Added "no network" detection, tweaked housekeeping timers to cap the

exponential fallback at different levels depending on that factor and
rejection status.

Also improved stability of frontend-choice in the face of local network
outages by only changing frontend choices if DNS is working.

include/pagekite.h

@@ -24,12 +24,12 @@ Note: For alternate license terms, see the file COPYING.md.
 #define _PAGEKITEC_DLL_H
 
 #ifdef ANDROID
-#define PK_VERSION "0.90.150209A"
+#define PK_VERSION "0.90.150224A"
 #else
 #ifdef _MSC_VER
-#define PK_VERSION "0.90.150209W"
+#define PK_VERSION "0.90.150224W"
 #else
-#define PK_VERSION "0.90.150209C"
+#define PK_VERSION "0.90.150224C"
 #endif
 #endif
 

libpagekite/pkblocker.c

@@ -213,7 +213,7 @@ void pkb_choose_tunnels(struct pk_manager* pkm)
   pk_log(PK_LOG_MANAGER_ERROR, "No front-end wanted! We are lame.");
 }
 
-void pkb_check_kites_dns(struct pk_manager* pkm)
+int pkb_check_kites_dns(struct pk_manager* pkm)
 {
   int i, j, rv;
   int in_dns = 0;
@@ -225,24 +225,27 @@ void pkb_check_kites_dns(struct pk_manager* pkm)
   struct addrinfo hints;
   struct addrinfo *result, *rp;
   char buffer[128];
+  int cleared_flags = 0;
 
   PK_TRACE_FUNCTION;
 
   memset(&hints, 0, sizeof(struct addrinfo));
   hints.ai_family = AF_UNSPEC;
   hints.ai_socktype = SOCK_STREAM;
 
-  /* Clear DNS flag... */
-  for (j = 0, fe = pkm->tunnels; j < pkm->tunnel_max; j++, fe++) {
-    fe->conn.status &= ~FE_STATUS_IN_DNS;
-  }
-
   /* Walk through kite list, look each up in DNS and update the
    * tunnel flags as appropriate.
    */
   for (i = 0, kite = pkm->kites; i < pkm->kite_max; i++, kite++) {
     rv = getaddrinfo(kite->public_domain, NULL, &hints, &result);
     if (rv == 0) {
+      if (!cleared_flags) {
+        /* Clear DNS flag everywhere, once we know DNS is responding. */
+        for (j = 0, fe = pkm->tunnels; j < pkm->tunnel_max; j++, fe++) {
+          fe->conn.status &= ~FE_STATUS_IN_DNS;
+        }
+        cleared_flags = 1;
+      }
       for (rp = result; rp != NULL; rp = rp->ai_next) {
         for (j = 0, fe = pkm->tunnels; j < pkm->tunnel_max; j++, fe++) {
           if (fe->ai && fe->fe_hostname) {
@@ -262,6 +265,11 @@ void pkb_check_kites_dns(struct pk_manager* pkm)
     }
   }
 
+  /*
+   * If flags weren't cleared, then the network is probably down: bail out!
+   */
+  if (!cleared_flags) return 1;
+
   /* FIXME: We should really get this from the TTL of the DNS record itself,
    *        not from a hard coded magic number.
    */
@@ -295,6 +303,7 @@ void pkb_check_kites_dns(struct pk_manager* pkm)
   }
 
   PK_CHECK_MEMORY_CANARIES;
+  return 0;
 }
 
 void* pkb_tunnel_ping(void* void_fe) {
@@ -528,16 +537,15 @@ void pkb_check_world(struct pk_manager* pkm)
 void pkb_check_tunnels(struct pk_manager* pkm)
 {
   int problems = 0;
+  int dns_is_down = 0;
   PK_TRACE_FUNCTION;
 
-  if (pkm->status == PK_STATUS_NO_NETWORK) return;
-  pk_log(PK_LOG_MANAGER_DEBUG, "Checking tunnels...");
+  pk_log(PK_LOG_MANAGER_DEBUG, "Checking network & tunnels...");
 
