dm-crypt/System configuration

From ArchWiki
Jump to: navigation, search

Tango-view-fullscreen.pngThis article or section needs expansion.Tango-view-fullscreen.png

Reason: Aggregate here all the generic information on system configuration from the other sub-articles of Dm-crypt. (Discuss in Talk:Dm-crypt/System configuration#)

Back to Dm-crypt.

Tip: If in need to remotely unlock root or other early-boot filesystems (headless machine, distant servers...), follow the specific instructions from Dm-crypt/Specialties#Remote unlocking of encrypted root.

mkinitcpio

When encrypting a system it is necessary to regenerate the initial ramdisk after properly configuring mkinitcpio. Depending on the particular scenarios, a subset of the following hooks will have to be enabled:

  • encrypt: always needed when encrypting the root partition, or a partition that needs to be mounted before root. It is not needed in all the other cases, as system initialization scripts like /etc/crypttab take care of unlocking other encrypted partitions. This hook must be placed after the udev hook, if that is used.
  • shutdown: deprecated, not necessary anymore.
  • keymap: provides support for foreign keymaps for typing encryption passwords; it must come before the encrypt hook. Setting your keymap is done in /etc/vconsole.conf.
  • keyboard: needed to make USB keyboards work in early userspace.

Other hooks needed should be clear from other manual steps followed during the installation of the system.

Boot loader

In order to enable booting an encrypted root partition, a subset of the following kernel parameters need to be set: see kernel parameters for instructions specific to your boot loader.

For example using GRUB the relevant parameters are best added to /etc/default/grub before generating the boot configuration. See also GRUB#Warning when installing in chroot as another point to be aware of when installing the GRUB loader.

Tango-view-fullscreen.pngThis article or section needs expansion.Tango-view-fullscreen.png

Reason: Users of the sd-encrypt hook require other kernel parameters. A subsection for it should be added. (FS#36265 is resolved, so the limitations it posed are gone). (Discuss in Talk:Dm-crypt/System configuration#)

cryptdevice

This parameter will make the system prompt for the passphrase to unlock the device containing the encrypted root on a cold boot. It is parsed by the encrypt hook to identify which device contains the encrypted system:

cryptdevice=device:dmname
  • device is the path to the device backing the encrypted device. Usage of Persistent block device naming is advisable.
  • dmname is the device-mapper name given to the device after decryption, which will be available as /dev/mapper/dmname.
  • If a LVM contains the encrypted root, the LVM gets activated first and the volume group containing the logical volume of the encrypted root serves as device. It is then followed by the respective volume group to be mapped to root. The parameter follows the form of cryptdevice=/dev/vgname/lvname:dmname.
Note: When using systemd & sd-encrypt hooks, use luks.uuid instead of cryptdevice, see systemd-cryptsetup-generator(8).

root

The root= parameter specifies the device of the actual (decrypted) root file system:

root=device
  • If the file system is formatted directly on the decrypted device file this will be /dev/mapper/dmname.
  • If a LVM gets activated first and contains an encrypted logical rootvolume, the above form applies as well.
  • If the root file system is contained in a logical volume of a fully encrypted LVM, the device mapper for it will be in the general form of root=/dev/mapper/volumegroup-logicalvolume.
Tip: This parameter is not needed to be specified manually when using GRUB. Executing grub-mkconfig is meant to determine the correct UUID of the decrypted root filesystem and specify it in the generated grub.cfg automatically.

resume

resume=device
  • device is the device file of the decrypted (swap) filesystem used for suspend2disk. If swap is on a separate partition, it will be in the form of /dev/mapper/swap. See also Dm-crypt/Swap encryption.

cryptkey

This parameter is required by the encrypt hook for reading a keyfile to unlock the cryptdevice. It can have three parameter sets, depending on whether the keyfile exists as a file in a particular device, a bitstream starting on a specific location, or a file in the initramfs.

For a file in a device the format is:

cryptkey=device:fstype:path
  • device is the raw block device where the key exists.
  • fstype is the filesystem type of device (or auto).
  • path is the absolute path of the keyfile within the device.

Example: cryptkey=/dev/usbstick:vfat:/secretkey

For a bitstream on a device the key's location is specified with the following:

cryptkey=device:offset:size 

where the offset is in bytes and the size in bits. Example: cryptkey=/dev/sdZ:0:512 reads a 512 bit keyfile starting at the beginning of the device.

For a file included in the initramfs the format is[1]:

cryptkey=rootfs:path

Example: cryptkey=rootfs:/secretkey

Also note that if cryptkey is not specified, it defaults to /crypto_keyfile.bin (in the initramfs).[2]

See also Dm-crypt/Device encryption#Keyfiles.

crypto

This parameter is specific to pass dm-crypt plain mode options to the encrypt hook.

