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