[Cluster-devel] cluster/fence/fence_tool fence_tool.c

[Cluster-devel] cluster/fence/fence_tool fence_tool.c

[rohara]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	STABLE
Changes by:	rohara at sourceware.org	2007-01-05 16:40:24

Modified files:
	fence/fence_tool: fence_tool.c 

Log message:
	Add timeout option for fence_tool leave.

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/fence/fence_tool/fence_tool.c.diff?cvsroot=cluster&only_with_tag=STABLE&r1=1.5.2.6.6.3&r2=1.5.2.6.6.4

--- cluster/fence/fence_tool/fence_tool.c	2006/02/09 15:38:09	1.5.2.6.6.3
+++ cluster/fence/fence_tool/fence_tool.c	2007/01/05 16:40:24	1.5.2.6.6.4
@@ -62,7 +62,7 @@
 int operation;
 int child_wait;
 int quorum_wait = TRUE;
-int fenced_start_timeout = 0;
+int fenced_timeout = 0;
 int signalled = 0;
 int cl_sock;
 char our_name[MAX_CLUSTER_MEMBER_NAME_LEN+1];
@@ -282,9 +282,9 @@
 
 	setup_sock();
 
-       	if (fenced_start_timeout) {
+       	if (fenced_timeout) {
                	signal(SIGALRM, sigalarm_handler);
-               	alarm(fenced_start_timeout);
+               	alarm(fenced_timeout);
        	}
 
 	if (!check_quorum()) {
@@ -373,8 +373,17 @@
 	check_mounted();
 	setup_sock();
 
-	if (!check_quorum())
+	if (fenced_timeout) {
+               	signal(SIGALRM, sigalarm_handler);
+               	alarm(fenced_timeout);
+       	}
+
+	if (!check_quorum()) {
+		if (errno == ETIMEDOUT)
+			printf("%s: Timed out waiting for cluster "
+			       "quorum to form.\n", prog_name);
 		return EXIT_FAILURE;
+	}
 
 	close(cl_sock);
 
@@ -487,7 +496,7 @@
 			break;
 
                 case 't':
-                        fenced_start_timeout = get_int_arg(optchar, optarg);
+                        fenced_timeout = get_int_arg(optchar, optarg);
                         break;
 
 		case 'c':

[Cluster-devel] cluster/fence/fence_tool fence_tool.c

[teigland]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL5
Changes by:	teigland at sourceware.org	2007-11-29 14:46:41

Modified files:
	fence/fence_tool: fence_tool.c 

Log message:
	[sync from HEAD]
	
	clean out some options that were only relevant to rhel4
	remove the monitor option which didn't do anything
	add the dump option to dump the fenced debug buffer
	(group_tool can still do this, but fence_tool wasn't oddly enough
	
	bz 404451

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/fence/fence_tool/fence_tool.c.diff?cvsroot=cluster&only_with_tag=RHEL5&r1=1.23.2.1&r2=1.23.2.2

--- cluster/fence/fence_tool/fence_tool.c	2007/01/23 16:54:09	1.23.2.1
+++ cluster/fence/fence_tool/fence_tool.c	2007/11/29 14:46:41	1.23.2.2
@@ -37,14 +37,18 @@
 #define FALSE 0
 #endif
 
-#define OPTION_STRING			("Vhcj:f:t:wQ")
+#define OPTION_STRING			("Vht:wQ")
 #define FENCED_SOCK_PATH                "fenced_socket"
 #define MAXLINE				256
 
 #define OP_JOIN  			1
 #define OP_LEAVE 			2
-#define OP_MONITOR			3
-#define OP_WAIT				4
+#define OP_WAIT				3
+#define OP_DUMP				4
+
+/* needs to match the same in cluster/group/daemon/gd_internal.h and
+   cluster/group/gfs_controld/lock_dlm.h and cluster/fence/fenced/fd.h */
+#define DUMP_SIZE                       (1024 * 1024)
 
 #define die(fmt, args...) \
 do \
@@ -63,10 +67,47 @@
 int signalled = 0;
 cman_handle_t ch;
 
