eCryptfs: a Stacked Cryptographic Filesystem

The media has been delivering a seemingly endless stream of reports of lost or stolen laptops, backup tapes, hard drives and servers from government and corporate facilities. These devices often contain medical, financial and other sensitive data. When the storage devices fall into the wrong hands, attackers can access the data directly, completely bypassing the access control mechanisms in place in the organization's network. Reports indicate that millions of people already have been affected by such compromises. As a result, customers and citizens are at an increasing risk of identify fraud and loss of privacy.

Although the cryptographic technology to protect data confidentiality has existed for decades, many organizations have failed to integrate this technology into their processes for handling sensitive data. In cases where cryptography is included in that process, it is frequently obtrusive, costly and complicated. Organizations sometimes neglect to establish data encryption policies, and employees often ignore such policies once they are in place.

In cases where employees attempt to utilize cryptography, they often use it ineffectively. For instance, they often select weak keys, and it is easy to save or transfer data inadvertently in unencrypted form through insecure media (such as Web e-mail or a USB Flash drive). Security strategies that depend on individual applications performing their own encryption often fail when the user copies and pastes sensitive information to other applications that do not have cryptographic capability.

Data encryption needs to be made ubiquitous, transparent, flexible, easily deployable, integrated into the data handling process and, of course, secure enough to counter sophisticated attacks. These properties need to be in effect regardless of the particular applications accessing the data. To make encryption services application-agnostic, the operating system kernel itself should provide a system-wide data encryption service for sensitive information written to secondary storage.

Several options exist for filesystem encryption under Linux, all with various advantages and disadvantages. Device mapper crypt (dm-crypt) ships with the Linux kernel and provides block device layer encryption. Loop-AES and TrueCrypt, which must be obtained separately from the official Linux kernel, also provide encryption at the block device layer. With block device layer encryption, the user creates the filesystem on the block device, and the encryption layer transparently encrypts the data before writing it to the actual lower block device.

The main advantage of block device layer encryption is that it is simple in concept and implementation. Another advantage of block device layer encryption is that attackers learn nothing about the filesystem unless they have the key; for instance, attackers will not even know the type of filesystem or the directory structure. Sparse files can be securely and efficiently supported in filesystems on encrypted block devices.

Block device encryption can have disadvantages that stem from the lack of integration with the filesystem itself:

EncFS is a user-space cryptographic filesystem that operates via FUSE. User-space filesystems are easier to implement than kernel-native filesystems, and they have the advantage of being able to utilize user-space libraries easily. This makes it simple to implement feature-rich filesystems with less time and effort on the part of the developer. Unlike block device encryption solutions, EncFS operates as an actual filesystem. EncFS encrypts and decrypts individual files. Disadvantages of user-space filesystems based on FUSE include performance overhead from frequent kernel/user-space context switches and a current lack of support for shared writable memory mappings.

eCryptfs is a kernel-native stacked cryptographic filesystem for Linux. Stacked filesystems layer on top of existing mounted filesystems that are referred to as lower filesystems. eCryptfs is a stacked filesystem that encrypts and decrypts the files as they are written to or read from the lower filesystem.

Applications in user space make filesystem system calls that go through the kernel Virtual Filesystem (VFS). Both eCryptfs and the lower filesystem (for example, ext3, JFS, NFS and so on) are registered in the kernel VFS. The operations under the eCryptfs mountpoint first go to eCryptfs. eCryptfs retrieves key material from the user session key ring and uses the kernel cryptographic API to perform encryption and decryption of file contents. eCryptfs may make key management requests with the user-space eCryptfs dæmon (ecryptfsd). eCryptfs reads and writes encrypted content stored in files in the lower filesystem (Figure 1).

Figure 1. Application file operations go through eCryptfs.

Application file operations go through eCryptfs, which communicates with the kernel crypto API, the kernel key ring and the user-space eCryptfs dæmon to perform encryption and decryption. eCryptfs manipulates files in lower filesystems, such as JFS or ext3.

eCryptfs aims to provide the flexibility of a Pretty Good Privacy (PGP) application as a transparent kernel service. For that reason, the OpenPGP (RFC 2440) specification inspires the basic key handling techniques in eCryptfs. This includes the common procedure of using a hierarchy of keys when performing cryptographic operations (Figure 2).

Figure 2. eCryptfs encrypts and decrypts individual data extents.

eCryptfs encrypts and decrypts individual data extents in each file using a unique randomly generated File Encryption Key (FEK). The FEK is encrypted with the File Encryption Key Encryption Key (FEKEK), and the resulting Encrypted File Encryption Key (EFEK) is stored in the header of each lower file.

The cryptographic metadata is in the header region of the encrypted lower file. Users can transmit the lower file as is to other users, and the recipients can access the decrypted contents of the file through eCryptfs, so long as they have the proper key. This provides a high degree of flexibility in how the files can be handled while maintaining strong security.