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 #include <sys/prctl.h>
  18
  19 #include "common.h"
  20 #include "log.h"
  21 #include "daemon.h"
  22 #include "web_server.h"
  23 #include "popen.h"
  24 #include "appconfig.h"
  25 #include "web_client.h"
  26 #include "rrd.h"
  27 #include "rrd2json.h"
  28
  29 #include "unit_test.h"
  30
  31 #include "plugins_d.h"
  32 #include "plugin_idlejitter.h"
  33 #include "plugin_tc.h"
  34 #include "plugin_checks.h"
  35 #include "plugin_proc.h"
  36 #include "plugin_nfacct.h"
  37
  38 #include "main.h"
  39 #include "../config.h"
  40
  41 int netdata_exit = 0;
  42
  43 void netdata_cleanup_and_exit(int ret)
  44 {
  45         netdata_exit = 1;
  46         rrdset_save_all();
  47         // kill_childs();
  48
  49         // let it log a few more error messages
  50         error_log_limit_reset();
  51
  52         if(pidfd != -1) {
  53                 if(ftruncate(pidfd, 0) != 0)
  54                         error("Cannot truncate pidfile '%s'.", pidfile);
  55
  56                 close(pidfd);
  57                 pidfd = -1;
  58         }
  59
  60         if(pidfile[0]) {
  61                 if(unlink(pidfile) != 0)
  62                         error("Cannot unlink pidfile '%s'.", pidfile);
  63         }
  64
  65         info("NetData exiting. Bye bye...");
  66         exit(ret);
  67 }
  68
  69 struct netdata_static_thread {
  70         char *name;
  71
  72         char *config_section;
  73         char *config_name;
  74
  75         int enabled;
  76
  77         pthread_t *thread;
  78
  79         void (*init_routine) (void);
  80         void *(*start_routine) (void *);
  81 };
  82
  83 struct netdata_static_thread static_threads[] = {
  84         {"tc",                  "plugins",      "tc",                   1, NULL, NULL,  tc_main},
  85         {"idlejitter",  "plugins",      "idlejitter",   1, NULL, NULL,  cpuidlejitter_main},
  86         {"proc",                "plugins",      "proc",                 1, NULL, NULL,  proc_main},
  87
  88 #ifdef INTERNAL_PLUGIN_NFACCT
  89         // nfacct requires root access
  90         // so, we build it as an external plugin with setuid to root
  91         {"nfacct",              "plugins",      "nfacct",               1, NULL, NULL,  nfacct_main},
  92 #endif
  93
  94         {"plugins.d",   NULL,           NULL,                   1, NULL, NULL,  pluginsd_main},
  95         {"check",               "plugins",      "checks",               0, NULL, NULL,  checks_main},
  96         {"web",                 NULL,           NULL,                   1, NULL, NULL,  socket_listen_main},
  97         {NULL,                  NULL,           NULL,                   0, NULL, NULL,  NULL}
  98 };
  99
 100 int killpid(pid_t pid, int sig)
 101 {
 102         int ret = -1;
 103         debug(D_EXIT, "Request to kill pid %d", pid);
 104
 105         errno = 0;
 106         if(kill(pid, 0) == -1) {
 107                 switch(errno) {
 108                         case ESRCH:
 109                                 error("Request to kill pid %d, but it is not running.", pid);
 110                                 break;
 111
 112                         case EPERM:
 113                                 error("Request to kill pid %d, but I do not have enough permissions.", pid);
 114                                 break;
 115
 116                         default:
 117                                 error("Request to kill pid %d, but I received an error.", pid);
 118                                 break;
 119                 }
 120         }
 121         else {
 122                 errno = 0;
 123
 124                 void (*old)(int);
 125                 old = signal(sig, SIG_IGN);
 126                 if(old == SIG_ERR) {
 127                         error("Cannot overwrite signal handler for signal %d", sig);
 128                         old = sig_handler;
 129                 }
 130
 131                 ret = kill(pid, sig);
 132
 133                 if(signal(sig, old) == SIG_ERR)
 134                         error("Cannot restore signal handler for signal %d", sig);
 135
 136                 if(ret == -1) {
 137                         switch(errno) {
 138                                 case ESRCH:
 139                                         error("Cannot kill pid %d, but it is not running.", pid);
