1 /* $OpenBSD: tcpbench.c,v 1.70 2024/03/21 16:46:04 bluhm Exp $ */
2
3 /*
4 * Copyright (c) 2008 Damien Miller <djm@mindrot.org>
5 * Copyright (c) 2011 Christiano F. Haesbaert <haesbaert@haesbaert.org>
6 *
7 * Permission to use, copy, modify, and distribute this software for any
8 * purpose with or without fee is hereby granted, provided that the above
9 * copyright notice and this permission notice appear in all copies.
10 *
11 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
12 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
13 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
14 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
15 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
16 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
17 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
18 */
19
20 #include <sys/types.h>
21 #include <sys/time.h>
22 #include <sys/socket.h>
23 #include <sys/socketvar.h>
24 #include <sys/resource.h>
25 #include <sys/queue.h>
26 #include <sys/un.h>
27
28 #include <net/route.h>
29
30 #include <netinet/in.h>
31 #include <netinet/ip.h>
32 #include <netinet/tcp.h>
33 #include <netinet/tcp_timer.h>
34 #include <netinet/tcp_fsm.h>
35 #include <netinet/in_pcb.h>
36 #include <netinet/tcp_var.h>
37
38 #include <arpa/inet.h>
39
40 #include <unistd.h>
41 #include <limits.h>
42 #include <stdlib.h>
43 #include <stdio.h>
44 #include <string.h>
45 #include <errno.h>
46 #include <event.h>
47 #include <netdb.h>
48 #include <signal.h>
49 #include <err.h>
50 #include <fcntl.h>
51 #include <poll.h>
52 #include <paths.h>
53 #include <math.h>
54
55 #define DEFAULT_PORT "12345"
56 #define DEFAULT_STATS_INTERVAL 1000 /* ms */
57 #define DEFAULT_BUF (256 * 1024)
58 #define DEFAULT_UDP_PKT (1500 - 28) /* TODO don't hardcode this */
59 #define TCP_MODE !ptb->uflag
60 #define UDP_MODE ptb->uflag
61 #define MAX_FD 1024
62
63 /* Our tcpbench globals */
64 struct {
65 int Dflag; /* Socket debug */
66 int Sflag; /* Socket buffer size */
67 u_int rflag; /* Report rate (ms) */
68 int sflag; /* True if server */
69 int Tflag; /* ToS if != -1 */
70 int vflag; /* Verbose */
71 int uflag; /* UDP mode */
72 int Uflag; /* UNIX (AF_LOCAL) mode */
73 int Rflag; /* randomize client write size */
74 char **kvars; /* Kvm enabled vars */
75 char *dummybuf; /* IO buffer */
76 size_t dummybuf_len; /* IO buffer len */
77 } tcpbench, *ptb;
78
79 struct tcpservsock {
80 struct event ev;
81 struct event evt;
82 int fd;
83 };
84
85 /* stats for a single tcp connection, udp uses only one */
86 struct statctx {
87 TAILQ_ENTRY(statctx) entry;
88 struct timeval t_start, t_last;
89 unsigned long long bytes;
90 int fd;
91 char *buf;
92 size_t buflen;
93 struct event ev;
94 /* TCP only */
95 struct tcpservsock *tcp_ts;
96 /* UDP only */
97 u_long udp_slice_pkts;
98 };
99
100 struct statctx *udp_sc; /* singleton */
101
102 static void signal_handler(int, short, void *);
103 static void saddr_ntop(const struct sockaddr *, socklen_t, char *, size_t);
104 static void set_slice_timer(int);
105 static void print_tcp_header(void);
106 static void list_kvars(void);
107 static void check_kvar(const char *);
108 static char ** check_prepare_kvars(char *);
109 static void stats_prepare(struct statctx *);
110 static void summary_display(void);
111 static void tcp_stats_display(unsigned long long, long double, float,
112 struct statctx *, struct tcp_info *);
113 static void tcp_process_slice(int, short, void *);
114 static void tcp_server_handle_sc(int, short, void *);
115 static void tcp_server_accept(int, short, void *);
116 static void server_init(struct addrinfo *);
117 static void client_handle_sc(int, short, void *);
118 static void client_init(struct addrinfo *, int, struct addrinfo *);
119 static int clock_gettime_tv(clockid_t, struct timeval *);
120 static void udp_server_handle_sc(int, short, void *);
121 static void udp_process_slice(int, short, void *);
122 static int map_tos(char *, int *);
123 static void quit(int, short, void *);
124 static void wrapup(int);
125
126 /*
127 * We account the mainstats here, that is the stats
128 * for all connections, all variables starting with slice
129 * are used to account information for the timeslice
130 * between each output. Peak variables record the highest
131 * between all slices so far.
