CN106294169A - The detection of a kind of data contention based on semiology analysis virtual machine and playback method - Google Patents

Abstract

The invention discloses a method for detecting and replaying parallel program data competition in an embedded environment. Combining with the requirements of software debugging and testing, a symbolic execution virtual machine scheme is adopted to dynamically monitor the running of the program, collect the execution information of the program, and Analyze data races. And it can also deterministically replay the program through the execution information, so that the execution track of the program can be reproduced. Functions include: data race detection based on symbolic execution virtual machine, data race replay function based on symbolic execution virtual machine. The invention can excavate the data competition hidden in the program, and prevent the multi-threaded program from running wrong.

Description

A Data Race Detection and Replay Method Based on Symbolic Execution Virtual Machine

technical field

The invention belongs to the technical field of software debugging, and more specifically relates to a data competition detection and playback method based on a symbolic execution virtual machine.

Background technique

With the increasing scale of software, software testing and program error detection become more and more important. Existing software testing and error detection methods can be divided into static methods and dynamic methods. The static method does not actually run the program. By analyzing the call graph, data flow graph, control flow graph, execution analysis and other information in the program source code, it compares these information with the established error model to detect the error code fragments that the program may contain. The dynamic method runs the program, analyzes the running track of the program, and detects the errors in the program.

At present, the mainstream dynamic detection methods at home and abroad include Valgrind developed by Nicholas Nethercott, and Pin developed by Intel Corporation, both of which are binary dynamic detection frameworks. The program is instrumented and then the internal virtual machine executes the source code.

However, there are many problems in the existing dynamic detection methods: First, both Valgrind and Pin belong to the general detection framework, and the binary code is compiled into an intermediate code, which cannot actively identify thread shared data, and monitors excessive redundant memory during data race inspection, thereby reducing In addition, Valgrind internally implements "shadow value" (shadow value) technology, and implementing lightweight detection tools under this framework will also cause relatively large overhead; finally, the detection under the Pin framework is based on the provided specific Interface, the function of the interface is relatively fixed, so it is difficult to meet the user's customization needs.

Contents of the invention

In view of the above defects or improvement needs of the prior art, the present invention provides a data race detection and replay method based on a symbolic execution virtual machine. Low detection efficiency, large detection process overhead, and poor scalability technical problems.

In order to achieve the above object, according to one aspect of the present invention, a data race detection and replay method based on a symbolic execution virtual machine is provided, including the following steps:

(1) Receive the program execution trajectory operation request from the host computer, and judge whether the request is a replay request or a detection request, if it is a replay request, then enter step (2), otherwise enter step (5);

(2) judge whether the recording file corresponding to this replay request exists in the disk, if exist then go to step (3), otherwise process ends;

(3) Obtain the program execution information file in the record file, and execute the global variable read and write instructions and/or multi-thread function call instructions in the virtual machine by obtaining symbols to call the detection function, so as to load the constraints between the thread instructions in the record file , and judge whether the global variable read and write instruction and/or the multi-threaded function call instruction can be executed according to the constraints, and if so, perform step (4), otherwise the process ends;

(4) Execute global variable read and write instructions and/or multi-thread function instructions, and the process ends;

(5) Execute the multi-threaded program corresponding to the detection request in the symbolic execution virtual machine, monitor the global variable access in the multi-threaded program between multiple threads, and record the order in which the multi-threads access the global variable, and record these Sequence information is recorded into a binary file to generate a program execution information file;

(6) Obtain the instruction information in the symbolic execution virtual machine, use the data competition detection algorithm to detect the reading and writing of global variables in the instruction information, and use the multi-thread function call interception function to call the multi-thread function in the instruction information to judge the global variable read and write Write whether there is a data race, if so, report the data race detection error information to the client, if the program execution is not over, obtain the next instruction information stored in the container of the symbolic execution virtual machine, and go to step (5) implement.

Preferably, the method also includes before step (1), carrying out multi-thread extension to the symbolic execution virtual machine, to generate a symbolic execution virtual machine with multi-threaded function call interception and internal emulation multi-threading function.

Preferably, the process of executing the multi-thread extension for the symbolic execution virtual machine is as follows: firstly, build a multi-thread simulation implementation framework through function pointers and structures, and then aim at the synchronization including thread creation, thread waiting, thread locking and unlocking The function is rewritten.