+static int do_write(int fd, void *buf, size_t count)
+{
+	int rv, off = 0;
+
+ retry:
+	rv = write(fd, buf + off, count);
+	if (rv == -1 && errno == EINTR)
+		goto retry;
+	if (rv < 0)
+		return rv;
+
+	if (rv != count) {
+		count -= rv;
+		off += rv;
+		goto retry;
+	}
+	return 0;
+}
+
+static int do_read(int fd, void *buf, size_t count)
+{
+	int rv, off = 0;
+
+	while (off < count) {
+		rv = read(fd, buf + off, count - off);
+		if (rv == 0)
+			return -1;
+		if (rv == -1 && errno == EINTR)
+			continue;
+		if (rv == -1)
+			return -1;
+		off += rv;
+	}
+	return 0;
+}
+
 static int get_int_arg(char argopt, char *arg)
 {
 	char *tmp;
-	int val;                                                                                 
+	int val;
+
 	val = strtol(arg, &tmp, 10);
 	if (tmp == arg || tmp != arg + strlen(arg))
 		die("argument to %c (%s) is not an integer", argopt, arg);
@@ -195,7 +236,7 @@
 	for (i=0; !fenced_start_timeout || i < fenced_start_timeout; i++) {
 		if (we_are_in_fence_domain() == joining)
 			return 0;
-		if (!(i % 5))
+		if (i && !(i % 5))
 			printf("Waiting for fenced to %s the fence group.\n",
 				   (joining?"join":"leave"));
 		sleep(1);
@@ -281,41 +322,43 @@
 	return EXIT_SUCCESS;
 }
 
-static int do_monitor(void)
+static int do_dump(void)
 {
+	char inbuf[DUMP_SIZE];
+	char outbuf[MAXLINE];
 	int fd, rv;
-	char *out, buf[256];
 
 	fd = fenced_connect();
-	if (!fd)
-		die("fenced not running");
 
-	out = "monitor";
+	memset(inbuf, 0, sizeof(inbuf));
+	memset(outbuf, 0, sizeof(outbuf));
+
+	sprintf(outbuf, "dump");
 
-	rv = write(fd, out, sizeof(out));
+	rv = do_write(fd, outbuf, sizeof(outbuf));
 	if (rv < 0)
 		die("can't communicate with fenced");
 
-	while (1) {
-		memset(buf, 0, sizeof(buf));
-		rv = read(fd, buf, sizeof(buf));
-		printf("%s", buf);
-	}
+	rv = do_read(fd, inbuf, sizeof(inbuf));
+	if (rv < 0)
+		printf("dump read: %s\n", strerror(errno));
+
+	do_write(STDOUT_FILENO, inbuf, sizeof(inbuf));
 
 	close(fd);
-	return EXIT_SUCCESS;
+	return 0;
 }
 
 static void print_usage(void)
 {
 	printf("Usage:\n");
 	printf("\n");
-	printf("%s <join|leave|wait> [options]\n", prog_name);
+	printf("%s <join|leave|dump> [options]\n", prog_name);
 	printf("\n");
 	printf("Actions:\n");
 	printf("  join             Join the default fence domain\n");
 	printf("  leave            Leave default fence domain\n");
-	printf("  wait             Wait for node to be member of default fence domain\n");
+	printf("  dump		   Dump debug buffer from fenced\n");
 	printf("\n");
 	printf("Options:\n");
 	printf("  -w               Wait for join to complete\n");
@@ -324,12 +367,6 @@
 	printf("  -t               Maximum time in seconds to wait\n");
 	printf("  -Q               Fail if cluster is not quorate, don't wait\n");
 	printf("\n");
-	printf("Fenced options:\n");
-	printf("  these are passed on to fenced when it's started\n");
-	printf("  -c               All nodes are in a clean state to start\n");
-	printf("  -j <secs>        Post-join fencing delay\n");
-	printf("  -f <secs>        Post-fail fencing delay\n");
-	printf("\n");
 }
 
 static void decode_arguments(int argc, char *argv[])
@@ -376,12 +413,6 @@
 			fenced_start_timeout = get_int_arg(optchar, optarg);
 			break;
 
-		case 'c':
-		case 'j':
-		case 'f':
-			/* Do nothing, just pass these options on to fenced */
-			break;
-
 		default:
 			die("unknown option: %c\n", optchar);
 			break;
@@ -393,8 +424,8 @@
 			operation = OP_JOIN;
 		} else if (strcmp(argv[optind], "leave") == 0) {
 			operation = OP_LEAVE;
-		} else if (strcmp(argv[optind], "monitor") == 0) {
-			operation = OP_MONITOR;
+		} else if (strcmp(argv[optind], "dump") == 0) {
+			operation = OP_DUMP;
 		} else
 			die("unknown option %s\n", argv[optind]);
 		optind++;
@@ -415,8 +446,8 @@
 		return do_join(argc, argv);
 	case OP_LEAVE:
 		return do_leave();
-	case OP_MONITOR:
-		return do_monitor();
+	case OP_DUMP:
+		return do_dump();
 	case OP_WAIT:
 		return -1;
 	}

[Cluster-devel] cluster/fence/fence_tool fence_tool.c

[teigland]

CVSROOT:	/cvs/cluster
Module name:	cluster
Changes by:	teigland at sourceware.org	2007-08-15 20:57:28

Modified files:
	fence/fence_tool: fence_tool.c 

Log message:
	clean out some options that were only relevant to rhel4
	remove the monitor option which didn't do anything
	add the dump option to dump the fenced debug buffer
	(group_tool can still do this, but fence_tool wasn't oddly enough)

