src/main.c

   1 #include <stdio.h>
   2 #include <stdlib.h>
   3 #include <unistd.h>
   4 #include <signal.h>
   5 #include <syslog.h>
   6 #include <pthread.h>
   7 #include <string.h>
   8 #include <errno.h>
   9 #include <sys/types.h>
  10 #include <sys/wait.h>
  11
  12 #include "common.h"
  13 #include "log.h"
  14 #include "daemon.h"
  15 #include "web_server.h"
  16 #include "popen.h"
  17 #include "config.h"
  18 #include "web_client.h"
  19 #include "rrd.h"
  20 #include "rrd2json.h"
  21
  22 #include "unit_test.h"
  23
  24 #include "plugins_d.h"
  25 #include "plugin_idlejitter.h"
  26 #include "plugin_tc.h"
  27 #include "plugin_checks.h"
  28 #include "plugin_proc.h"
  29 #include "plugin_nfacct.h"
  30
  31 #include "main.h"
  32
  33 struct netdata_static_thread {
  34         char *name;
  35
  36         char *config_section;
  37         char *config_name;
  38
  39         int enabled;
  40
  41         pthread_t *thread;
  42
  43         void (*init_routine) (void);
  44         void *(*start_routine) (void *);
  45 };
  46
  47 struct netdata_static_thread static_threads[] = {
  48         {"tc",                  "plugins",      "tc",                   1, NULL, NULL,  tc_main},
  49         {"idlejitter",  "plugins",      "idlejitter",   1, NULL, NULL,  cpuidlejitter_main},
  50         {"proc",                "plugins",      "proc",                 1, NULL, NULL,  proc_main},
  51
  52 #ifdef INTERNAL_PLUGIN_NFACCT
  53         // nfacct requires root access
  54         // so, we build it as an external plugin with setuid to root
  55         {"nfacct",              "plugins",      "nfacct",               1, NULL, NULL,  nfacct_main},
  56 #endif
  57
  58         {"plugins.d",   NULL,           NULL,                   1, NULL, NULL,  pluginsd_main},
  59         {"check",               "plugins",      "checks",               0, NULL, NULL,  checks_main},
  60         {"web",                 NULL,           NULL,                   1, NULL, NULL,  socket_listen_main},
  61         {NULL,                  NULL,           NULL,                   0, NULL, NULL,  NULL}
  62 };
  63
  64 int killpid(pid_t pid, int sig)
  65 {
  66         int ret = -1;
  67         debug(D_EXIT, "Request to kill pid %d", pid);
  68
  69         errno = 0;
  70         if(kill(pid, 0) == -1) {
  71                 switch(errno) {
  72                         case ESRCH:
  73                                 error("Request to kill pid %d, but it is not running.", pid);
  74                                 break;
  75
  76                         case EPERM:
  77                                 error("Request to kill pid %d, but I do not have enough permissions.", pid);
  78                                 break;
  79
  80                         default:
  81                                 error("Request to kill pid %d, but I received an error.", pid);
  82                                 break;
  83                 }
  84         }
  85         else {
  86                 errno = 0;
  87
  88                 void (*old)(int);
  89                 old = signal(sig, SIG_IGN);
  90                 if(old == SIG_ERR) {
  91                         error("Cannot overwrite signal handler for signal %d", sig);
  92                         old = sig_handler;
  93                 }
  94
  95                 ret = kill(pid, sig);
  96
  97                 if(signal(sig, old) == SIG_ERR)
  98                         error("Cannot restore signal handler for signal %d", sig);
  99
 100                 if(ret == -1) {
 101                         switch(errno) {
 102                                 case ESRCH:
 103                                         error("Cannot kill pid %d, but it is not running.", pid);
 104                                         break;
 105
 106                                 case EPERM:
 107                                         error("Cannot kill pid %d, but I do not have enough permissions.", pid);
 108                                         break;
 109
 110                                 default:
 111                                         error("Cannot kill pid %d, but I received an error.", pid);
 112                                         break;
 113                         }
 114                 }
 115         }
 116
 117         return ret;
 118 }
 119
 120 void kill_childs()
 121 {
 122         siginfo_t info;
 123
 124         struct web_client *w;
 125         for(w = web_clients; w ; w = w->next) {
 126                 debug(D_EXIT, "Stopping web client %s", w->client_ip);
 127                 pthread_cancel(w->thread);
 128                 pthread_join(w->thread, NULL);
 129         }
 130
 131         int i;
 132         for (i = 0; static_threads[i].name != NULL ; i++) {
 133                 if(static_threads[i].thread) {
 134                         debug(D_EXIT, "Stopping %s thread", static_threads[i].name);
 135                         pthread_cancel(*static_threads[i].thread);
 136                         pthread_join(*static_threads[i].thread, NULL);