Preferably, step (3) includes the following sub-steps:

(3-1) read the program execution information file, and write the record file in the program execution information file into the internal memory;

(3-2) According to the execution trajectory of the symbolic execution virtual machine, the instruction of the multi-threaded program corresponding to the detection request is cyclically selected, and it is judged whether the instruction is a read instruction, a write instruction, or a function call instruction, and if so, proceed to step (3-3) , otherwise continue to execute the instruction according to the execution track of the symbolic execution virtual machine;

(3-3) judge whether the memory address accessed by the instruction is a global variable, if it is, then enter the substep (3-4), otherwise if the instruction is the lock and unlock function in the multithreaded function call instruction, then enter the substep (3 -4), otherwise continue to execute the instruction according to the execution track of the symbolic execution virtual machine. .

(3-4) According to the file record information written into the internal memory, it is judged whether there is a dependency relationship in the current instruction, if there is no dependency relationship, then enter sub-step (3-5), otherwise the symbol is obtained to execute the next instruction in the virtual machine, And return sub-step (3-2) to carry out;

(3-5) Execute the current instruction, remove the dependencies between the instructions of the executed instructions, thereby update the file record information written into the memory, obtain the next instruction in the symbolic execution virtual machine, and return to the substep (3-5) when the program is not over 2) Execute until the end of the program.

Preferably, step (5) includes the following sub-steps:

(5-1) Detect the instruction that the corresponding multi-thread program of request is carried out, and judge whether this instruction is read instruction or write instruction, if then enter substep (5-2), otherwise if this instruction is lock instruction or For the unlock instruction, after executing the lock or unlock instruction, enter the sub-step (5-2), otherwise, continue to execute the instruction according to the execution track of the symbolic execution virtual machine;

(5-2) Determine whether the memory address accessed by the instruction is a global variable address, if so, obtain the address of the global variable, establish the mapping relationship between the global variable address and the number, and obtain the thread access global variable information, including the thread number and timestamp vector , if the instruction type is a write instruction, then proceed to substep (5-3), if the instruction type is a read instruction, then proceed to substep (5-4); otherwise, continue to execute the instruction according to the execution track of the symbolic execution virtual machine;

(5-3) Read the set of all read access records corresponding to the global variable according to the number corresponding to the global variable address, traverse all the read access records, and process each read access record in the following form: the thread of the read access record The timestamp value corresponding to the thread number is compared with the timestamp value corresponding to the same thread number of the current access record. If the less than or equal relationship is satisfied, no processing is performed. Otherwise, it indicates that there is a sequence relationship between the two operations and is recorded. Finally, merge the timestamp vectors of all read access records with the currently accessed timestamp vector;

(5-4) Read the collection of all write access records corresponding to the global variable according to the number corresponding to the global variable address, and process each write access record in the following form: compare the timestamp value corresponding to the thread number of the write access record with Compare the timestamp values corresponding to the same thread number of the current access record. If the less than or equal relationship is satisfied, no processing will be performed. Otherwise, it indicates that there is a sequence relationship between the two operations and will be recorded. Finally, merge the timestamp vectors of all write access records with the current access timestamp vector;

(5-5) Write the set of all read-write access records into the record file, so as to complete the recording process of the execution track.

Preferably, step (6) includes the following sub-steps:

(6-1) Execute the multi-threaded program corresponding to the request to obtain the executed instruction;

(6-2) execute the conditional branch structure in the virtual machine to judge the type of instruction by symbol, if the instruction is read instruction load or write instruction store, then proceed to substep (6-3); if the instruction is a function call instruction, then Go to sub-step (6-7); if the instruction is other instructions, continue to execute the instruction according to the execution track of the symbolic execution virtual machine, and then return to sub-step (6-1);

(6-3) Judging whether the memory address accessed by the instruction is a global variable, if so, proceed to substep (6-4), otherwise continue to execute the instruction according to the execution track of the symbolic execution virtual machine, and then return to substep (6-4) 1);

(6-4) Judge whether the instruction is a write access, if it is a write access, then call the write processing function callStore, and proceed to substep (6-5), if it is a read access, then call the read processing function callLoad, and turn to Into sub-steps (6-6);

(6-5) According to the memory address addr of the global variable, check whether the global variable is the first write access, if so, store the thread number and the timestamp of the thread accessing the global variable in the write record, and then return to the child Step (6-1); otherwise, take out the write record information of the memory address addr of the global variable, including thread number and time stamp vector, etc. Compare the timestamp value of the current thread recorded in the timestamp vector of the current thread with the timestamp value in the write record information. If the former is smaller than the latter, it means that data competition occurs, and report the write data competition error of the global variable to Client, otherwise do not process, and continue to judge whether there is a concurrent read access identifier between the two write accesses. If there is a concurrent read access, then compare the timestamp value of the current thread recorded in this write access with the The timestamp value of the current thread recorded in the timestamp vector of all concurrent read access records is compared, as long as the less than relationship is satisfied, it means that data competition occurs, and the read and write data competition error of the global variable is reported to the client, and the concurrent read access records are cleared and stored All the information in, update the write access record, and then return to substep (6-1), otherwise clear all the information stored in the concurrent read access record, update the write access record, and then return to substep (6-1).

