GHSA-GM9R-HVC9-96GV
Vulnerability from github – Published: 2026-11-19 22:50 – Updated: 2026-11-19 22:50Summary
_extract_from_forwarded_header() (guard_core/utils.py:172-181 in 3.12.0; :179 on master 08780f1) returns the selected X-Forwarded-For entry without any ipaddress validation. When that entry is attacker-controlled, an arbitrary string becomes client_ip and flows into BehaviorTracker (guard_core/handlers/behavior_handler.py:83-102), which interpolates it verbatim into both a Redis key name and a KEYS glob pattern:
key = f"behavior:usage:{endpoint_id}:{client_ip}"
await self.redis_handler.set_key("behavior_usage", f"{key}:{current_time}", "1", ttl=rule.window)
pattern = f"behavior_usage:{key}:*"
keys = await self.redis_handler.keys(pattern) # blocking KEYS, O(N) over keyspace
Same shape in track_return_pattern() (behavior_handler.py:135-142). redis_handler.keys() maps to the blocking Redis KEYS command (guard_core/handlers/redis_handler.py:265-275).
Details
Three consequences, in increasing severity:
- Cross-client counter contamination.
X-Forwarded-For: *makes the glob match every client's behavior-usage keys for the endpoint. Verified with a fake-Redis harness: a client9.9.9.9hit an endpoint 3 times (threshold=3 rule), then a single attacker request carryingXFF: *issued the patternbehavior_usage:behavior:usage:app.login:*:*, counted the victim's events, and tripped the rule on its first request. An embedded:also shifts which key segment is parsed as the timestamp (behavior_handler.py:94-100). - Unbounded key-space growth + blocking scans. Each request creates a new unique key (
f"{key}:{current_time}") and then issues a blockingKEYSscan whose cost grows with total keys — on the shared Redis instance, degrading every guard-core instance using it. Rotating arbitrary XFF strings multiplies distinct keys without bound. - Error path. A tripped
banaction passes the non-IP string toip_ban_manager.ban_ip(), whoseipaddressvalidation raisesValueErroruncaught in the behavioral processor — the request 500s.
PoC
import asyncio
from types import SimpleNamespace
from guard_core.models import SecurityConfig
from guard_core.utils import extract_client_ip
class CIHeaders(dict):
def get(self, key, default=None):
for k, v in self.items():
if k.lower() == key.lower():
return v
return default
class FakeRequest:
def __init__(self, xff):
self.state = SimpleNamespace()
self.client_host = "10.0.0.1" # the TRUSTED proxy
self.headers = CIHeaders({"X-Forwarded-For": xff})
async def main():
config = SecurityConfig(trusted_proxies=("10.0.0.1",), enable_redis=False)
ip = await extract_client_ip(FakeRequest("*"), config)
print(repr(ip)) # -> '*' (arbitrary string, no validation)
asyncio.run(main())
Then BehaviorTracker.track_endpoint_usage("app.login", "*", rule) issues KEYS behavior_usage:behavior:usage:app.login:*:* — matching other clients' keys (full fake-Redis script: poc_xff_redis.py, available on request).
Impact
Integrity and availability of the behavioral-analytics subsystem and the shared Redis instance, in deployments where the extracted XFF position is attacker-controllable. No data exposure; not reachable with default settings (trusted_proxies defaults empty).
Fix: validate the extracted XFF entry with ipaddress.ip_address() before returning it from extract_client_ip() (fall back to the connecting IP on failure); escape glob metacharacters (*?[]) in anything interpolated into a KEYS pattern — or replace the KEYS+per-request-key design with an atomic counter / Lua sliding window (the rate-limit Lua script already shows the right shape).
{
"affected": [
{
"package": {
"ecosystem": "PyPI",
"name": "guard-core"
},
"ranges": [
{
"events": [
{
"introduced": "1.0.0"
},
{
"fixed": "3.13.0"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-129304"
],
"database_specific": {
"cwe_ids": [
"CWE-20"
],
"github_reviewed": true,
"github_reviewed_at": "2026-11-19T22:50:33Z",
"nvd_published_at": "2026-11-16T20:19:01Z",
"severity": "MODERATE"
},
"details": "### Summary\n\n`_extract_from_forwarded_header()` (`guard_core/utils.py:172-181` in 3.12.0; `:179` on master `08780f1`) returns the selected `X-Forwarded-For` entry **without any `ipaddress` validation**. When that entry is attacker-controlled, an arbitrary string becomes `client_ip` and flows into `BehaviorTracker` (`guard_core/handlers/behavior_handler.py:83-102`), which interpolates it verbatim into both a Redis key name and a `KEYS` glob pattern:\n\n```python\nkey = f\"behavior:usage:{endpoint_id}:{client_ip}\"\nawait self.redis_handler.set_key(\"behavior_usage\", f\"{key}:{current_time}\", \"1\", ttl=rule.window)\npattern = f\"behavior_usage:{key}:*\"\nkeys = await self.redis_handler.keys(pattern) # blocking KEYS, O(N) over keyspace\n```\n\nSame shape in `track_return_pattern()` (`behavior_handler.py:135-142`). `redis_handler.keys()` maps to the blocking Redis `KEYS` command (`guard_core/handlers/redis_handler.py:265-275`).\n\n### Details\n\nThree consequences, in increasing severity:\n\n1. **Cross-client counter contamination.** `X-Forwarded-For: *` makes the glob match *every* client\u0027s behavior-usage keys for the endpoint. Verified with a fake-Redis harness: a client `9.9.9.9` hit an endpoint 3 times (threshold=3 rule), then a single attacker request carrying `XFF: *` issued the pattern `behavior_usage:behavior:usage:app.login:*:*`, counted the victim\u0027s events, and tripped the rule on its **first** request. An embedded `:` also shifts which key segment is parsed as the timestamp (`behavior_handler.py:94-100`).