A survey of confidential data storage and deletion methods

A Survey of Confidential Data Storage and Deletion Methods
SARAH M. DIESBURG and AN-I ANDY WANG

2

Florida State University

As the amount of digital data grows, so does the theft of sensitive data through the loss or misplacement
of laptops, thumb drives, external hard drives, and other electronic storage media. Sensitive data may also
be leaked accidentally due to improper disposal or resale of storage media. To protect the secrecy of the
entire data lifetime, we must have confidential ways to store and delete data. This survey summarizes and
compares existing methods of providing confidential storage and deletion of data in personal computing
environments.
Categories and Subject Descriptors: D.4.6 [Operating Systems]: Security and Protection—Cryptographic
controls; H.3.2 [Information Storage and Retrieval]: Information Storage; K.4.1 [Computers and
Society]: Public Policy Issues—Privacy; K.6.5 [Management of Computing and Information Systems]:
Security and Protection
General Terms: Human Factors, Legal Aspects, Performance, Security
Additional Key Words and Phrases: Data overwriting techniques, data theft, personal privacy, secure deletion, secure erasure
ACM Reference Format:
Diesburg, S. M. and Wang, A. A. 2010. A survey of confidential data storage and deletion methods. ACM
Comput. Surv. 43, 1, Article 2 (November 2010), 37 pages.
DOI = 10.1145/1824795.1824797 http://doi.acm.org/10.1145/1824795.1824797

1. INTRODUCTION

As the cost of electronic storage declines rapidly, more and more sensitive data is stored
on media such as hard disks, CDs, and thumb drives. The trend of the paperless office
also drives businesses toward converting sensitive documents, once stored in locked
filing cabinets, into digital forms. Today, an insurance agent can carry a laptop that
holds thousands of Social Security numbers, medical histories, and other confidential
information.
As early as 2003, the U.S. Census Bureau reported that two-thirds of American
households have at least one computer, with about one-third of adults using computers to manage household finances and make online purchases [U.S. Census Bureau

This research was supported in part by the U.S. Department of Energy, Grant P200A060279.
Authors’ address: Sarah Diesburg (contact author; email: diesburg@cs.fsu.edu) and An-1 Andy Wang, 253
Love Building, Department of Computer Science, Florida State University, 32306.
Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted
without fee provided that copies are not made or distributed for profit or commercial advantage and that
copies show this notice on the first page or initial screen of a display along with the full citation. Copyrights
for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, to republish, to post on servers, to redistribute to lists, or to use any component
of this work in other works requires prior specific permission and/or a fee. Permissions may be requested
from Publications Dept., ACM, Inc., 2 Penn Plaza, Suite 701, New York, NY 10121-0701 USA, fax +1 (212)
869-0481, or permissions@acm.org.
c 2010 ACM 0360-0300/2010/11-ART2 $10.00
DOI 10.1145/1824795.1824797 http://doi.acm.org/10.1145/1824795.1824797

ACM Computing Surveys, Vol. 43, No. 1, Article 2, Publication date: November 2010.

2:2

S. M. Diesburg and A. A. Wang

2005]. These statistics suggest that many computers store data on personal finances
and online transactions, not to mention other confidential data such as tax records,
passwords for bank accounts, and email. We can estimate that these figures have risen
dramatically since the 2003 survey.
Sensitive information stored in an insecure manner is vulnerable to theft. According to the most recent CSI Computer Crime and Security Survey [Richardson 2007],
50 percent of the respondents have been victims of laptop and mobile theft in the last
12 months. These respondents include 494 security practitioners in U.S. corporations,
government agencies, financial institutions, medical institutions, and universities. The
survey also shows that between the years 2001 and 2007, the theft of electronic storage
occurred much more frequently than other forms of attack or misuse such as denial of
service, telecom fraud, unauthorized access to information, financial fraud, abuse of
wireless networks, sabotage, and Web site defacement. Many incidences of theft continue to make headlines across many segments of society, including the government
[Hines 2007], academia [Square 2007], the health industry [McNevin 2007; Sullivan
2005], and large companies [WirelessWeek 2007].
Two major components exist to safeguard the privacy of data on electronic storage
media. First, data must be stored in a confidential manner to prevent unauthorized
access, and the solution should not impose significant inconvenience during normal
use. Second, at the time of disposal, confidential data must be removed from the storage media as well as the overall computing environment in an irrecoverable manner.
While the fundamental goals are clear, existing solutions tend to evolve independently
around each goal. This survey summarizes the advantages and challenges of various
confidential data storage and deletion techniques, with the aim of identifying underlying trends and lessons in order to arrive at an overarching solution.
Due to the size of the problem, the focus of this survey is on nondistributed, singleuser computing environments (e.g., desktops and laptops). The user is assumed to have
system administrative privileges to configure the confidentiality settings of storage
and deletion methods. The confidentiality threat model assumes that attacks to recover
sensitive data are staged after the computer has been powered off: in other words, the
attacker uses “dead” forensic methods. We assume that an attacker can gain access to
the original storage media either by booting the original machine with a live CD or
removing the storage media and placing it in another machine under the attacker’s
control. Other forms of attack (e.g., network-based attacks, memory-based attacks, or
privilege escalation attacks) are beyond the scope of this survey. Thus, the strength of
the confidentiality of the data storage or deletion is based directly upon the technique
used to store or delete the data at rest (for example, the strength of the encryption
technique).
2. SECURITY BACKGROUND

This section is designed for storage researchers; it provides the relevant security concepts when comparing storage designs.
The general concept of secure handling of data is composed of three aspects: confidentiality, integrity, and availability. Confidentiality involves ensuring that information is not read by unauthorized persons. Using encryption to store data or authenticating valid users are examples of means by which confidentiality is achieved. Integrity
ensures that the information is not altered by unauthorized persons. Storing a message authentication code or a digital signature computed on encrypted data is a way to
verify integrity. Finally, availability ensures that data is accessible when needed. Having multiple servers withstand a malicious shutdown of a server is one way to improve
availability.


2:3

This survey compares various confidential storage and deletion approaches in terms
of how each trades confidentiality with convenience (e.g., ease-of-use, performance,
and flexibility in setting security policies). Both integrity and availability goals are
assumed, but are beyond the scope of this survey.
The strength of confidentiality is the result of how well secure storage and deletion
mechanisms address the following questions:
—If encryption is used, how well are the keys protected?
—Do copies of sensitive data reside at multiple places in the system?
—If encryption is used, how strong is the encryption mechanism and mode of operation
in terms of the computational efforts to subvert the encryption?
—Can deleted data be recovered? If so, what are the time and resource costs?
—Is the entire file securely deleted, or is some portion left behind (such as the file name
or other metadata)?
In other words, the confidential data must not be accessed by unauthorized persons
after it is properly stored or deleted.
The ease-of-use of an approach reflects the level of inconvenience imposed on end
users. Methods of confidential storage and deletion that are too hard to use will either
encourage users to circumvent them or discourage users from using them entirely
[Whitten and Tygar 1999]. Some aspects of the user model examined include
—the number of times a person must enter an encryption key per session;
—the ease with which the method is invoked; and
—the number of encryption keys or passwords a person or a system must remember.
Levels of inconvenience vary depending on the type of applications used. For example,
entering a password to access each sensitive file in a text editor environment is less inconvenient than having to enter a password every time a compiler accesses a sensitive
file, since a compiler must generally access many files in a short period of time. In the
most ideal environment, users should behave similarly whether accessing secure files
or normal files.
Performance is another form of convenience, as methods that either take too long or
consume unreasonable amounts of system resources will not be used. For both confidential storage and deletion, performance can be measured by the latency and bandwidth of file access/erasure and overhead pertaining to the encryption algorithm and
the mode of operation used. Additionally, both methods can be measured by the time
taken per operation and by the total amount of system resources used.
Security policies are comprised of a set of rules, laws, and practices that regulate how
an individual or organization manages, distributes, and protects secure information. A
policy may be specific to a person or organization and may need to change frequently.
This survey compares the flexibility of the method, or ease of configuration, with regards to the implementation of various confidential storage and deletion policies. Some
aspects that we examine include
—the method compatibility with legacy applications and file systems;
—the ease of key or password revocation;
—how easily one may change the method’s configuration to fulfill a security policy (e.g.,
encryption algorithm and key size); and
—whether one can control the granularity (e.g., file and disk partition) of confidential
storage and deletion operations.

2:4


Many techniques for confidential storage and deletion involve cryptography. The following section briefly introduces and compares commonly used encryption algorithms
and their modes of operation.
2.1. Encryption Basics

Encryption is a procedure used in cryptography “to scramble information so that only
someone knowing the appropriate secret can obtain the original information (through
decryption)” [Kaufman et al. 2002]. The secret is often a key of n random bits of zeros
and ones, which can be derived through the use of a password or passphrase. A key’s
strength is often associated with the length of the key, which, if it consists of truly
random bits, requires a brute-force enumeration of the key space to decrypt the original message. Another measure of a key’s strength is its entropy, which measures the
degree of randomness associated with the bits in the key.
An encryption algorithm, or cipher, takes an input (referred to as plaintext) and
produces encrypted output (i.e., ciphertext); similarly, a decryption algorithm takes
a ciphertext as input and generates decrypted plaintext. Plaintext can be text files,
videos, music, executables, entire disk partitions, or even encrypted files for nested
encryptions. Thus, encrypting a music file is no different than encrypting a text file,
unless stated otherwise.
Encryption algorithms can be either symmetric or asymmetric. Symmetric algorithms use the same key for both encryption and decryption. Asymmetric algorithms
use two keys: one for encryption and another for decryption. For example, public-key
cryptography, which is a form of asymmetric encryption, uses two keys (public and
private keys), and is often used to establish secure communication across a network
where there is no way of exchanging a symmetric key beforehand. Symmetric encryption schemes can be many times faster than comparable asymmetric schemes, and are
therefore used more often in secure data storage, especially when the data in question
does not traverse through an insecure network.
Common symmetric key encryption algorithms include the Data Encryption Standard (DES), Triple-DES (3DES), and the Advanced Encryption Standard (AES). These
algorithms are block ciphers, meaning that they take a block of symbols of size n as
input and output a block of symbols of size n. DES was published in 1975, and was
developed as the U. S. standard for unclassified applications in 1977 [Stinson 2002].
DES uses a key size of 56 bits and a block size of 64 bits. The main criticism of DES
today is that the 56-bit key length is too short. With newer CPUs, the key space of 256
can be enumerated. Even with machines in 1998, a machine called the “DES Cracker”
could find a DES key in 56 hours.
Triple-DES was built to enlarge the DES key space without requiring users to switch
to a new encryption algorithm. 3DES operates by performing three DES operations on
the data with three keys: encryption with key one, decryption with key two, and encryption with key three. The three keys increase the key space to 2168 , but the strength
of 3DES is only twice as strong as DES, as demonstrated in the meet-in-the-middle
attack [Chaum and Evertse 1985]. Unfortunately, performing three cryptographic operations for every data access imposes a high performance penalty.
DES was replaced by the Advanced Encryption Standard (AES) algorithm in 2001.
AES has a block length of 128 bits and supports key lengths of 128, 192, and 256
bits. Among the five finalist algorithms to be chosen as AES (MARS, RC6, Rijndael,
Serpent, Twofish), Rijndael was chosen “because its combination of security, performance, efficiency, implementability, and flexibility was judged to be superior to the
other finalists” [Stinson 2002]. The National Security Agency (NSA) has reviewed and
concluded that all five finalists were secure enough for U.S. government nonclassified


2:5

Fig. 1. ECB-mode encryption and decryption.

data. In 2003, the U.S. government announced that AES, with all key lengths, was sufficient to protect classified information up to the level of secret. Only AES key lengths
of 192 or 256 bits can protect classified information at the top secret level [Ferguson et
al. 2001].
Key revocation is the act of invalidating an encryption key so that it may no longer
be used to decrypt data. This process often involves re-encrypting the data with a new
encryption key. Three types of key re-encryption modes exist: aggressive, lazy, and
periodic [Riedel 2002]. Aggressive re-encryption involves re-encrypting data directly
after key revocation. Lazy re-encryption involves re-encrypting selected data when it is
next modified or read. Periodic re-encryption involves changing keys and re-encrypting
data periodically. All cryptographic systems discussed in-depth in this survey force the
user to employ aggressive re-encryption when key revocation is possible. Some, due
to scope network-based cryptographic systems, listed briefly in Section 3.1.3, employ
lazy re-encryption (e.g., Cepheus [Fu 1999]; Plutus [Kallahalla et al. 2003]; and Secure
Network-Attached Disks (SNAD) [Miller et al. 2002]).
2.2. Traditional Modes of Operation

