Cryptopp: BLAKE2 produces wrong hashes for non-default digest sizes

[c0ff]

The constructor BLAKE2_Base(bool treeMode, unsigned int digestSize) uses an incorrect initialization block and produces wrong results if either treeMode or digestSize differ from default.
The patch is here:
blake2-init-patch.zip

In essense it replaces
UncheckedSetKey(NULL, 0, g_nullNameValuePairs);
with
UncheckedSetKey(NULL, 0, MakeParameters(Name::DigestSize(), (int)digestSize)(Name::TreeMode(), treeMode, false));

[noloader]

Thanks @c0ff.

Can you test this against master? I'm tied up at the moment with other testing.

git clone https://github.com/weidai11/cryptopp-fix
cd cryptopp-fix

./cryptest.sh

cryptest.sh is our test script from hell, and it torture tests the library under different configurations. On a modern desktop, like a 5th gen or 6th gen, it will take a couple of hours to run.

---

Stepping back to 10,000 feet, this is my fault. I did not generate test vectors for the configurations, so the bug passed without notice. I'll checkout Sam Neves' BLAKE2 GitHub (@sneves) and generate them.

[c0ff]

bool ValidateBLAKE2b() does not cover this case. That's why it has gone undetected. I'll try to cook a patch for the test as well.

[c0ff]

Adjusted and tested the patches against master: blake2-init-and-test-patches.zip

Though, I only did a short test with cryptest.exe v 74 and cryptest.exe v 75.

After applying blake2-test.patch:
Without blake2-init.patch:
failed_blake2s.txt
failed_blake2b.txt

With blake2-init.patch:
passed_blake2s.txt
passed_blake2b.txt

[c0ff]

The new test vectors were generated with pyblake2 which uses 'BLAKE2 reference source code package' by Samuel Neves.

[noloader]

Thanks @c0ff.

Adjusted and tested the patches against master: blake2-init-and-test-patches.zip

Do you want to make a Pull Request (see Creating a pull request)? If you do, then you become contributor. I like to see folks get credit for their work, so I'd like to see you make one. It also relieves me from maintaining a separate file with a list of contributors.

If you don't want to, then that's fine, too. I'll drop your patch by hand and push it.

---

The new test vectors were generated with pyblake2 which uses 'BLAKE2 reference source code package' by Samuel Neves.

Cool, thanks. That's good provenance.

[c0ff]

Ok, will try to make a Pull Request then, I need to get used to it anyway :)

[noloader]

Ok, will try to make a Pull Request then, I need to get used to it anyway :)

Just an FYI... The GitHub instructions for updating your clone from Wei's are kind of borked. They create a lot of unnecessary merges. Use these instructions instead: Keeping A GitHub Fork Updated.

[noloader]

Cleared with Pull Request 416.