132 */
133 static struct {
134 struct timeval t_first; /* first connect / packet */
135 unsigned long long total_bytes; /* bytes since t_first */
136 unsigned long long n_slices; /* slices since start */
137 unsigned long long slice_bytes; /* bytes since slice reset */
138 long double peak_mbps; /* peak mbps so far */
139 long double floor_mbps; /* floor mbps so far */
140 long double mean_mbps; /* mean mbps so far */
141 long double nvariance_mbps; /* for online std dev */
142 int nconns; /* connected clients */
143 struct event timer; /* process timer */
144 const char *host; /* remote server for display */
145 } mainstats;
146
147 /* When adding variables, also add to tcp_stats_display() */
148 static const char *allowed_kvars[] = {
149 "last_ack_recv",
150 "last_ack_sent",
151 "last_data_recv",
152 "last_data_sent",
153 "max_sndwnd",
154 "options",
155 "rcv_adv",
156 "rcv_mss",
157 "rcv_nxt",
158 "rcv_ooopack",
159 "rcv_space",
160 "rcv_up",
161 "rcv_wscale",
162 "rfbuf_cnt",
163 "rfbuf_ts",
164 "rtt",
165 "rttmin",
166 "rttvar",
167 "snd_cwnd",
168 "snd_max",
169 "snd_mss",
170 "snd_nxt",
171 "snd_rexmitpack",
172 "snd_ssthresh",
173 "snd_una",
174 "snd_wl1",
175 "snd_wl2",
176 "snd_wnd",
177 "snd_wscale",
178 "snd_zerowin",
179 "so_rcv_sb_cc",
180 "so_rcv_sb_hiwat",
181 "so_rcv_sb_lowat",
182 "so_rcv_sb_wat",
183 "so_snd_sb_cc",
184 "so_snd_sb_hiwat",
185 "so_snd_sb_lowat",
186 "so_snd_sb_wat",
187 "ts_recent",
188 "ts_recent_age",
189 NULL
190 };
191
192 TAILQ_HEAD(, statctx) sc_queue;
193
194 static void __dead
usage(void)195 usage(void)
196 {
197 fprintf(stderr,
198 "usage: tcpbench -l\n"
199 " tcpbench [-46DRUuv] [-B buf] [-b sourceaddr] [-k kvars] [-n connections]\n"
200 " [-p port] [-r interval] [-S space] [-T toskeyword]\n"
201 " [-t secs] [-V rtable] hostname\n"
202 " tcpbench -s [-46DUuv] [-B buf] [-k kvars] [-p port] [-r interval]\n"
203 " [-S space] [-T toskeyword] [-V rtable] [hostname]\n");
204 exit(1);
205 }
206
207 static void
signal_handler(int sig,short event,void * bula)208 signal_handler(int sig, short event, void *bula)
209 {
210 /*
211 * signal handler rules don't apply, libevent decouples for us
212 */
213 switch (sig) {
214 case SIGINFO:
215 printf("\n");
216 wrapup(-1);
217 break;
218 case SIGINT:
219 printf("\n");
220 wrapup(0);
221 break; /* NOTREACHED */
222 case SIGTERM:
223 case SIGHUP:
224 warnx("Terminated by signal %d", sig);
225 wrapup(0);
226 break; /* NOTREACHED */
227 default:
228 errx(1, "unexpected signal %d", sig);
229 break; /* NOTREACHED */
230 }
231 }
232
233 static void
saddr_ntop(const struct sockaddr * addr,socklen_t alen,char * buf,size_t len)234 saddr_ntop(const struct sockaddr *addr, socklen_t alen, char *buf, size_t len)
235 {
236 char hbuf[NI_MAXHOST], pbuf[NI_MAXSERV];
237 int herr;
238
239 if (addr->sa_family == AF_UNIX) {
240 struct sockaddr_un *sun = (struct sockaddr_un *)addr;
241 snprintf(buf, len, "%s", sun->sun_path);
242 return;
243 }
244 if ((herr = getnameinfo(addr, alen, hbuf, sizeof(hbuf),
245 pbuf, sizeof(pbuf), NI_NUMERICHOST|NI_NUMERICSERV)) != 0) {
246 if (herr == EAI_SYSTEM)
247 err(1, "getnameinfo");
248 else
249 errx(1, "getnameinfo: %s", gai_strerror(herr));
250 }
251 snprintf(buf, len, "[%s]:%s", hbuf, pbuf);
252 }
253
254 static void
set_slice_timer(int on)255 set_slice_timer(int on)
256 {
257 struct timeval tv;
258
259 if (ptb->rflag == 0)
260 return;
261
262 if (on) {
263 if (evtimer_pending(&mainstats.timer, NULL))
264 return;
265 /* XXX Is there a better way to do this ? */
266 tv.tv_sec = ptb->rflag / 1000;
267 tv.tv_usec = (ptb->rflag % 1000) * 1000;
268
269 evtimer_add(&mainstats.timer, &tv);
270 } else if (evtimer_pending(&mainstats.timer, NULL))
271 evtimer_del(&mainstats.timer);
272 }
273
274 static int
clock_gettime_tv(clockid_t clock_id,struct timeval * tv)275 clock_gettime_tv(clockid_t clock_id, struct timeval *tv)
276 {
277 struct timespec ts;
278
279 if (clock_gettime(clock_id, &ts) == -1)
280 return (-1);
281
282 TIMESPEC_TO_TIMEVAL(tv, &ts);
283
284 return (0);
285 }
286
287 static void
print_tcp_header(void)288 print_tcp_header(void)
289 {
290 char **kv;
291
292 if (ptb->rflag == 0)
293 return;
294
295 printf("%12s %14s %12s %8s ", "elapsed_ms", "bytes", "mbps",
296 "bwidth");
297 for (kv = ptb->kvars; ptb->kvars != NULL && *kv != NULL; kv++)
298 printf("%s%s", kv != ptb->kvars ? "," : "", *kv);
299 printf("\n");
300 }
301
302 static void
check_kvar(const char * var)303 check_kvar(const char *var)
304 {
305 u_int i;
306
307 for (i = 0; allowed_kvars[i] != NULL; i++)
308 if (strcmp(allowed_kvars[i], var) == 0)
309 return;
310 errx(1, "Unrecognised kvar: %s", var);
311 }
312
313 static void
list_kvars(void)314 list_kvars(void)
315 {
316 u_int i;
317
318 printf("Supported kernel variables:\n");
319 for (i = 0; allowed_kvars[i] != NULL; i++)
320 printf("\t%s\n", allowed_kvars[i]);
