/*
This file is part of GNUnet.
- (C) 2009 Christian Grothoff (and other contributing authors)
+ (C) 2009, 2010 Christian Grothoff (and other contributing authors)
GNUnet is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published
- by the Free Software Foundation; either version 2, or (at your
+ by the Free Software Foundation; either version 3, or (at your
option) any later version.
GNUnet is distributed in the hope that it will be useful, but
* @author Christian Grothoff
*
* TODO:
- * - multiple start-stop requests with RC>1 can result
- * in UP/DOWN signals based on "pending" that are inaccurate...
- * => have list of clients waiting for a resolution instead of
- * giving instant (but incorrect) replies
- * - code could go into restart-loop for a service
- * if service crashes instantly -- need exponential back-off
* - need to test auto-restart code on configuration changes;
* - should refine restart code to check if *relevant* parts of the
* configuration were changed (anything in the section for the service)
#include "gnunet_os_lib.h"
#include "gnunet_protocols.h"
#include "gnunet_service_lib.h"
+#include "gnunet_signal_lib.h"
+#include "gnunet-service-arm.h"
#include "arm.h"
/**
- * Run normal maintenance every 2s.
+ * Check for configuration file changes every 5s.
*/
-#define MAINT_FREQUENCY GNUNET_TIME_relative_multiply (GNUNET_TIME_UNIT_SECONDS, 2)
+#define MAINT_FREQUENCY GNUNET_TIME_relative_multiply (GNUNET_TIME_UNIT_SECONDS, 5)
/**
- * Run fast maintenance after 100ms. This is used for an extra-job
- * that is run to check for a process that we just killed.
+ * Threshold after which exponential backoff shouldn't increase (in ms); 30m
*/
-#define MAINT_FAST_FREQUENCY GNUNET_TIME_relative_multiply (GNUNET_TIME_UNIT_MILLISECONDS, 100)
+#define EXPONENTIAL_BACKOFF_THRESHOLD (1000 * 60 * 30)
-/**
- * How long do we wait until we decide that a service
- * did not start?
- */
-#define CHECK_TIMEOUT GNUNET_TIME_UNIT_MINUTES
+#define DELAY_SHUTDOWN GNUNET_NO
/**
* List of our services.
*/
struct ServiceList;
-/**
- * Function to call if waitpid informs us that
- * a process has died.
- *
- * @param cls closure
- * @param pos entry in the service list of the process that died
- */
-typedef void (*CleanCallback) (void *cls, struct ServiceList * pos);
/**
* List of our services.
char *config;
/**
- * Function to call upon kill completion (waitpid), NULL
+ * Client to notify upon kill completion (waitpid), NULL
* if we should simply restart the process.
*/
- CleanCallback kill_continuation;
-
- /**
- * Closure for kill_continuation.
- */
- void *kill_continuation_cls;
+ struct GNUNET_SERVER_Client *killing_client;
/**
* Process ID of the child.
time_t mtime;
/**
- * Reference counter (counts how many times we've been
- * asked to start the service). We only actually stop
- * it once rc hits zero.
+ * Process exponential backoff time
+ */
+ struct GNUNET_TIME_Relative backoff;
+
+ /**
+ * Absolute time at which the process is scheduled to restart in case of death
*/
- unsigned int rc;
+ struct GNUNET_TIME_Absolute restartAt;
};
*/
static char *prefix_command;
+/**
+ * Option to append to each actual command.
+ */
+static char *final_option;
+
+/**
+ * ID of task called whenever we get a SIGCHILD.
+ */
+static GNUNET_SCHEDULER_TaskIdentifier child_death_task;
+
+/**
+ * ID of task called whenever the timeout for restarting a child
+ * expires.
+ */
+static GNUNET_SCHEDULER_TaskIdentifier child_restart_task;
/**
- * Background task doing maintenance.
+ * Context for our SIGCHILD handler.
+ */
+static struct GNUNET_SIGNAL_Context *shc_chld;
+
+/**
+ * Pipe used to communicate shutdown via signal.
+ */
+static struct GNUNET_DISK_PipeHandle *sigpipe;
+
+/**
+ * Reading end of the signal pipe.
+ */
+static const struct GNUNET_DISK_FileHandle *pr;
+
+/**
+ * Are we in shutdown mode?
+ */
+static int in_shutdown;
+
+
+/**
+ * Handle to our server instance. Our server is a bit special in that
+ * its service is not immediately stopped once we get a shutdown
+ * request (since we need to continue service until all of our child
+ * processes are dead). This handle is used to shut down the server
+ * (and thus trigger process termination) once all child processes are
+ * also dead. A special option in the ARM configuration modifies the
+ * behaviour of the service implementation to not do the shutdown
+ * immediately.
+ */
+static struct GNUNET_SERVER_Handle *server;
+
+
+/**
+ * If the configuration file changes, restart tasks that depended on that
+ * option.
