restore.c

  1// SPDX-License-Identifier: GPL-2.0
  2/* Author: Dmitry Safonov <dima@arista.com> */
  3/* This is over-simplified TCP_REPAIR for TCP_ESTABLISHED sockets
  4 * It tests that TCP-AO enabled connection can be restored.
  5 * For the proper socket repair see:
  6 * https://github.com/checkpoint-restore/criu/blob/criu-dev/soccr/soccr.h
  7 */
  8#include <inttypes.h>
  9#include "aolib.h"
 10
 11const size_t nr_packets = 20;
 12const size_t msg_len = 100;
 13const size_t quota = nr_packets * msg_len;
 14#define fault(type)	(inj == FAULT_ ## type)
 15
 16static void try_server_run(const char *tst_name, unsigned int port,
 17			   fault_t inj, test_cnt cnt_expected)
 18{
 19	test_cnt poll_cnt = (cnt_expected == TEST_CNT_GOOD) ? 0 : cnt_expected;
 20	const char *cnt_name = "TCPAOGood";
 21	struct tcp_counters cnt1, cnt2;
 22	uint64_t before_cnt, after_cnt;
 23	int sk, lsk, dummy;
 24	ssize_t bytes;
 25
 26	if (fault(TIMEOUT))
 27		cnt_name = "TCPAOBad";
 28	lsk = test_listen_socket(this_ip_addr, port, 1);
 29
 30	if (test_add_key(lsk, DEFAULT_TEST_PASSWORD, this_ip_dest, -1, 100, 100))
 31		test_error("setsockopt(TCP_AO_ADD_KEY)");
 32	synchronize_threads(); /* 1: MKT added => connect() */
 33
 34	if (test_wait_fd(lsk, TEST_TIMEOUT_SEC, 0))
 35		test_error("test_wait_fd()");
 36
 37	sk = accept(lsk, NULL, NULL);
 38	if (sk < 0)
 39		test_error("accept()");
 40
 41	synchronize_threads(); /* 2: accepted => send data */
 42	close(lsk);
 43
 44	bytes = test_server_run(sk, quota, TEST_TIMEOUT_SEC);
 45	if (bytes != quota) {
 46		test_fail("%s: server served: %zd", tst_name, bytes);
 47		goto out;
 48	}
 49
 50	before_cnt = netstat_get_one(cnt_name, NULL);
 51	if (test_get_tcp_counters(sk, &cnt1))
 52		test_error("test_get_tcp_counters()");
 53
 54	bytes = test_skpair_server(sk, quota, poll_cnt, &dummy);
 55	if (fault(TIMEOUT)) {
 56		if (bytes > 0)
 57			test_fail("%s: server served: %zd", tst_name, bytes);
 58		else
 59			test_ok("%s: server couldn't serve", tst_name);
 60	} else {
 61		if (bytes != quota)
 62			test_fail("%s: server served: %zd", tst_name, bytes);
 63		else
 64			test_ok("%s: server alive", tst_name);
 65	}
 66	synchronize_threads(); /* 3: counters checks */
 67	if (test_get_tcp_counters(sk, &cnt2))
 68		test_error("test_get_tcp_counters()");
 69	after_cnt = netstat_get_one(cnt_name, NULL);
 70
 71	test_assert_counters(tst_name, &cnt1, &cnt2, cnt_expected);
 72
 73	if (after_cnt <= before_cnt) {
 74		test_fail("%s(server): %s counter did not increase: %" PRIu64 " <= %" PRIu64,
 75			  tst_name, cnt_name, after_cnt, before_cnt);
 76	} else {
 77		test_ok("%s(server): counter %s increased %" PRIu64 " => %" PRIu64,
 78			tst_name, cnt_name, before_cnt, after_cnt);
 79	}
 80
 81	/*
 82	 * Before close() as that will send FIN and move the peer in TCP_CLOSE
 83	 * and that will prevent reading AO counters from the peer's socket.
