proper shutdown
[oweals/gnunet.git] / src / util / client.c
index 4d158ab2fbd26a8ce8645540717a409f5b08cd25..7e688059044211cebf40dcea045f51534252b4a0 100644 (file)
 
 #define DEBUG_CLIENT GNUNET_NO
 
-
 /**
  * How often do we re-try tranmsitting requests before giving up?
  * Note that if we succeeded transmitting a request but failed to read
  * a response, we do NOT re-try.
  */
-#define MAX_ATTEMPTS 10
-
+#define MAX_ATTEMPTS 50
 
 /**
  * Handle for a transmission request.
@@ -78,7 +76,7 @@ struct GNUNET_CLIENT_TransmitHandle
   GNUNET_SCHEDULER_TaskIdentifier reconnect_task;
 
   /**
-   * Timeout.
+   * Timeout for the operation overall.
    */
   struct GNUNET_TIME_Absolute timeout;
 
@@ -137,7 +135,6 @@ struct TransmitGetResponseContext
   void *rn_cls;
 };
 
-
 /**
  * Struct to refer to a GNUnet TCP connection.
  * This is more than just a socket because if the server
@@ -152,13 +149,9 @@ struct GNUNET_CLIENT_Connection
    */
   struct GNUNET_CONNECTION_Handle *sock;
 
-  /**
-   * Our scheduler.
-   */
-  struct GNUNET_SCHEDULER_Handle *sched;
-
   /**
    * Our configuration.
+   * FIXME: why do we DUP the configuration? Avoid this!
    */
   struct GNUNET_CONFIGURATION_Handle *cfg;
 
@@ -194,6 +187,11 @@ struct GNUNET_CLIENT_Connection
    */
   GNUNET_SCHEDULER_Task test_cb;
 
+  /**
+   * Deadline for calling 'test_cb'.
+   */
+  struct GNUNET_TIME_Absolute test_deadline;
+
   /**
    * If we are re-trying and are delaying to do so,
    * handle to the scheduled task managing the delay.
@@ -215,6 +213,12 @@ struct GNUNET_CLIENT_Connection
    */
   struct GNUNET_TIME_Absolute receive_timeout;
 
+  /**
+   * Current value for our incremental back-off (for
+   * connect re-tries).
+   */
+  struct GNUNET_TIME_Relative back_off;
+
   /**
    * Number of bytes in received_buf that are valid.
    */
@@ -236,18 +240,62 @@ struct GNUNET_CLIENT_Connection
    */
   int in_receive;
 
+  /**
+   * Are we ignoring shutdown signals?
+   */
+  int ignore_shutdown;
+  
+  /**
+   * How often have we tried to connect?
+   */
+  unsigned int attempts;
+
 };
 
 
+/**
+ * Try to connect to the service.
+ *
+ * @param service_name name of service to connect to
+ * @param cfg configuration to use
+ * @param attempt counter used to alternate between IP and UNIX domain sockets
+ * @return NULL on error
+ */
 static struct GNUNET_CONNECTION_Handle *
-do_connect (struct GNUNET_SCHEDULER_Handle *sched,
-            const char *service_name,
-            const struct GNUNET_CONFIGURATION_Handle *cfg)
+do_connect (const char *service_name,
+            const struct GNUNET_CONFIGURATION_Handle *cfg,
+           unsigned int attempt)
 {
   struct GNUNET_CONNECTION_Handle *sock;
   char *hostname;
+  char *unixpath;
   unsigned long long port;
 