*
* @param cls closure, NULL if we need to self-restart
* @param tc context
*/
-static void
-maint (void *cls, const struct GNUNET_SCHEDULER_TaskContext *tc);
+static void
+config_change_task (void *cls, const struct GNUNET_SCHEDULER_TaskContext *tc)
+{
+ struct ServiceList *pos;
+ struct stat sbuf;
+
+ pos = running;
+ while (pos != NULL)
+ {
+ /* FIXME: this test for config change is a bit too coarse grained */
+ if ( (0 == STAT (pos->config, &sbuf)) &&
+ (pos->mtime < sbuf.st_mtime) &&
+ (pos->pid != 0) )
+ {
+ GNUNET_log (GNUNET_ERROR_TYPE_INFO,
+ _("Restarting service `%s' due to configuration file change.\n"));
+ if (0 != PLIBC_KILL (pos->pid, SIGTERM))
+ GNUNET_log_strerror (GNUNET_ERROR_TYPE_WARNING, "kill");
+ else
+ pos->backoff = GNUNET_TIME_UNIT_MILLISECONDS;
+ }
+ pos = pos->next;
+ }
+}
+
/**
struct GNUNET_MessageHeader *msg;
if (buf == NULL)
- return 0; /* error, not much we can do */
+ {
+ GNUNET_log (GNUNET_ERROR_TYPE_WARNING,
+ _("Could not send status result to client\n"));
+ return 0; /* error, not much we can do */
+ }
+#if DEBUG_ARM
+ GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+ "Sending status response %u to client\n",
+ (unsigned int) *res);
+#endif
GNUNET_assert (size >= sizeof (struct GNUNET_MessageHeader));
msg = buf;
msg->size = htons (sizeof (struct GNUNET_MessageHeader));
*/
static void
signal_result (struct GNUNET_SERVER_Client *client,
- const char *name, uint16_t result)
+ const char *name, uint16_t result)
{
uint16_t *res;
if (NULL == client)
- return;
+ {
+ GNUNET_log (GNUNET_ERROR_TYPE_INFO,
+ _
+ ("Not sending status result to client: no client known\n"));
+ return;
+ }
#if DEBUG_ARM
GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "Telling client that service `%s' is now %s\n",
- name,
- result == GNUNET_MESSAGE_TYPE_ARM_IS_DOWN ? "down" : "up");
+ "Telling client that service `%s' is now %s\n",
+ name,
+ result == GNUNET_MESSAGE_TYPE_ARM_IS_DOWN ? "down" : "up");
#endif
res = GNUNET_malloc (sizeof (uint16_t));
*res = result;
GNUNET_SERVER_notify_transmit_ready (client,
- sizeof (struct GNUNET_MessageHeader),
- GNUNET_TIME_UNIT_FOREVER_REL,
- &write_result, res);
+ sizeof (struct GNUNET_MessageHeader),
+ GNUNET_TIME_UNIT_FOREVER_REL,
+ &write_result, res);
}
while (pos != NULL)
{
if (0 == strcmp (pos->name, name))
- {
- if (prev == NULL)
- running = pos->next;
- else
- prev->next = pos->next;
- pos->next = NULL;
- return pos;
- }
+ {
+ if (prev == NULL)
+ running = pos->next;
+ else
+ prev->next = pos->next;
+ pos->next = NULL;
+ return pos;
+ }
prev = pos;
pos = pos->next;
}
GNUNET_free (pos);
}
+#include "do_start_process.c"
/**
* Actually start the process for the given service.
*
* @param sl identifies service to start
+ * @param lsocks -1 terminated list of listen sockets to pass (systemd style), or NULL
*/
static void
-start_process (struct ServiceList *sl)
+start_process (struct ServiceList *sl,
+ const int *lsocks)
{
char *loprefix;
char *options;
- char **argv;
- unsigned int argv_size;
- char *lopos;
char *optpos;
- const char *firstarg;
+ char *optend;
+ const char *next;
int use_debug;
+ char b;
+ char *val;
/* start service */
if (GNUNET_OK !=
GNUNET_CONFIGURATION_get_value_string (cfg,
- sl->name, "PREFIX", &loprefix))
+ sl->name, "PREFIX", &loprefix))
loprefix = GNUNET_strdup (prefix_command);
if (GNUNET_OK !=
GNUNET_CONFIGURATION_get_value_string (cfg,
- sl->name, "OPTIONS", &options))
- options = GNUNET_strdup ("");
+ sl->name, "OPTIONS", &options))
+ {
+ options = GNUNET_strdup (final_option);
+ if (NULL == strstr (options, "%"))
+ {
+ /* replace '{}' with service name */
+ while (NULL != (optpos = strstr (options, "{}")))
+ {
+ optpos[0] = '%';
+ optpos[1] = 's';
+ GNUNET_asprintf (&optpos,
+ options,
+ sl->name);
+ GNUNET_free (options);
+ options = optpos;
+ }
+ /* replace '$PATH' with value associated with "PATH" */
+ while (NULL != (optpos = strstr (options, "$")))
+ {
+ optend = optpos + 1;
+ while (isupper ( (unsigned char) *optend)) optend++;
+ b = *optend;
+ if ('\0' == b)
+ next = "";
+ else
+ next = optend+1;
+ *optend = '\0';
+ if (GNUNET_OK !=
+ GNUNET_CONFIGURATION_get_value_string (cfg, "PATHS",
+ optpos+1,
+ &val))
+ val = GNUNET_strdup ("");
+ *optpos = '\0';
+ GNUNET_asprintf (&optpos,
+ "%s%s%c%s",
+ options,
+ val,
+ b,
+ next);
+ GNUNET_free (options);
+ GNUNET_free (val);
+ options = optpos;
+ }
+ }
+ }
use_debug = GNUNET_CONFIGURATION_get_value_yesno (cfg, sl->name, "DEBUG");
- GNUNET_log (GNUNET_ERROR_TYPE_INFO, _("Starting service `%s'\n"), sl->name);
#if DEBUG_ARM
GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "Starting service `%s' using binary `%s' and configuration `%s'\n",
