EncFS Security Audit

-------------------------------------------------------------------------------
EncFS Security Audit
Taylor Hornby
January 14, 2014
-------------------------------------------------------------------------------

1. Introduction

This document describes the results of a 10-hour security audit of EncFS
1.7.4. The audit was performed on January 13th and 14th of 2014.

1.1. What is EncFS?

EncFS is a user-space encrypted file system. Unlike disk encryption software
like TrueCrypt, EncFS's ciphertext directory structure mirrors the plaintext's
directory structure. This introduces unique challenges, such as guaranteeing
unique IVs for file name and content encryption, while maintaining
performance.

1.2. Audit Results Summary

This audit finds that EncFS is not up to speed with modern cryptography
practices. Several previously known vulnerabilities have been reported [1, 2],
which have not been completely fixed. New issues were also discovered during
the audit.

The next section presents a list of the issues that were discovered. Each
issue is given a severity rating from 1 to 10. Due to lack of time, most
issues have not been confirmed with a proof-of-concept.

2. Issues

2.1. Same Key Used for Encryption and Authentication

SEVERITY: 3

EncFS uses the same key for encrypting data and computing MACs. This is
generally considered to be bad practice.

EncFS should use separate keys for encrypting data and computing MACs.

2.2. Stream Cipher Used to Encrypt Last File Block

SEVERITY: 7

As reported in [1], EncFS uses a stream cipher mode to encrypt the last file
block. The change log says that the ability to add random bytes to a block was
added as a workaround for this issue. However, it does not solve the problem,
and is not enabled by default.

EncFS needs to use a block mode to encrypt the last block.

EncFS's stream encryption is unorthodox:

1. Run "Shuffle Bytes" on the plaintext. N[J+1] = Xor-Sum(i = 0 TO J) {
P[i] } (N = "shuffled" plaintext value, P = plaintext) 2. Encrypt with
(IV, key) using CFB mode. 3. Run "Flip Bytes" on the ciphertext. This
reverses bytes in 64-byte chunks. 4. Run "Shuffle Bytes" on the
ciphertext. 5. Encrypt with (IV + 1, key) using CFB mode.

This should be removed and replaced with something more standard. As far as I
can see, this provides no useful security benefit, however, it is relied upon
to prevent the attacks in [1]. This is security by obscurity.

2.3. Generating Block IV by XORing Block Number

SEVERITY: 7

Given the File IV (an IV unique to a file), EncFS generates per-block IVs by
XORing the File IV with the Block Number. This is not a good solution, as it
leads to IV re-use when combined with the last-block stream cipher issue (see
the previous section):

The stream algorithm (see previous section) adds 1 to the IV, which could
*undo* the XOR with the block number, causing the IV to be re-used. Suppose
the file consists of one and a half blocks, and that the File IV's least
significant bit (LSB) is 1. The first block will be encrypted with the File IV
(block number = 0). The second (partial) block will be encrypted with File IV
XOR 1 (since block number = 1), making the LSB 0, using the stream algorithm.
The stream algorithm adds 1 to the IV, bringing the LSB back to 1, and hence
the same IV is used twice.

EncFS should use a mode like XTS for random-access block encryption, instead
of CBC mode with predictable IVs.

2.4. File Holes are Not Authenticated

SEVERITY: 5

File holes allow large files to contain "holes" of all zero bytes, which are
not saved to disk. EncFS supports these, but it determines if a file block is
part of a file hole by checking if it is all zeroes. If an entire block is
zeroes, it passes the zeroes on without decrypting it or verifying a MAC.

This allows an attacker to insert zero blocks inside a file (or append zero
blocks to the end of the file), without being detected when MAC headers are
enabled.

2.5. MACs Not Compared in Constant Time

SEVERITY: 6

MACs are not compared in constant time (MACFileIO.cpp, Line 209). This allows
an attacker with write access to the ciphertext to use a timing attack to
compute the MAC of arbitrary values.

A constant-time string comparison should be used.

2.6. 64-bit MACs

SEVERITY: 5

EncFS uses 64-bit MACs. This is not long enough, as they can be forged in 2^64
time, which is feasible today.

EncFS should use (at least) 128-bit MACs.

2.7. Editing Configuration File Disables MACs

SEVERITY: 7

The purpose of MAC headers is to prevent an attacker with read/write access to
the ciphertext from being able to make changes without being detected.
Unfortunately, this feature provides little security, since it is controlled
by an option in the .encfs6.xml configuration file (part of the ciphertext),
so the attacker can just disable it by setting "blockMACBytes" to 0 and adding
8 to "blockMACRandBytes" (so that the MAC is not interpreted as data).

EncFS needs to re-evaluate the purpose of MAC headers and come up with
something more robust. As a workaround, EncFS could add a command line option
--require-macs that will trigger an error if the configuration file does not
have MAC headers enabled.

3. Future Work

There were a few potential problems that I didn't have time to evaluate. This
section lists the most important ones. These will be prioritized in future
audits.

3.1. Padding Oracle

POSSIBLE SEVERITY: 8

EncFS uses Mac-then-Encrypt. This might make decryption padding oracles
possible through timing attacks.

3.2. Chosen Ciphertext Attacks

POSSIBLE SEVERITY: 10

Since the same key is used to encrypt all files, it may be possible for an
attacker with read/write access to the ciphertext and partial read access to
the plaintext (e.g. to one directory when --public is used) to perform a
chosen ciphertext attack and decrypt ciphertexts for which they have no
plaintext access.

EncFS should consider using XTS mode.

3.3. Possible Out of Bounds Write in StreamNameIO and BlockNameIO

POSSIBLE SEVERITY: 7

There is a possible buffer overflow in the encodeName method of StreamNameIO
and BlockNameIO. The methods write to the 'encodedName' argument without
checking its length. This may allow an attacker with control over file names
to crash EncFS or execute arbitrary code.

3.4. 64-bit Initialization Vectors

POSSIBLE SEVERITY: 5

Initialization vectors are only 64 bits, even when using AES instead of
Blowfish. This may lead to vulnerabilities when encrypting large (or lots of)
files.

4. Conclusion

In conclusion, while EncFS is a useful tool, it ignores many standard
best-practices in cryptography. This is most likely due to it's old age
(originally developed before 2005), however, it is still being used today, and
needs to be updated.

The EncFS author says that a 2.0 version is being developed [3]. This would be
a good time to fix the old problems.

As it is now, EncFS is not suitable for protecting mission-critical data.

5. References

[1] http://archives.neohapsis.com/archives/fulldisclosure/2010-08/0316.html

[2] http://code.google.com/p/encfs/issues/detail?id=128

[3] https://code.google.com/p/encfs/issues/detail?id=186

Source: https://defuse.ca/audits/encfs.htm