Mathieu Othacehe <[email protected]> writes:

> Hello,
>
>> Christine Lemmer-Webber <[email protected]> writes:
>>
>>> - It's painful with full disk encryption even then, because you have to
>>>   type your passphrase twice.
>>
>> This is not the case for quite some time now.  Sure, it requires some
>> configuration, but it can be done.  For all my machines I am typing my
>> passphrase just once.  Sure, it is the first one, so it takes *long*
>> time, though since it is just one passphrase, I usually just make a
>> coffee or something while waiting.
>
> I just performed a Guix System 1.5.0 installation with full-disk
> encryption in a VM. On the installed system, the password needs to be
> typed twice: in GRUB and in the initramfs.

I suppose no one adjusted the installer.

> Maybe for the next release, we should aim for an installer that creates
> a configuration where only one password input is necessary. The
> extra-initrd proposal feels a bit hacky to be the one proposed by the
> installer.

As original author of the extra-initrd approach I am obviously somewhat
biased, but it seems... fine?  It is based on the recommendations from
the Archlinux wiki, adjusted for our needs.

> Two alternatives come to my mind:
>
> 1. Make sure that all the kernels/initramfs of the live generations have
> a copy in /boot.

My understanding is that you need to enter the password twice due to:

1. GRUB needs to access its configuration                <-- Password #1
2. GRUB shows the menu and starts the boot process
3. The initrd is loaded
4. The initrd needs to pivot to the real root            <-- Password #2

So I admit I am unsure what having a copy directly in /boot solves.
GRUB already has access to /gnu/store after you unlock the root for the
first time.

> 2. Have the store in a dedicated, unencrypted partition.

This seems unwise.  The files in the store are not authenticated in any
way, so I would prefer having them encrypted.

> Any other alternatives :) ?

Not really :)

-- 
There are only two hard things in Computer Science:
cache invalidation, naming things and off-by-one errors.

Reply via email to