- sl->name, sl->binary, sl->config);
+ "Starting service `%s' using binary `%s' and configuration `%s'\n",
+ sl->name, sl->binary, sl->config);
#endif
- argv_size = 6;
- if (use_debug)
- argv_size += 2;
- lopos = loprefix;
- while ('\0' != *lopos)
- {
- if (*lopos == ' ')
- argv_size++;
- lopos++;
- }
- optpos = options;
- while ('\0' != *optpos)
- {
- if (*optpos == ' ')
- argv_size++;
- optpos++;
- }
- firstarg = NULL;
- argv = GNUNET_malloc (argv_size * sizeof (char *));
- argv_size = 0;
- lopos = loprefix;
-
- while ('\0' != *lopos)
- {
- while (*lopos == ' ')
- lopos++;
- if (*lopos == '\0')
- continue;
- if (argv_size == 0)
- firstarg = lopos;
- argv[argv_size++] = lopos;
- while (('\0' != *lopos) && (' ' != *lopos))
- lopos++;
- if ('\0' == *lopos)
- continue;
- *lopos = '\0';
- lopos++;
- }
- if (argv_size == 0)
- firstarg = sl->binary;
- argv[argv_size++] = sl->binary;
- argv[argv_size++] = "-c";
- argv[argv_size++] = sl->config;
if (GNUNET_YES == use_debug)
- {
- argv[argv_size++] = "-L";
- argv[argv_size++] = "DEBUG";
- }
- optpos = options;
- while ('\0' != *optpos)
- {
- while (*optpos == ' ')
- optpos++;
- if (*optpos == '\0')
- continue;
- argv[argv_size++] = optpos;
- while (('\0' != *optpos) && (' ' != *optpos))
- optpos++;
- if ('\0' == *optpos)
- continue;
- *optpos = '\0';
- optpos++;
- }
- argv[argv_size++] = NULL;
- sl->pid = GNUNET_OS_start_process_v (firstarg, argv);
- GNUNET_free (argv);
+ sl->pid = do_start_process (lsocks,
+ loprefix,
+ sl->binary,
+ "-c", sl->config,
+ "-L", "DEBUG",
+ options,
+ NULL);
+ else
+ sl->pid = do_start_process (lsocks,
+ loprefix,
+ sl->binary,
+ "-c", sl->config,
+ options,
+ NULL);
+ GNUNET_log (GNUNET_ERROR_TYPE_INFO,
+ _("Starting service `%s' (PID: %d)\n"),
+ sl->name,
+ (int) sl->pid);
GNUNET_free (loprefix);
GNUNET_free (options);
+ /* FIXME: should check sl->pid */
}
*
* @param client who is asking for this
* @param servicename name of the service to start
+ * @param lsocks -1 terminated list of listen sockets to pass (systemd style), or NULL
+ * @return GNUNET_OK on success, GNUNET_SYSERR on error
*/
-static void
-start_service (struct GNUNET_SERVER_Client *client, const char *servicename)
+int
+start_service (struct GNUNET_SERVER_Client *client,
+ const char *servicename,
+ const int *lsocks)
{
struct ServiceList *sl;
char *binary;
char *config;
struct stat sbuf;
+
+ if (GNUNET_YES == in_shutdown)
+ {
+ GNUNET_log (GNUNET_ERROR_TYPE_INFO,
+ _("ARM is shutting down, service `%s' not started.\n"),
+ servicename);
+ signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_DOWN);
+ return GNUNET_SYSERR;
+ }
sl = find_name (servicename);
if (sl != NULL)
{
- /* already running, just increment RC */
- GNUNET_log (GNUNET_ERROR_TYPE_INFO,
- _("Service `%s' already running.\n"), servicename);
- sl->rc++;
+ GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+ _("Service `%s' already running.\n"), servicename);
sl->next = running;
running = sl;
signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_UP);
- return;
+ return GNUNET_SYSERR;
}
if (GNUNET_OK !=
GNUNET_CONFIGURATION_get_value_string (cfg,
- servicename, "BINARY", &binary))
+ servicename, "BINARY", &binary))
{
GNUNET_log (GNUNET_ERROR_TYPE_WARNING,
- _("Binary implementing service `%s' not known!\n"),
- servicename);
+ _("Binary implementing service `%s' not known!\n"),
+ servicename);
signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_DOWN);
- return;
+ return GNUNET_SYSERR;
}
if ((GNUNET_OK !=
GNUNET_CONFIGURATION_get_value_filename (cfg,
- servicename,
- "CONFIG",
- &config)) ||
+ servicename,
+ "CONFIG",
+ &config)) ||
(0 != STAT (config, &sbuf)))
{
GNUNET_log (GNUNET_ERROR_TYPE_WARNING,
- _("Configuration file `%s' for service `%s' not known!\n"),
- config, servicename);
+ _("Configuration file `%s' for service `%s' not known!\n"),
+ config, servicename);
signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_DOWN);
GNUNET_free (binary);
- GNUNET_free (config);
- return;
+ GNUNET_free_non_null (config);
+ return GNUNET_SYSERR;
}
+ (void) stop_listening (servicename);
sl = GNUNET_malloc (sizeof (struct ServiceList));
sl->name = GNUNET_strdup (servicename);
sl->next = running;
- sl->rc = 1;
sl->binary = binary;
sl->config = config;
sl->mtime = sbuf.st_mtime;
+ sl->backoff = GNUNET_TIME_UNIT_MILLISECONDS;
+ sl->restartAt = GNUNET_TIME_UNIT_FOREVER_ABS;
+
running = sl;
- start_process (sl);
+ start_process (sl, lsocks);
if (NULL != client)
signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_UP);
-}
-
-
-/**
- * Free the given entry in the service list and signal
- * the given client that the service is now down.
