Secure information sharing on Decentralized Social Networks.

Decentralized Social Networks (DSNs) are web-based platforms built on distributed systems (federations) composed of multiple providers (pods) that run the same social networking service. DSNs have been presented as a valid alternative to Online Social Networks (OSNs), replacing the centralized paradigm of OSNs with a decentralized distribution of the features o↵ered by the social networking platform. Similarly to commercial OSNs, DSNs o↵er to their subscribed users a number of distinctive features, such as the possibility to share resources with other subscribed users or the possibility to establish virtual relationships with other DSN users. On the other hand, each DSN user takes part in the service, choosing to store personal data on his/her own trusted provider inside the federation or to deploy his/her own provider on a private machine. This, thus, gives each DSN user direct control of his/hers data and prevents the social network provider from performing data mining analysis over these information. Unfortunately, the deployment of a personal DSN pod is not as simple as it sounds. Indeed, each pod’s owner has to maintain the security, integrity, and reliability of all the data stored in that provider. Furthermore, given the amount of data produced each day in a social network service, it is reasonable to assume that the majority of users cannot a↵ord the upkeep of an hardware capable of handling such amount of information. As a result, it has been shown that most of DSN users prefer to subscribe to an existing provider despite setting up a new one, bringing to an indirect centralization of data that leads DSNs to su↵er of the same issues as centralized social network services. In order to overcome this issue in this thesis we have investigated the possibility for DSN providers to lean on modern cloud-based storage services so as to o↵er a cloudbased information sharing service. This has required to deal with many challenges. As such, we have investigated the definition of cryptographic protocols enabling DSN users to securely store their resources in the public cloud, along with the definition of communication protocols ensuring that decryption keys are distributed only to authorized users, that is users that satisfy at least one of the access control policies specified by data owner according to Relationship-based access control model (RelBAC) [20, 34]. In addition, it has emerged that even DSN users have the same difficulties as OSN users in defining RelBAC rules that properly express their attitude towards their own privacy. Indeed, it is nowadays well accepted that the definition of access control policies is an error-prone task. Then, since misconfigured RelBAC policies may lead to harmful data release and may expose the privacy of others as well, we believe that DSN users should be assisted in the RelBAC policy definition process. At this purpose, we have designed a RelBAC policy recommendation system such that it can learn from DSN users their own attitude towards privacy, and exploits all the learned data to assist DSN users in the definition of RelBAC policies by suggesting customized privacy rules. Nevertheless, despite the presence of the above mentioned policy recommender, it is reasonable to assume that misconfigured RelBAC rules may appear in the system. However, rather than considering all misconfigured policies as leading to potentially harmful situations, we have considered that they might even lead to an exacerbated data restriction that brings to a loss of utility to DSN users. As an example, assuming that a low resolution and an high resolution version of the same picture are uploaded in the network, we believe that the low-res version should be granted to all those users who are granted to access the hi-res version, even though, due to a misconfiurated system, no policy explicitly authorizes them on the low-res picture. As such, we have designed a technique capable of exploiting all the existing data dependencies (i.e., any correlation between data) as a mean for increasing the system utility, that is, the number of queries that can be safely answered. Then, we have defined a query rewriting technique capable of extending defined access control policy authorizations by exploiting data dependencies, in order to authorize unauthorized but inferable data. In this thesis we present a complete description of the above mentioned proposals, along with the experimental results of the tests that have been carried out so as to verify the feasibility of the presented techniques.