1 /*- 2 * Copyright (c) 2008-2009 Robert N. M. Watson 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 * 26 * $FreeBSD$ 27 */ 28 29 #include <sys/types.h> 30 #include <sys/endian.h> 31 #include <sys/event.h> 32 #include <sys/resource.h> 33 #include <sys/sched.h> 34 #include <sys/socket.h> 35 #include <sys/sysctl.h> 36 #include <sys/time.h> 37 #include <sys/wait.h> 38 39 #include <netinet/in.h> 40 41 #include <err.h> 42 #include <fcntl.h> 43 #include <inttypes.h> 44 #include <signal.h> 45 #include <stdio.h> 46 #include <stdlib.h> 47 #include <string.h> 48 #include <unistd.h> 49 50 #include "tcpp.h" 51 52 /* 53 * Server side -- create a pool of processes, each listening on its own TCP 54 * port number for new connections. The first 8 bytes of each connection 55 * will be a network byte order length, then there will be that number of 56 * bytes of data. We use non-blocking sockets with kqueue to to avoid the 57 * overhead of threading or more than one process per processor, which makes 58 * things a bit awkward when dealing with data we care about. As such, we 59 * read into a small character buffer which we then convert to a length once 60 * we have all the data. 61 */ 62 #define CONNECTION_MAGIC 0x6392af27 63 struct connection { 64 uint32_t conn_magic; /* Just magic. */ 65 int conn_fd; 66 struct tcpp_header conn_header; /* Header buffer. */ 67 u_int conn_header_len; /* Bytes so far. */ 68 u_int64_t conn_data_len; /* How much to sink. */ 69 u_int64_t conn_data_received; /* How much so far. */ 70 }; 71 72 static pid_t *pid_list; 73 static int kq; 74 75 static struct connection * 76 tcpp_server_newconn(int listen_fd) 77 { 78 struct connection *conn; 79 struct kevent kev; 80 int fd; 81 82 fd = accept(listen_fd, NULL, NULL); 83 if (fd < 0) { 84 warn("accept"); 85 return (NULL); 86 } 87 88 if (fcntl(fd, F_SETFL, O_NONBLOCK) < 0) 89 err(-1, "fcntl"); 90 91 conn = malloc(sizeof(*conn)); 92 if (conn == NULL) 93 return (NULL); 94 bzero(conn, sizeof(*conn)); 95 conn->conn_magic = CONNECTION_MAGIC; 96 conn->conn_fd = fd; 97 98 /* 99 * Register to read on the socket, and set our conn pointer as the 100 * udata so we can find it quickly in the future. 101 */ 102 EV_SET(&kev, fd, EVFILT_READ, EV_ADD, 0, 0, conn); 103 if (kevent(kq, &kev, 1, NULL, 0, NULL) < 0) 104 err(-1, "kevent"); 105 106 return (conn); 107 } 108 109 static void 110 tcpp_server_closeconn(struct connection *conn) 111 { 112 113 /* 114 * Kqueue cleans up after itself once we close the socket, and since 115 * we are processing only one kevent at a time, we don't need to 116 * worry about watching out for future kevents referring to it. 117 * 118 * ... right? 119 */ 120 close(conn->conn_fd); 121 bzero(conn, sizeof(*conn)); 122 free(conn); 123 } 124 125 static u_char buffer[256*1024]; /* Buffer in which to sink data. */ 126 static void 127 tcpp_server_handleconn(struct kevent *kev) 128 { 129 struct connection *conn; 130 ssize_t len; 131 132 conn = kev->udata; 133 if (conn->conn_magic != CONNECTION_MAGIC) 134 errx(-1, "tcpp_server_handleconn: magic"); 135 136 if (conn->conn_header_len < sizeof(conn->conn_header)) { 137 len = read(conn->conn_fd, 138 ((u_char *)&conn->conn_header) + conn->conn_header_len, 139 sizeof(conn->conn_header) - conn->conn_header_len); 140 if (len < 0) { 141 warn("tcpp_server_handleconn: header read"); 142 tcpp_server_closeconn(conn); 143 return; 144 } 145 if (len == 0) { 146 warnx("tcpp_server_handleconn: header premature eof"); 147 tcpp_server_closeconn(conn); 148 return; 149 } 150 conn->conn_header_len += len; 151 if (conn->conn_header_len == sizeof(conn->conn_header)) { 152 tcpp_header_decode(&conn->conn_header); 153 if (conn->conn_header.th_magic != TCPP_MAGIC) { 154 warnx("tcpp_server_handleconn: bad magic"); 155 tcpp_server_closeconn(conn); 156 return; 157 } 158 } 159 } else { 160 /* 161 * Drain up to a buffer from the connection, so that we pay 162 * attention to other connections too. 163 */ 164 len = read(conn->conn_fd, buffer, sizeof(buffer)); 165 if (len < 0) { 166 warn("tcpp_server_handleconn: data bad read"); 167 tcpp_server_closeconn(conn); 168 return; 169 } 170 if (len == 0 && conn->conn_data_received < 171 conn->conn_header.th_len) { 172 warnx("tcpp_server_handleconn: data premature eof"); 173 tcpp_server_closeconn(conn); 174 return; 175 } 176 