1 // SPDX-License-Identifier: GPL-2.0
2 
3 #define _GNU_SOURCE
4 
5 #include <assert.h>
6 #include <errno.h>
7 #include <fcntl.h>
8 #include <limits.h>
9 #include <string.h>
10 #include <stdarg.h>
11 #include <stdbool.h>
12 #include <stdint.h>
13 #include <inttypes.h>
14 #include <stdio.h>
15 #include <stdlib.h>
16 #include <strings.h>
17 #include <time.h>
18 #include <unistd.h>
19 
20 #include <sys/socket.h>
21 #include <sys/types.h>
22 #include <sys/wait.h>
23 
24 #include <netdb.h>
25 #include <netinet/in.h>
26 
27 #include <linux/tcp.h>
28 
29 static int pf = AF_INET;
30 
31 #ifndef IPPROTO_MPTCP
32 #define IPPROTO_MPTCP 262
33 #endif
34 #ifndef SOL_MPTCP
35 #define SOL_MPTCP 284
36 #endif
37 
38 #ifndef MPTCP_INFO
39 struct mptcp_info {
40 	__u8	mptcpi_subflows;
41 	__u8	mptcpi_add_addr_signal;
42 	__u8	mptcpi_add_addr_accepted;
43 	__u8	mptcpi_subflows_max;
44 	__u8	mptcpi_add_addr_signal_max;
45 	__u8	mptcpi_add_addr_accepted_max;
46 	__u32	mptcpi_flags;
47 	__u32	mptcpi_token;
48 	__u64	mptcpi_write_seq;
49 	__u64	mptcpi_snd_una;
50 	__u64	mptcpi_rcv_nxt;
51 	__u8	mptcpi_local_addr_used;
52 	__u8	mptcpi_local_addr_max;
53 	__u8	mptcpi_csum_enabled;
54 	__u32	mptcpi_retransmits;
55 	__u64	mptcpi_bytes_retrans;
56 	__u64	mptcpi_bytes_sent;
57 	__u64	mptcpi_bytes_received;
58 	__u64	mptcpi_bytes_acked;
59 };
60 
61 struct mptcp_subflow_data {
62 	__u32		size_subflow_data;		/* size of this structure in userspace */
63 	__u32		num_subflows;			/* must be 0, set by kernel */
64 	__u32		size_kernel;			/* must be 0, set by kernel */
65 	__u32		size_user;			/* size of one element in data[] */
66 } __attribute__((aligned(8)));
67 
68 struct mptcp_subflow_addrs {
69 	union {
70 		__kernel_sa_family_t sa_family;
71 		struct sockaddr sa_local;
72 		struct sockaddr_in sin_local;
73 		struct sockaddr_in6 sin6_local;
74 		struct __kernel_sockaddr_storage ss_local;
75 	};
76 	union {
77 		struct sockaddr sa_remote;
78 		struct sockaddr_in sin_remote;
79 		struct sockaddr_in6 sin6_remote;
80 		struct __kernel_sockaddr_storage ss_remote;
81 	};
82 };
83 
84 #define MPTCP_INFO		1
85 #define MPTCP_TCPINFO		2
86 #define MPTCP_SUBFLOW_ADDRS	3
87 #endif
88 
89 #ifndef MPTCP_FULL_INFO
90 struct mptcp_subflow_info {
91 	__u32				id;
92 	struct mptcp_subflow_addrs	addrs;
93 };
94 
95 struct mptcp_full_info {
96 	__u32		size_tcpinfo_kernel;	/* must be 0, set by kernel */
97 	__u32		size_tcpinfo_user;
98 	__u32		size_sfinfo_kernel;	/* must be 0, set by kernel */
99 	__u32		size_sfinfo_user;
100 	__u32		num_subflows;		/* must be 0, set by kernel (real subflow count) */
101 	__u32		size_arrays_user;	/* max subflows that userspace is interested in;
102 						 * the buffers at subflow_info/tcp_info
103 						 * are respectively at least:
104 						 *  size_arrays * size_sfinfo_user
105 						 *  size_arrays * size_tcpinfo_user
106 						 * bytes wide
107 						 */
108 	__aligned_u64		subflow_info;
109 	__aligned_u64		tcp_info;
110 	struct mptcp_info	mptcp_info;
111 };
112 
113 #define MPTCP_FULL_INFO		4
114 #endif
115 
116 struct so_state {
117 	struct mptcp_info mi;
118 	struct mptcp_info last_sample;
119 	struct tcp_info tcp_info;
120 	struct mptcp_subflow_addrs addrs;
121 	uint64_t mptcpi_rcv_delta;
122 	uint64_t tcpi_rcv_delta;
123 	bool pkt_stats_avail;
124 };
125 
126 #ifndef MIN
127 #define MIN(a, b) ((a) < (b) ? (a) : (b))
128 #endif
129 
130 static void die_perror(const char *msg)
131 {
132 	perror(msg);
133 	exit(1);
134 }
135 
136 static void die_usage(int r)
137 {
138 	fprintf(stderr, "Usage: mptcp_sockopt [-6]\n");
139 	exit(r);
140 }
141 
142 static void xerror(const char *fmt, ...)
