物聯網裝置上系統層級具隨選式記憶體追蹤機制

Abstract

現存物聯網環境快速成長，有許多資料不斷地被產生並上傳至網路。各種相關應用程式依照設計方案應用這些資料。然而這些程式也許存在著一些漏洞。一旦漏洞攻擊事件發生，敏感資料有外洩的危機。為了防範這類危機，我們提出了一個資料流追踪機制以追踪記憶體資料以及處理器內暫存器資料追蹤。記憶體資料追蹤部分監控敏感資料的操作，並且在具有敏感資料的暫存器上插⼊追蹤標記。當處理器完成敏感資料運算後，標記會被汙染回記憶體上。對於需要⼤量處理器時間的程式，此機制的負載落在 150 到 1000 倍之間，⽽等待輸⼊輸出的程式則需要 10 到 20% 的負載。

In current Internet-of-Thing (IoT) environment, a huge amount of data is produced by sensors every second and transmitted into internet. Data are normally operated within the original design of the IoT applications but a small part of them could potentially leak out to malicious servers due to program vulnerability and cyber-attack. To address the problem, we introduce Page-taint to track the data flow of sensitive data on a system. Page-taint is composed of on-demand memory tracing and register tracing. On-demand memory tracing tracks data flow by monitoring page-level data access. It imports the tag of interested data into CPU and tags are propagated at an instruction level. New tags are then set and are written back to memory. The path of interested data in a process is thus built by connecting partial memory access and corresponding data flow in the CPU. The overhead of page-taint is separated to CPU bound and I/O bound. The former overhead is from 150x to 1000x and the latter is 10 to 20%.