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