src/main.c

   1 #include <stdio.h>
   2 #include <stdlib.h>
   3 #include <unistd.h>
   4 #include <signal.h>
   5 #include <syslog.h>
   6 #include <pthread.h>
   7 #include <string.h>
   8 #include <errno.h>
   9 #include <sys/types.h>
  10 #include <sys/wait.h>
  11 #include <sys/time.h>
  12 #include <sys/resource.h>
  13
  14 #include "common.h"
  15 #include "log.h"
  16 #include "daemon.h"
  17 #include "web_server.h"
  18 #include "popen.h"
  19 #include "config.h"
  20 #include "web_client.h"
  21 #include "rrd.h"
  22 #include "rrd2json.h"
  23
  24 #include "unit_test.h"
  25
  26 #include "plugins_d.h"
  27 #include "plugin_idlejitter.h"
  28 #include "plugin_tc.h"
  29 #include "plugin_checks.h"
  30 #include "plugin_proc.h"
  31 #include "plugin_nfacct.h"
  32
  33 #include "main.h"
  34
  35 struct netdata_static_thread {
  36         char *name;
  37
  38         char *config_section;
  39         char *config_name;
  40
  41         int enabled;
  42
  43         pthread_t *thread;
  44
  45         void (*init_routine) (void);
  46         void *(*start_routine) (void *);
  47 };
  48
  49 struct netdata_static_thread static_threads[] = {
  50         {"tc",                  "plugins",      "tc",                   1, NULL, NULL,  tc_main},
  51         {"idlejitter",  "plugins",      "idlejitter",   1, NULL, NULL,  cpuidlejitter_main},
  52         {"proc",                "plugins",      "proc",                 1, NULL, NULL,  proc_main},
  53
  54 #ifdef INTERNAL_PLUGIN_NFACCT
  55         // nfacct requires root access
  56         // so, we build it as an external plugin with setuid to root
  57         {"nfacct",              "plugins",      "nfacct",               1, NULL, NULL,  nfacct_main},
  58 #endif
  59
  60         {"plugins.d",   NULL,           NULL,                   1, NULL, NULL,  pluginsd_main},
  61         {"check",               "plugins",      "checks",               0, NULL, NULL,  checks_main},
  62         {"web",                 NULL,           NULL,                   1, NULL, NULL,  socket_listen_main},
  63         {NULL,                  NULL,           NULL,                   0, NULL, NULL,  NULL}
  64 };
  65
  66 int killpid(pid_t pid, int sig)
  67 {
  68         int ret = -1;
  69         debug(D_EXIT, "Request to kill pid %d", pid);
  70
  71         errno = 0;
  72         if(kill(pid, 0) == -1) {
  73                 switch(errno) {
  74                         case ESRCH:
  75                                 error("Request to kill pid %d, but it is not running.", pid);
  76                                 break;
  77
  78                         case EPERM:
  79                                 error("Request to kill pid %d, but I do not have enough permissions.", pid);
  80                                 break;
  81
  82                         default:
  83                                 error("Request to kill pid %d, but I received an error.", pid);
  84                                 break;
  85                 }
  86         }
  87         else {
  88                 errno = 0;
  89
  90                 void (*old)(int);
  91                 old = signal(sig, SIG_IGN);
  92                 if(old == SIG_ERR) {
  93                         error("Cannot overwrite signal handler for signal %d", sig);
  94                         old = sig_handler;
  95                 }
  96
  97                 ret = kill(pid, sig);
  98
  99                 if(signal(sig, old) == SIG_ERR)
 100                         error("Cannot restore signal handler for signal %d", sig);
 101
 102                 if(ret == -1) {
 103                         switch(errno) {
 104                                 case ESRCH:
 105                                         error("Cannot kill pid %d, but it is not running.", pid);
 106                                         break;
 107
 108                                 case EPERM:
 109                                         error("Cannot kill pid %d, but I do not have enough permissions.", pid);
 110                                         break;
 111
 112                                 default:
 113                                         error("Cannot kill pid %d, but I received an error.", pid);
 114                                         break;
 115                         }
 116                 }
 117         }
 118
 119         return ret;
 120 }
 121
 122 void kill_childs()
 123 {
 124         siginfo_t info;
 125
 126         struct web_client *w;
 127         for(w = web_clients; w ; w = w->next) {
 128                 debug(D_EXIT, "Stopping web client %s", w->client_ip);
 129                 pthread_cancel(w->thread);
 130                 pthread_join(w->thread, NULL);
 131         }
 132
 133         int i;
 134         for (i = 0; static_threads[i].name != NULL ; i++) {
 135                 if(static_threads[i].thread) {
 136                         debug(D_EXIT, "Stopping %s thread", static_threads[i].name);
 137                         pthread_cancel(*static_threads[i].thread);
 138                         pthread_join(*static_threads[i].thread, NULL);
 139                         static_threads[i].thread = NULL;
 140                 }
 141         }
 142
 143         if(tc_child_pid) {
 144                 info("Killing tc-qos-helper procees");
 145                 if(killpid(tc_child_pid, SIGTERM) != -1)
 146                         waitid(tc_child_pid, 0, &info, WEXITED);
 147         }
 148         tc_child_pid = 0;
 149
 150         struct plugind *cd;
 151         for(cd = pluginsd_root ; cd ; cd = cd->next) {
 152                 debug(D_EXIT, "Stopping %s plugin thread", cd->id);
 153                 pthread_cancel(cd->thread);
 154                 pthread_join(cd->thread, NULL);
 155
 156                 if(cd->pid && !cd->obsolete) {
 157                         debug(D_EXIT, "killing %s plugin process", cd->id);
 158                         if(killpid(cd->pid, SIGTERM) != -1)
 159                                 waitid(cd->pid, 0, &info, WEXITED);
 160                 }
 161         }
 162
 163         debug(D_EXIT, "All threads/childs stopped.");
 164 }
 165
 166
 167 int main(int argc, char **argv)
 168 {
 169         int i;
 170         int config_loaded = 0;
 171         int dont_fork = 0;
 172
 173         // global initialization
 174         get_HZ();
 175
 176         // set the name for logging
 177         program_name = "netdata";
 178
 179         // parse  the arguments
 180         for(i = 1; i < argc ; i++) {
 181                 if(strcmp(argv[i], "-c") == 0 && (i+1) < argc) {
 182                         if(load_config(argv[i+1], 1) != 1) {
 183                                 fprintf(stderr, "Cannot load configuration file %s. Reason: %s\n", argv[i+1], strerror(errno));
 184                                 exit(1);
 185                         }
 186                         else {
 187                                 debug(D_OPTIONS, "Configuration loaded from %s.", argv[i+1]);
 188                                 config_loaded = 1;
 189                         }
 190                         i++;
 191                 }
 192                 else if(strcmp(argv[i], "-df") == 0 && (i+1) < argc) { config_set("global", "debug flags",  argv[i+1]); i++; }
 193                 else if(strcmp(argv[i], "-p")  == 0 && (i+1) < argc) { config_set("global", "port",         argv[i+1]); i++; }
 194                 else if(strcmp(argv[i], "-u")  == 0 && (i+1) < argc) { config_set("global", "run as user",  argv[i+1]); i++; }
 195                 else if(strcmp(argv[i], "-l")  == 0 && (i+1) < argc) { config_set("global", "history",      argv[i+1]); i++; }
 196                 else if(strcmp(argv[i], "-t")  == 0 && (i+1) < argc) { config_set("global", "update every", argv[i+1]); i++; }
 197                 else if(strcmp(argv[i], "-ch") == 0 && (i+1) < argc) { config_set("global", "host access prefix", argv[i+1]); i++; }
 198                 else if(strcmp(argv[i], "-nodeamon") == 0 || strcmp(argv[i], "-nd") == 0) dont_fork = 1;
 199                 else if(strcmp(argv[i], "--unittest")  == 0) {
 200                         if(unit_test_storage()) exit(1);
 201                         exit(0);
 202                         rrd_update_every = 1;
 203                         if(unit_test(1000000, 0)) exit(1);
