]> arthur.barton.de Git - netdata.git/blob - src/main.c
Merge pull request #327 from ktsaou/master
[netdata.git] / 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 extern void *cgroups_main(void *ptr);
41
42 int netdata_exit = 0;
43
44 void netdata_cleanup_and_exit(int ret)
45 {
46         netdata_exit = 1;
47         rrdset_save_all();
48         // kill_childs();
49
50         // let it log a few more error messages
51         error_log_limit_reset();
52
53         if(pidfd != -1) {
54                 if(ftruncate(pidfd, 0) != 0)
55                         error("Cannot truncate pidfile '%s'.", pidfile);
56
57                 close(pidfd);
58                 pidfd = -1;
59         }
60
61         if(pidfile[0]) {
62                 if(unlink(pidfile) != 0)
63                         error("Cannot unlink pidfile '%s'.", pidfile);
64         }
65
66         info("NetData exiting. Bye bye...");
67         exit(ret);
68 }
69
70 struct netdata_static_thread {
71         char *name;
72
73         char *config_section;
74         char *config_name;
75
76         int enabled;
77
78         pthread_t *thread;
79
80         void (*init_routine) (void);
81         void *(*start_routine) (void *);
82 };
83
84 struct netdata_static_thread static_threads[] = {
85         {"tc",                  "plugins",      "tc",                   1, NULL, NULL,  tc_main},
86         {"idlejitter",  "plugins",      "idlejitter",   1, NULL, NULL,  cpuidlejitter_main},
87         {"proc",                "plugins",      "proc",                 1, NULL, NULL,  proc_main},
88         {"cgroups",             "plugins",      "cgroups",              1, NULL, NULL,  cgroups_main},
89
90 #ifdef INTERNAL_PLUGIN_NFACCT
91         // nfacct requires root access
92         // so, we build it as an external plugin with setuid to root
93         {"nfacct",              "plugins",      "nfacct",               1, NULL, NULL,  nfacct_main},
94 #endif
95
96         {"plugins.d",   NULL,           NULL,                   1, NULL, NULL,  pluginsd_main},
97         {"check",               "plugins",      "checks",               0, NULL, NULL,  checks_main},
98         {"web",                 NULL,           NULL,                   1, NULL, NULL,  socket_listen_main},
99         {NULL,                  NULL,           NULL,                   0, NULL, NULL,  NULL}
100 };
101
102 int killpid(pid_t pid, int sig)
103 {
104         int ret = -1;
105         debug(D_EXIT, "Request to kill pid %d", pid);
106
107         errno = 0;
108         if(kill(pid, 0) == -1) {
109                 switch(errno) {
110                         case ESRCH:
111                                 error("Request to kill pid %d, but it is not running.", pid);
112                                 break;
113
114                         case EPERM:
115                                 error("Request to kill pid %d, but I do not have enough permissions.", pid);
116                                 break;
117
118                         default:
119                                 error("Request to kill pid %d, but I received an error.", pid);
120                                 break;
121                 }
122         }
123         else {
124                 errno = 0;
125
126                 void (*old)(int);
127                 old = signal(sig, SIG_IGN);
128                 if(old == SIG_ERR) {
129                         error("Cannot overwrite signal handler for signal %d", sig);
130                         old = sig_handler;
131                 }
132
133                 ret = kill(pid, sig);
134
135                 if(signal(sig, old) == SIG_ERR)
136                         error("Cannot restore signal handler for signal %d", sig);
137
138                 if(ret == -1) {
139                         switch(errno) {
140                                 case ESRCH:
141                                         error("Cannot kill pid %d, but it is not running.", pid);
142                                         break;
143
144                                 case EPERM:
145                                         error("Cannot kill pid %d, but I do not have enough permissions.", pid);
146                                         break;
147
148                                 default:
149                                         error("Cannot kill pid %d, but I received an error.", pid);
150                                         break;
151                         }
152                 }
153         }
154
155         return ret;
156 }
157
158 void kill_childs()
159 {
160         siginfo_t info;
161
162         struct web_client *w;
163         for(w = web_clients; w ; w = w->next) {
164                 debug(D_EXIT, "Stopping web client %s", w->client_ip);
165                 pthread_cancel(w->thread);
166                 pthread_join(w->thread, NULL);
167         }
168
169         int i;
170         for (i = 0; static_threads[i].name != NULL ; i++) {
171                 if(static_threads[i].thread) {
172                         debug(D_EXIT, "Stopping %s thread", static_threads[i].name);
