0, 0) can be calculated. The values of Y, W, S1, S2, S3, S4, S5, P1, P2 can be calculated. μ = H(MSG||Y). The current time stamp τ is attached to the ciphertext to prevent replay attacks. The signature σ is calculated as ABS.Sign. The ciphertext c = (C, τ, σ, Y) is then send to the cloud. The cloud verifies the signature using the function ABS.V erify as given in Equation (11). If Alice has valid credentials then the ciphertext (C, τ) is stored, else it is discarded. B. Reading from the cloud and modifying data Suppose Bob wants to access the records stored by Alice. Bob then decrypts the message MSG using his secret keys using function ABE.Decrypt. Writing proceeds like file creation. It is to be noted that the time τ is added to the data so that even if Bob’s credentials are revoked, he cannot write stale data in the cloud. VI. SECURITY OF THE PROTOCOL In this section we will prove the security of the protocol. We will show that our scheme authenticates a user who wants to write to the cloud. A user can only write provided the cloud is able to validate its access claim. An invalid user cannot receive attributes from a KDC, if it does not have the credentials from the trustee. If a user’s credentials are revoked, then it cannot replace data with previous stale data, thus preventing replay attacks.
• 8.
  8 Theorem 1: Ouraccess control scheme is secure (no outsider or cloud can decrypt ciphertexts), collusion resistant and allows access only to authorized users. Proof: We first show that no unauthorized user can access data from the cloud. We will first prove the validity of our scheme. A user can decrypt data if and only if it has a matching set of attributes. This follows from the fact that access structure S (and hence matrix R) is constructed if and only if there exists a set of rows X in R, and linear linear constants cx ∈ Zq, such that x∈X cxRx = (1, 0, . . ., 0). A proof of this appear in [25, Chapter 4]. We note that dec(x) = C1,xe(H(u),C3,x) e(skπ(x),u,C2,x) = e(g, g)λx e(H(u), g)ωx (14) Thus, Πx∈X dec(x) = Πx∈X (e(g, g)λx e(H(u), g)ωx )cx = e(g, g)s (15) Equation (15) above holds because λx = Rx · v and ωx = Rx · w, where v · (1, 0, . . . , 0) = r and ω · (1, 0, . . . , 0) = 0. C0/Πx∈X dec(x) = C0/e(g, g)s = M. For an invalid user, there does not exists attributes correspond- ing to rows x, such that x∈X cxRx = (1, 0, . . . , 0). Thus e(g, g)s cannot be calculated. We next show that two or more users cannot collude and gain access to data that they are not individually supposed to access. Suppose that there exist attributes π(x) from the colluders, such that x∈X cxRx = (1, 0, . . ., 0). However, e(H(u), g)ωx needs to be calculated according to Eq. (15). Since different users have different values of e(H(u), g), even if they combine their attributes, they cannot decrypt the message. We next observe that the cloud cannot decode stored data. This is because it does not posses the secret keys ski,u (by Eq.(3)). Even if it colludes with other users, it cannot decrypt data which the users cannot themselves decrypt, because of the above reason (same as collusion of users). The KDCs are located in different servers and are not owned by the cloud. For this reason, even if some (but not all) KDCs are compromised, the cloud cannot decode data. Theorem 2: Our authentication scheme is correct, collusion secure, resistant to replay attacks, and protects privacy of the user. Proof: We first note that only valid users registered with the trustee(s) receive attributes and keys from the KDCs. A user’s token is γ = (u, Kbase, K0, ρ), where ρ is signature on u||Kbase with T Sig belonging to the trustee. An invalid user with a different user-id cannot create the same signature because it does not know T Sig. We next show that only a valid user with valid access claim is only able to store the message in the cloud. This follows from the functions ABS.Sign and ABS.V erify given in Section III-E. A user who wants to create a file and tries to make a false access claim, cannot do so, because it will not have attribute keys Kx from the related KDCs. At the same time since the message is encrypted, a user without valid access policy cannot decrypt and change the information. Two users cannot collude and