321 }
322
323 static char **
check_prepare_kvars(char * list)324 check_prepare_kvars(char *list)
325 {
326 char *item, **ret = NULL;
327 u_int n = 0;
328
329 while ((item = strsep(&list, ", \t\n")) != NULL) {
330 check_kvar(item);
331 if ((ret = reallocarray(ret, (++n + 1), sizeof(*ret))) == NULL)
332 err(1, "reallocarray(kvars)");
333 if ((ret[n - 1] = strdup(item)) == NULL)
334 err(1, "strdup");
335 ret[n] = NULL;
336 }
337 return (ret);
338 }
339
340 static void
stats_prepare(struct statctx * sc)341 stats_prepare(struct statctx *sc)
342 {
343 sc->buf = ptb->dummybuf;
344 sc->buflen = ptb->dummybuf_len;
345
346 if (clock_gettime_tv(CLOCK_MONOTONIC, &sc->t_start) == -1)
347 err(1, "clock_gettime_tv");
348 sc->t_last = sc->t_start;
349 if (!timerisset(&mainstats.t_first))
350 mainstats.t_first = sc->t_start;
351 }
352
353 static void
summary_display(void)354 summary_display(void)
355 {
356 struct timeval t_cur, t_diff;
357 long double std_dev;
358 unsigned long long total_elapsed;
359 char *direction;
360
361 if (!ptb->sflag) {
362 direction = "sent";
363 printf("--- %s tcpbench statistics ---\n", mainstats.host);
364 } else {
365 direction = "received";
366 printf("--- tcpbench server statistics ---\n");
367 }
368
369 std_dev = sqrtl(mainstats.nvariance_mbps / mainstats.n_slices);
370
371 if (clock_gettime_tv(CLOCK_MONOTONIC, &t_cur) == -1)
372 err(1, "clock_gettime_tv");
373 timersub(&t_cur, &mainstats.t_first, &t_diff);
374 total_elapsed = t_diff.tv_sec * 1000 + t_diff.tv_usec / 1000;
375
376 printf("%llu bytes %s over %.3Lf seconds\n",
377 mainstats.total_bytes, direction, total_elapsed/1000.0L);
378 printf("bandwidth min/avg/max/std-dev = %.3Lf/%.3Lf/%.3Lf/%.3Lf Mbps\n",
379 mainstats.floor_mbps, mainstats.mean_mbps, mainstats.peak_mbps,
380 std_dev);
381 }
382
383 static void
tcp_stats_display(unsigned long long total_elapsed,long double mbps,float bwperc,struct statctx * sc,struct tcp_info * tcpi)384 tcp_stats_display(unsigned long long total_elapsed, long double mbps,
385 float bwperc, struct statctx *sc, struct tcp_info *tcpi)
386 {
387 int j;
388
389 printf("%12llu %14llu %12.3Lf %7.2f%% ", total_elapsed, sc->bytes,
390 mbps, bwperc);
391
392 if (ptb->kvars != NULL) {
393 for (j = 0; ptb->kvars[j] != NULL; j++) {
394 #define S(a) #a
395 #define P(b, v, f) \
396 if (strcmp(ptb->kvars[j], S(v)) == 0) { \
397 printf("%s"f, j > 0 ? "," : "", b->tcpi_##v); \
398 continue; \
399 }
400 P(tcpi, last_ack_recv, "%u")
401 P(tcpi, last_ack_sent, "%u")
402 P(tcpi, last_data_recv, "%u")
403 P(tcpi, last_data_sent, "%u")
404 P(tcpi, max_sndwnd, "%u")
405 P(tcpi, options, "%hhu")
406 P(tcpi, rcv_adv, "%u")
407 P(tcpi, rcv_mss, "%u")
408 P(tcpi, rcv_nxt, "%u")
409 P(tcpi, rcv_ooopack, "%u")
410 P(tcpi, rcv_space, "%u")
411 P(tcpi, rcv_up, "%u")
412 P(tcpi, rcv_wscale, "%hhu")
413 P(tcpi, rfbuf_cnt, "%u")
414 P(tcpi, rfbuf_ts, "%u")
415 P(tcpi, rtt, "%u")
416 P(tcpi, rttmin, "%u")
417 P(tcpi, rttvar, "%u")
418 P(tcpi, snd_cwnd, "%u")
419 P(tcpi, snd_max, "%u")
420 P(tcpi, snd_mss, "%u")
421 P(tcpi, snd_nxt, "%u")
422 P(tcpi, snd_rexmitpack, "%u")
423 P(tcpi, snd_ssthresh, "%u")
424 P(tcpi, snd_una, "%u")
425 P(tcpi, snd_wl1, "%u")
426 P(tcpi, snd_wl2, "%u")
427 P(tcpi, snd_wnd, "%u")
428 P(tcpi, snd_wscale, "%hhu")
429 P(tcpi, snd_zerowin, "%u")
430 P(tcpi, so_rcv_sb_cc, "%u")
431 P(tcpi, so_rcv_sb_hiwat, "%u")
432 P(tcpi, so_rcv_sb_lowat, "%u")
433 P(tcpi, so_rcv_sb_wat, "%u")
434 P(tcpi, so_snd_sb_cc, "%u")
435 P(tcpi, so_snd_sb_hiwat, "%u")
436 P(tcpi, so_snd_sb_lowat, "%u")
437 P(tcpi, so_snd_sb_wat, "%u")
438 P(tcpi, ts_recent, "%u")
439 P(tcpi, ts_recent_age, "%u")
440 #undef S
441 #undef P
442 }
443 }
444 printf("\n");
445 }
446
447 static void
tcp_process_slice(int fd,short event,void * bula)448 tcp_process_slice(int fd, short event, void *bula)
449 {
450 unsigned long long total_elapsed, since_last;
451 long double mbps, old_mean_mbps, slice_mbps = 0;
452 float bwperc;
453 struct statctx *sc;
454 struct timeval t_cur, t_diff;
455 struct tcp_info tcpi;
456 socklen_t tcpilen;
457
458 if (TAILQ_EMPTY(&sc_queue))
459 return; /* don't pollute stats */
460
461 mainstats.n_slices++;
462
463 TAILQ_FOREACH(sc, &sc_queue, entry) {
464 if (clock_gettime_tv(CLOCK_MONOTONIC, &t_cur) == -1)
465 err(1, "clock_gettime_tv");
466 if (ptb->kvars != NULL) { /* process kernel stats */
467 tcpilen = sizeof(tcpi);
468 if (getsockopt(sc->fd, IPPROTO_TCP, TCP_INFO,
469 &tcpi, &tcpilen) == -1)
470 err(1, "get tcp_info");
471 }
472
473 timersub(&t_cur, &sc->t_start, &t_diff);
474 total_elapsed = t_diff.tv_sec * 1000 + t_diff.tv_usec / 1000;
475 timersub(&t_cur, &sc->t_last, &t_diff);
476 since_last = t_diff.tv_sec * 1000 + t_diff.tv_usec / 1000;
477 if (since_last == 0)
478 continue;
479 bwperc = (sc->bytes * 100.0) / mainstats.slice_bytes;
480 mbps = (sc->bytes * 8) / (since_last * 1000.0);
