1 // SPDX-License-Identifier: GPL-2.0
11 #include <linux/netlink.h>
12 #include <linux/rtnetlink.h>
13 #include <netinet/if_ether.h>
14 #include <netinet/ip.h>
15 #include <netinet/ip6.h>
16 #include <netinet/udp.h>
21 #include <sys/ioctl.h>
22 #include <sys/socket.h>
25 #include <sys/types.h>
29 #define ETH_MAX_MTU 0xFFFFU
33 #define UDP_SEGMENT 103
36 #ifndef UDP_MAX_SEGMENTS
37 #define UDP_MAX_SEGMENTS (1 << 6UL)
40 #define CONST_MTU_TEST 1500
42 #define CONST_HDRLEN_V4 (sizeof(struct iphdr) + sizeof(struct udphdr))
43 #define CONST_HDRLEN_V6 (sizeof(struct ip6_hdr) + sizeof(struct udphdr))
45 #define CONST_MSS_V4 (CONST_MTU_TEST - CONST_HDRLEN_V4)
46 #define CONST_MSS_V6 (CONST_MTU_TEST - CONST_HDRLEN_V6)
48 #define CONST_MAX_SEGS_V4 (ETH_MAX_MTU / CONST_MSS_V4)
49 #define CONST_MAX_SEGS_V6 (ETH_MAX_MTU / CONST_MSS_V6)
51 static bool cfg_do_ipv4;
52 static bool cfg_do_ipv6;
53 static bool cfg_do_connected;
54 static bool cfg_do_connectionless;
55 static bool cfg_do_msgmore;
56 static bool cfg_do_setsockopt;
57 static int cfg_specific_test_id = -1;
59 static const char cfg_ifname[] = "lo";
60 static unsigned short cfg_port = 9000;
62 static char buf[ETH_MAX_MTU];
65 int tlen; /* send() buffer size, may exceed mss */
66 bool tfail; /* send() call is expected to fail */
67 int gso_len; /* mss after applying gso */
68 int r_num_mss; /* recv(): number of calls of full mss */
69 int r_len_last; /* recv(): size of last non-mss dgram, if any */
72 const struct in6_addr addr6 = IN6ADDR_LOOPBACK_INIT;
73 const struct in_addr addr4 = { .s_addr = __constant_htonl(INADDR_LOOPBACK + 2) };
75 struct testcase testcases_v4[] = {
77 /* no GSO: send a single byte */
82 /* no GSO: send a single MSS */
87 /* no GSO: send a single MSS + 1B: fail */
88 .tlen = CONST_MSS_V4 + 1,
92 /* send a single MSS: will fall back to no GSO */
94 .gso_len = CONST_MSS_V4,
98 /* send a single MSS + 1B */
99 .tlen = CONST_MSS_V4 + 1,
100 .gso_len = CONST_MSS_V4,
105 /* send exactly 2 MSS */
106 .tlen = CONST_MSS_V4 * 2,
107 .gso_len = CONST_MSS_V4,
111 /* send 2 MSS + 1B */
112 .tlen = (CONST_MSS_V4 * 2) + 1,
113 .gso_len = CONST_MSS_V4,
119 .tlen = (ETH_MAX_MTU / CONST_MSS_V4) * CONST_MSS_V4,
120 .gso_len = CONST_MSS_V4,
121 .r_num_mss = (ETH_MAX_MTU / CONST_MSS_V4),
126 .tlen = ETH_MAX_MTU - CONST_HDRLEN_V4,
127 .gso_len = CONST_MSS_V4,
128 .r_num_mss = CONST_MAX_SEGS_V4,
129 .r_len_last = ETH_MAX_MTU - CONST_HDRLEN_V4 -
130 (CONST_MAX_SEGS_V4 * CONST_MSS_V4),
133 /* send MAX + 1: fail */
134 .tlen = ETH_MAX_MTU - CONST_HDRLEN_V4 + 1,
135 .gso_len = CONST_MSS_V4,
139 /* send a single 1B MSS: will fall back to no GSO */
145 /* send 2 1B segments */
151 /* send 2B + 2B + 1B segments */
158 /* send max number of min sized segments */
159 .tlen = UDP_MAX_SEGMENTS,
161 .r_num_mss = UDP_MAX_SEGMENTS,
164 /* send max number + 1 of min sized segments: fail */
165 .tlen = UDP_MAX_SEGMENTS + 1,
175 #define IP6_MAX_MTU (ETH_MAX_MTU + sizeof(struct ip6_hdr))
178 struct testcase testcases_v6[] = {
180 /* no GSO: send a single byte */
185 /* no GSO: send a single MSS */
186 .tlen = CONST_MSS_V6,
190 /* no GSO: send a single MSS + 1B: fail */
191 .tlen = CONST_MSS_V6 + 1,
195 /* send a single MSS: will fall back to no GSO */