The operating mode of an encryption algorithm allows block ciphers to output messages of arbitrary length or turns block ciphers into self-synchronizing stream ciphers,
which generate a continuous key stream to produce ciphertext of arbitrary length. For
example, using AES alone, we may only input and output blocks of 128 bits each. Using AES with a mode of operation for a block cipher, we may input and output data of
any length.
An initialization vector (IV) is commonly used with many block ciphers; it is a small,
often random, but nonsecret value used to help introduce randomness into the block
cipher. The IV is often used at the beginning of the block cipher.
The most common modes of operation for block ciphers include electronic codebook
(ECB) mode, cipher-feedback (CFB) mode, cipher-block-chaining (CBC) mode, outputfeedback (OFB) mode, and counter (CTR) mode [Dworkin 2001].
2.2.1. Mode Examples and Specifications. ECB is the simplest mode of operation, and
does not use an IV (Figure 1). With a key Pi as the i th block of plaintext, and Ci as the
i th block of ciphertext, the encryption is performed as Ci = Ekey (Pi ), and decryption is
performed as Pi = Dkey (Ci ).
The CBC mode is slightly more complicated and uses an IV (Figure 2). Each block of
plaintext is first XORed with the previous block of ciphertext before being encrypted.
Therefore, each block of ciphertext relies on its previous block of ciphertext. Encryption
of the first block of plaintext is performed as C1 = Ekey (P1 ⊕ IV), where C1 is the 1st

2:6


Fig. 2. CBC-mode encryption and decryption.
Table I. Encryption and Decryption Specifications for Various Modes of Operations
Mode of Operation
ECB
CFB
CBC
OFB
CTR

Ci
Ci
Ci
Ci
Oi
Ci

Encryption
= Ekey (Pi )
= Ekey (Ci−1 ) ⊕ Pi , C0 = IV
= Ekey (Pi ⊕ Ci−1 ), C0 = IV
= Pi ⊕ Oi
= Ekey (Oi−1 ), O0 = IV
= Ekey (IV ⊕ CTRi ) ⊕ Pi

Decryption
Pi = Dkey (Ci )
Pi = Ekey (Ci−1 ) ⊕ Ci , C0 = IV
Pi = Dkey (Ci ) ⊕ Ci−1 , C0 = IV
Pi = Ci ⊕ Oi
Oi = Ekey (Oi−1 ), O0 = IV
Pi = Ekey (IV ⊕ CTRi ) ⊕ Ci

block of ciphertext; IV is the random, nonsecret initialization vector; and P1 is the 1st
block of plaintext. Subsequent blocks of plaintext are encrypted as Ci = Ekey (Pi ⊕Ci−1 ).
In the same manner, the first block of ciphertext is decrypted as P1 = Dkey (C1 )⊕ IV, and
the subsequent blocks of ciphertext are decrypted as Pi = Dkey (Ci ) ⊕ Ci−1 .
Other block ciphers can be specified in a similar way. Table I lists the various modes
of operation, along with the corresponding encryption and decryption specifications.
2.2.2. Padding. Certain modes of operation, such as ECB, CFB, and CBC, require the
plaintext to be partitioned into blocks of a static length. Therefore, whenever the last
block of the plaintext is partially filled, it is “padded” or appended with predefined bits
to form a full block. The padding is encrypted with the data, and after decryption, is
automatically removed from the data.
NIST [Dworkin 2001] recommends appending a single ‘1’ bit to the last block in
need of padding and then adding as few ‘0’ bits as possible (perhaps none) until the
block is full. The padding can be removed unambiguously by the decryption function
if either (1) every message is padded, even messages that have a full last block; or (2)
the decryption function knows the length of the data.
Another commonly used padding scheme is to pad the data with a string of one to
eight bytes to make the last block eight bytes [RSA Laboratories 1999, 1993; Housley
2004]. The value of each padding byte will be the number of bytes added to the data.
For example, if two bytes must be added to the last block to make the last block eight
bytes long, the padding bytes will consist of two 2s.
NIST has recently developed a proposal to extend CBC mode via ciphertext-stealing
[NIST 2007], which incorporates a padding mechanism which, unlike the two methods
just described, does not increase the size of the resulting ciphertext. The mode uses
the term “ciphertext-stealing” because padding bits are taken from the penultimate
ciphertext block whenever padding is necessary.
For example, consider an example of CBC mode extended with ciphertext-stealing
during encryption. Figure 3 illustrates what happens when the last plaintext block,
Pn, fills only a partial block, where 000.. is a temporary pad, Cn−1 equals the leftmost


2:7

Fig. 3. Encryption example using CBC and ciphertext stealing.
Table II. Random-Access Performance for Various Modes of Operation
Mode of Operation
ECB/CTR
CFB/CBC

OFB

Encryption Performance
Decryption Performance
Good: ECB and CTR do not depend on previous blocks. Multiple
blocks can be encrypted and decrypted in parallel.
Poor: Generating a CFB and CBC
Good: CFB and CBC decryption of one
ciphertext requires the previous
block requires only one previous
ciphertext block as input. In the case
ciphertext block as input. Multiple
of updates, CFB and CBC require
blocks can be decrypted in parallel.
re-encrypting the remainder of a file,
since all subsequent ciphertext
blocks depend on the current
ciphertext block. Thus, encryption is
not parallelizable.
Medium: The key stream (or encryption mask) can be pre-computed
independently of the ciphertext. This pre-computation can speed up random
file access, as the encryption and decryption of a block does not depend on
previous blocks. Once the key stream is computed, encryption and decryption
may be performed in parallel.

bits in the penultimate ciphertext block (with a size equal to Pn), and Cn−1 equals
the rightmost bits in the penultimate ciphertext block. Cn−1 is omitted from the final
ciphertext as it can be regenerated in the decryption process.
2.2.3. Performance Factors. Even when employing a fast block cipher (such as AES),
certain modes of operation may not interact well with the file system usage patterns.
Although most file accesses are sequential, where the accessed file locations are consecutive, a significant fraction of file references are not, and are loosely defined as random file accesses. Therefore, some modes may require decrypting an entire file before
a read can be performed at the very end of the file. Conversely, some modes may require re-encrypting an entire file after a write is performed at the beginning of the file.
To reduce random access time, certain solutions divide files into extents (contiguous
segments of blocks) for a finer encryption granularity. The main tradeoff is introducing
complexity in key and IV management.
Knowing the performance characteristics of various modes of operation helps the
storage designers understand how mode choices affect confidential storage and deletion later in this survey. Table II compares encryption and decryption performance of
the discussed modes of operation. Since the random file access pattern is more general compared to the sequential one, the table only characterizes the performance of
various operation modes under random accesses.

2:8

Table III. Recoverability and Security Characteristics for Various Modes of Operation
Mode of Operation
Error Resiliency
Overall Security
ECB
Good: Bit errors in a ciphertext Poor: Two plaintext blocks with
block will only corrupt the
the same content yield the
corresponding plaintext block.
same encrypted ciphertext.
Ciphertext blocks may also be
reordered by an adversary.
CFB
Good: A bit error in a
Good: As long as IVs are
ciphertext block affects only
randomly generated and not
two plaintext blocks—a
reused.
one-bit change in the
corresponding plaintext block,
and the corruption of the
following plaintext block.
Remaining plaintext blocks
are decrypted normally.
CBC
Good: A bit error in the
ciphertext will corrupt the
corresponding plaintext block,
reused.
and the corresponding bit in
the next plaintext block will
be flipped. Later plaintext
blocks are decrypted normally.
OFB/CTR
Good: Bit errors in a single
ciphertext block will corrupt
only the corresponding
reused.
plaintext block.

2.2.4. Security Caveats. Other considerations when choosing a mode of operation are
error resiliency and the overall security of the mode (Table III). Error resiliency is
concerned with the propagation of damage when errors occur in a ciphertext block
(e.g., the result of a damaged sector on a hard disk). Overall security is concerned with
the weaknesses for various modes of operation during attacks that attempt to recover
the key, plaintext, or both.
2.2.5. A Side Note about CTR Mode and Key Management. While CTR mode may seem to
excel in both security and performance, most encryption systems today use CBC mode.
The reason lies in the way the counters are generated and the keys are managed.
Each block encrypted or decrypted in CTR mode does not depend on previous blocks.
Instead, counters are generated and fed into the encryption algorithm along with the
file’s key. A counter can be just the block index within a file.
To prevent direct comparisons of ciphered blocks from different files that share a
key and the same counter indexing method, a per-file IV is often XORed with each
counter. Unfortunately, the counter XORed with IV is not guaranteed to be unique
(e.g., files with the same initial file header), giving attackers multiple counter-IV pairs
and ciphertext using the same key for analysis. Using a per-file unique key could be
one solution, but also introduces the hard problem of key management in encryption
systems, which is a large area of research beyond the scope of this survey.
2.3. SISWG Cipher Modes

This section summarizes modes of confidential encryption designed for hard disk encryption and data at rest. The IEEE Security in Storage Working Group (SISWG)
is currently looking to standardize narrow-block and wide-block encryption modes. A
narrow-block cipher mode operates on the block size of the underlying cipher (e.g., 16
bytes for AES). A wide-block cipher mode operates on more cipher blocks equal to the


2:9

Fig. 4. Standard block cipher contrasted with a tweakable block cipher.

size of an underlying disk sector (generally 512 bytes and above). At the time of this
writing, SISWG has already gotten the narrow-block encryption specification P16192007 [SISWG 2008a] approved through the IEEE, and is actively submitting the specification to NIST for it to become an approved mode of operation [NIST 2008]. SISWG
is also working on a draft for wide-block encryption P1619.2/D7 [SISWG 2008b] titled
“Draft Standard for Wide-Block Encryption for Shared Storage Media”.
Both narrow-block encryption and wide-block encryption are tweakable block
ciphers. A tweakable block cipher has similar inputs (plaintext and key) and outputs
(ciphertext) as the standard encryption algorithms discussed in Section 2.1, with the
addition of a nonsecret input called a “tweak” [Liskov et al. 2002]. This tweak serves a
similar randomization purpose as a nonce or IV in a mode of operation, except at the
block cipher level; see Figure 4. Using traditional algorithms, the ciphertext output
will not vary during multiple encryption operations using the same plaintext and key
as inputs. With a tweakable block cipher, however, using a different tweak value during multiple encryptions with the same key and plaintext values will create different
ciphertexts. The authors specify certain goals for tweakable block ciphers, namely, that
(1) any tweakable block ciphers designed should be just as efficient as non-weakable
block ciphers; and (2) a tweakable block cipher should also be secure, even if an adversary has control of the tweak value.
2.3.1. Narrow-Block Encryption. The published P1619-2007 draft standard discusses
the use of XTS-AES, which is a tweakable block cipher with ciphertext stealing. XTSAES can be applied to data units of 128 bits or more and uses AES as a subroutine
[Bohman 2007]. Data units are divided into blocks of 128 bits; but the last part of the
data unit may be shorter than that. With a key P as the 128-bit block of plaintext,
and C as the resulting 128-bit block of ciphertext for block i, t as the 128-bit tweak
value, and j as the sequential number of the 128-bit block inside the data unit, the
encryption and decryption operations are

Ci = Ekey (Pi , ti , ji ),
Pi = Dkey (Ci , ti , ji ).
Specifically, the XTS-AES encryption and decryption operations involve multiplication
of the 128-bit result of AES encryption or decryption by the jth power of a primitive
element of GF(2128 ), where GF stands for Galois Field [Menezes et al. 1997]. Unlike
other modes of operation, such as CBC, CFB, and OFB, the XTS-AES mode can operate
in parallel, which may increase performance.

