[Cryptography] A TRNG review per day (week?): ATSHA204A has low entropy

[Bill Cox]

I just did another simple test.  After cold-booting my Raspberry Pi, I
called "./haslet random" and recorded the 32-byte result.  I had the
update-seed parameter turned off.  I did this 4 times.  Then I generated
about 800 values in a loop, which resulted in only 161 unique 32-byte
values.  All 4 cold boot values were in the list of 161 unique values.

I then generated 5326 64-byte values, of which 1171 were unique.  Looking
at just the first 32 bytes, there were 171 unique values.  Looking at just
the second 32-byte values, there were 1171 unique values.

I then generated 5348 96 byte values, of which 3217 were unique.  The first
group had the same 171 unique values, and the second had 1179.  The 3rd
group of 32 had 3217 unique values.

My best guess as to what's going on here is that the device has a
ring-oscillator based entropy source, but that it generates only a few bits
of entropy for each use.  It seems to be called before generating each
32-byte "random" value, which is why the second set of 32-bit values have
more possible values, and the 3rd has even more.  However, the number of
unique values in the final column of 32*N byte values is always equal to
the number of unique values of the entire string of bytes.

In the most common use case, where a Linkys router or some other embedded
device calls the random function only a few times ever, then Atmel will be
able to easily guess the values generated, since they are the ones who
created the initial seeds.  For devices that always publish a public key
somewhere after using this random function to generate the secret, Atmel
would always be able to determine the private key.  The only way I see to
generate a secure key would be to disable the EEPROM and combine enough
entropy from the random function results, then whiten it, as I described in
my prior post.

Since this device does not seem to get any signals such as a clock from the
Raspberry Pi, yet it returns different values after a cold boot on the
first call to random, I think there is strong evidence that some sort of
entropy generator does exist on the part.  However, with that EEPROM, it
could still just be a PRNG.  There is no way to be sure, and Atmel does not
seem very forthcoming about what they actually put inside the part.

This remains the scariest TRNG I have reviewed!

Bill
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.metzdowd.com/pipermail/cryptography/attachments/20141209/060fd563/attachment.html>