- *
- * @param cls pointer to the client ("struct GNUNET_SERVER_Client*")
- * @param pos entry for the service
- */
-static void
-free_and_signal (void *cls, struct ServiceList *pos)
-{
- struct GNUNET_SERVER_Client *client = cls;
- /* find_name will remove "pos" from the list! */
- GNUNET_log (GNUNET_ERROR_TYPE_INFO, "Service `%s' stopped\n", pos->name);
- signal_result (client, pos->name, GNUNET_MESSAGE_TYPE_ARM_IS_DOWN);
- GNUNET_SERVER_receive_done (client, GNUNET_OK);
- GNUNET_SERVER_client_drop (client);
- free_entry (pos);
+ return GNUNET_OK;
}
const char *servicename)
{
struct ServiceList *pos;
- struct GNUNET_CLIENT_Connection *sc;
- unsigned long long port;
GNUNET_log (GNUNET_ERROR_TYPE_INFO,
- _("Preparing to stop `%s'\n"), servicename);
+ _("Preparing to stop `%s'\n"), servicename);
pos = find_name (servicename);
- if ((pos != NULL) && (pos->kill_continuation != NULL))
+ if (pos == NULL)
+ {
+ if (GNUNET_OK == stop_listening (servicename))
+ signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_DOWN);
+ else
+ signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_UNKNOWN);
+ GNUNET_SERVER_receive_done (client, GNUNET_OK);
+ return;
+ }
+ if (pos->killing_client != NULL)
{
/* killing already in progress */
#if DEBUG_ARM
"Service `%s' is already down\n", servicename);
#endif
signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_DOWN);
- return;
- }
- if ((pos != NULL) && (pos->rc > 1))
- {
- /* RC>1, just decrement RC */
- pos->rc--;
+ GNUNET_SERVER_receive_done (client, GNUNET_OK);
pos->next = running;
running = pos;
-#if DEBUG_ARM
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "Service `%s' still used by %u clients, will keep it running!\n",
- servicename,
- pos->rc);
-#endif
- signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_UP);
- GNUNET_SERVER_receive_done (client, GNUNET_OK);
return;
}
- if (pos != NULL)
+
+ if (GNUNET_YES == in_shutdown)
{
#if DEBUG_ARM
GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "Sending kill signal to service `%s', waiting for process to die.\n",
+ "Termination request already sent to `%s' (since ARM is in shutdown).\n",
servicename);
#endif
- if (0 != PLIBC_KILL (pos->pid, SIGTERM))
- GNUNET_log_strerror (GNUNET_ERROR_TYPE_WARNING, "kill");
+ signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_DOWN);
+ GNUNET_SERVER_receive_done (client, GNUNET_OK);
pos->next = running;
running = pos;
- pos->kill_continuation = &free_and_signal;
- pos->kill_continuation_cls = client;
- GNUNET_SERVER_client_keep (client);
- GNUNET_SCHEDULER_add_delayed (sched,
- GNUNET_YES,
- GNUNET_SCHEDULER_PRIORITY_IDLE,
- GNUNET_SCHEDULER_NO_TASK,
- MAINT_FAST_FREQUENCY, &maint, NULL);
+ return;
}
- else
+ if (pos->pid == 0)
{
-#if DEBUG_ARM
- GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "Sending termination request to service `%s'.\n",
- servicename);
-#endif
- if ( (GNUNET_OK ==
- GNUNET_CONFIGURATION_get_value_number (cfg,
- servicename,
- "PORT",
- &port)) &&
- (NULL != (sc = GNUNET_CLIENT_connect (sched, servicename, cfg))) )
- {
- GNUNET_CLIENT_service_shutdown (sc);
- signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_DOWN);
- }
- else
- {
- signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_UNKNOWN);
- }
+ /* process is in delayed restart, simply remove it! */
+ free_entry (pos);
+ signal_result (client, servicename, GNUNET_MESSAGE_TYPE_ARM_IS_DOWN);
GNUNET_SERVER_receive_done (client, GNUNET_OK);
+ return;
}
+#if DEBUG_ARM
+ GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+ "Sending kill signal to service `%s', waiting for process to die.\n",
+ servicename);
+#endif
+ if (0 != PLIBC_KILL (pos->pid, SIGTERM))
+ GNUNET_log_strerror (GNUNET_ERROR_TYPE_WARNING, "kill");
+ pos->next = running;
+ running = pos;
+ pos->killing_client = client;
+ GNUNET_SERVER_client_keep (client);
}
*/
static void
handle_start (void *cls,
- struct GNUNET_SERVER_Client *client,
- const struct GNUNET_MessageHeader *message)
+ struct GNUNET_SERVER_Client *client,
+ const struct GNUNET_MessageHeader *message)
{
const char *servicename;
uint16_t size;
GNUNET_SERVER_receive_done (client, GNUNET_SYSERR);
return;
}
- start_service (client, servicename);
+ start_service (client, servicename, NULL);
GNUNET_SERVER_receive_done (client, GNUNET_OK);
}
*/
static void
handle_stop (void *cls,
- struct GNUNET_SERVER_Client *client,
- const struct GNUNET_MessageHeader *message)
+ struct GNUNET_SERVER_Client *client,
+ const struct GNUNET_MessageHeader *message)
{
const char *servicename;
uint16_t size;
/**
- * Background task doing maintenance.