+  sock = NULL;
+#if AF_UNIX
+  if (0 == (attempt % 2))
+    {
+      /* on even rounds, try UNIX */
+      if ((GNUNET_OK ==
+         GNUNET_CONFIGURATION_get_value_string (cfg,
+                                                service_name,
+                                                "UNIXPATH", &unixpath)) &&
+          (0 < strlen (unixpath))) /* We have a non-NULL unixpath, does that mean it's valid? */
+       {
+          sock = GNUNET_CONNECTION_create_from_connect_to_unixpath (cfg, unixpath);
+         if (sock != NULL)
+           {
+#if DEBUG_CLIENT
+              GNUNET_log(GNUNET_ERROR_TYPE_DEBUG, "Connected to unixpath `%s'!\n", unixpath);
+#endif
+              GNUNET_free(unixpath);
+              return sock;
+           }
+       }
+      GNUNET_free_non_null (unixpath);
+    }
+#endif
+
   if ((GNUNET_OK !=
        GNUNET_CONFIGURATION_get_value_number (cfg,
                                               service_name,
@@ -260,8 +308,7 @@ do_connect (struct GNUNET_SCHEDULER_Handle *sched,
                                               "HOSTNAME", &hostname)))
     {
       GNUNET_log (GNUNET_ERROR_TYPE_WARNING,
-                  _
-                  ("Could not determine valid hostname and port for service `%s' from configuration.\n"),
+                  _("Could not determine valid hostname and port for service `%s' from configuration.\n"),
                   service_name);
       return NULL;
     }
@@ -273,11 +320,38 @@ do_connect (struct GNUNET_SCHEDULER_Handle *sched,
                   service_name);
       return NULL;
     }
-  sock = GNUNET_CONNECTION_create_from_connect (sched,
-                                                cfg,
+  if (port == 0)
+    {
+#if AF_UNIX
+      if (0 != (attempt % 2))
+       {
+         /* try UNIX */
+         if ((GNUNET_OK ==
+             GNUNET_CONFIGURATION_get_value_string (cfg,
+                                                    service_name,
+                                                    "UNIXPATH", &unixpath)) &&
+              (0 < strlen (unixpath)))
+           {
+             sock = GNUNET_CONNECTION_create_from_connect_to_unixpath (cfg,
+                                                                       unixpath);
+             GNUNET_free (unixpath);
+             if (sock != NULL)
+               return sock;            
+           }
+       }
+#endif
+#if DEBUG_CLIENT
+      GNUNET_log (GNUNET_ERROR_TYPE_DEBUG, 
+                 "Port is 0 for service `%s', UNIXPATH did not work, returning NULL!\n",
+                 service_name);
+#endif
+      GNUNET_free (hostname);
+      return NULL;
+    }
+
+  sock = GNUNET_CONNECTION_create_from_connect (cfg,
                                                 hostname,
-                                                port,
-                                                GNUNET_SERVER_MAX_MESSAGE_SIZE);
+                                                port);
   GNUNET_free (hostname);
   return sock;
 }
@@ -286,45 +360,66 @@ do_connect (struct GNUNET_SCHEDULER_Handle *sched,
 /**
  * Get a connection with a service.
  *
- * @param sched scheduler to use
  * @param service_name name of the service
  * @param cfg configuration to use
  * @return NULL on error (service unknown to configuration)
  */
 struct GNUNET_CLIENT_Connection *
-GNUNET_CLIENT_connect (struct GNUNET_SCHEDULER_Handle *sched,
-                       const char *service_name,
+GNUNET_CLIENT_connect (const char *service_name,
                        const struct GNUNET_CONFIGURATION_Handle *cfg)
 {
   struct GNUNET_CLIENT_Connection *ret;
   struct GNUNET_CONNECTION_Handle *sock;
 
-  sock = do_connect (sched, service_name, cfg);
+  sock = do_connect (service_name,
+                    cfg, 0);
   if (sock == NULL)
     return NULL;
   ret = GNUNET_malloc (sizeof (struct GNUNET_CLIENT_Connection));
+  ret->attempts = 1;
   ret->sock = sock;
-  ret->sched = sched;
   ret->service_name = GNUNET_strdup (service_name);
   ret->cfg = GNUNET_CONFIGURATION_dup (cfg);
+  ret->back_off = GNUNET_TIME_UNIT_MILLISECONDS;
   return ret;
 }
 
