Code review

[kpp]

fix: compare sensitive data with sodium_memcmp
fix: replace memset with sodium_memzero for sensitive data
fix: make increment_nonce & increment_nonce_number independent of user-controlled input
fix: make crypto_core more stable agains null ptr dereference

fix: update apt before installing anything
add: comments about hairy code

[towlie]

This breaks compatibility with vanilla NaCl https://nacl.cr.yp.to/
Which is something @irungentoo doesn't want?

[kpp]

@towlie , do you have any idea how to make memset be non-optimized (and deleted) by compiler? Only sodium_memzero is safe to clear sensitive data before free it. NaCl does not have such function, Sodium does.
@irungentoo , what do you prefer? memset or sodium_memzero?

[GrayHatter]

@kpp toxcore has never been shy to write it's own wrappers for functions. Why not create a tox_wrapper_memzero function and then #ifdef depending on if you're using NaCL or libsodium?

[kpp]

@GrayHatter I did not find a similar function in NaCl. Maybe you can do that? Or do you propose me to write something like:

#ifdef USE_SODIUM
sodium_memzero(ptr, size);
#else
memset(ptr, 0, size);
#endif

I will not do that, because that will be unsafe.

[GrayHatter]

@kpp less safe than it is right now?

[mrkiko]

I was thinking about libbsd and their explicit_bzero.

[alexbakker]

@irungentoo any comments?

[irungentoo]

A bit paranoid but that's fine.

I just need to check if using public_key_cmp everywhere doesn't slow down the code too much.

[kpp]

I just need to check if using public_key_cmp everywhere doesn't slow down the code too much.

@irungentoo Well, asymptotic complexity of comparing public keys using memcmp is O(N), using public_key_cmp is ~N, so there are such public keys that in the worst case compares with memcmp as "slow" as they are compared with public_key_cmp. There must be no slow down.

[irungentoo]

I fixed it already but here the array indexes should have been i - 1.

[kpp]

Nice, I will fix it in 1-2 days

[kpp]

@irungentoo thank you. That was a bad bug. I wrote unit tests for that case. See PR.