Files
moloapp/audit
..

audit

Scope: README ยท AGENTS

A fixed, named set of regression checks for vulnerability classes relevant to the mechanisms MoloApp actually has (an in-house HTTP/1 and HTTP/2 server, an in-house WebSocket/WebDAV/reverse-tunnel realtime stack, a shell-invoking CLI, in-house crypto and an ORM), each in its own directory with where the problem lives, why it matters to MoloApp specifically, and a check that reproduces it against real, built artifacts. Modeled on the audit/ convention used by the doos project: capped to the issues that target code paths MoloApp actually has, not "every CVE", consistent with the base rule that every automated process gets a fixed, named ceiling instead of an open-ended scope.

Most of these are in-house findings from reading this codebase, not public CVE assignments (MoloApp is not itself a CVE-tracked project); each README still cites the real-world vulnerability class it is modeled on, with sources, so the check's severity and shape can be judged against known precedent.

A check here answers one question: does this build, today, reproduce the exposure described. It is not a claim that MoloApp is unaffected by anything not listed here.

Running it

make                               # build/bin/runtime_interop_server, build/example/living-server, build/molo
make -C audit test                 # sequential
make -C audit -j8 test             # concurrent, one job per check
make -C audit report               # runs all, never stops, one line prefix per check
make -C audit list                 # names the check directories

Each check is independent, writes only under its own mktemp -d (Python checks) or is a pure in-memory harness (C++ checks), and cleans up on exit. A check that needs something the host does not have (an unbuilt binary) prints SKIP with the reason and exits 0; it never reports PASS for a precondition it could not test. Every check runs under a hard wall-clock watchdog (audit/lib/molo_audit.py's arm_watchdog, or an alarm() call in the C++ harnesses) so a check that finds a genuine hang reports FAIL instead of wedging the suite.

Verdicts

Verdict Meaning
PASS This build does not reproduce the exposure described, under the conditions this check creates.
FAIL It does. Treat as a regression; the check prints what it observed.
SKIP The check's precondition was not met on this host (missing build). Not a verdict on MoloApp.

How a check drives the real thing

Three kinds of real artifact, no mocks:

  • Over a real socket, against build/bin/runtime_interop_server (the runtime's own interop test server: real HTTP/1.1, real h2c HTTP/2, real compression, real static files, nothing stubbed) or build/example/living-server. These checks live in audit/lib/molo_audit.py's H2/h1 helpers: a minimal HTTP/2 client speaking real frames, and raw-socket HTTP/1.1, both stdlib-only Python.
  • Through the real CLI, build/molo, run as a subprocess against a scratch directory.
  • Against the real class directly, for the handful of classes (src/realtime/ websocket/server/handshake_validator.cpp, src/realtime/webdav/server/ resource_tree.cpp, src/realtime/expose/upstream.cpp) that are pure, I/O-free logic not yet wired to any route in the shipped example (example/fw/ is explicitly a skeleton still being re-bound to src/, per the main README). Each such check's README says so and compiles a tiny harness straight from the real .cpp file, the same code any real route would call, rather than skipping the class entirely.

What is deliberately not here

No kernel-level memory-corruption exploit code, and no check that requires outbound network access (DNS resolution, reaching a real external host) to prove its point. Everything here drives a MoloApp artifact through its real inputs inside a scratch directory or a loopback socket; nothing here targets any host other than the one running the check, and nothing here is a weaponized exploit for an unrelated target.

The list

Check Mechanism Modeled on
MOLO-H1-001 HTTP/1 chunked-body cap bypass via chunk-size addition overflow CWE-190 / request-smuggling-adjacent resource exhaustion
MOLO-H2-001 HTTP/2 stream-open/reset churn, no rate limit CVE-2023-44487 "Rapid Reset"
MOLO-H2-002 HTTP/2 server-triggered reset churn CVE-2025-8671 "MadeYouReset"
MOLO-H2-003 HTTP/2 empty-CONTINUATION flood, byte cap only CVE-2024-27316 "CONTINUATION flood"
MOLO-H2-004 HTTP/2 PING flood to an unread peer, no queue cap CVE-2019-9511/9512/9515/9518 HTTP/2 DoS batch
MOLO-STATIC-001 Static-file jail escape via a symlinked intermediate directory CWE-61 symlink following
MOLO-CLI-001 Shell command injection in molo new <name> CWE-78 OS command injection
MOLO-WS-001 WebSocket handshake has no Origin check CWE-1385 / CVE-2024-23898-class CSWSH
MOLO-DAV-001 WebDAV COPY into its own subtree never terminates CWE-674 uncontrolled recursion
MOLO-EXPOSE-001 expose upstream allow-list bypassed by hostname prefix CWE-184 incomplete allow-list / SSRF-adjacent

Back to README, AGENTS.