2:10


2.3.2. Wide-Block Encryption. The unpublished P1619.2/D7 draft currently specifies
Encryption with Associated Data (EAD) methods, such as the EME2-AES transform
and the XCB-AES transform. An EAD method consists of an encryption operation that
accepts three inputs: a secret key, a plaintext, and the associated data. Associated
data is data that characterizes the plaintext in some fine-grained way and does not
need to be encrypted. In other words, the associated data is used as a tweak value, in
that it serves to randomize the result of multiple encryptions with the same keys and
plaintexts. However, unlike normal tweakable encryption in which the tweak value
has a fixed size, the associated data may have a variable size. The encryption operation returns a single ciphertext value of the same size as the original plaintext. An
EAD method also consists of a decryption operation that accepts three inputs: a secret
key, a ciphertext, and the associated data. The decryption operation returns a single
plaintext value of the same size as the original ciphertext. Like XTS-AES, wide-block
encryption operations can be performed in parallel. With a key Pi as the i th block of
plaintext, and Ci as the i th block of ciphertext, and where C0 and P0 are initial values,
the encryption and decryption operations of an EAD method are

Ci = Ekey (ADi , Pi ),
Pi = Dkey (ADi , Ci ).
2.4. NIST-Proposed Encryption Modes of Operation

This section summarizes other modes of encryption currently being proposed by the
National Institute of Standards and Technology (NIST). These modes primarily deal
with protecting data confidentiality. Other modes, such as authenticated encryption,
are beyond the scope of this survey.
2.4.1. 2DEM. The 2D-Encryption Mode (2DEM) encrypts and decrypts binary data
in block structures with rows and columns [Belal and Abdel-Gawad 2001]. If DES is
chosen as the underlying encryption algorithm, the number of bytes n in each row and
column must be a multiple of 8; for AES, 16. 2DEM encrypts data by first performing
ECB encryption operations on each row of the block structure (traveling from top to
bottom) and producing an intermediary result. It then performs additional ECB encryption operations on each column of the intermediary result (traveling from left to
right). The resulting output is the ciphtertext. Decryption is performed on the ciphertext by reversing the encryption operation.
2DEM seems naturally suited for encrypting data of a 2D nature (such as images
and tables), but it can be applied to other forms of 1D data by simulating artificial
rows and columns. Since each block of data in the row-encrypting phase or the column
encrypting phase does not rely on any other block of data, these operations can be
parallelized. Bit errors are limited to the encryption/decryption block.
2.4.2. ABC. Traditional modes of operations have good error resiliency (Table III),
in that bit errors in the nth ciphertext block will affect at most up to the (n + 1)th
plaintext block. The Accumulated Block Chaining (ABC) mode, on the other hand,
was designed to have infinite error propagation [Knudsen 2000]. As Knudsen states,
error propagation modes are best suited “for situations where errors in transmission
are either unlikely to happen or taken care of by noncryptographic means like errorcorrecting codes, and/or situations where an erroneous data transmission is dealt with
by a retransmission.”
With a key Pi as the i th block of plaintext, and Ci as the i th block of ciphertext, h as
a mapping from n to n bits, and where H0 and C0 are initial values, the encryption


2:11

operations is
Hi = Pi ⊕ h(Hi−1 ),
Ci = Ekey (Hi ⊕ Ci−1 ) ⊕ Hi−1 .
The decryption operation is
Hi = Dkey (Ci ⊕ Hi−1 ) ⊕ Ci−1 ,
Pi = Hi ⊕ h(Hi−1 ).
h may be chosen as a hash function, but Knudsen also argues that choosing h(X) = X
or h(X) = X 1 (one-bit rotation) may be sufficient for most applications.
2.4.3. IGE. The Infinite Grappling Extension (IGE) is a special case of ABC where
h(X) = 0. The IGE mode was first proposed by Campbell [1978] and further analyzed
by Gligor and Donescu [2000]. IGE mode was conceived to prevent spoofing attacks in
which an attacker would intercept, modify, and retransmit a cipher in such a way that
the deception is not detected, even when the attacker does not know the secret key.
Any change made to the ciphertext in IGE mode will garble all remaining plaintext
during decryption. Campbell suggests placing an expected pattern at the end of the
message. If the recipient finds the expected pattern after decryption at the end of the
message, the recipient can be assured that the message has not been tampered with.
With a key Pi as the ith block of plaintext, and Ci as the i th block of ciphertext, and
where C0 and P0 are initial values, the encryption operation is

Ci = Ekey (Ci−1 ⊕ Pi ) ⊕ Pi−1 .
The decryption operation is
Pi = Dkey (Pi−1 ⊕ Ci ) ⊕ Ci−1 .
2.4.4. FFSEM. Some traditional modes of operation, such as ECB, CBC, and CTR,
will only accept as input fixed-size plaintext blocks and output fixed-size encryption
blocks. The Feistel Finite Set Encryption Mode (FFSEM) [Spies 2008] was designed
to encrypt arbitrarily-sized data using two components: cycle following and the Feistel
method. Cycle following uses a q-bit block cipher to encrypt and decrypt sets of size n
where n < 2q . The Feistel method uses a Luby–Rackoff construction [Luby and Rackoff
1988] to turn a fixed-width block cipher into an arbitrary-width block cipher.
An advantage is that FFSEM does not encrypt multiple blocks of data, but is designed to be used where data expansion is not acceptable. Some disadvantages are
that FFSEM needs multiple invocations of the block cipher for a single encryption,
and different data items can take different amounts of time to encrypt or decrypt due
to FFSEM’s cycling construction.
3. CONFIDENTIAL STORAGE

Many regulations and acts address the storage of sensitive data. The Gramm–Leach
Bliley Act [Federal Trade Commission 1999] requires financial institutions to have
a security plan for protecting the confidentiality and integrity of personal consumer
data. The Federal Information Security Management Act addresses the minimum security requirements for information and systems within the federal government and

2:12


Fig. 5. Unix and Windows storage data paths.

affiliated parties. The Health Insurance Portability and Accountability Act mandates
provisions to address the confidentiality and security of sensitive health data. These
acts and regulations and the threat of possible storage media theft motivates the need
for methods of secure data storage. For this survey, we focus on a major component of
secure data storage, namely, protecting data confidentiality.
The concept of confidential storage of data may be easy to understand, yet difficult
to implement. Achieving confidentiality means storing data in a way that can be read
or deciphered only by authorized persons. No unauthorized persons should be able to
read or otherwise obtain meaningful information from this data, even with physical
access to the storage media (e.g., a stolen laptop). To limit the scope of this article, we
do not cover cases where an attacker can infer information via indirect channels. For
example, we could infer that valuable information exists on a stolen laptop if we find
ciphertext on the laptop’s hard drive.
Confidential storage methods are difficult to implement for reasons including complexity of method setup; difficulty of conversion of prior methods to new secure methods; training, overhead, and latency in everyday tasks (e.g., reading and writing to
files); key management and password management.
As a brief storage background, Figure 5 shows the storage data paths for popular
Unix-based and Windows operating systems. For both platforms, applications reside
in user space. When a Unix application makes a call to a file system, the call crosses
the kernel boundary and is handled by the Virtual File System (VFS) layer [Kleiman
1986]. VFS provides functions commonly used in various file systems to ease individual
file system implementations, and allows different file systems to coexist, including
local file systems such as ext3 and network file systems such as NFS. Local file systems
then proceed to read and write to the block layer, which provides a unified API to access
block-layer devices.
When a Windows application makes a file system call, that call gets passed to the
I/O Manager. The I/O Manager translates application file system calls into I/O request
packets, which it then translates into device-specific calls. The File System Drivers
are high-level drivers such as FAT and NTFS. These drivers rely on the Storage Device


2:13

Drivers, which are lower-level drivers that directly access the storage media. Note that
both UNIX and Windows storage data paths share almost one-to-one mapping in terms
of their internal structures. Thus, a confidential storage solution designed for one can
be generalized to both platforms.
The following sections discuss specific aspects of confidential storage. Section 3.1
demonstrates different types of software-based confidential storage techniques, and
Section 3.2 delves into hardware-based confidential storage.
3.1. Software-Based Confidential Storage

Software-based solutions to confidential storage require no specialized hardware and
are widely available today. These solutions range from encryption programs to cryptographic file systems and block-based encryption systems. Encryption programs reside
in user space and employ cryptographic operations at a file granularity. Cryptographic
file systems include file systems that are either tailored for encryption or extended for
cryptographic functionality. Block-based solutions are tailored to encrypt and decrypt
large chunks of data at the granularity of a partition or disk. Each solution has its
strengths and limitations with regard to the level of confidentiality, ease-of-use, performance, and the flexibility to set policies. Riedel et al. [2002] and Wright et al. [2003]
include further references and discussion on performance and evaluation of softwarebased confidential storage.
3.1.1. Encryption Programs. Software encryption programs come in two flavors: generalized encryption programs and built-in encryption mechanisms in applications. Generalized encryption programs can encrypt and decrypt files using a variety of ciphers
and encryption modes; several examples are mcrypt, openssl, and gpg. mcrypt [Smith
2008] is a simple command-line program intended to replace the old Unix crypt program; openssl [Young and Hudson 2008] is an open-source toolkit that implements
the Secure Socket Layer and Transport Layer Security protocols as well as a generalpurpose cryptography library. Through use of the library, we may encrypt and decrypt
files through the command line. The GNU Privacy Guard, or GnuPG [Koch 2008], implements the OpenPGP standard, which features a versatile key management system
and allows the user to encrypt and sign data and communications using public and
private keys.
Many applications also include cryptographic options to protect the confidentiality
of files. Examples include the text editor vim [Moolenaar 2008] and Microsoft Office
products such as Word and Excel [Microsoft Corporation 2003]. These applications either derive the key from the user’s system information (such as a password) or prompt
for a key or passphrase at the beginning of the session.
While easiest to deploy, application-level solutions have their limitations in regards
to level of confidentiality. For example, temporary plaintext files may be created based
on the files that are being accessed. Therefore, if temporary files are not deleted, an
attacker can simply find them using dead forensic methods. If the temporary files have
been deleted but not securely erased (see Section 4), an attacker may use forensic tools
to recover the data. File names and metadata are also not encrypted, which may give
an attacker information about the type of encrypted data (e.g., size of file, date last
modified).
Encryption programs can vary widely in terms of the flexibility of security policies.
For example, generalized encryption programs often offer a wide array of encryption
algorithms and modes of operation. They can be used on general files, and possibly used as a filter in a command pipeline. On the other hand, application-specific
encryption programs tend to offer few ways to perform encryption, to limited files

2:14


Fig. 6. FUSE data path.

types, with limited compatibility with other applications. These characteristics could
limit application-specific encryption in terms of the flexibility of changing security
policies.
In terms of the user model, generalized encryption programs usually demand much
user participation and may not be easy to use. For example, a user must invoke these
programs along with the necessary key/password/passphrase every time encryption or
decryption takes place. A simple mistake, such as using an incorrect passphrase to
encrypt a file or accidental deletion of the private key, may render a file useless and
irrecoverable. While this is true of any method using encryption, the chances for error
tend to be higher when users can manipulate the encryption key directly. Conversely,
application-specific solutions typically prompt for a key/password/passphrase once and
tend to perform encryption in a “behind-the-scenes” manner.
In terms of performance, both generalized and application-specific solutions at the
user level are slower than other solutions because they do not take full advantage of
the VFS-layer caching. To be specific, since encryption occurs at the user level, VFS has
to cache encrypted data. Thus, unless a user space application caches plaintext itself,
it needs to perform decryption and encryption functions on every read and write.
3.1.2. User-Space File Systems. Many user-space file systems take advantage of the
Filesystem in Userspace (FUSE) module [Szeredi 2008], which is a Unix kernel module that allows a virtual file system to be built inside a user-space program without
having to write any kernel-level code. FUSE intercepts VFS calls and directs them to
a user-space file system with added security features before forwarding requests to an
underlying legacy file system in the kernel space (Figure 6).
Two examples of FUSE-based secure storage file systems include EncFS [Gough
2008] and CryptoFS [Hohmann 2008]. Both systems are similar in (1) storing encrypted files and file names in encrypted directories; (2) requiring users to mount
encrypted directories onto a special mount point with the correct key to see decrypted


2:15

files and file names; (3) prompting users for a password to generate the encryption key;
(4) typically supporting common encryption algorithms such as AES, DES, Blowfish,
Twofish, based on what is available in external encryption libraries; and (5) encrypting
files on a per-block basis. In cryptographic file systems, encrypted directories are similar to normal directories. They contain encrypted files and other encrypted directories,
but generally only directory names are encrypted. Thus, it is possible to traverse an
encrypted directory without mounting the encrypted directory by issuing encrypted
path names directly. File system blocks are encrypted with CBC mode in EncFS, and
CryptoFS uses CFB mode within extents. Neither file system document allows its
mode of operation to be a configurable option. In [Oprea 2007], EncFS is extended
to support the wide-block tweakable cipher CMC [Halevi and Rogaway 2003], as well
as other integrity constructions for encrypted storage that are beyond the scope of this
survey.
These file systems can provide strong confidentiality by employing good encryption
mechanisms and placing temporary files in encrypted directories. One drawback is
that user-space file systems still reveal the directory structure and file metadata information, which may help an attacker using dead forensic techniques gather information
about the type of data that is encrypted. User-space file systems do allow for some security policy flexibility, in that they generally allow for different encryption algorithms,
but neither eases key revocation nor changing the mode of operation.
User-space file systems tend to be easier to use than user encryption applications
because encryption and decryption happen transparently. Still, users must “mount”
encrypted directories to a special mount point to manipulate them. These file systems
tend to have higher performance overhead, in that requests may need to travel across
the kernel and user space multiple times. Since these file systems also rely on an
underlying file system, the performance overhead of using user-space file systems is in
addition to the overhead of using the underlying file system.
3.1.3. Local File Systems that leverage NFS Mechanisms. Some local file systems leverage
the existing design of Network File System (NFS) [Sandberg et al. 1985] to ease development, since NFS redirects the operating-system-level traffic to the user space. Two
examples that fit our nondistributed, single-user computing environment threat model
are the Cryptographic File System (CFS) and the Transparent Cryptographic File System (TCFS). Both systems are described for use on a local machine, even though they
could be mounted over a remote machine. However, both systems are somewhat dated.
As NFS was designed more for network use between multiple machines, mechanisms
such as tunneling NFS through SSH or SSL, or employing NFS over IPsec [Kent and
Atkinson 1998] may be employed to provide confidentiality over an external network.
However, these techniques will only protect data “on the wire,” and not the data at
rest. For examples of network-based file systems beyond the assumed local computing
environment of this article, please see the Self-Certifying File System (SFS) [Mazi` res
e
et al. 1999]; Cepheus [Fu 1999]; Secure Network-Attached Disks (SNAD) [Miller et al.
2002]; Plutus [Kallahalla et al. 2003]; SiRiUS [Goh et al. 2003]; FARSITE [Adya et al.
2002]; and the distributed IBM StorageTank file system SAN.FS [Pletka and Cachin
2007]. CFS [Blaze 1993] is implemented in the user space, which communicates with
the Unix kernel via NFS. Specifically, it uses an unmodified NFS client to talk with
a modified NFS server over the loopback network interface. Figure 7 shows the CFS
data path.
CFS allows users to “attach” cryptographic keys to directories to make the content within the directory available to the user until the user “detaches” them. While
directories are attached, files, filenames, and file paths inside those directories are