+ * Remove all entries for tasks that are not running
+ * (pid = 0) from the running list (they will no longer
+ * be restarted since we are shutting down).
+ */
+static void
+clean_up_running ()
+{
+ struct ServiceList *pos;
+ struct ServiceList *next;
+ struct ServiceList *prev;
+
+ pos = running;
+ prev = NULL;
+ while (NULL != pos)
+ {
+ next = pos->next;
+ if (pos->pid == 0)
+ {
+ if (prev == NULL)
+ running = next;
+ else
+ prev->next = next;
+ free_entry (pos);
+ }
+ else
+ prev = pos;
+ pos = next;
+ }
+}
+
+
+/**
+ * We are done with everything. Stop remaining
+ * tasks, signal handler and the server.
+ */
+static void
+do_shutdown ()
+{
+ GNUNET_SERVER_destroy (server);
+ server = NULL;
+ GNUNET_SIGNAL_handler_uninstall (shc_chld);
+ shc_chld = NULL;
+ GNUNET_SCHEDULER_cancel (sched, child_death_task);
+ child_death_task = GNUNET_SCHEDULER_NO_TASK;
+}
+
+#if DELAY_SHUTDOWN
+/**
+ * Dummy task to delay arm shutdown.
+ */
+void dummy_task (void *cls,
+ const struct GNUNET_SCHEDULER_TaskContext * tc)
+{
+ GNUNET_log(GNUNET_ERROR_TYPE_WARNING, "Dummy task executing\n");
+ return;
+}
+#endif
+
+/**
+ * Task run for shutdown.
*
* @param cls closure, NULL if we need to self-restart
* @param tc context
*/
static void
-maint (void *cls, const struct GNUNET_SCHEDULER_TaskContext *tc)
+shutdown_task (void *cls, const struct GNUNET_SCHEDULER_TaskContext *tc)
+{
+ struct ServiceList *pos;
+
+#if DEBUG_ARM
+ GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, _("Stopping all services\n"));
+#endif
+ stop_listening (NULL);
+ in_shutdown = GNUNET_YES;
+ pos = running;
+ while (NULL != pos)
+ {
+ if (pos->pid != 0)
+ {
+ GNUNET_log (GNUNET_ERROR_TYPE_INFO,
+ "Stopping service `%s' (PID: %d)\n",
+ pos->name,
+ pos->pid);
+ if (0 != PLIBC_KILL (pos->pid, SIGTERM))
+ GNUNET_log_strerror (GNUNET_ERROR_TYPE_WARNING, "kill");
+ }
+ pos = pos->next;
+ }
+#if DELAY_SHUTDOWN
+ GNUNET_SCHEDULER_add_delayed(sched, GNUNET_TIME_relative_multiply(GNUNET_TIME_UNIT_SECONDS, 2), &dummy_task, NULL);
+#endif
+ if (running == NULL)
+ do_shutdown ();
+}
+
+
+/**
+ * Task run whenever it is time to restart a child that died.
+ *
+ * @param cls closure, always NULL
+ * @param tc context
+ */
+static void
+delayed_restart_task (void *cls, const struct GNUNET_SCHEDULER_TaskContext *tc)
+{
+ struct ServiceList *pos;
+ struct GNUNET_TIME_Relative lowestRestartDelay;
+
+ child_restart_task = GNUNET_SCHEDULER_NO_TASK;
+ if (0 != (tc->reason & GNUNET_SCHEDULER_REASON_SHUTDOWN))
+ {
+ clean_up_running ();
+ if (NULL == running)
+ do_shutdown ();
+ return;
+ }
+ lowestRestartDelay = GNUNET_TIME_UNIT_FOREVER_REL;
+
+ /* check for services that need to be restarted due to
+ configuration changes or because the last restart failed */
+ pos = running;
+ while (pos != NULL)
+ {
+ if ( (pos->pid == 0) &&
+ (GNUNET_YES != in_shutdown) )
+ {
+ if (GNUNET_TIME_absolute_get_remaining (pos->restartAt).value == 0)
+ {
+ GNUNET_log (GNUNET_ERROR_TYPE_INFO,
+ _("Restarting service `%s'.\n"), pos->name);
+ start_process (pos, NULL);
+ }
+ else
+ {
+ lowestRestartDelay
+ = GNUNET_TIME_relative_min (lowestRestartDelay,
+ GNUNET_TIME_absolute_get_remaining
+ (pos->restartAt));
+ }
+ }
+ pos = pos->next;
+ }
+ if (lowestRestartDelay.value != GNUNET_TIME_UNIT_FOREVER_REL.value)
+ {
+#if DEBUG_ARM
+ GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+ "Will restart process in %llums\n",
+ (unsigned long long) lowestRestartDelay.value);
+#endif
+ child_restart_task
+ = GNUNET_SCHEDULER_add_delayed (sched,
+ lowestRestartDelay,
+ &delayed_restart_task,
+ NULL);
+ }
+}
+
+
+/**
+ * Task triggered whenever we receive a SIGCHLD (child
+ * process died).
