test_crypto_generichash fails with libsodium 1.0.13 about pysodium

[stef]

interesting! also on alpine linux. but not with 1.0.12. looking at libsodium what changed between those two versions.

[stef]

smallest reproducible testcase is this:

state = pysodium.crypto_generichash_init()

[stef]

maybe this change triggered it: jedisct1/libsodium@f72bcfa

[jedisct1]

Very likely.

But you shouldn't assume anything about the content of the structure. This may change, especially if BLAKE2X is added.

Use crypto_generichash_statebytes() to get the correct amount of bytes required to store the state, and you will always be safe.

[stef]

confirmed. just built a libsodium with that change reverted and it doesn't segfault anymore.

[stef]

not sure it's the number of bytes that's the problem, but the alignment itself?
this https://github.com/stef/pysodium/blob/master/pysodium/__init__.py#L160 seems to be exactly matching this https://github.com/jedisct1/libsodium/blob/master/src/libsodium/include/sodium/crypto_generichash_blake2b.h#L23

[jedisct1]

Use crypto_generichash_statebytes(), trust me :)

[stef]

would this be correct to pass to generichash_init?

state = sodium_malloc(crypto_generichash_statebytes());

and i would not have to mess with alignment?

[jedisct1]

That would work! But maybe CryptoGenericHashState can just be an array whose size is ctypes.c_uint8 * sodium.crypto_generichash_statebytes() bytes?

[stef]

seems to work. pushed a fix.

[stef]

thx to all of you, for reporting and contributing a fix

[stef]

i bumped the version and also uploaded a new version to pypi, if you agree to the fix please close this issue.

[jedisct1]

Awesome, thanks Stef!

[mans0954]

Thanks for the prompt response. Debian package updated:

https://packages.qa.debian.org/p/pysodium/news/20170811T211431Z.html

Christopher