-  pkb_check_kites_dns(pkm);
+  dns_is_down = (0 != pkb_check_kites_dns(pkm));
   pkb_choose_tunnels(pkm);
   pkb_log_fe_status(pkm);
-
-  problems += pkm_reconnect_all(pkm);
+  problems += pkm_reconnect_all(pkm, dns_is_down);
 
   if (!problems) pkm_disconnect_unused(pkm);
 
@@ -551,7 +559,10 @@ void pkb_check_tunnels(struct pk_manager* pkm)
     PKS_STATE(pkm->status = PK_STATUS_FLYING);
   }
   else if (pkm->status != PK_STATUS_REJECTED) {
-    PKS_STATE(pkm->status = PK_STATUS_PROBLEMS);
+    if (dns_is_down)
+      pk_log(PK_LOG_MANAGER_DEBUG, "Network appears to be down.");
+    PKS_STATE(pkm->status = (dns_is_down ? PK_STATUS_NO_NETWORK
+                                         : PK_STATUS_PROBLEMS));
   }
 }
 

libpagekite/pkmanager.c

@@ -664,7 +664,7 @@ static void pkm_listener_cb(EV_P_ ev_io* w, int revents)
   (void) revents;
 }
 
-int pkm_reconnect_all(struct pk_manager* pkm) {
+int pkm_reconnect_all(struct pk_manager* pkm, int ignore_errors) {
   struct pk_tunnel *fe;
   struct pk_kite_request *kite_r;
   unsigned int status;
@@ -742,7 +742,8 @@ int pkm_reconnect_all(struct pk_manager* pkm) {
         /* FIXME: Is this the right behavior? */
         pk_log(PK_LOG_MANAGER_INFO, "Connect failed: %d", fe->conn.sockfd);
         fe->request_count = 0;
-        if (fe->error_count < 999)
+
+        if (!ignore_errors && fe->error_count < 999)
           fe->error_count += 1;
 
         status = fe->conn.status;
@@ -848,9 +849,23 @@ static void pkm_tick_cb(EV_P_ ev_async* w, int revents)
     pk_log(PK_LOG_MANAGER_DEBUG, "Tick!  [repeating=%s, next=%d]",
            pkm->enable_timer ? "yes" : "no", pkm->next_tick);
 
-    /* We slow down exponentially by default, no matter what. */
+    /* We slow down exponentially by default... */
     next_tick += increment;
     max_tick = pkm->housekeeping_interval_max + pkm->interval_fudge_factor;
+
+    /* Fallback is tuned for normal operation: if the relay rejected
+       us that is not normal anymore and we'll happily go idle until it
+       is time to re-check the state of the world. */
+    if (pkm->status == PK_STATUS_REJECTED)
+      max_tick += pkm->check_world_interval;
+
+    /* Similarly, if there's no network AND we are relying on the built-in
+       timer and not external app logic, then retry more frequently. Power
+       constrained apps shouldn't be using the timer anyway! */
+    if (pkm->status == PK_STATUS_NO_NETWORK)
+      max_tick = (PK_HOUSEKEEPING_INTERVAL_MIN +
+                  PK_HOUSEKEEPING_INTERVAL_MAX_MIN) / 2;
+
     if (next_tick > max_tick)
       next_tick = max_tick;
   }

libpagekite/pkmanager.h

@@ -166,7 +166,7 @@ int pkm_run_in_thread               (struct pk_manager*);
 int pkm_wait_thread                 (struct pk_manager*);
 int pkm_stop_thread                 (struct pk_manager*);
 
-int pkm_reconnect_all               (struct pk_manager*);
+int pkm_reconnect_all               (struct pk_manager*, int);
 int pkm_disconnect_unused           (struct pk_manager*);
 
 int pkm_configure_lua               (struct pk_manager*);