A bug in the SmallVec::insert_many method caused it to allocate a buffer that was smaller than needed. It then wrote past the end of the buffer, causing a buffer overflow and memory corruption on the heap.

This bug was only triggered if the iterator passed to insert_many yielded more items than the lower bound returned from its size_hint method.

The flaw was corrected in smallvec 0.6.14 and 1.6.1, by ensuring that additional space is always reserved for each item inserted. The fix also simplified the implementation of insert_many to use less unsafe code, so it is easier to verify its correctness.

Thank you to Yechan Bae (@Qwaz) and the Rust group at Georgia Tech’s SSLab for finding and reporting this bug.

In affected versions, after a Report is constructed using wrap_err or wrap_err_with to attach a message of type D onto an error of type E, then using downcast to recover ownership of either the value of type D or the value of type E, one of two things can go wrong:

• If downcasting to E, there remains a value of type D to be dropped. It is incorrectly "dropped" by running E's drop behavior, rather than D's. For example if D is &str and E is std::io::Error, there would be a call of std::io::Error::drop in which the reference received by the Drop impl does not refer to a valid value of type std::io::Error, but instead to &str.

• If downcasting to D, there remains a value of type E to be dropped. When D and E do not happen to be the same size, E's drop behavior is incorrectly executed in the wrong location. The reference received by the Drop impl may point left or right of the real E value that is meant to be getting dropped.

In both cases, when the Report contains an error E that has nontrivial drop behavior, the most likely outcome is memory corruption.

When the Report contains an error E that has trivial drop behavior (for example a Utf8Error) but where D has nontrivial drop behavior (such as String), the most likely outcome is that downcasting to E would leak D.

The borsh serialization of the HashMap did not follow the borsh specification. It potentially produced non-canonical encodings dependent on insertion order. It also did not perform canonicty checks on decoding.

This can result in consensus splits and cause equivalent objects to be considered distinct.

This was patched in 0.15.1.

In the unique reclaim path of BytesMut::reserve, the condition

if v_capacity >= new_cap + offset

uses an unchecked addition. When new_cap + offset overflows usize in release builds, this condition may incorrectly pass, causing self.cap to be set to a value that exceeds the actual allocated capacity. Subsequent APIs such as spare_capacity_mut() then trust this corrupted cap value and may create out-of-bounds slices, leading to UB.

This behavior is observable in release builds (integer overflow wraps), whereas debug builds panic due to overflow checks.

PoC

use bytes::*;

fn main() {
    let mut a = BytesMut::from(&b"hello world"[..]);
    let mut b = a.split_off(5);

    // Ensure b becomes the unique owner of the backing storage
    drop(a);

    // Trigger overflow in new_cap + offset inside reserve
    b.reserve(usize::MAX - 6);

    // This call relies on the corrupted cap and may cause UB & HBO
    b.put_u8(b'h');
}

Workarounds

Users of BytesMut::reserve are only affected if integer overflow checks are configured to wrap. When integer overflow is configured to panic, this issue does not apply.

Impact

When user-provided input is provided to any type that parses with the RFC 2822 format, a denial of service attack via stack exhaustion is possible. The attack relies on formally deprecated and rarely-used features that are part of the RFC 2822 format used in a malicious manner. Ordinary, non-malicious input will never encounter this scenario.

Patches

A limit to the depth of recursion was added in v0.3.47. From this version, an error will be returned rather than exhausting the stack.

Workarounds

Limiting the length of user input is the simplest way to avoid stack exhaustion, as the amount of the stack consumed would be at most a factor of the length of the input.