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 const 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)
173 GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
174 "Telling client that service `%s' is now %s\n",
176 result == GNUNET_MESSAGE_TYPE_ARM_IS_DOWN ? "down" : "up");
178 res = GNUNET_malloc (sizeof (uint16_t));
180 GNUNET_SERVER_notify_transmit_ready (client,
181 sizeof (struct GNUNET_MessageHeader),
182 GNUNET_TIME_UNIT_FOREVER_REL,
188 * Find the process with the given service
189 * name in the given list, remove it and return it.
191 * @return NULL if it was not found
193 static struct ServiceList *
194 find_name (const char *name)
196 struct ServiceList *pos;
197 struct ServiceList *prev;
203 if (0 == strcmp (pos->name, name))
208 prev->next = pos->next;
220 free_entry (struct ServiceList *pos)
222 GNUNET_free_non_null (pos->config);
223 GNUNET_free_non_null (pos->binary);
224 GNUNET_free (pos->name);
232 * Actually start the process for the given service.
234 * @param sl identifies service to start
237 start_process (struct ServiceList *sl)
242 unsigned int argv_size;
245 const char *firstarg;
250 GNUNET_CONFIGURATION_get_value_string (cfg,
251 sl->name, "PREFIX", &loprefix))
252 loprefix = GNUNET_strdup (prefix_command);
254 GNUNET_CONFIGURATION_get_value_string (cfg,
255 sl->name, "OPTIONS", &options))
256 options = GNUNET_strdup ("");
257 use_debug = GNUNET_CONFIGURATION_get_value_yesno (cfg, sl->name, "DEBUG");
259 GNUNET_log (GNUNET_ERROR_TYPE_INFO, _("Starting service `%s'\n"), sl->name);
261 GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
262 "Starting service `%s' using binary `%s' and configuration `%s'\n",
263 sl->name, sl->binary, sl->config);
269 while ('\0' != *lopos)
276 while ('\0' != *optpos)
283 argv = GNUNET_malloc (argv_size * sizeof (char *));
287 while ('\0' != *lopos)
289 while (*lopos == ' ')
295 argv[argv_size++] = lopos;
296 while (('\0' != *lopos) && (' ' != *lopos))
304 firstarg = sl->binary;
305 argv[argv_size++] = sl->binary;
306 argv[argv_size++] = "-c";
307 argv[argv_size++] = sl->config;
308 if (GNUNET_YES == use_debug)
310 argv[argv_size++] = "-L";
311 argv[argv_size++] = "DEBUG";
314 while ('\0' != *optpos)
316 while (*optpos == ' ')
320 argv[argv_size++] = optpos;
321 while (('\0' != *optpos) && (' ' != *optpos))
328 argv[argv_size++] = NULL;
329 sl->pid = GNUNET_OS_start_process_v (firstarg, argv);
331 GNUNET_free (loprefix);
332 GNUNET_free (options);
337 * Start the specified service.
340 start_service (struct GNUNET_SERVER_Client *client, const char *servicename)
342 struct ServiceList *sl;
346 sl = find_name (servicename);
349 /* already running, just increment RC */
350 GNUNET_log (GNUNET_ERROR_TYPE_INFO,
351 _("Service `%s' already running.\n"), servicename);
355 signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_UP);
359 GNUNET_CONFIGURATION_get_value_string (cfg,
360 servicename, "BINARY", &binary))
362 GNUNET_log (GNUNET_ERROR_TYPE_WARNING,
363 _("Binary implementing service `%s' not known!\n"),
365 signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_DOWN);
369 GNUNET_CONFIGURATION_get_value_filename (cfg,
373 (0 != STAT (config, &sbuf)))
375 GNUNET_log (GNUNET_ERROR_TYPE_WARNING,
376 _("Configuration file `%s' for service `%s' not known!\n"),
377 config, servicename);
378 signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_DOWN);
379 GNUNET_free (binary);
380 GNUNET_free (config);
383 sl = GNUNET_malloc (sizeof (struct ServiceList));
384 sl->name = GNUNET_strdup (servicename);
389 sl->mtime = sbuf.st_mtime;
393 signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_UP);
398 free_and_signal (void *cls, struct ServiceList *pos)
400 struct GNUNET_SERVER_Client *client = cls;
401 /* find_name will remove "pos" from the list! */
402 GNUNET_log (GNUNET_ERROR_TYPE_INFO, "Service `%s' stopped\n", pos->name);
403 signal_result (client, pos->name, GNUNET_MESSAGE_TYPE_ARM_IS_DOWN);
404 GNUNET_SERVER_receive_done (client, GNUNET_OK);
405 GNUNET_SERVER_client_drop (client);
411 * Stop the specified service.
