Send a PING to the server before assuming a timeout is definite.

4 files changed, 58 insertions, 30 deletions

diff --git a/TODO b/TODO
index ead2645..618b2b9 100644
--- a/TODO
+++ b/TODO
@@ -7,4 +7,4 @@ Add various auto replay options:
 
 Might need to #include <limits.h> in blabouncer.c to make some operating systems and/or compilers happy.
 
-Send a PING to the server before assuming a timeout is definite.
+Handle connecting to the server failing.
diff --git a/blabouncer.c b/blabouncer.c
index d14223e..4986551 100644
--- a/blabouncer.c
+++ b/blabouncer.c
@@ -70,6 +70,7 @@
 #define MAXRCVSIZE 2417
 #define MAXTOKENS 100 // maximum number of (CRLF or space) separated tokens per server response we expect (TODO - check this is reasonable) (CRLF and spaces just grouped here due to laziness)
 #define SERVERTIMEOUT 300 // How many seconds to wait without hearing from the server before assuming a timeout
+#define SELECTTIMEOUT 60 // How many seconds to wait before our pselect() times out, useful for detecting server timeouts
 
 // Global debug control
 int debug = 0; // Debug verbosity ("0" for critical only, "1" for some extra info, "2" for full debug mode)
@@ -126,6 +127,7 @@ int connecttoircserver(SSL_CTX **serverctx, SSL **server_ssl, int *serversockfd,
   ircdstrings->capmultiprefix = 0;
   ircdstrings->autonicknum = 0;
   ircdstrings->lastmessagetime = time(NULL);
+  ircdstrings->timeoutcheck = 0;
   // ircdstrings.reconnecting is not set here as we want to track reconnections separately
 
   // Populate nick and username from our configuration file for now, real IRCd may change them later (TODO - Is this true of username?)
@@ -449,37 +451,12 @@ void dochat(int *serversockfd, int *clientsockfd, struct settings *settings) {
       }
     }
 
-    // Before we wait for a message, let's make sure the server is still responding
-    // TODO - Code duplication, make a function and share with socket error code below
-    if (ircdstrings.lastmessagetime < time(NULL) - SERVERTIMEOUT) {
-      debugprint(DEBUG_CRIT, "Server has timed out (%ld seconds), reconnecting!\n", time(NULL) - ircdstrings.lastmessagetime);
-      // Tell all clients if we timed out
-      char alertmsg[MAXDATASIZE];
-      snprintf(alertmsg, MAXDATASIZE, "NOTICE %s :Server has timed out (%ld seconds), reconnecting!", ircdstrings.ircnick, time(NULL) - ircdstrings.lastmessagetime);
-      sendtoallclients(clients, alertmsg, 0, settings);
-      if (settings->servertls) {
-        // Finish up with OpenSSL if using server TLS
-        SSL_free(server_ssl);
-      }
-      // Close the socket
-      close(*serversockfd);
-      // Make a new one
-      *serversockfd = createserversocket(settings->ircserver, settings->ircserverport);
-
-      // Set reconnection marker for other functions to know we're reconnecting
-      ircdstrings.reconnecting = 1;
-      // Set oldnick in case we change nick when reconnecting so we can inform existing clients
-      strcpy(ircdstrings.oldnick, ircdstrings.ircnick);
-      connecttoircserver(&serverctx, &server_ssl, serversockfd, &ircdstrings, settings, clients);
-
-      // Back to top of loop
-      continue;
-    }
-
     debugprint(DEBUG_FULL, "pselect()ing...\n");
     // Check to see if any fd in the fd_set is waiting or a signal happened - blocks here until one one of those things happens
-    // (pselect() to do signal handling in additiona to fd monitoring)
-    if (pselect(fdmax + 1, &rfds, NULL, NULL, NULL, &oldset) < 0) { // network socket + 1, rfds, no writes, no exceptions/errors, no timeout, original sigmask
+    // (pselect() to do signal handling in addition to fd monitoring)
+    struct timespec timeout = {SELECTTIMEOUT, 0};  // pselect() should timeout after SELECTTIMEOUT seconds
+    int selret = pselect(fdmax + 1, &rfds, NULL, NULL, &timeout, &oldset); // network socket + 1, rfds, no writes, no exceptions/errors, no timeout, original sigmask
+    if (selret < 0) { // Error or signal interrupt
       if (errno == EINTR) {
         // Signal caught, do signal handling
         debugprint(DEBUG_CRIT, "signal '%d' happened, exiting!\n", signum);
@@ -498,6 +475,54 @@ void dochat(int *serversockfd, int *clientsockfd, struct settings *settings) {
       }
     }
 
