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