414 stop_service (struct GNUNET_SERVER_Client *client, const char *servicename)
416 struct ServiceList *pos;
417 struct GNUNET_CLIENT_Connection *sc;
418 unsigned long long port;
420 GNUNET_log (GNUNET_ERROR_TYPE_INFO,
421 "Preparing to stop `%s'\n", servicename);
422 pos = find_name (servicename);
423 if ((pos != NULL) && (pos->kill_continuation != NULL))
425 /* killing already in progress */
426 signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_DOWN);
429 if ((pos != NULL) && (pos->rc > 1))
431 /* RC>1, just decrement RC */
435 signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_UP);
436 GNUNET_SERVER_receive_done (client, GNUNET_OK);
441 if (0 != PLIBC_KILL (pos->pid, SIGTERM))
442 GNUNET_log_strerror (GNUNET_ERROR_TYPE_WARNING, "kill");
445 pos->kill_continuation = &free_and_signal;
446 pos->kill_continuation_cls = client;
447 GNUNET_SERVER_client_keep (client);
452 GNUNET_CONFIGURATION_get_value_number (cfg,
456 (NULL != (sc = GNUNET_CLIENT_connect (sched, servicename, cfg))) )
458 GNUNET_CLIENT_service_shutdown (sc);
459 GNUNET_CLIENT_disconnect (sc);
460 signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_DOWN);
464 signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_UNKNOWN);
466 GNUNET_SERVER_receive_done (client, GNUNET_OK);
472 * Handle START-message.
474 * @param cls closure (always NULL)
475 * @param client identification of the client
476 * @param message the actual message
477 * @return GNUNET_OK to keep the connection open,
478 * GNUNET_SYSERR to close it (signal serious error)
481 handle_start (void *cls,
482 struct GNUNET_SERVER_Client *client,
483 const struct GNUNET_MessageHeader *message)
485 const char *servicename;
488 size = ntohs (message->size);
489 size -= sizeof (struct GNUNET_MessageHeader);
490 servicename = (const char *) &message[1];
491 if ((size == 0) || (servicename[size - 1] != '\0'))
494 GNUNET_SERVER_receive_done (client, GNUNET_SYSERR);
497 start_service (client, servicename);
498 GNUNET_SERVER_receive_done (client, GNUNET_OK);
503 * Handle STOP-message.
505 * @param cls closure (always NULL)
506 * @param client identification of the client
507 * @param message the actual message
508 * @return GNUNET_OK to keep the connection open,
509 * GNUNET_SYSERR to close it (signal serious error)
512 handle_stop (void *cls,
513 struct GNUNET_SERVER_Client *client,
514 const struct GNUNET_MessageHeader *message)
516 const char *servicename;
519 size = ntohs (message->size);
520 size -= sizeof (struct GNUNET_MessageHeader);
521 servicename = (const char *) &message[1];
522 if ((size == 0) || (servicename[size - 1] != '\0'))
525 GNUNET_SERVER_receive_done (client, GNUNET_SYSERR);
528 stop_service (client, servicename);
534 * Background task doing maintenance.
