Vulnerability GHSA-68w4-83fh-f2w8

Summary

Api.getUserData (legacy) and Api.get_userdata are declared with @permission(Perms.ANY) and are reachable at /api/getUserData and /api/get_userdata. Because Perms.ANY == 0 and pyLoad's permission check is a bitmask AND, that gate is a no-op: every authenticated account passes, including one holding zero permission bits.

Both methods are thin wrappers around check_auth(), which the maintainers deliberately restricted to administrators by omitting @permission (no entry in perm_map, so is_authorized() returns False for non-admins). The wrappers undo that protection.

An attacker holding the lowest-privileged account in the system therefore has a clean binary oracle on the administrator password, with no account lockout anywhere in the codebase, and with the 100 req/min rate limiter bypassable by rotating X-Forwarded-For.

Affected code

src/pyload/core/api/__init__.py:1446 and :1464

    #: Old API
    @permission(Perms.ANY)
    @get
    def getUserData(self, username: str, password: str) -> OldUserData:
        """
        similar to `check_auth` but returns UserData type.
        """
        user = self.check_auth(username, password)
        ...

    @permission(Perms.ANY)
    @get
    def get_userdata(self, username: str, password: str) -> UserData:
        user = self.check_auth(username, password)
        ...

Root cause

src/pyload/core/api/__init__.py:57

class Perms(IntFlag):
    ANY = 0  #: requires no permission, but login

src/pyload/core/api/__init__.py:108

def has_permission(user_perms: Perms, required_perms: Perms):
    return required_perms == (user_perms & required_perms)

For required_perms == 0 this evaluates to 0 == (user_perms & 0) → 0 == 0 → always True. The @permission(Perms.ANY) gate therefore admits every authenticated principal regardless of which permission bits they hold.

Contrast with the intended admin-only primitive, src/pyload/core/api/__init__.py:1396:

    @legacy("checkAuth")
    @get
    def check_auth(self, username: str, password: str) -> dict[str, Any]:

check_auth has no @permission; is_authorized() at api/__init__.py:1430 returns False for non-admins. The two wrappers carry Perms.ANY and restore access for everyone.

Because both wrappers also carry @get, they are placed in method_map and are directly routable via /api/<func>.

Amplifiers

1. No lockout. There is no failed-login counter, delay, or ban anywhere in the codebase. A failed guess costs the attacker only one PBKDF2 computation.

2. Rate limiting is bypassable. /api/* applies rate_limit(count=100, period=60) (src/pyload/webui/app/blueprints/api_blueprint.py:25), which buckets on a fully client-controlled header (src/pyload/webui/app/helpers.py:446):

client_ip = flask.request.headers.get("X-Forwarded-For", "").split(",")[0].strip() \
    or flask.request.remote_addr

Rotating X-Forwarded-For per request yields a fresh bucket each time. Notably, is_loopback_request() in the same file (helpers.py:288-294) explicitly treats the presence of X-Forwarded-For / X-Real-IP / Forwarded as untrustworthy — the same guard was evidently not applied inside rate_limit().

Impact

Online brute force of the administrator account leading to full administrative takeover. A successful response additionally discloses the target account's id, name, email, role, and permission bits.

Proof of concept

Verified against 0.5.0b3, commit a5b008958.

Setup: stock instance with admin pyload, plus a non-admin user bob created with role=USER and permission=0.

Step 1 — establish that bob is genuinely unprivileged:

GET /api/checkAuth?username=pyload&password=pyload   ->  401 {"error": "Access denied"}
GET /api/getAllUserData                              ->  401 {"error": "Access denied"}

Step 2 — the flaw. Same user, same session, Perms.ANY gate:

GET /api/getUserData?username=pyload&password=WRONG
  -> 200 {"name": null, "email": null, "role": null, "permission": null, "template_name": null}

GET /api/getUserData?username=pyload&password=pyload
  -> 200 {"name": "pyload", "email": "", "role": 0, "permission": 0, "template_name": "default"}

role: 0 is Role.ADMIN. get_userdata behaves identically. This is a perfect yes/no oracle.

Step 3 — measured brute force and recovery. Run as bob (permission = 0), admin password set to a 2-character value, X-Forwarded-For rotated on every request:

attempts         : 667
elapsed          : 39.7s   (16.8 guesses/sec)
429 rate-limits  : 0
account lockout  : NONE - same session authenticated throughout
RECOVERED SECRET : 'zq'   (true value 'zq')   match=True

Step 4 — full takeover with the recovered secret:

POST /login as pyload/<recovered>       ->  HTTP 302  (authenticated as admin)
GET  /api/getAllUserData (admin-only)  ->  HTTP 200  (full user dump)

Measured throughput is 17-47 guesses/sec/thread. The only bound is PBKDF2-HMAC-SHA256 at 100,000 iterations in src/pyload/core/database/user_database.py:14, not any rate limit.

Suggested remediation

• Remove @permission(Perms.ANY) from getUserData and get_userdata, or drop them entirely — they are legacy compatibility shims, and modern callers already use the admin-only check_auth.
• Structurally: Perms.ANY = 0 makes has_permission() vacuous, so any method decorated @permission(Perms.ANY) silently becomes public to every logged-in user. Give ANY a real bit value, or handle it explicitly in has_permission() as "requires an authenticated session".
• Do not trust X-Forwarded-For unless a trusted-proxy deployment is explicitly configured. Otherwise bucket rate_limit() on request.remote_addr, or add the same guard is_loopback_request() already uses.
• Add per-account failed-authentication throttling or lockout.
• Consider hmac.compare_digest() in _check_password() (src/pyload/core/database/user_database.py:61 uses a plain == on the derived hash, contradicting the "always use compare_digest" guidance two files away).

Notes for triage

There is no 0.5.0b3 release on PyPI. The develop branch auto-publishes dev builds (setup.py:86 appends .dev<build> to the VERSION file); the newest at time of testing was 0.5.0b3.dev101. The Perms.ANY = 0 design is long-standing, but only the 0.5.0b3.dev* line was verified here — please confirm whether 0.5.0b2.* and 0.4.x are affected before finalizing the version range.