(6-6) According to the memory address of the global variable, check whether the global variable is the first read access, if it is the first read access, store the thread number and the timestamp value of the thread in the read record, and then return to the child Step (6-1); if it is not the first read access, then take out the write record information of the memory address of the global variable, including thread number and timestamp vector, etc., and record the time of the current thread in the timestamp vector of the current thread The stamp value is compared with the recorded timestamp value. If the two satisfy the less than relationship, it means that a data competition has occurred, and the data competition will be reported to the client. Otherwise, it will not be processed. At the same time, continue to judge whether there is a concurrent read access identifier between the two write accesses. , if there is concurrent read access, add this read access directly to the concurrent read access record whose address is addr; otherwise, take out the read access record information, including thread number and timestamp vector, etc., and add the current thread’s timestamp vector to Compare the recorded timestamp value of the current thread with the recorded timestamp value, if the two satisfy the less than relationship, update the concurrent read access flag, add this read access to the concurrent read access record whose address is addr, and then Update the read access record whose address is addr, and return to substep (6-1);

(6-7) Check whether the current call function called by the function call instruction is a multi-thread function by means of character string matching, if yes, enter sub-step (6-8), otherwise continue to execute the instruction according to the execution track of the symbolic execution virtual machine ;

(6-8) At the function call instruction of the conditional branch structure in the symbolic execution virtual machine, the name of the currently called function is identified by string matching, and the parameters of the function are obtained, and the current call function is judged according to the name and parameters of the function. Thread creation function, thread waiting function, locking function, or unlocking function, if it is a thread creation function, then proceed to substep (6-9); if it is a thread waiting function, then proceed to substep (6-10) ; If it is a locking function, then proceed to substep (6-11); if it is an unlocking function, then proceed to substep (6-12);

(6-9) Call the thread creation function to create a child thread, the child thread initializes its own timestamp vector, and merges with the parent thread's timestamp vector, and the parent thread's timestamp value increases by 1;

(6-10) calling the thread waiting function to merge the timestamp vector of the waiting thread with the timestamp vector of the waiting thread;

(6-11) Call the locking function, and after successful locking, merge the timestamp vector of the thread and the timestamp vector of the lock object;

(6-12) Call the unlock function, and after successful unlocking, assign the thread's timestamp vector to the lock object's timestamp vector.

Generally speaking, compared with the prior art, the above technical solutions conceived by the present invention can achieve the following beneficial effects:

(1) the detection efficiency of the present invention is high: what monitor in the method for detecting data competition in step (6) is the read-write access of global variable and the calling access information of synchronous function, and only preserve last write access Record, when performing read-write data competition detection and write-write data competition detection of global variables, you only need to compare the timestamp value of the current read-write record with the timestamp value of the last write access record to determine whether there is data competition Error, the time complexity of detection is reduced from linear time complexity to constant time complexity, so the detection efficiency is improved.

(2) the detection expense of the present invention is little: because adopted the substep (5-2) in the step (5) only just set up the mapping relation of global variable address and numbering when judging that the memory address of instruction access is the global variable address In the detection process, only the access records of the global variables are stored to detect whether there is a data competition error in the access of the global variables, and the detection cost is reduced.

(3) The scalability of the present invention is good: due to the symbolic execution virtual machine adopted in the step (3) of the present invention, in the process of multi-thread extension to it, thread creation, thread waiting, thread locking and unlocking have been rewritten Handle synchronous functions, which can be expanded to implement more synchronous functions for detection, and the scalability is strong.

Description of drawings

FIG. 1 is an overall framework diagram of the data race detection and replay method based on a symbolic execution virtual machine in the present invention.

Fig. 2 is a flow chart of the data race detection and playback method based on symbolic execution virtual machine in the present invention.

Fig. 3 is a flowchart of data competition detection in the method of the present invention.

Fig. 4 is a flow chart of the running track of data competition recording in the method of the present invention.