 140                                         break;
 141
 142                                 case EPERM:
 143                                         error("Cannot kill pid %d, but I do not have enough permissions.", pid);
 144                                         break;
 145
 146                                 default:
 147                                         error("Cannot kill pid %d, but I received an error.", pid);
 148                                         break;
 149                         }
 150                 }
 151         }
 152
 153         return ret;
 154 }
 155
 156 void kill_childs()
 157 {
 158         siginfo_t info;
 159
 160         struct web_client *w;
 161         for(w = web_clients; w ; w = w->next) {
 162                 debug(D_EXIT, "Stopping web client %s", w->client_ip);
 163                 pthread_cancel(w->thread);
 164                 pthread_join(w->thread, NULL);
 165         }
 166
 167         int i;
 168         for (i = 0; static_threads[i].name != NULL ; i++) {
 169                 if(static_threads[i].thread) {
 170                         debug(D_EXIT, "Stopping %s thread", static_threads[i].name);
 171                         pthread_cancel(*static_threads[i].thread);
 172                         pthread_join(*static_threads[i].thread, NULL);
 173                         static_threads[i].thread = NULL;
 174                 }
 175         }
 176
 177         if(tc_child_pid) {
 178                 info("Killing tc-qos-helper procees");
 179                 if(killpid(tc_child_pid, SIGTERM) != -1)
 180                         waitid(P_PID, (id_t) tc_child_pid, &info, WEXITED);
 181         }
 182         tc_child_pid = 0;
 183
 184         struct plugind *cd;
 185         for(cd = pluginsd_root ; cd ; cd = cd->next) {
 186                 debug(D_EXIT, "Stopping %s plugin thread", cd->id);
 187                 pthread_cancel(cd->thread);
 188                 pthread_join(cd->thread, NULL);
 189
 190                 if(cd->pid && !cd->obsolete) {
 191                         debug(D_EXIT, "killing %s plugin process", cd->id);
 192                         if(killpid(cd->pid, SIGTERM) != -1)
 193                                 waitid(P_PID, (id_t) cd->pid, &info, WEXITED);
 194                 }
 195         }
 196
 197         // if, for any reason there is any child exited
 198         // catch it here
 199         waitid(P_PID, 0, &info, WEXITED|WNOHANG);
 200
 201         debug(D_EXIT, "All threads/childs stopped.");
 202 }
 203
 204
 205 int main(int argc, char **argv)
 206 {
 207         int i;
 208         int config_loaded = 0;
 209         int dont_fork = 0;
 210         size_t wanted_stacksize = 0, stacksize = 0;
 211         pthread_attr_t attr;
 212
 213         // global initialization
 214         get_HZ();
 215
 216         // set the name for logging
 217         program_name = "netdata";
 218
 219         // parse  the arguments
 220         for(i = 1; i < argc ; i++) {
 221                 if(strcmp(argv[i], "-c") == 0 && (i+1) < argc) {
 222                         if(load_config(argv[i+1], 1) != 1) {
 223                                 error("Cannot load configuration file %s.", argv[i+1]);
 224                                 exit(1);
 225                         }
 226                         else {
 227                                 debug(D_OPTIONS, "Configuration loaded from %s.", argv[i+1]);
 228                                 config_loaded = 1;
 229                         }
 230                         i++;
 231                 }
 232                 else if(strcmp(argv[i], "-df") == 0 && (i+1) < argc) { config_set("global", "debug flags",  argv[i+1]); debug_flags = strtoull(argv[i+1], NULL, 0); i++; }
 233                 else if(strcmp(argv[i], "-p")  == 0 && (i+1) < argc) { config_set("global", "port",         argv[i+1]); i++; }
 234                 else if(strcmp(argv[i], "-u")  == 0 && (i+1) < argc) { config_set("global", "run as user",  argv[i+1]); i++; }
 235                 else if(strcmp(argv[i], "-l")  == 0 && (i+1) < argc) { config_set("global", "history",      argv[i+1]); i++; }
 236                 else if(strcmp(argv[i], "-t")  == 0 && (i+1) < argc) { config_set("global", "update every", argv[i+1]); i++; }
 237                 else if(strcmp(argv[i], "-ch") == 0 && (i+1) < argc) { config_set("global", "host access prefix", argv[i+1]); i++; }
