Password Hashing: Scrypt, Bcrypt and ARGON2

Michele Preziuso
Jan 6, 2019 · 5 min read

There’s always a lot of debate in regards to how to safely store passwords and what algorithm to use: MD5, SHA1, SHA256, PBKDF2, Bcrypt, Scrypt, Argon2, plaintext??

So I tried to analyse and summarise the most recent and reasonable choices: Scrypt, Bcrypt and Argon2. …and yes, MD5, SHA1, SHA256 are not suitable for storing passwords! 😉

A summary

In 2015, I’ve published ‘Password Hashing: PBKDF2, Scrypt, Bcrypt’ intended as an extended reply to a friend’s question.

Summarily saying that:

Attackers have usually different and more specialised (powerful) hardware than ours;

Attackers use specialized hardware because it can be tailored to the algorithm, the different hardware architecture allows certain algorithm to run faster than on non-specialised hardware (CPU) and - overall - certain algorithms can be parallelised;

We rely on slow-hashing functions to hash passwords in order to fight the attacker on equal grounds: your power unit (CPU/GPU) against his GPU/FPGA/ASIC

All of the above is still true, however the race to cryptocurrencies added another facet to it: with multi-billion dollar market caps, the fastest software/hardware implementation of the underlying crypto algorithm of a cryptocurrency is also the one with an advantage against other miners and thus the most profitable one.

Whilst Bitcoin uses SHA256 as the underlying crypto function (which can therefore be greatly optimised on optimised hardware making it an ‘unfair’ coin for miners) other creators have tried to make new cryptocurrencies more fair to mine by relying on memory-hard: Litecoin (Scrypt) as an early example and Zcash (Equihash) as a more recent one.

This means that the same slow-functions that are used for password hashing are being used to protect millions or even billions of dollars in cryptocurrencies, making the building of the fastest implementation of the slow hashing function even more rewarding and usually publicly available!

What’s safe today then?

The principle is still the same: we need a slow function that’s been vetted by the crypto community and remains unbroken.

PBKDF2 has been out there for a long time and hasn’t aged very well as discussed in the previous article: easily parallelised on multi-core systems (GPUs) and trivial for tailored systems (FPGAs/ASICs). So it’s a No from me.

BCrypt has been out there since 1999 and does a better job at being GPU/ASIC resistant than PBKDF2 but I wouldn’t recommend it for new systems as it doesn’t shine in a threat model with offline cracking.
While some cryptocurrencies rely on it (i.e. NUD), it hasn’t gained a lot of popularity and - consequentially - enough interest from the FPGA/ASIC community to build a hardware implementation of it.
That being said, Solar Designer (OpenWall), Malvoni and Knezovic (University of Zagreb) have written a paper in 2014 describing a hybrid a system of ARM/FPGA SOCs to attack the algorithm.

SCrypt is a better choice today: better design than BCrypt (especially in regards to memory hardness) and has been in the field for 10 years.
On the other hand, it has been used for many cryptocurrencies and we have a few hardware (both FPGA and ASIC) implementation of it.

Argon2

Shortly after writing my initial article, Argon2 won the PHC in July 2015.

The competition

The competition was initiated in fall 2012, in Q1 2013 the board published a call for submissions and the deadline was end of March 2014. As part of the competition, the panelists thoroughly reviewed the submissions and published an initial short report where they describe their selection criteria and rationale.

An introduction

There are two main versions of Argon2: Argon2i which is the safest option against side-channel attacks and Argon2d which is the safest option against GPU cracking attacks.

Source code is available on GitHub, written in C89-compliant C, licensed under CC0 and compiles on most ARM, x86 and x64 architectures.

Parameters for fine-tuning

Both versions of the algorithm can be parameterised by:

  • A time cost, which defines the execution time
  • A memory cost, which defines the memory usage
  • A parallelism degree, which defines the number of threads

this means that you can separately tune these parameters and tailor the security bound to your use case, threat model and hardware specifications.

Tradeoff attacks

On top of this, Argon2 is particularly resistant to ranking tradeoff attacks making it much more difficult to cheaply optimise on FPGAs: even though recent FPGAs have embedded RAM blocks, memory bandwidth is still a constraint.

This and similar attacks are discussed in the specs (see chapter 5) as well as in a separate paper by the same authors where they also compare it with scrypt.

Argon2id

The below is a quote/paraphrase from the Argon2 IETF Draft.

Argon2d uses data-depending memory access, which makes it suitable for cryptocurrencies and PoW applications with no threats from side-channel timing attacks. Argon2i uses data-independent memory access, which is preferred for password hashing. Argon2id works as Argon2i for the first half of the first iteration over the memory and as Argon2d for the rest, thus providing both side-channel attack protection and bruteforce cost savings due to time-memory tradeoffs. Argon2i makes more passes over the memory to protect from tradeoff attacks.

If you fear side-channel attacks you should use the Argon2i version which is not vulnerable to side-channel attacks, otherwise Argon2d which is vulnerable to timing attacks, but offers the best resistance to TMTO.

If you are unsure or if you’re comfortable with a hybrid approach you can use Argon2id to have the best of two world.

Conclusion

In 2019 I’d recommend not to use PBKDF2 or BCrypt in the future and highly recommend Argon2 (preferrably Argon2id) for newer systems.

Scrypt can be a second choice on systems where Argon2 is not available, but keep in mind that it has the same issues with respect to side-channel leakage.

Updates:

  • May 16th 2020 (thanks @Sc00bzT): Argon2 is not AES-based, scrypt miners can’t easily be repurposed for password cracking and since Meltdown/Spectre allow attackers to read memory, this specific side-channel is the last of your worries: the attacker would be able to read the password in plaintext! 😃

Since you’re interested in security, you may also be interested in a series of posts I’m writing about common vulnerabilities and how to prevent them.

First two articles are about Preventing SQL injection (SQLi) and Preventing Cross-site Scripting (XSS).

Analytics Vidhya

Analytics Vidhya is a community of Analytics and Data…

Michele Preziuso

Written by

CEO @KaosDynamics. AWS Certified DevOps Engineer @Peach_video. Security fanatic. I build things, I break things. Also a human …sometimes.

Analytics Vidhya

Analytics Vidhya is a community of Analytics and Data Science professionals. We are building the next-gen data science ecosystem https://www.analyticsvidhya.com

Michele Preziuso

Written by

CEO @KaosDynamics. AWS Certified DevOps Engineer @Peach_video. Security fanatic. I build things, I break things. Also a human …sometimes.

Analytics Vidhya

Analytics Vidhya is a community of Analytics and Data Science professionals. We are building the next-gen data science ecosystem https://www.analyticsvidhya.com

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch

Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore

Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store