[Ejbca-develop] Suggestions around random serial number generation

[Jaime H. <hab...@gm...>]

First, I wanted to clarify that with EJBCA <7.0.1, the default 8 bytes
configuration for serial numbers generation wasn't delivering 63 bits of
entropy (as being declared) mainly because of the implementation filtering
out values that would produce a DER encoding in less than 8 bytes, so the
real entropy being provided would be around (or less than):

log2(0x7FFFFFFFFFFFFFFF - 0x80000000000000 + 1) = 62.99435343685886

Now, I would like to suggest to improve the documentation at
https://www.ejbca.org/docs/CA_Fields.html#src-42205242_id-.CAFieldsv7.0.1-Ca_Serial_Number_Octet_Size,
modifying it to something like this:

Sets the length in octets of certificate serial numbers generated, but note
> that the length of the serial number IS NOT the same that the entropy it
> contains (e.g. 8 octets or 64 bit serial number doesn't provide 64 bits of
> entropy) because some values are being filtered out:
> - Negative values (all with the most-significant-bit set to 1).
> - Values than would provide a shorter encoding than the requested length
> in octets.
> - Serial numbers for previously generated certificates.
> Now, CA/B Forum requires the use of 64 bit entropy when generating serial
> numbers, hence, for complying with that requirement, larger sizes than 8
> octets are required (for more information, refer to the CA/Browser Forum
> information on Ballot 164 - Certificate Serial Number Entropy).
> Possible values can range between 4 and 20 octets, and the default for all
> new CAs is 20 octets.


Finally, I have attached a patch with some improvements over the current
code at ECA-4991.

-- 
Jaime Hablutzel -  RPC 994690880