 
+/**
+ * Configure this connection to ignore shutdown signals.
+ *
+ * @param h client handle
+ * @param do_ignore GNUNET_YES to ignore, GNUNET_NO to restore default
+ */
+void
+GNUNET_CLIENT_ignore_shutdown (struct GNUNET_CLIENT_Connection *h,
+                              int do_ignore)
+{
+  h->ignore_shutdown = do_ignore;
+  if (h->sock != NULL)
+    GNUNET_CONNECTION_ignore_shutdown (h->sock,
+                                      do_ignore);
+}
+
+
 /**
  * Destroy connection with the service.  This will automatically
  * cancel any pending "receive" request (however, the handler will
  * *NOT* be called, not even with a NULL message).  Any pending
  * transmission request will also be cancelled UNLESS the callback for
  * the transmission request has already been called, in which case the
- * transmission is guaranteed to complete before the socket is fully
- * destroyed (unless, of course, there is an error with the server
- * in which case the message may still be lost).
+ * transmission 'finish_pending_write' argument determines whether or
+ * not the write is guaranteed to complete before the socket is fully
+ * destroyed (unless, of course, there is an error with the server in
+ * which case the message may still be lost).
  *
+ * @param finish_pending_write should a transmission already passed to the
+ *          handle be completed?
  * @param sock handle to the service connection
  */
 void
-GNUNET_CLIENT_disconnect (struct GNUNET_CLIENT_Connection *sock)
+GNUNET_CLIENT_disconnect (struct GNUNET_CLIENT_Connection *sock,
+                         int finish_pending_write)
 {
   GNUNET_assert (sock->sock != NULL);
   if (sock->in_receive == GNUNET_YES)
@@ -332,7 +427,7 @@ GNUNET_CLIENT_disconnect (struct GNUNET_CLIENT_Connection *sock)
       GNUNET_CONNECTION_receive_cancel (sock->sock);
       sock->in_receive = GNUNET_NO;
     }
-  GNUNET_CONNECTION_destroy (sock->sock);
+  GNUNET_CONNECTION_destroy (sock->sock, finish_pending_write);
   sock->sock = NULL;
   if (sock->tag != NULL)
     {
@@ -344,7 +439,7 @@ GNUNET_CLIENT_disconnect (struct GNUNET_CLIENT_Connection *sock)
     GNUNET_CLIENT_notify_transmit_ready_cancel (sock->th);
   if (sock->receive_task != GNUNET_SCHEDULER_NO_TASK)
     {
-      GNUNET_SCHEDULER_cancel (sock->sched, sock->receive_task);
+      GNUNET_SCHEDULER_cancel (sock->receive_task);
       sock->receive_task = GNUNET_SCHEDULER_NO_TASK;
     }
   GNUNET_array_grow (sock->received_buf, sock->received_size, 0);
@@ -362,8 +457,8 @@ check_complete (struct GNUNET_CLIENT_Connection *conn)
 {
   if ((conn->received_pos >= sizeof (struct GNUNET_MessageHeader)) &&
       (conn->received_pos >=
-       ntohs (((const struct GNUNET_MessageHeader *) conn->
-               received_buf)->size)))
+       ntohs (((const struct GNUNET_MessageHeader *) conn->received_buf)->
+              size)))
     conn->msg_complete = GNUNET_YES;
 }
 
@@ -396,6 +491,9 @@ receive_helper (void *cls,
   if ((available == 0) || (conn->sock == NULL) || (errCode != 0))
     {
       /* signal timeout! */
+#if DEBUG_CLIENT
+      GNUNET_log(GNUNET_ERROR_TYPE_DEBUG, "timeout in receive_helper, available %d, conn->sock %s, errCode %d\n", available, conn->sock == NULL ? "NULL" : "non-NULL", errCode);
+#endif
       if (NULL != (receive_handler = conn->receiver_handler))
         {
           receive_handler_cls = conn->receiver_handler_cls;
@@ -418,7 +516,7 @@ receive_helper (void *cls,
   check_complete (conn);
   /* check for timeout */
   remaining = GNUNET_TIME_absolute_get_remaining (conn->receive_timeout);
-  if (remaining.value == 0)
+  if (remaining.rel_value == 0)
     {
       /* signal timeout! */
       conn->receiver_handler (conn->receiver_handler_cls, NULL);
@@ -449,6 +547,12 @@ receive_task (void *cls, const struct GNUNET_SCHEDULER_TaskContext *tc)
   char mbuf[msize];
   struct GNUNET_MessageHeader *msg = (struct GNUNET_MessageHeader *) mbuf;
 