143 {
144 	va_list ap;
145 
146 	va_start(ap, fmt);
147 	vfprintf(stderr, fmt, ap);
148 	va_end(ap);
149 	fputc('\n', stderr);
150 	exit(1);
151 }
152 
153 static const char *getxinfo_strerr(int err)
154 {
155 	if (err == EAI_SYSTEM)
156 		return strerror(errno);
157 
158 	return gai_strerror(err);
159 }
160 
161 static void xgetaddrinfo(const char *node, const char *service,
162 			 const struct addrinfo *hints,
163 			 struct addrinfo **res)
164 {
165 	int err = getaddrinfo(node, service, hints, res);
166 
167 	if (err) {
168 		const char *errstr = getxinfo_strerr(err);
169 
170 		fprintf(stderr, "Fatal: getaddrinfo(%s:%s): %s\n",
171 			node ? node : "", service ? service : "", errstr);
172 		exit(1);
173 	}
174 }
175 
176 static int sock_listen_mptcp(const char * const listenaddr,
177 			     const char * const port)
178 {
179 	int sock = -1;
180 	struct addrinfo hints = {
181 		.ai_protocol = IPPROTO_TCP,
182 		.ai_socktype = SOCK_STREAM,
183 		.ai_flags = AI_PASSIVE | AI_NUMERICHOST
184 	};
185 
186 	hints.ai_family = pf;
187 
188 	struct addrinfo *a, *addr;
189 	int one = 1;
190 
191 	xgetaddrinfo(listenaddr, port, &hints, &addr);
192 	hints.ai_family = pf;
193 
194 	for (a = addr; a; a = a->ai_next) {
195 		sock = socket(a->ai_family, a->ai_socktype, IPPROTO_MPTCP);
196 		if (sock < 0)
197 			continue;
198 
199 		if (-1 == setsockopt(sock, SOL_SOCKET, SO_REUSEADDR, &one,
200 				     sizeof(one)))
201 			perror("setsockopt");
202 
203 		if (bind(sock, a->ai_addr, a->ai_addrlen) == 0)
204 			break; /* success */
205 
206 		perror("bind");
207 		close(sock);
208 		sock = -1;
209 	}
210 
211 	freeaddrinfo(addr);
212 
213 	if (sock < 0)
214 		xerror("could not create listen socket");
215 
216 	if (listen(sock, 20))
217 		die_perror("listen");
218 
219 	return sock;
220 }
221 
222 static int sock_connect_mptcp(const char * const remoteaddr,
223 			      const char * const port, int proto)
224 {
225 	struct addrinfo hints = {
226 		.ai_protocol = IPPROTO_TCP,
227 		.ai_socktype = SOCK_STREAM,
228 	};
229 	struct addrinfo *a, *addr;
230 	int sock = -1;
231 
232 	hints.ai_family = pf;
233 
234 	xgetaddrinfo(remoteaddr, port, &hints, &addr);
235 	for (a = addr; a; a = a->ai_next) {
236 		sock = socket(a->ai_family, a->ai_socktype, proto);
237 		if (sock < 0)
238 			continue;
239 
240 		if (connect(sock, a->ai_addr, a->ai_addrlen) == 0)
241 			break; /* success */
242 
243 		die_perror("connect");
244 	}
245 
246 	if (sock < 0)
247 		xerror("could not create connect socket");
248 
249 	freeaddrinfo(addr);
250 	return sock;
251 }
252 
253 static void parse_opts(int argc, char **argv)
254 {
255 	int c;
256 
257 	while ((c = getopt(argc, argv, "h6")) != -1) {
258 		switch (c) {
259 		case 'h':
260 			die_usage(0);
261 			break;
262 		case '6':
263 			pf = AF_INET6;