+    // pselect() timed out, let's see if a server timeout might be happening, if not, just continue to the top of the loop again
+    if (selret == 0) {
+      debugprint(DEBUG_CRIT, "pselect() timed out.\n", errno);
+
+      // SERVERTIMEOUT seconds have expired...
+      if (ircdstrings.lastmessagetime < time(NULL) - SERVERTIMEOUT && !FD_ISSET(*serversockfd, &rfds)) {
+        if (ircdstrings.timeoutcheck == 0) {
+          // ...and we haven't tried a PING yet, so let's PING the server to see if things are still working
+          debugprint(DEBUG_CRIT, "Server might have timed out after %ld seconds, PINGing it...\n", time(NULL) - ircdstrings.lastmessagetime);
+          ircdstrings.timeoutcheck = 1;
+          char outgoingmsg[MAXDATASIZE];
+          if (!snprintf(outgoingmsg, MAXDATASIZE, "PING %s", ircdstrings.ircdname)) {
+            fprintf(stderr, "Error while preparing timeout testing PING message!\n");
+            debugprint(DEBUG_CRIT, "Error while preparing timeout testing PING message\n");
+            snprintf(outgoingmsg, MAXDATASIZE, "PING timeouttest");
+          }
+          sendtoserver(server_ssl, outgoingmsg, strlen(outgoingmsg), 0, clients, settings);
+        } else {
+          // ...and we've already PINGed the server and haven't heard back yet, so let's assume we've timed out
+          // TODO - Code duplication, make a function and share with socket error code below
+          debugprint(DEBUG_CRIT, "Server has timed out (%ld seconds), reconnecting!\n", time(NULL) - ircdstrings.lastmessagetime);
+          // Tell all clients if we timed out
+          char alertmsg[MAXDATASIZE];
+          snprintf(alertmsg, MAXDATASIZE, "NOTICE %s :Server has timed out (%ld seconds), reconnecting!", ircdstrings.ircnick, time(NULL) - ircdstrings.lastmessagetime);
+          sendtoallclients(clients, alertmsg, 0, settings);
+          if (settings->servertls) {
+            // Finish up with OpenSSL if using server TLS
+            SSL_free(server_ssl);
+          }
+          // Close the socket
+          close(*serversockfd);
+          // Make a new one
+          *serversockfd = createserversocket(settings->ircserver, settings->ircserverport);
+
+          // Set reconnection marker for other functions to know we're reconnecting
+          ircdstrings.reconnecting = 1;
+          // Set oldnick in case we change nick when reconnecting so we can inform existing clients
+          strcpy(ircdstrings.oldnick, ircdstrings.ircnick);
+          connecttoircserver(&serverctx, &server_ssl, serversockfd, &ircdstrings, settings, clients);
+        }
+        // Back to top of loop
+        continue;
+      } else {
+        // No timeout is occuring, back to the top of the loop
+        continue;
+      }
+    }
+
     // TODO - switch around the serversockfd and STDIN FD_ISSET if-statements?  They feel the wrong way round.  Are they like this on purpose?  I can't remember.
     // (although actually stdin may go once its not wanted for possible direct interaction for debugging)
 
diff --git a/message.c b/message.c
index f837b83..b021347 100644
--- a/message.c
+++ b/message.c
@@ -23,6 +23,8 @@ int processservermessage(SSL *server_ssl, char *str, struct client *clients, int
                          struct channel *channels, struct settings *settings, char tokens[MAXTOKENS][MAXDATASIZE], int counter) {
   // Record that we received something from the server for timeout checking purposes
   ircdstrings->lastmessagetime = time(NULL); // snprintf(NULL, 0, "%ld", timenow);
+  // And we can't be timing out
+  ircdstrings->timeoutcheck = 0;
 
   // Server PING received?  If so, send a PONG back with the next element as the argument.
   if (strncmp(tokens[0], "PING", strlen(tokens[0])) == 0) {
diff --git a/structures.h b/structures.h
index 1de8fc8..5b9c2c0 100644
--- a/structures.h
+++ b/structures.h
@@ -46,6 +46,7 @@ struct ircdstrings {
   int capmultiprefix; // Whether the server approved our CAP multi-prefix request
   int autonicknum; // Number of attempts made at automatically setting a nick if all configured nicks were in use
   int lastmessagetime; // The last time we heard from the server
+  int timeoutcheck; // Whether we're checking to see if we've timed out from the server
   int reconnecting; // Whether or not we're reconnecting due to an earlier disconnection
   char oldnick[MAXNICKLENGTH]; // Set temporarily if we end up reconnecting in case we need to tell existing clients about a nick change
 };