196 .tlen = CONST_MSS_V6,
197 .gso_len = CONST_MSS_V6,
201 /* send a single MSS + 1B */
202 .tlen = CONST_MSS_V6 + 1,
203 .gso_len = CONST_MSS_V6,
208 /* send exactly 2 MSS */
209 .tlen = CONST_MSS_V6 * 2,
210 .gso_len = CONST_MSS_V6,
214 /* send 2 MSS + 1B */
215 .tlen = (CONST_MSS_V6 * 2) + 1,
216 .gso_len = CONST_MSS_V6,
222 .tlen = (IP6_MAX_MTU / CONST_MSS_V6) * CONST_MSS_V6,
223 .gso_len = CONST_MSS_V6,
224 .r_num_mss = (IP6_MAX_MTU / CONST_MSS_V6),
229 .tlen = IP6_MAX_MTU - CONST_HDRLEN_V6,
230 .gso_len = CONST_MSS_V6,
231 .r_num_mss = CONST_MAX_SEGS_V6,
232 .r_len_last = IP6_MAX_MTU - CONST_HDRLEN_V6 -
233 (CONST_MAX_SEGS_V6 * CONST_MSS_V6),
236 /* send MAX + 1: fail */
237 .tlen = IP6_MAX_MTU - CONST_HDRLEN_V6 + 1,
238 .gso_len = CONST_MSS_V6,
242 /* send a single 1B MSS: will fall back to no GSO */
248 /* send 2 1B segments */
254 /* send 2B + 2B + 1B segments */
261 /* send max number of min sized segments */
262 .tlen = UDP_MAX_SEGMENTS,
264 .r_num_mss = UDP_MAX_SEGMENTS,
267 /* send max number + 1 of min sized segments: fail */
268 .tlen = UDP_MAX_SEGMENTS + 1,
277 static unsigned int get_device_mtu(int fd, const char *ifname)
281 memset(&ifr, 0, sizeof(ifr));
283 strcpy(ifr.ifr_name, ifname);
285 if (ioctl(fd, SIOCGIFMTU, &ifr))
286 error(1, errno, "ioctl get mtu");
291 static void __set_device_mtu(int fd, const char *ifname, unsigned int mtu)
295 memset(&ifr, 0, sizeof(ifr));
298 strcpy(ifr.ifr_name, ifname);
300 if (ioctl(fd, SIOCSIFMTU, &ifr))
301 error(1, errno, "ioctl set mtu");
304 static void set_device_mtu(int fd, int mtu)
308 val = get_device_mtu(fd, cfg_ifname);
309 fprintf(stderr, "device mtu (orig): %u\n", val);
311 __set_device_mtu(fd, cfg_ifname, mtu);
312 val = get_device_mtu(fd, cfg_ifname);
314 error(1, 0, "unable to set device mtu to %u\n", val);
316 fprintf(stderr, "device mtu (test): %u\n", val);
319 static void set_pmtu_discover(int fd, bool is_ipv4)
321 int level, name, val;
325 name = IP_MTU_DISCOVER;
326 val = IP_PMTUDISC_DO;
329 name = IPV6_MTU_DISCOVER;
330 val = IPV6_PMTUDISC_DO;
333 if (setsockopt(fd, level, name, &val, sizeof(val)))
334 error(1, errno, "setsockopt path mtu");
337 static unsigned int get_path_mtu(int fd, bool is_ipv4)
343 vallen = sizeof(mtu);
345 ret = getsockopt(fd, SOL_IP, IP_MTU, &mtu, &vallen);
347 ret = getsockopt(fd, SOL_IPV6, IPV6_MTU, &mtu, &vallen);
350 error(1, errno, "getsockopt mtu");
353 fprintf(stderr, "path mtu (read): %u\n", mtu);
357 /* very wordy version of system("ip route add dev lo mtu 1500 127.0.0.3/32") */
358 static void set_route_mtu(int mtu, bool is_ipv4)
360 struct sockaddr_nl nladdr = { .nl_family = AF_NETLINK };
364 char data[NLMSG_ALIGN(sizeof(*nh)) +
365 NLMSG_ALIGN(sizeof(*rt)) +
366 NLMSG_ALIGN(RTA_LENGTH(sizeof(addr6))) +
367 NLMSG_ALIGN(RTA_LENGTH(sizeof(int))) +
368 NLMSG_ALIGN(RTA_LENGTH(0) + RTA_LENGTH(sizeof(int)))];
369 int fd, ret, alen, off = 0;
371 alen = is_ipv4 ? sizeof(addr4) : sizeof(addr6);
373 fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
375 error(1, errno, "socket netlink");
377 memset(data, 0, sizeof(data));
380 nh->nlmsg_type = RTM_NEWROUTE;
381 nh->nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE;
