The crypt-ssh dracut module allows remote unlocking of systems with full disk encryption via ssh.
There are a number of reasons why you would want to do this:
- It provides a way of entering encryption keys for a number of servers without console switching
- It allows booting of remote or co-located encrypted servers without console access
Users are strictly authenticated using their SSH public keys. These can be either:
/root/.ssh/authorized_keys or a custom file (dropbear_acl option). Depending
on your environment, it may make sense to make the preboot authorized_keys file
different from the normal one.
Plain text password authentication and port forwarding are disabled.
We recommend installation via packages, or build manually if unavailable. You are encouraged to get in contact with us if you are able to provide packages for other distributions.
We currently provide packages for RHEL+CentOS 7 as well as Fedora 22+ via a COPR repository.
Installation via dnf (Fedora):
# dnf copr enable rbu/dracut-crypt-ssh
# dnf install dracut-crypt-ssh
Installation via yum (CentOS 7):
# yum install epel-release
# wget -O /etc/yum.repos.d/rbu-dracut-crypt-ssh-epel-7.repo https://copr.fedorainfracloud.org/coprs/rbu/dracut-crypt-ssh/repo/epel-7/rbu-dracut-crypt-ssh-epel-7.repo
# yum install dracut-crypt-ssh
Install build dependencies (example commands for RPM systems). When building
on RHEL, be sure to enable EPEL (yum install epel-release).
# yum install dropbear dracut dracut-network openssh
# yum install libblkid-devel gcc
Retrieve a copy the source and build:
$ git clone https://github.com/dracut-crypt-ssh/dracut-crypt-ssh.git
$ cd dracut-crypt-ssh
$ ./configure
$ make
$ sudo make install
After the first installation and every time you update the dracut-crypt-ssh configuration,
it is required to rebuild the initramfs:
# dracut --force
Since you need network access before booting from the root partition, you need to configure this via GRUB, if you have not done so before.
The minimal change is enabling DHCP. Edit the /etc/default/grub file and append
to the end of the GRUB_CMDLINE_LINUX:
/etc/default/grub:
...
GRUB_CMDLINE_LINUX="... rd.neednet=1 ip=dhcp"
...
Afterwards, regenerate your GRUB config:
# grub2-mkconfig --output /etc/grub2.cfg
For a static network configuration, use something along the lines of
GRUB_CMDLINE_LINUX="... rd.neednet=1 ip=192.168.0.100::192.168.0.1:255.255.255.0:centos:enp0s8:off"
Refer to the network documentation of dracut
for more options (man dracut.cmdline).
When rebooting the system, dropbear sshd is started by the initramfs. You should be able to login and unlock the volumes:
% ssh -p 222 [email protected]
# console_peek # to see what's on the console
# console_auth # asks you for the passphrase and sends it to console
Passphrase:
If unlocking the device succeeded, the initramfs will clean up itself and dropbear terminates itself and your connection.
The unlock binary reads a passphrase from stdin, parses /etc/crypttab
and attempts to call cryptsetup luksOpen on all luks-encrypted drives that
don't have a keyfile, passing the passphrase that unlock got in stdin to luksOpen.
What this means in practice is you can do:
% ssh [email protected] -p 222 unlock < passwordFile
or:
% gpg -d password.gpg | ssh [email protected] -p 222 unlock
If you want to only unlock specific drives / LUKS volumes, you can provide wildcards on the command line, e.g.
% ssh [email protected] -p 222 unlock luks-3467c luks-34c13
unlock will search the crypttab for mapper names (first column in
/etc/crypttab) that start with the listed names. Volumes that match via this
method may have a keyfile listed in /etc/crypttab, it will be assumed that you
want to unlock the volume/s with an alternative key.
Note that the names provided are really wildcards, and by convention/default
all mappers start with luks-, so you can force unlock to try all drives simply
by doing something like 'unlock luks-'.
