Implementing Encrypted Home Directories
When used correctly, encrypted filesystems can be an effective way to protect sensitive data stored on a computer. Standard encryption packages, such as the GNU Privacy Guard (GPG), are good for encrypting e-mail. They are not as convenient, however, for encrypting files one wishes to read or modify many times throughout the files' lifetimes.
Unlike GPG, an encrypted filesystem is transparent to users. There is no hassle of manually decrypting a file before reading it or encrypting it again after modifying it. Potential user forgetfulness also is mitigated. After introducing some encrypted filesystems available for Linux, this article explains how to create an encrypted home directory that is automatically mounted at login time and unmounted at logout. Finally, this article introduces some concepts that demonstrate the dangers of improperly implemented encryption techniques.
Why would one want to encrypt the data stored on a computer? Isn't protecting sensitive data what filesystem permissions are for? Although useful, filesystem permissions quickly lose their effectiveness when an attacker has complete control of the storage medium the permissions are used to protect. For example, if someone steals my Linux laptop, an Apple iBook, its filesystems permissions are of little use against the thief who can simply boot from a diabolical CD-ROM. The same is true if I send my laptop to Apple for repairs. A dishonest employee conceivably could read my files. Correctly encrypting the files on a computer is a safe form of protection, because the process does not depend on the integrity of the operating system after the encryption takes place.
I have chosen to encrypt only the home directories on my iBook. Encrypting the entire filesystem, starting with root, was not acceptable in my case due to performance penalties and other factors. Information on implementing this technique can be found on the Internet—it requires using Linux's initial ramdisk capability. In my experiences with an x86-based system, the encryption technique I chose is around 50% slower than a non-encrypted XFS filesystem when writing to disk using buffered I/O. Encrypting only home directories obviously leaves many files, such as system logs, as plain text, but these were not significantly sensitive in my case. Encrypting only home directories was a sensible compromise for me.
Linux supports a few options for encrypting filesystems. Systems such as the Transparent Cryptographic File System provide an encrypted extension to NFS-served ext2 volumes. Other filesystems contain integrated cryptography in their design for local use. For my application, the best solution seemed to be the concept of loopback encrypted filesystems. As you will see, loopback encryption can be used to encrypt any filesystem type supported by Linux, including the proven ext2, XFS and ReiserFS.
Linux loopback filesystem support simply allows a user to mount an ordinary file as if it were a device, such as /dev/hda1. This traditionally is useful for doing things like mounting a CD-ROM filesystem image to populate and test before burning it to CD-R media or distributing bootable floppy disk images. Herbert Valerio Riedel's GNU/Linux Cryptographic API (CryptoAPI) and util-linux patches add support for mounting encrypted filesystem images to the loopback driver.
Before delving into the details of loopback encrypted filesystems, let's take a look at how to create and mount a vanilla loopback filesystem. First, create a file to contain the filesystem. This example creates a file large enough to host a 20MB filesystem:
dd if=/dev/zero of=plaintext.img bs=1M count=20
Second, associate the newly created file with a loopback device:
losetup /dev/loop0 plaintext.img
Next, create a filesystem within the file, using the newly associated loop device:
mkfs -t ext2 /dev/loop0
Finally, mount the filesystem and use it as if it were any other mounted volume:
mount /dev/loop0 mount point
Now, let us move on to the encrypted case. In order to use loopback encrypted filesystems, the kernel must support them. Most distributions do not include this functionality, so a custom-built kernel probably is necessary. A cryptographic API package similar to the one I use is being merged into the mainstream 2.5 kernel. However, for the stable 2.4 tree, the GNU/Linux CryptoAPI patches are necessary and available at www.kerneli.org. Once you apply the patch-int and loop-hvr patches, Cryptographic options should be available when you configure your kernel. The following options must be enabled:
cryptographic API support (CONFIG_CRYPTO)
generic loop cryptographic filter (CONFIG_CRYPTOLOOP)
cryptographic ciphers (CONFIG_CIPHERS)
You have to enable at least one cipher as well. Remember which cipher you choose. I chose AES, originally called Rijndael, and use AES in my examples.
Build and install the newly configured kernel. All of the kernel's encryption code may be compiled as modules. If you choose to build kernel modules, don't forget to insert them before you try to use their functionality. It also is necessary to patch util-linux, compile the tools and install them. The relevant util-linux patch is available at www.kernel.org/pub/linux/kernel/people/hvr/util-linux-patch-int. You should find that you end up with modified mount and losetup commands.
