CN109522320B

CN109522320B - Optimization method for serving homomorphic encryption of database

Info

Publication number: CN109522320B
Application number: CN201811339890.3A
Authority: CN
Inventors: 牛自宾; 滕海明; 李卫明
Original assignee: Hangzhou Flk Information Safety Technology Co ltd
Current assignee: Hangzhou Flk Information Safety Technology Co ltd
Priority date: 2018-11-12
Filing date: 2018-11-12
Publication date: 2022-08-02
Anticipated expiration: 2038-11-12
Also published as: CN109522320A

Abstract

The invention discloses an optimization method for serving homomorphic encryption of a database. The optimization method adopts a Split Client/Server model to Split a complex nested SQL statement into two parts, one part executes ciphertext homomorphic operation at a MySQL Server end by rewriting the SQL statement, and the other part decrypts an obtained result at the Client end and then further processes the result. Through the Split Client/Server model, the problem that all records are acquired from the MySQL Server and then are processed locally is avoided, the computing power of the cloud Server is utilized to the maximum extent, the data volume transmitted from the Server to the Client end through a network is reduced, and the processing at the Client end is combined, so that the query performance is greatly improved.

Description

Optimization method for serving homomorphic encryption of database

Technical Field

The invention belongs to the technical field of information security, relates to a database sensitive information protection technology, and particularly relates to an optimization method for serving homomorphic encryption of a database.

Background

The database system is one of the cores of the information system, and with the development of cloud computing technology, the stand-alone database and the distributed database which are traditionally deployed in a local environment are gradually migrating to the cloud database deployed in a data center. Compared with a traditional database system, the cloud database has the characteristics of high expandability and high availability, and users can enjoy the advantages of low price, convenience, elasticity, reliability and the like of cloud computing.

However, outsourcing data to a cloud database brings security risks, which are equivalent to users abandoning part of control rights on data, and have to rely on and trust cloud service providers. In recent years, the distrust of users to cloud service providers is greatly aggravated by frequently exploded cloud service data leakage events, and data security becomes a great concern for users to adopt cloud computing and cloud databases.

An effective means for solving data security is to encrypt data, but this brings a new problem that the operation on the encrypted ciphertext cannot be supported, so that the operation on the plaintext on the cloud database cannot be normally used originally. That is, the cloud server only stores the data in an encrypted manner, but does not process the data, and each time the user retrieves, calculates or analyzes the data, the user must transmit the corresponding data back to the client, and then the data is decrypted and processed. This approach gives up the computing power of the cloud server and inevitably requires a large amount of data to be transmitted back to the client, and thus is unacceptable from a performance standpoint.

If the computing power of the cloud server is to be utilized, a homomorphic encryption algorithm needs to be adopted. Homomorphic encryption is used as an encryption means and supports direct operation on a ciphertext, and the operation result is equal to the result of the same operation on a plaintext after being decrypted. For example, the encryption functions a and b having the homomorphic property satisfy Dec (En (a)/,) En (b)/,) a |. b, where En is an encryption operation, Dec is a decryption operation, and |, respectively correspond to operations in plaintext and ciphertext fields. When ≦ indicates addition, it is called addition homomorphic encryption, which indicates an addition operation on the ciphertext field; when [ ] indicates multiplication, it is called multiplication homomorphic encryption, and [ ] indicates a multiplication operation on the ciphertext domain.

The existing homomorphic encryption algorithm (homomorphic 'full' means that calculation of any function can be satisfied), and due to the efficiency problem, the algorithm cannot be applied to actual projects. The known mature encryption algorithm can satisfy homomorphic operation of a ciphertext on one aspect, namely partial homomorphic encryption. For example, OPE (Order-Preserving Encryption) algorithm is a homomorphic Encryption algorithm supporting Order preservation, and Paillier algorithm is a homomorphic Encryption algorithm supporting addition.