 84	 */
 85	synchronize_threads(); /* 4: verified => closed */
 86out:
 87	close(sk);
 88}
 89
 90static void *server_fn(void *arg)
 91{
 92	unsigned int port = test_server_port;
 93
 94	try_server_run("TCP-AO migrate to another socket (server)", port++,
 95		       0, TEST_CNT_GOOD);
 96	try_server_run("TCP-AO with wrong send ISN (server)", port++,
 97		       FAULT_TIMEOUT, TEST_CNT_BAD);
 98	try_server_run("TCP-AO with wrong receive ISN (server)", port++,
 99		       FAULT_TIMEOUT, TEST_CNT_BAD);
100	try_server_run("TCP-AO with wrong send SEQ ext number (server)", port++,
101		       FAULT_TIMEOUT, TEST_CNT_BAD);
102	try_server_run("TCP-AO with wrong receive SEQ ext number (server)",
103		       port++, FAULT_TIMEOUT, TEST_CNT_NS_BAD | TEST_CNT_GOOD);
104
105	synchronize_threads(); /* don't race to exit: client exits */
106	return NULL;
107}
108
109static void test_get_sk_checkpoint(unsigned int server_port, sockaddr_af *saddr,
110				   struct tcp_sock_state *img,
111				   struct tcp_ao_repair *ao_img)
112{
113	int sk;
114
115	sk = socket(test_family, SOCK_STREAM, IPPROTO_TCP);
116	if (sk < 0)
117		test_error("socket()");
118
119	if (test_add_key(sk, DEFAULT_TEST_PASSWORD, this_ip_dest, -1, 100, 100))
120		test_error("setsockopt(TCP_AO_ADD_KEY)");
121
122	synchronize_threads(); /* 1: MKT added => connect() */
123	if (test_connect_socket(sk, this_ip_dest, server_port) <= 0)
124		test_error("failed to connect()");
125
126	synchronize_threads(); /* 2: accepted => send data */
127	if (test_client_verify(sk, msg_len, nr_packets))
128		test_fail("pre-migrate verify failed");
129
130	test_enable_repair(sk);
131	test_sock_checkpoint(sk, img, saddr);
132	test_ao_checkpoint(sk, ao_img);
133	test_kill_sk(sk);
134}
135
136static void test_sk_restore(const char *tst_name, unsigned int server_port,
137			    sockaddr_af *saddr, struct tcp_sock_state *img,
138			    struct tcp_ao_repair *ao_img,
139			    fault_t inj, test_cnt cnt_expected)
140{
141	test_cnt poll_cnt = (cnt_expected == TEST_CNT_GOOD) ? 0 : cnt_expected;
142	const char *cnt_name = "TCPAOGood";
143	struct tcp_counters cnt1, cnt2;
144	uint64_t before_cnt, after_cnt;
145	int sk, dummy;
146
147	if (fault(TIMEOUT))
148		cnt_name = "TCPAOBad";
149
150	before_cnt = netstat_get_one(cnt_name, NULL);
151	sk = socket(test_family, SOCK_STREAM, IPPROTO_TCP);
152	if (sk < 0)
153		test_error("socket()");
154
155	test_enable_repair(sk);
156	test_sock_restore(sk, img, saddr, this_ip_dest, server_port);
157	if (test_add_repaired_key(sk, DEFAULT_TEST_PASSWORD, 0, this_ip_dest, -1, 100, 100))
158		test_error("setsockopt(TCP_AO_ADD_KEY)");
159	test_ao_restore(sk, ao_img);
160
161	if (test_get_tcp_counters(sk, &cnt1))
162		test_error("test_get_tcp_counters()");
163
164	test_disable_repair(sk);
165	test_sock_state_free(img);
166
167	if (test_skpair_client(sk, msg_len, nr_packets, poll_cnt, &dummy)) {
168		if (fault(TIMEOUT))
169			test_ok("%s: post-migrate connection is broken", tst_name);
170		else
171			test_fail("%s: post-migrate connection is working", tst_name);
172	} else {
173		if (fault(TIMEOUT))
174			test_fail("%s: post-migrate connection is working", tst_name);