conn->conn_data_received += len; 177 if (conn->conn_data_received > conn->conn_header.th_len) { 178 warnx("tcpp_server_handleconn: too much data"); 179 tcpp_server_closeconn(conn); 180 return; 181 } 182 if (conn->conn_data_received == conn->conn_header.th_len) { 183 /* 184 * All is well. 185 */ 186 tcpp_server_closeconn(conn); 187 return; 188 } 189 } 190 } 191 192 static void 193 tcpp_server_worker(int workernum) 194 { 195 int i, listen_sock, numevents; 196 struct kevent kev, *kev_array; 197 int kev_bytes; 198 #if defined(CPU_SETSIZE) && 0 199 cpu_set_t mask; 200 int ncpus; 201 ssize_t len; 202 203 len = sizeof(ncpus); 204 if (sysctlbyname(SYSCTLNAME_CPUS, &ncpus, &len, NULL, 0) < 0) 205 err(-1, "sysctlbyname: %s", SYSCTLNAME_CPUS); 206 if (len != sizeof(ncpus)) 207 errx(-1, "sysctlbyname: %s: len %jd", SYSCTLNAME_CPUS, 208 (intmax_t)len); 209 210 CPU_ZERO(&mask); 211 CPU_SET(workernum % ncpus, &mask); 212 if (sched_setaffinity(0, CPU_SETSIZE, &mask) < 0) 213 err(-1, "sched_setaffinity"); 214 #endif 215 setproctitle("tcpp_server %d", workernum); 216 217 /* Allow an extra kevent for the listen socket. */ 218 kev_bytes = sizeof(*kev_array) * (mflag + 1); 219 kev_array = malloc(kev_bytes); 220 if (kev_array == NULL) 221 err(-1, "malloc"); 222 bzero(kev_array, kev_bytes); 223 224 /* XXXRW: Want to set and pin the CPU here. */ 225 226 /* 227 * Add the worker number to the local port. 228 */ 229 localipbase.sin_port = htons(rflag + workernum); 230 231 listen_sock = socket(PF_INET, SOCK_STREAM, 0); 232 if (listen_sock < 0) 233 err(-1, "socket"); 234 i = 1; 235 if (setsockopt(listen_sock, SOL_SOCKET, SO_NOSIGPIPE, &i, sizeof(i)) 236 < 0) 237 err(-1, "setsockopt"); 238 i = 1; 239 if (setsockopt(listen_sock, SOL_SOCKET, SO_REUSEPORT, &i, sizeof(i)) 240 < 0) 241 err(-1, "setsockopt"); 242 if (bind(listen_sock, (struct sockaddr *)&localipbase, 243 sizeof(localipbase)) < 0) 244 err(-1, "bind"); 245 if (listen(listen_sock, 16384)) 246 err(-1, "listen"); 247 if (fcntl(listen_sock, F_SETFL, O_NONBLOCK) < 0) 248 err(-1, "fcntl"); 249 250 kq = kqueue(); 251 if (kq < 0) 252 err(-1, "kqueue"); 253 254 EV_SET(&kev, listen_sock, EVFILT_READ, EV_ADD, 0, 0, NULL); 255 if (kevent(kq, &kev, 1, NULL, 0, NULL) < 0) 256 err(-1, "kevent"); 257 258 while ((numevents = kevent(kq, NULL, 0, kev_array, mflag + 1, NULL)) 259 > 0) { 260 for (i = 0; i < numevents; i++) { 261 if (kev_array[i].ident == (u_int)listen_sock) 262 (void)tcpp_server_newconn(listen_sock); 263 else 264 tcpp_server_handleconn(&kev_array[i]); 265 } 266 } 267 printf("Worker %d done\n", workernum); 268 } 269 270 void 271 tcpp_server(void) 272 { 273 long cp_time_last[CPUSTATES], cp_time_now[CPUSTATES], ticks; 274 size_t size; 275 pid_t pid; 276 int i; 277 278 pid_list = malloc(sizeof(*pid_list) * pflag); 279 if (pid_list == NULL) 280 err(-1, "malloc pid_list"); 281 bzero(pid_list, sizeof(*pid_list) * pflag); 282 283 /* 284 * Start workers. 285 */ 286 for (i = 0; i < pflag; i++) { 287 pid = fork(); 288 if (pid < 0) { 289 warn("fork"); 290 for (i = 0; i < pflag; i++) { 291 if (pid_list[i] != 0) 292 (void)kill(pid_list[i], SIGKILL); 293 } 294 exit(-1); 295 } 296 if (pid == 0) { 297 tcpp_server_worker(i); 298 exit(0); 299 } 300 pid_list[i] = pid; 301 } 302 303 if (Tflag) { 304 size = sizeof(cp_time_last); 305 if (sysctlbyname(SYSCTLNAME_CPTIME, &cp_time_last, &size, 306 NULL, 0) < 0) 307 err(-1, "sysctlbyname: %s", SYSCTLNAME_CPTIME); 308 while (1) { 309 sleep(10); 310 size = sizeof(cp_time_last); 311 if (sysctlbyname(SYSCTLNAME_CPTIME, &cp_time_now, 312 &size, NULL, 0) < 0) 313 err(-1, "sysctlbyname: %s", 314 SYSCTLNAME_CPTIME); 315 ticks = 0; 316 for (i = 0; i < CPUSTATES; i++) { 317 cp_time_last[i] = cp_time_now[i] - 318 cp_time_last[i]; 319 ticks += cp_time_last[i]; 320 } 321 printf("user%% %lu nice%% %lu sys%% %lu intr%% %lu " 322 "idle%% %lu\n", 323 (100 * cp_time_last[CP_USER]) / ticks, 324 (100 * cp_time_last[CP_NICE]) / ticks, 325 (100 * cp_time_last[CP_SYS]) / ticks, 326 (100 * cp_time_last[CP_INTR]) / ticks, 327 (100 * cp_time_last[CP_IDLE]) / ticks); 328 bcopy(cp_time_now, cp_time_last, sizeof(cp_time_last)); 329 } 330 } 331 332 /* 333 * GC workers. 334 */ 335 for (i = 0; i < pflag; i++) { 336 if (pid_list[i] != 0) { 337 while (waitpid(pid_list[i], NULL, 0) != pid_list[i]); 338 } 339 } 340 } 341