 238                 else if(strcmp(argv[i], "-stacksize") == 0 && (i+1) < argc) { config_set("global", "pthread stack size", argv[i+1]); i++; }
 239                 else if(strcmp(argv[i], "-nodaemon") == 0 || strcmp(argv[i], "-nd") == 0) dont_fork = 1;
 240                 else if(strcmp(argv[i], "-pidfile") == 0 && (i+1) < argc) {
 241                         i++;
 242                         strncpy(pidfile, argv[i], FILENAME_MAX);
 243                         pidfile[FILENAME_MAX] = '\0';
 244                 }
 245                 else if(strcmp(argv[i], "--unittest")  == 0) {
 246                         rrd_update_every = 1;
 247                         if(run_all_mockup_tests()) exit(1);
 248                         if(unit_test_storage()) exit(1);
 249                         fprintf(stderr, "\n\nALL TESTS PASSED\n\n");
 250                         exit(0);
 251                 }
 252                 else {
 253                         fprintf(stderr, "Cannot understand option '%s'.\n", argv[i]);
 254                         fprintf(stderr, "\nUSAGE: %s [-d] [-l LINES_TO_SAVE] [-u UPDATE_TIMER] [-p LISTEN_PORT] [-df debug flags].\n\n", argv[0]);
 255                         fprintf(stderr, "  -c CONFIG FILE the configuration file to load. Default: %s.\n", CONFIG_DIR "/" CONFIG_FILENAME);
 256                         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);
 257                         fprintf(stderr, "  -t UPDATE_TIMER can be from 1 to %d seconds. Default: %d.\n", UPDATE_EVERY_MAX, UPDATE_EVERY);
 258                         fprintf(stderr, "  -p LISTEN_PORT can be from 1 to %d. Default: %d.\n", 65535, LISTEN_PORT);
 259                         fprintf(stderr, "  -u USERNAME can be any system username to run as. Default: none.\n");
 260                         fprintf(stderr, "  -ch path to access host /proc and /sys when running in a container. Default: empty.\n");
 261                         fprintf(stderr, "  -nd or -nodeamon to disable forking in the background. Default: unset.\n");
 262                         fprintf(stderr, "  -df FLAGS debug options. Default: 0x%08llx.\n", debug_flags);
 263                         fprintf(stderr, "  -stacksize BYTES to overwrite the pthread stack size.\n");
 264                         fprintf(stderr, "  -pidfile FILENAME to save a pid while running.\n");
 265                         exit(1);
 266                 }
 267         }
 268
 269         if(!config_loaded) load_config(NULL, 0);
 270
 271         // prepare configuration environment variables for the plugins
 272         setenv("NETDATA_CONFIG_DIR" , config_get("global", "config directory"   , CONFIG_DIR) , 1);
 273         setenv("NETDATA_PLUGINS_DIR", config_get("global", "plugins directory"  , PLUGINS_DIR), 1);
 274         setenv("NETDATA_WEB_DIR"    , config_get("global", "web files directory", WEB_DIR)    , 1);
 275         setenv("NETDATA_CACHE_DIR"  , config_get("global", "cache directory"    , CACHE_DIR)  , 1);
 276         setenv("NETDATA_LOG_DIR"    , config_get("global", "log directory"      , LOG_DIR)    , 1);
 277         setenv("NETDATA_HOST_PREFIX", config_get("global", "host access prefix" , "")         , 1);
 278
 279         // avoid extended to stat(/etc/localtime)
 280         // http://stackoverflow.com/questions/4554271/how-to-avoid-excessive-stat-etc-localtime-calls-in-strftime-on-linux
 281         setenv("TZ", ":/etc/localtime", 0);
 282
 283         // cd to /tmp to avoid any plugins writing files at random places
 284         if(chdir("/tmp")) error("netdata: ERROR: Cannot cd to /tmp");
 285
 286         char *input_log_file = NULL;
 287         char *output_log_file = NULL;
 288         char *error_log_file = NULL;
 289         char *access_log_file = NULL;
 290         char *user = NULL;
 291         {
 292                 char buffer[1024];
 293
 294                 // --------------------------------------------------------------------
 295
 296                 sprintf(buffer, "0x%08llx", 0ULL);
 297                 char *flags = config_get("global", "debug flags", buffer);
 298                 setenv("NETDATA_DEBUG_FLAGS", flags, 1);
 299
 300                 debug_flags = strtoull(flags, NULL, 0);
 301                 debug(D_OPTIONS, "Debug flags set to '0x%8llx'.", debug_flags);
