Re: kwallet crypto misuse

[Daniel Kahn Gillmor <dkg () fifthhorseman net>]

On 01/03/2014 01:13 PM, Simon McVittie wrote:
> My understanding was that kwallet is like gnome-keyring or the Firefox
> password store: it contains a large number of stored passwords, all
> encrypted with a (key derived from a) master password. (Terminology in
> this email is borrowed from Firefox, and might not match KWallet.)
> It's important to distinguish between the stored passwords and the
> master password.

yes, agreed.  thanks for proposing some clear terminology for the
conversation :)

so we have:

 * master password
 * master key (derived from master password)
 * stored passwords (and other data)

> Issue 1[1] described in that blog post: the *master* password is
> passed through a key derivation function (which does not need to be
> reversible) in order to turn it into an encryption key, but that KDF
> is not very good (hashing, but no salt), making it vulnerable to
> dictionary or brute-force attacks assisted by precomputation (rainbow
> tables), particularly if the password used is relatively weak. As far
> as I can see, MITRE has not allocated a CVE ID for this. (Or is it
> considered to be part of CVE-2013-7252?)

The dictionary or brute-force attacks you're talking about are often
used for people attempting to discover the master password from the
master key -- but that's not the threat we're talking about here,
because the master key isn't stored anywhere that an attacker can get to
it.  Furthermore, if an attacker *could* get the master key, then they
wouldn't care about the master password at all, because the master key
is sufficient to access the stored passwords.

So the threat you're describing is something subtly different: in this
case, the threat is that a pre-computed dictionary could be applied
against a kwallet data store to recover the stored passwords.  right?

This seems like a distinct flaw from the use of ECB, and i agree that i
don't see a particular CVE assigned for it.

> Issue 2 described in that blog post: when the stored passwords are
> encrypted using the key derived from the master password, KWallet uses
> ECB, which is bad at hiding patterns in data. For instance, if a
> password is stored more than once, an attacker can determine that this
> is likely to have been done, by noticing the corresponding pattern in
> the output. As far as I can see, this is now CVE-2013-7252.

yep, agreed.

> Issue 3 (which exacerbates issue 2 rather than being something
> separate, AIUI?) is that the encoding of the stored passwords is
> relatively low-entropy: if the stored password happens to be entirely
> Latin-1 (which is quite likely in practice), then it has a pattern,
> namely "odd-numbered bytes are zero".

if the password is in US-ASCII (also quite likely in practice for many
many people) then it also has a pattern, namely "every 8th bit is zero",
regardless of which character encoding scheme is used (there are
probably other patterns that we could come up with too).  I don't see
this as a vulnerability, though -- you should be able to feed structured
cleartext into a reasonable encryption mechanism without pre-whitening
it.  The problem is that ECB is not a reasonable mechanism.

        --dkg

Attachment: signature.asc
Description: OpenPGP digital signature