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/fence/fence_tool/fence_tool.c.diff?cvsroot=cluster&r1=1.24&r2=1.25

--- cluster/fence/fence_tool/fence_tool.c	2007/01/23 16:53:28	1.24
+++ cluster/fence/fence_tool/fence_tool.c	2007/08/15 20:57:28	1.25
@@ -37,14 +37,18 @@
 #define FALSE 0
 #endif
 
-#define OPTION_STRING			("Vhcj:f:t:wQ")
+#define OPTION_STRING			("Vht:wQ")
 #define FENCED_SOCK_PATH                "fenced_socket"
 #define MAXLINE				256
 
 #define OP_JOIN  			1
 #define OP_LEAVE 			2
-#define OP_MONITOR			3
-#define OP_WAIT				4
+#define OP_WAIT				3
+#define OP_DUMP				4
+
+/* needs to match the same in cluster/group/daemon/gd_internal.h and
+   cluster/group/gfs_controld/lock_dlm.h and cluster/fence/fenced/fd.h */
+#define DUMP_SIZE                       (1024 * 1024)
 
 #define die(fmt, args...) \
 do \
@@ -63,10 +67,47 @@
 int signalled = 0;
 cman_handle_t ch;
 
+static int do_write(int fd, void *buf, size_t count)
+{
+	int rv, off = 0;
+
+ retry:
+	rv = write(fd, buf + off, count);
+	if (rv == -1 && errno == EINTR)
+		goto retry;
+	if (rv < 0)
+		return rv;
+
+	if (rv != count) {
+		count -= rv;
+		off += rv;
+		goto retry;
+	}
+	return 0;
+}
+
+static int do_read(int fd, void *buf, size_t count)
+{
+	int rv, off = 0;
+
+	while (off < count) {
+		rv = read(fd, buf + off, count - off);
+		if (rv == 0)
+			return -1;
+		if (rv == -1 && errno == EINTR)
+			continue;
+		if (rv == -1)
+			return -1;
+		off += rv;
+	}
+	return 0;
+}
+
 static int get_int_arg(char argopt, char *arg)
 {
 	char *tmp;
-	int val;                                                                                 
+	int val;
+
 	val = strtol(arg, &tmp, 10);
 	if (tmp == arg || tmp != arg + strlen(arg))
 		die("argument to %c (%s) is not an integer", argopt, arg);
@@ -195,7 +236,7 @@
 	for (i=0; !fenced_start_timeout || i < fenced_start_timeout; i++) {
 		if (we_are_in_fence_domain() == joining)
 			return 0;
-		if (!(i % 5))
+		if (i && !(i % 5))
 			printf("Waiting for fenced to %s the fence group.\n",
 				   (joining?"join":"leave"));
 		sleep(1);
@@ -281,41 +322,43 @@
 	return EXIT_SUCCESS;
 }
 
-static int do_monitor(void)
+static int do_dump(void)
 {
+	char inbuf[DUMP_SIZE];
+	char outbuf[MAXLINE];
 	int fd, rv;
-	char *out, buf[256];
 
 	fd = fenced_connect();
-	if (!fd)
-		die("fenced not running");
 
-	out = "monitor";
+	memset(inbuf, 0, sizeof(inbuf));
+	memset(outbuf, 0, sizeof(outbuf));
+
+	sprintf(outbuf, "dump");
 
-	rv = write(fd, out, sizeof(out));
+	rv = do_write(fd, outbuf, sizeof(outbuf));
 	if (rv < 0)
 		die("can't communicate with fenced");
 
-	while (1) {
-		memset(buf, 0, sizeof(buf));
-		rv = read(fd, buf, sizeof(buf));
-		printf("%s", buf);
-	}
+	rv = do_read(fd, inbuf, sizeof(inbuf));
+	if (rv < 0)
+		printf("dump read: %s\n", strerror(errno));
+
+	do_write(STDOUT_FILENO, inbuf, sizeof(inbuf));
 
 	close(fd);
-	return EXIT_SUCCESS;
+	return 0;
 }
 
 static void print_usage(void)
 {
 	printf("Usage:\n");
 	printf("\n");
-	printf("%s <join|leave|wait> [options]\n", prog_name);
+	printf("%s <join|leave|dump> [options]\n", prog_name);
 	printf("\n");
 	printf("Actions:\n");
 	printf("  join             Join the default fence domain\n");
 	printf("  leave            Leave default fence domain\n");
-	printf("  wait             Wait for node to be member of default fence domain\n");
+	printf("  dump		   Dump debug buffer from fenced\n");
 	printf("\n");
 	printf("Options:\n");
 	printf("  -w               Wait for join to complete\n");
@@ -324,12 +367,6 @@
 	printf("  -t               Maximum time in seconds to wait\n");
 	printf("  -Q               Fail if cluster is not quorate, don't wait\n");
 	printf("\n");
-	printf("Fenced options:\n");
-	printf("  these are passed on to fenced when it's started\n");
-	printf("  -c               All nodes are in a clean state to start\n");
-	printf("  -j <secs>        Post-join fencing delay\n");
-	printf("  -f <secs>        Post-fail fencing delay\n");
-	printf("\n");
 }
 
 static void decode_arguments(int argc, char *argv[])
@@ -376,12 +413,6 @@
 			fenced_start_timeout = get_int_arg(optchar, optarg);
 			break;
 
