XTS mode revisited: high hopes for key scopes?
Milan Brož 1,2, Vladimír Sedláček 2 · 2025
Read XTS mode revisited: high hopes for key scopes? by Milan Brož 1,2, Vladimír Sedláček 2 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-18631.txt.
Introduction
Milan Brož and Vladimír Sedláček revisit the encryption mode that guards most disk drives. Their paper appeared in two thousand twenty-five. XTS mode encrypts data sector by sector on storage devices. It is the standard behind full-disk encryption tools. A proposed change to the IEEE standard would alter its key scope. Confusion about what XTS can and cannot do clouds the debate.
The authors summarize how the mode works in plain terms. Its limits are drawn sharply, especially for sector-based use. They aim to give the standards discussion a shared factual base. Wider encryption modes that might one day replace XTS are weighed. Storage security policy gains a clear reference point.
Catalog record
| CREATOR | Milan Brož 1,2, Vladimír Sedláček 2 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 16 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-18631
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
XTS mode revisited: high hopes for key scopes? Milan Brož 1,2, Vladimír Sedláček 2 Abstract This paper concisely summarizes the XTS block encryption mode for storage sector-based encryption applications and clarifies its limitations. In particular, we aim to provide a unified basis for much needed discussions about the newly proposed key scope change to the IEEE 1619 standard. 1 Introduction Since the introduction of XTS [ 3 ] in 2007, many software and hardware sector-based encryption systems have widely adopted it, e.g., BitLocker , VeraCrypt , Cryptsetup , and TCG Opal self-encrypting drives. Still, it seems to be surrounded by a shroud of misunderstandings, and some of its subtler points are still sometimes contested in the developer communities. It does not help that the existing NIST XTS-AES recommendation [ 1 ] only references the paid version of the IEEE standard [ 3 ] , making it probably the only NIST document that does not include a publicly available definition of the described cryptographic primitive (as already mentioned in Rogaway’s analysis [ 11 ] in 2011). The situation is especially problematic in light of the newly proposed key scope change to the standard [ 4 ] (again behind a paywall), which would render the vast majority of implementations non-compliant. This would force vendors to request (not only) open-source projects to make significant changes to adapt to the new version. In contrast, the reasons and security implications of the proposed change are not communicated at all. The goals of this paper are threefold: • Unify the terminology in the XTS context in a way easily understandable to different communities. • Describe XTS security limitations and clarify some contested points. • Encourage a constructive public discussion about future requirements and recommendations. 2 Terminology and definitions 2.1 Units and sectors The storage industry is known to mix the power-of-10 units (SI) and the power-of-2 units for a storage size. To avoid any confusion, we strictly use powers of two with IEC prefixes: a byte is 8 bits, a kibibyte (KiB) is 2^{10} bytes, a mebibyte (MiB) is 2^{20} bytes (1024 KiB), and a tebibyte (TiB) is 2^{40} bytes (1024 MiB). A block device is an abstraction of either a physical device (like a hard disk or a flash-based device) or a virtualized device that can be accessed (read, written to, or erased) only in units called sectors. A typical sector size is 4096 bytes (4 KiB) or 512 bytes (for older devices). In this text, we will only consider encryption of block devices (even though XTS can be used in other scenarios). Furthermore, we assume that the sector is always read or written to atomically, and individual sectors are accessed independently. 1 1 1 Modern storage devices are complex systems that can internally use different storage blocks for optimal performance, and can have specific requirements for optimal operations. The typical blocksize for modern ciphers is much smaller than the sector size (16 bytes for AES). Thus, to encrypt the whole sector, we need to use an encryption mode (such as XTS) that builds upon the original cipher. All modern real-world storage devices always use sector sizes that are multiples of 16 bytes, eliminating the need for padding during both encryption and decryption. 2 2 2 The only legacy exception is the 520-byte sector (a 512-byte data sector with an additional 8-byte integrity field) on enterprise devices. 