gh-144563: Fix remote debugging with duplicate libpython mappings from ctypes (#144595)

When _ctypes is imported, it may call dlopen on the libpython shared
library, causing the dynamic linker to load a second mapping of the
library into the process address space. The remote debugging code
iterates memory regions from low addresses upward and returns the first
mapping whose filename matches libpython. After _ctypes is imported, it
finds the dlopen'd copy first, but that copy's PyRuntime section was
never initialized, so reading debug offsets from it fails.

Fix this by validating each candidate PyRuntime address before accepting
it. The validation reads the first 8 bytes and checks for the "xdebugpy"
cookie that is only present in an initialized PyRuntime. Uninitialized
duplicate mappings will fail this check and be skipped, allowing the
search to continue to the real, initialized PyRuntime.

diff --git a/Lib/test/test_external_inspection.py b/Lib/test/test_external_inspection.py
index fe1b5fbe00bbc4650dfb29d8190d4264f0441984..890aa584cd192ca5bcd2e110c2f18699b9846596 100644 (file)
--- a/Lib/test/test_external_inspection.py
+++ b/Lib/test/test_external_inspection.py
@@ -516,6 +516,44 @@ class TestGetStackTrace(RemoteInspectionTestBase):
             finally:
                 _cleanup_sockets(client_socket, server_socket)
 
+    @skip_if_not_supported
+    @unittest.skipIf(
+        sys.platform == "linux" and not PROCESS_VM_READV_SUPPORTED,
+        "Test only runs on Linux with process_vm_readv support",
+    )
+    def test_self_trace_after_ctypes_import(self):
+        """Test that RemoteUnwinder works on the same process after _ctypes import.
+
+        When _ctypes is imported, it may call dlopen on the libpython shared
+        library, creating a duplicate mapping in the process address space.
+        The remote debugging code must skip these uninitialized duplicate
+        mappings and find the real PyRuntime. See gh-144563.
+        """
+        # Run the test in a subprocess to avoid side effects
+        script = textwrap.dedent("""\
+            import os
+            import _remote_debugging
+
+            # Should work before _ctypes import
+            unwinder = _remote_debugging.RemoteUnwinder(os.getpid())
+
+            import _ctypes
+
+            # Should still work after _ctypes import (gh-144563)
+            unwinder = _remote_debugging.RemoteUnwinder(os.getpid())
+            """)
+
+        result = subprocess.run(
+            [sys.executable, "-c", script],
+            capture_output=True,
+            text=True,
+            timeout=SHORT_TIMEOUT,
+        )
+        self.assertEqual(
+            result.returncode, 0,
+            f"stdout: {result.stdout}\nstderr: {result.stderr}"
+        )
+
     @skip_if_not_supported
     @unittest.skipIf(
         sys.platform == "linux" and not PROCESS_VM_READV_SUPPORTED,

diff --git a/Misc/NEWS.d/next/Core_and_Builtins/2026-02-08-18-13-38.gh-issue-144563.hb3kpp.rst b/Misc/NEWS.d/next/Core_and_Builtins/2026-02-08-18-13-38.gh-issue-144563.hb3kpp.rst
new file mode 100644 (file)
index 0000000..023f9dc
--- /dev/null
+++ b/Misc/NEWS.d/next/Core_and_Builtins/2026-02-08-18-13-38.gh-issue-144563.hb3kpp.rst
@@ -0,0 +1,4 @@
+Fix interaction of the Tachyon profiler and :mod:`ctypes` and other modules
+that load the Python shared library (if present) in an independent map as
+this was causing the mechanism that loads the binary information to be
+confused. Patch by Pablo Galindo

diff --git a/Modules/_remote_debugging/asyncio.c b/Modules/_remote_debugging/asyncio.c
index 3fcc939fd0e876b8e764a6652b0047cb0c5b5bed..fc059659511fd8d0c0668adcfe843b1841b7186a 100644 (file)
--- a/Modules/_remote_debugging/asyncio.c
+++ b/Modules/_remote_debugging/asyncio.c
@@ -18,7 +18,8 @@ _Py_RemoteDebug_GetAsyncioDebugAddress(proc_handle_t* handle)
 
