Affected versions contain multiple memory safety issues, such as:

• Unsoundly coercing immutable references to mutable references
• Unsoundly extending lifetimes of strings
• Adding the Send marker trait to objects that cannot be safely sent between threads

This may result in a variety of memory corruption scenarios, most likely use-after-free.

A significant refactoring effort has been conducted to resolve these issues.

Impact

Unix-like operating systems may segfault due to dereferencing a dangling pointer in specific circumstances. This requires an environment variable to be set in a different thread than the affected functions. This may occur without the user's knowledge, notably in a third-party library.

Workarounds

No workarounds are known.

References

• time-rs/time#293

The following Rust program demonstrates some strangeness in AES encryption - if you have an immutable key slice and then operate on that slice, you get different encryption output than if you operate on a copy of that key.

For these functions, we expect that extending a 16 byte key to a 32 byte key by repeating it gives the same encrypted data, because the underlying rust-crypto functions repeat key data up to the necessary key size for the cipher.

use crypto::{
    aes, blockmodes, buffer,
    buffer::{BufferResult, ReadBuffer, WriteBuffer},
    symmetriccipher,
};

fn encrypt(
    key: &[u8],
    iv: &[u8],
    data: &str,
) -> Result<String, symmetriccipher::SymmetricCipherError> {
    let mut encryptor =
        aes::cbc_encryptor(aes::KeySize::KeySize256, key, iv, blockmodes::PkcsPadding);

    let mut encrypted_data = Vec::<u8>::new();
    let mut read_buffer = buffer::RefReadBuffer::new(data.as_bytes());
    let mut buffer = [0; 4096];
    let mut write_buffer = buffer::RefWriteBuffer::new(&mut buffer);

    loop {
        let result = encryptor.encrypt(&mut read_buffer, &mut write_buffer, true)?;

        encrypted_data.extend(
            write_buffer
                .take_read_buffer()
                .take_remaining()
                .iter()
                .copied(),
        );

        match result {
            BufferResult::BufferUnderflow => break,
            BufferResult::BufferOverflow => {}
        }
    }

    Ok(hex::encode(encrypted_data))
}

fn working() {
    let data = "data";
    let iv = [
        0xF0, 0xF1, 0xF2, 0xF3, 0xF4, 0xF5, 0xF6, 0xF7, 0xF8, 0xF9, 0xFA, 0xFB, 0xFC, 0xFD, 0xFE,
        0xFF,
    ];
    let key = [
        0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E,
        0x0F,
    ];
    // The copy here makes the code work.
    let key_copy = key;
    let key2: Vec<u8> = key_copy.iter().cycle().take(32).copied().collect();
    println!("key1:{} key2: {}", hex::encode(&key), hex::encode(&key2));

    let x1 = encrypt(&key, &iv, data).unwrap();
    println!("X1: {}", x1);

    let x2 = encrypt(&key2, &iv, data).unwrap();
    println!("X2: {}", x2);

    assert_eq!(x1, x2);
}

fn broken() {
    let data = "data";
    let iv = [
        0xF0, 0xF1, 0xF2, 0xF3, 0xF4, 0xF5, 0xF6, 0xF7, 0xF8, 0xF9, 0xFA, 0xFB, 0xFC, 0xFD, 0xFE,
        0xFF,
    ];
    let key = [
        0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E,
        0x0F,
    ];
    // This operation shouldn't affect the contents of key at all.
    let key2: Vec<u8> = key.iter().cycle().take(32).copied().collect();
    println!("key1:{} key2: {}", hex::encode(&key), hex::encode(&key2));

    let x1 = encrypt(&key, &iv, data).unwrap();
    println!("X1: {}", x1);

    let x2 = encrypt(&key2, &iv, data).unwrap();
    println!("X2: {}", x2);

    assert_eq!(x1, x2);
}

fn main() {
    working();
    broken();
}

The output from this program:

     Running `target/host/debug/rust-crypto-test`
key1:000102030405060708090a0b0c0d0e0f key2: 000102030405060708090a0b0c0d0e0f000102030405060708090a0b0c0d0e0f
X1: 90462bbe32965c8e7ea0addbbed4cddb
X2: 90462bbe32965c8e7ea0addbbed4cddb
key1:000102030405060708090a0b0c0d0e0f key2: 000102030405060708090a0b0c0d0e0f000102030405060708090a0b0c0d0e0f
X1: 26e847e5e7df1947bf82a650548a7d5b
X2: 90462bbe32965c8e7ea0addbbed4cddb
thread 'main' panicked at 'assertion failed: `(left == right)`
  left: `"26e847e5e7df1947bf82a650548a7d5b"`,
 right: `"90462bbe32965c8e7ea0addbbed4cddb"`', src/main.rs:83:5

Notably, the X1 key in the broken() test changes every time after rerunning the program.

The following presentation at this year's DEF CON was brought to our attention on the Diesel Gitter Channel:

SQL Injection isn't Dead: Smuggling Queries at the Protocol Level
http://web.archive.org/web/20240812130923/https://media.defcon.org/DEF%20CON%2032/DEF%20CON%2032%20presentations/DEF%20CON%2032%20-%20Paul%20Gerste%20-%20SQL%20Injection%20Isn't%20Dead%20Smuggling%20Queries%20at%20the%20Protocol%20Level.pdf
(Archive link for posterity.) Essentially, encoding a value larger than 4GiB can cause the length prefix in the protocol to overflow, causing the server to interpret the rest of the string as binary protocol commands or other data.

It appears Diesel does perform truncating casts in a way that could be problematic, for example: https://github.com/diesel-rs/diesel/blob/ae82c4a5a133db65612b7436356f549bfecda1c7/diesel/src/pg/connection/stmt/mod.rs#L36

This code has existed essentially since the beginning, so it is reasonable to assume that all published versions <= 2.2.2 are affected.

Mitigation

The prefered migration to the outlined problem is to update to a Diesel version newer than 2.2.2, which includes fixes for the problem.

As always, you should make sure your application is validating untrustworthy user input. Reject any input over 4 GiB, or any input that could encode to a string longer than 4 GiB. Dynamically built queries are also potentially problematic if it pushes the message size over this 4 GiB bound.

For web application backends, consider adding some middleware that limits the size of request bodies by default.

Resolution

Diesel now uses #[deny] directives for the following Clippy lints:

• cast_possible_truncation
• cast_possible_wrap
• cast_sign_loss

to prevent casts that will lead to precision loss or other trunctations. Additionally we performed an audit of the relevant code.

A fix is included in the 2.2.3 release.