create an access policy consisting of attributes shared between them. Suppose, there are two users A and B who have attributes xA and xB respectively. They have the following information KbaseA , KxA and KbaseB , KxB , respectively. A new value of KxB = K 1/(a+bx ) baseA cannot be calculated by B, because it does not know the values of (a, b). Thus the authentication is collusion secure. Our scheme is resistant to replay attacks. If a writer’s access claims are revoked, it cannot replace a data with stale information from previous writes. This is because it has to attach a new time stamp τ and sign the message H(C)||τ again. Since it does not have attributes, it cannot have a valid signature. The mathematical proofs of security of our scheme follows from the security proofs of [35], [23] and has been omitted for the lack of space. VII. COMPUTATION COMPLEXITY In this section we present the computation complexity of the privacy preserving access control protocol. We will calculate the computations required by users (creator, reader, writer) and that by the cloud. Table II presents notations used for different operations. TABLE II NOTATIONS Symbols Computation Ex Exponentiation in group Gx τH Time to hash using function H τH Time to hash using function H τP /τ ˆP Time taken to perform 1 pairing operation in e/ˆe |G| Size of group G a Number of KDCs which contribute keys to user The creator needs to encrypt the message and sign it. Creator needs to calculate one pairing e(g, g). Encryption takes 2 ex- ponentiations to calculate each of C1,x. So this requires 2mET time, where m is the number of attributes. User needs to calculate 3 exponentiation to calculate C2,x and C3,x. So time taken for encryption is (3m + 1)E0 + 2mET + τP . To sign the message, Y, W, Sis and Pjs have to be calculated as well as H(C). So, time taken to sign is (2l + 2)E1 + 2tE2 + τH. The cloud needs to verify the signature. This requires checking for equation (11). Time taken to verify is (l + 2t)τ ˆP + l(E1 + E2) + τH. To read, a user needs only to decrypt the cipher- text. This requires 2m pairings to calculate e(H(u), C3,x) and e(skπ(x),u, C2,x) and O(mh) to find the vector c. Decryption takes 2mτP +τH +O(mh). Writing is similar to creating a record. The size of ciphertext with signature is 2m|G0|+m|GT |+m2 + |MSG| + (l + t + 2)|G1|. When revocation is required, C0 needs to be recalculated. e(g, g) is previously calculated. So, only one scalar multiplication is needed. If the user revoked is Uu, then for each x, C1,x has to be recomputed. e(g, g) is already computed. Thus, only two scalar multiplication needs to be done, for each x. So a total of
• 9.
  9 TABLE III COMPARISON OFOUR SCHEME WITH EXISTING ACCESS CONTROL SCHEMES Schemes Fine-grained Centralized/ Write/read Type of Privacy preserving User access control Decentralized access access control authentication revocation? [38] Yes Centralized 1-W-M-R Symmetric key cryptography No authentication No [12] Yes Centralized 1-W-M-R ABE No authentication No [13] Yes Centralized 1-W-M-R ABE No authentication No [16] Yes Decentralized 1-W-M-R ABE No authentication Yes [33] Yes Centralized 1-W-M-R ABE No authentication No [34] Yes Decentralized 1-W-M-R ABE Not privacy preserving Yes [15] Yes Centralized M-W-M-R ABE Authentication No Ours Yes Decentralized M-W-M-R ABE Authentication Yes TABLE IV COMPARISON OF COMPUTATION AND SIZE OF CIPHERTEXT WHILE CREATING A FILE Schemes Computation by creator Computation by cloud Size of ciphertext [12] (m + 2)E0 0 m log |G0| + |GT | + m log m + |MSG| [13] (m + 2)E0 0 m log |G0| + |G1| + |MSG| [16] (3m + 1)E0 + 2mET + τP (encrypt) 0 2m|G0| + m|GT | + m2 + |MSG| [33] (2m + 1)E0 + ET + τP (encrypt) mE0 + mET + (m + 1)Tp (2m + 1)|G0| + |GT | + m2 + |MSG| [34] (4m + 1)E0 + 2aET + τP (encrypt) 3maET + (3ma + 1)τP (3m + 1)|G0| + |GT | + m2 + |MSG| [15] E1 + (2m + 1)E0 + mτH (encrypt) (l + 2t)τ ˆP + l(E1 + E2) + τH (verify) |G2| + (2m + 1)|G1| + |MSG| (2l + 2)E1 + 2tE2 + τH (sign) +(l + t + 2)|G1| + m2 Our approach (3m + 1)E0 + 2mET + τP (encrypt) (l + 2t)τ ˆP + l(E1 + E2) + τH (verify) 2m|G0| + m|GT | + m2 + |MSG| (2l + 2)E1 + 2tE2 + τH (sign) +(l + t + 2)|G1| TABLE V COMPARISON OF COMPUTATION DURING READ AND WRITE BY USER AND CLOUD Schemes