2 This file is part of GNUnet.
3 (C) 2009 Christian Grothoff (and other contributing authors)
5 GNUnet is free software; you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published
7 by the Free Software Foundation; either version 2, or (at your
8 option) any later version.
10 GNUnet is distributed in the hope that it will be useful, but
11 WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 General Public License for more details.
15 You should have received a copy of the GNU General Public License
16 along with GNUnet; see the file COPYING. If not, write to the
17 Free Software Foundation, Inc., 59 Temple Place - Suite 330,
18 Boston, MA 02111-1307, USA.
22 * @file arm/gnunet-service-arm.c
23 * @brief the automated restart manager service
24 * @author Christian Grothoff
27 * - multiple start-stop requests with RC>1 can result
28 * in UP/DOWN signals based on "pending" that are inaccurate...
29 * => have list of clients waiting for a resolution instead of
30 * giving instant (but incorrect) replies
31 * - code could go into restart-loop for a service
32 * if service crashes instantly -- need exponential back-off
33 * - need to test auto-restart code on configuration changes;
34 * - should refine restart code to check if *relevant* parts of the
35 * configuration were changed (anything in the section for the service)
36 * - should have a way to specify dependencies between services and
37 * manage restarts of groups of services
40 #include "gnunet_client_lib.h"
41 #include "gnunet_getopt_lib.h"
42 #include "gnunet_os_lib.h"
43 #include "gnunet_protocols.h"
44 #include "gnunet_service_lib.h"
49 * Run maintenance every second.
51 #define MAINT_FREQUENCY GNUNET_TIME_UNIT_SECONDS
54 * How long do we wait until we decide that a service
57 #define CHECK_TIMEOUT GNUNET_TIME_UNIT_MINUTES
61 typedef void (*CleanCallback) (void *cls, struct ServiceList * pos);
64 * List of our services.
69 * This is a linked list.
71 struct ServiceList *next;
74 * Name of the service.
79 * Name of the binary used.
84 * Name of the configuration file used.
89 * Function to call upon kill completion (waitpid), NULL
90 * if we should simply restart the process.
92 CleanCallback kill_continuation;
95 * Closure for kill_continuation.
97 void *kill_continuation_cls;
100 * Process ID of the child.
105 * Last time the config of this service was
111 * Reference counter (counts how many times we've been
112 * asked to start the service). We only actually stop
113 * it once rc hits zero.
120 * List of running services.
122 static struct ServiceList *running;
127 static struct GNUNET_CONFIGURATION_Handle *cfg;
132 static struct GNUNET_SCHEDULER_Handle *sched;
135 * Command to prepend to each actual command.
137 static char *prefix_command;
141 write_result (void *cls, size_t size, void *buf)
144 struct GNUNET_MessageHeader *msg;
147 return 0; /* error, not much we can do */
148 GNUNET_assert (size >= sizeof (struct GNUNET_MessageHeader));
150 msg->size = htons (sizeof (struct GNUNET_MessageHeader));
151 msg->type = htons (*res);
153 return sizeof (struct GNUNET_MessageHeader);
159 * Signal our client that we will start or stop the
162 * @return NULL if it was not found
165 signal_result (struct GNUNET_SERVER_Client *client,
166 const char *name, uint16_t result)
171 GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
172 "Telling client that service `%s' is now %s\n",
174 result == GNUNET_MESSAGE_TYPE_ARM_IS_DOWN ? "down" : "up");
176 res = GNUNET_malloc (sizeof (uint16_t));
178 GNUNET_SERVER_notify_transmit_ready (client,
179 sizeof (struct GNUNET_MessageHeader),
180 GNUNET_TIME_UNIT_FOREVER_REL,
186 * Find the process with the given PID in the
189 * @return NULL if it was not found
191 static struct ServiceList *
194 struct ServiceList *pos;
208 * Find the process with the given service
209 * name in the given list, remove it and return it.
