
go-guerrilla <=1.6.7: XCLIENT handler sliced input[8:] but the verb is 7 bytes — a bare XCLIENT command panics (slice out of range) and crashes the SMTP daemon. Real daemon v1.6.7 vs fixed v1.6.7-1: panic on vuln, 250 OK on fixed.
| CVE | CVE-2026-115578 |
| Component | go-guerrilla — server.go XCLIENT command handler |
| Affected | 1.6.7 and earlier |
| Fixed in | 1.6.7-1 — commit f0d0f2c4 (GHSA-hghq-53p7-c582) |
| CWE | CWE-129 (Improper Validation of Array Index), CWE-193 (Off-by-one Error) |
| Impact | An unauthenticated SMTP client that sends the bare XCLIENT verb (no attributes) crashes the daemon — denial of service against any go-guerrilla deployment with XCLIENT enabled (proxy/front-end setups) |
| Vector | Network — SMTP command on the listening port |
| Trigger | XCLIENT\r\n with no trailing space or attributes |
The XCLIENT handler assumed at least one separator byte after the 7-byte
verb and sliced input[8:]:
// vulnerable v1.6.7
case sc.XClientOn && cmdXCLIENT.match(cmd):
if toks := bytes.Split(input[8:], []byte{' '}); len(toks) > 0 { // input may be 7 bytes
For the bare verb XCLIENT (7 bytes), input[8:] is out of range and
panics. The fix guards the slice:
// fixed v1.6.7-1
if len(input) > len(cmdXCLIENT) {
if toks := bytes.Split(input[len(cmdXCLIENT)+1:], []byte{' '}); ...
}
cmdXCLIENT is the 7-byte literal XCLIENT. The handler hard-coded
input[8:], presuming a space follows the verb. A client that sends only
the verb produces a 7-byte input, so the slice expression panics with
"slice bounds out of range", taking down the daemon. The fix checks the
input length before slicing and uses len(cmdXCLIENT)+1 instead of a
magic constant.
| service | image | role |
|---|---|---|
gg-vuln | cve-2026-115578:vuln | go-guerrilla v1.6.7 |
gg-fixed | cve-2026-115578:fixed | go-guerrilla v1.6.7-1 (vendor fix) |
poc | cve-2026-115578-poc:local | stdlib-only Python driver |
Both run guerrillad serve with xclient_on=true on port 2525 and the
default gateway backend. A supervisor auto-restarts the daemon and exposes
/status (panic flag + log) and /restart on :8081, so the crash is
observable even after the process is brought back.
./run-lab.sh build # builds both tags with Go 1.25
./run-lab.sh up # cve115578lab network: gg-vuln / gg-fixed
./run-lab.sh sweep
./run-lab.sh down
poc/poc.py (stdlib only) opens an SMTP session, sends EHLO, then the
bare XCLIENT verb:
250 OK; 3 no worker, 2 unreachable.xclient_on.A standalone full-chain exploit ships at exploit/exploit.py:
python3 exploit/exploit.py --host=gg-vuln
# exit 1 on confirmed daemon crash, 0 on a safe target
| step | gg-vuln (1.6.7) | gg-fixed (1.6.7-1) |
|---|---|---|
bare XCLIENT | response EOF, panic=True (slice bounds out of range) | 250 2.1.0 OK, panic=False |
Captured runs: evidence/check-*-run.txt, evidence/exploit-*-run.txt,
evidence/probe-*-run.txt, evidence/mitigation-*-run.txt,
evidence/exploit-standalone-run.txt, evidence/versions.txt,
evidence/0day-sweep.txt.
cve115578lab
network.xclient_on=true), as in proxy deployments.The fixed line accepts the bare XCLIENT verb with 250 OK and no panic;
the length guard prevents the out-of-range slice. No crash path remains for
the bare verb after the fix.
README.md this write-up
run-lab.sh environment driver (build / up / sweep / down)
Makefile convenience targets
poc/poc.py the PoC (check / exploit / probe / mitigation)
exploit/exploit.py standalone full-chain exploit
lab/lab.Dockerfile builds both upstream tags
lab/config.json xclient_on SMTP listener config
lab/supervisor.py auto-restart supervisor + /status
evidence/ captured runs of every mode on both sides
TRACKING.md excluded (working record)