Computation by user while write Computation by user while read Computation by cloud while write [12] No write access mτP No write access [13] No write access mτP No write access [16] No write access 2mτP + τH + O(mh) No write access [33] No write access E0 + τH + O(mh) No write access [34] No write access E0 + τH + O(mh) No write access [15] E1 + (2m + 1)E0 + mτH (encrypt) (2m + 1)τP (decrypt) (l + 2t)τ ˆP + l(E1 + E2) + τH (verify) (2l + 2)E1 + 2tE2 + τH (sign) Our approach (3m + 1)E0 + 2mET + τP (encrypt) 2mτP + τH + O(mh) (decrypt) (l + 2t)τ ˆP + l(E1 + E2) + τH (verify) (2l + 2)E1 + 2tE2 + τH (sign) 2m + 1 scalar multiplications are done by the cloud, where m is the number of attributes belonging to all revoked users. Users need not compute any scalar multiplication or pairing operations. Additional communication overhead is O((m + 1)|GT |). The curves chosen are either MNT curves (proposed by Miyaji, Nakabayashi and Takano) or supersingular curves. Considering the requirements, elliptic curve group of size 159, with an embedding degree 6 (type d curves of Pairing based Cryptography - PBC [36]) can be used. Pairing takes 14 ms on Intel Pentium D, 3.0GHz CPU [16]. Such operations are very suitable for a cloud computing environment. A new library for attribute based encryption is also available at [37]. PBC library (Pairing Based Cryptography) [36] is a C library which is built above GNU GMP (GNU Math Precision) library and contains functions to implement elliptic curves and pairing operations. Each element of G needs 512 bits at an 80-bit security level and 1536 bits when 128-bit of security are chosen [39]. Each cryptographic operation was implemented using the PBC library ver. 0.4.18 on a 3.0 GHZ processor PC. The public key parameters were selected to provide 80-bit security level. According to [39], implementation uses 160-bit elliptic curve group on the supersingular curve y2 = x3 + x over a 512-bit finite field. The computational cost for a pairing operation is 2.9 ms and that of exponentiation on G (and G0) and GT (and G2) are 1 ms and 0.2 ms respectively. We will compare our computation costs with existing schemes like [13], [12], [15] in next Section VIII. VIII. COMPARISON WITH OTHER ACCESS CONTROL SCHEMES IN CLOUD We compare our scheme with other access control schemes (in Table III) and show that our scheme supports many features that the other schemes did not support. 1-W-M-R means that only one user can write while many users can read. M-W-
• 10.
  10 M-R means thatmany users can write and read. We see that most schemes do not support many writes which is supported by our scheme. Our scheme is robust and decentralized, most of the others are centralized. Our scheme also supports privacy preserving authentication, which is not supported by others. Most of the schemes do not support user revocation, which our scheme does. In Tables IV and V we compare the computation and communication costs incurred by the users and clouds and show that our distributed approach has comparable costs to centralized approaches. The most expensive operations involving pairings and is done by the cloud. If we compare the computation load of user during read we see that our scheme has comparable costs. Our scheme also compares well with the other authenticated scheme of [15]. IX. CONCLUSION We have presented a decentralized access control technique with anonymous authentication, which provides user revocation and prevents replay attacks. The cloud does not know the identity of the user who stores information, but only verifies the user’s credentials. Key distribution is done in a decentralized way. One limitation is that the cloud knows the access policy for each record stored in the cloud. In future, we would like to hide the attributes and access policy of a user. ACKNOWLEDGEMENT This work is partially supported by NSERC Grant CRDPJ386874-09 and the grant: “Digital signal processing, and the synthesis of an information security system”, TR32054, Serbian Ministry of Science and Education. REFERENCES [1] S. Ruj, M. Stojmenovic and A. Nayak, “Privacy Preserving Access Control with Authentication for Securing Data in Clouds”, IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing, pp. 556–563, 