Now we are ready to create a loopback encrypted filesystem using a process similar to that which we used to create a vanilla loopback filesystem. First, in order to make it difficult to differentiate between encrypted blocks and unused disk space, the file that will hold the loopback filesystem is created using /dev/urandom instead of /dev/zero:
dd if=/dev/urandom of=ciphertext.img bs=1M count=20
After creating the host file, it must be temporarily associated with a loopback device, as before. This time, however, we must tell losetup that the loopback device is to be encrypted, in this case with the AES cipher:
losetup -e aes /dev/loop0 ciphertext.img
Enter the password and possibly—depending on the cipher you decided to use—the key size you wish to use for the volume when prompted by losetup.
Creating the filesystem is done in a manner identical to that for creating a vanilla loopback device. The encryption was set up in the previous step and is now handled by the loopback driver:
mkfs -t ext2 /dev/loop0
In addition to modifying losetup, the util-linux patch also makes the mount command crypto-aware. Mounting an encrypted loopback volume is a simple process, given the correct command parameters:
mount -o loop,encryption=aes ciphertext.img \ mount point
mount prompts you for a password and possibly for a key size.
Now that you understand how to mount and unmount encrypted loopback filesystems manually, an introduction to pam_mount is appropriate. pam_mount is a PAM module that simplifies the management of volumes and should be mounted when a user logs in to a system. It can handle mounting things like Samba-hosted volumes, Novell-hosted volumes and encrypted filesystems. The original author of pam_mount is Elvis Pftzenreuter. Mukesh Agrawal wrote the patch that first added support for loopback encrypted volumes. The author of this article now maintains pam_mount, which is available at www.flyn.org.
Instead of having to mount encrypted volumes manually, a system administrator can configure pam_mount to mount and unmount the volumes automatically when users log on and off. This can be configured so the system password also unlocks the encrypted volume, essentially creating a completely transparent encrypted volume.
pam_mount can employ three different techniques to unlock an encrypted volume. The first technique is rather boring. When the encrypted volume's key is unrelated to the system's login password, pam_mount simply prompts users for the key to their encrypted volume. In order to use this technique on a system, pam_mount.so and pmhelper must be installed and configured. The standard ./configure, make and make install commands build and install pam_mount's binaries and configuration file.
You should find the stock pam_mount.conf in /etc/security. Inspect and tailor it to your own system. The stock pam_mount.conf file is well documented. The most important change necessary is to add definitions for the volumes that should be mounted to the end of the file. The following is the definition format for encrypted loopback filesystems, as documented in the stock file:
volume user local ignored loopback file mount point mount options fs key cipher fs key path
Here is an example that mounts an AES-encrypted loopback filesystem hosted by /home/mike.img at /home/mike when Mike logs on:
volume mike local - /home/mike.img /home/mike loop,user,exec,encryption=aes,keybits=256 - -
Next, add the lines auth required pam_mount.so try_first_pass and session required pam_mount.so try_first_pass to the configuration files of the PAM-supporting services you want to support loopback encrypted filesystems. As an example, here is the /etc/pam.d/login file from my laptop:
auth requisite pam_securetty.so auth requisite pam_nologin.so auth required pam_env.so auth required pam_unix.so nullok account required pam_access.so account required pam_unix.so session required pam_unix.so session optional pam_lastlog.so session optional pam_motd.so session optional pam_mail.so standard noenv password required pam_unix.so nullok obscure \ min=4 max=8 md5 auth required pam_mount.so try_first_pass session required pam_mount.so try_first_pass
Finally, create the user's loopback encrypted filesystem using the steps listed in the introduction to encrypted loopback filesystems.
The second technique for pam_mount to unlock a volume is more convenient for users. If, when creating the encrypted volume using the same method as the first technique, a user specifies his or her login password as the volume key, then pam_mount unlocks the volume using the same password the user enters to login.
The third technique is the most flexible and requires a more sophisticated description. Here are a few terms to help you understand how this technique works:
sk: system key, the key or password used to log in to the system.
fsk: filesystem key, the key that allows you to use the filesystem you want pam_mount to mount for you.
