1 /*- 2 * Copyright (c) 2008-2009 Robert N. M. Watson 3 * Copyright (c) 2010 Juniper Networks, Inc. 4 * All rights reserved. 5 * 6 * This software was developed by Robert N. M. Watson under contract 7 * to Juniper Networks, Inc. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 * 30 * $FreeBSD$ 31 */ 32 33 #include <sys/types.h> 34 #include <sys/event.h> 35 #include <sys/resource.h> 36 #include <sys/sched.h> 37 #include <sys/socket.h> 38 #include <sys/sysctl.h> 39 #include <sys/time.h> 40 #include <sys/uio.h> 41 #include <sys/wait.h> 42 43 #include <netinet/in.h> 44 #include <netinet/tcp.h> 45 46 #include <err.h> 47 #include <errno.h> 48 #include <fcntl.h> 49 #include <inttypes.h> 50 #include <signal.h> 51 #include <stdio.h> 52 #include <stdlib.h> 53 #include <string.h> 54 #include <unistd.h> 55 56 #include "tcpp.h" 57 58 #define min(x, y) (x < y ? x : y) 59 60 61 /* 62 * Gist of each client worker: build up to mflag connections at a time, and 63 * pump data in to them somewhat fairly until tflag connections have been 64 * completed. 65 */ 66 #define CONNECTION_MAGIC 0x87a3f56e 67 struct connection { 68 uint32_t conn_magic; /* Just magic. */ 69 int conn_fd; 70 struct tcpp_header conn_header; /* Header buffer. */ 71 u_int conn_header_sent; /* Header bytes sent. */ 72 u_int64_t conn_data_sent; /* Data bytes sent.*/ 73 }; 74 75 static u_char buffer[256 * 1024]; /* Buffer to send. */ 76 static pid_t *pid_list; 77 static int kq; 78 static int started; /* Number started so far. */ 79 static int finished; /* Number finished so far. */ 80 static int counter; /* IP number offset. */ 81 static uint64_t payload_len; 82 83 static struct connection * 84 tcpp_client_newconn(void) 85 { 86 struct sockaddr_in sin; 87 struct connection *conn; 88 struct kevent kev; 89 int fd, i; 90 91 /* 92 * Spread load over available IPs, rotating through them as we go. No 93 * attempt to localize IPs to particular workers. 94 */ 95 sin = localipbase; 96 sin.sin_addr.s_addr = htonl(ntohl(localipbase.sin_addr.s_addr) + 97 (counter++ % Mflag)); 98 99 fd = socket(PF_INET, SOCK_STREAM, 0); 100 if (fd < 0) 101 err(-1, "socket"); 102 103 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) 104 err(-1, "fcntl"); 105 106 i = 1; 107 if (setsockopt(fd, SOL_SOCKET, SO_NOSIGPIPE, &i, sizeof(i)) < 0) 108 err(-1, "setsockopt"); 109 i = 1; 110 if (setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &i, sizeof(i)) < 0) 111 err(-1, "setsockopt"); 112 #if 0 113 i = 1; 114 if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &i, sizeof(i)) < 0) 115 err(-1, "setsockopt"); 116 #endif 117 118 if (lflag) { 119 if (bind(fd, (struct sockaddr *)&sin, sizeof(sin)) < 0) 120 err(-1, "bind"); 121 } 122 123 if (connect(fd, (struct sockaddr *)&remoteip, sizeof(remoteip)) < 0 && 124 errno != EINPROGRESS) 125 err(-1, "connect"); 126 127 conn = malloc(sizeof(*conn)); 128 if (conn == NULL) 129 return (NULL); 130 bzero(conn, sizeof(*conn)); 131 conn->conn_magic = CONNECTION_MAGIC; 132 conn->conn_fd = fd; 133 conn->conn_header.th_magic = TCPP_MAGIC; 134 conn->conn_header.th_len = payload_len; 135 tcpp_header_encode(&conn->conn_header); 136 137 