2.2 Notation In this text, E will denote a blockcipher (e.g., AES-128) of blocksize n bits, so that for each key K , E_{K} is a permutation on \{0,1\}^{n} – the set of binary n -bit strings. For s,s^{\prime}\in\{0,1\}^{n} , we will write s\|s^{\prime} and s\oplus s^{\prime} , for the concatenation and XOR respectively. The s=a_{n-1}\dots a_{1}a_{0} can be seen as the class of the polynomial a_{n-1}x^{n-1}+\dots+a_{1}x+a_{0} in the finite field \mathbb{F}_{2^{n}} . With this interpretation, we can multiply binary strings by elements of \mathbb{F}_{2^{n}} using shift and XOR operations (though it depends on the polynomial chosen to represent \mathbb{F}_{2^{n}} ). S will denote the total number of sectors on the device and J the number of n -bit blocks within a sector. Correspondingly, N will always denote a sector number and j the number of the n -bit block within the sector. We will denote the unencrypted data (plaintext) at block j within sector N by P_{N,j} and naturally extend this by concatenation: P_{N}:=P_{N,0}\|P_{N,1}\|\dots\|P_{N,J-1} will denote the unencrypted data at sector number N , while P:=P_{0}\|P_{1}\|\dots\|P_{S-1} will denote the unencrypted data of the whole device. Analogically, we denote the encrypted data by C_{N,j}:=E_{K}(P_{N,j}),C_{n},C , respectively. 3 XTS encryption mode The XTS mode – or rather XTS-AES – was first defined in Annex D.4.3 in [ 3 ] as an “instantiation” of Rogaway’s XEX [ 10 ] with the underlying blockcipher being AES-128 or AES-256. Though not quite, as there are several differences: • While XEX uses a single symmetric key to encrypt both the data and the sector number, XTS uses two different symmetric keys: K to encrypt the data, and K_{T} to encrypt the sector number. • While XEX starts at j=1 , XTS starts at j=0 . • XTS allows for ciphertext stealing to deal with encrypting data whose size is not a multiple of n . As explained in Section 2.1 , we focus only on full sector encryption, so we can ignore this. 3 3 3 XTS has been used in the Linux kernel since 2007 . In contrast, ciphertext stealing was only added in 2019 , and probably there is still no active in-kernel user. 3.1 XTS definition We propose a definition that seeks a compromise between Rogaway’s original XEX formality and the practical use case at hand. Definition 1 (XTS encryption mode) . Let K , K_{T} be symmetric keys for a blockcipher E of blocksize 4 4 4 This is not essential, but multiplication by \alpha differs for different block sizes. 128 bits. Then the XTS mode of E encrypts the storage data in the following way: C_{N,j}:=E_{K}(P_{N,j}\oplus T_{N,j})\oplus T_{N,j}, C_{N}:=C_{N,0}\|C_{N,1}\|\dots\|C_{N,J-1}, C:=C_{0}\|C_{1}\|\dots\|C_{S-1}, where T_{N,j}:=E_{K_{T}}(N)\cdot\alpha^{j} and \alpha:=0^{126}10 is the class of x in \mathbb{F}_{2^{128}}:=\mathbb{F}_{2}[x]/(x^{128}+x^{7}+x^{2}+x+1) . Note that for a binary string a_{127}a_{126}\dots a_{1}a_{0} , we have s\cdot\alpha=\begin{cases}a_{126}a_{125}\dots a_{0}0&if $a_{127}=0$,\\ a_{126}a_{125}\dots a_{0}0\oplus 0^{120}10^{4}1^{3}&if $a_{127}=1$,\end{cases} as x^{128} and x^{7}+x^{2}+x+1 represent the same element in \mathbb{F}_{2^{128}} . Figure 1: The principle of XTS sector encryption (without ciphertext stealing). Example 1 (AES-XTS-128 encryption of one sector with two blocks) . For \displaystyle N \displaystyle=0x00000000000000000000000000000001, \displaystyle K \displaystyle=0x11111111111111111111111111111111, \displaystyle K_{T} \displaystyle=0x22222222222222222222222222222222, \displaystyle P_{N,0} \displaystyle=0x44444444444444444444444444444444, \displaystyle P_{N,1} \displaystyle=0x88888888888888888888888888888888, the tweak values are generated in the following way: \displaystyle T_{N,0} \displaystyle=E_{K_{T}}(N)=0x6752ca5febca0f3fc8dc9dfc2a916295, \displaystyle T_{N,1} \displaystyle=T_{N,0}\cdot\alpha^{1}=0x49a494bfd6951f7e90b93bf95522c52a. The sector blocks are then encrypted as follows: \displaystyle C_{N,0} \displaystyle=E_{K}(P_{N,0}\oplus T_{N,0})\oplus T_{N,0} \displaystyle=E_{K}(0x6752ca5febca0f3fc8dc9dfc2a916295)\oplus T_{N,0} \displaystyle=0x13f084e65a7ca361f74957c9b11c7710\oplus T_{N,0} \displaystyle=0x74a24eb9b1b6ac5e3f95ca359b8d1585 \displaystyle C_{N,1} \displaystyle=E_{K}(P_{N,1}\oplus T_{N,1})\oplus T_{N,1} \displaystyle=E_{K}(0xc12c1c375e1d97f61831b371ddaa4da2)\oplus T_{N,1} \displaystyle=0x2cada9d22ad34bf19a226c2c824f0364\oplus T_{N,1} \displaystyle=0x65093d6dfc46548f0a9b57d5d76dc64e 3.2 Threat model The simplest threat model for length-preserving sector encryption usually considers a “stolen device”, allowing the attacker to access only one ciphertext snapshot at time. A typical example of such a threat model is the TCG Opal definition [ 2 ] : “protect the confidentiality of stored user data against unauthorized access once it leaves the owner’s control (following a power cycle and subsequent deauthentication).” This model expects the user to recognize when the