211 * @return NULL if it was not found
213 static struct ServiceList *
214 find_name (const char *name)
216 struct ServiceList *pos;
217 struct ServiceList *prev;
223 if (0 == strcmp (pos->name, name))
228 prev->next = pos->next;
240 free_entry (struct ServiceList *pos)
242 GNUNET_free_non_null (pos->config);
243 GNUNET_free_non_null (pos->binary);
244 GNUNET_free (pos->name);
252 * Actually start the process for the given service.
254 * @param sl identifies service to start
257 start_process (struct ServiceList *sl)
262 unsigned int argv_size;
265 const char *firstarg;
270 GNUNET_CONFIGURATION_get_value_string (cfg,
271 sl->name, "PREFIX", &loprefix))
272 loprefix = GNUNET_strdup (prefix_command);
274 GNUNET_CONFIGURATION_get_value_string (cfg,
275 sl->name, "OPTIONS", &options))
276 options = GNUNET_strdup ("");
277 use_debug = GNUNET_CONFIGURATION_get_value_yesno (cfg, sl->name, "DEBUG");
279 GNUNET_log (GNUNET_ERROR_TYPE_INFO, _("Starting service `%s'\n"), sl->name);
281 GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
282 "Starting service `%s' using binary `%s' and configuration `%s'\n",
283 sl->name, sl->binary, sl->config);
289 while ('\0' != *lopos)
296 while ('\0' != *optpos)
303 argv = GNUNET_malloc (argv_size * sizeof (char *));
307 while ('\0' != *lopos)
309 while (*lopos == ' ')
315 argv[argv_size++] = lopos;
316 while (('\0' != *lopos) && (' ' != *lopos))
324 firstarg = sl->binary;
325 argv[argv_size++] = sl->binary;
326 argv[argv_size++] = "-c";
327 argv[argv_size++] = sl->config;
328 if (GNUNET_YES == use_debug)
330 argv[argv_size++] = "-L";
331 argv[argv_size++] = "DEBUG";
334 while ('\0' != *optpos)
336 while (*optpos == ' ')
340 argv[argv_size++] = optpos;
341 while (('\0' != *optpos) && (' ' != *optpos))
348 argv[argv_size++] = NULL;
349 sl->pid = GNUNET_OS_start_process_v (firstarg, argv);
351 GNUNET_free (loprefix);
352 GNUNET_free (options);
357 * Start the specified service.
360 start_service (struct GNUNET_SERVER_Client *client, const char *servicename)
362 struct ServiceList *sl;
366 sl = find_name (servicename);
369 /* already running, just increment RC */
370 GNUNET_log (GNUNET_ERROR_TYPE_INFO,
371 _("Service `%s' already running.\n"), servicename);
375 signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_UP);
379 GNUNET_CONFIGURATION_get_value_string (cfg,
380 servicename, "BINARY", &binary))
382 GNUNET_log (GNUNET_ERROR_TYPE_WARNING,
383 _("Binary implementing service `%s' not known!\n"),
385 signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_DOWN);
389 GNUNET_CONFIGURATION_get_value_filename (cfg,
393 (0 != STAT (config, &sbuf)))
395 GNUNET_log (GNUNET_ERROR_TYPE_WARNING,
396 _("Configuration file `%s' for service `%s' not known!\n"),
397 config, servicename);
398 signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_DOWN);
399 GNUNET_free (binary);
400 GNUNET_free (config);
403 sl = GNUNET_malloc (sizeof (struct ServiceList));
404 sl->name = GNUNET_strdup (servicename);
409 sl->mtime = sbuf.st_mtime;
412 signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_UP);
417 free_and_signal (void *cls, struct ServiceList *pos)
419 struct GNUNET_SERVER_Client *client = cls;
420 /* find_name will remove "pos" from the list! */
421 GNUNET_assert (pos == find_name (pos->name));
422 GNUNET_log (GNUNET_ERROR_TYPE_INFO, "Service `%s' stopped\n", pos->name);
423 signal_result (client, pos->name, GNUNET_MESSAGE_TYPE_ARM_IS_DOWN);
424 GNUNET_SERVER_receive_done (client, GNUNET_OK);
425 GNUNET_SERVER_client_drop (client);
431 * Stop the specified service.