173                         pthread_cancel(*static_threads[i].thread);
174                         pthread_join(*static_threads[i].thread, NULL);
175                         static_threads[i].thread = NULL;
176                 }
177         }
178
179         if(tc_child_pid) {
180                 info("Killing tc-qos-helper procees");
181                 if(killpid(tc_child_pid, SIGTERM) != -1)
182                         waitid(P_PID, (id_t) tc_child_pid, &info, WEXITED);
183         }
184         tc_child_pid = 0;
185
186         struct plugind *cd;
187         for(cd = pluginsd_root ; cd ; cd = cd->next) {
188                 debug(D_EXIT, "Stopping %s plugin thread", cd->id);
189                 pthread_cancel(cd->thread);
190                 pthread_join(cd->thread, NULL);
191
192                 if(cd->pid && !cd->obsolete) {
193                         debug(D_EXIT, "killing %s plugin process", cd->id);
194                         if(killpid(cd->pid, SIGTERM) != -1)
195                                 waitid(P_PID, (id_t) cd->pid, &info, WEXITED);
196                 }
197         }
198
199         // if, for any reason there is any child exited
200         // catch it here
201         waitid(P_PID, 0, &info, WEXITED|WNOHANG);
202
203         debug(D_EXIT, "All threads/childs stopped.");
204 }
205
206
207 int main(int argc, char **argv)
208 {
209         int i;
210         int config_loaded = 0;
211         int dont_fork = 0;
212         size_t wanted_stacksize = 0, stacksize = 0;
213         pthread_attr_t attr;
214
215         // global initialization
216         get_HZ();
217
218         // set the name for logging
219         program_name = "netdata";
220
221         // parse  the arguments
222         for(i = 1; i < argc ; i++) {
223                 if(strcmp(argv[i], "-c") == 0 && (i+1) < argc) {
224                         if(load_config(argv[i+1], 1) != 1) {
225                                 error("Cannot load configuration file %s.", argv[i+1]);
226                                 exit(1);
227                         }
228                         else {
229                                 debug(D_OPTIONS, "Configuration loaded from %s.", argv[i+1]);
230                                 config_loaded = 1;
231                         }
232                         i++;
233                 }
234                 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++; }
235                 else if(strcmp(argv[i], "-p")  == 0 && (i+1) < argc) { config_set("global", "port",         argv[i+1]); i++; }
236                 else if(strcmp(argv[i], "-u")  == 0 && (i+1) < argc) { config_set("global", "run as user",  argv[i+1]); i++; }
237                 else if(strcmp(argv[i], "-l")  == 0 && (i+1) < argc) { config_set("global", "history",      argv[i+1]); i++; }
238                 else if(strcmp(argv[i], "-t")  == 0 && (i+1) < argc) { config_set("global", "update every", argv[i+1]); i++; }
239                 else if(strcmp(argv[i], "-ch") == 0 && (i+1) < argc) { config_set("global", "host access prefix", argv[i+1]); i++; }
240                 else if(strcmp(argv[i], "-stacksize") == 0 && (i+1) < argc) { config_set("global", "pthread stack size", argv[i+1]); i++; }
241                 else if(strcmp(argv[i], "-nodaemon") == 0 || strcmp(argv[i], "-nd") == 0) dont_fork = 1;
242                 else if(strcmp(argv[i], "-pidfile") == 0 && (i+1) < argc) {
243                         i++;
244                         strncpy(pidfile, argv[i], FILENAME_MAX);
245                         pidfile[FILENAME_MAX] = '\0';
246                 }
247                 else if(strcmp(argv[i], "--unittest")  == 0) {
248                         rrd_update_every = 1;
249                         if(run_all_mockup_tests()) exit(1);
250                         if(unit_test_storage()) exit(1);
251                         fprintf(stderr, "\n\nALL TESTS PASSED\n\n");
252                         exit(0);
253                 }
254                 else {
255                         fprintf(stderr, "Cannot understand option '%s'.\n", argv[i]);
256                         fprintf(stderr, "\nUSAGE: %s [-d] [-l LINES_TO_SAVE] [-u UPDATE_TIMER] [-p LISTEN_PORT] [-df debug flags].\n\n", argv[0]);
257                         fprintf(stderr, "  -c CONFIG FILE the configuration file to load. Default: %s.\n", CONFIG_DIR "/" CONFIG_FILENAME);
258                         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);
259                         fprintf(stderr, "  -t UPDATE_TIMER can be from 1 to %d seconds. Default: %d.\n", UPDATE_EVERY_MAX, UPDATE_EVERY);
260                         fprintf(stderr, "  -p LISTEN_PORT can be from 1 to %d. Default: %d.\n", 65535, LISTEN_PORT);
261                         fprintf(stderr, "  -u USERNAME can be any system username to run as. Default: none.\n");
262                         fprintf(stderr, "  -ch path to access host /proc and /sys when running in a container. Default: empty.\n");
263                         fprintf(stderr, "  -nd or -nodeamon to disable forking in the background. Default: unset.\n");
