src/main.c

   1 #ifdef HAVE_CONFIG_H
   2 #include <config.h>
   3 #endif
   4 #include <stdio.h>
   5 #include <stdlib.h>
   6 #include <unistd.h>
   7 #include <signal.h>
   8 #include <syslog.h>
   9 #include <pthread.h>
  10 #include <string.h>
  11 #include <errno.h>
  12 #include <sys/types.h>
  13 #include <sys/wait.h>
  14 #include <sys/time.h>
  15 #include <sys/resource.h>
  16 #include <sys/mman.h>
  17
  18 #include "common.h"
  19 #include "log.h"
  20 #include "daemon.h"
  21 #include "web_server.h"
  22 #include "popen.h"
  23 #include "appconfig.h"
  24 #include "web_client.h"
  25 #include "rrd.h"
  26 #include "rrd2json.h"
  27
  28 #include "unit_test.h"
  29
  30 #include "plugins_d.h"
  31 #include "plugin_idlejitter.h"
  32 #include "plugin_tc.h"
  33 #include "plugin_checks.h"
  34 #include "plugin_proc.h"
  35 #include "plugin_nfacct.h"
  36
  37 #include "main.h"
  38
  39 int netdata_exit = 0;
  40
  41 void netdata_cleanup_and_exit(int ret)
  42 {
  43         kill_childs();
  44         rrdset_free_all();
  45         //unlink("/var/run/netdata.pid");
  46         info("NetData exiting. Bye bye...");
  47         exit(ret);
  48 }
  49
  50 struct netdata_static_thread {
  51         char *name;
  52
  53         char *config_section;
  54         char *config_name;
  55
  56         int enabled;
  57
  58         pthread_t *thread;
  59
  60         void (*init_routine) (void);
  61         void *(*start_routine) (void *);
  62 };
  63
  64 struct netdata_static_thread static_threads[] = {
  65         {"tc",                  "plugins",      "tc",                   1, NULL, NULL,  tc_main},
  66         {"idlejitter",  "plugins",      "idlejitter",   1, NULL, NULL,  cpuidlejitter_main},
  67         {"proc",                "plugins",      "proc",                 1, NULL, NULL,  proc_main},
  68
  69 #ifdef INTERNAL_PLUGIN_NFACCT
  70         // nfacct requires root access
  71         // so, we build it as an external plugin with setuid to root
  72         {"nfacct",              "plugins",      "nfacct",               1, NULL, NULL,  nfacct_main},
  73 #endif
  74
  75         {"plugins.d",   NULL,           NULL,                   1, NULL, NULL,  pluginsd_main},
  76         {"check",               "plugins",      "checks",               0, NULL, NULL,  checks_main},
  77         {"web",                 NULL,           NULL,                   1, NULL, NULL,  socket_listen_main},
  78         {NULL,                  NULL,           NULL,                   0, NULL, NULL,  NULL}
  79 };
  80
  81 int killpid(pid_t pid, int sig)
  82 {
  83         int ret = -1;
  84         debug(D_EXIT, "Request to kill pid %d", pid);
  85
  86         errno = 0;
  87         if(kill(pid, 0) == -1) {
  88                 switch(errno) {
  89                         case ESRCH:
  90                                 error("Request to kill pid %d, but it is not running.", pid);
  91                                 break;
  92
  93                         case EPERM:
  94                                 error("Request to kill pid %d, but I do not have enough permissions.", pid);
  95                                 break;
  96
  97                         default:
  98                                 error("Request to kill pid %d, but I received an error.", pid);
  99                                 break;
 100                 }
 101         }
 102         else {
 103                 errno = 0;
 104
 105                 void (*old)(int);
 106                 old = signal(sig, SIG_IGN);
 107                 if(old == SIG_ERR) {
 108                         error("Cannot overwrite signal handler for signal %d", sig);
 109                         old = sig_handler;
 110                 }
 111
 112                 ret = kill(pid, sig);
 113
 114                 if(signal(sig, old) == SIG_ERR)
 115                         error("Cannot restore signal handler for signal %d", sig);
 116
 117                 if(ret == -1) {
 118                         switch(errno) {
 119                                 case ESRCH:
 120                                         error("Cannot kill pid %d, but it is not running.", pid);
 121                                         break;
 122
 123                                 case EPERM:
 124                                         error("Cannot kill pid %d, but I do not have enough permissions.", pid);
 125                                         break;
 126
 127                                 default:
 128                                         error("Cannot kill pid %d, but I received an error.", pid);