2:16


Fig. 7. Data flow in CFS.

transparently encrypted and decrypted. Files are encrypted using DES in a hybrid
ECB+OFB mode by default, but multiple encryption algorithms are supported. In the
standard encryption mode, no IV is used, and files are subject to an identical block
analysis attack. In high security mode, the IV is stored in the group ID field of each
file’s i-node data structure. This mode improves security but precludes different group
ownerships of files in the same directory, since the effective group ID is now inherited
by the root encrypted directory. Users are responsible for remembering keys, and keys
are not stored on disk.
TCFS [Cattaneo et al. 2001] uses a modified NFS client (in kernel mode) and an unmodified NFS server, which makes it possible (but not necessary) to work with remote
machines. TCFS encrypts files by employing a user-specified block encryption module
(e.g., AES) in CBC mode, and also includes authentication tags per file to insure data
integrity. Only file data and file names are encrypted; directory structures and other
metadata are not encrypted. Instead of requiring passphrases, TCFS uses the UNIX
authentication system. TCFS implements a threshold secret-sharing scheme [Shamir
1979] for reconstructing a group key if a member leaves the group. TCFS is available
under Linux with a 2.2.17 kernel or earlier and stores its keys on disk, which may not
be safe.
NFS-based local file systems can employ multiple confidential encryption mechanisms, confidential key solutions in which the decryption key is not stored on disk,
and confidential temporary file solutions in which temporary files can be placed in
encrypted directories. On the other hand, NFS-based local file systems reveal the directory structure and file metadata (except for the file name), and these file systems
are subject to the security vulnerabilities of the underlying network protocols (e.g., an
attack on NFS).
NFS-based local file systems enjoy certain flexibilities in the context of implementing
and changing security policy settings, including the ability to operate on top of any


2:17

Fig. 8. File system call path using Cryptfs.

file system and portability across different systems through the use of the network
stack and various network interfaces. These characteristics may aid existing security
policies by allowing the use of specific underlying file systems. Some limitations of
NFS-based local file systems include not allowing easy key revocation and changes in
the mode of operation.
NFS-based local file systems tend to be easier to use than user-space encryption
programs because encryption and decryption happen transparently to the user. On the
other hand, users must realize that they need to “mount” encrypted directories to a
special mount point in order to manipulate them. Thus, the encryption is not entirely
transparent.
Due to their need to cross the kernel boundary many times, NFS-based local file
systems have perhaps the worst performance numbers compared to other forms of
solutions in this survey.
3.1.4. Stackable File Systems. Stackable file systems use a stackable architecture to
extend functionality (such as cryptographic functions) by intercepting system calls and
routing them to an inserted file system layer. In other words, these file systems run
inside the kernel and can operate on top of any other file system without requiring
other user-level processes or daemons to run. Some examples of stackable file systems
include Cryptfs, NCryptfs, and eCryptfs.
Cryptfs and NCryptfs are applications of FiST, a File System Translator language
[Zadok and Nieh 2000]. FiST allows developers to describe stackable file systems at a
high level and generates kernel file system modules for various platforms (e.g., Solaris,
Linux, and FreeBSD).
Cryptfs is “a stackable v-node level encryption file system [Zadok et al. 1998].” The
term v-node refers to a data structure used to represent a virtual i-node in the virtual file system layer. Cryptfs uses the Blowfish encryption algorithm with 128-bit
keys in CBC mode. A tool prompts users for passphrases, which then form encryption keys. File names and directories are also encrypted. The key is not stored on disk.
Figure 8 demonstrates how Cryptfs is layered between the user and the underlying file
system.

2:18


NCryptfs is the successor to Cryptfs and improves on its base design in many ways
[Wright 2003]. NCryptfs supports multiple users, keys, authentication modes, and any
encryption algorithm that can encrypt arbitrary length buffers into a buffer of the
same size (e.g., Blowfish or AES) in CFB mode. It extends key management to support ad-hoc groups, per-process and per-session keys, and key timeouts. NCryptfs uses
“long-lived” keys to encrypt data to avoid the penalty of re-encrypting data when a key
is changed. Once the key is entered (through a passphrase), it is kept in core memory
by NCryptfs and is never revealed to other users. The key is not stored on disk.
One example of a nonFiST stackable file system is eCryptfs, which is a kernel-native
stacked cryptographic file system for Linux [Halcrow 2007]. Similar to the FiST stackable file systems, eCryptfs intercepts calls to the existing mounted file system. While
intercepting these calls, eCryptfs encrypts and decrypts file data. eCryptfs operates at
a lower granularity by encrypting and decrypting individual data extents in each file
using a uniquely generated File Encryption Key (FEK). That key is encrypted with the
user-supplied File Encryption Key Encryption Key (FEKEK), and the result is stored
inside the file’s metadata. Users supply the FEKEK either by passphrase or a public
key module.
eCryptfs supports multiple encryption algorithms and supports the CBC mode of
operation. Each data extent also has a unique IV associated with it. When data is
written to an extent, its corresponding IV is changed before the extent is re-encrypted.
Additional eCryptfs i-nodes are kept, which are mapped to the i-nodes in the underlying file system. eCryptfs i-nodes contain cryptographic contexts, including:
—the session key for the target file;
—the size of the extents for the file; and
—a flag specifying whether the file is encrypted.
Stackable file systems can employ confidential encryption mechanisms. The primary
file encryption key does not have to be stored on disk, and temporary files from applications can be placed in encrypted directories. In terms of limitations, existing stackable
file systems reveal the directory structure, as well as file metadata, and often do not
allow easy key revocation or a change in mode of operation.
Stackable file systems tend to be easier to use than user-level encryption programs
due to the transparency of cryptography to the user. Users must “mount” encrypted
directories to a special mount point. Since stackable file systems insert functionality
through a layer of indirection, the overhead can be higher than other methods (e.g., a
single-layer file system tailored to offer confidentiality). On the other hand, since these
file systems run in kernel space, they perform better than file systems that either run
in user space or require crossing the kernel boundary multiple times.
3.1.5. Disk-Based File Systems. Disk-based file systems operate at a lower level of abstraction than stackable file systems, software-based encryption programs, or NFSbased local file systems. A disk-based file system has full control over all of its directory
and file metadata and operations. One example is Microsoft’s Encryption File System
(EFS).
EFS extends the journaling NTFS file system and utilizes Windows’ authentication
methods as well as access control lists [Microsoft Corporation 2002, 2008]. EFS is supported by operating systems based on the Microsoft NT kernel, such as Windows 2000,
XP, and Vista. Figure 9 demonstrates how EFS extends NTFS inside and outside of
kernel space.
EFS utilizes both public key and private key encryption techniques. When a file
is encrypted for the first time, a unique symmetric per-file encryption key (FEK) is


2:19

Fig. 9. EFS data path.

Fig. 10. EFS file structure.

created and is used to encrypt the file. The FEK is embedded in the target file and
is then encrypted with the user’s public key. FEK can also be optionally encrypted
with the private key of a user designated as the “recovery agent” Figure 10 shows an
example of encrypted file structure. Decryption involves decrypting the FEK with the
user’s private key, which is stored in the user’s Windows profile. One weakness of this
method is that an attacker can recover the user’s or recovery agent’s private key if the
attacker can gain access to the user’s account [Microsoft Corporation 2007].
EFS encrypts files and FEKs using the DESX encryption algorithm, a variant of DES
that increases brute-force attack complexity on the key [Kilian and Rogaway 1996].

2:20


Encrypted folders are set with an encryption attribute, and by default all files and
folders inside an encrypted folder are also encrypted.
A disk-based file system, when configured correctly, can offer strong encryption
mechanisms. In addition, since a disk-based file system has control over its primitives, such as metadata, it can associate a unique key (and possibly even an IV) to
each file. Different encryption mechanisms are allowed for security policy creation and
implementation, and multiple users may be specified to be able to decrypt certain files.
However, a disk-based file system cannot easily change the mode of operation, and
often times some file structures and metadata are still revealed (such as with EFS).
The user experience is similar to NFS-based local and stackable file systems, in that
users must know enough of the encryption to place files into confidential directories
(or designate directories as confidential directories). Disk-based file systems remove
the level of indirection that other cryptographic file systems employ and perform encryption and decryption in the kernel, which utilizes lower level buffers and speed up
performance.
3.1.6. Block-Based Encryption Systems. Block-based encryption systems work at a
lower layer of abstraction than file systems. In other words, these systems work transparently below file systems to encrypt data at the disk-block level. Examples of blockbased encryption systems include dm-crypt, BestCrypt, the CryptoGraphic Disk driver,
the Encrypted Volume and File System, and Microsoft BitLocker Drive Encryption.
dm-crypt [Peters 2004] is a replacement for the Linux cryptoloop system that works
by using the Linux device mapper, an infrastructure introduced in Linux 2.6 to provide
a generic way to create virtual layers of block devices on top of real block devices that
interact with hardware directly. dm-crypt uses the Linux CryptoAPI and supports encrypting block devices such as disks, partitions, logical volumes, and RAID volumes.
If we write random blocks to a block device using dm-crypt, an attacker will not be
able to discern the locations of encrypted files and the amount of free space left on
the device. dm-crypt supports encryption algorithms and modes of operation present
in the Linux CryptoAPI, which include AES, DES, Serpent, Twofish, Blowfish, ECB
mode, and CBC mode. The IV is based on the sector number. An encrypted block device
will wait to be mounted until a user passphrase is given. Similar to cryptoloop and
dm-crypt, BestCrypt [Jetico, Inc. 2008] is a commercial example of a block-based encryption system. Other examples include the CryptoGraphic disk driver [Dowdeswell
and Ioannidis 2003] for NetBSD and the Encrypted Volume and File System (EVFS)
[Hewlett-Packard 2007] for HP-UX.
Microsoft BitLocker Drive Encryption provides encryption for hard disk volumes and
is available with Vista Enterprise, Vista Ultimate, and the upcoming Windows Server
operating systems [Microsoft Corporation 2006]. BitLocker drive encryption has two
goals:

(1) to encrypt the entire Windows operating system volume (and additional volumes
in the future);and
(2) to verify the integrity of the boot process using a Trusted Platform Module (TPM).
We are only concerned with the first goal, as boot process integrity is beyond our scope.
BitLocker encrypts the specified volume sector-by-sector using AES in CBC mode with
a diffuser called Elephant [Ferguson 2006]. The diffuser is stated as necessary due
to a weakness in CBC mode, which allows an attacker to flip an i th bit in the next
block’s plaintext by flipping the i th bit in the current block’s ciphertext at the risk of
randomizing the current block’s plaintext (Table III). This diffuser runs a series of XOR
and rotations on 32-bit words in a block to cause one flipped bit to cause many more


2:21

Fig. 11. BitLocker encryption of a block.

random bit flips in the same block. In fact, using this diffuser in a block with at least
128 words will cause full diffusion to occur within the active word with just one bit flip
in about one third of a cycle. This means that all bits in the next block’s plaintext will
be randomly affected by the bit flip, and the attacker cannot selectively control any
aspect of the plaintext if any ciphertext is modified. Figure 11 demonstrates Bitlocker
encryption of one block, where the drive sector key is the IV and the block size can
be any power of two within the range of 512-8,192 bytes, or 4,096-65,536 bits. Two
separate diffusers ( A and B) are used in Elephant. This is because one diffuser has a
property of diffusing well (quickly) during decryption, and the other has a property of
diffusing well during encryption.
A BitLocker encryption key can be retrieved either automatically from the TPM chip
without the user entering a PIN or from a USB device. If the key is retrieved from the
TPM chip, the device will boot up into the operating system.
Block-based encryption systems have many advantages. Once they are set up, they
are transparent to the user (except when the user must enter the passphrase) and to
the file system. Swap and temporary files are automatically encrypted if the entire
block device or volume is encrypted, which is a large improvement in confidentiality
over other methods.
Block-based encryption systems can employ confidential encryption algorithms and
modes of encryption. Encryption must be performed over the entire disk on all types
of data, which makes it very hard for an attacker using dead forensic techniques to
discern file structures from data. An attacker may not even be able to discern partition size if the entire disk is first filled with random bits. Keys, encryption algorithms,
and mode of operation generally cannot be changed. This characteristic makes it difficult to control the extra encryption granularity (e.g., everything must be encrypted),
which may be a challenging obstacle for certain security policies. Performance is rated
similarly to disk-based file systems, in that encryption and decryption operations take
place in kernel-space and take advantage of lower-level disk buffering.
3.2. Hardware-Based Confidential Storage

Hardware-based confidential storage mechanisms differ from software ones in that the
cryptographic functionality is either hard-coded into the hardware or into an external
specialty device. This method is more rigid in that a user cannot change authentication

