VLDB2024_SecuDB：An In-enclave Privacy-preserving and Tamper-resistant Relational Database_字节跳动.pdf

迹部景吾

661

14页

4次

2024-09-09

免费下载

SecuDB: An In-enclave Privacy-preser ving and Tamper-resistant

Relational Database

Xinying Yang

ByteDance Inc

xinying.yang@bytedance.com

Cong Yue

National University of Singapore

yuecong@comp.nus.edu.sg

Wenhui Zhang

ByteDance Inc

wenhui.zhang@bytedance.com

Yang Liu

ByteDance Inc

liuyang.007@bytedance.com

Beng Chin Ooi

National University of Singapore

ooibc@comp.nus.edu.sg

Jianjun Chen

ByteDance Inc

jianjun.chen@bytedance.com

ABSTRACT

With the escalation in the demand for privacy-preserving and

tamper-resistant data management and processing on the pub-

lic cloud, an increasing number of mainstream databases start to

provide always-encrypted and blockchain-like features, including

Microsoft SQL Server, MongoDB, and Alibaba PolarDB. The recent

progress in Trusted Execution Environment (TEE) technology has

enabled the deployment of the complete database engine within

TEE. This implementation ensures that data stored in memory,

cache, and registers is encrypted, thereby maintaining the con-

dentiality of information. In this paper, we present SecuDB, a

multi-granularity privacy-preserving and tamper-resistant rela-

tional database by placing the entire RDBMS in Intel TDX. We

propose a novel visibility control mechanism incorporating column

masking, log masking, and statistics masking to realize ne-grained

privacy preservation and devise an isolated TEE-endorsed tempo-

ral table method to support ecient data and query veriability,

without aecting insertion and selection performance. We evaluate

SecuDB using Sysbench, TPC-C and TikTok copyright workloads.

The results show that compared with a system without an enclave,

SecuDB hits 84.7% and 94.7% of the performance when providing

coarse-grained and ne-grained privacy preservation, respectively.

While the overhead for tamper-resistance is less than 22.6%.

PVLDB Reference Format:

Xinying Yang, Cong Yue, Wenhui Zhang, Yang Liu, Beng Chin Ooi,

and Jianjun Chen. SecuDB: An In-enclave Privacy-preserving and

Tamper-resistant Relational Database. PVLDB, 17(12): 3906-3919, 2024.

doi:10.14778/3685800.3685815

1 INTRODUCTION

With the migration of sensitive data to the cloud, businesses, in-

stitutions, and individuals have endured higher risks associated

with data breaches, unauthorized access, and other security threats.

Consequently, the focus on both data security and privacy in the

cloud has become paramount. For example, enterprises are digitiz-

ing their business documents using a database system. It is essential

to ensure the contents of condential business documents are kept

This work is licensed under the Creative Commons BY-NC-ND 4.0 International

License. Visit https://creativecommons.org/licenses/by-nc-nd/4.0/ to view a copy of

this license. For any use beyond those covered by this license, obtain permission by

emailing info@vldb.org. Copyright is held by the owner/author(s). Publication rights

licensed to the VLDB Endowment.

Proceedings of the VLDB Endowment, Vol. 17, No. 12 ISSN 2150-8097.

doi:10.14778/3685800.3685815

secret and cannot be revealed to unauthorized users, attackers, and

database administrators. Moreover, the system must ensure the

documents are authentic and stored correctly.

Many database systems have been developed by mainstream

database service providers with the primary goal of safeguarding

data security and preserving data privacy. On the one hand, veri-

able databases [

–

] are built to ensure the integrity of the

data content. These systems often leverage cryptographic functions

and authenticated data structures to summarize database states into

a digest. The users can perform client-side verication with the

digest and proofs generated by the server to verify that the data

has not been tampered with. Moreover, systems can support server-

side verication with secure hardware to eliminate the network

transmission of proofs and relieve the burden of clients [

However, these systems do not protect data privacy. On the other

hand, encrypted databases are developed to ensure data privacy

[

]. In these systems, the clients employ encryption algo-

rithms such as the Advanced Encryption Standard (AES) [

] to en-

crypt data before transmitting it to the server and decrypting it upon

retrieval. The cryptographic algorithm ensures data privacy and

security, but the notable challenge limits its utilization in encrypted

databases due to its performance bottleneck and decient func-

tionality for ubiquitous cipher data-based computation; examples

include fully homomorphic encryption (FHE) techniques [

partially homomorphic encryption [

], and property-preserving

encryption methods [

]. Furthermore, systems [

]

employ secure hardware to support more operations on encrypted

data within the enclave. Once a ciphertext is delivered to the en-

clave, it rst decrypts the data, computes on the plaintext, and then

encrypts the data before replying to DBMS. We call this architec-

ture as partially hardware encrypted (P-HE). These systems are

designed with the assumption of a limited Enclave Page Cache

(EPC) size, making it impractical to run the entire DBMS within

the enclave. Consequently, they suer from signicant I/O costs

between the enclave and the DBMS. Besides, the DBMS run out-

side the enclave can be compromised, and therefore, data security

cannot be guaranteed.

To address the limitations of existing designs and inspired by

recent advancements in trusted execution environment technolo-

gies [

], such as Intel TDX, AMD SEV-SNP,

and ARM CCA, we have conceived and implemented SecuDB. It

stands as an in-enclave relational database utilizing full hardware

encryption (F-HE), to enable privacy-preserving and veriable func-

tionalities. This endeavor involves migrating the entire database

3906

management system (DBMS) into TEE, thus reshaping the prevail-

ing paradigm of partial hardware encryption (P-HE). We invent

a column mask-enabled visibility control method for select lists,

predicates, logs, and statistics, to realize ne-grained privacy preser-

vation, thereby eliminating the need for conventional client-side

encryption and decryption solutions. We isolate TEE endorsement

from general queries such as insertion and selection, which signi-

cantly outperforms conventional P-HE, and support both data and

query-oriented verication using a native temporal table.

