0

Whenever i tried generate √N randomly(possible combinations of password length), i always find few collisons of the hash password no matter the length, if the password was made of entirely random characters of course, so i test this case on hashcat. I noticed its generate method always keep some digits intact & try all other positions possible combinations like this

caefnNHJKn854557 --> plOcmawJKn854557

This make the chance for collison to happen very low. Is there any way to set up or configure hashcat to generate an entire random combinations eachtime it try candidates?

i tried the expriment with C++ & C# using their random generator function & many times got the collision. i want to try this with hashcat because its random generator was the fastest but don't know hot to configure it

closed as unclear what you're asking by Anders, Xander, forest, Gilles, Teun Vink Dec 31 '18 at 20:30

Please clarify your specific problem or add additional details to highlight exactly what you need. As it's currently written, it’s hard to tell exactly what you're asking. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.

  • 1
    Note that password hashes such as PBKDF2, Argon2i, bcrypt, etc are designed to be slow generally by iteration. The general design is to require about 100ms to hash a password. The best attack is using a large list of passwords ordered by frequency of use and this will fail against rather simple password creation methods, typically word-number-word such as "blup4handy" will not be in the list nor will fuzzing help. – zaph Dec 31 '18 at 14:21
3

For any halfway decent hash function, even hashes that are completely unsuitable for cryptography, if you look for collisions by brute force, it doesn't matter how you explore the search space. Keeping a fixed prefix and changing the suffix, or keeping a fixed suffix and changing the prefix, are as good as any other method. That's because all hash functions are designed to produce distinct results for similar inputs.

If there are N possible hash values, you need about to calculate and store about √N hashes until you have a good chance of finding a collision (the birthday bound). For a 32-bit hash as used in hash tables, N = 232 so √N = 216 which is reached very quickly. For a 128-bit hash such as older cryptographic hashes, √N = 264. A GPU these days can calculate something like 234 (16 billion) MD5 hashes per second, which means it calculates 264 hashes in about 230 seconds ≈ 34 years).

But that's not all: to find a collision, you need to store the result of all these calculations, so the limiting factor will be memory bandwidth. And once your RAM fills up, that'll be storage bandwidth. You aren't going to find a collision in a 128-bit hash in your lifetime.

Hashcat is designed to find hash preimages, not to look for collisions. Hashcat is designed to reverse password hashes (and ordinary hashes that are misused for passwords), and collisions are not relevant for that. But even with an optimized tool, what you're trying to do is not feasible for any cryptography-sized hash.

If you want to find a collision for a 128-bit hash, you need to understand its structure. There are known methods to find collisions for MD4 and MD5, for example. These methods are practical because they don't use brute force over the whole search space.

By the way, you're calling the hash inputs ”passwords“, but password hashes are different from “ordinary” cryptographic hashes. They're slower, and salted. The use of a random salt means that with a password hash, you don't get the same output twice, not even if you use the same password twice.

Not the answer you're looking for? Browse other questions tagged or ask your own question.