Eric,

Excellent test with very useful results!

Modern security practice would indicate elimination of the cleartext
passwords.  How difficult would it be compile vpopmail without cleartext
passwords?

I propose two reasons for this path:
1)  The one of the biggest reasons to use qmail is it's security record.
 We should try as much a possible to make the accessories as secure as
qmail.  There are a lot of people who would not use QMT because storage
of cleartext passwords SOUNDS medieval and dangerous at this point,
regardless of whether or not it is.  My understanding is that DJB wrote
the qmail modules with each assuming that the others had been
compromised.  I think that the qmail accessories should be configured
using this same philosophy as much as possible.  Assuming that no
intruder could ever get in or get root access is a path to making lots
poor decisions which could eventually be regretted.

2) Using only the hash allows infinite size passwords.  Admittedly, 40
characters sounds adequate.  But that might be eventually outgrown and
we would have the same problem again.  But any size password will always
yield the same size hash.

One extra thought:

If vpopmail were to be patched or recompiled, one thought might be to
check what hash routine it uses.  If it is weaker than SHA256, it
probably ought to be updated to this or SHA512.  I've not looked at the
vpopmail source code so I have no idea if that is feasible.  But it
might be quite easy.

-Andy


On 10/2/2018 9:28 PM, Eric Broch wrote:
> The solution might be to either patch dovecot with our own QMT patch at
> compile time to avoid the clear text password altogether during
> authentication,
> 
> Or compile vpopmail clear text password field disabled,
> 
> Or, Another solution would be for users to clear all clear text password
> fields from the vpopmail database before migration,
> 
> Or as Tony brought up changing the size of the clear password field to
> 40 chars.
> 
> Any opinions?
> 
> 
> On 10/2/2018 11:21 PM, Eric Broch wrote:
>> Dovecot will authenticate against the clear text password if it is
>> present.
>>
>> Upon updating the clear text password (encrypted 17 characters) to
>> 'null', I authenticated using Dovecot against the 17 character password.
>>
>> Here's the command I used to set the clear text password to null:
>> mysql> update mydomain_tld set pw_clear_passwd='' where pw_name ='user';
>>
>> Then Dovecot authenticated fine against the 17 character
>> password...now encrypted to 40 chars.
>>
>>
>> On 10/2/2018 11:09 PM, Andrew Swartz wrote:
>>> On further review, your debug output looks pretty definitive for Dovecot
>>> directly accessing the database.  Given that the hash cannot be
>>> reversed, the only way to get the cleartext password is direct database
>>> access.
>>>
>>> This seems a pretty substantial problem, as it means that the hash and
>>> cleartext will be discordant for passwords >16 characters. But nothing
>>> stops users from choosing such passwords.
>>>
>>> Or alternately, could be an interesting bug to capitalize on. It allows
>>> creation of relay-only passwords.  I could use this for accounts which
>>> only send mail but which should never check it (like my UPS or system
>>> monitoring scripts).  From one perspective, this could be a security
>>> advantage.  But forcing real users to use small passwords is probably a
>>> much bigger disadvantage.
>>>
>>> -Andy
>>>
>>>
>>> On 10/2/2018 8:47 PM, Eric Broch wrote:
>>>> And when debugging authentication with Dovecot, I get...
>>>>
>>>> CLEARTEXT(password entered in webmail) != 'password in the database'
>>>>
>>>>  From dovecot.log the actual output:
>>>>
>>>> Oct 02 22:05:45 auth-worker(19953): Debug:
>>>> vpopmail(someu...@domain.tld,127.0.0.1,<dVsPJ0t3GIl/AAAB>):
>>>> CLEARTEXT(xxxxxxxxxxxxxxxxx) != 'xxxxxxxxxxxxxxxx'
>>>> Oct 02 22:05:47 auth: Debug: client passdb out: FAIL    1
>>>> user=someu...@domain.tld
>>>>
>>>>
>>>>
>>>>
>>>> On 10/2/2018 10:25 PM, Eric Broch wrote:
>>>>> So, I think Dovecot MAY be authenticating against the plain text
>>>>> password. I can't be sure until I look at the code or ask on the
>>>>> Dovecot mailing list.
>>>>>
>>>>>
>>>>> On 10/2/2018 10:22 PM, Eric Broch wrote:
>>>>>> Okay,
>>>>>>
>>>>>> 17 character password works with Submission port. Not with IMAP which
>>>>>> is authenticated through Dovecot.
>>>>>>
>>>>>> Eric
>>>>>>
>>>>>>
>>>>>> On 10/2/2018 9:21 PM, Andrew Swartz wrote:
>>>>>>> Eric,
>>>>>>>
>>>>>>> Regarding the hash: difficult to answer because of the atypical
>>>>>>> storage
>>>>>>> method (in the database).  It looks like two items (username and
>>>>>>> password???), each stored in an atypical base64 (using "."
>>>>>>> instead of
>>>>>>> "+" for the 64th character) and each prefixed with a "$" and then
>>>>>>> concatenated. Unfortunately, this makes it difficult to know what
>>>>>>> the
>>>>>>> hash "should" be.  Each of these could also be salted. Browsing the
