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

[3.14] gh-119452: Fix a potential virtual memory allocation denial of service in http.server (GH-142216)

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,
therefore the memory consumption is proportional to the amount of sent
data.
(cherry picked from commit 0e4f4f1a4633f2d215fb5a803cae278aeea31845)

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

diff --git a/Lib/http/server.py b/Lib/http/server.py
index da07f110f1778f5ba03bc169064904bf70410f4e..cdeb8f097db6b3b89aed0ea9d015ffc050c97093 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
@@ -1217,7 +1221,18 @@ 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:
+                    cursize = len(data)
+                    # This is a geometric increase in read size (never more
+                    # than doubling out the current length of data per loop
+                    # iteration).
+                    delta = min(cursize, nbytes - cursize)
+                    try:
+                        data += self.rfile.read(delta)
+                    except TimeoutError:
+                        break
             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 cfd8a101dcc1c1e202acad4e528358390c6c22fb..b0b09daab003f0f4dfb364c91a307184e1b344f6 100644 (file)
--- a/Lib/test/test_httpservers.py
+++ b/Lib/test/test_httpservers.py
@@ -694,6 +694,20 @@ 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).")
@@ -723,6 +737,8 @@ 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.
@@ -777,6 +793,11 @@ 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):
@@ -798,6 +819,8 @@ 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)
@@ -867,6 +890,22 @@ 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):
+        with support.swap_attr(self.request_handler, 'timeout', 0.001):
+            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.