+ *
+ * @param cls closure, NULL if we need to self-restart
+ * @param tc context
+ */
+static void
+maint_child_death (void *cls, const struct GNUNET_SCHEDULER_TaskContext *tc)
{
struct ServiceList *pos;
struct ServiceList *prev;
struct ServiceList *next;
const char *statstr;
int statcode;
- struct stat sbuf;
int ret;
+ char c[16];
+ enum GNUNET_OS_ProcessStatusType statusType;
+ unsigned long statusCode;
- if (0 != (tc->reason & GNUNET_SCHEDULER_REASON_SHUTDOWN))
+ child_death_task = GNUNET_SCHEDULER_NO_TASK;
+ if (0 == (tc->reason & GNUNET_SCHEDULER_REASON_READ_READY))
{
- GNUNET_log (GNUNET_ERROR_TYPE_INFO, _("Stopping all services\n"));
- while (NULL != (pos = running))
- {
- running = pos->next;
- if (0 != PLIBC_KILL (pos->pid, SIGTERM))
- GNUNET_log_strerror (GNUNET_ERROR_TYPE_WARNING, "kill");
- if (GNUNET_OK != GNUNET_OS_process_wait(pos->pid))
- GNUNET_log_strerror (GNUNET_ERROR_TYPE_WARNING, "waitpid");
- free_entry (pos);
- }
- return;
+ child_death_task =
+ GNUNET_SCHEDULER_add_read_file (sched, GNUNET_TIME_UNIT_FOREVER_REL, pr,
+ &maint_child_death, NULL);
+ return;
}
- if (cls == NULL)
- GNUNET_SCHEDULER_add_delayed (tc->sched,
- GNUNET_YES,
- GNUNET_SCHEDULER_PRIORITY_IDLE,
- GNUNET_SCHEDULER_NO_TASK,
- MAINT_FREQUENCY, &maint, NULL);
+ /* consume the signal */
+ GNUNET_break (0 < GNUNET_DISK_file_read (pr, &c, sizeof (c)));
/* check for services that died (WAITPID) */
prev = NULL;
next = running;
while (NULL != (pos = next))
{
- enum GNUNET_OS_ProcessStatusType statusType;
- unsigned long statusCode;
-
next = pos->next;
- if (pos->pid == 0)
+ if (pos->pid == 0)
{
- if (NULL != pos->kill_continuation)
- {
- if (prev == NULL)
- running = next;
- else
- prev->next = next;
- pos->kill_continuation (pos->kill_continuation_cls, pos);
- }
+ prev = pos;
continue;
}
- if ( (GNUNET_SYSERR == (ret = GNUNET_OS_process_status(pos->pid,
+ if ((GNUNET_SYSERR == (ret = GNUNET_OS_process_status (pos->pid,
&statusType,
&statusCode))) ||
- ( (ret == GNUNET_NO) ||
- (statusType == GNUNET_OS_PROCESS_STOPPED) ||
- (statusType == GNUNET_OS_PROCESS_RUNNING) ) )
+ ( (ret == GNUNET_NO) ||
+ (statusType == GNUNET_OS_PROCESS_STOPPED) ||
+ (statusType == GNUNET_OS_PROCESS_RUNNING)) )
{
prev = pos;
continue;
}
+
if (statusType == GNUNET_OS_PROCESS_EXITED)
{
statstr = _( /* process termination method */ "exit");
{
statstr = _( /* process termination method */ "unknown");
statcode = 0;
- }
- if (NULL != pos->kill_continuation)
- {
+ }
+ pos->pid = 0;
+ if (NULL != pos->killing_client)
+ {
if (prev == NULL)
running = next;
else
prev->next = next;
- pos->kill_continuation (pos->kill_continuation_cls, pos);
+ GNUNET_log (GNUNET_ERROR_TYPE_INFO,
+ _("Service `%s' stopped\n"),
+ pos->name);
+ signal_result (pos->killing_client,
+ pos->name, GNUNET_MESSAGE_TYPE_ARM_IS_DOWN);
+ GNUNET_SERVER_receive_done (pos->killing_client, GNUNET_OK);
+ GNUNET_SERVER_client_drop (pos->killing_client);
+ free_entry (pos);
continue;
- }
- GNUNET_log (GNUNET_ERROR_TYPE_WARNING,
- _("Service `%s' terminated with status %s/%d, will try to restart it!\n"),
- pos->name, statstr, statcode);
- /* schedule restart */
- pos->pid = 0;
+ }
+ if (GNUNET_YES != in_shutdown)
+ {
+ GNUNET_log (GNUNET_ERROR_TYPE_WARNING,
+ _
+ ("Service `%s' terminated with status %s/%d, will try to restart it!\n"),
+ pos->name, statstr, statcode);