 137                         static_threads[i].thread = NULL;
 138                 }
 139         }
 140
 141         if(tc_child_pid) {
 142                 info("Killing tc-qos-helper procees");
 143                 if(killpid(tc_child_pid, SIGTERM) != -1)
 144                         waitid(tc_child_pid, 0, &info, WEXITED);
 145         }
 146         tc_child_pid = 0;
 147
 148         struct plugind *cd;
 149         for(cd = pluginsd_root ; cd ; cd = cd->next) {
 150                 debug(D_EXIT, "Stopping %s plugin thread", cd->id);
 151                 pthread_cancel(cd->thread);
 152                 pthread_join(cd->thread, NULL);
 153
 154                 if(cd->pid && !cd->obsolete) {
 155                         debug(D_EXIT, "killing %s plugin process", cd->id);
 156                         if(killpid(cd->pid, SIGTERM) != -1)
 157                                 waitid(cd->pid, 0, &info, WEXITED);
 158                 }
 159         }
 160
 161         debug(D_EXIT, "All threads/childs stopped.");
 162 }
 163
 164
 165 int main(int argc, char **argv)
 166 {
 167         int i;
 168         int config_loaded = 0;
 169         int dont_fork = 0;
 170
 171         // global initialization
 172         get_HZ();
 173
 174         // set the name for logging
 175         program_name = "netdata";
 176
 177         // parse  the arguments
 178         for(i = 1; i < argc ; i++) {
 179                 if(strcmp(argv[i], "-c") == 0 && (i+1) < argc) {
 180                         if(load_config(argv[i+1], 1) != 1) {
 181                                 fprintf(stderr, "Cannot load configuration file %s. Reason: %s\n", argv[i+1], strerror(errno));
 182                                 exit(1);
 183                         }
 184                         else {
 185                                 debug(D_OPTIONS, "Configuration loaded from %s.", argv[i+1]);
 186                                 config_loaded = 1;
 187                         }
 188                         i++;
 189                 }
 190                 else if(strcmp(argv[i], "-df") == 0 && (i+1) < argc) { config_set("global", "debug flags",  argv[i+1]); i++; }
 191                 else if(strcmp(argv[i], "-p")  == 0 && (i+1) < argc) { config_set("global", "port",         argv[i+1]); i++; }
 192                 else if(strcmp(argv[i], "-u")  == 0 && (i+1) < argc) { config_set("global", "run as user",  argv[i+1]); i++; }
 193                 else if(strcmp(argv[i], "-l")  == 0 && (i+1) < argc) { config_set("global", "history",      argv[i+1]); i++; }
 194                 else if(strcmp(argv[i], "-t")  == 0 && (i+1) < argc) { config_set("global", "update every", argv[i+1]); i++; }
 195                 else if(strcmp(argv[i], "-ch") == 0 && (i+1) < argc) { config_set("global", "host access prefix", argv[i+1]); i++; }
 196                 else if(strcmp(argv[i], "-nodeamon") == 0 || strcmp(argv[i], "-nd") == 0) dont_fork = 1;
 197                 else if(strcmp(argv[i], "--unittest")  == 0) {
 198                         if(unit_test_storage()) exit(1);
 199                         exit(0);
 200                         rrd_update_every = 1;
 201                         if(unit_test(1000000, 0)) exit(1);
 202                         if(unit_test(1000000, 500000)) exit(1);
 203                         if(unit_test(1000000, 100000)) exit(1);
 204                         if(unit_test(1000000, 900000)) exit(1);
 205                         if(unit_test(2000000, 0)) exit(1);
 206                         if(unit_test(2000000, 500000)) exit(1);
 207                         if(unit_test(2000000, 100000)) exit(1);
 208                         if(unit_test(2000000, 900000)) exit(1);
 209                         if(unit_test(500000, 500000)) exit(1);
 210                         //if(unit_test(500000, 100000)) exit(1); // FIXME: the expected numbers are wrong
 211                         //if(unit_test(500000, 900000)) exit(1); // FIXME: the expected numbers are wrong
 212                         if(unit_test(500000, 0)) exit(1);
 213                         fprintf(stderr, "\n\nALL TESTS PASSED\n\n");
 214                         exit(0);
 215                 }
 216                 else {
 217                         fprintf(stderr, "Cannot understand option '%s'.\n", argv[i]);
 218                         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]);
 219                         fprintf(stderr, "  -c CONFIG FILE the configuration file to load. Default: %s.\n", CONFIG_DIR "/" CONFIG_FILENAME);
 220                         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);
 221                         fprintf(stderr, "  -t UPDATE_TIMER can be from 1 to %d seconds. Default: %d.\n", UPDATE_EVERY_MAX, UPDATE_EVERY);
 222                         fprintf(stderr, "  -p LISTEN_PORT can be from 1 to %d. Default: %d.\n", 65535, LISTEN_PORT);
 223                         fprintf(stderr, "  -u USERNAME can be any system username to run as. Default: none.\n");
 224                         fprintf(stderr, "  -ch path to access host /proc and /sys when running in a container. Default: empty.\n");