481 slice_mbps += mbps;
482
483 tcp_stats_display(total_elapsed, mbps, bwperc, sc, &tcpi);
484
485 sc->t_last = t_cur;
486 sc->bytes = 0;
487 }
488
489 /* process stats for this slice */
490 if (slice_mbps > mainstats.peak_mbps)
491 mainstats.peak_mbps = slice_mbps;
492 if (slice_mbps < mainstats.floor_mbps)
493 mainstats.floor_mbps = slice_mbps;
494 old_mean_mbps = mainstats.mean_mbps;
495 mainstats.mean_mbps += (slice_mbps - mainstats.mean_mbps) /
496 mainstats.n_slices;
497
498 /* "Welford's method" for online variance
499 * see Knuth, TAoCP Volume 2, 3rd edn., p232 */
500 mainstats.nvariance_mbps += (slice_mbps - old_mean_mbps) *
501 (slice_mbps - mainstats.mean_mbps);
502
503 printf("Conn: %3d Mbps: %12.3Lf Peak Mbps: %12.3Lf Avg Mbps: %12.3Lf\n",
504 mainstats.nconns, slice_mbps, mainstats.peak_mbps,
505 mainstats.nconns ? slice_mbps / mainstats.nconns : 0);
506
507 mainstats.slice_bytes = 0;
508 set_slice_timer(mainstats.nconns > 0);
509 }
510
511 static void
udp_process_slice(int fd,short event,void * bula)512 udp_process_slice(int fd, short event, void *bula)
513 {
514 unsigned long long total_elapsed, since_last, pps;
515 long double old_mean_mbps, slice_mbps;
516 struct timeval t_cur, t_diff;
517
518 mainstats.n_slices++;
519
520 if (clock_gettime_tv(CLOCK_MONOTONIC, &t_cur) == -1)
521 err(1, "clock_gettime_tv");
522
523 timersub(&t_cur, &udp_sc->t_start, &t_diff);
524 total_elapsed = t_diff.tv_sec * 1000 + t_diff.tv_usec / 1000;
525
526 timersub(&t_cur, &udp_sc->t_last, &t_diff);
527 since_last = t_diff.tv_sec * 1000 + t_diff.tv_usec / 1000;
528 if (since_last == 0)
529 return;
530
531 slice_mbps = (udp_sc->bytes * 8) / (since_last * 1000.0);
532 pps = (udp_sc->udp_slice_pkts * 1000) / since_last;
533
534 if (slice_mbps > mainstats.peak_mbps)
535 mainstats.peak_mbps = slice_mbps;
536 if (slice_mbps < mainstats.floor_mbps)
537 mainstats.floor_mbps = slice_mbps;
538 old_mean_mbps = mainstats.mean_mbps;
539 mainstats.mean_mbps += (slice_mbps - mainstats.mean_mbps) /
540 mainstats.n_slices;
541
542 /* "Welford's method" for online variance
543 * see Knuth, TAoCP Volume 2, 3rd edn., p232 */
544 mainstats.nvariance_mbps += (slice_mbps - old_mean_mbps) *
545 (slice_mbps - mainstats.mean_mbps);
546
547 printf("Elapsed: %11llu Mbps: %11.3Lf Peak Mbps: %11.3Lf %s PPS: %7llu\n",
548 total_elapsed, slice_mbps, mainstats.peak_mbps,
549 ptb->sflag ? "Rx" : "Tx", pps);
550
551 /* Clean up this slice time */
552 udp_sc->t_last = t_cur;
553 udp_sc->bytes = 0;
554 udp_sc->udp_slice_pkts = 0;
555
556 mainstats.slice_bytes = 0;
557 set_slice_timer(1);
558 }
559
560 static void
udp_server_handle_sc(int fd,short event,void * bula)561 udp_server_handle_sc(int fd, short event, void *bula)
562 {
563 static int first_read = 1;
564 ssize_t n;
565
566 n = read(fd, ptb->dummybuf, ptb->dummybuf_len);
567 if (n == 0)
568 return;
569 else if (n == -1) {
570 if (errno != EINTR && errno != EWOULDBLOCK)
571 warn("fd %d read error", fd);
572 return;
573 }
574
575 if (ptb->vflag >= 3)
576 fprintf(stderr, "read: %zd bytes\n", n);
577 if (first_read) {
578 first_read = 0;
579 stats_prepare(udp_sc);
580 set_slice_timer(1);
581 }
582 /* Account packet */
583 udp_sc->udp_slice_pkts++;
584 udp_sc->bytes += n;
585 mainstats.slice_bytes += n;
586 mainstats.total_bytes += n;
587 }
588
589 static void
tcp_server_handle_sc(int fd,short event,void * v_sc)590 tcp_server_handle_sc(int fd, short event, void *v_sc)
591 {
592 struct statctx *sc = v_sc;
593 ssize_t n;
594
595 n = read(sc->fd, sc->buf, sc->buflen);
596 if (n == -1) {
597 if (errno != EINTR && errno != EWOULDBLOCK)
598 warn("fd %d read error", sc->fd);
599 return;
600 } else if (n == 0) {
601 if (ptb->vflag)
602 fprintf(stderr, "%8d closed by remote end\n", sc->fd);
603
604 TAILQ_REMOVE(&sc_queue, sc, entry);
605
606 event_del(&sc->ev);
607 close(sc->fd);
608
609 /* Some file descriptors are available again. */
610 if (evtimer_pending(&sc->tcp_ts->evt, NULL)) {
611 evtimer_del(&sc->tcp_ts->evt);
612 event_add(&sc->tcp_ts->ev, NULL);
613 }
614
615 free(sc);
616 mainstats.nconns--;
617 return;
618 }
619 if (ptb->vflag >= 3)
620 fprintf(stderr, "read: %zd bytes\n", n);
621 sc->bytes += n;
622 mainstats.slice_bytes += n;
623 mainstats.total_bytes += n;
624 }
625
626 static void
tcp_server_accept(int fd,short event,void * arg)627 tcp_server_accept(int fd, short event, void *arg)
628 {
629 struct tcpservsock *ts = arg;
630 int sock;
631 struct statctx *sc;
632 struct sockaddr_storage ss;
633 socklen_t sslen;
634 char tmp[NI_MAXHOST + 2 + NI_MAXSERV];
635
636 sslen = sizeof(ss);
637
638 event_add(&ts->ev, NULL);
639 if (event & EV_TIMEOUT)
640 return;
641 if ((sock = accept4(fd, (struct sockaddr *)&ss, &sslen, SOCK_NONBLOCK))
642 == -1) {
643 /*
644 * Pause accept if we are out of file descriptors, or
645 * libevent will haunt us here too.