SecuDB deploys the entire VeDB [

] engine into TDX with

ecient attestation and optimized system tuning. It is deployed

in TikTok [

] and Lark [

] to support privacy-preserving and

tamper-resistant management of the copyright and grounding, re-

spectively (detailed in Section 3.1). We conduct detailed perfor-

mance evaluation on SecuDB using two synthesis workloads, Sys-

bench and TPC-C, and a real-world workload consists of TikTok

•

We design and develop SecuDB, which, to the best of our

knowledge, is the rst in-enclave DMBS that serves both

multi-granularity privacy-preserving and tamper-resistant

functionalities.

•

We comprehensively analyze the threat model and database

kernel design principle for in-enclave privacy-preserving

and tamper-resistant databases.

•

We design a novel framework using column mask-enabled

visibility control to realize ecient ne-grained privacy

preservation in the F-HE system, which reforms client-side

encryption in conventional P-HE systems.

•

We enhance security and accountability within the isolated

environment by extending the attestation and trust chain

establishment processes between the Trust Domain (TD)

and the operating software components.

•

We devise TEE-endorsed advanced temporal tables to sup-

port ecient data and query veriability, signicantly en-

hancing the serviceability of conventional veriable databases.

•

We conduct experiments to evaluate the performance of Se-

cuDB. The experimental results show that SecuDB achieves

84.7% and 94.7% of the baseline throughput when provid-

ing coarse-grained and ne-grained privacy preservation,

respectively, and incurs 22.6% for tamper-resistance func-

tionalities.

The rest of this paper is organized as follows. Section 2 introduces

the background of TEE and databases in TEE. Section 3 outlines

system architecture with table and SQL enhancement of SecuDB.

Section 4 presents our design of attestation and trust chain in TDX.

Section 5 discusses ne-grained privacy-preserving in SecuDB real-

ized by visibility control. Section 6 describes the tamper-resistant

framework in our system. We then provide our experimental evalu-

ation in Section 7 before we conclude in Section 8.

2 BACKGROUND

In this section, we rst introduce the general principles of Trusted

Execution Environment (TEE) and functions in Intel Trust Domain

Extensions (TDX) and then present privacy-preserving and tamper-

resistant database-related works.

2.1 Trusted Execution Environment

A Trusted Execution Environment (TEE) oers an isolated envi-

ronment (i.e., secure enclave) for protecting data-in-use. TEE safe-

guards data and computations against threats, including from a

malicious host kernel or hypervisor. TEE has been a prominent fo-

cus of academic and industrial research for the past two decades [

], based on various TEE hardware platforms,

such as Intel Software Guard Extensions (Intel® SGX) [

], Intel

Trust Domain Extensions (Intel TDX) [

], ARM Trustzone [

Arm Condential Compute Architecture (ARM CCA) [

], AMD

Secure Encrypted Virtualization (SEV) [

], AMD Secure Encrypted

Virtualization and Secure Nested Paging (AMD SEV-SNP) [

] and

RISC-V Keystone [37].

In general, dierent hardware supports dierent levels of isola-

tion. For example, Intel SGX [

] and ARM Trustzone [

]

provide application code and data isolation through Enclaves and

Secure World, respectively. Intel TDX [

], ARM CCA [

AMD SEV [

] and AMD SEV-SNP [

] provides secure execution

environments that are completely opaque to privileged, untrusted

system software such as host OSes and hypervisors.

SecuDB relies on Intel Trusted Domain Extension to provide

a TEE-based virtual machine (VM) environment, which provides

execution domain isolation by encryption of memory and regis-

ters, integrity measurement, and remote attestation to ensure data

condentiality. Intel TDX VM instances, unlike Intel SGX, do not re-

quire additional development of a library OS to support application

workloads, thereby conserving engineering resources. Moreover,

Intel TDX VMs have the ability to fully utilize all CPU and memory

resources available on a physical node. This advantage facilitates

the management of large-memory workloads entirely within se-

cure memory, minimizing I/O operations and boosting performance

signicantly [15, 27, 59].

2.2 Intel Trust Domain Extensions

Intel’s TDX [

] provides isolation, condentiality, and integrity

at the VM level. It forties the condentiality of guest VMs against

the host system and physical security threats. This is achieved

through the isolation of guest register states and the encryption of

guest memory via secure page management. TDX module operates

in a privileged mode, acting as an intermediary between the host

and guest environments to oversee the separation between the two.

Intel TDX delivers two major functionalities. First, it ensures

the condentiality and integrity of memory and CPU states, safe-

guarding sensitive intellectual property and workload data against

threats from the host OS. Secondly, it enables remote attestation,

allowing a verifying entity, be it the workload’s proprietor or a user

of the workload’s services, to ascertain that the workload is opera-

tional on an Intel-TDX-enabled platform within a trusted domain

(TD) before sharing any workload-related data.

Memory Encryption by TDX. TDX introduces a new CPU

operating mode and utilizes memory encryption techniques to en-

sure isolation between two VMs. These VMs are encrypted using

dierent keys, directly managed by the TDX Module [

]. TDX

employs two complementary technical mechanisms: the Secure Ar-

bitration Mode (SEAM) CPU mode and the Multi-key Total-Memory

Encryption (MKTME) [58].

3907

of 14

免费下载

文档被以下合辑收录

VLDB2024 数据库顶会论文（共31篇）

本合辑收录了VLDB2024 数据库顶会论文。

关注

文档被以下合辑收录

评论