382 off += NLMSG_ALIGN(sizeof(*nh));
384 rt = (void *)(data + off);
385 rt->rtm_family = is_ipv4 ? AF_INET : AF_INET6;
386 rt->rtm_table = RT_TABLE_MAIN;
387 rt->rtm_dst_len = alen << 3;
388 rt->rtm_protocol = RTPROT_BOOT;
389 rt->rtm_scope = RT_SCOPE_UNIVERSE;
390 rt->rtm_type = RTN_UNICAST;
391 off += NLMSG_ALIGN(sizeof(*rt));
393 rta = (void *)(data + off);
394 rta->rta_type = RTA_DST;
395 rta->rta_len = RTA_LENGTH(alen);
397 memcpy(RTA_DATA(rta), &addr4, alen);
399 memcpy(RTA_DATA(rta), &addr6, alen);
400 off += NLMSG_ALIGN(rta->rta_len);
402 rta = (void *)(data + off);
403 rta->rta_type = RTA_OIF;
404 rta->rta_len = RTA_LENGTH(sizeof(int));
405 *((int *)(RTA_DATA(rta))) = 1; //if_nametoindex("lo");
406 off += NLMSG_ALIGN(rta->rta_len);
408 /* MTU is a subtype in a metrics type */
409 rta = (void *)(data + off);
410 rta->rta_type = RTA_METRICS;
411 rta->rta_len = RTA_LENGTH(0) + RTA_LENGTH(sizeof(int));
412 off += NLMSG_ALIGN(rta->rta_len);
414 /* now fill MTU subtype. Note that it fits within above rta_len */
415 rta = (void *)(((char *) rta) + RTA_LENGTH(0));
416 rta->rta_type = RTAX_MTU;
417 rta->rta_len = RTA_LENGTH(sizeof(int));
418 *((int *)(RTA_DATA(rta))) = mtu;
422 ret = sendto(fd, data, off, 0, (void *)&nladdr, sizeof(nladdr));
424 error(1, errno, "send netlink: %uB != %uB\n", ret, off);
427 error(1, errno, "close netlink");
429 fprintf(stderr, "route mtu (test): %u\n", mtu);
432 static bool __send_one(int fd, struct msghdr *msg, int flags)
436 ret = sendmsg(fd, msg, flags);
438 (errno == EMSGSIZE || errno == ENOMEM || errno == EINVAL))
441 error(1, errno, "sendmsg");
442 if (ret != msg->msg_iov->iov_len)
443 error(1, 0, "sendto: %d != %llu", ret,
444 (unsigned long long)msg->msg_iov->iov_len);
446 error(1, 0, "sendmsg: return flags 0x%x\n", msg->msg_flags);
451 static bool send_one(int fd, int len, int gso_len,
452 struct sockaddr *addr, socklen_t alen)
454 char control[CMSG_SPACE(sizeof(uint16_t))] = {0};
455 struct msghdr msg = {0};
456 struct iovec iov = {0};
466 msg.msg_namelen = alen;
468 if (gso_len && !cfg_do_setsockopt) {
469 msg.msg_control = control;
470 msg.msg_controllen = sizeof(control);
472 cm = CMSG_FIRSTHDR(&msg);
473 cm->cmsg_level = SOL_UDP;
474 cm->cmsg_type = UDP_SEGMENT;
475 cm->cmsg_len = CMSG_LEN(sizeof(uint16_t));
476 *((uint16_t *) CMSG_DATA(cm)) = gso_len;
479 /* If MSG_MORE, send 1 byte followed by remainder */
480 if (cfg_do_msgmore && len > 1) {
482 if (!__send_one(fd, &msg, MSG_MORE))
483 error(1, 0, "send 1B failed");
486 iov.iov_len = len - 1;
489 return __send_one(fd, &msg, 0);
492 static int recv_one(int fd, int flags)
496 ret = recv(fd, buf, sizeof(buf), flags);
497 if (ret == -1 && errno == EAGAIN && (flags & MSG_DONTWAIT))
500 error(1, errno, "recv");
505 static void run_one(struct testcase *test, int fdt, int fdr,
506 struct sockaddr *addr, socklen_t alen)
508 int i, ret, val, mss;
511 fprintf(stderr, "ipv%d tx:%d gso:%d %s\n",
512 addr->sa_family == AF_INET ? 4 : 6,
513 test->tlen, test->gso_len,
514 test->tfail ? "(fail)" : "");
517 if (cfg_do_setsockopt) {
518 if (setsockopt(fdt, SOL_UDP, UDP_SEGMENT, &val, sizeof(val)))
519 error(1, errno, "setsockopt udp segment");
522 sent = send_one(fdt, test->tlen, test->gso_len, addr, alen);