2:22


mechanisms (such as an encryption key or smart card) or add extra functionality to the
storage mechanism, yet it is often much faster than any software-based solution.
Some present-day examples of hardware-based secure storage include secure flash
drives, extended cryptographic instruction sets used on specialty CPUs, and hard disk
enclosures, as well as PCI/PCMCIA extension cards that act as cryptographic interceptors between the hard disk and the operating system. We discuss these mechanisms,
as well as their strengths and limitations, in the following sections.
3.2.1. Secure Flash Drives. The secure flash drive is a relatively new phenomenon
on the market today, apparently in response to data and identity theft. Some secure
flash drives provide only software encryption using block-based encryption methods,
as mentioned. Other flash drives protect data through cryptographic mechanisms provided on the flash drive itself. This means that encryption and decryption are transparent to the host operating system, yet a mechanism must be provided (usually via
operating system driver or application) to allow the user to input an encryption key
(or a password that unlocks the encryption key or keys). Two example products are
Ironkey and the Kingston Data Traveler Secure.
Ironkey [2007] uses hardware-based AES encryption in CBC mode with 128-bit randomly generated keys. These keys are generated inside the flash drive in a CryptoChip
and are unlocked by a user password. Ten wrong attempts will trigger a “self-destruct”
of the encryption keys. Two volumes become available when the Ironkey is inserted:
one software volume and one encrypted volume where user data is stored. Password
entering (or “unlocking”) software is installed on the software volume, and not on the
host computer, yet it must be executed on the flash drive using the host operating
system. The Kingston Data Traveler Secure [Kingston Technology 2008] is similar to
Ironkey, except that it uses 256-bit keys with AES encryption and allows users to store
data on either the encrypted or nonencrypted partition.
Hardware-based encryption flash drives can employ good encryption techniques and,
similarly to software block-based encryption systems, directory structure is not revealed. Confidential policy changing is not supported, as encryption algorithms and
mode of operation cannot be changed. Keys may only be changed by completely resetting the device. In other words, the hardware cannot be reconfigured to meet changes
in confidential policy. Usability is good, in that most of the password-entering software
seems easy to use and requires the user to enter a password once per session. In terms
of overhead, all encryption operations are performed on the flash drives themselves
and do not consume CPU cycles and memory on the host machine. Therefore, the performance depends on the speeds of the on-flash cryptographic processing, flash data
access times, and the interface used to access secure flash drives.
3.2.2. Enclosures and Extension Cards. Hard disk enclosures and extension cards (either PCI or PCMCIA) have been used for several years as a fast, transparent encryption mechanism for sensitive data. Examples include SecureDisk Hardware [SecureDisk 2008] and RocSecure Hard Drives [RocSecure 2008]. These solutions intercept
and encrypt/decrypt data going to and from the hard drive real-time and use a specialized USB thumb drive as the encryption key. The encryption key is generated and
placed on the thumb drive by the manufacturer, and often the method of key generation (i.e., randomness technique) is not disclosed.
Enclosures and extension cards can employ good encryption techniques and do not
divulge any information about files or the structure of the file system on disk. Similarly to secure flash drives, policy regarding confidentiality changing is not supported,
since keys, encryption algorithms, and mode of operation cannot be changed. Secure


2:23

storage is not performed on a per-file level, so the entire hard disk must be encrypted.
This characteristic may not be flexible in regards to security policy. A USB thumb drive
acting as a key is simple to use. Similar to secure flash drives, enclosures and extension cards do not consume host CPU cycles and memory, as encryption is done in the
enclosure or on the extension card.
3.2.3. Encrypted Hard Drives. Seagate [2006] is introducing “DriveTrust Technology”
into their Momentus 5400 FDE series notebook hard drives, which implement full
disk encryption. This technology is implemented in the hard drive firmware and provides encryption, decryption, hashing (for passwords), digital signature, and randomnumber generation functions. Extra-sensitive data, such as keys, can be stored in separate, secure partitions on the hard disk. Any time a user application attempts to access
information on a secure partition, it must present its credentials to the “administrator
function” on the hard drive, which “authenticates the application, activates the appropriate secure partition, and allows the application to interact with the secure partition
through the trusted send/receive command set” [Seagate 2006]. The current hard disk
supports the AES encryption algorithm.
Once again, confidentiality is good if encryption keys are properly encrypted and
stored in a secure manner, and the encryption algorithm, mode of operation, and key
size are strong. Flexibility of policy settings is coarse-grained, as entire partitions must
be encrypted, and specifications such as encryption algorithm, mode of operation, and
key size cannot be changed. Keys also cannot be easily revoked. The user model is
simple, in the sense that cryptographic functions occur transparently, and the user
must enter a password/passphrase when mounting the hard drive. All encryption and
decryption operations happen in the hard disk firmware with a claim of little to no
performance penalty.
4. CONFIDENTIAL DATA ERASURE

When the time comes to remove confidential data, we must be sure that once deleted,
the data can no longer be restored. A full secure data lifecycle implies that data is
not only stored securely, but deleted in a secure manner as well. However, typical
file deletion (encrypted or not) only removes a file name from its directory or folder,
while a file’s content is still stored on the physical media until the data blocks are
overwritten. Unfortunately, average users believe that a file’s content is erased with
its name [Rosenbaum 2000].
Many forensic techniques are available to the determined (and well-funded) attacker
to recover the data. CMRR scanning microscopes [Hughes 2004] can recover data on
a piece of a destroyed disk if any remaining pieces are larger than a single 512-byte
record block in size, which is about 1/125” on today’s drives. Magnetic force microscopy
and magnetic force scanning tunneling microscopy [Gomez et al. 1992] analyze the
polarity of the magnetic domains of the electronic storage medium and can recover
data in minutes. For example, when a zero overwrites a one, the actual value will
become .95, and when a one overwrites a one it will be 1.05. Another approach is
to use a spin-stand to collect several concentric and radial magnetic surface images,
which can be processed to form a single surface image [Mayergoyz et al. 2000]. A less
well-funded attacker can resort to many drive-independent data recovery techniques
[Sobey et al. 2006], which may be used on most hard drives independently of their
make. The existence of these recovery techniques makes it mandatory that sensitive
data be securely deleted from its storage media.
Another issue is that true erasure may incur high overhead; therefore, security policy should have the flexibility to allow less-sensitive data to use conventional deletion

2:24


techniques. For example, a user might want to securely delete tax information from
his or her computer’s hard drive, and yet not mind if other files such as cookie recipes
could be recovered.
Confidential data deletion can be accomplished in three ways: physical destruction
of the storage medium, overwriting all of the sensitive data, and secure overwriting
the key of encrypted sensitive data. Each method has its relative strengths and will be
addressed in the following sections.
4.1. Physical Destruction

One way of deleting sensitive data is through physical destruction. For example, the
Department of Defense government document DoD 522.22M [1995] states that classified material may be destroyed by numerous methods including smelting, shredding,
sanding, pulverization, or acid bath. Needless to say, these methods will leave the storage medium unusable. The following definitions are supplied by Bennison and Lasher
for destruction of hard drives [Bennison and Lasher 2004]:
—With smelting, the hard drive is melted down into liquid metal, effectively destroying
any data contained therein.
—Shredding grinds the hard drive down into small pieces of scrap metal that cannot
be reconstructed.
—The sanding process grinds the hard drive platter down with an emery wheel or disk
sander until the recordable surface is removed completely.
—Pulverization is the act of pounding or crushing a hard drive into smaller pieces
through a mechanical process.
—An acid bath can be used for destruction of data on hard drive platters. A 58% concentration of hydriodic acid will remove the recordable surface of the platter.
Magnetic degaussing is another option that erases data by exposing a hard drive platter to an inverted magnetic field, which leaves data unrecoverable by software or laboratory attempts [OSS-Spectrum Project 2008]. This method also renders the storage
media unusable.
Physical destruction methods provide great confidentiality (the physical media is destroyed). On the other hand, the granularity of data destruction is the entire drive. For
example, we cannot securely delete only one file using these methods. Therefore, this
method does not support flexible security policies. Many of the discussed physical destruction methods require specialized equipment (which may not be easy to obtain) and
potential physical removal of the storage media (which may not be easy to perform), so
physical destruction may not be straightforward to perform. Conversely, since physical
destruction can destroy large amounts of data in a relatively short amount of time, the
performance in this sense is quite good (not including the time to acquire equipment
for physical destruction).
4.2. Data Overwriting

Another way to remove confidential data is to overwrite the data. Several standards
exist for overwriting data on electronic media. NIST recommends that magnetic media
be degaussed or overwritten at least three times [Grance et al. 2003]. The Department
of Defense document DoD 522.22M [1995] suggests an overwrite with a character,
its compliment, then a random character, as well as other software-based, overwrite
methods that refer to nonvolatile electronic storage as listed in Table IV.
Peter Gutmann [1996] developed a 35-pass data overwriting scheme to work on older
disks that use error-correcting-encoding patterns, referred to as run-length-limited


2:25

Table IV. Software-Based Methods of Erasing Data on Nonvolatile Storage, Defined in the
National Industrial Security Program Operating Manual
ID
Erasure Method
C
Overwrite all addressable locations with a character.
D
Overwrite all addressable locations with a character, its complement, then a
random character and verify.
E
Overwrite all addressable locations with a character, its complement, then a
random character.
H
Overwrite all locations with a random pattern, with binary zeros, and then with
binary ones.

Table V. Peter Gutmann’s 35-Pass Overwrite Technique [1996]
Pass Number
Data Written
Hex Code
1
Random
Random
2
Random
Random
3
Random
Random
4
Random
Random
5
01010101
01010101
01010101
0 × 55
0 × 55
6
10101010
10101010
10101010
0 × AA
0 × AA
7
10010010
01001001
00100100
0 × 92
0 × 49
8
01001001
00100100
10010010
0 × 49
0 × 24
9
00100100
10010010
01001001
0 × 24
0 × 92
10
00000000
00000000
00000000
0 × 00
0 × 00
11
00010001
00010001
00010001
0 × 11
0 × 11
12
00100010
00100010
00100010
0 × 22
0 × 22
13
00110011
00110011
00110011
0 × 33
0 × 33
14
01000100
01000100
01000100
0 × 44
0 × 44
15
01010101
01010101
01010101
0 × 55
0 × 55
16
01100110
01100110
01100110
0 × 66
0 × 66
17
01110111
01110111
01110111
0 × 77
0 × 77
18
10001000
10001000
10001000
0 × 88
0 × 88
19
10011001
10011001
10011001
0 × 99
0 × 99
20
10101010
10101010
10101010
0 × AA
0 × AA
21
10111011
10111011
10111011
0 × BB
0 × BB
22
11001100
11001100
11001100
0 × CC
0 × CC
23
11011101
11011101
11011101
0 × DD
0 × DD
24
11101110
11101110
11101110
0 × EE
0 × EE
25
11111111
11111111
11111111
0 × FF
0 × FF
26
10010010
01001001
00100100
0 × 92
0 × 49
27
01001001
00100100
10010010
0 × 49
0 × 24
28
00100100
10010010
01001001
0 × 24
0 × 92
29
01101101
10110110
11011011
0 × 6D
0 × B6
30
10110110
11011011
01101101
0 × B6
0 × DB
31
11011011
01101101
10110110
0 × DB
0 × 6D
32
Random
Random
33
Random
Random
34
Random
Random
35
Random
Random

0 × 55
0 × AA
0 × 24
0 × 92
0 × 49
0 × 00
0 × 11
0 × 22
0 × 33
0 × 44
0 × 55
0 × 66
0 × 77
0 × 88
0 × 99
0 × AA
0 × BB
0 × CC
0 × DD
0 × EE
0 × FF
0 × 24
0 × 92
0 × 49
0 × DB
0 × 6D
0 × B6

encodings. The basic idea is to flip each magnetic domain on the disk back and forth
as much as possible without writing the same pattern twice in a row and to saturate
the disk surface to the greatest depth possible. Peter Gutmann’s 35-pass overwrite
technique is demonstrated in Table V. Modern hard drives use a different encoding
scheme referred to as Partial-Response Maximum-Likelihood (PRML) encoding [Bauer
and Priyantha 2001]. Specific overwrite patterns have not yet been developed for the
newer PRML encoding.
The number of overwrite passes, thought necessary to delete data securely, is controversial. Some believe that various governmental agencies can recover data that has

