Research And Implementation Of GPU Covert Channel

The Covert channel has always been a key issue in the field of computer security.Covert channel have become an important indicator of the security of computerized information systems above the B2 level since its introduction in 1973.With the development of modern society,computer information system has been deeply penetrated into various fields,the traditional graphics hardware has evolved into a programmable graphics processor(GPU).Its powerful computing performance,flexible programming ability and low prices for the consumer market have attracted more and more researchers to use GPU in Solve the graphics rendering general computing tasks,its areas include physical simulation,scientific computing,signal processing,database operations.More and more computer systems are using GPU to help with their operations,but most of the current GPU based systems are less comprehensive in terms of hidden lanes.This is still a threat that is currently on the market for GPU.Based on the status research of both GPU and covert channel at domestic and abroad,this paper analyzes the three different memory of GPU in detail and proposes related algorithms to realize the covert channel.The main research work is divided into three parts:First of all,the basics of GPU architecture and CUDA threading model and memory model are discussed.From the perspective of hardware and software,the working methods and principles of the GPU are analyzed.Then,use CUDA stream technology to start two application processes(Send process and Receive process),and then control the position of the two processes,and finally get the shared memory size parameters by loading the array with a fixed stride.Secondly,based on the GPU memory model,this paper selects three representative memories from three aspects: location,size and nature of memory.Analyze the shared memory,texture memory,and TLB shared resource of the GPU,measure their respective hierarchical structures and sizes,and then propose an algorithm to establish covert channels through memory contention.The resultobtained is a number of representative delay values.Finally,the experiment results are analyzed,where the high delay values represents binary 1 and the low delay values represents binary 0.With such rules for data conversion,we can get the information received by application.Experiments show that can establish a covert channel and send information successfully in the GPU three different memory.At the same time,the paper makes a detailed description of the bit error rate and communication bandwidth of the covert channel,and gives a covert channel detection and defense method.