Google Chrome 115.0.5790.102 produces incorrect SHA-256 hashes #40

randombit · 2023-08-03T15:47:01Z

I don't think this is a bug in this code but opening this issue to maybe help out someone else trying to debug this in the future:

When this library is used with Google Chrome 115.0.5790.102 (released on July 20 2023), it will produce correct hashes for a while and then after a certain number of hashes are computed, all future hash outputs will be incorrect.

This was seen on x86 and ARM machines running Linux and macOS.

Presumably this is a JIT bug in this particular version of Chrome. It is apparently fixed in 115.0.5790.110 which was released to stable channel 5 days later.

partical · 2023-08-21T01:58:08Z

I have a similar problem with Google Chrome 116.0.0.0, it will produce incorrect hashes for a while and then after a certain number of hashes are computed, all future hash outputs will be correct.

emn178 · 2023-08-21T02:09:18Z

Can you give a sample code?

slisson · 2023-08-24T19:34:33Z

I can reproduce it in Chrome 116.0.5845.110 using this code:

<html>
<body>
<pre id="output"></pre>
<script src="https://cdn.jsdelivr.net/gh/emn178/js-sha256/build/sha256.min.js"></script>
<script>
    for (let i = 0; i < 100; i++) {
        document.getElementById("output").innerText += i + ": " + sha256.hex("a") + "\n"
    }
</script>
</body>
</html>

After 33 iterations it starts returning a different value.

slisson · 2023-08-24T19:43:13Z

An input of "b" is even more interesting. Then it produces more than two different values.

0tt0sson · 2023-08-25T07:15:07Z

We're experiencing slowly increasing error rate based some Chrome version rollout.

For now, we changed to @aws-crypto/sha256-js

import { Sha256 } from '@aws-crypto/sha256-js';

export const sha256 = (s: string): string => {
  const hash = new Sha256();
  hash.update(s);
  const hashUint8Array = hash.digestSync();

  return Buffer.from(hashUint8Array).toString('hex');
};

Replaced it js-sha256 with @aws-crypto/sha256-js. Also added consistency checks on server and client side to ensure both side compute the same hash value. see emn178/js-sha256#40

emn178 · 2023-08-25T16:03:45Z

I can't reproduce in my Chrome 116.0.5845.110 (arm64)

ggzzyy112233 · 2023-08-26T02:55:37Z

I had the same problem and needed to open a traceless browser to solve it.my Chrome 116.0.5845.96

NicolasFlamel1 · 2023-08-27T05:36:12Z

I'm having the same issue with Chrome 116.0.5845.110 on x86-64 Linux. The JIT is doing some weird optimizations to the a variable in the Sha256.prototype.hash function which causes its value to be inconsistent.

As a workaround, using the a variable immediately after setting it in each pass seems to fix this. I accomplish this by passing it to an empty unnamed function here.

djouf007 · 2023-08-29T14:45:44Z

Same problem, could you help how to fixe that in angular 16 pls, do i need to go to the file js-sha256? is it possible to explain how we do it

thanks

hanzheliva · 2023-08-29T14:56:45Z

We've replaced this library with @aws-crypto/sha256-browser as a solution

emn178 · 2023-08-29T15:08:36Z

I use @slisson 's sample code and tried on many versions and devices. I still can't reproduce this issue. Can you guys reproduce by the sample code?

hanzheliva · 2023-08-29T15:56:08Z

https://commondatastorage.googleapis.com/chromium-browser-snapshots/index.html?prefix=Mac/1160321/

The link is located using https://www.chromium.org/getting-involved/download-chromium/

If you run this Chromium 116.0.5845.96 on Macbook x86 version, you will get a different hash after 132 tries.

And this behavior is not stable if you run the sample code another time

@emn178

phoenixeliot · 2023-08-29T16:58:01Z

My coworker and I narrowed at least part of it down to this i++ operation acting very strangely — if you hover over it and then other expressions on this line repeatedly in the debugger, it causes i to increment just by previewing that value, which is extremely not-correct behavior. At this state in the debugger i simultaneously has a value of 0, and 2 depending on which part of the devtools you ask, and it seems that the incremented value is the one used in the calculation.

Sharing in case this is helpful in updating this library's code to work around this problem. Perhaps not inlining the combined var and ++ parts might help? In particular it's weird that i is declared much higher up than where it's used in this loop.

phoenixeliot · 2023-08-29T20:51:34Z

FWIW, even though the above debugger behavior was reproducible on my computer, the simple repro case above does not repro for me, but does repro for my coworker consistently. I haven't run into this general problem happening in production, but he does very consistently. IIRC we use the same model of computer (m1 pro macbook pro 16")

emn178 · 2023-08-30T02:53:09Z

@hanzheliva I can reproduce by download old version. And I use @NicolasFlamel1 's workaround but change to an assignment

{
  //...
  e = a + t1 << 0;
  a = t1 + t2 << 0;
  this.chromeBugWorkAround = true;
}

This seems work and won't lose performance too much. I will use this change to fix this issue in next version.

But I can't sign in my npm account to publish. still working on it...