264 			break;
265 		default:
266 			die_usage(1);
267 			break;
268 		}
269 	}
270 }
271 
272 static void do_getsockopt_bogus_sf_data(int fd, int optname)
273 {
274 	struct mptcp_subflow_data good_data;
275 	struct bogus_data {
276 		struct mptcp_subflow_data d;
277 		char buf[2];
278 	} bd;
279 	socklen_t olen, _olen;
280 	int ret;
281 
282 	memset(&bd, 0, sizeof(bd));
283 	memset(&good_data, 0, sizeof(good_data));
284 
285 	olen = sizeof(good_data);
286 	good_data.size_subflow_data = olen;
287 
288 	ret = getsockopt(fd, SOL_MPTCP, optname, &bd, &olen);
289 	assert(ret < 0); /* 0 size_subflow_data */
290 	assert(olen == sizeof(good_data));
291 
292 	bd.d = good_data;
293 
294 	ret = getsockopt(fd, SOL_MPTCP, optname, &bd, &olen);
295 	assert(ret == 0);
296 	assert(olen == sizeof(good_data));
297 	assert(bd.d.num_subflows == 1);
298 	assert(bd.d.size_kernel > 0);
299 	assert(bd.d.size_user == 0);
300 
301 	bd.d = good_data;
302 	_olen = rand() % olen;
303 	olen = _olen;
304 	ret = getsockopt(fd, SOL_MPTCP, optname, &bd, &olen);
305 	assert(ret < 0);	/* bogus olen */
306 	assert(olen == _olen);	/* must be unchanged */
307 
308 	bd.d = good_data;
309 	olen = sizeof(good_data);
310 	bd.d.size_kernel = 1;
311 	ret = getsockopt(fd, SOL_MPTCP, optname, &bd, &olen);
312 	assert(ret < 0); /* size_kernel not 0 */
313 
314 	bd.d = good_data;
315 	olen = sizeof(good_data);
316 	bd.d.num_subflows = 1;
317 	ret = getsockopt(fd, SOL_MPTCP, optname, &bd, &olen);
318 	assert(ret < 0); /* num_subflows not 0 */
319 
320 	/* forward compat check: larger struct mptcp_subflow_data on 'old' kernel */
321 	bd.d = good_data;
322 	olen = sizeof(bd);
323 	bd.d.size_subflow_data = sizeof(bd);
324 
325 	ret = getsockopt(fd, SOL_MPTCP, optname, &bd, &olen);
326 	assert(ret == 0);
327 
328 	/* olen must be truncated to real data size filled by kernel: */
329 	assert(olen == sizeof(good_data));
330 
331 	assert(bd.d.size_subflow_data == sizeof(bd));
332 
333 	bd.d = good_data;
334 	bd.d.size_subflow_data += 1;
335 	bd.d.size_user = 1;
336 	olen = bd.d.size_subflow_data + 1;
337 	_olen = olen;
338 
339 	ret = getsockopt(fd, SOL_MPTCP, optname, &bd, &_olen);
340 	assert(ret == 0);
341 
342 	/* no truncation, kernel should have filled 1 byte of optname payload in buf[1]: */
343 	assert(olen == _olen);
344 
345 	assert(bd.d.size_subflow_data == sizeof(good_data) + 1);
346 	assert(bd.buf[0] == 0);
347 }
348 
349 static void do_getsockopt_mptcp_info(struct so_state *s, int fd, size_t w)
350 {
351 	struct mptcp_info i;
352 	socklen_t olen;
353 	int ret;
354 
355 	olen = sizeof(i);
356 	ret = getsockopt(fd, SOL_MPTCP, MPTCP_INFO, &i, &olen);
357 
358 	if (ret < 0)
359 		die_perror("getsockopt MPTCP_INFO");
360 
361 	s->pkt_stats_avail = olen >= sizeof(i);
362 
363 	s->last_sample = i;
364 	if (s->mi.mptcpi_write_seq == 0)