646 */
647 if (errno == ENFILE || errno == EMFILE) {
648 struct timeval evtpause = { 1, 0 };
649
650 event_del(&ts->ev);
651 evtimer_add(&ts->evt, &evtpause);
652 } else if (errno != EWOULDBLOCK && errno != EINTR &&
653 errno != ECONNABORTED)
654 warn("accept");
655 return;
656 }
657 saddr_ntop((struct sockaddr *)&ss, sslen,
658 tmp, sizeof(tmp));
659 if (ptb->Tflag != -1 && ss.ss_family == AF_INET) {
660 if (setsockopt(sock, IPPROTO_IP, IP_TOS,
661 &ptb->Tflag, sizeof(ptb->Tflag)))
662 err(1, "setsockopt IP_TOS");
663 }
664 if (ptb->Tflag != -1 && ss.ss_family == AF_INET6) {
665 if (setsockopt(sock, IPPROTO_IPV6, IPV6_TCLASS,
666 &ptb->Tflag, sizeof(ptb->Tflag)))
667 err(1, "setsockopt IPV6_TCLASS");
668 }
669 /* Alloc client structure and register reading callback */
670 if ((sc = calloc(1, sizeof(*sc))) == NULL)
671 err(1, "calloc");
672 sc->tcp_ts = ts;
673 sc->fd = sock;
674 stats_prepare(sc);
675
676 event_set(&sc->ev, sc->fd, EV_READ | EV_PERSIST,
677 tcp_server_handle_sc, sc);
678 event_add(&sc->ev, NULL);
679 TAILQ_INSERT_TAIL(&sc_queue, sc, entry);
680
681 mainstats.nconns++;
682 if (mainstats.nconns == 1)
683 set_slice_timer(1);
684 if (ptb->vflag)
685 fprintf(stderr, "Accepted connection from %s, fd = %d\n",
686 tmp, sc->fd);
687 }
688
689 static void
server_init(struct addrinfo * aitop)690 server_init(struct addrinfo *aitop)
691 {
692 int sock, on = 1;
693 struct addrinfo *ai;
694 struct event *ev;
695 struct tcpservsock *ts;
696 nfds_t lnfds;
697
698 lnfds = 0;
699 for (ai = aitop; ai != NULL; ai = ai->ai_next) {
700 char tmp[NI_MAXHOST + 2 + NI_MAXSERV];
701
702 saddr_ntop(ai->ai_addr, ai->ai_addrlen, tmp, sizeof(tmp));
703 if (ptb->vflag)
704 fprintf(stderr, "Try to bind to %s\n", tmp);
705 if ((sock = socket(ai->ai_family, ai->ai_socktype,
706 ai->ai_protocol)) == -1) {
707 if (ai->ai_next == NULL)
708 err(1, "socket");
709 if (ptb->vflag)
710 warn("socket");
711 continue;
712 }
713 if (ptb->Dflag) {
714 if (setsockopt(sock, SOL_SOCKET, SO_DEBUG,
715 &ptb->Dflag, sizeof(ptb->Dflag)))
716 err(1, "setsockopt SO_DEBUG");
717 }
718 if (ptb->Tflag != -1 && ai->ai_family == AF_INET) {
719 if (setsockopt(sock, IPPROTO_IP, IP_TOS,
720 &ptb->Tflag, sizeof(ptb->Tflag)))
721 err(1, "setsockopt IP_TOS");
722 }
723 if (ptb->Tflag != -1 && ai->ai_family == AF_INET6) {
724 if (setsockopt(sock, IPPROTO_IPV6, IPV6_TCLASS,
725 &ptb->Tflag, sizeof(ptb->Tflag)))
726 err(1, "setsockopt IPV6_TCLASS");
727 }
728 if (setsockopt(sock, SOL_SOCKET, SO_REUSEADDR,
729 &on, sizeof(on)) == -1)
730 warn("reuse port");
731 if (bind(sock, ai->ai_addr, ai->ai_addrlen) != 0) {
732 if (ai->ai_next == NULL)
733 err(1, "bind");
734 if (ptb->vflag)
735 warn("bind");
736 close(sock);
737 continue;
738 }
739 if (ptb->Sflag) {
740 if (setsockopt(sock, SOL_SOCKET, SO_RCVBUF,
741 &ptb->Sflag, sizeof(ptb->Sflag)) == -1)
742 warn("set receive socket buffer size");
743 }
744 if (TCP_MODE) {
745 if (listen(sock, 64) == -1) {
746 if (ai->ai_next == NULL)
747 err(1, "listen");
748 if (ptb->vflag)
749 warn("listen");
750 close(sock);
751 continue;
752 }
753 }
754 if (UDP_MODE) {
755 if ((ev = calloc(1, sizeof(*ev))) == NULL)
756 err(1, "calloc");
757 event_set(ev, sock, EV_READ | EV_PERSIST,
758 udp_server_handle_sc, NULL);
759 event_add(ev, NULL);
760 } else {
761 if ((ts = calloc(1, sizeof(*ts))) == NULL)
762 err(1, "calloc");
763
764 ts->fd = sock;
765 evtimer_set(&ts->evt, tcp_server_accept, ts);
766 event_set(&ts->ev, ts->fd, EV_READ,
767 tcp_server_accept, ts);
768 event_add(&ts->ev, NULL);
769 }
770 if (ptb->vflag >= 3)
771 fprintf(stderr, "bound to fd %d\n", sock);
772 lnfds++;
773 }
774 if (!ptb->Uflag)
775 freeaddrinfo(aitop);
776 if (lnfds == 0)
777 errx(1, "No working listen addresses found");
778 }
779
780 static void
client_handle_sc(int fd,short event,void * v_sc)781 client_handle_sc(int fd, short event, void *v_sc)
782 {
783 struct statctx *sc = v_sc;
784 ssize_t n;
785 size_t blen = sc->buflen;
786
787 if (ptb->Rflag)
788 blen = arc4random_uniform(blen) + 1;
789 if ((n = write(sc->fd, sc->buf, blen)) == -1) {
790 if (errno == EINTR || errno == EWOULDBLOCK ||
791 (UDP_MODE && errno == ENOBUFS))
792 return;
793 warn("write");
794 wrapup(1);
795 }
796 if (TCP_MODE && n == 0) {
797 fprintf(stderr, "Remote end closed connection");
798 wrapup(1);
799 }
800 if (ptb->vflag >= 3)
