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