264                         fprintf(stderr, "  -df FLAGS debug options. Default: 0x%08llx.\n", debug_flags);
265                         fprintf(stderr, "  -stacksize BYTES to overwrite the pthread stack size.\n");
266                         fprintf(stderr, "  -pidfile FILENAME to save a pid while running.\n");
267                         exit(1);
268                 }
269         }
270
271         if(!config_loaded) load_config(NULL, 0);
272
273         // prepare configuration environment variables for the plugins
274         setenv("NETDATA_CONFIG_DIR" , config_get("global", "config directory"   , CONFIG_DIR) , 1);
275         setenv("NETDATA_PLUGINS_DIR", config_get("global", "plugins directory"  , PLUGINS_DIR), 1);
276         setenv("NETDATA_WEB_DIR"    , config_get("global", "web files directory", WEB_DIR)    , 1);
277         setenv("NETDATA_CACHE_DIR"  , config_get("global", "cache directory"    , CACHE_DIR)  , 1);
278         setenv("NETDATA_LOG_DIR"    , config_get("global", "log directory"      , LOG_DIR)    , 1);
279         setenv("NETDATA_HOST_PREFIX", config_get("global", "host access prefix" , "")         , 1);
280         setenv("HOME"               , config_get("global", "home directory"     , CACHE_DIR)  , 1);
281
282         // avoid extended to stat(/etc/localtime)
283         // http://stackoverflow.com/questions/4554271/how-to-avoid-excessive-stat-etc-localtime-calls-in-strftime-on-linux
284         setenv("TZ", ":/etc/localtime", 0);
285
286         // cd to /tmp to avoid any plugins writing files at random places
287         if(chdir("/tmp")) error("netdata: ERROR: Cannot cd to /tmp");
288
289         char *input_log_file = NULL;
290         char *output_log_file = NULL;
291         char *error_log_file = NULL;
292         char *access_log_file = NULL;
293         char *user = NULL;
294         {
295                 char buffer[1024];
296
297                 // --------------------------------------------------------------------
298
299                 sprintf(buffer, "0x%08llx", 0ULL);
300                 char *flags = config_get("global", "debug flags", buffer);
301                 setenv("NETDATA_DEBUG_FLAGS", flags, 1);
302
303                 debug_flags = strtoull(flags, NULL, 0);
304                 debug(D_OPTIONS, "Debug flags set to '0x%8llx'.", debug_flags);
305
306                 if(debug_flags != 0) {
307                         struct rlimit rl = { RLIM_INFINITY, RLIM_INFINITY };
308                         if(setrlimit(RLIMIT_CORE, &rl) != 0)
309                                 info("Cannot request unlimited core dumps for debugging... Proceeding anyway...");
310                         prctl(PR_SET_DUMPABLE, 1, 0, 0, 0);
311                 }
312
313                 // --------------------------------------------------------------------
314
315 #ifdef MADV_MERGEABLE
316                 enable_ksm = config_get_boolean("global", "memory deduplication (ksm)", enable_ksm);
317 #else
318 #warning "Kernel memory deduplication (KSM) is not available"
319 #endif
320
321                 // --------------------------------------------------------------------
322
323
324                 global_host_prefix = config_get("global", "host access prefix", "");
325                 setenv("NETDATA_HOST_PREFIX", global_host_prefix, 1);
326
327                 // --------------------------------------------------------------------
328
329                 output_log_file = config_get("global", "debug log", LOG_DIR "/debug.log");
330                 if(strcmp(output_log_file, "syslog") == 0) {
331                         output_log_syslog = 1;
332                         output_log_file = NULL;
333                 }
334                 else if(strcmp(output_log_file, "none") == 0) {
335                         output_log_syslog = 0;
336                         output_log_file = NULL;
337                 }
338                 else output_log_syslog = 0;
339
340                 // --------------------------------------------------------------------
341
342                 error_log_file = config_get("global", "error log", LOG_DIR "/error.log");
343                 if(strcmp(error_log_file, "syslog") == 0) {
344                         error_log_syslog = 1;
345                         error_log_file = NULL;
346                 }
347                 else if(strcmp(error_log_file, "none") == 0) {
348                         error_log_syslog = 0;
349                         error_log_file = NULL;
350                         // optimization - do not even generate debug log entries
351                 }
352                 else error_log_syslog = 0;
353
354                 error_log_throttle_period = config_get_number("global", "errors flood protection period", error_log_throttle_period);
355                 setenv("NETDATA_ERRORS_THROTTLE_PERIOD", config_get("global", "errors flood protection period"    , ""), 1);
356
357                 error_log_errors_per_period = config_get_number("global", "errors to trigger flood protection", error_log_errors_per_period);