The crypto DB cloud database encryption system of Ma province's engineering university provides an effective scheme for processing database ciphertext homomorphic operation. The idea for solving the problems is as follows: and a plurality of partial homomorphic encryption algorithms are adopted to achieve the effect of fully homomorphic encryption. The rough treatment steps are as follows: when a table is created, an original table building statement is rewritten, one field is expanded into a plurality of fields, for example, a salary field in a staff table is expanded into salaryeq, salryoorder and salryoaadd, and the encryption and decryption are respectively performed by using DET, OPE and HOM (i.e., Paillier) algorithms to support the equality, size comparison and addition and subtraction homomorphic operation of ciphertexts. When the salary field is inserted and updated, a DET (digital Top removal), OPE (optical edge inspection) and HOM (home position management) homomorphic encryption algorithm is respectively called to encrypt the value of the salary as the value of the salaryoEQ, the salaryoORDER and the salaryoADD, a certain expansion field is dynamically selected in the SQL statement query stage, and ciphertext homomorphic operation is performed in the MySQL Server. Such as a typical query statement: select name from staff work shop salary >5000, name is rewritten as query salaryoEQ field, filtering condition selects salaryoORDER field, and uses OPE algorithm to encrypt constant 5000, and rewrite as select nameoEQ from staff work shop salaryorder >6291927251322894654, and makes size comparison operation of cipher text in MySQL Server, filters out expected record, and then uses DET algorithm to decrypt result locally. Similarly, select sum (salary) from staff; rewriting as select sum (salaryoADD) from staff, performing sum aggregation operation in MySQL Server, and then locally decrypting the aggregation result by using the HOM algorithm.

However, there is a class of query statements, and CryptDB cannot be rewritten effectively, such as: the sum (salary) as total from staff group by means of special settlement total >100000 is grouped according to departments, and the departments with the total salary more than 100000 are searched, and sum (salary) aggregation requires that salary ADD is summed in the MySQL Server, but the summed result total does not have the property of ciphertext order preservation, namely, the operation of size comparison cannot be carried out at the MySQL Server.

Aiming at the problem, the invention provides a Split Client/Server model, an SQL statement is divided into two parts, one part of the SQL statement is rewritten, and homomorphic operation of a ciphertext is executed at a MySQL Server end; and acquiring an execution result at the Client terminal, decrypting the execution result, and locally filtering the other part of SQL statements. The Split Client/Server model serves for homomorphic encryption and decryption of a database realized by SQL rewriting, provides an optimized query execution plan, and performs query operation step by step according to the plan, so that the computing capability of a cloud Server is utilized to the maximum extent, the data volume transmitted from a Server to a Client network is reduced, and the query performance is greatly improved.

Disclosure of Invention

The invention aims to solve the defects of the prior art and provides a method which is reasonable in design, safe and efficient, namely serves homomorphic encryption and decryption realized by SQL rewriting and provides a Split Client/Server model. Aiming at the complex SQL query requirement of a user, the query statement is split, homomorphic operation of a ciphertext is executed at the MySQL Server end, and encryption, decryption and filtering processing are carried out at the Client end. The whole query statement is divided into a plurality of steps to be executed and finished, and finally the result expected by the user is returned.

The reason for causing the problems of the CryptDB cloud database is that: in order to support fully homomorphic ciphertext operation, the original field is expanded into a plurality of fields, and homomorphic operation is supported on a certain operation. This results in complex queries where the same expression may involve operating on different extension fields. Such as: a select name from staff name, a select sum, an equal comparison of the two EQ extension fields is performed. And a select name from staff name local is compared equally to the EQ extension field and the ADD extension field. In the first case, the SQL sentence can be directly rewritten, and the ciphertext homomorphic operation is carried out on the rewritten SQL sentence at the MySQL Server. In the second case, the SQL statement cannot be rewritten effectively, and step processing needs to be performed by adopting a Split Client/Server model.

In order to determine whether the query SQL statement supports direct rewrite, the method herein is: the MySQL statement is parsed into an abstract syntax tree AST, each tree node is considered to be an item, and the expression itself is also an item. An item type is defined herein and is denoted as item Ω, and is characterized in that operations are performed on different extension fields in an expression, such as a sample sum (equal) extension field, the EQ extension field is used before and after the requirement of the equal number, the sample can support operations of EQ, ORDER, and ADD extension fields in MySQL Server, and the sum (sample) can only support operations of ADD extension field in MySQL Server, so that the SQL statement cannot support direct rewriting. That is, if the query SQL statement contains item Ω, step processing needs to be performed by using a Split Client/Server model.

The Split Client/Server model is divided into two steps: the first step is to analyze SQL sentences into abstract syntax trees, and recursively traverse the syntax trees, namely recursively traverse the sub-queries from the father queries to generate call stacks of SQL sentence execution plans; the second step is that a sub-query is popped from the stack top, an SQL statement is rewritten and sent to the MySQL Server for execution, the obtained result is decrypted in the Client, then the next execution plan of the stack top is executed until the whole stack is empty, and the final result is obtained.

Through the scheme provided by the invention, the technical effects are as follows: and screening out the parts supporting the ciphertext homomorphic operation in the complex query SQL sentence, executing the ciphertext homomorphic operation at the MySQL Server terminal, and combining the processing at the Client terminal to obtain a query result. The method can utilize the computing power of the cloud server to the maximum extent, and reduce the data volume transmitted from the server side to the client side through the network, thereby greatly improving the query performance.

