AsyncListener.handle_query_or_defer retained every truncated (TC-bit) incoming query in self._deferred[addr] and armed a per-addr timer in self._timers[addr] that flushed the reassembled query within ~500 ms (RFC 6762 §18.5). Neither the per-addr list nor the number of distinct addr keys was capped, and the dedup check (for incoming in reversed(deferred): if incoming.data == msg.data) ran O(N) over the per-addr list on every arrival.
Any unauthenticated host on the local link (UDP/5353, 224.0.0.251 / ff02::fb) can stream byte-distinct TC-flagged mDNS queries — each up to _MAX_MSG_ABSOLUTE = 8966 bytes, with DNSIncoming retaining the raw data buffer plus parsed-record state. Trivially spoofed source IPs multiply the effect across _deferred / _timers, and the O(N) data compare burns CPU quadratically as each per-addr queue grows. On memory-constrained deployments (Home Assistant on Raspberry-Pi-class hardware is the canonical victim) sustained traffic OOM-kills the process; under lighter load, the per-arrival scan and event-loop scheduler starvation break unrelated zeroconf consumers (discovery, registration, ServiceBrowser callbacks).
Fixed in zeroconf 0.149.12 (PR #1751). Upgrade to >= 0.149.12.
There is no in-process workaround; upgrading is the fix. Otherwise, restrict mDNS (UDP/5353) to trusted Layer-2 segments via AP client isolation, guest-network separation, or host firewall rules.
{
"github_reviewed_at": "2026-06-11T13:28:55Z",
"severity": "MODERATE",
"cwe_ids": [
"CWE-400",
"CWE-770"
],
"github_reviewed": true,
"nvd_published_at": null
}