 204                         if(unit_test(1000000, 500000)) exit(1);
 205                         if(unit_test(1000000, 100000)) exit(1);
 206                         if(unit_test(1000000, 900000)) exit(1);
 207                         if(unit_test(2000000, 0)) exit(1);
 208                         if(unit_test(2000000, 500000)) exit(1);
 209                         if(unit_test(2000000, 100000)) exit(1);
 210                         if(unit_test(2000000, 900000)) exit(1);
 211                         if(unit_test(500000, 500000)) exit(1);
 212                         //if(unit_test(500000, 100000)) exit(1); // FIXME: the expected numbers are wrong
 213                         //if(unit_test(500000, 900000)) exit(1); // FIXME: the expected numbers are wrong
 214                         if(unit_test(500000, 0)) exit(1);
 215                         fprintf(stderr, "\n\nALL TESTS PASSED\n\n");
 216                         exit(0);
 217                 }
 218                 else {
 219                         fprintf(stderr, "Cannot understand option '%s'.\n", argv[i]);
 220                         fprintf(stderr, "\nUSAGE: %s [-d] [-l LINES_TO_SAVE] [-u UPDATE_TIMER] [-p LISTEN_PORT] [-dl debug log file] [-df debug flags].\n\n", argv[0]);
 221                         fprintf(stderr, "  -c CONFIG FILE the configuration file to load. Default: %s.\n", CONFIG_DIR "/" CONFIG_FILENAME);
 222                         fprintf(stderr, "  -l LINES_TO_SAVE can be from 5 to %d lines in JSON data. Default: %d.\n", RRD_HISTORY_ENTRIES_MAX, RRD_DEFAULT_HISTORY_ENTRIES);
 223                         fprintf(stderr, "  -t UPDATE_TIMER can be from 1 to %d seconds. Default: %d.\n", UPDATE_EVERY_MAX, UPDATE_EVERY);
 224                         fprintf(stderr, "  -p LISTEN_PORT can be from 1 to %d. Default: %d.\n", 65535, LISTEN_PORT);
 225                         fprintf(stderr, "  -u USERNAME can be any system username to run as. Default: none.\n");
 226                         fprintf(stderr, "  -ch path to access host /proc and /sys when running in a container. Default: empty.\n");
 227                         fprintf(stderr, "  -nd or -nodeamon to disable forking in the background. Default: unset.\n");
 228                         fprintf(stderr, "  -df FLAGS debug options. Default: 0x%08llx.\n", debug_flags);
 229                         exit(1);
 230                 }
 231         }
 232
 233         if(!config_loaded) load_config(NULL, 0);
 234
 235         char *input_log_file = NULL;
 236         char *output_log_file = NULL;
 237         char *error_log_file = NULL;
 238         char *access_log_file = NULL;
 239         {
 240                 char buffer[1024];
 241
 242                 // --------------------------------------------------------------------
 243
 244                 sprintf(buffer, "0x%08llx", 0ULL);
 245                 char *flags = config_get("global", "debug flags", buffer);
 246                 debug_flags = strtoull(flags, NULL, 0);
 247                 debug(D_OPTIONS, "Debug flags set to '0x%8llx'.", debug_flags);
 248
 249                 if(debug_flags != 0) {
 250                         struct rlimit rl = { RLIM_INFINITY, RLIM_INFINITY };
 251                         if(setrlimit(RLIMIT_CORE, &rl) != 0)
 252                                 info("Cannot request unlimited core dumps for debugging... Proceeding anyway...");
 253                 }
 254
 255                 // --------------------------------------------------------------------
 256
 257                 global_host_prefix = config_get("global", "host access prefix", "");
 258
 259                 // --------------------------------------------------------------------
 260
 261                 output_log_file = config_get("global", "debug log", LOG_DIR "/debug.log");
 262                 if(strcmp(output_log_file, "syslog") == 0) {
 263                         output_log_syslog = 1;
 264                         output_log_file = NULL;
 265                 }
 266                 else if(strcmp(output_log_file, "none") == 0) {
 267                         output_log_syslog = 0;
 268                         output_log_file = NULL;
 269                 }