+#if DEBUG_CLIENT
+  GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+             "Received message of type %u and size %u\n",
+             ntohs (cmsg->type),
+             msize);
+#endif
   sock->receive_task = GNUNET_SCHEDULER_NO_TASK;
   GNUNET_assert (GNUNET_YES == sock->msg_complete);
   GNUNET_assert (sock->received_pos >= msize);
@@ -489,75 +593,30 @@ GNUNET_CLIENT_receive (struct GNUNET_CLIENT_Connection *sock,
   sock->receive_timeout = GNUNET_TIME_relative_to_absolute (timeout);
   if (GNUNET_YES == sock->msg_complete)
     {
-      sock->receive_task = GNUNET_SCHEDULER_add_after (sock->sched,
-                                                       GNUNET_SCHEDULER_NO_TASK,
+      sock->receive_task = GNUNET_SCHEDULER_add_after (GNUNET_SCHEDULER_NO_TASK,
                                                        &receive_task, sock);
     }
   else
     {
+      GNUNET_assert (sock->in_receive == GNUNET_NO);
       sock->in_receive = GNUNET_YES;
+#if DEBUG_CLIENT
+      GNUNET_log(GNUNET_ERROR_TYPE_DEBUG, "calling GNUNET_CONNECTION_receive\n");
+#endif
       GNUNET_CONNECTION_receive (sock->sock,
-                                 GNUNET_SERVER_MAX_MESSAGE_SIZE,
+                                 GNUNET_SERVER_MAX_MESSAGE_SIZE - 1,
                                  timeout, &receive_helper, sock);
     }
 }
 
 
-/**
- * If possible, write a shutdown message to the target
- * buffer and destroy the client connection.
- *
- * @param cls the "struct GNUNET_CLIENT_Connection" to destroy
- * @param size number of bytes available in buf
- * @param buf NULL on error, otherwise target buffer
- * @return number of bytes written to buf
- */
-static size_t
-write_shutdown (void *cls, size_t size, void *buf)
-{
-  struct GNUNET_MessageHeader *msg;
-  struct GNUNET_CLIENT_Connection *sock = cls;
-
-  GNUNET_CLIENT_disconnect (sock);
-  if (size < sizeof (struct GNUNET_MessageHeader))
-    {
-      GNUNET_log (GNUNET_ERROR_TYPE_INFO,
-                  _("Failed to transmit shutdown request to client.\n"));
-      return 0;                 /* client disconnected */
-    }
-  msg = (struct GNUNET_MessageHeader *) buf;
-  msg->type = htons (GNUNET_MESSAGE_TYPE_SHUTDOWN);
-  msg->size = htons (sizeof (struct GNUNET_MessageHeader));
-  return sizeof (struct GNUNET_MessageHeader);
-}
-
-
-/**
- * Request that the service should shutdown.
- * Afterwards, the connection should be disconnected.
- *
- * @param sock the socket connected to the service
- */
-void
-GNUNET_CLIENT_service_shutdown (struct GNUNET_CLIENT_Connection *sock)
-{
-  GNUNET_CONNECTION_notify_transmit_ready (sock->sock,
-                                           sizeof (struct
-                                                   GNUNET_MessageHeader),
-                                           GNUNET_TIME_UNIT_FOREVER_REL,
-                                           &write_shutdown, sock);
-}
-
-
 /**
  * Report service unavailable.
  */
 static void
-service_test_error (struct GNUNET_SCHEDULER_Handle *s,
-                    GNUNET_SCHEDULER_Task task, void *task_cls)
+service_test_error (GNUNET_SCHEDULER_Task task, void *task_cls)
 {
-  GNUNET_SCHEDULER_add_continuation (s,
-                                     task,
+  GNUNET_SCHEDULER_add_continuation (task,
                                      task_cls,
                                      GNUNET_SCHEDULER_REASON_TIMEOUT);
 }
@@ -582,22 +641,22 @@ confirm_handler (void *cls, const struct GNUNET_MessageHeader *msg)
       GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
                   "Received confirmation that service is running.\n");
 #endif
-      GNUNET_SCHEDULER_add_continuation (conn->sched,
-                                         conn->test_cb,
+      GNUNET_SCHEDULER_add_continuation (conn->test_cb,
                                          conn->test_cb_cls,
                                          GNUNET_SCHEDULER_REASON_PREREQ_DONE);
     }
   else
     {
-      service_test_error (conn->sched, conn->test_cb, conn->test_cb_cls);
+      service_test_error (conn->test_cb, conn->test_cb_cls);
     }
-  GNUNET_CLIENT_disconnect (conn);
+  GNUNET_CLIENT_disconnect (conn, GNUNET_NO);
 }
 