365 		s->mi = i;
366 
367 	assert(s->mi.mptcpi_write_seq + w == i.mptcpi_write_seq);
368 
369 	s->mptcpi_rcv_delta = i.mptcpi_rcv_nxt - s->mi.mptcpi_rcv_nxt;
370 }
371 
372 static void do_getsockopt_tcp_info(struct so_state *s, int fd, size_t r, size_t w)
373 {
374 	struct my_tcp_info {
375 		struct mptcp_subflow_data d;
376 		struct tcp_info ti[2];
377 	} ti;
378 	int ret, tries = 5;
379 	socklen_t olen;
380 
381 	do {
382 		memset(&ti, 0, sizeof(ti));
383 
384 		ti.d.size_subflow_data = sizeof(struct mptcp_subflow_data);
385 		ti.d.size_user = sizeof(struct tcp_info);
386 		olen = sizeof(ti);
387 
388 		ret = getsockopt(fd, SOL_MPTCP, MPTCP_TCPINFO, &ti, &olen);
389 		if (ret < 0)
390 			xerror("getsockopt MPTCP_TCPINFO (tries %d, %m)");
391 
392 		assert(olen <= sizeof(ti));
393 		assert(ti.d.size_kernel > 0);
394 		assert(ti.d.size_user ==
395 		       MIN(ti.d.size_kernel, sizeof(struct tcp_info)));
396 		assert(ti.d.num_subflows == 1);
397 
398 		assert(olen > (socklen_t)sizeof(struct mptcp_subflow_data));
399 		olen -= sizeof(struct mptcp_subflow_data);
400 		assert(olen == ti.d.size_user);
401 
402 		s->tcp_info = ti.ti[0];
403 
404 		if (ti.ti[0].tcpi_bytes_sent == w &&
405 		    ti.ti[0].tcpi_bytes_received == r)
406 			goto done;
407 
408 		if (r == 0 && ti.ti[0].tcpi_bytes_sent == w &&
409 		    ti.ti[0].tcpi_bytes_received) {
410 			s->tcpi_rcv_delta = ti.ti[0].tcpi_bytes_received;
411 			goto done;
412 		}
413 
414 		/* wait and repeat, might be that tx is still ongoing */
415 		sleep(1);
416 	} while (tries-- > 0);
417 
418 	xerror("tcpi_bytes_sent %" PRIu64 ", want %zu. tcpi_bytes_received %" PRIu64 ", want %zu",
419 		ti.ti[0].tcpi_bytes_sent, w, ti.ti[0].tcpi_bytes_received, r);
420 
421 done:
422 	do_getsockopt_bogus_sf_data(fd, MPTCP_TCPINFO);
423 }
424 
425 static void do_getsockopt_subflow_addrs(struct so_state *s, int fd)
426 {
427 	struct sockaddr_storage remote, local;
428 	socklen_t olen, rlen, llen;
429 	int ret;
430 	struct my_addrs {
431 		struct mptcp_subflow_data d;
432 		struct mptcp_subflow_addrs addr[2];
433 	} addrs;
434 
435 	memset(&addrs, 0, sizeof(addrs));
436 	memset(&local, 0, sizeof(local));
437 	memset(&remote, 0, sizeof(remote));
438 
439 	addrs.d.size_subflow_data = sizeof(struct mptcp_subflow_data);
440 	addrs.d.size_user = sizeof(struct mptcp_subflow_addrs);
441 	olen = sizeof(addrs);
442 
443 	ret = getsockopt(fd, SOL_MPTCP, MPTCP_SUBFLOW_ADDRS, &addrs, &olen);
444 	if (ret < 0)
445 		die_perror("getsockopt MPTCP_SUBFLOW_ADDRS");
446 
447 	assert(olen <= sizeof(addrs));
448 	assert(addrs.d.size_kernel > 0);
449 	assert(addrs.d.size_user ==
450 	       MIN(addrs.d.size_kernel, sizeof(struct mptcp_subflow_addrs)));
451 	assert(addrs.d.num_subflows == 1);
452 
453 	assert(olen > (socklen_t)sizeof(struct mptcp_subflow_data));
