Back to Dm-crypt.
Before encrypting a drive, you should perform a secure erase of the disk by overwriting the entire drive with random data. To prevent cryptographic attacks or unwanted file recovery, this data is ideally indistinguishable from data later written by dm-crypt. For a more comprehensive discussion see Disk encryption#Preparing_the_disk.
Secure erasure of the hard disk drive
In deciding which method to use for secure erasure of a hard disk drive, remember that this needs only to be performed once for as long as the drive is used as an encrypted drive.
For detailed instructions on how to erase and prepare a drive consult Securely wipe disk.
dm-crypt specific methods
The following methods are specific for dm-crypt and are mentioned complementarily, because they are very fast and can be performed after a partition setup too.
The cryptsetup FAQ mentions a very simple procedure to use an existing dm-crypt-volume to wipe all free space accessible on the underlying block device with random data by acting as a simple pseudorandom number generator. It is also claimed to protect against disclosure of usage patterns.
dm-crypt wipe before installation
First, create a temporary encrypted container on the partition (
sdXY) or the full disk (
sdX) you want to encrypt, e.g. using default parameters
# cryptsetup open --type plain /dev/sdXY container
Second, check it exists
# fdisk -l
Disk /dev/mapper/container: 1000 MB, 1000277504 bytes ... Disk /dev/mapper/container does not contain a valid partition table
Finally, wipe it with pseudorandom (encrypted data), a use of
/dev/urandom is not required as the encryption cipher is used for randomness:
# dd if=/dev/zero of=/dev/mapper/container
dd: writing to ‘/dev/mapper/container’: No space left on device
Now the next step is #Partitioning.
dm-crypt wipe free space after installation
The same effect can be achieved if a file is created on an encrypted partition that fills the free space of the partition completely after the system is installed, booted and filesystems mounted. That is because encrypted data is practically indistinguishable from random.
# dd if=/dev/zero of=/file/in/container # rm /file/in/container
The above process has to be repeated for every partition blockdevice created.
Wipe LUKS header
The partitions formatted with dm-crypt/LUKS contain a header with the cipher and crypt-options used, which is referred to
dm-mod when opening the blockdevice. After the header the actual random data partition starts. Hence, when re-installing on an already randomised drive, or de-commissioning one (e.g. sale of PC, switch of drives, etc.) it may be just enough to wipe the header of the partition, rather than overwriting the whole drive - which can be a lengthy process.
Wiping the LUKS header will delete the PBKDF2-encrypted (AES) master key, salts and so on.
/dev/sda1in this example) and not directly to the disks device node. If you did set up encryption as a device-mapper layer on top of others, e.g. LVM on LUKS on RAID then write to RAID respectively.
A header with one single default 256 bit size keyslot is 1024KB in size. It is advised to also overwrite the first 4KB written by dm-crypt, so 1028KB have to be wiped. That is
For zero offset use:
#head -c 1052672 /dev/zero > /dev/sda1; sync
For 512 bit key length (e.g. for aes-xts-plain with 512 bit key) the header is 2MB.
If in doubt, just be generous and overwrite the first 10MB or so.
#dd if=/dev/zero of=/dev/sda1 bs=512 count=20480
When wiping the header with random data everything left on the device is encrypted data. An exception to this may occur for an SSD, because of cache blocks SSDs employ. In theory it may happen that the header was cached in these some time before and that copy may consequently be still available after wiping the original header. For strong security concerns, a secure ATA erase of the SSD should be done (procedure please see the cryptsetup FAQ 5.19).
This section only applies when encrypting an entire system. After the drive(s) has/have been securely overwritten, a proper partitioning scheme will have to be accurately chosen, taking into account the requirements of dm-crypt, and the effects that the various choices will have on the management of the resulting system.
It is important to note from now that in every case there has to be a separate partition for
/boot that must remain unencrypted, because the bootloader needs to access the
/boot directory where it will load the initramfs/encryption modules needed to load the rest of the system (see mkinitcpio for details). If this raises security concerns, see dm-crypt/Specialties#Securing the unencrypted boot partition.
Another important factor to take into account is how the swap area and system suspension will be handled, see dm-crypt/Swap encryption.
In the simplest case, the encrypted layers can be based directly on the physical partitions; see Partitioning for the methods to create them. Just like in an unencrypted system, a root partition is sufficient, besides another for
/boot as noted above. This method allows deciding which partitions to encrypt and which to leave unencrypted, and works the same regardless of the number of disks involved. It will also be possible to add or remove partitions in the future, but resizing a partition will be limited by the size of the disk that is hosting it. Finally note that separate passphrases or keys are required to open each encrypted partition, even though this can be automated during boot using the
crypttab file, see Dm-crypt/System configuration#crypttab.
Stacked block devices
If more flexibility is needed, though, dm-crypt can coexist with other stacked block devices like LVM and RAID. The encrypted containers can either reside below or on top of other stacked block devices:
- If the LVM/RAID devices are created on top of the encrypted layer, it will be possible to add, remove and resize the file systems of the same encrypted partition liberally, and only one key or passphrase will be required for all of them. Since the encrypted layer resides on a physical partition, though, it will not be possible to exploit the ability of LVM and RAID to span multiple disks.
- If the encrypted layer is created on top of LVM/RAID devices, it will still be possible to reorganize the file systems in the future, but with added complexity, since the encryption layers will have to be adjusted accordingly. Moreover, separate passphrases or keys will be required to open each encrypted device. This, however, is the only choice for systems that need encrypted file systems to span multiple disks.
Btrfs's built-in subvolumes feature can be used with dm-crypt, fully replacing the need for LVM if no other file systems are required. However, note that an encrypted swap is not possible this way and swap files are not supported by btrfs up to now.