Could you salt and perform a one-way hash on the IP address and store that? It would alleviate a large amount of leakage issues while still giving you uniqueness counts.
The IP is a an identifier, so unlike password salt (where the user is the identifier) you need a way to know what the salt is to hash the IP, and it needs to be consistent.
You can do a lookup table of IP-to-salt, but this either gives away your list of addresses (if only containing IPs you've seen) or is huge (entire ipv4 range), and either way doesn't prevent rainbow tables.
You can have a static salt for the entire site, but again this is not really helping much against rainbow tables (beyond requiring recalculating the table, once).
Is there a mitigation I'm not thinking of?
If you need the ability to group ciphertexts without decrypting them, you could create a scheme which will make cryptographers cringe, but could be justified in this specific case.
For instance, you can probably assert things like which IP blocks are likely to comprise most of the entries in the table or which IP blocks or addresses cannot be in the table.
That just makes me all sorts of uncomfortable.
Modern GPUs can manage several thousand million SHA256 hashes/sec, so even with a salt per hash it’s not going to take long to get a given entry, given the 32bit address space of IPv4
If I am got a DoS attack or Spam, I need the IP to find out to whom I should file abuse complain.
Do we need to sanitize SMTP header too? How about shuting down DNSBL?
From security point of view, if you use a fixed (unrelated to input) salt, the attacker will have a harder time to discover the function f (unless you store the salt next to your IP hashes). But from privacy point of view, in relationship between me (user) and you (service provider), you are the attacker. And you know your function f. Hashing IPv4 addresses, salt or not, gives me no privacy protection, since you can trivially reverse the hash - just due to small domain size. With IPv6, this problem will resolve itself somewhat; till then, I'd prefer if you encrypted those IPs with keys that have finite and short life time, in a way that a third party could audit if need be.
Simply out of curiosity, what do you mean by that?
All my life experience and knowledge tells me it's exactly how you get around the law, unless court has its own agenda or strong bias.
However I'm not sure myself it makes sense. Some people will be identified by just a partial IP or even a partial hash.