\n2. **Unbounded key-space growth + blocking scans.** Each request creates a **new unique key** (`f\"{key}:{current_time}\"`) and then issues a blocking `KEYS` scan whose cost grows with total keys \u2014 on the shared Redis instance, degrading every guard-core instance using it. Rotating arbitrary XFF strings multiplies distinct keys without bound.\n3. **Error path.** A tripped `ban` action passes the non-IP string to `ip_ban_manager.ban_ip()`, whose `ipaddress` validation raises `ValueError` uncaught in the behavioral processor \u2014 the request 500s.\n\n### PoC\n\n```python\nimport asyncio\nfrom types import SimpleNamespace\nfrom guard_core.models import SecurityConfig\nfrom guard_core.utils import extract_client_ip\n\nclass CIHeaders(dict):\n def get(self, key, default=None):\n for k, v in self.items():\n if k.lower() == key.lower():\n return v\n return default\n\nclass FakeRequest:\n def __init__(self, xff):\n self.state = SimpleNamespace()\n self.client_host = \"10.0.0.1\" # the TRUSTED proxy\n self.headers = CIHeaders({\"X-Forwarded-For\": xff})\n\nasync def main():\n config = SecurityConfig(trusted_proxies=(\"10.0.0.1\",), enable_redis=False)\n ip = await extract_client_ip(FakeRequest(\"*\"), config)\n print(repr(ip)) # -\u003e \u0027*\u0027 (arbitrary string, no validation)\n\nasyncio.run(main())\n```\n\nThen `BehaviorTracker.track_endpoint_usage(\"app.login\", \"*\", rule)` issues `KEYS behavior_usage:behavior:usage:app.login:*:*` \u2014 matching other clients\u0027 keys (full fake-Redis script: `poc_xff_redis.py`, available on request).\n\n### Impact\n\nIntegrity and availability of the behavioral-analytics subsystem and the shared Redis instance, in deployments where the extracted XFF position is attacker-controllable. No data exposure; not reachable with default settings (`trusted_proxies` defaults empty).\n\n**Fix:** validate the extracted XFF entry with `ipaddress.ip_address()` before returning it from `extract_client_ip()` (fall back to the connecting IP on failure); escape glob metacharacters (`*?[]`) in anything interpolated into a `KEYS` pattern \u2014 or replace the KEYS+per-request-key design with an atomic counter / Lua sliding window (the rate-limit Lua script already shows the right shape).",
"id": "GHSA-gm9r-hvc9-96gv",
"modified": "2026-11-19T22:50:33Z",
"published": "2026-11-19T22:50:33Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/Guard-Core/guard-core/security/advisories/GHSA-gm9r-hvc9-96gv"
},
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-129304"
},
{
"type": "WEB",
"url": "https://github.com/Guard-Core/guard-core/pull/68"
},
{
"type": "WEB",
"url": "https://github.com/Guard-Core/guard-core/commit/c07239fa1c5aa93e3f179440e79dad652977d3e4"
},
{
"type": "WEB",
"url": "https://github.com/Guard-Core/guard-core/releases/tag/3.13.0"
},
{
"type": "PACKAGE",
"url": "https://github.com/rennf93/guard-core"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:C/C:N/I:L/A:L",
"type": "CVSS_V3"
}
],
"summary": "Guard Core: Unvalidated X-Forwarded-For-derived string used as a Redis key component and interpolated into a KEYS glob pattern"
}
Drag nodes to arrange them. Double-click a related vulnerability to open it, or a source to visit its link. The Physics button on the left tunes the layout and the simulation; the legend at the bottom left hides or shows nodes by kind and by sighting type.
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.
Vulnerability timeline
Sightings and recorded dates, grouped by day · UTC
How durations are calculated
Following the VTEM framework (Black Hat USA 2015), time to patch starts when the vendor or a coordinator was informed, total time to patch at discovery, and disclosure to patch at public disclosure; vendor response time runs from the report to its acknowledgement. Time to exploit and time unpatched start at the earliest discovery or report, falling back to public disclosure. Unpatched time ends at the first recorded patch, or the calculation time when no patch is recorded. Exploitation to patch shows the interval between those two events; a published proof of concept is a public exploit, not exploitation.
Each result shows its source dates. A negative interval means the end event came first; when date-only inputs leave the order open, the result says so. Date-only inputs are approximate. The earliest evidence wins: KEV report dates and sightings are labelled as such. Records describing another vulnerability are listed but never calculated from, and a PoC alone does not establish exploitation. Missing patch evidence does not prove that no patch exists. Filters below change the event list only.
Loading timeline…
The approach is described in our paper Mapping CVEs to MITRE ATT&CK Techniques: A Curated Gold-Set Classifier and the Limits of LLM-Assisted Label Expansion.
Browse all ATT&CK techniques and the vulnerabilities related to each.
Related by attack behaviour
Vulnerabilities whose description is nearest to this one in the vector space of the CIRCL/vulnerability-attack-technique-biencoder model. This is a similarity search over the bi-encoder space (plain cosine), not a classification, and it has no measured accuracy.