gh-78502: Add a trackfd parameter to mmap.mmap() (GH-25425)

If *trackfd* is False, the file descriptor specified by *fileno*
will not be duplicated.

Co-authored-by: Erlend E. Aasland <[email protected]>
Co-authored-by: Petr Viktorin <[email protected]>
Co-authored-by: Serhiy Storchaka <[email protected]>

Author: 4 people

Doc/library/mmap.rst

@@ -48,7 +48,7 @@ update the underlying file.
 
 To map anonymous memory, -1 should be passed as the fileno along with the length.
 
-.. class:: mmap(fileno, length, tagname=None, access=ACCESS_DEFAULT[, offset])
+.. class:: mmap(fileno, length, tagname=None, access=ACCESS_DEFAULT, offset=0)
 
    **(Windows version)** Maps *length* bytes from the file specified by the
    file handle *fileno*, and creates a mmap object.  If *length* is larger
@@ -71,7 +71,8 @@ To map anonymous memory, -1 should be passed as the fileno along with the length
 
    .. audit-event:: mmap.__new__ fileno,length,access,offset mmap.mmap
 
-.. class:: mmap(fileno, length, flags=MAP_SHARED, prot=PROT_WRITE|PROT_READ, access=ACCESS_DEFAULT[, offset])
+.. class:: mmap(fileno, length, flags=MAP_SHARED, prot=PROT_WRITE|PROT_READ, \
+                access=ACCESS_DEFAULT, offset=0, *, trackfd=True)
    :noindex:
 
    **(Unix version)** Maps *length* bytes from the file specified by the file
@@ -102,10 +103,20 @@ To map anonymous memory, -1 should be passed as the fileno along with the length
    defaults to 0. *offset* must be a multiple of :const:`ALLOCATIONGRANULARITY`
    which is equal to :const:`PAGESIZE` on Unix systems.
 
+   If *trackfd* is ``False``, the file descriptor specified by *fileno* will
+   not be duplicated, and the resulting :class:`!mmap` object will not
+   be associated with the map's underlying file.
+   This means that the :meth:`~mmap.mmap.size` and :meth:`~mmap.mmap.resize`
+   methods will fail.
+   This mode is useful to limit the number of open file descriptors.
+
    To ensure validity of the created memory mapping the file specified
    by the descriptor *fileno* is internally automatically synchronized
    with the physical backing store on macOS.
 
+   .. versionchanged:: 3.13
+      The *trackfd* parameter was added.
+
    This example shows a simple way of using :class:`~mmap.mmap`::
 
       import mmap
@@ -254,9 +265,12 @@ To map anonymous memory, -1 should be passed as the fileno along with the length
 
    .. method:: resize(newsize)
 
-      Resizes the map and the underlying file, if any. If the mmap was created
-      with :const:`ACCESS_READ` or :const:`ACCESS_COPY`, resizing the map will
-      raise a :exc:`TypeError` exception.
+      Resizes the map and the underlying file, if any.
+
+      Resizing a map created with *access* of :const:`ACCESS_READ` or
+      :const:`ACCESS_COPY`, will raise a :exc:`TypeError` exception.
+      Resizing a map created with with *trackfd* set to ``False``,
+      will raise a :exc:`ValueError` exception.
 
       **On Windows**: Resizing the map will raise an :exc:`OSError` if there are other
       maps against the same named file. Resizing an anonymous map (ie against the

Doc/whatsnew/3.13.rst

@@ -254,6 +254,9 @@ mmap
   that can be used where it requires a file-like object with seekable and
   the :meth:`~mmap.mmap.seek` method return the new absolute position.
   (Contributed by Donghee Na and Sylvie Liberman in :gh:`111835`.)
+* :class:`mmap.mmap` now has a *trackfd* parameter on Unix; if it is ``False``,
+  the file descriptor specified by *fileno* will not be duplicated.
+  (Contributed by Zackery Spytz and Petr Viktorin in :gh:`78502`.)
 
 opcode
 ------

Lib/test/test_mmap.py

@@ -4,6 +4,7 @@
 from test.support.import_helper import import_module
 from test.support.os_helper import TESTFN, unlink
 import unittest
+import errno
 import os
 import re
 import itertools
@@ -266,6 +267,62 @@ def test_access_parameter(self):
                     self.assertRaises(TypeError, m.write_byte, 0)
                     m.close()
 
