CN110543419A

CN110543419A - intelligent contract code vulnerability detection method based on deep learning technology

Info

Publication number: CN110543419A
Application number: CN201910802157.9A
Authority: CN
Inventors: 邱炜伟; 李伟; 李启雷; 梁秀波; 尹可挺; 周健
Original assignee: Hangzhou Qulian Technology Co Ltd
Current assignee: Hangzhou Qulian Technology Co Ltd
Priority date: 2019-08-28
Filing date: 2019-08-28
Publication date: 2019-12-06
Anticipated expiration: 2039-08-28
Also published as: CN110543419B; WO2021037196A1

Abstract

The invention discloses an intelligent contract code vulnerability detection method based on a deep learning technology, which specifically comprises the following steps: serializing and preprocessing grammar nodes of the intelligent contract source code to obtain node vectors; the obtained node vector is used as the input of a bidirectional long-short term memory network, the output of a global maximum pooling layer in a neural network is obtained and is used as the intermediate representation of an intelligent contract source code; and taking the obtained intermediate representation of the intelligent contract source code as the input of a random forest classifier, and training the random forest classifier to obtain a trained classifier to perform new intelligent contract code vulnerability detection. Compared with the conventional code vulnerability detection mode which is detected by adopting a security rule-based technology or formal verification, the method is more sensitive to new code vulnerability detection, and developers do not need to make and add corresponding rules or formal specifications in time.

Description

Intelligent contract code vulnerability detection method based on deep learning technology

Technical Field

the invention relates to the technical field of intelligent contract and code vulnerability detection, in particular to an intelligent contract code vulnerability detection method based on a deep learning technology.

Background

the block chain is a distributed data management technology for realizing decentralization based on data encryption, timestamps and a distributed consensus mechanism, and has the characteristics of traceability, no tampering and high availability. Smart contracts are a computer protocol intended to propagate, verify or execute contracts in an informational manner that allows trusted transactions to be conducted without third parties, which transactions are traceable and irreversible. The emergence of the block chain technology provides a set of digital systems capable of supporting programmable for intelligent contracts, the intelligent contracts cannot be changed once the intelligent contracts are successfully issued on a block chain, and huge loss is brought if security vulnerabilities exist, so that code vulnerability detection on the intelligent contracts is very necessary.

At present, the mainstream solutions for intelligent contract vulnerability detection all adopt security rule-based technologies or formal verification, which means that when a new code vulnerability is exposed, developers need to make and add corresponding rules or formal specifications in time to update a vulnerability detection system. Based on the above deficiency of intelligent contract vulnerability detection, a more flexible and powerful vulnerability detection mechanism is needed to make up for. At present, deep learning technology under artificial intelligence is broken through, more and more available training data are generated by popularization of block chain technology application, and how to design a set of detection system which is more sensitive to the security vulnerability of the intelligent contract according to the characteristics of the intelligent contract and the writing language thereof is also a breakthrough point for researching the security problem of the intelligent contract.

Disclosure of Invention

The invention aims to provide an intelligent contract code vulnerability detection method based on a deep learning technology aiming at the defects of the prior art, and the method can improve the accuracy and comprehensiveness of intelligent contract code vulnerability detection. The specific technical scheme is as follows:

an intelligent contract code vulnerability detection method based on a deep learning technology is characterized by specifically comprising the following steps:

S1: serializing and preprocessing grammar nodes of the intelligent contract source code to obtain node vectors;

S2: taking the node vector obtained in the step S1 as the input of a bidirectional long-short term memory network, and obtaining the output of a global maximum pooling layer in a neural network as the intermediate representation of an intelligent contract source code;

S3: and (4) taking the middle representation of the intelligent contract source code obtained in the step (S2) as the input of a random forest classifier, training the random forest classifier, and carrying out new intelligent contract code vulnerability detection on the trained classifier.

Further, the S1 specifically includes:

S1.1: carrying out syntax analysis on the intelligent contract source code character string by using a syntax analysis tool to generate an abstract syntax tree;

S1.2: performing depth-first traversal on the abstract syntax tree, and serializing the abstract syntax tree into a one-dimensional vector consisting of syntax nodes;

s1.3: preprocessing the one-dimensional vector, and generating a preprocessed node vector through grammar marking and filling truncation; in the preprocessing process, the integrity of a code structure and semantics needs to be kept.

further, the one-dimensional vectors in S1.2 are an ordered list composed of syntax nodes, and each one-dimensional vector represents a corresponding intelligent contract function.

Further, the padding truncation bit length c in S1.3 should be between the longest one-dimensional vector list length and the shortest one-dimensional vector list length.

Further, the padding truncation rule in S1.3 is to perform last list bit truncation on a list length greater than the padding truncation bit length c, so that the length of the list is equal to the bit length c; and carrying out zero filling on the last bit of the list to make the length of the list equal to the bit length c, wherein the length of the list is less than the length c of the filling truncation bit.

