SqlCipherKeyring hangs on passphrase open: default Argon2 m_cost is a bytes value (~19 GB), only correct under #[cfg(test)]

[dougcooper]

Version: 2.3.0 (also present on main)

Summary

Opening a SqlCipherKeyring via the passphrase path (no explicit key) hangs / OOMs. ConnectionParams::default().memory is initialized from get_default_memory_cost(), whose non-test definition returns a bytes value that is then used as Argon2's KiB m_cost.

#[cfg(test)]
fn get_default_memory_cost() -> u32 { Argon2Params::DEFAULT_M_COST } // 19456 (KiB) ✓

#[cfg(not(test))]
fn get_default_memory_cost() -> u32 { 19_917_824 } // "19456 KiB converted to bytes" ✗

derive_key passes params.memory straight into Argon2Params::new(memory, …), where m_cost is measured in KiB blocks. So the non-test default requests 19_917_824 KiB ≈ 19 GB, and the derivation stalls allocating/initializing that buffer.

Because the #[cfg(test)] variant returns the correct Argon2Params::DEFAULT_M_COST (19456), the crate's own test suite never exercises the broken value — only real (non-test) builds do.

Repro

use cryptex::sqlcipher::{ConnectionParams, SqlCipherKeyring};

let mut p = ConnectionParams::default();
p.password = b"pw".to_vec();
p.salt = b"saltsaltsaltsalt".to_vec();
// hangs on the ~19 GB Argon2 allocation:
let _ = SqlCipherKeyring::with_params(&p, Some(dir)).unwrap();

Setting p.memory = 19_456; (the correct KiB value) before the call makes it return in tens of milliseconds.

Also affected

• The ConnectionParams doc example uses memory=19917824, propagating the bytes value.
• The FromStr range check (Argon2Params::DEFAULT_M_COST..Argon2Params::MAX_M_COST) accepts the oversized value, so it isn't caught during parsing.

Suggested fix

• Make #[cfg(not(test))] get_default_memory_cost() return Argon2Params::DEFAULT_M_COST (KiB), matching the test variant.
• Update the doc example to a KiB value.
• Optionally document that ConnectionParams::memory is in KiB blocks, not bytes.

[mikelodder7]

Hmm, it was supposed to be 19MB not 19GB. I'll look at argon2 and see if something changed recently or if I read it wrong.