454 	olen -= sizeof(struct mptcp_subflow_data);
455 	assert(olen == addrs.d.size_user);
456 
457 	llen = sizeof(local);
458 	ret = getsockname(fd, (struct sockaddr *)&local, &llen);
459 	if (ret < 0)
460 		die_perror("getsockname");
461 	rlen = sizeof(remote);
462 	ret = getpeername(fd, (struct sockaddr *)&remote, &rlen);
463 	if (ret < 0)
464 		die_perror("getpeername");
465 
466 	assert(rlen > 0);
467 	assert(rlen == llen);
468 
469 	assert(remote.ss_family == local.ss_family);
470 
471 	assert(memcmp(&local, &addrs.addr[0].ss_local, sizeof(local)) == 0);
472 	assert(memcmp(&remote, &addrs.addr[0].ss_remote, sizeof(remote)) == 0);
473 	s->addrs = addrs.addr[0];
474 
475 	memset(&addrs, 0, sizeof(addrs));
476 
477 	addrs.d.size_subflow_data = sizeof(struct mptcp_subflow_data);
478 	addrs.d.size_user = sizeof(sa_family_t);
479 	olen = sizeof(addrs.d) + sizeof(sa_family_t);
480 
481 	ret = getsockopt(fd, SOL_MPTCP, MPTCP_SUBFLOW_ADDRS, &addrs, &olen);
482 	assert(ret == 0);
483 	assert(olen == sizeof(addrs.d) + sizeof(sa_family_t));
484 
485 	assert(addrs.addr[0].sa_family == pf);
486 	assert(addrs.addr[0].sa_family == local.ss_family);
487 
488 	assert(memcmp(&local, &addrs.addr[0].ss_local, sizeof(local)) != 0);
489 	assert(memcmp(&remote, &addrs.addr[0].ss_remote, sizeof(remote)) != 0);
490 
491 	do_getsockopt_bogus_sf_data(fd, MPTCP_SUBFLOW_ADDRS);
492 }
493 
494 static void do_getsockopt_mptcp_full_info(struct so_state *s, int fd)
495 {
496 	size_t data_size = sizeof(struct mptcp_full_info);
497 	struct mptcp_subflow_info sfinfo[2];
498 	struct tcp_info tcp_info[2];
499 	struct mptcp_full_info mfi;
500 	socklen_t olen;
501 	int ret;
502 
503 	memset(&mfi, 0, data_size);
504 	memset(tcp_info, 0, sizeof(tcp_info));
505 	memset(sfinfo, 0, sizeof(sfinfo));
506 
507 	mfi.size_tcpinfo_user = sizeof(struct tcp_info);
508 	mfi.size_sfinfo_user = sizeof(struct mptcp_subflow_info);
509 	mfi.size_arrays_user = 2;
510 	mfi.subflow_info = (unsigned long)&sfinfo[0];
511 	mfi.tcp_info = (unsigned long)&tcp_info[0];
512 	olen = data_size;
513 
514 	ret = getsockopt(fd, SOL_MPTCP, MPTCP_FULL_INFO, &mfi, &olen);
515 	if (ret < 0) {
516 		if (errno == EOPNOTSUPP) {
517 			perror("MPTCP_FULL_INFO test skipped");
518 			return;
519 		}
520 		xerror("getsockopt MPTCP_FULL_INFO");
521 	}
522 
523 	assert(olen <= data_size);
524 	assert(mfi.size_tcpinfo_kernel > 0);
525 	assert(mfi.size_tcpinfo_user ==
526 	       MIN(mfi.size_tcpinfo_kernel, sizeof(struct tcp_info)));
527 	assert(mfi.size_sfinfo_kernel > 0);
528 	assert(mfi.size_sfinfo_user ==
529 	       MIN(mfi.size_sfinfo_kernel, sizeof(struct mptcp_subflow_info)));
530 	assert(mfi.num_subflows == 1);
531 
532 	/* Tolerate future extension to mptcp_info struct and running newer
533 	 * test on top of older kernel.