2:26


been overwritten any number of times, but most data recovery companies say they
cannot recover data that has been overwritten even once. It is probably safe to say,
though, that the more times the data is overwritten, the more secure the deletion.
Three main methods exist to delete data securely from electronic storage media.
These methods involve software applications, file systems, and hard disk mechanisms.
Their characteristics and relative strengths are discussed in the following sections.
4.2.1. Software Applications. Three main software methods exist for overwriting sen-

sitive data:
(1) Overwrite the contents of a file.
(2) Delete the file normally, and then overwrite all free space in the partition.
(3) Erase the entire partition or disk.
The first method is probably the quickest if only a few small files are to be securely
overwritten. Many utilities, both free and commercial, are available to perform this
operation. Two common UNIX utilities are shred, made by the Free Software Foundation, Inc., and wipe [Nester 2008]. The shred utility will overwrite a file’s content
with random data for a configurable number of passes (default 25). However, shred
will not work on file systems that do not overwrite data in place. This can include
log-structured file systems [Rosenblum and Ousterhout 1991]; journaling file systems
(such as JFS, reiserFS, ext3); RAID-based file systems [Hartman and Ousterhout
1995]; file systems that take snapshots [Peterson and Burns 2005]; and compressed
file systems [Woodhouse 2001; Dumesnil 2008]. The shred utility will not overwrite a
file’s metadata.
In contrast, the wipe utility will write over file data using the 35-bit patterns recommended by Peter Gutmann [1996]. It will also attempt to remove filenames by renaming them, although this does not guarantee that the old filename (or metadata) will be
overwritten. The wipe utility has file system limitations similar to those of shred.
Overwriting all the free space in the partition is more of an afterthought method and
might be employed after files have been deleted the normal way. One example is scrub,
a Unix open-source utility [Garlick 2008], which erases free space in a partition by creating a file that extends to all the free space. A user needs to remember to remove the
file after the application is done. The scrub utility implements user-selectable pattern
algorithms compliant with the U.S. Department of Defense document 522.22M [1995].
Erasing the entire partition or disk will securely delete all confidential information
on the partition or disk such as data, metadata, and directory structures. One such
software utility is Darik’s Boot and Nuke, or DBAN [2008], which is a self-contained
boot floppy that wipes a hard drive by filling it with random data. Depending on the
size of the drive, the erasure process can take a long time.
Neither the software file-erasure nor the free-space-erasure methods will write over
previously deleted metadata. Therefore, these methods can still leak confidential information. On the other hand, partition overwriting software will erase all data and
metadata, as well as the structure of the file system.
Flexibility of confidentiality policy settings varies among these methods due to different granularities of deletion. For example, it is possible to erase only sensitive files
with software file erasure, while partition overwriting securely removes all files and
metadata, regardless of their confidentiality requirements.
All three methods are relatively easy to use. The user needs only to input a command
in order for the secure erasure to take place. However, the user still needs to initiate
secure erasure explicitly.


2:27

The level of performance can vary with software file erasure, since the user has to
wait for only chosen files (hopefully small) to be securely overwritten. The other two
methods may incur a considerable wait time, depending on the size of the free space
and storage partition.
4.2.2. File Systems. Two examples of data overwriting file systems are FoSgen
[Joukov et al. 2006] and Purgefs [Joukov and Zadok 2005], which are stackable file
systems built in FiST [Zadok and Nieh 2000]. Purgefs can overwrite file data and
metadata when deleting or truncating a file. Alternatively, to increase efficiency, the
purge delete option can be chosen using a special file attribute, for which only files with
such an attribute will be purge-deleted. Purgefs will delete data one or more times, and
supports the NIST standards and all NISPOM overwrite patterns (Table IV).
FoSgen consists of two components: a file system extension and the user mode shred
tool (Section 4.2.1). FoSgen intercepts files that require overwriting and moves them to
a special directory. The shred tool, invoked either manually or periodically, eventually
writes over the data in the special directory.
The authors of FoSgen have also created patches to add secure deletion functionality
to the ext3 file system. The first patch adds one-pass secure deletion functionality, and
the second patch supports multiple overwrites and securely deletes a file’s metadata.
Both implementations work in all three of ext3’s journaling modes, and erase either a
specially marked file’s data or all files.
Overwriting file systems can confidentially erase files and metadata using a variety of methods and passes. Users can specify the files and the number of passes and
writing patterns for security policies. These file systems are easy to use because a user
only needs to mount the file system with specific options. Unfortunately, depending on
the file size, overwriting files may incur a heavy performance penalty.
4.2.3. Semantically-Aware Hard Disks. A semantically-smart disk system (SDS)
[Sivathanu et al. 2003] tries to understand how the file system uses the disk beyond
the information made available by the interface between the two components. In other
words, an SDS can have access to certain knowledge of the file system’s on-disk data
structures and policies and can make intelligent decisions based on this knowledge,
such as prefetching blocks on a per-file basis. The authors describe a “secure-deleting”
SDS that guarantees that deleted file data will be rendered unrecoverable by recognizing deleted blocks through “operation inferencing” and overwriting those blocks with
different data patterns a specified number of times. Since it is possible that the file system might immediately reallocate those deleted blocks to another file, the SDS must
track those deleted blocks and queue up pending write requests to those blocks until
the secure overwrites have finished. The authors also note that the ext2 file system
must be mounted synchronously to operate correctly.
Using implicit detection techniques, which enable the storage system to infer block
liveness information without modification of the file system, Sivathanu et al. [2004]
make a next-generation SDS prototype called FADED (A File-Aware Data-Erasing
Disk) that operates under asynchronous file systems. Sometimes, though, this erasure
method can be incomplete. [Arpaci-Dusseau 2006], explain that due to the reordering
and reuse inherent in the ext2 file system, the SDS cannot definitively know whether
the current contents of a reallocated block are those of the new or old file. Here the
authors use a conservative overwrite method to deal with this problem, which erases
past layers of data on the block but leaves the current contents on the block. Because of
this, the user can never be positive that all deleted data has been confidentially erased
immediately after user file deletion.

2:28


SDS overwrites deleted files regardless of their need to be deleted securely, which
could result in unnecessary performance overhead and limited flexibility with security
policies. On the other hand, SDS operates transparently to the user (and operating
system), which make them easy to use.
A type-safe disk [Sivathanu et al. 2006] is similar to a SDS system in that it too can
perform semantically-aware optimizations by understanding the pointer relationships
between disk blocks that are imposed by high-level layers, such as the file system. One
difference between a type-safe disk, and semantically smart disk systems is that, when
using a type-safe disk, the file system must be modified in multiple ways, including the
use of the expanded type-safe disk API. The authors also describe a secure deletion
type-safe disk, which tracks when a block is garbage-collected and overwrites it one
or more times. It ensures that the block is not reallocated before it is overwritten
by postponing associated updates to the free block bitmap until the overwriting is
complete.
Since the file system is modified Sivathanu et al. [2006] guarantee that all previously
deleted data is overwritten. This characteristic also requires the use of a modified file
system, which may not be conducive to the policies in many environments. A typesafe disk’s overwriting granularity characteristics also result in the same unnecessary
performance overhead and limited flexibility with security policies as SDS due to the
overwriting of all data (sensitive or not). Like SDS systems, though, type-safe disks
are easy to use in that they operate in a completely transparent fashion to the user.
4.3. Encryption with Key Erasure

The third way to delete data securely is to encrypt the data and then securely erase
the key. The encryption key is often securely deleted using overwriting methods. This
combination allows for much faster secure deletion, in that only a small key is overwritten instead of the entire file (which could be very large). The downside is the extra
encryption/decryption overhead of regular file operations until the file is deleted. Not
many specialized solutions exist. One solution [Peterson et al. 2005] is built on top of
the versioning file system, ext3cow [Peterson and Burns 2005], and is based on the
all-or-nothing (AON) transform [Rivest 1997; Boyko 1999]. AON is defined as a cryptographic transform that, given only partial output, reveals nothing about its input.
AON is leveraged in the secure versioning file system to make decryption impossible if
one or more of the ciphertext blocks belonging to a file (or a file version) is deleted. No
commonly used solution of encryption with key erasure that we are aware of exists for
general-use file systems.
The policy and performance characteristics of any encryption method with the addition of key erasure are inherited from the base encryption method. The confidentiality characteristic is also inherited from the base encryption method, with one caveat:
the encrypted data may not stay “deleted” forever if the encryption method used to
initially encrypt the data is ever broken. For example, this may occur if a weakness
is ever found in the encryption method, or exhaustive search of the key space becomes possible. Also note that if the encryption key is protected by a password and
the password is merely forgotten, the strength of the secure deletion is directly correlated to the strength of the password. It is best to delete the encryption key(s) securely
through physical destruction or overwriting methods. The ease-of-use characteristic is
degraded in that the user must destroy the key explicitly.
5. OTHER CHALLENGES

This section discusses other challenges for implementing confidential storage and
deletion.


2:29

Fig. 12. Demonstration of bad sector forwarding.

5.1. Hard-Disk Issues

Two hard-disk-specific issues we must consider in relation to confidential data deletion
include bad sector forwarding and storage-persistent caches.
Bad sectors are disk locations that cannot be accessed consistently, developed during
the normal use of a hard disk. Bad sector forwarding is performed transparently at the
hardware level, in which the firmware identifies and remaps a bad sector to a reserved
area hidden from the user through the hard-disk defects table (G-List) [Shirobokov
2006]. In other words, the defective sector is replaced with a sector on a different part
of the hard disk. The defective sector cannot be accessed again by the hard disk itself.
Figure 12 demonstrates bad sector forwarding. Data sector 2 has gone bad and has
been detected by the hard disk firmware. The hard disk firmware remaps sector 2 to
the reserved area sector 0. Now whenever a read or write operation is performed on
sector 2, the operation will be mapped to reserved area sector 0.
The problem with bad sector forwarding is that the sector might still be partially
readable with only a small number of error bytes. This presents a problem if a bad
sector contains a key or IV that could still be read using other forensic methods. SDS
systems [Sivathanu et al. 2003, 2004; Arpaci-Dusseau et al. 2006] and type-safe disks
[Sivathanu et al. 2006] can address this problem (Section 4.2.3). Unfortunately the
ATA specification does not have a command to turn off bad sector forwarding, so
vendor-specific ATA commands must be used [Shirobokov 2006].
In addition to bad sector forwarding, persistent caches have been placed in diskstorage systems to improve performance [Joukov et al. 2006]. These caches may not
only defer writing to the actual physical media, but may also aggregate multiple writes
to the same location on the disk as a single write. In this case, the write cache of the
disk must be disabled.
5.2. Data Lifetime Problem

The data lifetime problem addresses the phenomenon of various copies of sensitive
data, such as passwords or encryption keys, being scattered all over a computer system during normal system operation [Garfinkel et al. 2004, Chow et al. 2005, 2004].
These locations include numerous buffers (such as string buffers, network buffers, or
operating system input queues), core dumps of memory, virtual memory, swap, hibernation values, and unintended leakage through application logs or features. A strategy
for secure deallocation of memory is presented in [Chow et al. 2005]. Garfinkel et al.
[2004] and Chow et al. [2005] argue that data lifetime is a system-wide issue that must
be addressed at every level of the software stack.
The attack model in this survey assumes that any attacks to recover sensitive data
are staged after the computer has been powered off, so volatile leakage of data such
as buffers, queues, and memory are beyond the scope of this survey. However, the recent cold boot suite of attacks [Halderman et al. 2008] demonstrate that encryption

2:30


keys and other data can be recovered from DRAMs used in most modern computers in
between cold reboots. The authors suggest four ways to partially mitigate the attack:
continuously discarding or obscuring encryption keys in memory; preventing any sort
of memory dumping software from being executed on the physical machine; physically
protecting the DRAM chips; and making the rate of memory decay faster. Hibernation files and swap are generally stored on the hard disk and may not go away once
the system is powered down. Some block-based encryption methods from Section 3.1.5
may be used to encrypt swap partitions and hibernation files. Hardware encryption
enclosures and extensions from Section 3.2.2 and encrypted hard drives from Section
3.2.3 can protect both swap and hibernation as the data is decrypted upon load transparently from the operating system.
6. OVERALL TRENDS

Quite a few general trends emerge when comparing secure storage techniques. These
techniques range from solutions closest to the user (application programs) to solutions
farthest from the user (hardware-based encryption). We can see that the level of confidentiality becomes higher as we move responsibility away from the user, which leads
to a lower possibility of human mistakes. On the other hand, the flexibility of policy
decreases as the solutions move away from the user. In the extreme, all policy decisions
are hard-coded in hardware, with no room for user configurations. Ease-of-use seems
to be correlated to the degree of user involvement and therefore, indirectly, the confidence rating. Performance gains steadily as the method is moved toward the hardware,
and then stabilizes when it is in hardware.
Table VI lists the strengths and weaknesses of the confidential storage approaches
discussed in this survey. Many of these observations are high-level generalizations, not
absolutes. The purpose of the table is to help the reader become aware of the issues
involved when designing solutions at different levels of the storage stack.
Confidential deletion techniques contain many tradeoffs. For example, data overwriting techniques have the potential to take a long time. Data encryption with key
erasure solves this problem, but introduces cryptography overhead. Solutions that are
farther away from user space and require little involvement from users, once again
tend to be easier to use if the necessary equipment is readily available.
Table VII lists the strengths and weaknesses of the confidential erasure approaches
discussed in this survey. Again, these observations are high-level generalizations to
help the reader become aware of the issues involved when designing solutions at different levels of the storage stack.
Clearly, a combined solution that can store and remove confidential information
should have the following ideal characteristics:
—high confidential storage and deletion granularity;
—acceptable performance overhead in terms of storage and deletion;
—enhanced security policy support to enable key revocation, encryption algorithm/mode of operation change and mitigation, and erasure technique;
—confidential storage and erasure of file and directory metadata; and
—easy to use with minimal user awareness.
While an overarching solution is currently elusive, we hope this survey sheds light
on the ground work and other considerations to handle confidential data storage and
deletion.