Fig. 5 is a flow chart of the data contention replay method in the method of the present invention.

detailed description

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not constitute a conflict with each other.

As shown in Figure 1, the overall framework based on the data race detection and replay method based on the symbolic execution virtual machine of the present invention is composed of a host machine end and a target machine end. The host machine runs a graphical interface, sends relevant commands for data competition detection to the target machine through the network port, and processes the data sent from the target machine, performs relevant processing and screening, and finally displays it to the user; The command sent by the host machine performs data race detection or replay, and sends the execution information and detection results to the host machine through the network port. After receiving the detection result of data competition and the execution information of the program from the target machine, the detection result of data competition is displayed on the panel by analyzing and classifying the information, and the execution track of the program is drawn on another panel for the user view.

As shown in Figure 2, the data race detection and replay method based on the symbolic execution virtual machine of the present invention includes the following steps:

(1) Receive the program execution track operation request from the host computer, and judge whether the request is a replay request or a detection request, if it is a replay request, then enter step (2), otherwise enter step (5); specifically , if the flag in the request is replay, the request is a replay request, otherwise it is a detection request;

(2) judge whether the recording file corresponding to this replay request exists in the disk, if exist then go to step (3), otherwise process ends;

(3) Obtain the program execution information file in the record file, and execute the global variable read and write instructions and/or multi-thread function call instructions in the symbolic virtual machine (symbolic virtual machine) to call the detection function to load the thread instructions in the record file. Between constraints, and according to the constraints to determine whether the global variable read and write instructions and/or multi-threaded function call instructions can be executed, if it can then perform step (4), otherwise the process ends; in the present invention, the symbolic execution used The virtual machine is KLEE;

It should be noted that, before the process of the present invention is executed, the symbolic execution virtual machine needs to be multi-threaded, so as to generate a symbolic execution virtual machine with functions of multi-threaded function call interception and internal emulation of multi-threading.

Specifically, the process of executing multi-thread extension on the symbolic execution virtual machine is as follows: firstly, build the implementation framework of multi-thread simulation through function pointers and structures, and then target synchronization functions (thread creation, thread waiting, thread locking and unlocking) processing) to rewrite, where:

For the thread creation function, each instruction in the symbolic execution virtual machine is used as an execution path. The execution path has an independent execution environment and is stored in the container. Each execution path can be regarded as a thread. The process of thread creation is first Obtain instruction information, number of parameters, function address, and specific parameter information, and then create a new thread by calling the symbol to execute the internal function of the virtual machine, initialize the thread number and other information, and then initialize the timestamp vector of each thread, mainly The operation is to set the timestamp value corresponding to this thread to 1, and set the timestamp value of other threads to 0, and then merge the timestamp vector of the child thread and the parent thread to obtain a new timestamp vector and assign it to the child thread. Finally, the created thread is stored in the container of the symbolic execution virtual machine.

For the thread waiting function, take out the information of the waiting thread from the storage thread container, judge whether the waiting thread has ended, if not, then for the unfinished waiting thread, add the waiting thread to the waiting set of the calling thread, and wait for the thread to end When it is woken up, if it ends, scan the thread collection, wake up those threads waiting for it, and call the thread to do related exit cleanup operations.

For the thread locking function, first determine whether the lock is in the lock set. For a lock that does not exist in the lock set, it means that the lock is accessed for the first time, a lock object is created, and the lock is set to a busy state, and the lock holding thread Set as the current thread, for the lock in the idle state, set the lock to the busy state, set the lock holding thread as the current thread; for the busy state of the lock, set the current thread to the blocked state, and add the current thread to the lock In the waiting queue of the lock, it will be acquired after the lock is released.

For the thread unlock function, for the object of the lock in the busy state, an error is returned when the current thread does not hold the current lock; for the waiting queue of the lock that is not empty, the thread at the beginning of the queue is taken out, and the thread holding the lock Set it to this thread, and reset the waiting state of this thread, indicating that the thread is no longer in a blocked state. For a lock whose waiting queue is empty, set the lock to an idle state, and clear the thread holding the lock.