534 	 * Anyway any kernel supporting MPTCP_FULL_INFO must at least include
535 	 * the following in mptcp_info.
536 	 */
537 	assert(olen > (socklen_t)__builtin_offsetof(struct mptcp_full_info, tcp_info));
538 	assert(mfi.mptcp_info.mptcpi_subflows == 0);
539 	assert(mfi.mptcp_info.mptcpi_bytes_sent == s->last_sample.mptcpi_bytes_sent);
540 	assert(mfi.mptcp_info.mptcpi_bytes_received == s->last_sample.mptcpi_bytes_received);
541 
542 	assert(sfinfo[0].id == 1);
543 	assert(tcp_info[0].tcpi_bytes_sent == s->tcp_info.tcpi_bytes_sent);
544 	assert(tcp_info[0].tcpi_bytes_received == s->tcp_info.tcpi_bytes_received);
545 	assert(!memcmp(&sfinfo->addrs, &s->addrs, sizeof(struct mptcp_subflow_addrs)));
546 }
547 
548 static void do_getsockopts(struct so_state *s, int fd, size_t r, size_t w)
549 {
550 	do_getsockopt_mptcp_info(s, fd, w);
551 
552 	do_getsockopt_tcp_info(s, fd, r, w);
553 
554 	do_getsockopt_subflow_addrs(s, fd);
555 
556 	if (r)
557 		do_getsockopt_mptcp_full_info(s, fd);
558 }
559 
560 static void connect_one_server(int fd, int pipefd)
561 {
562 	char buf[4096], buf2[4096];
563 	size_t len, i, total;
564 	struct so_state s;
565 	bool eof = false;
566 	ssize_t ret;
567 
568 	memset(&s, 0, sizeof(s));
569 
570 	len = rand() % (sizeof(buf) - 1);
571 
572 	if (len < 128)
573 		len = 128;
574 
575 	for (i = 0; i < len ; i++) {
576 		buf[i] = rand() % 26;
577 		buf[i] += 'A';
578 	}
579 
580 	buf[i] = '\n';
581 
582 	do_getsockopts(&s, fd, 0, 0);
583 
584 	/* un-block server */
585 	ret = read(pipefd, buf2, 4);
586 	assert(ret == 4);
587 	close(pipefd);
588 
589 	assert(strncmp(buf2, "xmit", 4) == 0);
590 
591 	ret = write(fd, buf, len);
592 	if (ret < 0)
593 		die_perror("write");
594 
595 	if (ret != (ssize_t)len)
596 		xerror("short write");
597 
598 	total = 0;
599 	do {
600 		ret = read(fd, buf2 + total, sizeof(buf2) - total);
601 		if (ret < 0)
602 			die_perror("read");
603 		if (ret == 0) {
604 			eof = true;
605 			break;
606 		}
607 
608 		total += ret;
609 	} while (total < len);
610 
611 	if (total != len)
612 		xerror("total %lu, len %lu eof %d\n", total, len, eof);
613 
614 	if (memcmp(buf, buf2, len))
615 		xerror("data corruption");
616 
617 	if (s.tcpi_rcv_delta)
618 		assert(s.tcpi_rcv_delta <= total);
619 
620 	do_getsockopts(&s, fd, ret, ret);
621 
622 	if (eof)
623 		total += 1; /* sequence advances due to FIN */
624 
625 	assert(s.mptcpi_rcv_delta == (uint64_t)total);
626 	close(fd);
627 }
628 
629 static void process_one_client(int fd, int pipefd)
630 {
631 	ssize_t ret, ret2, ret3;
632 	struct so_state s;
633 	char buf[4096];
634 
635 	memset(&s, 0, sizeof(s));
636 	do_getsockopts(&s, fd, 0, 0);
637 
638 	ret = write(pipefd, "xmit", 4);