E and D: an OpenSSL-supported synchronous encryption/decryption algorithm, bf-ecb, for example.
efsk: encrypted filesystem key, efsk = E_sk (fsk), stored somewhere on the local filesystem (that is, /home/user.key).
pam_mount reads efsk from the local filesystem, performs fsk = D_sk (efsk) and uses fsk to mount the filesystem. This technique has the advantage of allowing users to change their login passwords without having to re-encrypt their home directories using this new key. If the login password is changed, simply regenerate efsk (that is, /home/user.key) using efsk = E_newsk (D_oldsk (efsk)). A script named passwdehd is included in pam_mount to do this for you.
To implement this third technique, begin by creating the file to host the encrypted filesystem (as before):
dd if=/dev/urandom of=/home/user.img \ bs=1M count=image size in MB
Then, create a file (efsk) containing the volume's password (fsk) using /dev/urandom, encrypted using the user's login password as the key:
dd if=/dev/urandom bs=1c count=keysize / 8 | \ openssl enc -bf-ecb > /home/user.key
Next, create an encrypted loopback filesystem. The filesystem's key should be fsk (generated using /dev/urandom, encrypted and stored as /home/user.key in step 2).
openssl enc -d -bf-ecb -in /home/user.key | \ losetup -e aes -k keysize -p0 /dev/loop0 \ /home/user.img mkfs -t ext2 /dev/loop0 umount /dev/loop0 losetup -d /dev/loop0
Finally, in pam_mount.conf, set the fs key cipher variable to the cipher used to encrypt fsk, in this case bf-ecb, and set the fs key path variable to efsk's path, in this case, /home/user.key.
In his definitive text, Applied Cryptography, Bruce Schneier states, “Software encryption is scary.” What he means is, it is difficult to design truly secure encryption software for computers running general-purpose operating systems such as Linux. For example, modern operating systems can swap memory to disk at any time, and this memory could contain plain text or encryption keys. An encrypted volume is useless if its key has been written to disk by the operating system. One way to reduce the effects of this is to encrypt one's swap volume. CryptoAPI still cannot do this safely, but it is in development. A similar project, LoopAES, already can encrypt a system's swap space.
Consider again the example where I sent my iBook to Apple for repairs. Though my home directory is encrypted, my data still may not be completely safe. A dishonest employee could boot his or her diabolical CD-ROM and replace, for example, the login binary on my system with the employee's own design. When my computer is returned and I log in, my encryption key could be shipped off to a remote computer by the newly installed login program. An intrusion detection system would make this scenario much less likely.
Another possible weak point in a system employing encrypted home directories using pam_mount is the system's login password. Because the login password is used, directly or indirectly, to unlock an encrypted filesystem, it must be strong. Countless studies have shown that most passwords chosen by users are quite weak. Rather than blindly increasing the required length of passwords, spend some time reading Bruce Schneier's Secrets and Lies. A strong passphrase, written down and stored in your wallet may be more secure than a memorized password. The addition of a physical authentication token might be even better. Remember, if your system login password is not secure, your encrypted filesystem is as good as read.
Finally, encrypted filesystems can be a double-edged sword. What if you forget your encryption key? What if you use the third technique described above and accidentally delete all records of your encrypted filesystem key? What if my or someone else's encryption-related software is buggy? All of these problems can result in you having to try 2128 or so different encryption keys to get your filesystem back. Your data may be as good as gone. As with any system administration endeavor, make filesystem backups. Ideally, these backups are not encrypted and are physically locked up somewhere secure.
The bottom line is many subtle considerations and procedures are required to administer a reasonably secure system beyond the use of a modern encryption algorithm like AES. To quote Matt Blaze's contribution to Applied Cryptography:
High-quality ciphers and protocols are important tools, but by themselves poor substitutes for realistic, critical thinking about what is being protected and how various defenses might fail (attackers, after all, rarely restrict themselves to the clean, well-defined threat models of the academic world).
After reading this article, you should be familiar with the concept of an encrypted loopback filesystem. In addition, you should be able to deploy encrypted filesystems on the systems you administer and manage them with the pam_mount PAM module. In the future, I would like to see the CryptoAPI patches and pam_mount supported by the major Linux distributors. I also would like to see the CryptoAPI patch rolled into the mainstream util-linux package. Properly administered encrypted home directories are a powerful security tool.
Mike Petullo is a platoon leader in the US Army, stationed in Germany. He jumps out of airplanes by day, fights C code bugs by night and has been tinkering with Linux since early 1997. He welcomes your comments sent to firstname.lastname@example.org.