434 stop_service (struct GNUNET_SERVER_Client *client, const char *servicename)
436 struct ServiceList *pos;
437 struct GNUNET_CLIENT_Connection *sc;
439 GNUNET_log (GNUNET_ERROR_TYPE_INFO,
440 "Preparing to stop `%s'\n", servicename);
441 pos = find_name (servicename);
442 if ((pos != NULL) && (pos->kill_continuation != NULL))
444 /* killing already in progress */
445 signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_DOWN);
448 if ((pos != NULL) && (pos->rc > 1))
450 /* RC>1, just decrement RC */
454 signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_UP);
455 GNUNET_SERVER_receive_done (client, GNUNET_OK);
460 if (0 != PLIBC_KILL (pos->pid, SIGTERM))
461 GNUNET_log_strerror (GNUNET_ERROR_TYPE_WARNING, "kill");
464 pos->kill_continuation = &free_and_signal;
465 pos->kill_continuation_cls = client;
466 GNUNET_SERVER_client_keep (client);
470 sc = GNUNET_CLIENT_connect (sched, servicename, cfg);
471 GNUNET_CLIENT_service_shutdown (sc);
472 GNUNET_CLIENT_disconnect (sc);
473 signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_DOWN);
474 GNUNET_SERVER_receive_done (client, GNUNET_OK);
480 * Handle START-message.
482 * @param cls closure (always NULL)
483 * @param client identification of the client
484 * @param message the actual message
485 * @return GNUNET_OK to keep the connection open,
486 * GNUNET_SYSERR to close it (signal serious error)
489 handle_start (void *cls,
490 struct GNUNET_SERVER_Client *client,
491 const struct GNUNET_MessageHeader *message)
493 const char *servicename;
496 size = ntohs (message->size);
497 size -= sizeof (struct GNUNET_MessageHeader);
498 servicename = (const char *) &message[1];
499 if ((size == 0) || (servicename[size - 1] != '\0'))
502 GNUNET_SERVER_receive_done (client, GNUNET_SYSERR);
505 start_service (client, servicename);
506 GNUNET_SERVER_receive_done (client, GNUNET_OK);
511 * Handle STOP-message.
513 * @param cls closure (always NULL)
514 * @param client identification of the client
515 * @param message the actual message
516 * @return GNUNET_OK to keep the connection open,
517 * GNUNET_SYSERR to close it (signal serious error)
520 handle_stop (void *cls,
521 struct GNUNET_SERVER_Client *client,
522 const struct GNUNET_MessageHeader *message)
524 const char *servicename;
527 size = ntohs (message->size);
528 size -= sizeof (struct GNUNET_MessageHeader);
529 servicename = (const char *) &message[1];
530 if ((size == 0) || (servicename[size - 1] != '\0'))
533 GNUNET_SERVER_receive_done (client, GNUNET_SYSERR);
536 stop_service (client, servicename);
542 * Background task doing maintenance.