 225                         fprintf(stderr, "  -nd or -nodeamon to disable forking in the background. Default: unset.\n");
 226                         fprintf(stderr, "  -df FLAGS debug options. Default: 0x%08llx.\n", debug_flags);
 227                         exit(1);
 228                 }
 229         }
 230
 231         if(!config_loaded) load_config(NULL, 0);
 232
 233         char *input_log_file = NULL;
 234         char *output_log_file = NULL;
 235         char *error_log_file = NULL;
 236         char *access_log_file = NULL;
 237         {
 238                 char buffer[1024];
 239
 240                 // --------------------------------------------------------------------
 241
 242                 sprintf(buffer, "0x%08llx", 0ULL);
 243                 char *flags = config_get("global", "debug flags", buffer);
 244                 debug_flags = strtoull(flags, NULL, 0);
 245                 debug(D_OPTIONS, "Debug flags set to '0x%8llx'.", debug_flags);
 246
 247                 // --------------------------------------------------------------------
 248
 249                 global_host_prefix = config_get("global", "host access prefix", "");
 250
 251                 // --------------------------------------------------------------------
 252
 253                 output_log_file = config_get("global", "debug log", LOG_DIR "/debug.log");
 254                 if(strcmp(output_log_file, "syslog") == 0) {
 255                         output_log_syslog = 1;
 256                         output_log_file = NULL;
 257                 }
 258                 else if(strcmp(output_log_file, "none") == 0) {
 259                         output_log_syslog = 0;
 260                         output_log_file = NULL;
 261                 }
 262                 else output_log_syslog = 0;
 263
 264                 // --------------------------------------------------------------------
 265
 266                 silent = 0;
 267                 error_log_file = config_get("global", "error log", LOG_DIR "/error.log");
 268                 if(strcmp(error_log_file, "syslog") == 0) {
 269                         error_log_syslog = 1;
 270                         error_log_file = NULL;
 271                 }
 272                 else if(strcmp(error_log_file, "none") == 0) {
 273                         error_log_syslog = 0;
 274                         error_log_file = NULL;
 275                         silent = 1; // optimization - do not even generate debug log entries
 276                 }
 277                 else error_log_syslog = 0;
 278
 279                 // --------------------------------------------------------------------
 280
 281                 access_log_file = config_get("global", "access log", LOG_DIR "/access.log");
 282                 if(strcmp(access_log_file, "syslog") == 0) {
 283                         access_log_syslog = 1;
 284                         access_log_file = NULL;
 285                 }
 286                 else if(strcmp(access_log_file, "none") == 0) {
 287                         access_log_syslog = 0;
 288                         access_log_file = NULL;
 289                 }
 290                 else access_log_syslog = 0;
 291
 292                 // --------------------------------------------------------------------
 293
 294                 rrd_memory_mode = rrd_memory_mode_id(config_get("global", "memory mode", rrd_memory_mode_name(rrd_memory_mode)));
 295
 296                 // --------------------------------------------------------------------
 297
 298                 if(gethostname(buffer, HOSTNAME_MAX) == -1)
 299                         error("WARNING: Cannot get machine hostname.");
 300                 hostname = config_get("global", "hostname", buffer);
 301                 debug(D_OPTIONS, "hostname set to '%s'", hostname);
 302
 303                 // --------------------------------------------------------------------
 304
 305                 rrd_default_history_entries = config_get_number("global", "history", RRD_DEFAULT_HISTORY_ENTRIES);
 306                 if(rrd_default_history_entries < 5 || rrd_default_history_entries > RRD_HISTORY_ENTRIES_MAX) {
 307                         fprintf(stderr, "Invalid save lines %d given. Defaulting to %d.\n", rrd_default_history_entries, RRD_DEFAULT_HISTORY_ENTRIES);
 308                         rrd_default_history_entries = RRD_DEFAULT_HISTORY_ENTRIES;
 309                 }
 310                 else {
 311                         debug(D_OPTIONS, "save lines set to %d.", rrd_default_history_entries);
 312                 }
 313
 314                 // --------------------------------------------------------------------
 315
 316                 rrd_update_every = config_get_number("global", "update every", UPDATE_EVERY);
 317                 if(rrd_update_every < 1 || rrd_update_every > 600) {
 318                         fprintf(stderr, "Invalid update timer %d given. Defaulting to %d.\n", rrd_update_every, UPDATE_EVERY_MAX);
 319                         rrd_update_every = UPDATE_EVERY;