EV_SET(&kev, fd, EVFILT_WRITE, EV_ADD, 0, 0, conn); 138 if (kevent(kq, &kev, 1, NULL, 0, NULL) < 0) 139 err(-1, "newconn kevent"); 140 141 started++; 142 return (conn); 143 } 144 145 static void 146 tcpp_client_closeconn(struct connection *conn) 147 { 148 149 close(conn->conn_fd); 150 bzero(conn, sizeof(*conn)); 151 free(conn); 152 finished++; 153 } 154 155 static void 156 tcpp_client_handleconn(struct kevent *kev) 157 { 158 struct connection *conn; 159 struct iovec iov[2]; 160 ssize_t len, header_left; 161 162 conn = kev->udata; 163 if (conn->conn_magic != CONNECTION_MAGIC) 164 errx(-1, "tcpp_client_handleconn: magic"); 165 166 if (conn->conn_header_sent < sizeof(conn->conn_header)) { 167 header_left = sizeof(conn->conn_header) - 168 conn->conn_header_sent; 169 iov[0].iov_base = ((u_char *)&conn->conn_header) + 170 conn->conn_header_sent; 171 iov[0].iov_len = header_left; 172 iov[1].iov_base = buffer; 173 iov[1].iov_len = min(sizeof(buffer), payload_len); 174 len = writev(conn->conn_fd, iov, 2); 175 if (len < 0) { 176 tcpp_client_closeconn(conn); 177 err(-1, "tcpp_client_handleconn: header write"); 178 } 179 if (len == 0) { 180 tcpp_client_closeconn(conn); 181 errx(-1, "tcpp_client_handleconn: header write " 182 "premature EOF"); 183 } 184 if (len > header_left) { 185 conn->conn_data_sent += (len - header_left); 186 conn->conn_header_sent += header_left; 187 } else 188 conn->conn_header_sent += len; 189 } else { 190 len = write(conn->conn_fd, buffer, min(sizeof(buffer), 191 payload_len - conn->conn_data_sent)); 192 if (len < 0) { 193 tcpp_client_closeconn(conn); 194 err(-1, "tcpp_client_handleconn: data write"); 195 } 196 if (len == 0) { 197 tcpp_client_closeconn(conn); 198 errx(-1, "tcpp_client_handleconn: data write: " 199 "premature EOF"); 200 } 201 conn->conn_data_sent += len; 202 } 203 if (conn->conn_data_sent >= payload_len) { 204 /* 205 * All is well. 206 */ 207 tcpp_client_closeconn(conn); 208 } 209 } 210 211 static void 212 tcpp_client_worker(int workernum) 213 { 214 struct kevent *kev_array; 215 int i, numevents, kev_bytes; 216 #if defined(CPU_SETSIZE) && 0 217 cpu_set_t mask; 218 int ncpus; 219 size_t len; 220 221 if (Pflag) { 222 len = sizeof(ncpus); 223 if (sysctlbyname(SYSCTLNAME_CPUS, &ncpus, &len, NULL, 0) < 0) 224 err(-1, "sysctlbyname: %s", SYSCTLNAME_CPUS); 225 if (len != sizeof(ncpus)) 226 errx(-1, "sysctlbyname: %s: len %jd", SYSCTLNAME_CPUS, 227 (intmax_t)len); 228 229 CPU_ZERO(&mask); 230 CPU_SET(workernum % ncpus, &mask); 231 if (sched_setaffinity(0, CPU_SETSIZE, &mask) < 0) 232 err(-1, "sched_setaffinity"); 233 } 234 #endif 235 setproctitle("tcpp_client %d", workernum); 236 237 /* 238 * Add the worker number to the remote port. 239 */ 240 remoteip.sin_port = htons(rflag + workernum); 241 242 kev_bytes = sizeof(*kev_array) * mflag; 243 kev_array = malloc(kev_bytes); 244 if (kev_array == NULL) 245 err(-1, "malloc"); 246 bzero(kev_array, kev_bytes); 247 248 kq = kqueue(); 249 if (kq < 0) 250 err(-1, "kqueue"); 251 252 while (finished < tflag) { 253 while ((started - finished < mflag) && (started < tflag)) 254 (void)tcpp_client_newconn(); 255 numevents = kevent(kq, NULL, 