 #ifdef MS_WINDOWS
     // On Windows, search for asyncio debug in executable or DLL
-    address = search_windows_map_for_section(handle, "AsyncioD", L"_asyncio");
+    address = search_windows_map_for_section(handle, "AsyncioD", L"_asyncio",
+                                             NULL);
     if (address == 0) {
         // Error out: 'python' substring covers both executable and DLL
         PyObject *exc = PyErr_GetRaisedException();
@@ -27,7 +28,8 @@ _Py_RemoteDebug_GetAsyncioDebugAddress(proc_handle_t* handle)
     }
 #elif defined(__linux__) && HAVE_PROCESS_VM_READV
     // On Linux, search for asyncio debug in executable or DLL
-    address = search_linux_map_for_section(handle, "AsyncioDebug", "python");
+    address = search_linux_map_for_section(handle, "AsyncioDebug", "python",
+                                           NULL);
     if (address == 0) {
         // Error out: 'python' substring covers both executable and DLL
         PyObject *exc = PyErr_GetRaisedException();
@@ -36,10 +38,12 @@ _Py_RemoteDebug_GetAsyncioDebugAddress(proc_handle_t* handle)
     }
 #elif defined(__APPLE__) && TARGET_OS_OSX
     // On macOS, try libpython first, then fall back to python
-    address = search_map_for_section(handle, "AsyncioDebug", "libpython");
+    address = search_map_for_section(handle, "AsyncioDebug", "libpython",
+                                     NULL);
     if (address == 0) {
         PyErr_Clear();
-        address = search_map_for_section(handle, "AsyncioDebug", "python");
+        address = search_map_for_section(handle, "AsyncioDebug", "python",
+                                         NULL);
     }
     if (address == 0) {
         // Error out: 'python' substring covers both executable and DLL

diff --git a/Python/remote_debug.h b/Python/remote_debug.h
index dba6da3bad4197785d6a4dab10f16f806ce71088..4ae1166e885485fec867d605beb06be632e67811 100644 (file)
--- a/Python/remote_debug.h
+++ b/Python/remote_debug.h
@@ -150,6 +150,31 @@ typedef struct {
     Py_ssize_t page_size;
 } proc_handle_t;
 
+// Forward declaration for use in validation function
+static int
+_Py_RemoteDebug_ReadRemoteMemory(proc_handle_t *handle, uintptr_t remote_address, size_t len, void* dst);
+
+// Optional callback to validate a candidate section address found during
+// memory map searches. Returns 1 if the address is valid, 0 to skip it.
+// This allows callers to filter out duplicate/stale mappings (e.g. from
+// ctypes dlopen) whose sections were never initialized.
+typedef int (*section_validator_t)(proc_handle_t *handle, uintptr_t address);
+
+// Validate that a candidate address starts with _Py_Debug_Cookie.
+static int
+_Py_RemoteDebug_ValidatePyRuntimeCookie(proc_handle_t *handle, uintptr_t address)
+{
+    if (address == 0) {
+        return 0;
+    }
+    char buf[sizeof(_Py_Debug_Cookie) - 1];
+    if (_Py_RemoteDebug_ReadRemoteMemory(handle, address, sizeof(buf), buf) != 0) {
+        PyErr_Clear();
+        return 0;
+    }
+    return memcmp(buf, _Py_Debug_Cookie, sizeof(buf)) == 0;
+}
+
 static void
 _Py_RemoteDebug_FreePageCache(proc_handle_t *handle)
 {
@@ -509,7 +534,8 @@ pid_to_task(pid_t pid)
 }
 
 static uintptr_t
-search_map_for_section(proc_handle_t *handle, const char* secname, const char* substr) {
+search_map_for_section(proc_handle_t *handle, const char* secname, const char* substr,
+                       section_validator_t validator) {
     mach_vm_address_t address = 0;
     mach_vm_size_t size = 0;
     mach_msg_type_number_t count = sizeof(vm_region_basic_info_data_64_t);
@@ -561,7 +587,9 @@ search_map_for_section(proc_handle_t *handle, const char* secname, const char* s
         if (strncmp(filename, substr, strlen(substr)) == 0) {
             uintptr_t result = search_section_in_file(
                 secname, map_filename, address, size, proc_ref);
-            if (result != 0) {
+            if (result != 0
+                && (validator == NULL || validator(handle, result)))
+            {
                 return result;
             }
         }
@@ -678,7 +706,8 @@ exit:
 }
 
