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