(4) Execute global variable read and write instructions and/or multi-thread function instructions, and the process ends;

(5) Execute the multi-threaded program corresponding to the detection request in the symbolic execution virtual machine, monitor the global variable access in the multi-threaded program between multiple threads, and record the order in which the multi-threads access the global variable, and record these The sequence information is recorded into the binary file to generate the program execution information file;

(6) Obtain the instruction information in the symbolic execution virtual machine, use the data competition detection algorithm to detect the reading and writing of global variables in the instruction information, and use the multi-thread function call interception function to call the multi-thread function in the instruction information to judge the global variable read and write Write whether there is a data race, if so, report the data race detection error information to the client, if the program execution is not over, obtain the next instruction information stored in the container of the symbolic execution virtual machine, and go to step (5) implement;

As shown in Figure 3, the step (6) of the inventive method specifically includes the following sub-steps:

(6-1) Execute the multi-threaded program corresponding to the request to obtain the executed instruction;

(6-2) execute the conditional branch structure in the virtual machine to judge the type of instruction by symbol, if the instruction is read instruction load or write instruction store, then proceed to substep (6-3); if the instruction is a function call instruction, then Go to sub-step (6-7); if the instruction is other instructions, continue to execute the instruction according to the execution track of the symbolic execution virtual machine, and then return to sub-step (6-1);

(6-3) Judging whether the memory address accessed by the instruction is a global variable, if so, proceed to substep (6-4), otherwise continue to execute the instruction according to the execution track of the symbolic execution virtual machine, and then return to substep (6-4) 1);

(6-4) Judge whether the instruction is a write access, if it is a write access, then call the write processing function callStore, and proceed to substep (6-5), if it is a read access, then call the read processing function callLoad, and turn to Into sub-steps (6-6);

(6-5) According to the memory address of the global variable (for the convenience of description, the address is assumed to be addr), check whether the global variable is the first write access, if so, the thread number that will access the global variable, and the timestamp of the thread Store it in the write record, and then return to substep (6-1); otherwise, take out the write record information of the memory address addr of the global variable, including the thread number and the timestamp vector. Compare the timestamp value of the current thread recorded in the timestamp vector of the current thread with the timestamp value in the write record information. If the former is smaller than the latter, it means that data competition occurs, and report the write data competition error of the global variable to Client, otherwise do not process, and continue to judge whether there is a concurrent read access identifier between the two write accesses. If there is a concurrent read access, then compare the timestamp value of the current thread recorded in this write access with the The timestamp value of the current thread recorded in the timestamp vector of all concurrent read access records is compared, as long as the less than relationship is satisfied, it means that data competition occurs, and the read and write data competition error of the global variable is reported to the client, and the concurrent read access records are cleared and stored All the information in, update the write access record, and then return to substep (6-1), otherwise clear all the information stored in the concurrent read access record, update the write access record, and then return to substep (6-1).

(6-6) According to the memory address of the global variable, check whether the global variable is the first read access, if it is the first read access, store the thread number and the timestamp value of the thread in the read record, and then return to the child Step (6-1); if it is not the first read access, then take out the write record information of the memory address of the global variable, including thread number and timestamp vector, etc., and record the time of the current thread in the timestamp vector of the current thread The stamp value is compared with the recorded timestamp value. If the two satisfy the less than relationship, it means that a data competition has occurred, and the data competition will be reported to the client. Otherwise, it will not be processed. At the same time, continue to judge whether there is a concurrent read access identifier between the two write accesses. , if there is concurrent read access, add this read access directly to the concurrent read access record whose address is addr; otherwise, take out the read access record information, including thread number and timestamp vector, etc., and add the current thread’s timestamp vector to Compare the recorded timestamp value of the current thread with the recorded timestamp value, if the two satisfy the less than relationship, update the concurrent read access flag, add this read access to the concurrent read access record whose address is addr, and then Update address is the read access record of addr, then return to substep (6-1);

(6-7) Check whether the current call function called by the function call instruction is a multi-thread function by means of character string matching, if yes, enter sub-step (6-8), otherwise continue to execute the instruction according to the execution track of the symbolic execution virtual machine ;

(6-8) At the function call instruction of the conditional branch structure in the symbolic execution virtual machine, the name of the currently called function is identified by string matching, and the parameters of the function are obtained, and the current call function is judged according to the name and parameters of the function. Thread creation function, thread waiting function, locking function, or unlocking function, if it is a thread creation function, then proceed to substep (6-9); if it is a thread waiting function, then proceed to substep (6-10) ; If it is a locking function, then proceed to substep (6-11); if it is an unlocking function, then proceed to substep (6-12);

(6-9) Call the thread creation function to create a child thread, the child thread initializes its own timestamp vector, and merges with the parent thread's timestamp vector, and the parent thread's timestamp value increases by 1;

(6-10) calling the thread waiting function to merge the timestamp vector of the waiting thread with the timestamp vector of the waiting thread;

(6-11) Call the locking function, and after successful locking, merge the timestamp vector of the thread and the timestamp vector of the lock object;

(6-12) Call the unlock function, and after successful unlocking, assign the thread's timestamp vector to the lock object's timestamp vector.