540 maint (void *cls, const struct GNUNET_SCHEDULER_TaskContext *tc)
542 struct ServiceList *pos;
543 struct ServiceList *prev;
544 struct ServiceList *next;
550 if (0 != (tc->reason & GNUNET_SCHEDULER_REASON_SHUTDOWN))
552 GNUNET_log (GNUNET_ERROR_TYPE_INFO, _("Stopping all services\n"));
553 while (NULL != (pos = running))
556 if (0 != PLIBC_KILL (pos->pid, SIGTERM))
557 GNUNET_log_strerror (GNUNET_ERROR_TYPE_WARNING, "kill");
558 if (GNUNET_OK != GNUNET_OS_process_wait(pos->pid))
559 GNUNET_log_strerror (GNUNET_ERROR_TYPE_WARNING, "waitpid");
564 GNUNET_SCHEDULER_add_delayed (tc->sched,
566 GNUNET_SCHEDULER_PRIORITY_IDLE,
567 GNUNET_SCHEDULER_NO_TASK,
568 MAINT_FREQUENCY, &maint, NULL);
570 /* check for services that died (WAITPID) */
573 while (NULL != (pos = next))
575 enum GNUNET_OS_ProcessStatusType statusType;
576 unsigned long statusCode;
581 if (NULL != pos->kill_continuation)
587 pos->kill_continuation (pos->kill_continuation_cls, pos);
591 if ( (GNUNET_SYSERR == (ret = GNUNET_OS_process_status(pos->pid,
594 ( (ret == GNUNET_NO) ||
595 (statusType == GNUNET_OS_PROCESS_STOPPED) ||
596 (statusType == GNUNET_OS_PROCESS_RUNNING) ) )
601 if (statusType == GNUNET_OS_PROCESS_EXITED)
603 statstr = _( /* process termination method */ "exit");
604 statcode = statusCode;
606 else if (statusType == GNUNET_OS_PROCESS_SIGNALED)
608 statstr = _( /* process termination method */ "signal");
609 statcode = statusCode;
613 statstr = _( /* process termination method */ "unknown");
616 if (NULL != pos->kill_continuation)
622 pos->kill_continuation (pos->kill_continuation_cls, pos);
625 GNUNET_log (GNUNET_ERROR_TYPE_WARNING,
626 _("Service `%s' terminated with status %s/%d, will try to restart it!\n"),
627 pos->name, statstr, statcode);
628 /* schedule restart */
633 /* check for services that need to be restarted due to
634 configuration changes or because the last restart failed */
638 if ((0 == STAT (pos->config, &sbuf)) && (pos->mtime < sbuf.st_mtime))
640 GNUNET_log (GNUNET_ERROR_TYPE_INFO,
641 _("Restarting service `%s' due to configuration file change.\n"));
642 if (0 != PLIBC_KILL (pos->pid, SIGTERM))
643 GNUNET_log_strerror (GNUNET_ERROR_TYPE_WARNING, "kill");
647 GNUNET_log (GNUNET_ERROR_TYPE_INFO,
648 _("Restarting service `%s'.\n"), pos->name);
649 /* FIXME: should have some exponentially
650 increasing timer to avoid tight restart loops */
659 * List of handlers for the messages understood by this
662 static struct GNUNET_SERVER_MessageHandler handlers[] = {
663 {&handle_start, NULL, GNUNET_MESSAGE_TYPE_ARM_START, 0},
664 {&handle_stop, NULL, GNUNET_MESSAGE_TYPE_ARM_STOP, 0},
670 * Process arm requests.
673 * @param s scheduler to use
674 * @param server the initialized server
675 * @param c configuration to use
679 struct GNUNET_SCHEDULER_Handle *s,
680 struct GNUNET_SERVER_Handle *server,
681 const struct GNUNET_CONFIGURATION_Handle *c)
683 char *defaultservices;
686 GNUNET_log (GNUNET_ERROR_TYPE_INFO, "Starting...\n");
690 GNUNET_CONFIGURATION_get_value_string (cfg,
694 prefix_command = GNUNET_strdup ("");
695 /* start default services... */
697 GNUNET_CONFIGURATION_get_value_string (cfg,
703 GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
704 "Starting default services `%s'\n", defaultservices);
706 pos = strtok (defaultservices, " ");
709 start_service (NULL, pos);
710 pos = strtok (NULL, " ");
712 GNUNET_free (defaultservices);
717 GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
718 "No default services configured.\n");
722 /* process client requests */
723 GNUNET_SERVER_add_handlers (server, handlers);
725 /* manage services */
726 GNUNET_SCHEDULER_add_delayed (sched,
728 GNUNET_SCHEDULER_PRIORITY_IDLE,
729 GNUNET_SCHEDULER_NO_TASK,
730 MAINT_FREQUENCY, &maint, NULL);
735 * The main function for the arm service.
737 * @param argc number of arguments from the command line
738 * @param argv command line arguments
739 * @return 0 ok, 1 on error
742 main (int argc, char *const *argv)
745 GNUNET_SERVICE_run (argc,
746 argv, "arm", &run, NULL, NULL, NULL)) ? 0 : 1;
749 /* end of gnunet-service-arm.c */