gh-159875: Keep the SSL protocol of a closed transport after connection loss (GH-160069)

* gh-159875: Add tests for an SSL transport closed after connection loss

An application that closes its transport from connection_lost(), as aiohttp
does, must be left with a transport whose write(), get_write_buffer_size()
and set_protocol() behave as after closing a live connection, and
loop.sendfile()'s fallback must then raise ConnectionError rather than
AttributeError from set_protocol().

The sendfile test reads the file synchronously, so that the fallback is
waiting in drain() when the connection is lost; a read in the executor that
completes in the same loop iteration as the reset lets the fallback finish
first and hides the bug.

test_create_connection_memory_leak now keeps the transport past its check,
as its comment says it covers, so that it would see an SSLContext kept alive
by a transport the application still holds.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* gh-159875: Keep the SSL protocol of a closed transport after connection loss

_SSLProtocolTransport.close() dropped its _ssl_protocol when called a second
time, and the transport is already marked closed once the connection is lost,
so an application closing it from connection_lost() left a transport whose
write(), set_protocol(), get_write_buffer_size(), ... raise AttributeError.
That branch was also what released the SSLContext of a transport the
application keeps. Keep the protocol, and release the TLS state from
SSLProtocol.connection_lost() instead: one loop iteration later when the
application's connection_lost() was scheduled, so that it can still read
get_extra_info('ssl_object') and ('sslcontext').

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

---------
(cherry picked from commit 0626b14)

Co-authored-by: Mauricio Paez <jose.m.paez@icloud.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

Author: 2 people

Lib/asyncio/sslproto.py

@@ -114,8 +114,6 @@ def close(self):
         if not self._closed:
             self._closed = True
             self._ssl_protocol._start_shutdown()
-        else:
-            self._ssl_protocol = None
 
     def __del__(self, _warnings=warnings):
         if not self._closed:
@@ -412,12 +410,14 @@ def connection_lost(self, exc):
         # The app protocol gets connection_lost() only if it got
         # connection_made().  After start_tls() that is already the case
         # while the handshake is in progress.
+        app_notified = False
         if (
             self._app_state == AppProtocolState.STATE_CON_MADE or
             self._app_state == AppProtocolState.STATE_EOF
         ):
             self._app_state = AppProtocolState.STATE_CON_LOST
             self._loop.call_soon(self._app_protocol.connection_lost, exc)
+            app_notified = True
         self._set_state(SSLProtocolState.UNWRAPPED)
         self._transport = None
         self._app_transport = None
@@ -430,6 +430,21 @@ def connection_lost(self, exc):
         if self._handshake_timeout_handle:
             self._handshake_timeout_handle.cancel()
             self._handshake_timeout_handle = None
+        if app_notified:
+            # After the app protocol's connection_lost() (scheduled above),
+            # so that it can still read the TLS extra info.
+            self._loop.call_soon(self._release_tls_state)
+        else:
+            self._release_tls_state()
+
+    def _release_tls_state(self):
+        # The connection is gone: drop everything that refers to the
+        # SSLContext, so that an application holding the transport does
+        # not keep the context alive.
+        self._sslobj = None
+        self._sslcontext = None
+        self._extra.pop('sslcontext', None)
+        self._extra.pop('ssl_object', None)
 
     def get_buffer(self, n):
         want = n

Lib/test/test_asyncio/test_sendfile.py

@@ -512,6 +512,38 @@ def test_sendfile_ssl_close_peer_after_receiving(self):
         self.assertEqual(srv_proto.data, self.DATA)
         self.assertEqual(self.file.tell(), len(self.DATA))
 