As shown in Figure 4, the step (5) of the inventive method comprises the following sub-steps:

(5-1) Detect the instruction that the corresponding multi-thread program of request is carried out, and judge whether this instruction is read instruction or write instruction, if then enter substep (5-2), otherwise if this instruction is lock instruction or For the unlock instruction, after executing the lock or unlock instruction, enter the sub-step (5-2), otherwise, continue to execute the instruction according to the execution track of the symbolic execution virtual machine;

(5-2) Determine whether the memory address accessed by the instruction is a global variable address, if so, obtain the address of the global variable, establish the mapping relationship between the global variable address and the number, and obtain the thread access global variable information, including the thread number and timestamp vector , if the instruction type is a write instruction, then proceed to substep (5-3), if the instruction type is a read instruction, then proceed to substep (5-4); otherwise, continue to execute the instruction according to the execution track of the symbolic execution virtual machine;

(5-3) Read the set of all read access records corresponding to the global variable according to the number corresponding to the global variable address, traverse all the read access records, and process each read access record in the following form: the thread of the read access record The timestamp value corresponding to the thread number is compared with the timestamp value corresponding to the same thread number of the current access record. If the less than or equal relationship is satisfied, no processing is performed. Otherwise, it indicates that there is a sequence relationship between the two operations and is recorded. Finally, the timestamp vector of all read access records is merged with the timestamp vector of the current access. Merging is called: traversing the two arrays in order of array subscripts from small to large, assuming that the subscript is i, the two arrays The larger value with the subscript i in the middle is assigned to the timestamp value with the subscript i of the currently accessed timestamp vector.

(5-4) Read the collection of all write access records corresponding to the global variable according to the number corresponding to the global variable address, and process each write access record in the following form: compare the timestamp value corresponding to the thread number of the write access record with Compare the timestamp values corresponding to the same thread number of the current access record. If the less than or equal relationship is satisfied, no processing will be performed. Otherwise, it indicates that there is a sequence relationship between the two operations and will be recorded. Finally, merge the timestamp vectors of all write access records with the currently accessed timestamp vector. The merging process is as follows: traverse the two arrays according to the order of the array subscripts from small to large. Assuming that the subscript is i, the two arrays The larger value with the subscript i is assigned to the timestamp value with the subscript i of the currently accessed timestamp vector;

(5-5) Write the set of all read-write access records into the record file, so as to complete the recording process of the execution track.

As shown in Figure 5, the step (3) of the inventive method comprises the following sub-steps:

(3-1) read the program execution information file, and write the record file in the program execution information file into the internal memory;

(3-2) According to the execution trajectory of the symbolic execution virtual machine, the instruction of the multi-threaded program corresponding to the detection request is cyclically selected, and it is judged whether the instruction is a read instruction, a write instruction, or a function call instruction, and if so, proceed to step (3-3) , otherwise continue to execute the instruction according to the execution track of the symbolic execution virtual machine;

(3-3) judge whether the memory address accessed by the instruction is a global variable, if it is, then enter the substep (3-4), otherwise if the instruction is the lock and unlock function in the multithreaded function call instruction, then enter the substep (3 -4), otherwise continue to execute the instruction according to the execution track of the symbolic execution virtual machine. .

(3-4) According to the file record information written into the internal memory, it is judged whether there is a dependency relationship in the current instruction, if there is no dependency relationship, then enter sub-step (3-5), otherwise the symbol is obtained to execute the next instruction in the virtual machine, And return sub-step (3-2) to carry out;

(3-5) Execute the current instruction, remove the dependencies between the instructions of the executed instructions, thereby update the file record information written into the memory, obtain the next instruction in the symbolic execution virtual machine, and return to the substep (3-5) when the program is not over 2) Execute until the end of the program.

Those skilled in the art can easily understand that the above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention, All should be included within the protection scope of the present invention.