801 fprintf(stderr, "write: %zd bytes\n", n);
802 sc->bytes += n;
803 mainstats.slice_bytes += n;
804 mainstats.total_bytes += n;
805 if (UDP_MODE)
806 sc->udp_slice_pkts++;
807 }
808
809 static void
client_init(struct addrinfo * aitop,int nconn,struct addrinfo * aib)810 client_init(struct addrinfo *aitop, int nconn, struct addrinfo *aib)
811 {
812 struct statctx *sc;
813 struct addrinfo *ai;
814 int i, r, sock;
815
816 for (i = 0; i < nconn; i++) {
817 for (sock = -1, ai = aitop; ai != NULL; ai = ai->ai_next) {
818 char tmp[NI_MAXHOST + 2 + NI_MAXSERV];
819
820 saddr_ntop(ai->ai_addr, ai->ai_addrlen, tmp,
821 sizeof(tmp));
822 if (ptb->vflag && i == 0)
823 fprintf(stderr, "Trying %s\n", tmp);
824 if ((sock = socket(ai->ai_family, ai->ai_socktype,
825 ai->ai_protocol)) == -1) {
826 if (ai->ai_next == NULL)
827 err(1, "socket");
828 if (ptb->vflag)
829 warn("socket");
830 continue;
831 }
832 if (ptb->Dflag) {
833 if (setsockopt(sock, SOL_SOCKET, SO_DEBUG,
834 &ptb->Dflag, sizeof(ptb->Dflag)))
835 err(1, "setsockopt SO_DEBUG");
836 }
837 if (aib != NULL) {
838 saddr_ntop(aib->ai_addr, aib->ai_addrlen,
839 tmp, sizeof(tmp));
840 if (ptb->vflag)
841 fprintf(stderr,
842 "Try to bind to %s\n", tmp);
843 if (bind(sock, (struct sockaddr *)aib->ai_addr,
844 aib->ai_addrlen) == -1)
845 err(1, "bind");
846 }
847 if (ptb->Tflag != -1 && ai->ai_family == AF_INET) {
848 if (setsockopt(sock, IPPROTO_IP, IP_TOS,
849 &ptb->Tflag, sizeof(ptb->Tflag)))
850 err(1, "setsockopt IP_TOS");
851 }
852 if (ptb->Tflag != -1 && ai->ai_family == AF_INET6) {
853 if (setsockopt(sock, IPPROTO_IPV6, IPV6_TCLASS,
854 &ptb->Tflag, sizeof(ptb->Tflag)))
855 err(1, "setsockopt IPV6_TCLASS");
856 }
857 if (ptb->Sflag) {
858 if (setsockopt(sock, SOL_SOCKET, SO_SNDBUF,
859 &ptb->Sflag, sizeof(ptb->Sflag)) == -1)
860 warn("set send socket buffer size");
861 }
862 if (connect(sock, ai->ai_addr, ai->ai_addrlen) != 0) {
863 if (ai->ai_next == NULL)
864 err(1, "connect");
865 if (ptb->vflag)
866 warn("connect");
867 close(sock);
868 sock = -1;
869 continue;
870 }
871 break;
872 }
873 if (sock == -1)
874 errx(1, "No host found");
875 if ((r = fcntl(sock, F_GETFL)) == -1)
876 err(1, "fcntl(F_GETFL)");
877 r |= O_NONBLOCK;
878 if (fcntl(sock, F_SETFL, r) == -1)
879 err(1, "fcntl(F_SETFL, O_NONBLOCK)");
880 /* Alloc and prepare stats */
881 if (TCP_MODE) {
882 if ((sc = calloc(1, sizeof(*sc))) == NULL)
883 err(1, "calloc");
884 } else
885 sc = udp_sc;
886
887 sc->fd = sock;
888 stats_prepare(sc);
889
890 event_set(&sc->ev, sc->fd, EV_WRITE | EV_PERSIST,
891 client_handle_sc, sc);
892 event_add(&sc->ev, NULL);
893 TAILQ_INSERT_TAIL(&sc_queue, sc, entry);
894
895 mainstats.nconns++;
896 if (mainstats.nconns == 1)
897 set_slice_timer(1);
898 }
899 if (!ptb->Uflag)
900 freeaddrinfo(aitop);
901 if (aib != NULL)
902 freeaddrinfo(aib);
903
904 if (ptb->vflag && nconn > 1)
905 fprintf(stderr, "%d connections established\n",
906 mainstats.nconns);
907 }
908
909 static int
map_tos(char * s,int * val)910 map_tos(char *s, int *val)
911 {
912 /* DiffServ Codepoints and other TOS mappings */
913 const struct toskeywords {
914 const char *keyword;
915 int val;
916 } *t, toskeywords[] = {
917 { "af11", IPTOS_DSCP_AF11 },
918 { "af12", IPTOS_DSCP_AF12 },
919 { "af13", IPTOS_DSCP_AF13 },
920 { "af21", IPTOS_DSCP_AF21 },
921 { "af22", IPTOS_DSCP_AF22 },
922 { "af23", IPTOS_DSCP_AF23 },
923 { "af31", IPTOS_DSCP_AF31 },
924 { "af32", IPTOS_DSCP_AF32 },
925 { "af33", IPTOS_DSCP_AF33 },
926 { "af41", IPTOS_DSCP_AF41 },
927 { "af42", IPTOS_DSCP_AF42 },
928 { "af43", IPTOS_DSCP_AF43 },
929 { "critical", IPTOS_PREC_CRITIC_ECP },
930 { "cs0", IPTOS_DSCP_CS0 },
931 { "cs1", IPTOS_DSCP_CS1 },
932 { "cs2", IPTOS_DSCP_CS2 },
933 { "cs3", IPTOS_DSCP_CS3 },
934 { "cs4", IPTOS_DSCP_CS4 },
935 { "cs5", IPTOS_DSCP_CS5 },
936 { "cs6", IPTOS_DSCP_CS6 },
937 { "cs7", IPTOS_DSCP_CS7 },
938 { "ef", IPTOS_DSCP_EF },
939 { "inetcontrol", IPTOS_PREC_INTERNETCONTROL },
940 { "lowdelay", IPTOS_LOWDELAY },
941 { "netcontrol", IPTOS_PREC_NETCONTROL },
942 { "reliability", IPTOS_RELIABILITY },
943 { "throughput", IPTOS_THROUGHPUT },
944 { NULL, -1 },
945 };
946
947 for (t = toskeywords; t->keyword != NULL; t++) {
948 if (strcmp(s, t->keyword) == 0) {
949 *val = t->val;
950 return (1);
951 }
952 }
953
954 return (0);
955 }
956
