Fix handling of duplicate _PyRuntime symbols when libraries dlopen Python

.github/workflows/build_wheels.yml

@@ -116,6 +116,41 @@ jobs:
           PYTHON_TEST_VERSION: ${{ matrix.python_version }}
         run: python3 -m pytest tests -n auto -vvv
 
+  test_attaching_to_uv_interpreters:
+    needs: [build_wheels]
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        python_version: ["3.13"]
+    steps:
+      - uses: actions/checkout@v5
+      - uses: actions/download-artifact@v5
+        with:
+          name: "manylinux_x86_64-wheels"
+          path: dist
+      - name: Install uv
+        uses: astral-sh/setup-uv@v6
+        with:
+          version: latest
+          python-version: ${{ matrix.python_version }}
+          activate-environment: true
+      - name: Set up dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install -qy gdb
+      - name: Install Python dependencies
+        run: |
+          uv pip install -r requirements-test.txt
+          uv pip install --no-index --find-links=dist/ --only-binary=pystack pystack
+      - name: Disable ptrace security restrictions
+        run: |
+          echo 0 | sudo tee /proc/sys/kernel/yama/ptrace_scope
+      - name: Run pytest
+        env:
+          PYTHON_TEST_VERSION: ${{ matrix.python_version }}
+        run: python3 -m pytest tests -n auto -vvv
+
   test_wheels:
     needs: [build_wheels]
     runs-on: ubuntu-22.04

news/258.bugfix.rst

@@ -0,0 +1 @@
+Fix handling of duplicate ``_PyRuntime`` symbols when ctypes mmaps a statically linked Python interpreter's binary in order to create a trampoline. Previously this led to "Invalid address in remote process" errors.

src/pystack/_pystack/process.cpp

@@ -80,7 +80,7 @@ operator<<(std::ostream& out, const ParsedPyVersion& version)
     // Use a temporary stringstream in case `out` is using hex or showbase
     std::ostringstream oss;
     oss << version.major << "." << version.minor << "." << version.patch;
-    if (version.release_level) {
+    if (version.release_level[0]) {
         oss << version.release_level << version.serial;
     }
 
@@ -97,7 +97,7 @@ parsePyVersionHex(uint64_t version, ParsedPyVersion& parsed)
     int level = (version >> 4) & 0x0F;
     int count = (version >> 0) & 0x0F;
 
-    const char* level_str = nullptr;
+    const char* level_str = "(unknown release level)";
     if (level == 0xA) {
         level_str = "a";
     } else if (level == 0xB) {

src/pystack/_pystack/unwinder.cpp

@@ -378,7 +378,7 @@ module_callback(
             module_arg->addr = addr;
             LOG(INFO) << "Symbol '" << sname << "' found at address " << std::hex << std::showbase
                       << addr;
-            break;
+            return DWARF_CB_ABORT;
         }
     }
     return DWARF_CB_OK;
@@ -389,7 +389,7 @@ AbstractUnwinder::getAddressforSymbol(const std::string& symbol, const std::stri
 {
     LOG(DEBUG) << "Trying to find address for symbol " << symbol;
     ModuleArg arg = {symbol.c_str(), modulename.c_str(), 0};
-    if (dwfl_getmodules(Dwfl(), module_callback, &arg, 0) != 0) {
+    if (dwfl_getmodules(Dwfl(), module_callback, &arg, 0) == -1) {
         throw UnwinderError("Failed to fetch modules!");
     }
     LOG(DEBUG) << "Address for symbol " << symbol << " resolved to: " << std::hex << std::showbase

tests/conftest.py

@@ -0,0 +1,8 @@
+import shutil
+
+import pytest
+
+
+def pytest_sessionstart(session):
+    if not shutil.which("gcore"):
+        pytest.exit("gcore not found (you probably forgot to install gdb)")

tests/integration/ctypes_program.py

@@ -0,0 +1,24 @@
+import ctypes
+import sys
+import time
+
+
+def first_func():
+    second_func()
+
+
+def second_func():
+    third_func()
+
+
+def third_func():
+    # Trigger libffi to re-import the Python binary
+    global gil_check
+    gil_check = ctypes.CFUNCTYPE(ctypes.c_int)(ctypes.CDLL(None).PyGILState_Check)
+
+    with open(sys.argv[1], "w") as fifo:
+        fifo.write("ready")
+    time.sleep(1000)
+
+
+first_func()

tests/integration/test_shenanigans.py

@@ -0,0 +1,35 @@
+import sys
+from pathlib import Path
+
+from pystack.engine import get_process_threads
+from tests.utils import spawn_child_process
+
+TEST_DUPLICATE_SYMBOLS_FILE = Path(__file__).parent / "ctypes_program.py"
+
+
+def test_duplicate_pyruntime_symbol_handling(tmpdir):
+    """Test that pystack correctly handles duplicate _PyRuntime symbols.
+
+    This can occur when ctypes uses libffi to dlopen the Python binary
+    in order to create a trampoline (which it only does if the Python binary
+    was statically linked against libpython).
+    """
+    # GIVEN
+    with spawn_child_process(
+        sys.executable, TEST_DUPLICATE_SYMBOLS_FILE, tmpdir
+    ) as child_process:
+        # WHEN
+        threads = list(get_process_threads(child_process.pid, stop_process=True))
+
+    # THEN
+    # We should have successfully resolved threads without "Invalid address" errors
+    assert threads is not None
+    assert len(threads) > 0
+
+    # Verify we can get stack traces (which requires correct _PyRuntime)
+    for thread in threads:
+        # Just ensure we can get frames without crashing
+        frames = list(thread.frames)
+        # The main thread should have at least one frame
+        if thread.tid == child_process.pid:
+            assert len(frames) > 0