The invention has the following beneficial effects:

according to the intelligent contract code vulnerability detection method based on the deep learning technology, disclosed by the invention, the intelligent contract code vulnerability can be more flexibly and accurately predicted and positioned by training the neural network model for the intelligent contract code vulnerability. Compared with the conventional code vulnerability detection mode which is detected by adopting a security rule-based technology or formal verification, the method is more sensitive to new code vulnerability detection, and developers do not need to make and add corresponding rules or formal specifications in time. Meanwhile, along with the popularization of the application of the block chain technology, more and more available intelligent contract code data sets are helpful to the perfection of the model. Under limited experimental data (about 3000 data sets are subjected to 10-fold cross validation), ideally, the model prediction accuracy can reach 85% or even more than 90%.

Drawings

FIG. 1 is a flow chart of an intelligent contract code vulnerability detection method based on deep learning technology.

Detailed Description

The present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments, and the objects and effects of the present invention will become more apparent, it being understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

As shown in fig. 1, the intelligent contract code vulnerability detection method based on the deep learning technology specifically includes the following steps:

S1.1: the method comprises the steps of compiling a matching list according with intelligent contract grammatical rules by using a grammatical parsing tool such as Antrl4, Yacc and the like, and carrying out grammatical parsing on intelligent contract source code character strings to generate an abstract syntax tree AST;

s1.2: and sequencing each grammar node by adopting a depth-first traversal (DFT) mode, wherein each grammar node has a one-dimensional vector representation. Meanwhile, the serialized vectors well keep the structure and semantic information of the nodes. Each one-dimensional vector corresponds to a function in the intelligent contract one-to-one, and each one-dimensional vector can represent the corresponding intelligent contract function.

s1.3: and preprocessing the one-dimensional vector, wherein the operation is divided into a mode of syntax marking and filling truncation. The syntax notation refers to mapping each component node of the one-dimensional vector into a plurality of integer components, such as "public" for number 1, and "private" for number 2, etc. to improve the processing capability; the filling truncation is to standardize the length of each one-bit vector, facilitate the processing of data, select a proper value c for filling truncation, perform filling operation of last bit supplement 0 on the vector with the bit length smaller than c, and perform truncation operation on the vector with the bit length larger than c.

The padding truncation bit length c should be between the longest one-dimensional vector list length and the shortest one-dimensional vector list length.

the filling truncation rule is that the list end bit truncation is carried out on the list length which is greater than the filling truncation bit length c, so that the length of the list end bit truncation is equal to the bit length c; and carrying out zero filling on the last bit of the list to make the length of the list equal to the bit length c, wherein the length of the list is less than the length c of the filling truncation bit.

S2: processing node information by the LSTM neural network;

Taking the node vector obtained in the step S1 as the input of a bidirectional long-short term memory network, and obtaining the output of a global maximum pooling layer in a neural network as the intermediate representation of an intelligent contract source code;

the LSTM is a modified recurrent neural network, and can solve the problem that RNN cannot handle long-distance dependence. The LSTM neural network is adopted, the problem of vulnerability of part of intelligent contracts is contained in a grammatical structure of the source code, and the structure and semantic information of the source code are reserved to a certain extent by constructing an abstract grammatical tree and performing depth-first traversal. By adopting the representation form, the processing advantages of the bidirectional LSTM neural network on the time series data can be fully utilized.

The operation needs proper granularity to measure the grade of the vulnerability, and the vulnerability library and the code library are researched and identified by the vulnerability.

it will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and although the invention has been described in detail with reference to the foregoing examples, it will be apparent to those skilled in the art that various changes in the form and details of the embodiments may be made and equivalents may be substituted for elements thereof. All modifications, equivalents and the like which come within the spirit and principle of the invention are intended to be included within the scope of the invention.

Claims

1. an intelligent contract code vulnerability detection method based on a deep learning technology is characterized by specifically comprising the following steps:

s2: and taking the node vector obtained in the step S1 as the input of the bidirectional long-short term memory network, and obtaining the output of the global maximum pooling layer in the neural network as the intermediate representation of the intelligent contract source code.

2. The intelligent contract code vulnerability detection method based on deep learning technology according to claim 1, wherein the S1 is specifically:

3. The intelligent contract code vulnerability detection method based on deep learning technology of claim 2, wherein the one-dimensional vectors in S1.2 are an ordered list composed of grammar nodes, each one-dimensional vector representing a corresponding intelligent contract function.

4. the method of claim 2, wherein the padding truncation bit length c in S1.3 is between the longest one-dimensional vector list length and the shortest one-dimensional vector list length.

5. the intelligent contract code vulnerability detection method based on deep learning technology of claim 2, wherein the filling truncation rule in S1.3 is to perform list last bit truncation for the length of the list being greater than the filling truncation bit length c to make the length equal to the bit length c; and carrying out zero filling on the last bit of the list to make the length of the list equal to the bit length c, wherein the length of the list is less than the length c of the filling truncation bit.