 
 static size_t
 write_test (void *cls, size_t size, void *buf)
 {
+  struct GNUNET_CLIENT_Connection *conn = cls;
   struct GNUNET_MessageHeader *msg;
 
   if (size < sizeof (struct GNUNET_MessageHeader))
@@ -606,6 +665,8 @@ write_test (void *cls, size_t size, void *buf)
       GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
                   _("Failure to transmit TEST request.\n"));
 #endif
+      service_test_error (conn->test_cb, conn->test_cb_cls);
+      GNUNET_CLIENT_disconnect (conn, GNUNET_NO);
       return 0;                 /* client disconnected */
     }
 #if DEBUG_CLIENT
@@ -615,6 +676,10 @@ write_test (void *cls, size_t size, void *buf)
   msg = (struct GNUNET_MessageHeader *) buf;
   msg->type = htons (GNUNET_MESSAGE_TYPE_TEST);
   msg->size = htons (sizeof (struct GNUNET_MessageHeader));
+  GNUNET_CLIENT_receive (conn, 
+                        &confirm_handler, 
+                        conn, 
+                        GNUNET_TIME_absolute_get_remaining (conn->test_deadline));
   return sizeof (struct GNUNET_MessageHeader);
 }
 
@@ -622,7 +687,6 @@ write_test (void *cls, size_t size, void *buf)
 /**
  * Wait until the service is running.
  *
- * @param sched scheduler to use
  * @param service name of the service to wait for
  * @param cfg configuration to use
  * @param timeout how long to wait at most in ms
@@ -632,8 +696,7 @@ write_test (void *cls, size_t size, void *buf)
  * @param task_cls closure for task
  */
 void
-GNUNET_CLIENT_service_test (struct GNUNET_SCHEDULER_Handle *sched,
-                            const char *service,
+GNUNET_CLIENT_service_test (const char *service,
                             const struct GNUNET_CONFIGURATION_Handle *cfg,
                             struct GNUNET_TIME_Relative timeout,
                             GNUNET_SCHEDULER_Task task, void *task_cls)
@@ -644,32 +707,33 @@ GNUNET_CLIENT_service_test (struct GNUNET_SCHEDULER_Handle *sched,
   GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
               "Testing if service `%s' is running.\n", service);
 #endif
-  conn = GNUNET_CLIENT_connect (sched, service, cfg);
+  conn = GNUNET_CLIENT_connect (service, cfg);
   if (conn == NULL)
     {
       GNUNET_log (GNUNET_ERROR_TYPE_INFO,
                   _
                   ("Could not connect to service `%s', must not be running.\n"),
                   service);
-      service_test_error (sched, task, task_cls);
+      service_test_error (task, task_cls);
       return;
     }
   conn->test_cb = task;
   conn->test_cb_cls = task_cls;
-  if (NULL ==
-      GNUNET_CONNECTION_notify_transmit_ready (conn->sock,
-                                               sizeof (struct
-                                                       GNUNET_MessageHeader),
-                                               timeout, &write_test, NULL))
+  conn->test_deadline = GNUNET_TIME_relative_to_absolute (timeout);
+
+  if (NULL == GNUNET_CLIENT_notify_transmit_ready (conn,
+                                                  sizeof (struct GNUNET_MessageHeader),
+                                                  timeout,
+                                                  GNUNET_YES,
+                                                  &write_test, conn))  
     {
       GNUNET_log (GNUNET_ERROR_TYPE_WARNING,
                   _("Failure to transmit request to service `%s'\n"),
                   service);
-      service_test_error (sched, task, task_cls);
-      GNUNET_CLIENT_disconnect (conn);
+      service_test_error (task, task_cls);
+      GNUNET_CLIENT_disconnect (conn, GNUNET_NO);
       return;
     }
-  GNUNET_CLIENT_receive (conn, &confirm_handler, conn, timeout);
 }
 
