Re: [dm-crypt] Empty key files vs empty passwords in plain mode

[Quentin Lefebvre <qlefebvre_pro@yahoo.com>]

On 23/11/2014 14:16, Milan Broz wrote :
> On 11/23/2014 01:44 PM, Quentin Lefebvre wrote:
>
>>> I experienced some troubles recently with Debian's cryptsetup package
>>> (testing version), which version is 1.6.6.
>>> I found out that empty key files get refused by cryptsetup, for example:
>>> cat empty_file | cryptsetup --debug --key-file=- open --type plain
>>> /test1.loop test1
>>> gets rejected.
>>> The debug output directly leads to a test in utils_crypt.c that, I
>>> think, should be removed.
>>>
>>> Indeed, empty passwords are accepted, so it make sense to accept also
>>> empty inputs.
>>> Especially in Debian, where cryptdisks_start script calls:
>>> /lib/cryptsetup/askpass | cryptsetup --key-file=- open --type [type]
>>> [src] [dst]
>>>
>>> What do you think about this issue?
>>> Shall I send a patch for that?
>
> Well, logically it should be the same. But reading empty keyfile never worked AFAIK

Right, and this is just because of a test that returns an error code in 
case the key file is empty.

> and IMHO the case that you encrypt device by empty keyfile by mistake
> is more common...

I agree and I think there should be at least a warning.

> I am tempting to say it is a safety feature than bug :-)
>
> Anyway, please create issue on project page, https://code.google.com/p/cryptsetup/issues/list
> If you have a patch, attach it there as well.

Sure, I'll do that. But which tool is preferred to write a patch for 
cryptsetup?

> I will get to it but this is not really urgent issue to solve.

In my opinion, the problem is, for example, when a user enters a blank 
password (not secured, sure) and then reboot and can't mount his file 
system because of a script piped to cryptsetup.

> BTW I would better suggest that Debian uses pwquality library with some sane
> defaults and will not allow users to enter so weak passwords in the first place...
>
> (There is always --force-password switch so your issue is still kind of problem
> for testing though.)

OK.
Thanks for this answer.

Cheers,
Quentin