+ /* schedule restart */
+ pos->restartAt
+ = GNUNET_TIME_relative_to_absolute (pos->backoff);
+ if (pos->backoff.value < EXPONENTIAL_BACKOFF_THRESHOLD)
+ pos->backoff
+ = GNUNET_TIME_relative_multiply (pos->backoff, 2);
+ if (GNUNET_SCHEDULER_NO_TASK != child_restart_task)
+ GNUNET_SCHEDULER_cancel (sched, child_restart_task);
+ child_restart_task
+ = GNUNET_SCHEDULER_add_with_priority (sched,
+ GNUNET_SCHEDULER_PRIORITY_IDLE,
+ &delayed_restart_task,
+ NULL);
+ }
+#if DEBUG_ARM
+ else
+ GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+ "Service `%s' terminated with status %s/%d\n",
+ pos->name, statstr, statcode);
+#endif
prev = pos;
}
+ if (in_shutdown)
+ clean_up_running ();
+ if ( (running == NULL) &&
+ (in_shutdown) )
+ {
+ GNUNET_SERVER_destroy (server);
+ GNUNET_SIGNAL_handler_uninstall (shc_chld);
+ shc_chld = NULL;
+ }
+ else
+ {
+ child_death_task =
+ GNUNET_SCHEDULER_add_read_file (sched, GNUNET_TIME_UNIT_FOREVER_REL, pr,
+ &maint_child_death, NULL);
+ }
+}
- /* check for services that need to be restarted due to
- configuration changes or because the last restart failed */
- pos = running;
- while (pos != NULL)
+
+static size_t
+transmit_shutdown_ack (void *cls, size_t size, void *buf)
+{
+ struct GNUNET_SERVER_Client *client = cls;
+ struct GNUNET_MessageHeader *msg;
+
+ if (size < sizeof (struct GNUNET_MessageHeader))
{
- if ((0 == STAT (pos->config, &sbuf)) && (pos->mtime < sbuf.st_mtime))
- {
- GNUNET_log (GNUNET_ERROR_TYPE_INFO,
- _("Restarting service `%s' due to configuration file change.\n"));
- if (0 != PLIBC_KILL (pos->pid, SIGTERM))
- GNUNET_log_strerror (GNUNET_ERROR_TYPE_WARNING, "kill");
- }
- if (pos->pid == 0)
- {
- GNUNET_log (GNUNET_ERROR_TYPE_INFO,
- _("Restarting service `%s'.\n"), pos->name);
- /* FIXME: should have some exponentially
- increasing timer to avoid tight restart loops */
- start_process (pos);
- }
- pos = pos->next;
+ GNUNET_log (GNUNET_ERROR_TYPE_INFO,
+ _("Failed to transmit shutdown ACK.\n"));
+ GNUNET_SERVER_receive_done (client, GNUNET_SYSERR);
+ return 0; /* client disconnected */
}
+
+ GNUNET_log (GNUNET_ERROR_TYPE_INFO,
+ _("Transmitting shutdown ACK.\n"));
+
+ msg = (struct GNUNET_MessageHeader *) buf;
+ msg->type = htons (GNUNET_MESSAGE_TYPE_ARM_SHUTDOWN_ACK);
+ msg->size = htons (sizeof (struct GNUNET_MessageHeader));
+ GNUNET_SERVER_receive_done (client, GNUNET_OK);
+ GNUNET_SERVER_client_drop(client);
+ return sizeof (struct GNUNET_MessageHeader);
+}
+
+/**
+ * Handler for SHUTDOWN message.
+ *
+ * @param cls closure (refers to service)
+ * @param client identification of the client
+ * @param message the actual message
+ */
+static void
+handle_shutdown (void *cls,
+ struct GNUNET_SERVER_Client *client,
+ const struct GNUNET_MessageHeader *message)
+{
+ GNUNET_SERVER_client_keep(client);
+ GNUNET_log (GNUNET_ERROR_TYPE_INFO,
+ _("Initiating shutdown as requested by client.\n"));
+
+ GNUNET_SERVER_notify_transmit_ready (client,
+ sizeof(struct GNUNET_MessageHeader),
+ GNUNET_TIME_UNIT_FOREVER_REL,
+ &transmit_shutdown_ack, client);
+ GNUNET_SERVER_client_persist_ (client);
+ GNUNET_SCHEDULER_shutdown (sched);
}
/**
- * List of handlers for the messages understood by this service.
+ * Signal handler called for SIGCHLD. Triggers the
+ * respective handler by writing to the trigger pipe.
