cluster: STABLE3 - cman: drastically improve startup errors

[Christine Caulfield <chrissie at fedoraproject dot org>]

Gitweb:        http://git.fedorahosted.org/git/cluster.git?p=cluster.git;a=commitdiff;h=1205c26952556376284a1092bd89ed530645f98f
Commit:        1205c26952556376284a1092bd89ed530645f98f
Parent:        a2ba1474b8ed32a6eb796da2c0630816f3d2087d
Author:        Christine Caulfield <ccaulfie@redhat.com>
AuthorDate:    Thu Dec 18 10:03:32 2008 +0000
Committer:     Christine Caulfield <ccaulfie@redhat.com>
CommitterDate: Thu Dec 18 10:03:32 2008 +0000

cman: drastically improve startup errors

cman_tool join has a nasty habit of just exiting with
"corosync failed to start" or some such unhelpful error message.

This patch improves on these by trapping the corosync exit code and
attempting to interpret it for the user.

Signed-off-by: Christine Caulfield <ccaulfie@redhat.com>
---
 cman/cman_tool/join.c |  116 ++++++++++++++++++++++++++++++++++++++-----------
 1 files changed, 91 insertions(+), 25 deletions(-)

diff --git a/cman/cman_tool/join.c b/cman/cman_tool/join.c
index 72462d2..cef6619 100644
--- a/cman/cman_tool/join.c
+++ b/cman/cman_tool/join.c
@@ -39,6 +39,77 @@ static void be_daemon(int close_stderr)
 	setsid();
 }
 
+
+static char *corosync_exit_reason(signed char status)
+{
+	static char reason[256];
+	switch (status) {
+	case -2:
+		return "Could not determine UID to run as";
+		break;
+	case -3:
+		return "Could not determine GID to run as";
+		break;
+	case -4:
+		return "Error initialising memory pool";
+		break;
+	case -5:
+		return "Could not fork";
+		break;
+	case -6:
+		return "Could not bind to libais socket";
+		break;
+	case -7:
+		return "Could not bind to network socket";
+		break;
+	case -8:
+		return "Could not read security key for communications";
+		break;
+	case -9:
+		return "Could not read cluster configuration";
+		break;
+	case -10:
+		return "Could not set up logging";
+		break;
+	case -11:
+		return "Could not dynamically load modules";
+		break;
+	case -12:
+		return "Could not load and initialise object database";
+		break;
+	case -13:
+		return "Could not initialise all required services";
+		break;
+	case -14:
+		return "Out of memory";
+		break;
+	default:
+		sprintf(reason, "Error, reason code is %d", status);
+		return reason;
+		break;
+	}
+}
+
+static int check_corosync_status(pid_t pid)
+{
+	int status;
+	int pidstatus;
+
+	status = waitpid(pid, &pidstatus, WNOHANG);
+	if (status == -1 && errno == ECHILD) {
+
+		return 0;
+	}
+	if ((status == 0 || status == pid) && pidstatus != 0) {
+		if (WIFEXITED(pidstatus))
+			fprintf(stderr, "corosync died: %s\n", corosync_exit_reason(WEXITSTATUS(pidstatus)));
+		if (WIFSIGNALED(pidstatus))
+			fprintf(stderr, "corosync died with signal: %d\n", WTERMSIG(pidstatus));
+		exit(1);
+	}
+	return status;
+}
+
 int join(commandline_t *comline, char *main_envp[])
 {
 	int i, err;
@@ -127,9 +198,10 @@ int join(commandline_t *comline, char *main_envp[])
 	envp[envptr++] = strdup(scratch);
 	envp[envptr++] = NULL;
 
+	/* Always run corosync -f because we have already forked twice anyway, and
+	   we want to return any exit code that might happen */
 	argv[0] = "corosync";
-	if (comline->verbose & ~DEBUG_STARTUP_ONLY)
-		argv[++argvptr] = "-f";
+	argv[++argvptr] = "-f";
 	if (comline->nosetpri_opt)
 		argv[++argvptr] = "-p";
 	argv[++argvptr] = NULL;
@@ -153,6 +225,10 @@ int join(commandline_t *comline, char *main_envp[])
 			}
 		}
 		be_daemon(!(comline->verbose & ~DEBUG_STARTUP_ONLY));
+
+		sprintf(scratch, "FORKED: %d", getpid());
+		err = write(p[1], scratch, strlen(scratch));
+
 		execve(COROSYNCBIN, argv, envp);
 
 		/* exec failed - tell the parent process */
@@ -182,43 +258,31 @@ int join(commandline_t *comline, char *main_envp[])
 
 		status = select(p[0]+1, &fds, NULL, NULL, &tv);
 
-		/* Did we get an error? */
+		/* Did we get a cman-reported error? */
 		if (status == 1) {
 			int len;
 			if ((len = read(p[0], message, sizeof(message)) > 0)) {
 
+				/* Forked OK - get the real corosync pid */
+				if (sscanf(message, "FORKED: %d", &corosync_pid) == 1) {
+					if (comline->verbose & DEBUG_STARTUP_ONLY)
+						fprintf(stderr, "forked process ID is %d\n", corosync_pid);
+					status = 1;
+					continue;
+				}
 				/* Success! get the new PID of double-forked corosync */
 				if (sscanf(message, "SUCCESS: %d", &corosync_pid) == 1) {
 					if (comline->verbose & DEBUG_STARTUP_ONLY)
 						fprintf(stderr, "corosync running, process ID is %d\n", corosync_pid);
 					status = 0;
+					break;
 				}
-				else {
-					fprintf(stderr, "cman not started: %s\n", message);
-				}
-				break;
 			}
 			else if (len < 0 && errno == EINTR) {
 				continue;
 			}
 			else { /* Error or EOF - check the child status */
-				int pidstatus;
-				status = waitpid(corosync_pid, &pidstatus, WNOHANG);
-				if (status == -1 && errno == ECHILD) {
-					fprintf(stderr, "cman not started\n");
-					break;
-				}
-				if (status == 0 && pidstatus != 0) {
-					if (WIFEXITED(pidstatus))
-						fprintf(stderr, "corosync died with status: %d\n", WEXITSTATUS(pidstatus));
-					if (WIFSIGNALED(pidstatus))
-						fprintf(stderr, "corosync died with signal: %d\n", WTERMSIG(pidstatus));
-					status = -1;
-					break;
-				}
-				else {
-					status = 0; /* Try to connect */
-				}
+				status = check_corosync_status(corosync_pid);
 			}
 		}
 
@@ -230,13 +294,15 @@ int join(commandline_t *comline, char *main_envp[])
 		do {
 			if (status == 0) {
 				if (kill(corosync_pid, 0) < 0) {
+					status = check_corosync_status(corosync_pid);
 					die("corosync died during startup\n");
 				}
 
 				h = cman_admin_init(NULL);
 				if (!h && comline->verbose & DEBUG_STARTUP_ONLY)
 				{
-					fprintf(stderr, "waiting for corosync to start\n");
+					fprintf(stderr, "waiting for cman to start\n");
+					status = check_corosync_status(corosync_pid);
 				}
 			}
 			sleep (1);