2012. [2] C. Wang, Q. Wang, K. Ren, N. Cao and W. Lou, “Toward Secure and Dependable Storage Services in Cloud Computing”, IEEE T. Services Computing, vol. 5, no. 2, pp. 220–232, 2012. [3] J. Li, Q. Wang, C. Wang, N. Cao, K. Ren, and W. Lou, “Fuzzy keyword search over encrypted data in cloud computing,” in IEEE INFOCOM. , pp. 441–445, 2010. [4] S. Kamara and K. Lauter, “Cryptographic cloud storage,” in Financial Cryptography Workshops, ser. Lecture Notes in Computer Science, vol. 6054. Springer, pp. 136–149, 2010. [5] H. Li, Y. Dai, L. Tian, and H. Yang, “Identity-based authentication for cloud computing,” in CloudCom, ser. Lecture Notes in Computer Science, vol. 5931. Springer, pp. 157–166, 2009. [6] C. Gentry, “A fully homomorphic encryption scheme,” Ph.D. dissertation, Stanford University, 2009, http://www.crypto.stanford.edu/craig. [7] A.-R. Sadeghi, T. Schneider, and M. Winandy, “Token-based cloud com- puting,” in TRUST, ser. Lecture Notes in Computer Science, vol. 6101. Springer, pp. 417–429, 2010. [8] R. K. L. Ko, P. Jagadpramana, M. Mowbray, S. Pearson, M. Kirchberg, Q. Liang, and B. S. Lee, “Trustcloud: A framework for accountability and trust in cloud computing,” HP Technical Report HPL-2011-38. Available at http://www.hpl.hp.com/techreports/2011/HPL-2011-38.html. [9] R. Lu, X. Lin, X. Liang, and X. Shen, “Secure Provenance: The Essential of Bread and Butter of Data Forensics in Cloud Computing,” in ACM ASIACCS, pp. 282–292, 2010. [10] D. F. Ferraiolo and D. R. Kuhn, “Role-based access controls,” in 15th National Computer Security Conference, 1992. [11] D. R. Kuhn, E. J. Coyne, and T. R. Weil, “Adding attributes to role-based access control,” IEEE Computer, vol. 43, no. 6, pp. 79–81, 2010. [12] M. Li, S. Yu, K. Ren, and W. Lou, “Securing personal health records in cloud computing: Patient-centric and fine-grained data access control in multi- owner settings,” in SecureComm, pp. 89–106, 2010. [13] S. Yu, C. Wang, K. Ren, and W. Lou, “Attribute based data sharing with attribute revocation,” in ACM ASIACCS, pp. 261–270, 2010. [14] G. Wang, Q. Liu, and J. Wu, “Hierarchical attribute-based encryption for fine-grained access control in cloud storage services,” in ACM CCS, , pp. 735–737, 2010. [15] F. Zhao, T. Nishide, and K. Sakurai, “Realizing fine-grained and flexible access control to outsourced data with attribute-based cryptosystems,” in ISPEC, ser. Lecture Notes in Computer Science, vol. 6672. Springer, pp. 83–97, 2011. [16] S. Ruj, A. Nayak, and I. Stojmenovic, “DACC: Distributed access control in clouds,” in IEEE TrustCom, 2011. [17] http://docs.oasis-open.org/xacml/3.0/xacml-3.0-core-spec-cs-01-en.pdf. [18] http://securesoftwaredev.com/2012/08/20/xacml-in-the-cloud. [19] S. Jahid, P. Mittal, and N. Borisov, “EASiER: Encryption-based access control in social networks with efficient revocation,” in ACM ASIACCS, 2011. [20] R. L. Rivest, A. Shamir, and Y. Tauman, “How to leak a secret,” in ASIACRYPT, ser. Lecture Notes in Computer Science, vol. 2248. Springer, pp. 552–565, 2001. [21] X. Boyen, “Mesh signatures,” in EUROCRYPT, ser. Lecture Notes in Computer Science, vol. 4515. Springer, pp. 210–227, 2007. [22] D. Chaum and E. van Heyst, “Group signatures,” in EUROCRYPT, pp. 257– 265, 1991. [23] H. K. Maji, M. Prabhakaran, and M. Rosulek, “Attribute-based signatures: Achieving attribute-privacy and collusion-resistance,” IACR Cryptology ePrint Archive, 2008. [24] ——, “Attribute-based signatures,” in CT-RSA, ser. Lecture Notes in Com- puter Science, vol. 6558. Springer, pp. 376–392, 2011. [25] A. Beimel, “Secure Schemes for Secret Sharing and Key Distribution,” Ph D Thesis. Technion, Haifa, 1996. [26] A. Sahai and B. Waters, “Fuzzy identity-based encryption,” in EUROCRYPT, ser. Lecture Notes in Computer Science, vol. 3494. Springer, pp. 457–473, 2005. [27] V. Goyal, O. Pandey, A. Sahai, and B. Waters, “Attribute-based encryption for fine-grained access control of encrypted data,” in ACM Conference on Computer and Communications Security, pp. 89–98, 2006. [28] J. Bethencourt, A. Sahai, and B. Waters, “Ciphertext-policy attribute-based encryption,” in IEEE Symposium on Security and Privacy. , pp. 321–334, 2007. [29] X. Liang, Z. Cao, H. Lin and D. Xing, “Provably Secure and