Fine Grained Transaction Log for Data Recovery in Database Systems

Abstract

Data recovery for malicious committed transactions after attacks increasingly becomes an important issue. Damage assessment for data recovery requires a transaction log which record data items read or written by all malicious and benign transactions. Unfortunately, conventional undo/redo log could not record read operations for transactions; and existing auditing mechanisms in DBMS could not capture operations for data items. In this paper, we introduce a concept of "extended read operations" and illustrate how the extended read operations would cause the damage spreading, and then a fine grained transaction log (FGTL) is proposed. The log records all the data items of the read only and update-involved operations (read and write) for the committed transactions, and even extracts data items read by the subqueries in the SQL statements. A prototype system denoted FGTL generator is developed to generate the FGTL. Experiments based on TPC-W benchmark show the availability for FGTL generator.