548 maint (void *cls, const struct GNUNET_SCHEDULER_TaskContext *tc)
550 struct ServiceList *pos;
557 if (0 != (tc->reason & GNUNET_SCHEDULER_REASON_SHUTDOWN))
559 GNUNET_log (GNUNET_ERROR_TYPE_INFO, _("Stopping all services\n"));
560 while (NULL != (pos = running))
563 if (0 != PLIBC_KILL (pos->pid, SIGTERM))
564 GNUNET_log_strerror (GNUNET_ERROR_TYPE_WARNING, "kill");
565 if (pos->pid != waitpid (pos->pid, NULL, 0))
566 GNUNET_log_strerror (GNUNET_ERROR_TYPE_WARNING, "waitpid");
571 GNUNET_SCHEDULER_add_delayed (tc->sched,
573 GNUNET_SCHEDULER_PRIORITY_IDLE,
574 GNUNET_SCHEDULER_NO_PREREQUISITE_TASK,
575 MAINT_FREQUENCY, &maint, cfg);
577 /* check for services that died (WAITPID) */
578 while (0 < (pid = waitpid (0, &status, WNOHANG)))
580 if (WIFSTOPPED (status) || WIFCONTINUED (status))
582 pos = find_pid (pid);
585 /* we killed the service */
588 if (WIFEXITED (status))
590 statstr = _( /* process termination method */ "exit");
591 statcode = WEXITSTATUS (status);
593 else if (WTERMSIG (status))
595 statstr = _( /* process termination method */ "signal");
596 statcode = WTERMSIG (status);
600 statstr = _( /* process termination method */ "unknown");
603 if (NULL != pos->kill_continuation)
605 pos->kill_continuation (pos->kill_continuation_cls, pos);
609 GNUNET_log (GNUNET_ERROR_TYPE_WARNING,
611 ("Service `%s' terminated with status %s/%d, will try to restart it!\n"),
612 pos->name, statstr, statcode);
613 /* schedule restart */
618 /* check for services that need to be restarted due to
619 configuration changes or because the last restart failed */
623 if ((0 == STAT (pos->config, &sbuf)) && (pos->mtime < sbuf.st_mtime))
625 GNUNET_log (GNUNET_ERROR_TYPE_INFO,
627 ("Restarting service `%s' due to configuration file change.\n"));
628 if (0 != PLIBC_KILL (pos->pid, SIGTERM))
629 GNUNET_log_strerror (GNUNET_ERROR_TYPE_WARNING, "kill");
633 GNUNET_log (GNUNET_ERROR_TYPE_INFO,
634 _("Restarting service `%s'.\n"), pos->name);
635 /* FIXME: should have some exponentially
636 increasing timer to avoid tight restart loops */
645 * List of handlers for the messages understood by this
648 static struct GNUNET_SERVER_MessageHandler handlers[] = {
649 {&handle_start, NULL, GNUNET_MESSAGE_TYPE_ARM_START, 0},
650 {&handle_stop, NULL, GNUNET_MESSAGE_TYPE_ARM_STOP, 0},
656 * Process arm requests.
659 * @param s scheduler to use
660 * @param server the initialized server
661 * @param c configuration to use
665 struct GNUNET_SCHEDULER_Handle *s,
666 struct GNUNET_SERVER_Handle *server,
667 struct GNUNET_CONFIGURATION_Handle *c)
669 char *defaultservices;
672 GNUNET_log (GNUNET_ERROR_TYPE_INFO, "Starting...\n");
676 GNUNET_CONFIGURATION_get_value_string (cfg,
680 prefix_command = GNUNET_strdup ("");
681 /* start default services... */
683 GNUNET_CONFIGURATION_get_value_string (cfg,
689 GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
690 "Starting default services `%s'\n", defaultservices);
692 pos = strtok (defaultservices, " ");
695 start_service (NULL, pos);
696 pos = strtok (NULL, " ");
698 GNUNET_free (defaultservices);
703 GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
704 "No default services configured.\n");
708 /* process client requests */
709 GNUNET_SERVER_add_handlers (server, handlers);
711 /* manage services */
712 GNUNET_SCHEDULER_add_delayed (sched,
714 GNUNET_SCHEDULER_PRIORITY_IDLE,
715 GNUNET_SCHEDULER_NO_PREREQUISITE_TASK,
716 MAINT_FREQUENCY, &maint, NULL);
721 * The main function for the arm service.
723 * @param argc number of arguments from the command line
724 * @param argv command line arguments
725 * @return 0 ok, 1 on error
728 main (int argc, char *const *argv)
731 GNUNET_SERVICE_run (argc,
732 argv, "arm", &run, NULL, NULL, NULL)) ? 0 : 1;
735 /* end of gnunet-service-arm.c */