639 	assert(ret == 4);
640 
641 	ret = read(fd, buf, sizeof(buf));
642 	if (ret < 0)
643 		die_perror("read");
644 
645 	assert(s.mptcpi_rcv_delta <= (uint64_t)ret);
646 
647 	if (s.tcpi_rcv_delta)
648 		assert(s.tcpi_rcv_delta == (uint64_t)ret);
649 
650 	ret2 = write(fd, buf, ret);
651 	if (ret2 < 0)
652 		die_perror("write");
653 
654 	/* wait for hangup */
655 	ret3 = read(fd, buf, 1);
656 	if (ret3 != 0)
657 		xerror("expected EOF, got %lu", ret3);
658 
659 	do_getsockopts(&s, fd, ret, ret2);
660 	if (s.mptcpi_rcv_delta != (uint64_t)ret + 1)
661 		xerror("mptcpi_rcv_delta %" PRIu64 ", expect %" PRIu64, s.mptcpi_rcv_delta, ret + 1, s.mptcpi_rcv_delta - ret);
662 
663 	/* be nice when running on top of older kernel */
664 	if (s.pkt_stats_avail) {
665 		if (s.last_sample.mptcpi_bytes_sent != ret2)
666 			xerror("mptcpi_bytes_sent %" PRIu64 ", expect %" PRIu64,
667 			       s.last_sample.mptcpi_bytes_sent, ret2,
668 			       s.last_sample.mptcpi_bytes_sent - ret2);
669 		if (s.last_sample.mptcpi_bytes_received != ret)
670 			xerror("mptcpi_bytes_received %" PRIu64 ", expect %" PRIu64,
671 			       s.last_sample.mptcpi_bytes_received, ret,
672 			       s.last_sample.mptcpi_bytes_received - ret);
673 		if (s.last_sample.mptcpi_bytes_acked != ret)
674 			xerror("mptcpi_bytes_acked %" PRIu64 ", expect %" PRIu64,
675 			       s.last_sample.mptcpi_bytes_acked, ret2,
676 			       s.last_sample.mptcpi_bytes_acked - ret2);
677 	}
678 
679 	close(fd);
680 }
681 
682 static int xaccept(int s)
683 {
684 	int fd = accept(s, NULL, 0);
685 
686 	if (fd < 0)
687 		die_perror("accept");
688 
689 	return fd;
690 }
691 
692 static int server(int pipefd)
693 {
694 	int fd = -1, r;
695 
696 	switch (pf) {
697 	case AF_INET:
698 		fd = sock_listen_mptcp("127.0.0.1", "15432");
699 		break;
700 	case AF_INET6:
701 		fd = sock_listen_mptcp("::1", "15432");
702 		break;
703 	default:
704 		xerror("Unknown pf %d\n", pf);
705 		break;
706 	}
707 
708 	r = write(pipefd, "conn", 4);
709 	assert(r == 4);
710 
711 	alarm(15);
712 	r = xaccept(fd);
713 
714 	process_one_client(r, pipefd);
715 
716 	return 0;
717 }
718 
719 static void test_ip_tos_sockopt(int fd)
720 {
721 	uint8_t tos_in, tos_out;
722 	socklen_t s;
723 	int r;
724 
725 	tos_in = rand() & 0xfc;
726 	r = setsockopt(fd, SOL_IP, IP_TOS, &tos_in, sizeof(tos_out));
727 	if (r != 0)
728 		die_perror("setsockopt IP_TOS");
729 
730 	tos_out = 0;
731 	s = sizeof(tos_out);
732 	r = getsockopt(fd, SOL_IP, IP_TOS, &tos_out, &s);
733 	if (r != 0)
734 		die_perror("getsockopt IP_TOS");
735 
736 	if (tos_in != tos_out)
737 		xerror("tos %x != %x socklen_t %d\n", tos_in, tos_out, s);
738 
739 	if (s != 1)
740 		xerror("tos should be 1 byte");
741 
