[3.13] gh-119452: Fix a potential virtual memory allocation denial of service in...

The CGI server on Windows could consume the amount of memory specified
in the Content-Length header of the request even if the client does not
send such much data. Now it reads the POST request body by chunks,
so that the memory consumption is proportional to the amount of sent
data.
(cherry picked from commit 29c657a1f231c0908796e0c9ff6967e15ab20d9b)

Co-authored-by: Serhiy Storchaka <storchaka@gmail.com>

diff --git a/Lib/http/server.py b/Lib/http/server.py
index 758a725f1fd540741e9b9c679af196796a48b84b..f9d676c8eb1861a7f2f334130dd7f459c6443e43 100644 (file)
--- a/Lib/http/server.py
+++ b/Lib/http/server.py
@@ -127,6 +127,10 @@ DEFAULT_ERROR_MESSAGE = """\
 
 DEFAULT_ERROR_CONTENT_TYPE = "text/html;charset=utf-8"
 
+# Data larger than this will be read in chunks, to prevent extreme
+# overallocation.
+_MIN_READ_BUF_SIZE = 1 << 20
+
 class HTTPServer(socketserver.TCPServer):
 
     allow_reuse_address = 1    # Seems to make sense in testing environment
@@ -1234,7 +1238,16 @@ class CGIHTTPRequestHandler(SimpleHTTPRequestHandler):
                                  env = env
                                  )
             if self.command.lower() == "post" and nbytes > 0:
-                data = self.rfile.read(nbytes)
+                cursize = 0
+                data = self.rfile.read(min(nbytes, _MIN_READ_BUF_SIZE))
+                while (len(data) < nbytes and len(data) != cursize and
+                       select.select([self.rfile._sock], [], [], 0)[0]):
+                    cursize = len(data)
+                    # This is a geometric increase in read size (never more
+                    # than doubling our the current length of data per loop
+                    # iteration).
+                    delta = min(cursize, nbytes - cursize)
+                    data += self.rfile.read(delta)
             else:
                 data = None
             # throw away additional data [see bug #427345]

diff --git a/Lib/test/test_httpservers.py b/Lib/test/test_httpservers.py
index c369baf28fe823153cb67c7f9e15896813dceac2..9e69f6f6d75dec96faa82ce28f6ca960bc147c30 100644 (file)
--- a/Lib/test/test_httpservers.py
+++ b/Lib/test/test_httpservers.py
@@ -802,6 +802,20 @@ for k, v in os.environ.items():
 print("</pre>")
 """
 
+cgi_file7 = """\
+#!%s
+import os
+import sys
+
+print("Content-type: text/plain")
+print()
+
+content_length = int(os.environ["CONTENT_LENGTH"])
+body = sys.stdin.buffer.read(content_length)
+
+print(f"{content_length} {len(body)}")
+"""
+
 
 @unittest.skipIf(hasattr(os, 'geteuid') and os.geteuid() == 0,
         "This test can't be run reliably as root (issue #13308).")
@@ -841,6 +855,8 @@ class CGIHTTPServerTestCase(BaseTestCase):
         self.file3_path = None
         self.file4_path = None
         self.file5_path = None
+        self.file6_path = None
+        self.file7_path = None
 
         # The shebang line should be pure ASCII: use symlink if possible.
         # See issue #7668.
@@ -895,6 +911,11 @@ class CGIHTTPServerTestCase(BaseTestCase):
             file6.write(cgi_file6 % self.pythonexe)
         os.chmod(self.file6_path, 0o777)
 
+        self.file7_path = os.path.join(self.cgi_dir, 'file7.py')
+        with open(self.file7_path, 'w', encoding='utf-8') as file7:
+            file7.write(cgi_file7 % self.pythonexe)
+        os.chmod(self.file7_path, 0o777)
+
         os.chdir(self.parent_dir)
 
     def tearDown(self):
@@ -917,6 +938,8 @@ class CGIHTTPServerTestCase(BaseTestCase):
                 os.remove(self.file5_path)
             if self.file6_path:
                 os.remove(self.file6_path)
+            if self.file7_path:
+                os.remove(self.file7_path)
             os.rmdir(self.cgi_child_dir)
             os.rmdir(self.cgi_dir)
             os.rmdir(self.cgi_dir_in_sub_dir)
@@ -989,6 +1012,21 @@ class CGIHTTPServerTestCase(BaseTestCase):
 
         self.assertEqual(res.read(), b'1, python, 123456' + self.linesep)
 
+    def test_large_content_length(self):
+        for w in range(15, 25):
+            size = 1 << w
+            body = b'X' * size
+            headers = {'Content-Length' : str(size)}
+            res = self.request('/cgi-bin/file7.py', 'POST', body, headers)
+            self.assertEqual(res.read(), b'%d %d' % (size, size) + self.linesep)
+
+    def test_large_content_length_truncated(self):
+        for w in range(18, 65):
+            size = 1 << w
+            headers = {'Content-Length' : str(size)}
+            res = self.request('/cgi-bin/file1.py', 'POST', b'x', headers)
+            self.assertEqual(res.read(), b'Hello World' + self.linesep)
+
     def test_invaliduri(self):
         res = self.request('/cgi-bin/invalid')
         res.read()

diff --git a/Misc/NEWS.d/next/Security/2024-05-23-11-44-41.gh-issue-119452.PRfsSv.rst b/Misc/NEWS.d/next/Security/2024-05-23-11-44-41.gh-issue-119452.PRfsSv.rst
new file mode 100644 (file)
index 0000000..9895662
--- /dev/null
+++ b/Misc/NEWS.d/next/Security/2024-05-23-11-44-41.gh-issue-119452.PRfsSv.rst
@@ -0,0 +1,5 @@
+Fix a potential memory denial of service in the :mod:`http.server` module.
+When a malicious user is connected to the CGI server on Windows, it could cause
+an arbitrary amount of memory to be allocated.
+This could have led to symptoms including a :exc:`MemoryError`, swapping, out
+of memory (OOM) killed processes or containers, or even system crashes.