Skip to content

Safe Boot subcommands

Usage: safeboot subcommand [options...]

key-init

Usage:

safeboot key-init "subject"

Generate a new x509 signing key with openssl, writing storing the public key in /etc/safeboot/cert.pem and the password protected private key in /etc/safeboot/signing.key. This is not as secure as storing it in a hardware token, although if the key is moved to an external device and the $KEY variable in /etc/safeboot/local.conf updated to point to it, then it will prevent a software-only attack.

The subject must be written as a "distinguished name":

       /CN=host.example.com/OU=test/O=example.com/

yubikey-init

Usage:

safeboot yubikey-init "subject"

Generate a new x509 signing key on an attached Yubikey device and set the certificate subject to the provided argument. The public key certificate will be written to /etc/safeboot/cert.pem and will also be used for the UEFI SecureBoot variables.

Due to a limitation in the OpenSSL PKCS11 engine, you have to enter the password multiple times (once for the engine, and then once for the signature).

The subject must be written as a "distinguished name":

       /CN=host.example.com/OU=test/O=example.com/

yubikey-pubkey

Usage:

safeboot yubikey-pubkey cert.pem

Extract the public key certificate in PEM, DER and PUB format. The sbsign tool wants PEM, the kmodsign tool wants DER, the tpm2-tools wants a raw public key. The best part about standards...

uefi-sign-keys

Usage:

safeboot uefi-sign-keys

Create three signed certificates for the PK, KEK, and db using the attached Yubikey or x509 key stored in /etc/safeboot/signing.key and store them in the UEFI SecureBoot configuration. You should have run safeboot yubikey-init or safeboot key-init to have already generated the keys.

Due to an issue with the OpenSSL PKCS11 engine, you will have to authenticate to the Yubikey multiple times during this process.

uefi-set-keys

Usage:

safeboot uefi-set-keys

Store the PK, KEK, and db into the UEFI Secure Boot configuration variables. This must be done once during system setup or if a new key is generated. The uefi-sign-key subcommand attempts to do this automatically.

pcrs-sign

Usage: safeboot pcrs-sign [prevent-rollback] [path-to-unified-kernel]

Generate a signature for the PCRs that can be used to unseal the LUKS key according to the policy created by safeboot luks-seal. The PCRs used are specified in the /etc/safeboot/safeboot.conf or local.conf files, and must match the values that were configured during luks-seal.

If the prevent-rollback argument is prevent-rollback, the TPM version counter will be incremented, which will invalidate all previous PCR signatures and prevent the older unified kernel images from being able to unseal the PCR data.

The signature is persisted in a UEFI NVRAM variable, defined in safeboot.conf.

luks-seal

Usage:

safeboot luks-seal

This will generate a new LUKS encryption key for the block device in /etc/crypttab and requires an existing recovery key to install the new key slot. You will also be prompted for an unlock PIN, which will be required on the next normal boot in place of the recovery code.

If this is the first time the disk has been sealed, /etc/crypttab will be updated to include a call to the unsealing script to retrieve the keys from the TPM, and a counter will be created to prevent rollbacks.

After sealing the secret, the initrd will be rebuild, the kernel signed, and the new predicted PCRs signed. Any previous sealed data will be invalidated since the version counter will be incremented.

Right now only a single crypt disk is supported.

sign

Usage:

safeboot sign exe [exe.signed]

Sign an EFI executable with the safeboot keys.

This is the raw command; you might want to use safeboot linux-sign or safeboot recovery-sign instead.

unify-kernel

Usage:

safeboot unify-kernel linux.efi kernel=path-to-kernel initrd=path-to-initrd ...

Creates a unified kernel image with the named sections and files (typically kernel, initrd, cmdline, and osrel) bundled into an EFI executable.

This is the raw command; you might want to use safeboot linux-sign or safeboot recovery-sign instead to add the EFI boot manager entry.

sign-kernel

Usage:

safeboot sign-kernel linux.efi [linux.signed.efi]

Sign a unified EFI executable with the safeboot keys. If no destination is specified it will be the same name as the input kernel with .signed.efi added.

This is the raw command; you might want to use safeboot linux-sign or safeboot recovery-sign instead to add the EFI boot manager entry.

install-kernel

Usage:

safeboot install-kernel boot-name [extra kernel parameters...]

Create an EFI boot menu entry for boot-name, with the specified kernel, initrd and command line bundled into an executable and signed. This command requires the Yubikey or x509 password to be able to sign the merged EFI executable.

This is the raw command; you might want to use safeboot linux-sign or safeboot recovery-sign instead.

linux-sign

Usage:

safeboot linux-sign [target-name [parameters...]]

Generate dm-verity hashes and then sign the Linux with the root hash added to the kernel command line. The default target for the EFI boot manager is linux. You will need the Yubikey or x509 password to sign the new hashes and kernel.

If the environment variable $HASH is set to the hash value, or if the $HASHFILE variable points to the previous dmverity log (typically /boot/efi/EFI/linux/verity.log), then the precomputed value will be used instead of recomputing the dmverity hashes (which can take some time). If the hashes are out-of-date, this might render the linux target unbootable and require a recovery reboot to re-hash the root filesystem.

recovery-sign

Usage:

safeboot recovery-sign [kernel command line...]

Sign the Linux kernel and initrd into the EFI boot manager recovery entry. Typically this only needs to be done once and after validating that the system can boot with it, you should not have to re-run this command.

You will need the Yubikey or x509 password as well as root accesss to perform this action.

If SIP is enabled the root device will be marked read-only for the reboot and fscked will not been run on boot.

Use safeboot remount to remount / as read-write when in recovery mode, and then safeboot remount ro to restore it to read-only mode before signing the hashes.

recovery-reboot

Usage:

safeboot recovery-reboot

Configure the EFI boot manager so that the BootNext is the recovery target and reboot the machine. This command requires root access to update the EFI variables and will also require the disk encryption recovery key since the TPM will not unseal the disk automatically for recovery mode.

NOTE! This will reboot the machine!

bootnext

Usage:

safeboot bootnext Setup

Configure the EFI boot manager BootNext variable to select an alternate boot menu item. This command requires root access to update the EFI variables.

remount

Usage:

safeboot remount [ro]

Attempt to remount the root filesystem read/write or read-only. If SIP is enabled this will likely invalidate any hashes and require a re-signing of the root filesystem.

If ro is specified, then the file system will be re-mounted read-only If there are processes blocking the remount, they will be listed.

sip-init

Usage:

safeboot sip-init [home-size-in-GB [var-size]]

DANGER! This command can mess up your root filesystem. There must be space in the volume group for the new entries It will create the volume groups for /var and /home, add entries to /etc/fstab for them with secure mount parameters, and makes /tmp a symlink into /var/tmp.

unlock

Usage:

safeboot unlock

This is a recovery shell command to scan the /etc/crypttab for devices and call cryptsetup luksOpen on each of them, and then scan the LVM groups for volumes. After it succeeds you can call safeboot mount to mount the root filesystem (read-only) on /root.

mount-all

Usage:

safeboot mount-all

This is a recovery shell command to attempt to mount the root disk read-only on /root, as well as the /boot and /boot/efi if they exist in /root/etc/fstab.


Last update: July 23, 2021