It takes the form

crypto=<hash>:<cipher>:<keysize>:<offset>:<skip>

The arguments relate directly to the cryptsetup options. See Dm-crypt/Device encryption#Encryption options for plain mode.

For a disk encrypted with just plain default options, the crypto arguments must be specified, but each entry can be left blank:

crypto=::::

A specific example of arguments is

crypto=sha512:twofish-xts-plain64:512:0:

crypttab

The /etc/crypttab (encrypted device table) file is similar to the fstab file and contains a list of encrypted devices to be unlocked during system boot up. This file can be used for automatically mounting encrypted swap devices or secondary file systems.

crypttab is read before fstab, so that dm-crypt containers can be unlocked before the file system inside is mounted. Note that crypttab is read after the system has booted up, therefore it is not a replacement for unlocking encrypted partitions by using mkinitcpio hooks and boot loader options as in the case of encrypting the root partition. crypttab processing at boot time is made by the systemd-cryptsetup-generator automatically.

See the crypttab man page for details, read below for some examples, and the #Mounting at boot time section for instructions on how to use UUIDs to mount an encrypted device.

Warning: There are issues with systemd when processing crypttab entries for dm-crypt plain mode (--type plain) devices:
  • For --type plain devices with a keyfile, it is necessary to add the hash=plain option to crypttab due to a systemd incompatibility. Do not use systemd-cryptsetup manually for device creation to work around it.
  • It may be further required to add the plain option explicitly to force systemd-cryptsetup to recognize a --type plain) device at boot. GitHub issue in question.
/etc/crypttab
 # Example crypttab file. Fields are: name, underlying device, passphrase, cryptsetup options.

 # Mount /dev/lvm/swap re-encrypting it with a fresh key each reboot
 swap	/dev/lvm/swap	/dev/urandom	swap,cipher=aes-xts-plain64,size=256

 # Mount /dev/lvm/tmp as /dev/mapper/tmp using plain dm-crypt with a random passphrase, making its contents unrecoverable after it is dismounted.
 tmp	/dev/lvm/tmp	/dev/urandom	tmp,cipher=aes-xts-plain64,size=256 

 # Mount /dev/lvm/home as /dev/mapper/home using LUKS, and prompt for the passphrase at boot time.
 home   /dev/lvm/home

 # Mount /dev/sdb1 as /dev/mapper/backup using LUKS, with a passphrase stored in a file.
 backup /dev/sdb1       /home/alice/backup.key

Mounting at boot time

If you want to mount an encrypted drive at boot time, enter the device's UUID in /etc/crypttab. You get the UUID (partition) by using the command lsblk -f and adding it to crypttab in the form:

/etc/crypttab
 externaldrive         UUID=2f9a8428-ac69-478a-88a2-4aa458565431        none    luks,timeout=180

The first parameter is your preferred device mapper's name for the encrypted drive. The option none will trigger a prompt during boot to type the passphrase for unlocking the partition. The timeout option defines a timeout in seconds for entering the decryption password during boot.

A keyfile can also be set up and referenced instead of none. This results in an automatic unlocking, if the keyfile is accessible during boot. Since LUKS offers the option to have multiple keys, the chosen option can also be changed later.

Use the device mapper's name you've defined in /etc/crypttab in /etc/fstab as follows:

/etc/fstab
 /dev/mapper/externaldrive      /mnt/backup               ext4    defaults,errors=remount-ro  0  2

Since /dev/mapper/externaldrive already is the result of a unique partition mapping, there is no need to specify an UUID for it. In any case, the mapper with the filesystem will have a different UUID than the partition it is encrypted in.

Mounting a stacked blockdevice

The systemd generators also automatically process stacked block devices at boot.

For example, you can create a RAID setup, use cryptsetup on it and create an LVM logical volume with respective filesystem inside the encrypted block device. A resulting:

$ lsblk -f
─sdXX                  linux_raid_member    
│ └─md0                 crypto_LUKS   
│   └─cryptedbackup     LVM2_member 
│     └─vgraid-lvraid   ext4              /mnt/backup
└─sdYY                  linux_raid_member    
  └─md0                 crypto_LUKS       
    └─cryptedbackup     LVM2_member 
      └─vgraid-lvraid   ext4              /mnt/backup

will ask for the passphrase and mount automatically at boot.

Given you specify the correct corresponding crypttab (e.g. UUID for the crypto_LUKS device) and fstab (/dev/mapper/vgraid-lvraid) entries, there is no need to add additional mkinitcpio hooks/configuration, because /etc/crypttab processing applies to non-root mounts only. One exception is when the mdadm_udev hook is used already (e.g. for the root device). In this case /etc/madadm.conf and the initramfs need updating to achieve the correct root raid is picked first.