 129                                         break;
 130                         }
 131                 }
 132         }
 133
 134         return ret;
 135 }
 136
 137 void kill_childs()
 138 {
 139         siginfo_t info;
 140
 141         struct web_client *w;
 142         for(w = web_clients; w ; w = w->next) {
 143                 debug(D_EXIT, "Stopping web client %s", w->client_ip);
 144                 pthread_cancel(w->thread);
 145                 pthread_join(w->thread, NULL);
 146         }
 147
 148         int i;
 149         for (i = 0; static_threads[i].name != NULL ; i++) {
 150                 if(static_threads[i].thread) {
 151                         debug(D_EXIT, "Stopping %s thread", static_threads[i].name);
 152                         pthread_cancel(*static_threads[i].thread);
 153                         pthread_join(*static_threads[i].thread, NULL);
 154                         static_threads[i].thread = NULL;
 155                 }
 156         }
 157
 158         if(tc_child_pid) {
 159                 info("Killing tc-qos-helper procees");
 160                 if(killpid(tc_child_pid, SIGTERM) != -1)
 161                         waitid(P_PID, tc_child_pid, &info, WEXITED);
 162         }
 163         tc_child_pid = 0;
 164
 165         struct plugind *cd;
 166         for(cd = pluginsd_root ; cd ; cd = cd->next) {
 167                 debug(D_EXIT, "Stopping %s plugin thread", cd->id);
 168                 pthread_cancel(cd->thread);
 169                 pthread_join(cd->thread, NULL);
 170
 171                 if(cd->pid && !cd->obsolete) {
 172                         debug(D_EXIT, "killing %s plugin process", cd->id);
 173                         if(killpid(cd->pid, SIGTERM) != -1)
 174                                 waitid(P_PID, cd->pid, &info, WEXITED);
 175                 }
 176         }
 177
 178         // if, for any reason there is any child exited
 179         // catch it here
 180         waitid(P_PID, 0, &info, WEXITED|WNOHANG);
 181
 182         debug(D_EXIT, "All threads/childs stopped.");
 183 }
 184
 185
 186 int main(int argc, char **argv)
 187 {
 188         int i;
 189         int config_loaded = 0;
 190         int dont_fork = 0;
 191
 192         // global initialization
 193         get_HZ();
 194
 195         // set the name for logging
 196         program_name = "netdata";
 197
 198         // parse  the arguments
 199         for(i = 1; i < argc ; i++) {
 200                 if(strcmp(argv[i], "-c") == 0 && (i+1) < argc) {
 201                         if(load_config(argv[i+1], 1) != 1) {
 202                                 error("Cannot load configuration file %s.", argv[i+1]);
 203                                 exit(1);
 204                         }
 205                         else {
 206                                 debug(D_OPTIONS, "Configuration loaded from %s.", argv[i+1]);
 207                                 config_loaded = 1;
 208                         }
 209                         i++;
 210                 }
 211                 else if(strcmp(argv[i], "-df") == 0 && (i+1) < argc) { config_set("global", "debug flags",  argv[i+1]); i++; }
 212                 else if(strcmp(argv[i], "-p")  == 0 && (i+1) < argc) { config_set("global", "port",         argv[i+1]); i++; }
 213                 else if(strcmp(argv[i], "-u")  == 0 && (i+1) < argc) { config_set("global", "run as user",  argv[i+1]); i++; }
 214                 else if(strcmp(argv[i], "-l")  == 0 && (i+1) < argc) { config_set("global", "history",      argv[i+1]); i++; }
 215                 else if(strcmp(argv[i], "-t")  == 0 && (i+1) < argc) { config_set("global", "update every", argv[i+1]); i++; }
 216                 else if(strcmp(argv[i], "-ch") == 0 && (i+1) < argc) { config_set("global", "host access prefix", argv[i+1]); i++; }
 217                 else if(strcmp(argv[i], "-nodeamon") == 0 || strcmp(argv[i], "-nd") == 0) dont_fork = 1;
 218                 else if(strcmp(argv[i], "--unittest")  == 0) {
 219                         if(unit_test_storage()) exit(1);
 220                         exit(0);
 221                         rrd_update_every = 1;
 222                         if(unit_test(1000000, 0)) exit(1);
 223                         if(unit_test(1000000, 500000)) exit(1);
 224                         if(unit_test(1000000, 100000)) exit(1);
 225                         if(unit_test(1000000, 900000)) exit(1);
 226                         if(unit_test(2000000, 0)) exit(1);
 227                         if(unit_test(2000000, 500000)) exit(1);
 228                         if(unit_test(2000000, 100000)) exit(1);