### Fixed - Chrome bug by workaround. #40 - deprecated `new Buffer`, replace with `Buffer.from`. #34 - dependencies and security issues. #32, #36 ### Changed - TypeScript interface, secretKey can be bytes like message. #23, #25 - remove `eval` and use `require` directly. #18, #26

emn178 · 2023-08-30T15:21:09Z

workaround in v0.10.0. Please check if it fixed.

djouf007 · 2023-08-30T15:41:58Z

Thanks Sir but i have this erros with npm install js-sha256@0.10.0

./node_modules/js-sha256/src/sha256.js:68:17-34 - Error: Module not found: Error: Can't resolve 'crypto' in '/var/www/produits/neoshop/neoV16/node_modules/js-sha256/src'

BREAKING CHANGE: webpack < 5 used to include polyfills for node.js core modules by default.
This is no longer the case. Verify if you need this module and configure a polyfill for it.

If you want to include a polyfill, you need to:
        - add a fallback 'resolve.fallback: { "crypto": require.resolve("crypto-browserify") }'
        - install 'crypto-browserify'
If you don't want to include a polyfill, you can use an empty module like this:
        resolve.fallback: { "crypto": false }

How do i can fixe this

Thanks again

emn178 · 2023-08-31T00:49:03Z

This version remove eval, so webpack detects "require" and causes this. Please check
https://stackoverflow.com/questions/54162297/module-not-found-error-cant-resolve-crypto

emn178 · 2023-08-31T01:47:57Z

@djouf007 I found that I can disable polyfills in my package. use v0.10.1 should work.

djouf007 · 2023-08-31T09:41:49Z

Thank you so much; it's all working perfectly @emn178

dskloetd · 2023-09-05T07:32:45Z

Has anyone opened an issue on crbug.com for this?

marcos-statsig mentioned this issue Aug 28, 2023

Patch js-sha256 bug on some Chrome versions statsig-io/js-client#18

Closed

tore-statsig mentioned this issue Aug 28, 2023

[Bug Fix] Patch js-sha256 on some chrome versions statsig-io/js-client#19

Merged

This was referenced Aug 29, 2023

fix: 🐛 fix sha256 incorrect in google chrome AElfProject/aelf-web3.js#114

Closed

Feature/test chain util AElfProject/aelf-web3.js#115

Closed

potter-sun mentioned this issue Aug 30, 2023

fix: 🐛 sha256 incorrect in google chrome AElfProject/aelf-web3.js#123

Merged

acheroncrypto mentioned this issue Aug 30, 2023

vue + vite = broken sha256 for anchor discriminator ? coral-xyz/anchor#2619

Closed

emn178 closed this as completed Sep 9, 2023

neeboo mentioned this issue Oct 30, 2023

Dependencies with js-sha256 is risky unisat-wallet/bitcoin-simple-keyring#1

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Google Chrome 115.0.5790.102 produces incorrect SHA-256 hashes #40

Google Chrome 115.0.5790.102 produces incorrect SHA-256 hashes #40

randombit commented Aug 3, 2023

partical commented Aug 21, 2023

emn178 commented Aug 21, 2023

slisson commented Aug 24, 2023

slisson commented Aug 24, 2023

0tt0sson commented Aug 25, 2023

emn178 commented Aug 25, 2023

ggzzyy112233 commented Aug 26, 2023 •

edited

Loading

NicolasFlamel1 commented Aug 27, 2023

djouf007 commented Aug 29, 2023

hanzheliva commented Aug 29, 2023

emn178 commented Aug 29, 2023

hanzheliva commented Aug 29, 2023 •

edited

Loading

phoenixeliot commented Aug 29, 2023

phoenixeliot commented Aug 29, 2023 •

edited

Loading

emn178 commented Aug 30, 2023

emn178 commented Aug 30, 2023

djouf007 commented Aug 30, 2023 •

edited

Loading

emn178 commented Aug 31, 2023

emn178 commented Aug 31, 2023

djouf007 commented Aug 31, 2023

dskloetd commented Sep 5, 2023 •

edited

Loading

Google Chrome 115.0.5790.102 produces incorrect SHA-256 hashes #40

Google Chrome 115.0.5790.102 produces incorrect SHA-256 hashes #40

Comments

randombit commented Aug 3, 2023

partical commented Aug 21, 2023

emn178 commented Aug 21, 2023

slisson commented Aug 24, 2023

slisson commented Aug 24, 2023

0tt0sson commented Aug 25, 2023

emn178 commented Aug 25, 2023

ggzzyy112233 commented Aug 26, 2023 • edited Loading

NicolasFlamel1 commented Aug 27, 2023

djouf007 commented Aug 29, 2023

hanzheliva commented Aug 29, 2023

emn178 commented Aug 29, 2023

hanzheliva commented Aug 29, 2023 • edited Loading

phoenixeliot commented Aug 29, 2023

phoenixeliot commented Aug 29, 2023 • edited Loading

emn178 commented Aug 30, 2023

emn178 commented Aug 30, 2023

djouf007 commented Aug 30, 2023 • edited Loading

emn178 commented Aug 31, 2023

emn178 commented Aug 31, 2023

djouf007 commented Aug 31, 2023

dskloetd commented Sep 5, 2023 • edited Loading

ggzzyy112233 commented Aug 26, 2023 •

edited

Loading

hanzheliva commented Aug 29, 2023 •

edited

Loading

phoenixeliot commented Aug 29, 2023 •

edited

Loading

djouf007 commented Aug 30, 2023 •

edited

Loading

dskloetd commented Sep 5, 2023 •

edited

Loading