175		else
176			test_ok("%s: post-migrate connection is alive", tst_name);
177	}
178
179	synchronize_threads(); /* 3: counters checks */
180	if (test_get_tcp_counters(sk, &cnt2))
181		test_error("test_get_tcp_counters()");
182	after_cnt = netstat_get_one(cnt_name, NULL);
183
184	test_assert_counters(tst_name, &cnt1, &cnt2, cnt_expected);
185
186	if (after_cnt <= before_cnt) {
187		test_fail("%s: %s counter did not increase: %" PRIu64 " <= %" PRIu64,
188				tst_name, cnt_name, after_cnt, before_cnt);
189	} else {
190		test_ok("%s: counter %s increased %" PRIu64 " => %" PRIu64,
191			tst_name, cnt_name, before_cnt, after_cnt);
192	}
193	synchronize_threads(); /* 4: verified => closed */
194	close(sk);
195}
196
197static void *client_fn(void *arg)
198{
199	unsigned int port = test_server_port;
200	struct tcp_sock_state tcp_img;
201	struct tcp_ao_repair ao_img;
202	sockaddr_af saddr;
203
204	test_get_sk_checkpoint(port, &saddr, &tcp_img, &ao_img);
205	test_sk_restore("TCP-AO migrate to another socket (client)", port++,
206			&saddr, &tcp_img, &ao_img, 0, TEST_CNT_GOOD);
207
208	test_get_sk_checkpoint(port, &saddr, &tcp_img, &ao_img);
209	ao_img.snt_isn += 1;
210	trace_ao_event_expect(TCP_AO_MISMATCH, this_ip_addr, this_ip_dest,
211			      -1, port, 0, -1, -1, -1, -1, -1, 100, 100, -1);
212	trace_ao_event_expect(TCP_AO_MISMATCH, this_ip_dest, this_ip_addr,
213			      port, -1, 0, -1, -1, -1, -1, -1, 100, 100, -1);
214	test_sk_restore("TCP-AO with wrong send ISN (client)", port++,
215			&saddr, &tcp_img, &ao_img, FAULT_TIMEOUT, TEST_CNT_BAD);
216
217	test_get_sk_checkpoint(port, &saddr, &tcp_img, &ao_img);
218	ao_img.rcv_isn += 1;
219	trace_ao_event_expect(TCP_AO_MISMATCH, this_ip_addr, this_ip_dest,
220			      -1, port, 0, -1, -1, -1, -1, -1, 100, 100, -1);
221	trace_ao_event_expect(TCP_AO_MISMATCH, this_ip_dest, this_ip_addr,
222			      port, -1, 0, -1, -1, -1, -1, -1, 100, 100, -1);
223	test_sk_restore("TCP-AO with wrong receive ISN (client)", port++,
224			&saddr, &tcp_img, &ao_img, FAULT_TIMEOUT, TEST_CNT_BAD);
225
226	test_get_sk_checkpoint(port, &saddr, &tcp_img, &ao_img);
227	ao_img.snd_sne += 1;
228	trace_ao_event_expect(TCP_AO_MISMATCH, this_ip_addr, this_ip_dest,
229			      -1, port, 0, -1, -1, -1, -1, -1, 100, 100, -1);
230	/* not expecting server => client mismatches as only snd sne is broken */
231	test_sk_restore("TCP-AO with wrong send SEQ ext number (client)",
232			port++, &saddr, &tcp_img, &ao_img, FAULT_TIMEOUT,
233			TEST_CNT_NS_BAD | TEST_CNT_GOOD);
234
235	test_get_sk_checkpoint(port, &saddr, &tcp_img, &ao_img);
236	ao_img.rcv_sne += 1;
237	/* not expecting client => server mismatches as only rcv sne is broken */
238	trace_ao_event_expect(TCP_AO_MISMATCH, this_ip_dest, this_ip_addr,
239			      port, -1, 0, -1, -1, -1, -1, -1, 100, 100, -1);
240	test_sk_restore("TCP-AO with wrong receive SEQ ext number (client)",
241			port++, &saddr, &tcp_img, &ao_img, FAULT_TIMEOUT,
242			TEST_CNT_NS_GOOD | TEST_CNT_BAD);
243
244	return NULL;
245}
246
247int main(int argc, char *argv[])
248{
249	test_init(21, server_fn, client_fn);
250	return 0;
251}