2:31

Table VI. Pros and Cons of Confidential Storage Approaches

Application

Layer

Approach
Generalized
encryption
programs

Pros
• Easy to deploy
• Offer a wide array of encryption
algorithms and modes of
operation
• Can be used on general files
Application-specific • Easy to deploy
encryption
• Easy to use through the
application

• Generally support multiple
encryption algorithms
• Easy to use but not completely
transparent
• Users may separate encrypted
files and non-encrypted files via
directories

NFS-based local
file systems

VFS/File System

User-space file
systems

• Operate on top of existing file
system
• Easily portable
transparent
mount points

Stackable file
systems

• Operate on top of existing file
system
transparent
mount points
transparent
• Users may mix encrypted and
non-encrypted files in the same
directory
• Good performance
• Directory structure and
metadata are not revealed
• Easy to use and transparent to
the user
• Good performance

Block

Disk-based file
systems

Block-based
encryption
systems

Cons
• Must be careful about
temporary files
• May be difficult to use
• Slower than lower-level
approaches
• Must be careful about
temporary files
• Offer few ways to perform
encryption
• Limited compatibility with
other applications
• Often limited to certain file
types
• Slower than lower-level
approaches
• Reveal directory structure
and file metadata information
• No easy key revocation
• No easy way to change
encryption or mode of
operation once started
• Higher performance overhead
due to kernel boundary
crossings
• Subject to vulnerabilities of
underlying network protocol
• Higher performance overhead
due to kernel boundary
crossings
• Slight performance overhead
due to layer of indirection
• Reveal some
directory-structure and
file-metadata information
• All files on the volume must
be encrypted
(Continued on next page)



Storage Media

2:32

Table VI. Continued
Secure flash drives, • Generally incorporate a
• No way (or very hard) to
enclosures and
strong confidential encryption
change encryption key
extension cards,
algorithm
• No way to change encryption
and encrypted
• Directory structure and
or mode of operation
hard drives
metadata are not revealed
• Generally all files on the
• Easy to use and transparent
volume must be encrypted
to the user
• Performance not tied to host
system

Table VII. Pros and Cons of Confidential Erasure Approaches

Semanticallyaware hard
disks

Pros
• Overwrite data using
standards-compliant patterns
• High level of deletion
granularity
• Easy to use, but user must
initiate process
• Overwrites data using
• May erase files after they
have been normally deleted
initiate process
• Will overwrite previously
deleted metadata
• Overwrites data using
initiate process
• Overwrites data and
metadata using
• Easy to use, but not
completely transparent
• Completely transparent
• Can erase sectors that can
only be accessed by the disk

Encryption with
key erasure

Application

Layer

Approach
Software file
erasure

• Fast confidential erasure of
files

Software
free-space
erasure

Partitionoverwriting
software

Any layer

Storage media

VFS/File
system

File systems

Cons
• Will not overwrite previously
deleted metadata
• Erasure wait time depends on
size of file to erase and
erasure pattern
• Will not overwrite previously
deleted metadata
size of remaining free space
on partition and erasure
pattern
• Low level of deletion
granularity
size of the partition and
erasure pattern
size of file to erase and
erasure pattern
• Confidentiality of erasure
depends on modification of
the file system and disk API
size of the partition and
erasure pattern
• Low erasure granularity
• Extra encryption/decryption
overhead of regular file
operations
• No generalized solutions exist
• Data could be recovered in the
future if encryption method
develops weakness or
exhaustive key search
becomes feasible

7. CONCLUSION

This survey took a look at the methods, advantages, and limitations of confidential
storage and deletion methods for electronic media in a nondistributed, single-user environment, with a dead forensic attack model. We compared confidential data-handling
methods using characteristics associated with confidentiality, policy, ease-of-use, and


2:33

performance. Additionally, we discussed challenges such as hard-disk issues and the
data lifetime problem, as well as the overall trends of various approaches. By compiling
experiences and constraints of various confidential storage and deletion techniques, we
hope that knowledge from research areas that have been evolving independently can
cross disseminate, to form solutions that are tolerant to a broader range of constraints.
ACKNOWLEDGMENTS
We thank Theodore Baker, Lois Hawkes, Steve Bellenot, and Mike Burmester for their advice and guidance
during this process. Chris Meyers, Cory Fox, Ryan Fishel, Dragan Lojpur, and Mark Stanovich have also
contributed to this work. We also thank Peter Reiher and Geoff Kuenning for reviewing an early draft of
this survey.

REFERENCES
ADYA, A., BOLOSKY, W. J., CASTRO, M., CERMAK, G., CHAIKEN, R., DOUCEUR, J. R., HOWELL, J., LORCH, J. R.,
THEIMER, M., AND WATTENHOFER, R. P. 2002. FARSITE: Federated, available, and reliable storage
for an incompletely trusted environment. In Proceedings of the 5th Symposium on Operating Systems
Design and Implementation. 1–14.
ARPACI-DUSSEAU, A. C., ARPACI-DUSSEAU, R. H., BAIRAVASUNDARAM, L. N., DENEHY, T. E., POPOVICI, F. I.,
PRABHAKARAN, V., AND SIVATHANU, M. 2006. Semantically-smart disk systems: past, present, and future. ACM SIGMETRICS Perform. Eval. Rev. 33 , 4.
BAUER, S. AND PRIYANTHA, N. B. 2001. Secure data deletion for Linux file systems. In Procceding of the
10th USENIX Security Symposium. USENIX Association, Berkeley, CA, 153–164.
BELAL, A. AND ABDEL-GAWAD, M. 2001. 2D-encryption mode. In Procceding of the 2nd NIST Modes of
Operation Workshop.
BENNISON, P. AND LASHER, P. 2004. Data security issues relating to end of life equipment. In Procceding
of the IEEE International Symposium on Electronics and the Environment. IEEE, Los Alamitos, CA,
317–320.
BLAZE, M. 1993. A cryptographic file system for UNIX. In Proceedings of the 1st ACM Conference on
Computer and Communications Security (CCS’93). ACM, New York, 9–16.
BOHMAN, T. 2007. Critical recording benefits from cryptic measures. VME Critical Syst. 26–28.
BOYKO, V. 1999. On the security properties of OAEP as an all-or-nothing transform. In Proceedings of
Advances in Cryptology (Crypto’99). Springer, Berlin, 503–518.
CAMPBELL, C. 1978. Design and specification of cryptographic capabilities. In National Bureau of Standards Special Publications, D. Branstad et al. Eds., U.S. Department of Commerce, Washington, D.C.,
54–66.
CATTANEO, G., CATUOGNO, L., DEL SORBO, A., AND PERSIANO, P. 2001. The design and implementation of a
transparent cryptographic filesystem for UNIX. In Proceedings of the Annual USENIX Technical Conference. USENIX Association, Berkeley, CA, 245–252.
CHAUM, D., AND EVERTSE, J.-H. 1985. Cryptanalysis of DES with a reduced number of rounds sequences
of linear factors in block ciphers. In Proceedings of Advances in Cryptology (Crypto’85). Lecture Notes
in Computer Science, vol. LNCS 218, Springer, Berlin, 192–211.
CHOW, J., PFAFF, B., GARFINKEL, T., CHRISTOPHER, K., AND ROSENBLUM, M. 2004. Understanding data lifetime via whole system simulation. In Proceedings of the 12th USENIX Security Symposium. USENIX
Association, Berkeley, CA, 2004.
CHOW, J., PFAFF, B., GARFINKEL, T., AND ROSENBLUM, M. 2005. Shredding your garbage: Reducing data
lifetime through secure deallocation. In Proceedings of the USENIX Security Symposium. USENIX Association, Berkeley, CA, 331–346.
DARIK’S BOOT AND NUKE. 2008. Homepage. http://dban.sourceforge.net/.
DOWDESWELL, R. AND IOANNIDIS, J. 2003. The CryptoGraphic disk driver. In Proceedings of the Annual
USENIX Technical Conference (FREENIX Track). USENIX Association, Berkeley, CA, 179–186.
DUMESNIL, A. 2008. e2compr homepage. http://e2compr.sourceforge.net.
DWORKIN, M. 2001. Recommendation for block cipher modes of operation, methods and techniques. NIST
Special Publication 800-38A. National Institute of Standards and Technology.
http://www.csrc.nist.gov/publications/nistpubs/800-38a/sp800-38a.pdf.


2:34


FEDERAL TRADE COMMISSION. 1999. Gramm-Leach-Bliley Financial Services Modernization Act, Pub. L.
No. 106-102, 113 Stat. 1338, codified at 15 U.S.C. §§6801-09.
FERGUSON, N., SCHROEPPEL, R., AND WHITING, D. 2001. A simple algebraic representation of Rijndael. In
Proceedings of Selected Areas in Cryptography. Lecture Notes in Computer Science, vol. 2259, Springer,
Berlin, 103–111.
FERGUSON, N. 2006. AES-CBC + Elephant diffuser: A disk encryption algorithm for Windows
Vista. Tech. rep. http://www.microsoft.com/downloads/details.aspx?FamilyID=131dae03-39ae-48bea8d6-8b0034c92555&DisplayLang=en.
FU, K. 1999. Group sharing and random access in cryptographic storage file systems. Master’s thesis,
MIT, Cambridge, MA.
GARFINKEL, T., PFAFF, B., CHOW, J., AND ROSENBLUM, M. 2004. Data lifetime is a systems problem. In
Proceedings of the 11th Workshop on ACM SIGOPS European Workshop: Beyond the PC. ACM, New
York.
GARLICK, J. 2008. Scrub utility project homepage. http://www.llnl.gov/linux/scrub/scrub.html.
GLIGOR, V. AND DONESCU, P. 2000. On message integrity in symmetric encryption. In Proceeding of the 1st
NIST Workshop on AES Modes of Operation.
GOH, E., SHACHAM, H., MODADUGU, N., AND BONEH, D. 2003. SiRiUS: Securing remote untrusted storage.
In Proceedings of the ISOC Network and Distributed Systems Security Symposium (NDSS). 131–145.
GOMEZ, R., ADLY, A., MAYERGOYZ, I., AND BURKE., E. 1992. Magnetic force scanning tunneling microscope
imaging of overwritten data. IEEE Trans. Magnetics 28, 5, 3141–3143.
GOUGH, V. 2008. EncFS: Encrypted file system. http://arg0.net/wiki/encfs.
GRANCE, T., STEVENS, M., AND MYERS, M. 2003. Guide to selecting information security products.
NIST special publication 800-36, National Institute of Standards and Technology, Gaithersburg, MD.
http://csrc.nist.gov/publications/nistpubs/800-36/NIST-SP800-36.pdf.
GUTMANN, P. 1996. Secure deletion of data from magnetic and solid-state memory. In Proceedings of the
6th USENIX Security Symposium. USENNIX Association, Berkeley, CA, 77–90.
HALCROW, M. 2007. eCryptfs: A stacked cryptographic filesystem. Linux J. 156, 2.
HARTMAN, J. AND OUSTERHOUT, J. 1995. The Zebra striped network file system. In ACM Trans. Comput.
Syst. 13, 3, 274–310.
HEWLETT-PACKARD. 2007. Encrypted volume and file system v1.0.02 release notes. Hewlett-Packard.
http://docs.hp.com/en/5992-3353/5992-3353.pdf.
HALDERMAN, J. A., SCHOEN, S. D., HENINGER, N., CLARKSON, W., PAUL, W., CALANDRINO, J. A., FELDMAN, A. J.,
APPELBAUM, J., AND FELTEN, E. W. 2008. Lest we remember: Cold boot attacks on encryption keys. In
Proceedings of the 17th USENIX Security Symposium (Sec’08). USENIX Association, Berkeley, CA.
HALEVI, S. AND ROGAWAY, P. 2003. A tweakable enciphering mode. In Proceedings of the Crypto’03. Lecture
Notes in Computer Science, vol. 2729, Springer, Berlin, 482–499.
HINES, M. 2007. IRS still losing laptops. InfoWorld.
http://www.infoworld.com/article/07/04/05/HNirslostlaptops 1.html.
HOHMANN, C. 2008. CryptoFS. http://reboot.animeirc.de/cryptofs/.
HOUSLEY, R. 2004. Cryptographic message syntax. Network Working Group RFC 3825, Standards Track.
http://www.ietf.org/rfc/rfc3852.txt.
HUGHES, G. 2004. CMRR Protocols for disk drive secure erase.
http://cmrr.ucsd.edu/people/Hughes/CmrrSecureEraseProtocols.pdf.
IRONKEY. 2007. Benefits of hardware-based encryption.
https://learn.ironkey.com/docs/IronKey WhitepaperBenefits of Hardware Encryption.pdf.
JETICO, INC. 2008. BestCrypt software home page. http://www.jetico.com/.
JOUKOV, N. AND ZADOK, E. 2005. Adding secure deletion to your favorite file system. In Proceedings of the
3rd International IEEE Security In Storage Workshop. IEEE, Los Alamitos, CA.
JOUKOV, N., PAPAXENOPOULOS, H., AND ZADOK, E. 2006. Secure deletion myths, issues, and solutions. In
Proceedings of the 2nd ACM Workshop on Storage Security and Survivability. ACM, New York.
KALLAHALLA, M., RIEDEL, E., SWAMINATHAN, R., WANG, Q., AND FU, K. 2003. Plutus: Scalable secure file
sharing on untrusted storage. In Proceedings of the 2nd USENIX Conference on File and Storage Technologies (FAST). USENIX Association, Berkeley, CA.
KAUFMAN, C., PERLMAN, R., AND SPECINER, M. 2002. Network Security: Private Communication in a Public
World 2nd Ed. Prentice Hall, Englewood Cliffs, NJ.
KENT, S. AND ATKINSON, R. 1998. Security architecture for the Internet Protocol. Technical rep. RFC 2401.
http://www.ietf.org/rfc/rfc2401.txt.