 
@@ -727,9 +791,8 @@ client_delayed_retry (void *cls,
 
 
 /**
- * Connection notifies us about failure or success of
- * a transmission request.  Either pass it on to our
- * user or, if possible, retry.
+ * Connection notifies us about failure or success of a transmission
+ * request.  Either pass it on to our user or, if possible, retry.
  *
  * @param cls our "struct GNUNET_CLIENT_TransmissionHandle"
  * @param size number of bytes available for transmission
@@ -747,13 +810,12 @@ client_notify (void *cls, size_t size, void *buf)
   th->sock->th = NULL;
   if (buf == NULL)
     {
-      // FIXME: need a way to check if the
-      // reason is SHUTDOWN (not timeout) and
-      // if so NOT retry!
       delay = GNUNET_TIME_absolute_get_remaining (th->timeout);
-      delay.value /= 2;
-      if ((GNUNET_YES != th->auto_retry) ||
-          (0 == --th->attempts_left) || (delay.value < 1))
+      delay.rel_value /= 2;
+      if ( (0 != (GNUNET_SCHEDULER_REASON_SHUTDOWN & GNUNET_SCHEDULER_get_reason ())) ||
+          (GNUNET_YES != th->auto_retry) ||
+          (0 == --th->attempts_left) || 
+          (delay.rel_value < 1) )
         {
 #if DEBUG_CLIENT
           GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
@@ -765,19 +827,24 @@ client_notify (void *cls, size_t size, void *buf)
           return 0;
         }
       /* auto-retry */
-      GNUNET_CONNECTION_destroy (th->sock->sock);
-      th->sock->sock = do_connect (th->sock->sched,
-                                   th->sock->service_name, th->sock->cfg);
+      GNUNET_CONNECTION_destroy (th->sock->sock, GNUNET_NO);
+      th->sock->sock = do_connect (th->sock->service_name,
+                                  th->sock->cfg,
+                                  th->sock->attempts++);
       GNUNET_assert (NULL != th->sock->sock);
-      delay = GNUNET_TIME_relative_min (delay, GNUNET_TIME_UNIT_SECONDS);
+      GNUNET_CONNECTION_ignore_shutdown (th->sock->sock,
+                                        th->sock->ignore_shutdown);
+      delay = GNUNET_TIME_relative_min (delay, th->sock->back_off);
+      th->sock->back_off 
+         = GNUNET_TIME_relative_min (GNUNET_TIME_relative_multiply (th->sock->back_off, 2),
+                                   GNUNET_TIME_UNIT_SECONDS);
 #if DEBUG_CLIENT
       GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
                   "Transmission failed %u times, trying again in %llums.\n",
                   MAX_ATTEMPTS - th->attempts_left,
-                  (unsigned long long) delay.value);
+                  (unsigned long long) delay.rel_value);
 #endif
-      th->reconnect_task = GNUNET_SCHEDULER_add_delayed (th->sock->sched,
-                                                         delay,
+      th->reconnect_task = GNUNET_SCHEDULER_add_delayed (delay,
                                                          &client_delayed_retry,
                                                          th);
       th->sock->th = th;
@@ -856,12 +923,12 @@ GNUNET_CLIENT_notify_transmit_ready_cancel (struct
   if (th->reconnect_task != GNUNET_SCHEDULER_NO_TASK)
     {
       GNUNET_break (NULL == th->th);
-      GNUNET_SCHEDULER_cancel (th->sock->sched, th->reconnect_task);
+      GNUNET_SCHEDULER_cancel (th->reconnect_task);
       th->reconnect_task = GNUNET_SCHEDULER_NO_TASK;
     }
   else
     {
-      GNUNET_break (NULL != th->th);
+      GNUNET_assert (NULL != th->th);
       GNUNET_CONNECTION_notify_transmit_ready_cancel (th->th);
     }
   th->sock->th = NULL;
@@ -890,6 +957,10 @@ transmit_for_response (void *cls, size_t size, void *buf)
   msize = ntohs (tc->hdr->size);
   if (NULL == buf)
     {
+#if DEBUG_CLIENT
+      GNUNET_log (GNUNET_ERROR_TYPE_DEBUG,
+                 _("Could not submit request, not expecting to receive a response.\n"));
+#endif
       tc->rn (tc->rn_cls, NULL);
       GNUNET_free (tc);
       return 0;