 229                         if(unit_test(2000000, 900000)) exit(1);
 230                         if(unit_test(500000, 500000)) exit(1);
 231                         //if(unit_test(500000, 100000)) exit(1); // FIXME: the expected numbers are wrong
 232                         //if(unit_test(500000, 900000)) exit(1); // FIXME: the expected numbers are wrong
 233                         if(unit_test(500000, 0)) exit(1);
 234                         fprintf(stderr, "\n\nALL TESTS PASSED\n\n");
 235                         exit(0);
 236                 }
 237                 else {
 238                         fprintf(stderr, "Cannot understand option '%s'.\n", argv[i]);
 239                         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]);
 240                         fprintf(stderr, "  -c CONFIG FILE the configuration file to load. Default: %s.\n", CONFIG_DIR "/" CONFIG_FILENAME);
 241                         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);
 242                         fprintf(stderr, "  -t UPDATE_TIMER can be from 1 to %d seconds. Default: %d.\n", UPDATE_EVERY_MAX, UPDATE_EVERY);
 243                         fprintf(stderr, "  -p LISTEN_PORT can be from 1 to %d. Default: %d.\n", 65535, LISTEN_PORT);
 244                         fprintf(stderr, "  -u USERNAME can be any system username to run as. Default: none.\n");
 245                         fprintf(stderr, "  -ch path to access host /proc and /sys when running in a container. Default: empty.\n");
 246                         fprintf(stderr, "  -nd or -nodeamon to disable forking in the background. Default: unset.\n");
 247                         fprintf(stderr, "  -df FLAGS debug options. Default: 0x%08llx.\n", debug_flags);
 248                         exit(1);
 249                 }
 250         }
 251
 252         if(!config_loaded) load_config(NULL, 0);
 253
 254         // prepare configuration environment variables for the plugins
 255         setenv("NETDATA_CONFIG_DIR" , config_get("global", "config directory"   , CONFIG_DIR) , 1);
 256         setenv("NETDATA_PLUGINS_DIR", config_get("global", "plugins directory"  , PLUGINS_DIR), 1);
 257         setenv("NETDATA_WEB_DIR"    , config_get("global", "web files directory", WEB_DIR)    , 1);
 258         setenv("NETDATA_CACHE_DIR"  , config_get("global", "cache directory"    , CACHE_DIR)  , 1);
 259         setenv("NETDATA_LOG_DIR"    , config_get("global", "log directory"      , LOG_DIR)    , 1);
 260
 261         // avoid extended to stat(/etc/localtime)
 262         // http://stackoverflow.com/questions/4554271/how-to-avoid-excessive-stat-etc-localtime-calls-in-strftime-on-linux
 263         setenv("TZ", ":/etc/localtime", 0);
 264
 265         // cd to /tmp to avoid any plugins writing files at random places
 266         if(chdir("/tmp")) error("netdata: ERROR: Cannot cd to /tmp");
 267
 268         char *input_log_file = NULL;
 269         char *output_log_file = NULL;
 270         char *error_log_file = NULL;
 271         char *access_log_file = NULL;
 272         char *user = NULL;
 273         {
 274                 char buffer[1024];
 275
 276                 // --------------------------------------------------------------------
 277
 278                 sprintf(buffer, "0x%08llx", 0ULL);
 279                 char *flags = config_get("global", "debug flags", buffer);
 280                 setenv("NETDATA_DEBUG_FLAGS", flags, 1);
 281
 282                 debug_flags = strtoull(flags, NULL, 0);
 283                 debug(D_OPTIONS, "Debug flags set to '0x%8llx'.", debug_flags);
 284
 285                 if(debug_flags != 0) {
 286                         struct rlimit rl = { RLIM_INFINITY, RLIM_INFINITY };
 287                         if(setrlimit(RLIMIT_CORE, &rl) != 0)
 288                                 info("Cannot request unlimited core dumps for debugging... Proceeding anyway...");
 289                 }
 290
 291                 // --------------------------------------------------------------------
 292
 293 #ifdef MADV_MERGEABLE
 294                 enable_ksm = config_get_boolean("global", "memory deduplication (ksm)", enable_ksm);
 295 #else
 296 #warning "Kernel memory deduplication (KSM) is not available"
 297 #endif
 298
 299                 // --------------------------------------------------------------------
 300
 301
 302                 global_host_prefix = config_get("global", "host access prefix", "");
 303                 setenv("NETDATA_HOST_PREFIX", global_host_prefix, 1);
 304
 305                 // --------------------------------------------------------------------
 306
