[3.9] gh-119511: Fix a potential denial of service in imaplib (GH-119514) (#130248)

The IMAP4 client could consume an arbitrary amount of memory when trying
to connect to a malicious server, because it read a "literal" data with a
single read(size) call, and BufferedReader.read() allocates the bytes
object of the specified size before reading. Now the IMAP4 client reads data
by chunks, therefore the amount of used memory is limited by the
amount of the data actually been sent by the server.
(cherry picked from commit 735f25c5e3a0f74438c86468ec4dfbe219d93c91)

Co-authored-by: Serhiy Storchaka <storchaka@gmail.com>
Co-authored-by: Gregory P. Smith <greg@krypto.org>

diff --git a/Lib/imaplib.py b/Lib/imaplib.py
index d9720f20c390256322cc207bf89fb4981dd0b218..ced7203e0c47ea1eafb5fc338f918d34831eace3 100644 (file)
--- a/Lib/imaplib.py
+++ b/Lib/imaplib.py
@@ -52,6 +52,9 @@ AllowedVersions = ('IMAP4REV1', 'IMAP4')        # Most recent first
 # search command can be quite large, so we now use 1M.
 _MAXLINE = 1000000
 
+# Data larger than this will be read in chunks, to prevent extreme
+# overallocation.
+_SAFE_BUF_SIZE = 1 << 20
 
 #       Commands
 
@@ -315,7 +318,13 @@ class IMAP4:
 
     def read(self, size):
         """Read 'size' bytes from remote."""
-        return self.file.read(size)
+        cursize = min(size, _SAFE_BUF_SIZE)
+        data = self.file.read(cursize)
+        while cursize < size and len(data) == cursize:
+            delta = min(cursize, size - cursize)
+            data += self.file.read(delta)
+            cursize += delta
+        return data
 
 
     def readline(self):

diff --git a/Lib/test/test_imaplib.py b/Lib/test/test_imaplib.py
index 70559383e26994ff053617db02ee71cbf9ecf7dd..057e4e65f03e78f1b40942416ae074e62d1c10cf 100644 (file)
--- a/Lib/test/test_imaplib.py
+++ b/Lib/test/test_imaplib.py
@@ -906,6 +906,21 @@ class ThreadedNetworkedTests(unittest.TestCase):
             self.assertRaises(imaplib.IMAP4.error,
                               self.imap_class, *server.server_address)
 
+    @reap_threads
+    def test_truncated_large_literal(self):
+        size = 0
+        class BadHandler(SimpleIMAPHandler):
+            def handle(self):
+                self._send_textline('* OK {%d}' % size)
+                self._send_textline('IMAP4rev1')
+
+        for exponent in range(15, 64):
+            size = 1 << exponent
+            with self.subTest(f"size=2e{size}"):
+                with self.reaped_server(BadHandler) as server:
+                    with self.assertRaises(imaplib.IMAP4.abort):
+                        self.imap_class(*server.server_address)
+
     @reap_threads
     def test_simple_with_statement(self):
         # simplest call

diff --git a/Misc/NEWS.d/next/Security/2024-05-24-21-00-52.gh-issue-119511.jKrXQ8.rst b/Misc/NEWS.d/next/Security/2024-05-24-21-00-52.gh-issue-119511.jKrXQ8.rst
new file mode 100644 (file)
index 0000000..f7b4031
--- /dev/null
+++ b/Misc/NEWS.d/next/Security/2024-05-24-21-00-52.gh-issue-119511.jKrXQ8.rst
@@ -0,0 +1,7 @@
+Fix a potential denial of service in the :mod:`imaplib` module. When connecting
+to a malicious server, it could cause an arbitrary amount of memory to be
+allocated. On many systems this is harmless as unused virtual memory is only a
+mapping, but if this hit a virtual address size limit it could lead to a
+:exc:`MemoryError` or other process crash. On unusual systems or builds where
+all allocated memory is touched and backed by actual ram or storage it could've
+consumed resources doing so until similarly crashing.