 270                 else output_log_syslog = 0;
 271
 272                 // --------------------------------------------------------------------
 273
 274                 silent = 0;
 275                 error_log_file = config_get("global", "error log", LOG_DIR "/error.log");
 276                 if(strcmp(error_log_file, "syslog") == 0) {
 277                         error_log_syslog = 1;
 278                         error_log_file = NULL;
 279                 }
 280                 else if(strcmp(error_log_file, "none") == 0) {
 281                         error_log_syslog = 0;
 282                         error_log_file = NULL;
 283                         silent = 1; // optimization - do not even generate debug log entries
 284                 }
 285                 else error_log_syslog = 0;
 286
 287                 // --------------------------------------------------------------------
 288
 289                 access_log_file = config_get("global", "access log", LOG_DIR "/access.log");
 290                 if(strcmp(access_log_file, "syslog") == 0) {
 291                         access_log_syslog = 1;
 292                         access_log_file = NULL;
 293                 }
 294                 else if(strcmp(access_log_file, "none") == 0) {
 295                         access_log_syslog = 0;
 296                         access_log_file = NULL;
 297                 }
 298                 else access_log_syslog = 0;
 299
 300                 // --------------------------------------------------------------------
 301
 302                 rrd_memory_mode = rrd_memory_mode_id(config_get("global", "memory mode", rrd_memory_mode_name(rrd_memory_mode)));
 303
 304                 // --------------------------------------------------------------------
 305
 306                 if(gethostname(buffer, HOSTNAME_MAX) == -1)
 307                         error("WARNING: Cannot get machine hostname.");
 308                 hostname = config_get("global", "hostname", buffer);
 309                 debug(D_OPTIONS, "hostname set to '%s'", hostname);
 310
 311                 // --------------------------------------------------------------------
 312
 313                 rrd_default_history_entries = config_get_number("global", "history", RRD_DEFAULT_HISTORY_ENTRIES);
 314                 if(rrd_default_history_entries < 5 || rrd_default_history_entries > RRD_HISTORY_ENTRIES_MAX) {
 315                         fprintf(stderr, "Invalid save lines %d given. Defaulting to %d.\n", rrd_default_history_entries, RRD_DEFAULT_HISTORY_ENTRIES);
 316                         rrd_default_history_entries = RRD_DEFAULT_HISTORY_ENTRIES;
 317                 }
 318                 else {
 319                         debug(D_OPTIONS, "save lines set to %d.", rrd_default_history_entries);
 320                 }
 321
 322                 // --------------------------------------------------------------------
 323
 324                 rrd_update_every = config_get_number("global", "update every", UPDATE_EVERY);
 325                 if(rrd_update_every < 1 || rrd_update_every > 600) {
 326                         fprintf(stderr, "Invalid update timer %d given. Defaulting to %d.\n", rrd_update_every, UPDATE_EVERY_MAX);
 327                         rrd_update_every = UPDATE_EVERY;
 328                 }
 329                 else debug(D_OPTIONS, "update timer set to %d.", rrd_update_every);
 330
 331                 // --------------------------------------------------------------------
 332
 333                 for (i = 0; static_threads[i].name != NULL ; i++) {
 334                         struct netdata_static_thread *st = &static_threads[i];
 335
 336                         if(st->config_name) st->enabled = config_get_boolean(st->config_section, st->config_name, st->enabled);
 337                         if(st->enabled && st->init_routine) st->init_routine();
 338                 }
 339
 340                 // --------------------------------------------------------------------
 341
 342                 prepare_rundir();
 343                 char *user = config_get("global", "run as user", (getuid() == 0)?"nobody":"");
 344                 if(*user) {