 320                 }
 321                 else debug(D_OPTIONS, "update timer set to %d.", rrd_update_every);
 322
 323                 // --------------------------------------------------------------------
 324
 325                 for (i = 0; static_threads[i].name != NULL ; i++) {
 326                         struct netdata_static_thread *st = &static_threads[i];
 327
 328                         if(st->config_name) st->enabled = config_get_boolean(st->config_section, st->config_name, st->enabled);
 329                         if(st->enabled && st->init_routine) st->init_routine();
 330                 }
 331
 332                 // --------------------------------------------------------------------
 333
 334                 prepare_rundir();
 335                 char *user = config_get("global", "run as user", (getuid() == 0)?"nobody":"");
 336                 if(*user) {
 337                         if(become_user(user) != 0) {
 338                                 fprintf(stderr, "Cannot become user %s.\n", user);
 339                                 exit(1);
 340                         }
 341                         else debug(D_OPTIONS, "Successfully became user %s.", user);
 342                 }
 343
 344                 // --------------------------------------------------------------------
 345
 346                 listen_backlog = config_get_number("global", "http port listen backlog", LISTEN_BACKLOG);
 347
 348                 listen_port = config_get_number("global", "port", LISTEN_PORT);
 349                 if(listen_port < 1 || listen_port > 65535) {
 350                         fprintf(stderr, "Invalid listen port %d given. Defaulting to %d.\n", listen_port, LISTEN_PORT);
 351                         listen_port = LISTEN_PORT;
 352                 }
 353                 else debug(D_OPTIONS, "Listen port set to %d.", listen_port);
 354
 355                 int ip = 0;
 356                 char *ipv = config_get("global", "ip version", "any");
 357                 if(!strcmp(ipv, "any") || !strcmp(ipv, "both") || !strcmp(ipv, "all")) ip = 0;
 358                 else if(!strcmp(ipv, "ipv4") || !strcmp(ipv, "IPV4") || !strcmp(ipv, "IPv4") || !strcmp(ipv, "4")) ip = 4;
 359                 else if(!strcmp(ipv, "ipv6") || !strcmp(ipv, "IPV6") || !strcmp(ipv, "IPv6") || !strcmp(ipv, "6")) ip = 6;
 360                 else fprintf(stderr, "Cannot understand ip version '%s'. Assumming 'any'.", ipv);
 361
 362                 if(ip == 0 || ip == 6) listen_fd = create_listen_socket6(listen_port, listen_backlog);
 363                 if(listen_fd < 0) {
 364                         listen_fd = create_listen_socket4(listen_port, listen_backlog);
 365                         if(listen_fd >= 0 && ip != 4) fprintf(stderr, "Managed to open an IPv4 socket on port %d.", listen_port);
 366                 }
 367
 368                 if(listen_fd < 0) fatal("Cannot listen socket.");
 369         }
 370
 371         // never become a problem
 372         if(nice(20) == -1) fprintf(stderr, "Cannot lower my CPU priority. Error: %s.\n", strerror(errno));
 373
 374 #ifndef NETDATA_NO_DAEMON
 375         if(become_daemon(dont_fork, 0, input_log_file, output_log_file, error_log_file, access_log_file, &access_fd, &stdaccess) == -1) {
 376                 fprintf(stderr, "Cannot demonize myself (%s).", strerror(errno));
 377                 exit(1);
 378         }
 379 #endif
 380
 381         if(output_log_syslog || error_log_syslog || access_log_syslog)
 382                 openlog("netdata", LOG_PID, LOG_DAEMON);
 383
 384         info("NetData started on pid %d", getpid());
 385
 386
 387         // catch all signals
 388         for (i = 1 ; i < 65 ;i++) {
 389                 switch(i) {
 390                         case SIGSEGV:
 391                         case SIGFPE:
 392                         case SIGCHLD:
 393                                 signal(i, SIG_DFL);
 394                                 break;
 395
 396                         default:
 397                                 signal(i,  sig_handler);
 398                                 break;
 399                 }
 400         }
 401
 402         for (i = 0; static_threads[i].name != NULL ; i++) {
 403                 struct netdata_static_thread *st = &static_threads[i];
 404
 405                 if(st->enabled) {
 406                         st->thread = malloc(sizeof(pthread_t));
 407
 408                         info("Starting thread %s.", st->name);
 409
 410                         if(pthread_create(st->thread, NULL, st->start_routine, NULL))
 411                                 error("failed to create new thread for %s.", st->name);
 412
 413                         else if(pthread_detach(*st->thread))
 414                                 error("Cannot request detach of newly created %s thread.", st->name);
 415                 }
 416                 else info("Not starting thread %s.", st->name);
 417         }
 418
 419         // for future use - the main thread
 420         while(1) sleep(60);
 421
 422         exit(0);
 423 }