 302
 303                 if(debug_flags != 0) {
 304                         struct rlimit rl = { RLIM_INFINITY, RLIM_INFINITY };
 305                         if(setrlimit(RLIMIT_CORE, &rl) != 0)
 306                                 info("Cannot request unlimited core dumps for debugging... Proceeding anyway...");
 307                         prctl(PR_SET_DUMPABLE, 1, 0, 0, 0);
 308                 }
 309
 310                 // --------------------------------------------------------------------
 311
 312 #ifdef MADV_MERGEABLE
 313                 enable_ksm = config_get_boolean("global", "memory deduplication (ksm)", enable_ksm);
 314 #else
 315 #warning "Kernel memory deduplication (KSM) is not available"
 316 #endif
 317
 318                 // --------------------------------------------------------------------
 319
 320
 321                 global_host_prefix = config_get("global", "host access prefix", "");
 322                 setenv("NETDATA_HOST_PREFIX", global_host_prefix, 1);
 323
 324                 // --------------------------------------------------------------------
 325
 326                 output_log_file = config_get("global", "debug log", LOG_DIR "/debug.log");
 327                 if(strcmp(output_log_file, "syslog") == 0) {
 328                         output_log_syslog = 1;
 329                         output_log_file = NULL;
 330                 }
 331                 else if(strcmp(output_log_file, "none") == 0) {
 332                         output_log_syslog = 0;
 333                         output_log_file = NULL;
 334                 }
 335                 else output_log_syslog = 0;
 336
 337                 // --------------------------------------------------------------------
 338
 339                 error_log_file = config_get("global", "error log", LOG_DIR "/error.log");
 340                 if(strcmp(error_log_file, "syslog") == 0) {
 341                         error_log_syslog = 1;
 342                         error_log_file = NULL;
 343                 }
 344                 else if(strcmp(error_log_file, "none") == 0) {
 345                         error_log_syslog = 0;
 346                         error_log_file = NULL;
 347                         // optimization - do not even generate debug log entries
 348                 }
 349                 else error_log_syslog = 0;
 350
 351                 error_log_throttle_period = config_get_number("global", "errors flood protection period", error_log_throttle_period);
 352                 setenv("NETDATA_ERRORS_THROTTLE_PERIOD", config_get("global", "errors flood protection period"    , ""), 1);
 353
 354                 error_log_errors_per_period = config_get_number("global", "errors to trigger flood protection", error_log_errors_per_period);
 355                 setenv("NETDATA_ERRORS_PER_PERIOD"     , config_get("global", "errors to trigger flood protection", ""), 1);
 356
 357                 // --------------------------------------------------------------------
 358
 359                 access_log_file = config_get("global", "access log", LOG_DIR "/access.log");
 360                 if(strcmp(access_log_file, "syslog") == 0) {
 361                         access_log_syslog = 1;
 362                         access_log_file = NULL;
 363                 }
 364                 else if(strcmp(access_log_file, "none") == 0) {
 365                         access_log_syslog = 0;
 366                         access_log_file = NULL;
 367                 }
 368                 else access_log_syslog = 0;
 369
 370                 // --------------------------------------------------------------------
 371
 372                 rrd_memory_mode = rrd_memory_mode_id(config_get("global", "memory mode", rrd_memory_mode_name(rrd_memory_mode)));
 373
 374                 // --------------------------------------------------------------------
 375
 376                 if(gethostname(buffer, HOSTNAME_MAX) == -1)
 377                         error("WARNING: Cannot get machine hostname.");
 378                 hostname = config_get("global", "hostname", buffer);
 379                 debug(D_OPTIONS, "hostname set to '%s'", hostname);
 380
 381                 // --------------------------------------------------------------------
 382
 383                 rrd_default_history_entries = (int) config_get_number("global", "history", RRD_DEFAULT_HISTORY_ENTRIES);