2:35

KILIAN, J. AND ROGAWAY, P. 1996. How to protect DES against exhaustive key search. In Proceedings of
the 16th Annual International Conference on Advances in Cryptology (CRYPTO ’96). N. Koblitz Ed.,
Springer, Berlin, 252–267.
KINGSTON TECHNOLOGY. 2008. Data traveler secure. http://www.kingston.com/flash/dt secure.asp.
KLEIMAN, S. R. 1986. Vnodes: An architecture for multiple file system types in Sun UNIX. In Proceedings
of the USENIX Annual Technical Conference. USENIX Association, Berkeley, CA, 238–247.
KNUDSEN, L. 2000. Block chaining modes of operation. In Proceeding of the Symmetric Key Block Cipher
Modes of Operation Workshop.
KOCH, W. 2008. The GNU privacy guard. http://www.gnupg.org/.
LISKOV, M., RIVEST, R., AND WAGNER, D. 2002. Tweakable block ciphers. In Advances in Cryptology
(CRYPTO ’02). Lecture Notes in Computer Science, Springer, Berlin.
LUBY, M. AND RACKOFF, C. 1988. How to construct pseudorandom permutations from pseudorandom functions. SIAM J. Comput. 17, 2, 373–386.
`
MAZIERES, D., KAMINSKY, M., KAASHOEK M., AND WITCHEL, E. 1999. Separating key management from file
system security. In Proceedings of the 17th ACM Symposium on Operating System Principles. ACM,
New York.
MAYERGOYZ, I., SEPRICO, C., KRAFFT, C., AND TSE, C. 2000. Magnetic imaging on a spin-stand. J. Appl. Phys.
87, 9, 6824–6826.
MCNEVIN, G. 2007. 2.9 million US residents at risk of ID theft. Image Data Manager.
http://www.idm.net.au/story.asp?id=8270.
MENEZES, A., OORSHOT, P., AND VANSTONE, S. 1997. Handbook of Applied Cryptography. CRC Press.
MICROSOFT CORP. 2002. Encrypting file system in Windows XP and Windows Server 2003. Tech. rep.
http://technet.microsoft.com/en-us/library/bb457065.aspx.
MICROSOFT CORP. 2003. Microsoft Office 2003 Eds. Security whitepaper.
http://www.microsoft.com/technet/prodtechnol/office/office2003/operate/o3secdet.mspx.
MICROSOFT CORP. 2006. Bitlocker drive encryption: Executive overview. Tech. rep.
http://technet.microsoft.com/en-us/windowsvista/aa906018.aspx#EQB.
MICROSOFT CORP. 2007. How to back up the recovery agent Encrypting File System (EFS) private key in
Windows Server 2003, in Windows 2000, and in Windows XP.
http://support.microsoft.com/kb/241201.
MICROSOFT CORP. 2008. How EFS works. Tech. Rep. 2008.
http://www.microsoft.com/technet/prodtechnol/windows2000serv/reskit/distrib/dsck efs duwf.mspx?
mfr=true.
MILLER, E., LONG, D., FREEMAN, W., AND REED, B. 2002. Strong security for network-attached storage. In
Proceedings of the 1st USENIX Conference on File and Storage Technologies (FAST). USENIX Association, Berkeley, CA.
MOOLENAAR, B. 2008. Vim the editor. http://www.vim.org/.
NESTER. 2008. Wipe homepage. http://wipe.sourceforge.net/.
NIST. 2007. Proposal to extend CBC mode by “Ciphertext stealing”.
http://csrc.nist.gov/groups/ST/toolkit/BCM/documents/ciphertext%20stealing%20proposal.pdf.
NIST. 2008. Request for public comments on XTS. National Institute of Standards and Technology,
Gaithersburg, MD. http://csrc.nist.gov/groups/ST/documents/Request-for-PublicComment-on XTS.pdf.
OPREA, A. 2007. Efficient cryptographic techniques for securing storage systems. Tech. rep. CMU-CS-07119, Carnegie Mellon University.
OSS-SPECTRUM PROJECT. 2008. Disposition of computer hard drives: Specifications for sanitization of hard
drives, Attachment 2. http://www.spectrumwest.com/Attach2.htm.
PATTERSON, D., GIBSON, G., AND KATZ, R. 1988. A case for redundant arrays of inexpensive disks. In Proceeding of the International Conference on Management of Data (ACM SIGMOD). ACM, New York,
109–116.
PETERS, M. 2004. Encrypting partitions using dm-crypt and the 2.6 series kernel. Linux.com.
http://www.linux.com/articles/36596.
PETERSON, Z., BURNS, R., HERRING, J., STUBBLEFIELD, A., AND RUBIN, A. 2005. Secure deletion for a versioning file system. In Proceedings of the 4th USENIX Conference on File and Storage Technologies.
USENIX Association, Berkeley, CA,143–154.
PETERSON, Z. AND BURNS, R. 2005. Ext3cow: A time-shifting file system for regulatory compliance. ACM
Trans. Storage 1, 2, 190–212.


2:36


PLETKA, R. AND CACHIN, C. 2007. Cryptographic security for a high-performance distributed file system.
In Proceedings of Mass Storage Systems and Technologies (MSST). 227–232.
RICHARDSON, R. 2007. CSI survey 2007: The 12th annual computer crime and security survey. Computer
Security Institute. http://www.gocsi.com/forms/csi survey.jhtml.
RIEDEL, E., KALLAHALLA, M., AND SWAMINATHAN, R. 2002. A framework for evaluating storage system security. In Proceedings of the 1st USENIX Conference on File and Storage Technologies (FAST’02). USENIX
Association, Berkeley, CA, 15–30.
RIVEST, R. L. 1997. All-or-nothing encryption and the package transform. In Proceedings of the Fast Software Encryption Conference.
ROCSECURE. 2008. RocSecure security encrypted hard drives.
http://www.datamediastore.com/rocstor-rocsecure-security-encripted-hard-drives.html.
ROSENBAUM, J. 2000. In defense of the DELETE key. The Green Bag 3, 4. www.greenbag.org/rosenbaum
deletekey.pdf.
ROSENBLUM, M. AND OUSTERHOUT, J. 1991. The design and implementation of a log-structured file system.
In Proceedings of the 13th Symposium on Operating Systems Principles.
RSA LABORATORIES. 1993. Cryptographic message syntax standard. Tech. note, PKCS 7, v1.5.
ftp://ftp.rsasecurity.com/pub/pkcs/ps/pkcs-7.ps.
RSA LABORATORIES. 1999. PKCS 5 v2.0: Password-based cryptography standard. ftp://ftp.rsasecurity.com/
pub/pkcs/pkcs-5v2/pkcs5v2-0.pdf.
SANDBERG, R., GOLDBERG, D., KLEIMAN, S., WALSH, D., AND LYON, B. 1985. Design and implementation of
the Sun network file system. In Proceedings of the USENIX Annual Technical Conference. USENIX
Association, Berkeley, CA, 119–130.
SEAGATE. 2006. DriveTrust technology: A technical overview.
http://www.seagate.com/docs/pdf/whitepaper/TP564 DriveTrust Oct06.pdf.
SECUREDISK. 2008. Hardware SecureDisk encryption.
http://www.usbgear.com/secure-disk-hardware-encryptedusbdrive.html.
SHAMIR, A. 1979. How to share a secret. Comm. ACM 22, 11, 612–613.
SHIROBOKOV, A. 2006. Drive imaging as part of data recovery. Whitepaper, ACE Data Recovery Engineering Inc. http://www.acedre.com.
SISWG. 2008a. IEEE Standard for Cryptographic Protection of Data on Block-Oriented Storage Devices.
IEEE Std 1619-2007, vol., no., pp.c1-32.
http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=4493450&isnumber=4493449.
SISWG. 2008b. Standard for Wide-Block Encryption for Shared Storage Media. Draft Standard
P1619.2/D7.
https://siswg.net/index.php?option=com docman&task=doc download&gid=134&Itemid=41.
SIVATHANU, M., PRABHAKARAN, V., POPOVICI, F. I., DENEHY, T. E., ARPACI-DUSSEAU, A. C., AND ARPACI-DUSSEAU,
R. H. 2003. Semantically-smart disk systems. In Proceedings of the 2nd USENIX Conference on File
and Storage Technologies. USENIX Association, Berkeley, CA.
SIVANTHANU, M., BAIRAVASUNDARAM, L. N., ARPACI-DUSSEAU, A. C., AND ARPACI-DUSSEAU, R. H. 2004. Life or
death at block level. In Proceedings of the 6th Symposium on Operating Systems Design and Implementation (OSDI’04). 379–394.
SIVATHANU, G., SUNDARARAMAN, S., AND ZADOK, E. 2006. Type-safe disks. In Proceedings of the 7th Symposium on Operating Systems Design and Implementation (OSDI’06).
SMITH, J. Mcrypt home page. http://mcrypt.sourceforge.net/.
SOBEY, C., ORTO, L., AND SAKAGUCHI, G. 2006. Drive-independent data recovery: The current state-of-theart. IEEE Trans. Magnetics 42, 2, 1, 188–193.
SPIES, T. 2008. Feistel finite set encryption mode. NIST proposed encryption mode.
http://csrc.nist.gov/groups/ST/toolkit/BCM/documents/proposedmodes/ffsem/ffsem-spec.pdf.
SQUARE, L. 2007. Stolen laptop may hold ID numbers. The Daily Reveille.
http://media.www.lsureveille.com/media/storage/paper868/news/2007/05/03/News/Stolen.Laptop.May.
Hold.Id.Numbers-2892874.shtml.
STINSON, D. 2002. Cryptography: Theory and Practice. 2nd Ed. Chapman & Hall/CRC, Boca Raton, FL.
SULLIVAN, B. 2005. Help! I left my identity in the backseat of a taxi. MSNBC: The Red Tape Chronicles.
http://redtape.msnbc.com/2005/11/why you should .html.
SZEREDI, M. 2008. Filesystem in USEr space. http://sourceforge.net/projects/avf.
U.S. CENSUS BUREAU. 2005. Computer and Internet use in the United States: 2003.
http://www.census.gov/prod/2005pubs/p23-208.pdf.



2:37

U. S. DEPARTMENT OF DEFENSE. 1995. National Industrial Security Program Operating Manual. 5220.22M, U.S. Government Printing Office, Washington, D.C.
WIRELESSWEEK. 2007. Alcatel-Lucent loses computer disk. Wirel. Week
http://www.wirelessweek.com/article.aspx?id=148070.
WHITTEN, A. AND TYGAR, J. D. 1999. Why Johnny can’t encrypt: A usability evaluation of pgp 5.0. In
Proceedings of the 8th USENIX Security Symposium. USENIX Association, Berkeley, CA, 169–184.
WOODHOUSE, D. 2001. JFFS: The journaling flash file system. In Proceedings of the Ottawa Linux Symposium. RedHat Inc.
WRIGHT, C., DAVE, J., AND ZADOK, E. 2003. Cryptographic file systems performance: What you don’t know
can hurt you. In Proceedings of the IEEE Security in Storage Workshop (SISW). IEEE, Los Alamitos,
CA, 47–61.
WRIGHT, C. P., MARTINO, M., AND ZADOK, E. 2003. NCryptfs: A secure and convenient cryptographic file
system. In Proceedings of the Annual USENIX Technical Conference. USENIX Association, Berkeley,
CA, 197–210.
YOUNG, E. AND HUDSON, T. 2008. OpenSSL project home page. http://www.openssl.org/.
ZADOK, E., BADULESCU, I., AND SHENDER, A. 1998. Cryptfs: A stackable vnode level encryption file system.
Tech. rep. CUCS-021-98, Computer Science Dept., Columbia University.
ZADOK, E. AND NIEH, J. 2000. FiST: A language for stackable file systems. In Proceedings of the Annual
USENIX Technical Conference. USENIX Association, Berkeley, CA, 55–70.
Received October 2008; accepted February 2009


A survey of confidential data storage and deletion methods

More Related Content

What's hot (20)

Similar to A survey of confidential data storage and deletion methods (20)

More from unyil96 (20)

Recently uploaded (20)

A survey of confidential data storage and deletion methods