Passwords are a popular authentication method in the field of information technology. Passwords were created for humans to be remembered. Sometimes they are not ideal for usage in encryption software. Therefore, there exist key derivation functions, which transform a password into more suitable cryptographic key. This thesis deals with such functions, in particular considering their usage in disk encryption . The most popular function PBKDF2 is described together with its vulnerabilities and attacks. Memory-hard functions have started being used as a mitigation of time-memory trade-off attacks. One of such functions is Argon2 selected as a winner of Password Hashing Competition. The thesis describes Argon2 in detail. The practical part of the thesis deals with simulating of an attack on a disk encrypted with LUKS2 encryption scheme using Argon2 as PBKDF. It includes collecting Argon2 parameters benchmarked by Cryptsetup software. Attack is devised through CPUs and GPUs using high-performance hardware provided by MetaCentrum VO. The last part of the thesis introduces a price model for an attacker using either physical hardware or on-demand allocation of computing resources in the cloud. This model is then applied to real world prices and data obtained during the attack simulation. The thesis shows that it can take thousands of machines and hundreds of millions of dollars to crack a LUKS2 password eight characters long in ten years.