[dm-crypt] SHAx and LUKS/cryptsetup

[dm-crypt] SHAx and LUKS/cryptsetup

[Heinz Diehl]

Hi,

while experimenting with LUKS/dmcrypt, I prepared a partition
using "-h ripemd160 -c aes-xts-plain64". LuksDump shows:

Version:       	    1
Cipher name:   	    aes
Cipher mode:   	    xts-plain64
Hash spec:     	    ripemd160

However, lsmod shows that both sha1_generic and sha1_ssse3 modules are
loaded. So I did a reboot without touching the encrypted device, and
there was no SHAx module loaded. After accessing the encrypted drive
it was, hence this is clearly the cause that these modules get loaded.

My question: why are there any SHA1 modules loaded when then encrypted
drive is NOT using it?

Re: [dm-crypt] SHAx and LUKS/cryptsetup

[Milan Broz]

On 9.3.2014 18:42, Heinz Diehl wrote:
> Hi,
>
> while experimenting with LUKS/dmcrypt, I prepared a partition
> using "-h ripemd160 -c aes-xts-plain64". LuksDump shows:
>
> Version:       	    1
> Cipher name:   	    aes
> Cipher mode:   	    xts-plain64
> Hash spec:     	    ripemd160
>
> However, lsmod shows that both sha1_generic and sha1_ssse3 modules are
> loaded. So I did a reboot without touching the encrypted device, and
> there was no SHAx module loaded. After accessing the encrypted drive
> it was, hence this is clearly the cause that these modules get loaded.
>
> My question: why are there any SHA1 modules loaded when then encrypted
> drive is NOT using it?

Probably some dependence inside kernel, cryptsetup should not cause
loading of these modules (moreover it uses hash from userspace gcrypt
for LUKS header processing, not from kernel).

Probably trace which exact command caused it, maybe there is some sha
checksum somewhere triggering it. Isn't it filesystem on top loading it?

How exactly do you activate device and "touch" it?

Milan

p.s.
If you are using kernel backend (not gcrypt one), sha1 is used as test
that interface works. It was the simple way at that time. So in this case
it is cryptsetup causing it ;-) Maybe it could by changed now by some more
clever test.

Re: [dm-crypt] SHAx and LUKS/cryptsetup

[Heinz Diehl]

On 09.03.2014, Milan Broz wrote: 

> If you are using kernel backend (not gcrypt one)

I do :-)

> sha1 is used as test that interface works.

Ok, all good! So this is it. Thanks a lot!

Re: [dm-crypt] SHAx and LUKS/cryptsetup

[Milan Broz]

On 9.3.2014 19:32, Heinz Diehl wrote:
> On 09.03.2014, Milan Broz wrote:
>
>> If you are using kernel backend (not gcrypt one)
>
> I do :-)
>
>> sha1 is used as test that interface works.
>
> Ok, all good! So this is it. Thanks a lot!

Just to clarity it little bit:

Kernel userspace crypto API was (and still is)
quite undocumented, and testing SHA1 (which is mandatory
for LUKS backend support) was the simplest way how
to verify kernel backend works reliably.
(In some kernel versions it was impossible to check if just algorithm
is missing or the whole kernel socket interface is not available.)

It actually does not compute any sha1 hash, it just tries
to initialize it.

BTW I found some problems with kernel backend so use with care.

One problem is e.g. backend cannot use longer
key for HMAC than 20480 bytes (at least on my 32bit VM),
which can cause problems for larger keyfiles in PBKDF2.

I have workaround for this but will need some time to finish
it (I do not want to touch internal PBKDF2 without adding test
vectors and other tests.)

Milan

Re: [dm-crypt] SHAx and LUKS/cryptsetup

[Arno Wagner]

On Sun, Mar 09, 2014 at 20:53:02 CET, Milan Broz wrote:
> On 9.3.2014 19:32, Heinz Diehl wrote:
> >On 09.03.2014, Milan Broz wrote:
> >
> >>If you are using kernel backend (not gcrypt one)
> >
> >I do :-)
> >
> >>sha1 is used as test that interface works.
> >
> >Ok, all good! So this is it. Thanks a lot!
> 
> Just to clarity it little bit:
> 
> Kernel userspace crypto API was (and still is) quite undocumented, 

This is one thing I really do not understand. Doing crypto
right is already hard. With bad documentation it gets worse.
Yet the documentation for kernel, OpenSSL, commercial
libraries I have looked at, Java API, etc. is bad. (for Java 
so bad that recently 30'000 Apps on Android were insecure).

I really do not get it. Systematic sabotage of the documentation
seems unlikely, even after Snowden, so I can only conclude
many people implementing crypto have a problem writing 
documentation.

> and testing SHA1 (which is mandatory
> for LUKS backend support) was the simplest way how
> to verify kernel backend works reliably.
> (In some kernel versions it was impossible to check if just algorithm
> is missing or the whole kernel socket interface is not available.)
> 
> It actually does not compute any sha1 hash, it just tries
> to initialize it.
> 
> BTW I found some problems with kernel backend so use with care.

Bad documentation and unreliable. Urgh.
 
> One problem is e.g. backend cannot use longer
> key for HMAC than 20480 bytes (at least on my 32bit VM),
> which can cause problems for larger keyfiles in PBKDF2.
> 
> I have workaround for this but will need some time to finish
> it (I do not want to touch internal PBKDF2 without adding test
> vectors and other tests.)

Very sensible. I completely support this approach.

Arno

> 
> Milan
> _______________________________________________
> dm-crypt mailing list
> dm-crypt@saout.de
> http://www.saout.de/mailman/listinfo/dm-crypt

-- 
Arno Wagner,     Dr. sc. techn., Dipl. Inform.,    Email: arno@wagner.name
GnuPG: ID: CB5D9718  FP: 12D6 C03B 1B30 33BB 13CF  B774 E35C 5FA1 CB5D 9718
----
A good decision is based on knowledge and not on numbers. -  Plato