 307                 output_log_file = config_get("global", "debug log", LOG_DIR "/debug.log");
 308                 if(strcmp(output_log_file, "syslog") == 0) {
 309                         output_log_syslog = 1;
 310                         output_log_file = NULL;
 311                 }
 312                 else if(strcmp(output_log_file, "none") == 0) {
 313                         output_log_syslog = 0;
 314                         output_log_file = NULL;
 315                 }
 316                 else output_log_syslog = 0;
 317
 318                 // --------------------------------------------------------------------
 319
 320                 silent = 0;
 321                 error_log_file = config_get("global", "error log", LOG_DIR "/error.log");
 322                 if(strcmp(error_log_file, "syslog") == 0) {
 323                         error_log_syslog = 1;
 324                         error_log_file = NULL;
 325                 }
 326                 else if(strcmp(error_log_file, "none") == 0) {
 327                         error_log_syslog = 0;
 328                         error_log_file = NULL;
 329                         silent = 1; // optimization - do not even generate debug log entries
 330                 }
 331                 else error_log_syslog = 0;
 332
 333                 // --------------------------------------------------------------------
 334
 335                 access_log_file = config_get("global", "access log", LOG_DIR "/access.log");
 336                 if(strcmp(access_log_file, "syslog") == 0) {
 337                         access_log_syslog = 1;
 338                         access_log_file = NULL;
 339                 }
 340                 else if(strcmp(access_log_file, "none") == 0) {
 341                         access_log_syslog = 0;
 342                         access_log_file = NULL;
 343                 }
 344                 else access_log_syslog = 0;
 345
 346                 // --------------------------------------------------------------------
 347
 348                 rrd_memory_mode = rrd_memory_mode_id(config_get("global", "memory mode", rrd_memory_mode_name(rrd_memory_mode)));
 349
 350                 // --------------------------------------------------------------------
 351
 352                 if(gethostname(buffer, HOSTNAME_MAX) == -1)
 353                         error("WARNING: Cannot get machine hostname.");
 354                 hostname = config_get("global", "hostname", buffer);
 355                 debug(D_OPTIONS, "hostname set to '%s'", hostname);
 356
 357                 // --------------------------------------------------------------------
 358
 359                 rrd_default_history_entries = config_get_number("global", "history", RRD_DEFAULT_HISTORY_ENTRIES);
 360                 if(rrd_default_history_entries < 5 || rrd_default_history_entries > RRD_HISTORY_ENTRIES_MAX) {
 361                         fprintf(stderr, "Invalid save lines %d given. Defaulting to %d.\n", rrd_default_history_entries, RRD_DEFAULT_HISTORY_ENTRIES);
 362                         rrd_default_history_entries = RRD_DEFAULT_HISTORY_ENTRIES;
 363                 }
 364                 else {
 365                         debug(D_OPTIONS, "save lines set to %d.", rrd_default_history_entries);
 366                 }
 367
 368                 // --------------------------------------------------------------------
 369
 370                 rrd_update_every = config_get_number("global", "update every", UPDATE_EVERY);
 371                 if(rrd_update_every < 1 || rrd_update_every > 600) {
 372                         fprintf(stderr, "Invalid update timer %d given. Defaulting to %d.\n", rrd_update_every, UPDATE_EVERY_MAX);
 373                         rrd_update_every = UPDATE_EVERY;
 374                 }
 375                 else debug(D_OPTIONS, "update timer set to %d.", rrd_update_every);
 376
 377                 // let the plugins know the min update_every
 378                 {
 379                         char buffer[50];
 380                         snprintf(buffer, 50, "%d", rrd_update_every);
 381                         setenv("NETDATA_UPDATE_EVERY", buffer, 1);
 382                 }
 383
 384                 // --------------------------------------------------------------------
 385
 386                 for (i = 0; static_threads[i].name != NULL ; i++) {
 387                         struct netdata_static_thread *st = &static_threads[i];
 388
 389                         if(st->config_name) st->enabled = config_get_boolean(st->config_section, st->config_name, st->enabled);
