Multiverse databases for secure web applications

Abstract

Most modern web applications authenticate users and enforce security policies in the application logic. Therefore, buggy applications can easily leak sensitive data. MultiverseDB addresses this problem in a new database architecture, where each user has her own private view of the database and declarative security policies restrict data-flow into a user's private universe. To support multi-user universes, MultiverseDB builds on ideas of streaming data-flow systems and low-overhead materialized views. When a new user session starts, the system creates data-flow nodes to support the user queries and automatically inserts special nodes to enforce security policies at universe boundaries. MultiverseDB provides fast reads by storing the pre-computed results of user queries with policies already applied in incrementally maintained materialized views. To reduce space overheads created by these views and avoid redundant processing, MultiverseDB reuses views and allows system administrators to specify security groups for users subjected to the same security policies.