If the hashing is done on the client and then sent to the server, then the server is effectively just processing as a plaintext password. If an attacker gets hold of the server password database, then they can just connect to the server and pretend to be the client and hand it the hashed password that they read from the database breach.
If you hash the password on the server instead, then if the password database is breached, then an attacker needs to actually reverse the hash[0] and find the original password in order to log in, because that's all that the server will accept.
[0] Note, this should be difficult[1]
[1] In crypto, "difficult" should mean "impossible before the end of the universe"
No it's not. Did you ever think that you can hash something twice? Hash it once on the client, then hash and salt it server side, like normal. It means that the server never actually knows your password, but that's about all it gives you.
> It means that the server never actually knows your password
If the client is hashing it without a salt the server could simply check a Rainbow table (https://en.wikipedia.org/wiki/Rainbow_table) to know which password it is. For short inputs this could be trivial.
Sure, but I still think this is preferable to sending the password in clear text even over HTTPS. You're trusting the server doesn't do anything with the password and immediately hashes it, but it might not. It might store it, or even if it doesn't, your password will stick around in RAM for an indeterminate amount of time.
If the server is compromised in any way, passwords could be exfiltrated. Companies are, sometimes, wildly incompetent. Zoom historically stored private keys on the same server as their "encrypted" data. I would not be surprised if your password is just stored for "convenience" or some other bullshit reason and just waiting to be breached.
> Sure, but I still think this is preferable to sending the password in clear text even over HTTPS. You're trusting the server doesn't do anything with the password
My point is in both cases the server has access to the password. As I mentioned, without salt the server can get the original password (by checking the pre-computed rainbow table of hashes up to n length), so the trust issue is the same.
If this is slightly better (more obfuscated) or the same thing + a false sense of security is debatable, but I could agree.
Well no, because rainbow tables are quite small. You don't have precomputed hashes for all passwords 24 characters and under that contain numbers and symbols.
I mean, even with just letters, you're looking at 620448401733239439360000 hashes required. x 128 / 8 bytes, you're looking at ~ 9000 zettabytes. So, a few order of magnitude larger than the entire internet.
If you have a strong password, it's not comparable. In scenario one the server has the password immediately. In scenario two, it would require the heat death of the universe to precompute the hash to find out the password.
Very interesting point. I did not think of that, thanks. I was looking at the speed of calculation in different scenarios (calculated on the fly)based on which characters are part of the password and it is still very difficult, if you have a strong password. I am curious which percentage one could match just checking for hashes of common passwords, or common patterns like exclamation marks at the end. In any case it is better than the unhashed version, I agree.
If you hash the password on the server instead, then if the password database is breached, then an attacker needs to actually reverse the hash[0] and find the original password in order to log in, because that's all that the server will accept.
[0] Note, this should be difficult[1] [1] In crypto, "difficult" should mean "impossible before the end of the universe"