957 static void
quit(int sig,short event,void * arg)958 quit(int sig, short event, void *arg)
959 {
960 wrapup(0);
961 }
962
963 static void
wrapup(int err)964 wrapup(int err)
965 {
966 const int transfers = timerisset(&mainstats.t_first);
967 const int stats = (mainstats.floor_mbps != INFINITY);
968
969 if (transfers) {
970 if (!stats) {
971 if (UDP_MODE)
972 udp_process_slice(0, 0, NULL);
973 else
974 tcp_process_slice(0, 0, NULL);
975 }
976
977 summary_display();
978 }
979
980 if (err != -1)
981 exit(err);
982 }
983
984 int
main(int argc,char ** argv)985 main(int argc, char **argv)
986 {
987 struct timeval tv;
988 unsigned int secs, rtable;
989 char *tmp;
990 struct addrinfo *aitop, *aib, hints;
991 const char *errstr;
992 struct rlimit rl;
993 int ch, herr, nconn;
994 int family = PF_UNSPEC;
995 const char *host = NULL, *port = DEFAULT_PORT, *srcbind = NULL;
996 struct event ev_sigint, ev_sigterm, ev_sighup, ev_siginfo, ev_progtimer;
997 struct sockaddr_un sock_un;
998
999 /* Init world */
1000 setvbuf(stdout, NULL, _IOLBF, 0);
1001 ptb = &tcpbench;
1002 ptb->dummybuf_len = 0;
1003 ptb->Dflag = 0;
1004 ptb->Sflag = ptb->sflag = ptb->vflag = ptb->Rflag = ptb->Uflag = 0;
1005 ptb->kvars = NULL;
1006 ptb->rflag = DEFAULT_STATS_INTERVAL;
1007 ptb->Tflag = -1;
1008 nconn = 1;
1009 aib = NULL;
1010 secs = 0;
1011
1012 while ((ch = getopt(argc, argv, "46b:B:Dhlk:n:p:Rr:sS:t:T:uUvV:"))
1013 != -1) {
1014 switch (ch) {
1015 case '4':
1016 family = PF_INET;
1017 break;
1018 case '6':
1019 family = PF_INET6;
1020 break;
1021 case 'b':
1022 srcbind = optarg;
1023 break;
1024 case 'D':
1025 ptb->Dflag = 1;
1026 break;
1027 case 'l':
1028 list_kvars();
1029 exit(0);
1030 case 'k':
1031 if ((tmp = strdup(optarg)) == NULL)
1032 err(1, "strdup");
1033 ptb->kvars = check_prepare_kvars(tmp);
1034 free(tmp);
1035 break;
1036 case 'R':
1037 ptb->Rflag = 1;
1038 break;
1039 case 'r':
1040 ptb->rflag = strtonum(optarg, 0, 60 * 60 * 24 * 1000,
1041 &errstr);
1042 if (errstr != NULL)
1043 errx(1, "statistics interval is %s: %s",
1044 errstr, optarg);
1045 break;
1046 case 'p':
1047 port = optarg;
1048 break;
1049 case 's':
1050 ptb->sflag = 1;
1051 break;
1052 case 'S':
1053 ptb->Sflag = strtonum(optarg, 0, 1024*1024*1024,
1054 &errstr);
1055 if (errstr != NULL)
1056 errx(1, "socket buffer size is %s: %s",
1057 errstr, optarg);
1058 break;
1059 case 'B':
1060 ptb->dummybuf_len = strtonum(optarg, 0, 1024*1024*1024,
1061 &errstr);
1062 if (errstr != NULL)
1063 errx(1, "read/write buffer size is %s: %s",
1064 errstr, optarg);
1065 break;
1066 case 'v':
1067 ptb->vflag++;
1068 break;
1069 case 'V':
1070 rtable = (unsigned int)strtonum(optarg, 0,
1071 RT_TABLEID_MAX, &errstr);
1072 if (errstr)
1073 errx(1, "rtable value is %s: %s",
1074 errstr, optarg);
1075 if (setrtable(rtable) == -1)
1076 err(1, "setrtable");
1077 break;
1078 case 'n':
1079 nconn = strtonum(optarg, 0, 65535, &errstr);
1080 if (errstr != NULL)
1081 errx(1, "number of connections is %s: %s",
1082 errstr, optarg);
1083 break;
1084 case 'u':
1085 ptb->uflag = 1;
1086 break;
1087 case 'U':
1088 ptb->Uflag = 1;
1089 break;
1090 case 'T':
1091 if (map_tos(optarg, &ptb->Tflag))
1092 break;
1093 errstr = NULL;
1094 if (strlen(optarg) > 1 && optarg[0] == '0' &&
1095 optarg[1] == 'x')
1096 ptb->Tflag = (int)strtol(optarg, NULL, 16);
1097 else
1098 ptb->Tflag = (int)strtonum(optarg, 0, 255,
1099 &errstr);
1100 if (ptb->Tflag == -1 || ptb->Tflag > 255 || errstr)
1101 errx(1, "illegal tos value %s", optarg);
1102 break;
1103 case 't':
1104 secs = strtonum(optarg, 1, UINT_MAX, &errstr);
1105 if (errstr != NULL)
1106 errx(1, "secs is %s: %s",
1107 errstr, optarg);
1108 break;
1109 case 'h':
1110 default:
1111 usage();
1112 }
1113 }
1114
1115 if (pledge("stdio unveil rpath dns inet unix id", NULL) == -1)
1116 err(1, "pledge");
1117
1118 argv += optind;
1119 argc -= optind;
1120 if ((argc != (ptb->sflag && !ptb->Uflag ? 0 : 1)) ||
1121 (UDP_MODE && (ptb->kvars || nconn != 1)))
1122 usage();
1123
1124 if (!ptb->sflag || ptb->Uflag)
1125 mainstats.host = host = argv[0];
1126
1127 if (ptb->Uflag)
1128 if (unveil(host, "rwc") == -1)
1129 err(1, "unveil %s", host);
1130
1131 if (pledge("stdio id dns inet unix", NULL) == -1)
1132 err(1, "pledge");
1133
1134 /*
1135 * Rationale,
1136 * If TCP, use a big buffer with big reads/writes.