358                 setenv("NETDATA_ERRORS_PER_PERIOD"     , config_get("global", "errors to trigger flood protection", ""), 1);
359
360                 // --------------------------------------------------------------------
361
362                 access_log_file = config_get("global", "access log", LOG_DIR "/access.log");
363                 if(strcmp(access_log_file, "syslog") == 0) {
364                         access_log_syslog = 1;
365                         access_log_file = NULL;
366                 }
367                 else if(strcmp(access_log_file, "none") == 0) {
368                         access_log_syslog = 0;
369                         access_log_file = NULL;
370                 }
371                 else access_log_syslog = 0;
372
373                 // --------------------------------------------------------------------
374
375                 rrd_memory_mode = rrd_memory_mode_id(config_get("global", "memory mode", rrd_memory_mode_name(rrd_memory_mode)));
376
377                 // --------------------------------------------------------------------
378
379                 if(gethostname(buffer, HOSTNAME_MAX) == -1)
380                         error("WARNING: Cannot get machine hostname.");
381                 hostname = config_get("global", "hostname", buffer);
382                 debug(D_OPTIONS, "hostname set to '%s'", hostname);
383
384                 // --------------------------------------------------------------------
385
386                 rrd_default_history_entries = (int) config_get_number("global", "history", RRD_DEFAULT_HISTORY_ENTRIES);
387                 if(rrd_default_history_entries < 5 || rrd_default_history_entries > RRD_HISTORY_ENTRIES_MAX) {
388                         info("Invalid save lines %d given. Defaulting to %d.", rrd_default_history_entries, RRD_DEFAULT_HISTORY_ENTRIES);
389                         rrd_default_history_entries = RRD_DEFAULT_HISTORY_ENTRIES;
390                 }
391                 else {
392                         debug(D_OPTIONS, "save lines set to %d.", rrd_default_history_entries);
393                 }
394
395                 // --------------------------------------------------------------------
396
397                 rrd_update_every = (int) config_get_number("global", "update every", UPDATE_EVERY);
398                 if(rrd_update_every < 1 || rrd_update_every > 600) {
399                         info("Invalid update timer %d given. Defaulting to %d.", rrd_update_every, UPDATE_EVERY_MAX);
400                         rrd_update_every = UPDATE_EVERY;
401                 }
402                 else debug(D_OPTIONS, "update timer set to %d.", rrd_update_every);
403
404                 // let the plugins know the min update_every
405                 {
406                         char buf[50];
407                         snprintf(buf, 50, "%d", rrd_update_every);
408                         setenv("NETDATA_UPDATE_EVERY", buf, 1);
409                 }
410
411                 // --------------------------------------------------------------------
412
413                 i = pthread_attr_init(&attr);
414                 if(i != 0)
415                         fatal("pthread_attr_init() failed with code %d.", i);
416
417                 i = pthread_attr_getstacksize(&attr, &stacksize);
418                 if(i != 0)
419                         fatal("pthread_attr_getstacksize() failed with code %d.", i);
420                 else
421                         debug(D_OPTIONS, "initial pthread stack size is %zu bytes", stacksize);
422
423                 wanted_stacksize = config_get_number("global", "pthread stack size", stacksize);
424
425                 // --------------------------------------------------------------------
426
427                 for (i = 0; static_threads[i].name != NULL ; i++) {
428                         struct netdata_static_thread *st = &static_threads[i];
429
430                         if(st->config_name) st->enabled = config_get_boolean(st->config_section, st->config_name, st->enabled);
431                         if(st->enabled && st->init_routine) st->init_routine();
432                 }
433
434                 // --------------------------------------------------------------------
435
436                 // get the user we should run
437                 // IMPORTANT: this is required before web_files_uid()
438                 user = config_get("global", "run as user"    , (getuid() == 0)?NETDATA_USER:"");
439
440                 // IMPORTANT: these have to run once, while single threaded
441                 web_files_uid(); // IMPORTANT: web_files_uid() before web_files_gid()
442                 web_files_gid();
443
444                 // --------------------------------------------------------------------
445
446                 listen_backlog = (int) config_get_number("global", "http port listen backlog", LISTEN_BACKLOG);
447
448                 listen_port = (int) config_get_number("global", "port", LISTEN_PORT);