Drawings

FIG. 1 is a flow chart of SQL statement execution plan generation

FIG. 2 is a flow chart of the execution of the SQL statement execution plan

FIG. 3 is a diagram of an example of an execution plan for SQL statement splitting

FIG. 4 is a diagram illustrating an example of the execution flow of an SQL statement

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in detail below with reference to the accompanying drawings and embodiments.

As shown in fig. 1, the method for generating an execution plan of an SQL statement by using a Split Client/Server model includes the following specific steps:

in step s101, recursively checking items one by one starting from the root node of the parent query;

in step s102, checking ORDER by which requires ORDER fields to be sorted, and comparing whether the object of select supports the ORDER fields;

in step s103, judging whether the current item is an item omega, if so, entering step s104, otherwise, skipping step s104 and checking the next item;

in step s104, putting the item of order by into a stack, and marking the item as being sequenced after the Client decrypts;

at step s105, checking group by hashing, which may require an ORDER field (such as hashing max.) or an ADD field (such as hashing sum.), checking a packet filtering condition to determine whether the packet filtering condition matches the field required by group by hashing, and if the object compared by hashing is a sub-query, recursively processing the sub-query;

in step s106, judging whether the current item is an item omega, if so, entering step s107, and if not, skipping step s107 and checking the next item;

in step s107, the item of group by enjoying is put into a stack and marked as packet filtering after Client decryption;

in step s108, checking the where condition, if the object compared by the where is the sub-query, recursively processing the sub-query;

in step s109, judging whether the current item is item Ω, if yes, entering step s110, and if no, skipping step s110 and checking the next item;

at step s110, place this item of where into the stack, and mark as filtering after the Client decrypts;

in step s111, the SQL statement for the rest of the query is rewritten and sent to the MySQL service to execute the ciphertext homomorphic operation.

And (3) forming a call stack by using the Split Client/Server model, wherein the stack top is the execution plan of the innermost layer of the sub-queries. And rewriting SQL sentences of the sub-queries at the stack top, sending the SQL sentences to the MySQL Server to execute homomorphic operation of the ciphertext, decrypting after the Client obtains a result, and processing the sub-queries of the next layer.

FIG. 2 shows, by way of a typical example, a multi-level nested sub-query composed of where conditions, illustrating the execution flow as follows:

in step s201, it is an example of a multi-layered nested sub-query composed of where conditions, consider that where 1> xxx, the greater than number requires the ORDER field, and the object of the internal sub-query is sum (c1), i.e., the ADD field, the expression item is item Ω, which does not support direct rewrite. Similarly, where a1 is equal to the number requirement EQ field, and the object of the internal sub-query is sum (b1), i.e. ADD field, the expression item is also item Ω, and direct rewriting is not supported;

in step s202, rewriting the innermost sub-query select sum (C1) from t3 where C1 ═ const _ C, executing the rewritten SQL statement in MySQL Server, performing homomorphic operation on the ciphertext, obtaining an execution result in the Client, and decrypting to obtain const _ B;

at step s203, the sub-query of the innermost layer is eliminated and replaced with const _ B;

in step s204, rewriting select sum (B1) from t2 where B1> const _ B in the inner-layer sub-query, executing the rewritten SQL statement in the MySQL Server, performing homomorphic operation on the ciphertext, obtaining an execution result in the Client, and decrypting the execution result to obtain const _ A;

at step s205, the sub-query of the inner layer is eliminated, and const _ A is used instead;

in step s206, the parent query select a1 from t1 where a1 is rewritten as const _ a, the rewritten SQL statement is executed in MySQL Server, the ciphertext is homomorphically operated, the execution result is obtained in the Client, and the Client decrypts the execution result to obtain the final query result.

In the above example, only the sub-queries formed by the where condition are considered, the actual scene often includes sub-queries and sorting of packet filtering, and the judgment logic and the execution flow are similar, which are not described herein again.

The Split Client/Server model proposed herein is further explained with reference to an example in a real-world scenario. Consider the SQL statement that creates the table: create table status (id int, name varchar (64), salary float, local varchar (64), subcompany varchar (64), primary key (id)); the table staff is used for storing information of the staff, including staff names, salaries, departments where the staff are located and affiliated subsidiaries, each field in the table is expanded into a plurality of fields, and ciphertext homomorphic operation is supported.

