How difficult/easy is it to decode a custom base 62 encoding?

Question

Here's what I am going to do to obfuscate database id's in permalinks:

1) XOR the id with a lengthy secret key

2) Scramble (rotate, flip, reverse) bits around a little in the XOR'ed integer in a reversible way

3) Base 62 encode the resulting integer with my own secret scrambled up sequence of all alphanumeric characters (A-Za-z0-9)

How difficult would it be to convert my Base 62 encoding back to base 10?

Also How difficult is it to reverse engineer the whole process? (obviously without taking a peak at source or compiled code) I know 'only XOR' is pretty susceptible to basic analysis.

EDIT: the result should be not more than 8-9 chars long, 3DES and AES seem to produce very long encrypted texts and can't be practically used in URLs

Resulting strings look something like:

In [2]: for i in range(1, 11):
    print code(i)
   ...:     
9fYgiSHq
MdKx0tZu
vjd0Dipm
6dDakK9x
Ph7DYBzp
sfRUFaRt
jkmg0hl
dBbX9nHk4
ifqBZwLW
WdaQE630

As you can see 1 looks nothing like 2 so this seems to works great for obfuscation of id's.

I'd consider using a standard blockcipher with a blocksize of 64 bits giving you 11 Base62 chars. For example 3DES would do the job. That gives you actual security and doesn't rely on your code staying secret. — CodesInChaos, Feb 22 '13 at 14:37
The core assumption of modern cryptography is that the attacker knows everything about your system except the key. By that standard your system is broken. — CodesInChaos, Feb 22 '13 at 14:38
@CodesInChaos ok I get it, i'll see if 3DES works for me as I wanted encrypted id to be not more than 8-9 chars long. but I also really wanted to know out of curiosity, as to how difficult is it to revert custom base62 encoding (with secret scrambled up sequence of alphanumerics that only i possess) — Optimus, Feb 22 '13 at 14:46
@DanW than please suggest me an encryption where the result is not more than 8-9 chars long, 3DES and AES seem to produce very long encrypted texts and can't be practically used in URLs — Optimus, Feb 22 '13 at 14:59
I think you'd get better responses if you had posted this question in [Cryptography](http://crypto.stackexchange.com/). Before you do that, I can already tell you that this type of cipher is considered extremely weak and insecure, however complicated it may seem. — John Willemse, Feb 22 '13 at 15:06
With this scheme, you don't expect people to type the URL. Therefore, limiting the length is not necessary. You might do better using a (secure, standard) hash of the URL you're obscuring, and at the server, simply look up the hash (in a database) to find the real URL (hashes are not reversible). — Jonathan Leffler, Feb 22 '13 at 15:12

score 2 · Answer 1 · edited Apr 13 '17 at 12:48

If the attacker is allowed to play around with the input, it will be trivial for a skilled attacker to "decrypt" the data. A crucial property of modern crypto systems is the "Avalanche effect" which your system lacks. Basically it means that every bit of the output is connected with every bit of the input.

If an attacker of your system is allowed to see that, for example, id = 1000 produces the output "AAAAAA" and id=1001 produces "ABAAA" and id=1002 produces "ACAAA" the algorithm can be easily reversed, and the value of the key obtained.

That said, this question is a better fit for https://security.stackexchange.com/ or https://crypto.stackexchange.com/

score 2 · Answer 2 · answered Feb 22 '13 at 16:19

The standard advice for anyone trying to develop their own cryptography is, "Don't". The advanced advice is to read Bruce Schneier's Memo to the Amateur Cipher Designer and then don't.

You are not the first person to need to obfuscate IDs, so there are already methods available. @CodesInChaos suggested a good method above; you should try that first to see if it meets your needs.

How difficult/easy is it to decode a custom base 62 encoding?

2 Answers2