 345                         if(become_user(user) != 0) {
 346                                 fprintf(stderr, "Cannot become user %s.\n", user);
 347                                 exit(1);
 348                         }
 349                         else debug(D_OPTIONS, "Successfully became user %s.", user);
 350                 }
 351
 352                 // --------------------------------------------------------------------
 353
 354                 listen_backlog = config_get_number("global", "http port listen backlog", LISTEN_BACKLOG);
 355
 356                 listen_port = config_get_number("global", "port", LISTEN_PORT);
 357                 if(listen_port < 1 || listen_port > 65535) {
 358                         fprintf(stderr, "Invalid listen port %d given. Defaulting to %d.\n", listen_port, LISTEN_PORT);
 359                         listen_port = LISTEN_PORT;
 360                 }
 361                 else debug(D_OPTIONS, "Listen port set to %d.", listen_port);
 362
 363                 int ip = 0;
 364                 char *ipv = config_get("global", "ip version", "any");
 365                 if(!strcmp(ipv, "any") || !strcmp(ipv, "both") || !strcmp(ipv, "all")) ip = 0;
 366                 else if(!strcmp(ipv, "ipv4") || !strcmp(ipv, "IPV4") || !strcmp(ipv, "IPv4") || !strcmp(ipv, "4")) ip = 4;
 367                 else if(!strcmp(ipv, "ipv6") || !strcmp(ipv, "IPV6") || !strcmp(ipv, "IPv6") || !strcmp(ipv, "6")) ip = 6;
 368                 else fprintf(stderr, "Cannot understand ip version '%s'. Assumming 'any'.", ipv);
 369
 370                 if(ip == 0 || ip == 6) listen_fd = create_listen_socket6(listen_port, listen_backlog);
 371                 if(listen_fd < 0) {
 372                         listen_fd = create_listen_socket4(listen_port, listen_backlog);
 373                         if(listen_fd >= 0 && ip != 4) fprintf(stderr, "Managed to open an IPv4 socket on port %d.", listen_port);
 374                 }
 375
 376                 if(listen_fd < 0) fatal("Cannot listen socket.");
 377         }
 378
 379         // never become a problem
 380         if(nice(20) == -1) fprintf(stderr, "Cannot lower my CPU priority. Error: %s.\n", strerror(errno));
 381
 382 #ifndef NETDATA_NO_DAEMON
 383         if(become_daemon(dont_fork, 0, input_log_file, output_log_file, error_log_file, access_log_file, &access_fd, &stdaccess) == -1) {
 384                 fprintf(stderr, "Cannot demonize myself (%s).", strerror(errno));
 385                 exit(1);
 386         }
 387 #endif
 388
 389         if(output_log_syslog || error_log_syslog || access_log_syslog)
 390                 openlog("netdata", LOG_PID, LOG_DAEMON);
 391
 392         info("NetData started on pid %d", getpid());
 393
 394
 395         // catch all signals
 396         for (i = 1 ; i < 65 ;i++) {
 397                 switch(i) {
 398                         case SIGSEGV:
 399                         case SIGFPE:
 400                         case SIGCHLD:
 401                                 signal(i, SIG_DFL);
 402                                 break;
 403
 404                         default:
 405                                 signal(i,  sig_handler);
 406                                 break;
 407                 }
 408         }
 409
 410         for (i = 0; static_threads[i].name != NULL ; i++) {
 411                 struct netdata_static_thread *st = &static_threads[i];
 412
 413                 if(st->enabled) {
 414                         st->thread = malloc(sizeof(pthread_t));
 415
 416                         info("Starting thread %s.", st->name);
 417
 418                         if(pthread_create(st->thread, NULL, st->start_routine, NULL))
 419                                 error("failed to create new thread for %s.", st->name);
 420
 421                         else if(pthread_detach(*st->thread))
 422                                 error("Cannot request detach of newly created %s thread.", st->name);
 423                 }
 424                 else info("Not starting thread %s.", st->name);
 425         }
 426
 427         // for future use - the main thread
 428         while(1) sleep(60);
 429
 430         exit(0);
 431 }