523 if (sent && test->tfail)
524 error(1, 0, "send succeeded while expecting failure");
525 if (!sent && !test->tfail)
526 error(1, 0, "send failed while expecting success");
533 mss = addr->sa_family == AF_INET ? CONST_MSS_V4 : CONST_MSS_V6;
536 /* Recv all full MSS datagrams */
537 for (i = 0; i < test->r_num_mss; i++) {
538 ret = recv_one(fdr, 0);
540 error(1, 0, "recv.%d: %d != %d", i, ret, mss);
543 /* Recv the non-full last datagram, if tlen was not a multiple of mss */
544 if (test->r_len_last) {
545 ret = recv_one(fdr, 0);
546 if (ret != test->r_len_last)
547 error(1, 0, "recv.%d: %d != %d (last)",
548 i, ret, test->r_len_last);
551 /* Verify received all data */
552 ret = recv_one(fdr, MSG_DONTWAIT);
554 error(1, 0, "recv: unexpected datagram");
557 static void run_all(int fdt, int fdr, struct sockaddr *addr, socklen_t alen)
559 struct testcase *tests, *test;
561 tests = addr->sa_family == AF_INET ? testcases_v4 : testcases_v6;
563 for (test = tests; test->tlen; test++) {
564 /* if a specific test is given, then skip all others */
565 if (cfg_specific_test_id == -1 ||
566 cfg_specific_test_id == test - tests)
567 run_one(test, fdt, fdr, addr, alen);
571 static void run_test(struct sockaddr *addr, socklen_t alen)
573 struct timeval tv = { .tv_usec = 100 * 1000 };
576 fdr = socket(addr->sa_family, SOCK_DGRAM, 0);
578 error(1, errno, "socket r");
580 if (bind(fdr, addr, alen))
581 error(1, errno, "bind");
583 /* Have tests fail quickly instead of hang */
584 if (setsockopt(fdr, SOL_SOCKET, SO_RCVTIMEO, &tv, sizeof(tv)))
585 error(1, errno, "setsockopt rcv timeout");
587 fdt = socket(addr->sa_family, SOCK_DGRAM, 0);
589 error(1, errno, "socket t");
591 /* Do not fragment these datagrams: only succeed if GSO works */
592 set_pmtu_discover(fdt, addr->sa_family == AF_INET);
594 if (cfg_do_connectionless) {
595 set_device_mtu(fdt, CONST_MTU_TEST);
596 run_all(fdt, fdr, addr, alen);
599 if (cfg_do_connected) {
600 set_device_mtu(fdt, CONST_MTU_TEST + 100);
601 set_route_mtu(CONST_MTU_TEST, addr->sa_family == AF_INET);
603 if (connect(fdt, addr, alen))
604 error(1, errno, "connect");
606 val = get_path_mtu(fdt, addr->sa_family == AF_INET);
607 if (val != CONST_MTU_TEST)
608 error(1, 0, "bad path mtu %u\n", val);
610 run_all(fdt, fdr, addr, 0 /* use connected addr */);
614 error(1, errno, "close t");
616 error(1, errno, "close r");
619 static void run_test_v4(void)
621 struct sockaddr_in addr = {0};
623 addr.sin_family = AF_INET;
624 addr.sin_port = htons(cfg_port);
625 addr.sin_addr = addr4;
627 run_test((void *)&addr, sizeof(addr));
630 static void run_test_v6(void)
632 struct sockaddr_in6 addr = {0};
634 addr.sin6_family = AF_INET6;
635 addr.sin6_port = htons(cfg_port);
636 addr.sin6_addr = addr6;
638 run_test((void *)&addr, sizeof(addr));
641 static void parse_opts(int argc, char **argv)
645 while ((c = getopt(argc, argv, "46cCmst:")) != -1) {
654 cfg_do_connected = true;
657 cfg_do_connectionless = true;
660 cfg_do_msgmore = true;
663 cfg_do_setsockopt = true;
666 cfg_specific_test_id = strtoul(optarg, NULL, 0);
669 error(1, 0, "%s: parse error", argv[0]);
674 int main(int argc, char **argv)
676 parse_opts(argc, argv);
683 fprintf(stderr, "OK\n");