tls: send the second request once in the post-handshake auth test

[glaziermag]

Description

test_tls_client_native() hands REQS[1] to the client automaton twice, with send() and then with sr1() (

scapy/test/scapy/layers/tls/tlsclientserver.uts

Lines 706 to 708 in e4742a5

	# Second request
	a.send(REQS[1])
	pkt = a.sr1(REQS[1], timeout=1, verbose=0)

). The ssl_server thread reads one request, answers it and shuts the connection down (

scapy/test/scapy/layers/tls/tlsclientserver.uts

Lines 664 to 674 in e4742a5

	t = ssl_client_socket.verify_client_post_handshake()
	# Receive / send data
	resp = ssl_client_socket.read(len(REQS[1]))
	assert resp == bytes(REQS[1])
	ssl_client_socket.send(bytes(RESPS[1]))
	# close socket
	try:
	ssl_client_socket.shutdown(socket.SHUT_RDWR)
	finally:
	ssl_client_socket.close()
	try:

), which can happen while the client is still building its post-handshake Certificate. That flight then lands on the shut-down socket and is answered with a RST, so the send of the second copy raises BrokenPipeError in flush_records() under should_send_ClientData() (

scapy/scapy/layers/tls/automaton_cli.py

Lines 665 to 668 in e4742a5

	@ATMT.condition(ADDED_CLIENTDATA)
	def should_send_ClientData(self):
	self.flush_records()
	raise self.SENT_CLIENTDATA()

), the automaton stops, and the 200 Welcome record already queued on the socket is never read, so sr1() returns None.

That is the macos-14 3.14 failure of Testing TLS client against ssl.SSLContext server with TLS 1.3 and a post-handshake authentication on master (AttributeError: 'NoneType' object has no attribute 'load'), e.g. https://github.com/secdev/scapy/actions/runs/35789038699/job/106956973992 and https://github.com/secdev/scapy/actions/runs/35743207536/job/106801186588; locally on macOS it fails about 1 run in 100. The OSError: [Errno 57] traceback printed by the server thread in those logs is server.shutdown() on the listening socket and appears in passing runs too.

Sending the request once removes the write on the closed connection.

This message was written with the help of AI (Claude Fable 5.1).

🤖 Generated with Claude Code