From db6eea60f23c0bca5da76469aaa8bc15a1310fd9 Mon Sep 17 00:00:00 2001
From: arpita-1111 <arpitasoni649@gmail.com>
Date: Tue, 28 Jul 2026 00:30:59 +0530
Subject: [PATCH] fix(security): sandbox repo verification in Docker to prevent
 RCE (#211)

---
 backend/app/sandbox/verify.py | 28 +++++++------
 backend/app/utils/exec.py     | 77 +++++++++++++++++++++++++++++++++++
 2 files changed, 92 insertions(+), 13 deletions(-)

diff --git a/backend/app/sandbox/verify.py b/backend/app/sandbox/verify.py
index 8d27d0d..07c6a01 100644
--- a/backend/app/sandbox/verify.py
+++ b/backend/app/sandbox/verify.py
@@ -7,7 +7,7 @@
 from typing import Any, Dict, List, Optional
 
 from ..models import VerifyResponse
-from ..utils.exec import run_cmd
+from ..utils.exec import run_cmd, run_cmd_sandboxed
 
 
 def _find_npm_command() -> Optional[str]:
@@ -52,11 +52,10 @@ def verify_repo(repo_dir: Path) -> VerifyResponse:
     checks: Dict[str, Any] = {}
 
     if (repo_dir / "package.json").exists():
-        npm_path = _find_npm_command()
-        if not npm_path:
-            checks["npm"] = {
+        if shutil.which("docker") is None:
+            checks["docker"] = {
                 "ok": False,
-                "reason": "npm not found on PATH for backend process",
+                "reason": "docker not found on PATH; sandboxed verification requires Docker",
             }
             return VerifyResponse(ok=False, checks=checks)
 
@@ -65,12 +64,12 @@ def verify_repo(repo_dir: Path) -> VerifyResponse:
                 repo_dir / "npm-shrinkwrap.json"
             ).exists()
             if has_lock:
-                r_ci = run_cmd([npm_path, "ci"], cwd=repo_dir, timeout_s=1200)
+                r_ci = run_cmd_sandboxed(["npm", "ci"], cwd=repo_dir, timeout_s=1200)
                 checks["npm_ci"] = r_ci
                 if r_ci.get("returncode") != 0:
                     return VerifyResponse(ok=False, checks=checks)
             else:
-                r_i = run_cmd([npm_path, "install"], cwd=repo_dir, timeout_s=1200)
+                r_i = run_cmd_sandboxed(["npm", "install"], cwd=repo_dir, timeout_s=1200)
                 checks["npm_install"] = r_i
                 if r_i.get("returncode") != 0:
                     return VerifyResponse(ok=False, checks=checks)
@@ -96,7 +95,7 @@ def verify_repo(repo_dir: Path) -> VerifyResponse:
             }
             return VerifyResponse(ok=True, checks=checks)
 
-        r = run_cmd([npm_path, *chosen], cwd=repo_dir, timeout_s=1200)
+        r = run_cmd_sandboxed(["npm", *chosen], cwd=repo_dir, timeout_s=1200)
         checks["npm_verify"] = {
             "selected": chosen,
             "available_scripts": sorted(list(scripts.keys())),
@@ -116,15 +115,18 @@ def verify_repo(repo_dir: Path) -> VerifyResponse:
             }
             return VerifyResponse(ok=True, checks=checks)
 
-        r = run_cmd(["pytest", "-q"], cwd=repo_dir, timeout_s=600)
-        checks["pytest"] = r
-        ok = r.get("returncode") == 0
-        return VerifyResponse(ok=ok, checks=checks)
+    r = run_cmd_sandboxed(
+                ["pytest", "-q"], cwd=repo_dir, timeout_s=600, image="python:3.11-slim"
+            )
+    checks["pytest"] = r
+    ok = r.get("returncode") == 0
+    return VerifyResponse(ok=ok, checks=checks)
 
-    r = run_cmd(
+    r = run_cmd_sandboxed(
         ["python", "-c", "print('verify: no tests detected')"],
         cwd=repo_dir,
         timeout_s=30,
+        image="python:3.11-slim"
     )
     checks["fallback"] = r
     return VerifyResponse(ok=True, checks=checks)
diff --git a/backend/app/utils/exec.py b/backend/app/utils/exec.py
index 6432341..f7b8cc0 100644
--- a/backend/app/utils/exec.py
+++ b/backend/app/utils/exec.py
@@ -48,3 +48,80 @@ def run_cmd(cmd: List[str], cwd: Path, timeout_s: int = 300) -> CmdResult:
             stderr=f"TimeoutExpired: {e}",
             ok=False,
         )
+
+
+def run_cmd_sandboxed(
+    cmd: List[str],
+    cwd: Path,
+    timeout_s: int = 300,
+    image: str = "node:20-alpine",
+    network: bool = False,
+) -> CmdResult:
+    """
+    Run `cmd` inside an ephemeral, non-root, resource-limited Docker
+    container instead of directly on the host. Prevents untrusted
+    repository code from achieving remote code execution (issue #211).
+    """
+    docker_cmd = [
+        "docker",
+        "run",
+        "--rm",
+        "--user",
+        "1000:1000",
+        "--memory",
+        "512m",
+        "--cpus",
+        "1",
+        "--pids-limit",
+        "256",
+        "--security-opt",
+        "no-new-privileges",
+        "--cap-drop",
+        "ALL",
+        "-v",
+        f"{cwd.resolve()}:/workspace",
+        "-w",
+        "/workspace",
+    ]
+
+    if not network:
+        docker_cmd += ["--network", "none"]
+
+    docker_cmd += [image, *cmd]
+
+    env = os.environ.copy()
+
+    try:
+        p = subprocess.run(
+            docker_cmd,
+            capture_output=True,
+            text=True,
+            timeout=timeout_s,
+            env=env,
+        )
+        return CmdResult(
+            cmd=cmd,
+            returncode=p.returncode,
+            stdout=p.stdout,
+            stderr=p.stderr,
+            ok=True,
+            sandboxed=True,
+        )
+    except OSError as e:
+        return CmdResult(
+            cmd=cmd,
+            returncode=127,
+            stdout="",
+            stderr=str(e),
+            ok=False,
+            sandboxed=True,
+        )
+    except subprocess.TimeoutExpired as e:
+        return CmdResult(
+            cmd=cmd,
+            returncode=124,
+            stdout=getattr(e, "stdout", "") or "",
+            stderr=f"TimeoutExpired: {e}",
+            ok=False,
+            sandboxed=True,
+        )
\ No newline at end of file