449                 if(listen_port < 1 || listen_port > 65535) {
450                         info("Invalid listen port %d given. Defaulting to %d.", listen_port, LISTEN_PORT);
451                         listen_port = LISTEN_PORT;
452                 }
453                 else debug(D_OPTIONS, "Listen port set to %d.", listen_port);
454
455                 int ip = 0;
456                 char *ipv = config_get("global", "ip version", "any");
457                 if(!strcmp(ipv, "any") || !strcmp(ipv, "both") || !strcmp(ipv, "all")) ip = 0;
458                 else if(!strcmp(ipv, "ipv4") || !strcmp(ipv, "IPV4") || !strcmp(ipv, "IPv4") || !strcmp(ipv, "4")) ip = 4;
459                 else if(!strcmp(ipv, "ipv6") || !strcmp(ipv, "IPV6") || !strcmp(ipv, "IPv6") || !strcmp(ipv, "6")) ip = 6;
460                 else info("Cannot understand ip version '%s'. Assuming 'any'.", ipv);
461
462                 if(ip == 0 || ip == 6) listen_fd = create_listen_socket6(config_get("global", "bind socket to IP", "*"), listen_port, listen_backlog);
463                 if(listen_fd < 0) {
464                         listen_fd = create_listen_socket4(config_get("global", "bind socket to IP", "*"), listen_port, listen_backlog);
465                         if(listen_fd >= 0 && ip != 4) info("Managed to open an IPv4 socket on port %d.", listen_port);
466                 }
467
468                 if(listen_fd < 0) fatal("Cannot listen socket.");
469         }
470
471         // never become a problem
472         if(nice(20) == -1) error("Cannot lower my CPU priority.");
473
474         if(become_daemon(dont_fork, 0, user, input_log_file, output_log_file, error_log_file, access_log_file, &access_fd, &stdaccess) == -1) {
475                 fatal("Cannot demonize myself.");
476                 exit(1);
477         }
478
479         if(debug_flags != 0) {
480                 struct rlimit rl = { RLIM_INFINITY, RLIM_INFINITY };
481                 if(setrlimit(RLIMIT_CORE, &rl) != 0)
482                         info("Cannot request unlimited core dumps for debugging... Proceeding anyway...");
483
484                 prctl(PR_SET_DUMPABLE, 1, 0, 0, 0);
485         }
486
487         if(output_log_syslog || error_log_syslog || access_log_syslog)
488                 openlog("netdata", LOG_PID, LOG_DAEMON);
489
490         info("NetData started on pid %d", getpid());
491
492
493         // catch all signals
494         for (i = 1 ; i < 65 ;i++) {
495                 switch(i) {
496                         case SIGKILL: // not catchable
497                         case SIGSTOP: // not catchable
498                                 break;
499
500                         case SIGSEGV:
501                         case SIGFPE:
502                         case SIGCHLD:
503                                 signal(i, SIG_DFL);
504                                 break;
505
506                         default:
507                                 signal(i,  sig_handler);
508                                 break;
509                 }
510         }
511
512         // ------------------------------------------------------------------------
513         // get default pthread stack size
514
515         if(stacksize < wanted_stacksize) {
516                 i = pthread_attr_setstacksize(&attr, wanted_stacksize);
517                 if(i != 0)
518                         fatal("pthread_attr_setstacksize() to %zu bytes, failed with code %d.", wanted_stacksize, i);
519                 else
520                         info("Successfully set pthread stacksize to %zu bytes", wanted_stacksize);
521         }
522
523         // ------------------------------------------------------------------------
524         // spawn the threads
525
526         for (i = 0; static_threads[i].name != NULL ; i++) {
527                 struct netdata_static_thread *st = &static_threads[i];
528
529                 if(st->enabled) {
530                         st->thread = malloc(sizeof(pthread_t));
531                         if(!st->thread)
532                                 fatal("Cannot allocate pthread_t memory");
533
534                         info("Starting thread %s.", st->name);
535
536                         if(pthread_create(st->thread, &attr, st->start_routine, NULL))
537                                 error("failed to create new thread for %s.", st->name);
538
539                         else if(pthread_detach(*st->thread))
540                                 error("Cannot request detach of newly created %s thread.", st->name);
541                 }
542                 else info("Not starting thread %s.", st->name);
543         }
544
545         // for future use - the main thread
546         while(1) {
547                 if(netdata_exit != 0) {
548                         netdata_exit++;
549
550                         if(netdata_exit > 5) {
551                                 netdata_cleanup_and_exit(0);
552                                 exit(0);
553                         }
554                 }
555                 sleep(2);
556         }
557
558         exit(0);
559 }