 static uintptr_t
-search_linux_map_for_section(proc_handle_t *handle, const char* secname, const char* substr)
+search_linux_map_for_section(proc_handle_t *handle, const char* secname, const char* substr,
+                             section_validator_t validator)
 {
     char maps_file_path[64];
     sprintf(maps_file_path, "/proc/%d/maps", handle->pid);
@@ -753,9 +782,12 @@ search_linux_map_for_section(proc_handle_t *handle, const char* secname, const c
 
         if (strstr(filename, substr)) {
             retval = search_elf_file_for_section(handle, secname, start, path);
-            if (retval) {
+            if (retval
+                && (validator == NULL || validator(handle, retval)))
+            {
                 break;
             }
+            retval = 0;
         }
     }
 
@@ -859,7 +891,8 @@ static void* analyze_pe(const wchar_t* mod_path, BYTE* remote_base, const char*
 
 
 static uintptr_t
-search_windows_map_for_section(proc_handle_t* handle, const char* secname, const wchar_t* substr) {
+search_windows_map_for_section(proc_handle_t* handle, const char* secname, const wchar_t* substr,
+                               section_validator_t validator) {
     HANDLE hProcSnap;
     do {
         hProcSnap = CreateToolhelp32Snapshot(TH32CS_SNAPMODULE, handle->pid);
@@ -882,8 +915,11 @@ search_windows_map_for_section(proc_handle_t* handle, const char* secname, const
     for (BOOL hasModule = Module32FirstW(hProcSnap, &moduleEntry); hasModule; hasModule = Module32NextW(hProcSnap, &moduleEntry)) {
         // Look for either python executable or DLL
         if (wcsstr(moduleEntry.szModule, substr)) {
-            runtime_addr = analyze_pe(moduleEntry.szExePath, moduleEntry.modBaseAddr, secname);
-            if (runtime_addr != NULL) {
+            void *candidate = analyze_pe(moduleEntry.szExePath, moduleEntry.modBaseAddr, secname);
+            if (candidate != NULL
+                && (validator == NULL || validator(handle, (uintptr_t)candidate)))
+            {
+                runtime_addr = candidate;
                 break;
             }
         }
@@ -904,7 +940,8 @@ _Py_RemoteDebug_GetPyRuntimeAddress(proc_handle_t* handle)
 
 #ifdef MS_WINDOWS
     // On Windows, search for 'python' in executable or DLL
-    address = search_windows_map_for_section(handle, "PyRuntime", L"python");
+    address = search_windows_map_for_section(handle, "PyRuntime", L"python",
+                                             _Py_RemoteDebug_ValidatePyRuntimeCookie);
     if (address == 0) {
         // Error out: 'python' substring covers both executable and DLL
         PyObject *exc = PyErr_GetRaisedException();
@@ -915,7 +952,8 @@ _Py_RemoteDebug_GetPyRuntimeAddress(proc_handle_t* handle)
     }
 #elif defined(__linux__) && HAVE_PROCESS_VM_READV
     // On Linux, search for 'python' in executable or DLL
-    address = search_linux_map_for_section(handle, "PyRuntime", "python");
+    address = search_linux_map_for_section(handle, "PyRuntime", "python",
+                                           _Py_RemoteDebug_ValidatePyRuntimeCookie);
     if (address == 0) {
         // Error out: 'python' substring covers both executable and DLL
         PyObject *exc = PyErr_GetRaisedException();
@@ -929,7 +967,8 @@ _Py_RemoteDebug_GetPyRuntimeAddress(proc_handle_t* handle)
     const char* candidates[] = {"libpython", "python", "Python", NULL};
     for (const char** candidate = candidates; *candidate; candidate++) {
         PyErr_Clear();
-        address = search_map_for_section(handle, "PyRuntime", *candidate);
+        address = search_map_for_section(handle, "PyRuntime", *candidate,
+                                         _Py_RemoteDebug_ValidatePyRuntimeCookie);
         if (address != 0) {
             break;
         }