[patch 46/56] fix kcomd_thread error handling behavior

Subject: [patch @num@/@total@] fix kcomd_thread error handling behavior

kcomd_thread should not exit in case of packet error, instead, it should just 
reset the connection to the corresponding node

kcomd_thread should only quit when the module is unloaded

FIXME: for now, the kcomd module does not send back all tasks (remote) and
does not backup the deputy tasks when exiting.... It just delays it for 120sec
and exit ... 

(this problem will be addressed in another patch, I'll do that soon I hope)
Index: linux/hpc/kcomd.c
===================================================================
--- linux.orig/hpc/kcomd.c	2006-11-02 22:52:47.000000000 +0100
+++ linux/hpc/kcomd.c	2006-11-02 22:52:51.000000000 +0100
@@ -29,7 +29,8 @@
 #include <hpc/prototype.h>
 #include <hpc/debug.h>
 
-static int kcomd_done=0;
+static int kcomd_done = 0;
+static int kcomd_module_exit = 0;
 
 /**
  * socket_listen - Creates the network socket and maps it to a file descriptor
@@ -493,6 +494,49 @@
 }
 
 /**
+ * kcomd_leave_cluster
+ *
+ * Description:
+ *    This function is called before the kcomd_thread daeomon quits
+ *    because of module unloading (FIXME or kernel stop ??) and
+ *    return 0 if kcomd_thread can stop operations.
+ *    This function is responsible for sending back all remote
+ *    and retreiving all deputies back ...
+ **/
+
+int kcomd_leave_cluster(void)
+{
+
+	int ret = 0;
+	struct kcom_node* node;
+
+	read_lock(&kcom_nodes_lock);
+	list_for_each_entry(node, &kcom_nodes, list) {
+
+	    if (!list_empty(&node->tasks)) {
+	    /*
+	     * fdy, FIXME: when exiting kcomd we should ask all tasks to comeback ...
+	     * send back all tasks, then prevent new one to comeback ... this needs an
+	     * upgrade of this thread ... Moreover, there seems to have kcache_*alloc
+	     * misusage, before exit ... (for now just prevent exit ..., with a message
+	     */
+
+		OMBUG("kcomd FIXME, there are still remote/deputy "
+		       "tasks, I can't die now!\n");
+		ret = 1;
+		break;
+	    }
+
+	}
+
+	read_unlock(&kcom_nodes_lock);
+
+
+	return ret;
+}
+
+
+/**
  * kcomd_thread
  *
  * Description:
@@ -543,8 +587,6 @@
 
 	while (kcomd_done==0)
 	{
-
-
 		n = maxfds;
 
 		kcomd_thread_set_fds(n);
@@ -559,7 +601,7 @@
 		/* Now wait for a signal or packet to arrive */
 
 		allow_signal (SIGHUP);
-		timeout=-1;
+		timeout = -1;
 		ret = do_select(n + 1, &sockets_fds, &timeout);
 		disallow_signal (SIGHUP);
 		/* SIGHUP is sent by kcom_send to wake up kcomd so it can send the packet*/
@@ -582,7 +624,7 @@
 			continue;
 
 		} else if (ret < 0) {/* -1=error; 0=signal*/
-			printk(KERN_ERR "do_select returned an error. \n");
+			OMBUG("do_select returned an error. \n");
 			schedule_timeout_interruptible(HZ);
 		}
 
@@ -605,8 +647,7 @@
 			if (test_bit(node->fd, sockets_fds.res_in)) {
 				err=pkt_read(node);
 				if (err!=0) {
-					kcomd_done=1;
-					printk(KERN_ERR "ERROR receiving data.\nQuitting.\n");
+					OMBUG(KERN_ERR "ERROR receiving data => ignoring packet.\n");
 					continue;
 				}
 			}
@@ -615,21 +656,22 @@
 			}
 		}
 		write_unlock(&kcom_nodes_lock);
+
+ 		/* Check if we must leave ... */
+ 		if (kcomd_module_exit) {
+ 			if (0 == kcomd_leave_cluster()) kcomd_done = 1;
+ 		}
 	}
 
 /* DONE*/
-	printk(KERN_INFO "kcomd: cleaning up.\n");
+ 	printk(KERN_INFO "kcomd: openMosix kernel communication daemon cleaning up.\n");
 	if (fd4!=-1) {
+ 		if (lsock4) sock_release(lsock4);
 		sys_close(fd4);
-		if (lsock4) {  /* doesn't like this.  Help????????*/
-			sock_release(lsock4);
-		}
 	}
 
 	if (fd6!=-1) {
-		if (lsock6) { /* doesn't like this.  Help????????*/
-			sock_release(lsock6);
-		}
+ 		if (lsock6) sock_release(lsock6);
 		sys_close(fd6);
 	}
 
@@ -708,19 +750,50 @@
 	return ret;
 }
 
+
+/*
+ * Set the exit condition and signal kcomd
+ * kcomd cleans up after itself
+ */
 static void __exit kcomd_exit(void)
 {
+	int retries = 0;
+
+	if (!kcomd_task) {
+		OMBUG("Unable to find the kcomd_thread task !!\n");
+		return;
+	}
+
+
+	printk(KERN_INFO "openMosix, asking kcomd to exit gracefuly\n");
+
+	kcomd_module_exit = 1;
+
+	send_sig(SIGHUP, kcomd_task, 0);
+
+wait_for_kcomd:
+	if (0 == kcomd_done) {
+		if (120 > retries) {
+			goto kill_kcomd;
+		}
+
+		schedule_timeout(HZ);
+		retries ++;
+		goto wait_for_kcomd;
+	}
+
+	printk(KERN_INFO "openMosix, module unloading ... \n");
+
+	return;
+
+kill_kcomd:
+
+	OMBUG("kcomd did not gracefuly exited, killing it !\n");
+
+	kcomd_done = 1;
+	send_sig(SIGHUP, kcomd_task, 0);
 
-	/*
-	 * Set the exit condition and signal kcomd
-	 * kcomd cleans up after itself
-	 */
-	kcomd_done=1;
-
-	if (kcomd_task)
-		send_sig(SIGHUP, kcomd_task, 0);
-	else
-		printk(KERN_WARNING "Unable to to find kcomd daemon\n");
+	schedule_timeout(HZ);
 
 }