>>>>>>> vpopmail source code would likely clear this up. Unsure when I'll
>>>>>>> have
>>>>>>> time for that.
>>>>>>>
>>>>>>> I was testing passwords using Squirrelmail, which goes through IMAP,
>>>>>>> which means that Dovecot does the authentication (I believe). It is
>>>>>>> possible that dovecot (Centos7) is authenticating differently
>>>>>>> than did
>>>>>>> courier-IMAP (Centos5).  There are two places in
>>>>>>> /etc/dovecot/toaster.conf which specify "driver = vpopmail". I
>>>>>>> have no
>>>>>>> idea what the detailed implications of that setting are.
>>>>>>>
>>>>>>> It would be interesting to see if the 16 or 17 character
>>>>>>> passwords work
>>>>>>> for qmail-smtp.  Could try to telnet to port 25 and see if qmail
>>>>>>> accepts
>>>>>>> the 16 or 17 character password for relay.  If qmail takes the 17
>>>>>>> character password and not the 16, it would indicate a different
>>>>>>> authentication method than via IMAP.  This would mean that the
>>>>>>> database
>>>>>>> is not the problem.
>>>>>>>
>>>>>>> Unfortunately, and not somewhere that allows me to try this right
>>>>>>> now.
>>>>>>>
>>>>>>> -Andy
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> On 10/2/2018 6:47 PM, Eric Broch wrote:
>>>>>>>> Okay,
>>>>>>>>
>>>>>>>> Set user's password to 17 x's, eg: xxxxxxxxxxxxxxxxx
>>>>>>>>
>>>>>>>> I could not log in with 17x password but I could with 16x password.
>>>>>>>>
>>>>>>>> Not sure what this means, I'm open to enlightenment. Could it be
>>>>>>>> the hash?
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> On 10/2/2018 8:41 PM, Eric Broch wrote:
>>>>>>>>> Will do.
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> On 10/2/2018 8:40 PM, Andrew Swartz wrote:
>>>>>>>>>> Eric,
>>>>>>>>>>
>>>>>>>>>> Before I do that, can you see if you can replicate the
>>>>>>>>>> problem: On
>>>>>>>>>> Centos7, create an account with a long password and see if you
>>>>>>>>>> can then
>>>>>>>>>> log in with the long password.  If that fails, then try with the
>>>>>>>>>> first
>>>>>>>>>> 16 characters of that password.
>>>>>>>>>>
>>>>>>>>>> -Andy
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> On 10/2/2018 6:28 PM, Eric Broch wrote:
>>>>>>>>>>> Andrew,
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> On 10/2/2018 7:34 PM, Andrew Swartz wrote:
>>>>>>>>>>>> 1.  vpopmail (or something else) is NOW authenticating
>>>>>>>>>>>> against the
>>>>>>>>>>>> cleartext password instead of the hash.
>>>>>>>>>>> I don't think so, or I hope not. I've done nothing except
>>>>>>>>>>> compile
>>>>>>>>>>> vpopmail on CentOS 7 back in 2015 no patches.
>>>>>>>>>>> The only change, if I remember correctly, is MariaDB
>>>>>>>>>>> requirements
>>>>>>>>>>> rather
>>>>>>>>>>> the MySQL.
>>>>>>>>>>>> 2.  vpopmail (or something else) is NOW truncating the password
>>>>>>>>>>>> at 16
>>>>>>>>>>>> characters when it is set (i.e. hashed), but not during
>>>>>>>>>>>> subsequent
>>>>>>>>>>>> authentication.
>>>>>>>>>>> I hope it's something else.
>>>>>>>>>>>> 3.  mysql was storing something in the cleartext password field
>>>>>>>>>>>> which it
>>>>>>>>>>>> did not export.  This seems unlikely, as I can see 16
>>>>>>>>>>>> characters
>>>>>>>>>>>> and the
>>>>>>>>>>>> field type is "char(16)".  I went through the database export
>>>>>>>>>>>> file,
>>>>>>>>>>>> and
>>>>>>>>>>>> its contents appear the same as those of the running mysql
>>>>>>>>>>>> database on
>>>>>>>>>>>> Centos5, which is the same as the running mariadb database on
>>>>>>>>>>>> Centos7 (I
>>>>>>>>>>>> view the contents with WebMin).  Therefore it appears that the
>>>>>>>>>>>> backup/restore worked properly.
>>>>>>>>>>> Maybe something worth my time: Bring up two qmail (w/vpopmail)
>>>>>>>>>>> VM's on
>>>>>>>>>>> COS5 and COS7.
>>>>>>>>>>> Next, Create a domain and user entry on COS5 with >16 length
>>>>>>>>>>> password.
>>>>>>>>>>> Dump the vpopmail db on COS5 (vpopmail-cos5db), and import it on
>>>>>>>>>>> COS7.
>>>>>>>>>>> Dump the vpopmail db on COS7 (vpopmail-cos7db), and compare
>>>>>>>>>>> (diff) the
>>>>>>>>>>> two dumps.
>>>>>>>>>>> If they're the same it could possibly be an issue with the
>>>>>>>>>>> vpopmail
>>>>>>>>>>> program.
>>>>>>>>>>>
>>>>>>>>>>> If you were up to it, you could also create a database called
>>>>>>>>>>> vpopmail1
>>>>>>>>>>> on your COS7 machine,
>>>>>>>>>>> and import the COS5 vpopmail db into it (that way it doesn't
>>>>>>>>>>> mess with
>>>>>>>>>>> your regular vpopmail db), and
>>>>>>>>>>> dump it and compare the two (COS5/COS7) dumps.
>>>>>>>>>>>> Does anyone know the details of how vpopmail interacts with the
>>>>>>>>>>>> database
>>>>>>>>>>>> server?  Or if any authentication is done by some means
>>>>>>>>>>>> other than
>>>>>>>>>>>> through vpopmail?
>>>>>>>>>>> Interaction with db by vpopmail is done at compile time.
>>>>>>>>>>>
>>
> 

Attachment: smime.p7s
Description: S/MIME Cryptographic Signature

Reply via email to