*/
-static struct GNUNET_SERVER_MessageHandler handlers[] = {
- {&handle_start, NULL, GNUNET_MESSAGE_TYPE_ARM_START, 0},
- {&handle_stop, NULL, GNUNET_MESSAGE_TYPE_ARM_STOP, 0},
- {NULL, NULL, 0, 0}
-};
+static void
+sighandler_child_death ()
+{
+ static char c;
+ int old_errno = errno; /* back-up errno */
+ GNUNET_break (1 ==
+ GNUNET_DISK_file_write (GNUNET_DISK_pipe_handle
+ (sigpipe, GNUNET_DISK_PIPE_END_WRITE), &c,
+ sizeof (c)));
+ errno = old_errno; /* restore errno */
+}
/**
*
* @param cls closure
* @param s scheduler to use
- * @param server the initialized server
+ * @param serv the initialized server
* @param c configuration to use
*/
static void
run (void *cls,
struct GNUNET_SCHEDULER_Handle *s,
- struct GNUNET_SERVER_Handle *server,
+ struct GNUNET_SERVER_Handle *serv,
const struct GNUNET_CONFIGURATION_Handle *c)
{
+ static const struct GNUNET_SERVER_MessageHandler handlers[] = {
+ {&handle_start, NULL, GNUNET_MESSAGE_TYPE_ARM_START, 0},
+ {&handle_stop, NULL, GNUNET_MESSAGE_TYPE_ARM_STOP, 0},
+ {&handle_shutdown, NULL, GNUNET_MESSAGE_TYPE_ARM_SHUTDOWN,
+ sizeof (struct GNUNET_MessageHeader)},
+ {NULL, NULL, 0, 0}
+ };
char *defaultservices;
char *pos;
cfg = c;
sched = s;
+ server = serv;
+ GNUNET_assert (serv != NULL);
+ shc_chld = GNUNET_SIGNAL_handler_install (GNUNET_SIGCHLD, &sighandler_child_death);
+ GNUNET_assert (sigpipe == NULL);
+ sigpipe = GNUNET_DISK_pipe (GNUNET_NO);
+ GNUNET_assert (sigpipe != NULL);
+ pr = GNUNET_DISK_pipe_handle (sigpipe, GNUNET_DISK_PIPE_END_READ);
+ GNUNET_assert (pr != NULL);
+ GNUNET_SERVER_ignore_shutdown (serv, GNUNET_YES);
+ GNUNET_SCHEDULER_add_delayed (sched,
+ GNUNET_TIME_UNIT_FOREVER_REL,
+ &shutdown_task,
+ NULL);
+ child_death_task =
+ GNUNET_SCHEDULER_add_read_file (sched, GNUNET_TIME_UNIT_FOREVER_REL, pr,
+ &maint_child_death, NULL);
+
if (GNUNET_OK !=
GNUNET_CONFIGURATION_get_value_string (cfg,
- "ARM",
- "GLOBAL_PREFIX",
- &prefix_command))
+ "ARM",
+ "GLOBAL_PREFIX",
+ &prefix_command))
prefix_command = GNUNET_strdup ("");
+ if (GNUNET_OK !=
+ GNUNET_CONFIGURATION_get_value_string (cfg,
+ "ARM",
+ "GLOBAL_POSTFIX",
+ &final_option))
+ final_option = GNUNET_strdup ("");
/* start default services... */
if (GNUNET_OK ==
GNUNET_CONFIGURATION_get_value_string (cfg,
- "ARM",
- "DEFAULTSERVICES",
- &defaultservices))
+ "ARM",
+ "DEFAULTSERVICES",
+ &defaultservices))
{
#if DEBUG_ARM
GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "Starting default services `%s'\n", defaultservices);
+ "Starting default services `%s'\n", defaultservices);
#endif
- pos = strtok (defaultservices, " ");
- while (pos != NULL)
- {
- start_service (NULL, pos);
- pos = strtok (NULL, " ");
- }
+ if (0 < strlen (defaultservices))
+ {
+ pos = strtok (defaultservices, " ");
+ while (pos != NULL)
+ {
+ start_service (NULL, pos, NULL);
+ pos = strtok (NULL, " ");
+ }
+ }
GNUNET_free (defaultservices);
}
else
{
#if DEBUG_ARM
GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
- "No default services configured.\n");
+ "No default services configured.\n");
#endif
}
+ /* create listening sockets for future services*/
+ prepareServices (cfg, sched);
+
/* process client requests */
GNUNET_SERVER_add_handlers (server, handlers);
/* manage services */
- GNUNET_SCHEDULER_add_delayed (sched,
- GNUNET_YES,
- GNUNET_SCHEDULER_PRIORITY_IDLE,
- GNUNET_SCHEDULER_NO_TASK,
- MAINT_FREQUENCY, &maint, NULL);
+ GNUNET_SCHEDULER_add_with_priority (sched,
+ GNUNET_SCHEDULER_PRIORITY_IDLE,
+ &config_change_task, NULL);
}
main (int argc, char *const *argv)
{
return (GNUNET_OK ==
- GNUNET_SERVICE_run (argc,
- argv, "arm", &run, NULL, NULL, NULL)) ? 0 : 1;
+ GNUNET_SERVICE_run (argc,
+ argv, "arm", GNUNET_YES, &run, NULL)) ? 0 : 1;
}
/* end of gnunet-service-arm.c */