 390                         if(st->enabled && st->init_routine) st->init_routine();
 391                 }
 392
 393                 // --------------------------------------------------------------------
 394
 395                 prepare_rundir();
 396                 user = config_get("global", "run as user"    , (getuid() == 0)?NETDATA_USER:"");
 397                 web_files_uid();
 398
 399                 // --------------------------------------------------------------------
 400
 401                 listen_backlog = config_get_number("global", "http port listen backlog", LISTEN_BACKLOG);
 402
 403                 listen_port = config_get_number("global", "port", LISTEN_PORT);
 404                 if(listen_port < 1 || listen_port > 65535) {
 405                         fprintf(stderr, "Invalid listen port %d given. Defaulting to %d.\n", listen_port, LISTEN_PORT);
 406                         listen_port = LISTEN_PORT;
 407                 }
 408                 else debug(D_OPTIONS, "Listen port set to %d.", listen_port);
 409
 410                 int ip = 0;
 411                 char *ipv = config_get("global", "ip version", "any");
 412                 if(!strcmp(ipv, "any") || !strcmp(ipv, "both") || !strcmp(ipv, "all")) ip = 0;
 413                 else if(!strcmp(ipv, "ipv4") || !strcmp(ipv, "IPV4") || !strcmp(ipv, "IPv4") || !strcmp(ipv, "4")) ip = 4;
 414                 else if(!strcmp(ipv, "ipv6") || !strcmp(ipv, "IPV6") || !strcmp(ipv, "IPv6") || !strcmp(ipv, "6")) ip = 6;
 415                 else fprintf(stderr, "Cannot understand ip version '%s'. Assumming 'any'.", ipv);
 416
 417                 if(ip == 0 || ip == 6) listen_fd = create_listen_socket6(listen_port, listen_backlog);
 418                 if(listen_fd < 0) {
 419                         listen_fd = create_listen_socket4(listen_port, listen_backlog);
 420                         if(listen_fd >= 0 && ip != 4) fprintf(stderr, "Managed to open an IPv4 socket on port %d.", listen_port);
 421                 }
 422
 423                 if(listen_fd < 0) fatal("Cannot listen socket.");
 424         }
 425
 426         // never become a problem
 427         if(nice(20) == -1) error("Cannot lower my CPU priority.");
 428
 429         if(become_daemon(dont_fork, 0, user, input_log_file, output_log_file, error_log_file, access_log_file, &access_fd, &stdaccess) == -1) {
 430                 fatal("Cannot demonize myself.");
 431                 exit(1);
 432         }
 433
 434         if(output_log_syslog || error_log_syslog || access_log_syslog)
 435                 openlog("netdata", LOG_PID, LOG_DAEMON);
 436
 437         info("NetData started on pid %d", getpid());
 438
 439
 440         // catch all signals
 441         for (i = 1 ; i < 65 ;i++) {
 442                 switch(i) {
 443                         case SIGKILL: // not catchable
 444                         case SIGSTOP: // not catchable
 445                                 break;
 446
 447                         case SIGSEGV:
 448                         case SIGFPE:
 449                         case SIGCHLD:
 450                                 signal(i, SIG_DFL);
 451                                 break;
 452
 453                         default:
 454                                 signal(i,  sig_handler);
 455                                 break;
 456                 }
 457         }
 458
 459         for (i = 0; static_threads[i].name != NULL ; i++) {
 460                 struct netdata_static_thread *st = &static_threads[i];
 461
 462                 if(st->enabled) {
 463                         st->thread = malloc(sizeof(pthread_t));
 464
 465                         info("Starting thread %s.", st->name);
 466
 467                         if(pthread_create(st->thread, NULL, st->start_routine, NULL))
 468                                 error("failed to create new thread for %s.", st->name);
 469
 470                         else if(pthread_detach(*st->thread))
 471                                 error("Cannot request detach of newly created %s thread.", st->name);
 472                 }
 473                 else info("Not starting thread %s.", st->name);
 474         }
 475
 476         // for future use - the main thread
 477         while(1) {
 478                 if(netdata_exit != 0) {
 479                         netdata_exit++;
 480
 481                         if(netdata_exit > 5) {
 482                                 netdata_cleanup_and_exit(0);
 483                                 exit(0);
 484                         }
 485                 }
 486                 sleep(2);
 487         }
 488
 489         exit(0);
 490 }