+    @unittest.skipIf(os.name == 'nt', 'trackfd not present on Windows')
+    def test_trackfd_parameter(self):
+        size = 64
+        with open(TESTFN, "wb") as f:
+            f.write(b"a"*size)
+        for close_original_fd in True, False:
+            with self.subTest(close_original_fd=close_original_fd):
+                with open(TESTFN, "r+b") as f:
+                    with mmap.mmap(f.fileno(), size, trackfd=False) as m:
+                        if close_original_fd:
+                            f.close()
+                        self.assertEqual(len(m), size)
+                        with self.assertRaises(OSError) as err_cm:
+                            m.size()
+                        self.assertEqual(err_cm.exception.errno, errno.EBADF)
+                        with self.assertRaises(ValueError):
+                            m.resize(size * 2)
+                        with self.assertRaises(ValueError):
+                            m.resize(size // 2)
+                        self.assertEqual(m.closed, False)
+
+                        # Smoke-test other API
+                        m.write_byte(ord('X'))
+                        m[2] = ord('Y')
+                        m.flush()
+                        with open(TESTFN, "rb") as f:
+                            self.assertEqual(f.read(4), b'XaYa')
+                        self.assertEqual(m.tell(), 1)
+                        m.seek(0)
+                        self.assertEqual(m.tell(), 0)
+                        self.assertEqual(m.read_byte(), ord('X'))
+
+                self.assertEqual(m.closed, True)
+                self.assertEqual(os.stat(TESTFN).st_size, size)
+
+    @unittest.skipIf(os.name == 'nt', 'trackfd not present on Windows')
+    def test_trackfd_neg1(self):
+        size = 64
+        with mmap.mmap(-1, size, trackfd=False) as m:
+            with self.assertRaises(OSError):
+                m.size()
+            with self.assertRaises(ValueError):
+                m.resize(size // 2)
+            self.assertEqual(len(m), size)
+            m[0] = ord('a')
+            assert m[0] == ord('a')
+
+    @unittest.skipIf(os.name != 'nt', 'trackfd only fails on Windows')
+    def test_no_trackfd_parameter_on_windows(self):
+        # 'trackffd' is an invalid keyword argument for this function
+        size = 64
+        with self.assertRaises(TypeError):
+            mmap.mmap(-1, size, trackfd=True)
+        with self.assertRaises(TypeError):
+            mmap.mmap(-1, size, trackfd=False)
+
     def test_bad_file_desc(self):
         # Try opening a bad file descriptor...
         self.assertRaises(OSError, mmap.mmap, -2, 4096)

Misc/NEWS.d/next/Library/2021-04-15-10-41-51.bpo-34321.36m6_l.rst

@@ -0,0 +1,2 @@
+:class:`mmap.mmap` now has a *trackfd* parameter on Unix; if it is
+``False``, the file descriptor specified by *fileno* will not be duplicated.

Modules/mmapmodule.c

@@ -117,6 +117,7 @@ typedef struct {
 
 #ifdef UNIX
     int fd;
+    _Bool trackfd;
 #endif
 
     PyObject *weakreflist;
@@ -393,6 +394,13 @@ is_resizeable(mmap_object *self)
             "mmap can't resize with extant buffers exported.");
         return 0;
     }
+#ifdef UNIX
+    if (!self->trackfd) {
+        PyErr_SetString(PyExc_ValueError,
+            "mmap can't resize with trackfd=False.");
+        return 0;
+    }
+#endif
     if ((self->access == ACCESS_WRITE) || (self->access == ACCESS_DEFAULT))
         return 1;
     PyErr_Format(PyExc_TypeError,
@@ -1154,7 +1162,7 @@ is 0, the maximum length of the map is the current size of the file,\n\
 except that if the file is empty Windows raises an exception (you cannot\n\
 create an empty mapping on Windows).\n\
 \n\
-Unix: mmap(fileno, length[, flags[, prot[, access[, offset]]]])\n\
+Unix: mmap(fileno, length[, flags[, prot[, access[, offset[, trackfd]]]]])\n\
 \n\
 Maps length bytes from the file specified by the file descriptor fileno,\n\
 and returns a mmap object.  If length is 0, the maximum length of the map\n\
@@ -1221,15 +1229,17 @@ new_mmap_object(PyTypeObject *type, PyObject *args, PyObject *kwdict)
     off_t offset = 0;
     int fd, flags = MAP_SHARED, prot = PROT_WRITE | PROT_READ;
     int devzero = -1;
-    int access = (int)ACCESS_DEFAULT;
+    int access = (int)ACCESS_DEFAULT, trackfd = 1;
     static char *keywords[] = {"fileno", "length",
                                "flags", "prot",
-                               "access", "offset", NULL};
+                               "access", "offset", "trackfd", NULL};
 
-    if (!PyArg_ParseTupleAndKeywords(args, kwdict, "in|iii" _Py_PARSE_OFF_T, keywords,
+    if (!PyArg_ParseTupleAndKeywords(args, kwdict,
+                                     "in|iii" _Py_PARSE_OFF_T "$p", keywords,
                                      &fd, &map_size, &flags, &prot,
-                                     &access, &offset))
+                                     &access, &offset, &trackfd)) {
         return NULL;
+    }
     if (map_size < 0) {
         PyErr_SetString(PyExc_OverflowError,
                         "memory mapped length must be positive");
@@ -1325,6 +1335,7 @@ new_mmap_object(PyTypeObject *type, PyObject *args, PyObject *kwdict)
     m_obj->weakreflist = NULL;
     m_obj->exports = 0;
     m_obj->offset = offset;
+    m_obj->trackfd = trackfd;
     if (fd == -1) {
         m_obj->fd = -1;
         /* Assume the caller wants to map anonymous memory.
@@ -1350,13 +1361,16 @@ new_mmap_object(PyTypeObject *type, PyObject *args, PyObject *kwdict)
         }
 #endif
     }
-    else {
+    else if (trackfd) {
         m_obj->fd = _Py_dup(fd);
         if (m_obj->fd == -1) {
             Py_DECREF(m_obj);
             return NULL;
         }
     }
+    else {
+        m_obj->fd = -1;
+    }
 
     Py_BEGIN_ALLOW_THREADS
     m_obj->data = mmap(NULL, map_size, prot, flags, fd, offset);