Claims (6)

1. data contention based on a semiology analysis virtual machine detection and playback method, it is characterised in that comprise the following steps:

(1) receive the program from host end and perform track operation request, and judge that this request is playback request, or detection Request, if playback request, then enters step (2), otherwise enters step (5)；

(2) judge whether log file corresponding to this playback request is present in disk, if there is then proceeding to step (3), no Then process terminates；

(3) program obtained in log file performs message file, reads by obtaining the global variable in semiology analysis virtual machine Write command and/or multithreading function call instruction call detection function, to load the constraint between the instruction of log file thread Condition, and judge whether global variable read write command and/or multithreading function call instruction can be performed according to constraints, If can, performing step (4), else process terminates；

(4) performing global variable read write command and/or multithreading function instruction, process terminates；

(5) in semiology analysis virtual machine, perform the multithread programs that detection request is corresponding, monitor between multiple thread multi-thread Global variable access in Cheng Chengxu, and record the multithreading sequencing to global variable access, and by these sequencings Information recorded in binary file, to generate program execution message file；

(6) obtaining the command information in semiology analysis virtual machine, use in data contention detection algorithm detection command information is complete Office's variable read-write, and utilize the multithreading function in multithreading function call interception funcall command information, to judge the overall situation Whether variable read-write exists data contention, if it is reports to client by the information of data contention detection mistake, if program Perform not terminate, then obtain next command information deposited of container obtaining semiology analysis virtual machine, and forward step (5) to and hold OK.

Data contention the most according to claim 1 detection and playback method, it is characterised in that be additionally included in step (1) it Before, semiology analysis virtual machine is carried out multithreading extension, with generation, there is multithreading function call interception and internal emulation is multi-thread The semiology analysis virtual machine of Cheng Gongneng.

Data contention the most according to claim 2 detection and playback method, it is characterised in that semiology analysis virtual machine is held The process of row multithreading extension particularly as follows: first, realizes framework, so by what function pointer and structure built multithreading emulation After for include thread creation, thread wait, thread lock and unblock process synchronous function be written over.

Data contention the most according to claim 1 detection and playback method, it is characterised in that step (3) includes following son Step:

(3-1) reading program performs message file, and program performs the log file write internal memory in message file；

(3-2) instruction of multithread programs corresponding to detection request is chosen according to the track circulation that performs of semiology analysis virtual machine, Whether decision instruction is reading instruction, write command, function call instruction, the most then proceed to step (3-3), otherwise hold according to symbol The execution track of row virtual machine continues executing with this instruction；

(3-3) whether the memory address that decision instruction accesses is global variable, if yes then enter sub-step (3-4), the most such as Locking and de-locking function during really this instruction is multithreading function call instruction then enters sub-step (3-4), otherwise holds according to symbol The execution track of row virtual machine continues executing with this instruction..

(3-4) according to the file record information of write internal memory, it is judged that whether present instruction exists dependence, if there is not dependence Relation, then enter sub-step (3-5), otherwise obtains next instruction in semiology analysis virtual machine, and returns sub-step (3-2) Perform；

(3-5) perform present instruction, release the inter-instruction dependency relationship of executed instructions, thus the file of more newly written internal memory is remembered Record information, obtains next instruction in semiology analysis virtual machine, program not at the end of return sub-step (3-2) and perform, until EP (end of program).

Data contention the most according to claim 4 detection and playback method, it is characterised in that step (5) includes following son Step:

(5-1) instruction performed by multithread programs that detection request is corresponding, and judge whether this instruction is reading instruction or writes finger Order, if yes then enter sub-step (5-2), this instruction is to lock instruction or unlock instruction else if, then performing locking Or after unlocking instruction, enter sub-step (5-2), otherwise continue executing with this instruction according to the execution track of semiology analysis virtual machine；

(5-2) whether the memory address that decision instruction accesses is global variable address, if it is, obtain the address of global variable, Set up the mapping relations of global variable address and numbering, obtain thread accesses global variable information, including thread number and time Stamp vector, if instruction type is write command, then proceeds to sub-step (5-3), if instruction type is reading instruction, proceeds to sub-step (5- 4)；Otherwise continue executing with this instruction according to the execution track of semiology analysis virtual machine；

(5-3) set of all read access records corresponding to this global variable is read according to the numbering that global variable address is corresponding, Travel through all of read access record, to each read access record according to following formal layout: by the thread number pair of read access record Timestamp value corresponding to thread number that the timestamp value answered is identical with current accessed record compares, and is less than or equal to if meeting Relation, does not processes, and otherwise shows that two operations exist successively order relation, and carries out record.Finally all read accesses are remembered The vector time stamp of record merges with the vector time stamp of current accessed；

(5-4) set of all write access records corresponding to this global variable is read according to the numbering that global variable address is corresponding, To each write access record according to following formal layout: by timestamp value corresponding for the thread number of write access record and current accessed The timestamp value that the identical thread number of record is corresponding compares, if meeting less than or equal to relation, does not deals with, otherwise table There is successively order relation in bright two operations, and carries out record.Finally the vector time stamp of all write access records is visited with current The vector time stamp asked merges；

(5-5) set of all read and write access records is write log file, thus complete to perform the recording process of track.