 384                 if(rrd_default_history_entries < 5 || rrd_default_history_entries > RRD_HISTORY_ENTRIES_MAX) {
 385                         info("Invalid save lines %d given. Defaulting to %d.", rrd_default_history_entries, RRD_DEFAULT_HISTORY_ENTRIES);
 386                         rrd_default_history_entries = RRD_DEFAULT_HISTORY_ENTRIES;
 387                 }
 388                 else {
 389                         debug(D_OPTIONS, "save lines set to %d.", rrd_default_history_entries);
 390                 }
 391
 392                 // --------------------------------------------------------------------
 393
 394                 rrd_update_every = (int) config_get_number("global", "update every", UPDATE_EVERY);
 395                 if(rrd_update_every < 1 || rrd_update_every > 600) {
 396                         info("Invalid update timer %d given. Defaulting to %d.", rrd_update_every, UPDATE_EVERY_MAX);
 397                         rrd_update_every = UPDATE_EVERY;
 398                 }
 399                 else debug(D_OPTIONS, "update timer set to %d.", rrd_update_every);
 400
 401                 // let the plugins know the min update_every
 402                 {
 403                         char buf[50];
 404                         snprintf(buf, 50, "%d", rrd_update_every);
 405                         setenv("NETDATA_UPDATE_EVERY", buf, 1);
 406                 }
 407
 408                 // --------------------------------------------------------------------
 409
 410                 i = pthread_attr_init(&attr);
 411                 if(i != 0)
 412                         fatal("pthread_attr_init() failed with code %d.", i);
 413
 414                 i = pthread_attr_getstacksize(&attr, &stacksize);
 415                 if(i != 0)
 416                         fatal("pthread_attr_getstacksize() failed with code %d.", i);
 417                 else
 418                         debug(D_OPTIONS, "initial pthread stack size is %zu bytes", stacksize);
 419
 420                 wanted_stacksize = config_get_number("global", "pthread stack size", stacksize);
 421
 422                 // --------------------------------------------------------------------
 423
 424                 for (i = 0; static_threads[i].name != NULL ; i++) {
 425                         struct netdata_static_thread *st = &static_threads[i];
 426
 427                         if(st->config_name) st->enabled = config_get_boolean(st->config_section, st->config_name, st->enabled);
 428                         if(st->enabled && st->init_routine) st->init_routine();
 429                 }
 430
 431                 // --------------------------------------------------------------------
 432
 433                 // get the user we should run
 434                 // IMPORTANT: this is required before web_files_uid()
 435                 user = config_get("global", "run as user"    , (getuid() == 0)?NETDATA_USER:"");
 436
 437                 // IMPORTANT: these have to run once, while single threaded
 438                 web_files_uid(); // IMPORTANT: web_files_uid() before web_files_gid()
 439                 web_files_gid();
 440
 441                 // --------------------------------------------------------------------
 442
 443                 listen_backlog = (int) config_get_number("global", "http port listen backlog", LISTEN_BACKLOG);
 444
 445                 listen_port = (int) config_get_number("global", "port", LISTEN_PORT);
 446                 if(listen_port < 1 || listen_port > 65535) {
 447                         info("Invalid listen port %d given. Defaulting to %d.", listen_port, LISTEN_PORT);
 448                         listen_port = LISTEN_PORT;
 449                 }
 450                 else debug(D_OPTIONS, "Listen port set to %d.", listen_port);
 451
 452                 int ip = 0;
 453                 char *ipv = config_get("global", "ip version", "any");
 454                 if(!strcmp(ipv, "any") || !strcmp(ipv, "both") || !strcmp(ipv, "all")) ip = 0;
 455                 else if(!strcmp(ipv, "ipv4") || !strcmp(ipv, "IPV4") || !strcmp(ipv, "IPv4") || !strcmp(ipv, "4")) ip = 4;
 456                 else if(!strcmp(ipv, "ipv6") || !strcmp(ipv, "IPV6") || !strcmp(ipv, "IPv6") || !strcmp(ipv, "6")) ip = 6;
