Malicious Wear Detection And Defense Mechanism For Persistent Memory File Systems

With the rapid development of emerging persistent memories(PMs),many novel file systems have been designed using PMs as storage devices.They make full use of the byte addressing,low latency,non-volatile,and high storage density advantages of PMs,greatly improve the performance of the file system.However,the existing persistent memory file system does not consider that persistent memory generally has the problem of limited write endurance.Malicious wear attacks can wear out the underlying persistent memory storage cell through simple file operations,which seriously threatens the data reliability of persistent file systems.Existing malicious wear detection and defense mechanisms at the hardware layer is unable to correlate the data update characteristics of upper-layer applications and file systems,which cannot achieve efficient malicious wear detection and defense,causing problems such as high data migration overhead.Therefore,implementing malicious wear detection and defense mechanism at the persistent memory file system layer is a key issue that needs to be resolved urgently.This thesis fully considers the data update characteristics of the file system and the expected service life of the persistent memory.It proposes a malicious wear detection and defense mechanism named Themis for the persistent memory file system.Themis includes the following three key technologies.First,in order to improve the identification accuracy of malicious wear processes and reduce costs,this thesis proposes a malicious wear detection model based on the service life of persistent memory and designs a random selection of detection windows for global detection.Once abnormalities are found,the malicious wear detection model immediately detects all processes to identify malicious wear processes.Secondly,this thesis proposes a defense cache strategy based on dynamic random access memory(DRAM),which caches the write operation of malicious wear attacks to the DRAM to avoid the continuous wear of the persistent memory fixed storage cell by the malicious wear process.Finally,in order to further improve the service life of persistent memory,this thesis proposes a wear-aware data management strategy.Based on the page wear degree of persistent memory,it dynamically migrates the file system metadata and designs a space allocation strategy that considers wear degrees to achieve wear leveling of the persistent memory.This thesis implements the Themis prototype system in the Linux kernel based on the typical persistent memory file system NOVA to verify the effectiveness of Themis.Experimental results show that Themis can significantly reduce the wear of persistent memory under malicious wear attacks.When three typical workloads continuously write the same number of operations on different schemes,the persistent memory service life is 5774× that of the typical solution DWARM.Besides,the performance experimental result shows that the performance of Themis is 1.13× that of DWARM.