742 	s = 0;
743 	r = getsockopt(fd, SOL_IP, IP_TOS, &tos_out, &s);
744 	if (r != 0)
745 		die_perror("getsockopt IP_TOS 0");
746 	if (s != 0)
747 		xerror("expect socklen_t == 0");
748 
749 	s = -1;
750 	r = getsockopt(fd, SOL_IP, IP_TOS, &tos_out, &s);
751 	if (r != -1 && errno != EINVAL)
752 		die_perror("getsockopt IP_TOS did not indicate -EINVAL");
753 	if (s != -1)
754 		xerror("expect socklen_t == -1");
755 }
756 
757 static int client(int pipefd)
758 {
759 	int fd = -1;
760 
761 	alarm(15);
762 
763 	switch (pf) {
764 	case AF_INET:
765 		fd = sock_connect_mptcp("127.0.0.1", "15432", IPPROTO_MPTCP);
766 		break;
767 	case AF_INET6:
768 		fd = sock_connect_mptcp("::1", "15432", IPPROTO_MPTCP);
769 		break;
770 	default:
771 		xerror("Unknown pf %d\n", pf);
772 	}
773 
774 	test_ip_tos_sockopt(fd);
775 
776 	connect_one_server(fd, pipefd);
777 
778 	return 0;
779 }
780 
781 static pid_t xfork(void)
782 {
783 	pid_t p = fork();
784 
785 	if (p < 0)
786 		die_perror("fork");
787 
788 	return p;
789 }
790 
791 static int rcheck(int wstatus, const char *what)
792 {
793 	if (WIFEXITED(wstatus)) {
794 		if (WEXITSTATUS(wstatus) == 0)
795 			return 0;
796 		fprintf(stderr, "%s exited, status=%d\n", what, WEXITSTATUS(wstatus));
797 		return WEXITSTATUS(wstatus);
798 	} else if (WIFSIGNALED(wstatus)) {
799 		xerror("%s killed by signal %d\n", what, WTERMSIG(wstatus));
800 	} else if (WIFSTOPPED(wstatus)) {
801 		xerror("%s stopped by signal %d\n", what, WSTOPSIG(wstatus));
802 	}
803 
804 	return 111;
805 }
806 
807 static void init_rng(void)
808 {
809 	int fd = open("/dev/urandom", O_RDONLY);
810 
811 	if (fd >= 0) {
812 		unsigned int foo;
813 		ssize_t ret;
814 
815 		/* can't fail */
816 		ret = read(fd, &foo, sizeof(foo));
817 		assert(ret == sizeof(foo));
818 
819 		close(fd);
820 		srand(foo);
821 	} else {
822 		srand(time(NULL));
823 	}
824 }
825 
826 int main(int argc, char *argv[])
827 {
828 	int e1, e2, wstatus;
829 	pid_t s, c, ret;
830 	int pipefds[2];
831 
832 	parse_opts(argc, argv);
833 
834 	init_rng();
835 
836 	e1 = pipe(pipefds);
837 	if (e1 < 0)
838 		die_perror("pipe");
839 
840 	s = xfork();
841 	if (s == 0)
842 		return server(pipefds[1]);
843 
844 	close(pipefds[1]);
845 
846 	/* wait until server bound a socket */
847 	e1 = read(pipefds[0], &e1, 4);
848 	assert(e1 == 4);
849 
850 	c = xfork();
851 	if (c == 0)
852 		return client(pipefds[0]);
853 
854 	close(pipefds[0]);
855 
856 	ret = waitpid(s, &wstatus, 0);
857 	if (ret == -1)
858 		die_perror("waitpid");
859 	e1 = rcheck(wstatus, "server");
860 	ret = waitpid(c, &wstatus, 0);
861 	if (ret == -1)
862 		die_perror("waitpid");
863 	e2 = rcheck(wstatus, "client");
864 
865 	return e1 ? e1 : e2;
866 }
867