Considering a typical SQL query, in Hangzhou branch companies, the total salary of the query department is larger than that of the market department and the total salaries of the department, and the query SQL statements are arranged in descending order according to the total salaries of the department as follows:

select department,sum(salary)as department_total

from staff

where

subcompany＝'Hangzhou Branch'

group by department

having department_total>

(

select sum(salary)from staff

where subcompany＝'Hangzhou Branch'and department＝'Market department'

)

order by department_total desc；

for the above query of the SQL statement, the query is first parsed into an abstract syntax tree (abstract syntax tree AST), each part is regarded as an item, the abstract syntax tree is recursively traversed, and an execution plan for querying the SQL statement is generated by using a Split Client/Server model.

Fig. 3 shows an execution plan generated by the query SQL statement.

As shown in fig. 4, for the generated execution plan, the specific execution steps are as follows:

in step s401, at the Client, rewriting the query conditions of the sub-query, using the EQ extension field, and encrypting the constant using the DET algorithm, where the constant is rewritten as subcomma nyeq ═ n.. q. ═ d..4\ rO.. 'and deparattoeq ═ 2.. L. ].1.>. d. _ b.1.. g.. ski.t';

at step s402, rewriting the object of the sub-query at the Client end, and for the aggregation result, using ADD extension field to obtain a rewritten SQL statement of the sub-query, where the rewritten SQL statement is a select sum (salaryoad) from staff work subbasic eq ═ n.q.. d.4 \ rO... and deparattmoeq ═ 2.. L.. 1.>. d.. b.1.. g.. ski.t';

in step s403, the sub-query rewritten SQL statement is executed at the MySQL Server, and homomorphic operation is performed on the ciphertext;

in step s404, at the Client, obtaining a result of the sub-query, and decrypting the result by using an HOM algorithm to obtain a plaintext record set;

in step s405, rewriting a parent query at the Client end, similar to the process of rewriting a child query, where the SQL statement rewritten by the parent query is a select deparatmentoeq, sum (salaryoadd) from staff work subbasic eq '. n.q.d. 4\ rO..' group by deparatmentoeq;

in step s406, the SQL statement rewritten by the parent query is executed at the MySQL Server, and homomorphic operation is performed on the ciphertext;

in step s407, at the Client, obtaining a result of the sub-query, and decrypting the result by using an HOM algorithm to obtain a plaintext record set;

in step s408, traversing the result of the parent query in the Client, and filtering out the records meeting the conditions;

in step s409, the priority _ total is sorted in descending order and output in the Client.

The above description is only exemplary of the present invention and should not be taken as limiting the invention, as any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An optimization method for serving database homomorphic encryption, comprising the steps of:

the method comprises the following steps: analyzing and inquiring SQL sentences into an abstract syntax tree, wherein each tree node is an item, the expression is also an item, recursively traversing the syntax tree, checking whether each item supports ciphertext homomorphic operation of rewriting SQL sentences in a MySQL Server, and marking the item type as an item omega, and the method is characterized in that different extension fields are operated in the expression to generate a calling stack of an execution plan, and the execution plan of the SQL sentences is generated by adopting a Split Client/Server model, and the method specifically comprises the following steps:

step s104, putting the item of order by into a stack, and marking the item as being sequenced after the Client decrypts;

step s105, checking group by hashing, wherein the group by hashing requires an ORDER field or an ADD field, checking whether a grouping filtering condition is matched with the field required by the group by hashing, and if the object compared by hashing is a sub-query, recursively processing the sub-query;

step s108, checking the where condition, if the object of the where comparison is the sub-query, recursively processing the sub-query;

in step s111, rewriting the SQL sentences of the rest part of the query, sending the rewritten SQL sentences to the MySQL service to execute the cryptograph homomorphic operation, and forming a call stack by adopting a Split Client/Server model and the generated execution plan, wherein the top of the stack is the execution plan of the innermost layer of the sub-query;

step two: and rewriting the sub-query of the innermost layer at the stack top according to the call stack of the execution plan, executing ciphertext homomorphic operation at the MySQL Server end, acquiring and decrypting the result at the Client end, popping the sub-query at the stack top, and processing the next sub-query at the stack top until the whole stack is empty to acquire the final query result.

2. The optimization method for serving the homomorphic encryption of the database according to claim 1, characterized in that, in the first step, the complicated nested SQL statement is Split by using a Split Client/Server model, one part of the rewritten SQL statement performs ciphertext homomorphic operation at the MySQL Server end, and the other part of the rewritten SQL statement performs sorting and filtering processing after the Client end obtains the result from the MySQL Server and decrypts the result.