1137 * If UDP, use a big buffer in server and a buffer the size of a
1138 * ethernet packet.
1139 */
1140 if (!ptb->dummybuf_len) {
1141 if (ptb->sflag || TCP_MODE)
1142 ptb->dummybuf_len = DEFAULT_BUF;
1143 else
1144 ptb->dummybuf_len = DEFAULT_UDP_PKT;
1145 }
1146
1147 bzero(&hints, sizeof(hints));
1148 hints.ai_family = family;
1149 if (UDP_MODE) {
1150 hints.ai_socktype = SOCK_DGRAM;
1151 hints.ai_protocol = IPPROTO_UDP;
1152 } else {
1153 hints.ai_socktype = SOCK_STREAM;
1154 hints.ai_protocol = IPPROTO_TCP;
1155 }
1156 if (ptb->Uflag) {
1157 hints.ai_family = AF_UNIX;
1158 hints.ai_protocol = 0;
1159 sock_un.sun_family = AF_UNIX;
1160 if (strlcpy(sock_un.sun_path, host, sizeof(sock_un.sun_path)) >=
1161 sizeof(sock_un.sun_path))
1162 errx(1, "socket name '%s' too long", host);
1163 hints.ai_addr = (struct sockaddr *)&sock_un;
1164 hints.ai_addrlen = sizeof(sock_un);
1165 aitop = &hints;
1166 } else {
1167 if (ptb->sflag)
1168 hints.ai_flags = AI_PASSIVE;
1169 if (srcbind != NULL) {
1170 hints.ai_flags |= AI_NUMERICHOST;
1171 herr = getaddrinfo(srcbind, NULL, &hints, &aib);
1172 hints.ai_flags &= ~AI_NUMERICHOST;
1173 if (herr != 0) {
1174 if (herr == EAI_SYSTEM)
1175 err(1, "getaddrinfo");
1176 else
1177 errx(1, "getaddrinfo: %s",
1178 gai_strerror(herr));
1179 }
1180 }
1181 if ((herr = getaddrinfo(host, port, &hints, &aitop)) != 0) {
1182 if (herr == EAI_SYSTEM)
1183 err(1, "getaddrinfo");
1184 else
1185 errx(1, "getaddrinfo: %s", gai_strerror(herr));
1186 }
1187 }
1188
1189 if (pledge("stdio id inet unix", NULL) == -1)
1190 err(1, "pledge");
1191
1192 if (getrlimit(RLIMIT_NOFILE, &rl) == -1)
1193 err(1, "getrlimit");
1194 if (rl.rlim_cur < MAX_FD)
1195 rl.rlim_cur = MAX_FD;
1196 if (setrlimit(RLIMIT_NOFILE, &rl))
1197 err(1, "setrlimit");
1198 if (getrlimit(RLIMIT_NOFILE, &rl) == -1)
1199 err(1, "getrlimit");
1200
1201 if (pledge("stdio inet unix", NULL) == -1)
1202 err(1, "pledge");
1203
1204 /* Init world */
1205 TAILQ_INIT(&sc_queue);
1206 if ((ptb->dummybuf = malloc(ptb->dummybuf_len)) == NULL)
1207 err(1, "malloc");
1208 arc4random_buf(ptb->dummybuf, ptb->dummybuf_len);
1209
1210 timerclear(&mainstats.t_first);
1211 mainstats.floor_mbps = INFINITY;
1212
1213 /* Setup libevent and signals */
1214 event_init();
1215 signal_set(&ev_sigterm, SIGTERM, signal_handler, NULL);
1216 signal_set(&ev_sighup, SIGHUP, signal_handler, NULL);
1217 signal_set(&ev_sigint, SIGINT, signal_handler, NULL);
1218 signal_set(&ev_siginfo, SIGINFO, signal_handler, NULL);
1219 signal_add(&ev_sigint, NULL);
1220 signal_add(&ev_sigterm, NULL);
1221 signal_add(&ev_sighup, NULL);
1222 signal_add(&ev_siginfo, NULL);
1223 signal(SIGPIPE, SIG_IGN);
1224
1225 if (UDP_MODE) {
1226 if ((udp_sc = calloc(1, sizeof(*udp_sc))) == NULL)
1227 err(1, "calloc");
1228 udp_sc->fd = -1;
1229 evtimer_set(&mainstats.timer, udp_process_slice, NULL);
1230 } else {
1231 print_tcp_header();
1232 evtimer_set(&mainstats.timer, tcp_process_slice, NULL);
1233 }
1234
1235 if (ptb->sflag)
1236 server_init(aitop);
1237 else {
1238 if (secs > 0) {
1239 timerclear(&tv);
1240 tv.tv_sec = secs + 1;
1241 evtimer_set(&ev_progtimer, quit, NULL);
1242 evtimer_add(&ev_progtimer, &tv);
1243 }
1244 client_init(aitop, nconn, aib);
1245
1246 if (pledge("stdio inet", NULL) == -1)
1247 err(1, "pledge");
1248 }
1249
1250 /* libevent main loop*/
1251 event_dispatch();
1252
1253 return (0);
1254 }
1255