Defining Injection Attacks

This paper defines and analyzes injection attacks. The definition is based on the NIE property, which states that an application’s untrusted inputs must only produce Noncode Insertions or Expansions in output programs (e.g., SQL queries). That is, when applications generate output programs based on untrusted inputs, the NIE property requires that inputs only affect output programs by inserting or expanding noncode tokens (e.g., string and float literals, lambda values, pointers, etc). This paper calls attacks based on violating the NIE property BroNIEs (i.e., Broken NIEs) and shows that all code-injection attacks are BroNIEs. In addition, BroNIEs contain many malicious injections that do not involve injections of code; we call such attacks noncode-injection attacks. In order to mitigate both codeand noncode-injection attacks, this paper presents an algorithm for detecting and preventing BroNIEs.