Re: Should Roundup replace current password hash with a new one.

Hi Thomas:

In message <[email protected]>,
Thomas Arendsen Hein writes:
>* John P. Rouillard <[email protected]> [20230228 23:38]:
>> work-factor which increases the number of rounds used
>> when hashing the stored password. For a long time that
>> factor was 10000.
>
>> The 2.3.0 release will increase it to 2,000,000 and admins
>> will have to make the change when they upgrade.
>> However PBKDF2 can be used with SHA512 rather than
>> SHA1. The number of rounds (work-factor) needed for that
>> is less, but more computation is required for SHA512
>> versus SHA1.
>
>How long will logins take with this increased settings on a
>hardware that is a bit older or less powerful in general
>(does anyone use Roundup on Raspberry Pis?)

I happen to use a Rasp-Pi 2B as a trial/demo server for
clients. It has an ARMv6-compatible processor rev 7 (v6l).

With 10000 rounds a verifyPassword takes ~ .2 seconds With
2000000 rounds it takes 30-40 seconds. So yes, it's a major
issue.  It is roughly linear though. 2,000,000/10,000 is 200
and .2s * 200 = 40s.

>100000 rounds with SHA1 on an AMD Opteron 6128 are fine, we
>use this in Moin wikis, I'm a bit scared to try 2000000
>though :)

Understandable. If you know the time to calculate 100,000
rounds multiply by 20 to see what 2,000,000 will be. You
can at least calculate the expected login time.

On my underpowered 5 year old desktop runs an Intel(R)
Core(TM) i3-7100 CPU @ 3.90GHz. 2,000,000 rounds of PBKDF2
with *SHA512* takes 1.3+/-0.03 seconds.  With 300,000 rounds
of PBKDF2-SHA512 takes 0.2+/-0.02 seconds.

The ratio of 2,000,000/300,000 is 6.6 and 0.2sx6.6=1.3s. So
this looks pretty predictable. I expect using SHA1 rather
than SHA512 to be even faster.

>> I would also like to deprecate SSHA (salted sha-1) hashes
>> at the same time. IIRC it was a heavily used hashing
>> method for LDAP passwords. I don't know if anybody is
>> using SSHA these days.
>
>SSHA is still used in some places, but regarding Roundup
>deprecating it should be fine.

Thanks for the confirmation.

I am auto-upgrading the 10,000 value to 2,000,000 at this
point.  There is a section in the upgrading docs about this
change.

But maybe I should disable the auto-upgrade. I doubt people
are deploying in production using a rasp-pi but...

Thoughts?
 
--
				-- rouilj
John Rouillard
===========================================================================
My employers don't acknowledge my existence much less my opinions.