Efficient Bounded Ciphertext Policy Attribute Based Encryption,” in ACM ASIACCS, pp 343–352, 2009. [30] M. Chase, “Multi-authority attribute based encryption,” in TCC, ser. Lecture Notes in Computer Science, vol. 4392. Springer, pp. 515–534, 2007. [31] H. Lin, Z. Cao, X. Liang and J. Shao, “Secure Threshold Multi-authority Attribute Based Encryption without a Central Authority,” in INDOCRYPT, ser. Lecture Notes in Computer Science, vol. 5365, Springer, pp. 426–436, 2008. [32] M. Chase and S. S. M. Chow, “Improving privacy and security in multi- authority attribute-based encryption,” in ACM Conference on Computer and Communications Security, pp. 121–130, 2009. [33] Matthew Green, Susan Hohenberger and Brent Waters, “Outsourcing the Decryption of ABE Ciphertexts,” in USENIX Security Symposium, 2011. [34] Kan Yang, Xiaohua Jia and Kui Ren, “ DAC-MACS: Effective Data Access Control for Multi-Authority Cloud Storage Systems”, IACR Cryptology ePrint Archive, 419, 2012. [35] A. B. Lewko and B. Waters, “Decentralizing attribute-based encryption,” in EUROCRYPT, ser. Lecture Notes in Computer Science, vol. 6632. Springer, pp. 568–588, 2011. [36] “http://crypto.stanford.edu/pbc/.” [37] “libfenc: The functional encryption library. http://code.google.com/p/libfenc/.” [38] W. Wang, Z. Li, R. Owens, and B. Bhargava, “Secure and efficient access to outsourced data,” in ACM Cloud Computing Security Workshop (CCSW), 2009. [39] J. Hur and D. Kun Noh, “Attribute-Based Access Control with Efficient Revocation in Data Outsourcing Systems”, IEEE Trans. Parallel Distrib. Syst., vol. 22, no. 7, pp. 1214–1221, 2011. IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication.
• 11.
  11 Sushmita Ruj receivedher B.E. degree in Computer Science from Bengal Engineering and Science University, Shibpur, India in 2004, and Masters and Ph D in Computer Science from Indian Statistical Institute, India in 2006 and 2010, respectively. Between 2009 and 2010, she was a Erasmus Mundus Post Doctoral Fellow at Lund University, Sweden and between 2010 and 2012, she was a Post Doctoral Fellow at University of Ottawa, Canada. She is currently an Assistant Professor at Indian Institute of Technology, Indore, India. Her research interests are in security in mobile ad hoc networks, vehicular networks, cloud security, security in smart grids, combinatorics and cryptography. She is on the Editorial Board of Ad Hoc and Sensor Wireless Networks. Milos Stojmenovic received the Bachelor of Computer Science degree at the School of Information Technology and Engineering, University of Ottawa, in 2003. He obtained his Masters degree in computer science at Carleton University in Ottawa, Canada in 2005, and completed his PhD in the same field at the University of Ottawa, Canada in 2008. Currently he is an Assistant Professor at the Singidunum University, Serbia. He was a visiting researcher at Japans National Institute of Advanced Industrial Science and Technology in 2009. He published over thirty articles in the fields of computer vision, image processing, and wireless networks. His work implements machine learning techniques such as AdaBoost and SVM classification which in turn use Higher order Auto- correlation features (HLAC) to perform Image segmentation and classification. Amiya Nayak received his B.Math. degree in Computer Science and Combinatorics & Optimization from University of Waterloo in 1981, and Ph.D. in Systems and Computer En- gineering from Carleton University in 1991. He has over 17 years of industrial experience in software engineering, avionics and navigation systems, simulation and system level performance analysis. He is in the Editorial Board of several journals, in- cluding IEEE Transactions on Parallel & Distributed Systems, International Journal of Parallel, Emergent and Distributed Sys- tems, International Journal of Computers and Applications, and EURASIP Journal of Wireless Communications and Networking. Currently, he is a Full Professor at the School of Electrical Engineering and Computer Science at the University of Ottawa. His research interests are in the area of fault tolerance, distributed systems/algorithms, and mobile ad hoc networks with over 150 publications in refereed journals and conference proceedings.