 457                 else info("Cannot understand ip version '%s'. Assuming 'any'.", ipv);
 458
 459                 if(ip == 0 || ip == 6) listen_fd = create_listen_socket6(config_get("global", "bind socket to IP", "*"), listen_port, listen_backlog);
 460                 if(listen_fd < 0) {
 461                         listen_fd = create_listen_socket4(config_get("global", "bind socket to IP", "*"), listen_port, listen_backlog);
 462                         if(listen_fd >= 0 && ip != 4) info("Managed to open an IPv4 socket on port %d.", listen_port);
 463                 }
 464
 465                 if(listen_fd < 0) fatal("Cannot listen socket.");
 466         }
 467
 468         // never become a problem
 469         if(nice(20) == -1) error("Cannot lower my CPU priority.");
 470
 471         if(become_daemon(dont_fork, 0, user, input_log_file, output_log_file, error_log_file, access_log_file, &access_fd, &stdaccess) == -1) {
 472                 fatal("Cannot demonize myself.");
 473                 exit(1);
 474         }
 475
 476         if(debug_flags != 0) {
 477                 struct rlimit rl = { RLIM_INFINITY, RLIM_INFINITY };
 478                 if(setrlimit(RLIMIT_CORE, &rl) != 0)
 479                         info("Cannot request unlimited core dumps for debugging... Proceeding anyway...");
 480
 481                 prctl(PR_SET_DUMPABLE, 1, 0, 0, 0);
 482         }
 483
 484         if(output_log_syslog || error_log_syslog || access_log_syslog)
 485                 openlog("netdata", LOG_PID, LOG_DAEMON);
 486
 487         info("NetData started on pid %d", getpid());
 488
 489
 490         // catch all signals
 491         for (i = 1 ; i < 65 ;i++) {
 492                 switch(i) {
 493                         case SIGKILL: // not catchable
 494                         case SIGSTOP: // not catchable
 495                                 break;
 496
 497                         case SIGSEGV:
 498                         case SIGFPE:
 499                         case SIGCHLD:
 500                                 signal(i, SIG_DFL);
 501                                 break;
 502
 503                         default:
 504                                 signal(i,  sig_handler);
 505                                 break;
 506                 }
 507         }
 508
 509         // ------------------------------------------------------------------------
 510         // get default pthread stack size
 511
 512         if(stacksize < wanted_stacksize) {
 513                 i = pthread_attr_setstacksize(&attr, wanted_stacksize);
 514                 if(i != 0)
 515                         fatal("pthread_attr_setstacksize() to %zu bytes, failed with code %d.", wanted_stacksize, i);
 516                 else
 517                         info("Successfully set pthread stacksize to %zu bytes", wanted_stacksize);
 518         }
 519
 520         // ------------------------------------------------------------------------
 521         // spawn the threads
 522
 523         for (i = 0; static_threads[i].name != NULL ; i++) {
 524                 struct netdata_static_thread *st = &static_threads[i];
 525
 526                 if(st->enabled) {
 527                         st->thread = malloc(sizeof(pthread_t));
 528                         if(!st->thread)
 529                                 fatal("Cannot allocate pthread_t memory");
 530
 531                         info("Starting thread %s.", st->name);
 532
 533                         if(pthread_create(st->thread, &attr, st->start_routine, NULL))
 534                                 error("failed to create new thread for %s.", st->name);
 535
 536                         else if(pthread_detach(*st->thread))
 537                                 error("Cannot request detach of newly created %s thread.", st->name);
 538                 }
 539                 else info("Not starting thread %s.", st->name);
 540         }
 541
 542         // for future use - the main thread
 543         while(1) {
 544                 if(netdata_exit != 0) {
 545                         netdata_exit++;
 546
 547                         if(netdata_exit > 5) {
 548                                 netdata_cleanup_and_exit(0);
 549                                 exit(0);
 550                         }
 551                 }
 552                 sleep(2);
 553         }
 554
 555         exit(0);
 556 }