In all cases, unlock will only consider the process a success only if all
eligible volumes are unlocked successfully. This means:
- All the associated devices must be available at boot / unlock time
- The passphrase must be accepted for all eligible volumes
- cryptsetup luksOpen should not exit for any other reason.
In short, if you have more than one volume in /etc/crypttab, you will need to
be careful about how use this tool.
If the process is successful, unlock will launch the script
/sbin/unlock-reap-success. This will attempt to kill systemd-cryptsetup, and
failing that, attempt to kill cryptroot-ask. On RHEL6 & 7, this aborts the
builtin decrypt password request processes and allows the boot process to
proceed. Note that the plymouth splash screen on RHEL6 (if you happen to be
watching the console...) will still appear to ask for your password, but this
is an artifact. Disable plymouth (rhgb command line) if this annoys you.
You might want to limit access only to the unlock binary, just add command="unlock" to your authorized_keys before the key, e.g.
command="unlock" ssh-rsa .....
The configuration is stored in the crypt-ssh.conf, usually located in /etc/dracut.conf.d/.
The following options are available (see the config file for detailed description):
dropbear_port(default:222) - port ssh daemon should listen ondropbear_rsa_key,dropbear_ecdsa_key(default:GENERATE) - Source of the keys, possible options:SYSTEM- copy the private keys from the encrypted system (not recommended)GENERATE- generate a new keys (during the creation of initramfs)- path - key file in openssh format to use (a public file with '.pub' ending should be present too)
dropbear_acl(default:/root/.ssh/authorized_keys) - Keys which allowed to login into initramfs
After any configuration change, you have to rebuild the initramfs as the configuration takes effect during the building the initramfs.
The integrity, confidentiality and authenticity of your encrypted data relies on the physical integrity of your device. If someone else has access to the device that you are unlocking, it is entirely possible to replace the executable files handling your key material, or steal your initramfs's SSH private keys. Arguably, this kind of attack is possible without the "crypt-ssh" module, but using automated or remote access could make such an attack easier to conceal.
If this is a concern for you, consider keeping your devices offline and on your person.
If this is not a concern for you, i.e. you place a certain amount of trust in your hosting provider or physical integrity, this tool might still protect against accidental data leaks (i.e. VM deprovisioning, replaced hard disks).
If (or rather "when") something goes wrong and you can't access just-booted machine over network and can't get to console (hence sshd in initramfs), don't panic - it's fixable if machine can be rebooted into some rescue system remotely.
Usually it's some dhcp+tftp netboot thing from co-located machine (good idea to setup/test in advance) plus whoever is there occasionally pushing the power button, or maybe some fancy hw/interface for that (e.g. hetzner "rescue" interface).
To see what was going on during initramfs, open "modules.d/99base/rdsosreport.sh" in dracut, append this (to the end):
set -x
netstat -lnp
netstat -np
netstat -s
netstat -i
ip addr
ip ro
set +x
exec >/dev/null 2>&1
mkdir /tmp/myboot
mount /dev/sda2 /tmp/myboot
cp /run/initramfs/rdsosreport.txt /tmp/myboot/
umount /tmp/myboot
rmdir /tmp/myboot
Be sure to replace /dev/sda2 with whatever device is used for /boot, rebuild
dracut and add rd.debug to cmdline (e.g. in grub.cfg's "linux" line).
Upon next reboot, wait for at least a minute, since dracut should give up on trying to boot the system first, then it will store full log of all the stuff modules run ("set -x") and their output in "/boot/rdsosreport.txt".
Naturally, to access that, +1 reboot into some "rescue" system might be needed.
In case of network-related issues - e.g. if "rdsosreport.txt" file gets created
with "rd.debug", but host can't be pinged/connected-to for whatever reason -
either enable "debug" dracut module or add dracut_install netstat ip line to
install() section of "modules.d/60dropbear-sshd/module-setup.sh" and check
"rdsosreport.txt" or console output for whatever netstat + ip commands above
(for "rdsosreport.sh") show - there can be no default route, whatever interface
naming mixup, no traffic (e.g. unrelated connection issue), etc.