Data contention the most according to claim 5 detection and playback method, it is characterised in that step (6) includes following son Step:

(6-1) multithread programs that request is corresponding is performed, to obtain the instruction performed；

(6-2) by the type of the conditional branching structure decision instruction in semiology analysis virtual machine, if instruction is reading instruction Load or write command store, then proceed to sub-step (6-3)；If instruction is function call instruction, then proceed to sub-step (6- 7)；Instruct for other if instructed, then continue executing with this instruction according to the execution track of semiology analysis virtual machine, be then back to son Step (6-1)；

(6-3) judge whether the memory address that this instruction accesses is global variable, is, proceeds to sub-step (6-4), otherwise basis The execution track of semiology analysis virtual machine continues executing with this instruction, is then back to sub-step (6-1)；

(6-4) judge whether this instruction is write access, if write access, then call and write process function callStore, and proceed to son Step (6-5), if read access, then calls reading and processes function callLoad, and proceed to sub-step (6-6)；

(6-5) according to the memory address addr of global variable, check whether global variable is first time write access, if it is will The timestamp of the thread number and thread that access this global variable is stored in write record, is then back to sub-step (6-1)；No Then take out the write record information of the memory address addr of global variable, including thread number and vector time stamp etc..By current thread Vector time stamp in the timestamp value of current thread of record compare with the timestamp value in write record information, if the former Less than the latter, represent and data contention occur, then report global variable write data contention mistake to client, do not make to locate Reason, continues to judge whether there is concurrent read access mark between twice write access, if there is concurrent read access, then by this simultaneously The timestamp value of the current thread of write access record with deposit twice write access all concurrent read access record timestamp to The timestamp value recording current thread in amount compares, as long as meeting less than relation, representing and data contention occurs, then report is complete The read-write data contention mistake of office's variable, to client, empties all information deposited in concurrent read access record, updates and writes visit Ask record, be then back to sub-step (6-1), otherwise empty all information deposited in concurrent read access record, update write access Record, is then back to sub-step (6-1).

(6-6) according to the memory address of global variable, check whether global variable is read access for the first time, if reading for the first time Access, then the timestamp value of thread number and thread is stored in read record, is then back to sub-step (6-1)；If not Read access for the first time, then take out the write record information of the memory address of global variable, including thread number and vector time stamp etc., and will In the vector time stamp of current thread, the timestamp value of current thread of record compares with the timestamp value of record, if both Meet less than relation, represent and data contention occurs, then report data contention is to client, does not deals with, continues to sentence simultaneously Whether there is concurrent read access mark between disconnected twice write access, if there is concurrent read access, then this read access directly being added It is added in the concurrent read access record that address is addr；Otherwise take out read access record information, including thread number and timestamp to Amounts etc., compare the timestamp value of current thread of record in the vector time stamp of current thread with the timestamp value of record Relatively, if both meet less than relation, then update concurrent read access flag, this read access is added to address be addr also Send out in read access record, then update the read access record that address is addr, and return sub-step (6-1)；

(6-7) by the way of string matching, check that what function call instruction called currently calls whether function is multi-thread eikonal Number, is to enter sub-step (6-8), otherwise continue executing with this instruction according to the execution track of semiology analysis virtual machine；

(6-8) at the function call instruction of semiology analysis virtual machine conditional branched structure, by the way of string matching Identifying and currently call the title of function, and obtain the parameter of function, title and parameter according to function judge currently to call function It is thread creation function, or thread waits function, or locking function, or unlock function, if thread creation function Then proceed to sub-step (6-9)；If thread waits function, then proceed to sub-step (6-10)；If locking function, then proceed to sub-step Suddenly (6-11)；If unblock function, then proceed to sub-step (6-12)；

(6-9) calling thread creation function creation sub-line journey, sub-line journey initializes own timestamp vector, and with father's thread time Between stab vector merge, simultaneously the timestamp value of father's thread from increase 1；

(6-10) call thread and wait function, so that the vector time stamp waiting thread is entered with by the vector time stamp waiting thread Row merges；

(6-11) locking function is called, and after successfully locking, by the vector time stamp of thread and the vector time stamp of lock object Merge；

(6-12) call unblock function, and after successfully unlocking, the vector time stamp of thread is given the timestamp of lock object to Amount.