0, kev_array, mflag, NULL); 256 if (numevents < 0) 257 err(-1, "kevent"); 258 if (numevents > mflag) 259 errx(-1, "kevent: %d", numevents); 260 for (i = 0; i < numevents; i++) 261 tcpp_client_handleconn(&kev_array[i]); 262 } 263 /* printf("Worker %d done - %d finished\n", workernum, finished); */ 264 } 265 266 void 267 tcpp_client(void) 268 { 269 struct timespec ts_start, ts_finish; 270 long cp_time_start[CPUSTATES], cp_time_finish[CPUSTATES]; 271 long ticks; 272 size_t size; 273 pid_t pid; 274 int i, failed, status; 275 276 if (bflag < sizeof(struct tcpp_header)) 277 errx(-1, "Can't use -b less than %zu\n", 278 sizeof(struct tcpp_header)); 279 payload_len = bflag - sizeof(struct tcpp_header); 280 281 pid_list = malloc(sizeof(*pid_list) * pflag); 282 if (pid_list == NULL) 283 err(-1, "malloc pid_list"); 284 bzero(pid_list, sizeof(*pid_list) * pflag); 285 286 /* 287 * Start workers. 288 */ 289 size = sizeof(cp_time_start); 290 if (sysctlbyname(SYSCTLNAME_CPTIME, &cp_time_start, &size, NULL, 0) 291 < 0) 292 err(-1, "sysctlbyname: %s", SYSCTLNAME_CPTIME); 293 if (clock_gettime(CLOCK_REALTIME, &ts_start) < 0) 294 err(-1, "clock_gettime"); 295 for (i = 0; i < pflag; i++) { 296 pid = fork(); 297 if (pid < 0) { 298 warn("fork"); 299 for (i = 0; i < pflag; i++) { 300 if (pid_list[i] != 0) 301 (void)kill(pid_list[i], SIGKILL); 302 } 303 exit(-1); 304 } 305 if (pid == 0) { 306 tcpp_client_worker(i); 307 exit(0); 308 } 309 pid_list[i] = pid; 310 } 311 312 /* 313 * GC workers. 314 */ 315 failed = 0; 316 for (i = 0; i < pflag; i++) { 317 if (pid_list[i] != 0) { 318 while (waitpid(pid_list[i], &status, 0) != pid_list[i]); 319 if (WEXITSTATUS(status) != 0) 320 failed = 1; 321 } 322 } 323 if (clock_gettime(CLOCK_REALTIME, &ts_finish) < 0) 324 err(-1, "clock_gettime"); 325 size = sizeof(cp_time_finish); 326 if (sysctlbyname(SYSCTLNAME_CPTIME, &cp_time_finish, &size, NULL, 0) 327 < 0) 328 err(-1, "sysctlbyname: %s", SYSCTLNAME_CPTIME); 329 timespecsub(&ts_finish, &ts_start, &ts_finish); 330 331 if (failed) 332 errx(-1, "Too many errors"); 333 334 if (hflag) 335 printf("bytes,seconds,conn/s,Gb/s,user%%,nice%%,sys%%," 336 "intr%%,idle%%\n"); 337 338 /* 339 * Configuration parameters. 340 */ 341 printf("%jd,", bflag * tflag * pflag); 342 printf("%jd.%09jd,", (intmax_t)ts_finish.tv_sec, 343 (intmax_t)(ts_finish.tv_nsec)); 344 345 /* 346 * Effective transmit rates. 347 */ 348 printf("%f,", (double)(pflag * tflag)/ 349 (ts_finish.tv_sec + ts_finish.tv_nsec * 1e-9)); 350 printf("%f,", (double)(bflag * tflag * pflag * 8) / 351 (ts_finish.tv_sec + ts_finish.tv_nsec * 1e-9) * 1e-9); 352 353 /* 354 * CPU time (est). 355 */ 356 ticks = 0; 357 for (i = 0; i < CPUSTATES; i++) { 358 cp_time_finish[i] -= cp_time_start[i]; 359 ticks += cp_time_finish[i]; 360 } 361 printf("%0.02f,", (float)(100 * cp_time_finish[CP_USER]) / ticks); 362 printf("%0.02f,", (float)(100 * cp_time_finish[CP_NICE]) / ticks); 363 printf("%0.02f,", (float)(100 * cp_time_finish[CP_SYS]) / ticks); 364 printf("%0.02f,", (float)(100 * cp_time_finish[CP_INTR]) / ticks); 365 printf("%0.02f", (float)(100 * cp_time_finish[CP_IDLE]) / ticks); 366 printf("\n"); 367 } 368