Python PoC for five pre-auth CVEs in Copeland/Dixell XWEB EVO: OS command injection via .lua sinks, setuid-root wlua escalation, hardcoded master password, and unauth config write, with a Docker lab rehost.
Copeland/Dixell XWEB EVO is the web brain of supermarket refrigeration and HACCP food-safety monitoring: an appliance that sits in a store back office and drives the whole cold-chain estate. On 2026-11-12 VulnCheck disclosed 14 vulnerabilities in it (CNA: VulnCheck, researcher Valentin Lobstein), every one proven to uid=0(root) against appliance labs rebuilt from vendor firmware. This repo covers the five pre-authentication members of that cluster with one PoC script against a lab rehost: a pre-auth OS command injection in the login stack that turns into root through the setuid-root wlua interpreter (CVE-2026-124002, with the password-field variant CVE-2026-124007), a hardcoded master password for the Administrator account (CVE-2026-124004), a single-GET root command execution through the setuid-root xwconsole.cgi (CVE-2026-124010), and an unauthenticated write of the system configuration file (CVE-2026-124014). The product line went End-of-Software-Support on 2026-03-16, so the exposed fleet gets nothing newer than what shipped before that date - and CVE-2026-124014 has no fixed release at all. VulnCheck canaries recorded in-the-wild exploitation of 124014 within 100 minutes of the NVD record going live, and of 124010 the next morning; both sit in VulnCheck KEV. The line was also named by CERT-FR among the controllers compromised in the October 2026 attack that forced ~700 French supermarkets into defrost overnight (CERTFR-2026-ALE-012). This is lab-only work: the lab rehost reproduces the documented causal chain, nothing here was aimed at any real unit.
unsupported-when-assignedCVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H 9.8 critical; 124004 9.8; 124010 9.8; 124007 8.1 (CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N 9.2); 124014 9.1 (CNA 4.0: 8.8)unsupported-when-assigned)All five primitives are verified against the lab rehost (lab/), each with a must-NOT-trigger control run (lab/control, see Notes for why the control is config-gated):
| CVE | Primitive | Status | Evidence |
|---|---|---|---|
| CVE-2026-124002 | username field of login.lua / cleanout.lua / api.lua -> www-data -> setuid-root wlua -> root | VERIFIED (all three sinks) | evidence/02-cve-124002-root-vuln.txt |
| CVE-2026-124007 | password field (hashed-mode accounts) -> hashpass invocation -> root | VERIFIED (login.lua + api.lua) | evidence/03-cve-124007-password-root-vuln.txt |
| CVE-2026-124004 | hardcoded master password -> Administrator session | VERIFIED | evidence/04-cve-124004-backdoor.txt |
| CVE-2026-124010 | unauthenticated GET xwconsole.cgi?sh= -> root, output inline | VERIFIED | evidence/05-cve-124010-rootexec.txt |
| CVE-2026-124014 | unauthenticated POST config.lua?action=set -> /etc/xwebconfig.json overwrite + SYSTEM_NOT_CONFIGURED state | VERIFIED (write, read-back, DoS state) | evidence/06-cve-124014-config-write.txt |
Honest lab-fidelity split, per endpoint: the sink lines for 124002, the wlua main, the xwconsole popen sink and the backdoor branch are reproduced verbatim from the code excerpts published in the VulnCheck analysis. What is NOT public is the full firmware tree, so the lab rehost implements the documented behaviour (see Notes): the 124007 hashpass invocation is rebuilt from the advisory wording (unescaped password interpolation into hashpass), and the 124014 endpoint shape (action=set/action=get, /etc/xwebconfig.json, raw-body write) follows the advisory text with appliance-style response JSON. The lab proves the causal chain (pre-auth injection -> www-data foothold -> setuid honored on direct wlua execve -> root); it does not claim byte-equality with vendor firmware.
docker compose up -d # vulnerable box on 127.0.0.1:8301, hardened control on 127.0.0.1:8302
# wait ~30 s for the two images to build and boot, then:
python3 poc.py --box vuln --port 8301 --mode check
python3 poc.py --box vuln --port 8301 --mode inject --endpoint login # 124002, root via wlua
python3 poc.py --box vuln --port 8301 --mode inject --endpoint cleanout
python3 poc.py --box vuln --port 8301 --mode inject --endpoint api
python3 poc.py --box vuln --port 8301 --mode password --endpoint login # 124007, root via wlua
python3 poc.py --box vuln --port 8301 --mode backdoor # 124004, admin session
python3 poc.py --box vuln --port 8301 --mode rootexec --command 'id; uname -s -m' # 124010, inline root
python3 poc.py --box vuln --port 8301 --mode config # 124014, write + DoS state
python3 poc.py --box control --port 8302 --mode inject # control: refused
Flags: --target host (default 127.0.0.1 - run it on the lab host, keep it off other people's networks), --port 8301 vulnerable / 8302 control, --box vuln|control sets the expectation the verdict is judged against, --endpoint picks which of the three advisory-named .lua sinks to fire (login/cleanout/api), --command the command run as root (travels base64-wrapped so no quoting breaks the wrapper layers), --config-body raw body for the 124014 write (default: tamper JSON with tampered_by marker), --no-dos skip the SYSTEM_NOT_CONFIGURED step of config mode. Exit code 0 = expectation met. Root markers are written to /opt/xweb/web/proof/ inside the target container and read back over HTTP, so every trigger is provable in-band.
Vulnerable box, CVE-2026-124002 through login.lua (evidence/02-cve-124002-root-vuln.txt):