+    def test_sendfile_ssl_close_peer_in_the_middle_closes_transport(self):
+        # The protocol closes its transport when the connection is lost, as
+        # aiohttp's does.  That close() must leave a transport on which the
+        # fallback can still restore the protocol: sendfile() raises
+        # ConnectionError, not AttributeError.
+        srv_proto, cli_proto = self.prepare_sendfile(
+            is_ssl=True, close_after=1024)
+        connection_lost = cli_proto.connection_lost
+
+        def close_on_lost(exc):
+            cli_proto.transport.close()
+            connection_lost(exc)
+
+        cli_proto.connection_lost = close_on_lost
+        # Read the file synchronously, so that the fallback only waits in
+        # drain().  A read in the executor that completes in the same loop
+        # iteration as the reset lets the fallback finish before close() is
+        # called, which hides the bug.
+        def run_in_executor(executor, func, *args):
+            fut = self.loop.create_future()
+            fut.set_result(func(*args))
+            return fut
+
+        with mock.patch.object(self.loop, 'run_in_executor', run_in_executor):
+            with self.assertRaises(ConnectionError):
+                self.run_loop(
+                    self.loop.sendfile(cli_proto.transport, self.file))
+        self.run_loop(srv_proto.done)
+        self.run_loop(cli_proto.done)
+        self.assertTrue(1024 <= srv_proto.nbytes < len(self.DATA),
+                        srv_proto.nbytes)
+
     # On Solaris, lowering SO_RCVBUF on a TCP connection after it has been
     # established has no effect. Due to its age, this bug affects both Oracle
     # Solaris as well as all other OpenSolaris forks (unless they fixed it

Lib/test/test_asyncio/test_ssl.py

@@ -791,16 +791,21 @@ async def client(addr):
             await on_eof
 
             tr.close()
+            return tr
 
         with self.tcp_server(serve, timeout=self.TIMEOUT) as srv:
-            self.loop.run_until_complete(
+            tr = self.loop.run_until_complete(
                 asyncio.wait_for(client(srv.addr),
                                  timeout=support.SHORT_TIMEOUT))
+        # Let the protocol's connection_lost() run.
+        self.loop.run_until_complete(asyncio.sleep(0))
 
         # No garbage is left for SSL client from loop.create_connection, even
         # if user stores the SSLTransport in corresponding protocol instance
+        # (and keeps the transport after the connection is lost).
         client_context = weakref.ref(client_context)
         self.assertIsNone(client_context())
+        del tr
 
     def test_start_tls_client_buf_proto_1(self):
         HELLO_MSG = b'1' * self.PAYLOAD_SIZE

Lib/test/test_asyncio/test_sslproto.py

@@ -289,6 +289,26 @@ def test_write_after_closing(self):
         # should not raise
         self.assertIsNone(transp.write(b'data'))
 
+    def test_transport_usable_after_connection_lost_and_close(self):
+        # connection_lost() marks the app transport closed, so the
+        # application's own close() afterwards (often made from its
+        # protocol's connection_lost()) must not be taken as a second
+        # close() that leaves the transport without its SSL protocol.
+        ssl_proto = self.ssl_protocol()
+        self.connection_made(ssl_proto)
+        transp = ssl_proto._app_transport
+        ssl_proto.connection_lost(None)
+        transp.close()
+
+        # should not raise, as after closing a live connection
+        self.assertIsNone(transp.write(b'data'))
+        self.assertEqual(transp.get_write_buffer_size(), 0)
+        new_proto = asyncio.Protocol()
+        transp.set_protocol(new_proto)  # as loop.sendfile()'s fallback does
+        self.assertIs(transp.get_protocol(), new_proto)
+        transp.abort()
+        self.assertTrue(transp.is_closing())
+
     def test_pause_reading_if_ssl_replies_are_not_read(self):
         # Output which OpenSSL produces on its own while reading (alerts,
         # KeyUpdate, ...) must not pile up if the peer does not read it.

Misc/NEWS.d/next/Library/2026-10-21-15-42-29.gh-issue-159875.p6j3yW.rst

@@ -0,0 +1,7 @@
+Fix :exc:`AttributeError` from the methods of an :mod:`asyncio` SSL transport,
+such as :meth:`~asyncio.WriteTransport.write` and
+:meth:`~asyncio.BaseTransport.set_protocol`, when the transport is closed after
+its connection was lost. This also affected the fallback of
+:meth:`loop.sendfile() <asyncio.loop.sendfile>` over SSL when the peer went
+away mid-transfer. The SSL context is still released once the connection is
+lost.