-		case 'c':
-		case 'j':
-		case 'f':
-			/* Do nothing, just pass these options on to fenced */
-			break;
-
 		default:
 			die("unknown option: %c\n", optchar);
 			break;
@@ -393,8 +424,8 @@
 			operation = OP_JOIN;
 		} else if (strcmp(argv[optind], "leave") == 0) {
 			operation = OP_LEAVE;
-		} else if (strcmp(argv[optind], "monitor") == 0) {
-			operation = OP_MONITOR;
+		} else if (strcmp(argv[optind], "dump") == 0) {
+			operation = OP_DUMP;
 		} else
 			die("unknown option %s\n", argv[optind]);
 		optind++;
@@ -415,8 +446,8 @@
 		return do_join(argc, argv);
 	case OP_LEAVE:
 		return do_leave();
-	case OP_MONITOR:
-		return do_monitor();
+	case OP_DUMP:
+		return do_dump();
 	case OP_WAIT:
 		return -1;
 	}

[Cluster-devel] cluster/fence/fence_tool fence_tool.c

[rpeterso]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL50
Changes by:	rpeterso at sourceware.org	2007-01-23 17:21:11

Modified files:
	fence/fence_tool: fence_tool.c 

Log message:
	Resolves: bz 222933: regression: fence_tool no longer times out
	after 300 seconds

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/fence/fence_tool/fence_tool.c.diff?cvsroot=cluster&only_with_tag=RHEL50&r1=1.23&r2=1.23.4.1

--- cluster/fence/fence_tool/fence_tool.c	2006/10/13 14:57:55	1.23
+++ cluster/fence/fence_tool/fence_tool.c	2007/01/23 17:21:11	1.23.4.1
@@ -2,7 +2,7 @@
 *******************************************************************************
 **
 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
-**  Copyright (C) 2004 Red Hat, Inc.  All rights reserved.
+**  Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
 **  
 **  This copyrighted material is made available to anyone wishing to use,
 **  modify, copy, or redistribute it subject to the terms and conditions
@@ -29,6 +29,7 @@
 
 #include "ccs.h"
 #include "copyright.cf"
+#include "libcman.h"
 #include "libgroup.h"
 
 #ifndef TRUE
@@ -36,7 +37,7 @@
 #define FALSE 0
 #endif
 
-#define OPTION_STRING			("Vhcj:f:t:w")
+#define OPTION_STRING			("Vhcj:f:t:wQ")
 #define FENCED_SOCK_PATH                "fenced_socket"
 #define MAXLINE				256
 
@@ -57,7 +58,10 @@
 char *prog_name;
 int operation;
 int child_wait = FALSE;
+int quorum_wait = TRUE;
 int fenced_start_timeout = 300; /* five minutes */
+int signalled = 0;
+cman_handle_t ch;
 
 static int get_int_arg(char argopt, char *arg)
 {
@@ -97,6 +101,11 @@
 	return 0;
 }
 
+static void sigalarm_handler(int sig)
+{
+	signalled = 1;
+}
+
 int fenced_connect(void)
 {
 	struct sockaddr_un sun;
@@ -135,6 +144,50 @@
 	return gdata.member;
 }
 
+/*
+ * We wait for the cluster to be quorate in this program because it's easy to
+ * kill this program if we want to quit waiting.  If we just started fenced
+ * without waiting for quorum, fenced's join would then wait for quorum in SM
+ * but we can't kill/cancel it at that point -- we have to wait for it to
+ * complete.
+ *
+ * A second reason to wait for quorum is that the unfencing step involves
+ * cluster.conf lookups through ccs, but ccsd may wait for the cluster to be
+ * quorate before responding to the lookups.  There wouldn't be a problem
+ * blocking there per se, but it's cleaner I think to just wait here first.
+ *
+ * In the case where we're leaving, we want to wait for quorum because if we go
+ * ahead and shut down fenced, the fence domain leave will block in SM where it
+ * will wait for quorum before the leave can be processed.  We can't
+ * kill/cancel the leave at that point, but we can if we're waiting here.
+ *
+ * Waiting here doesn't guarantee we won't end up blocking in SM on the join or
+ * leave, but it avoids it in some common cases which can be helpful.  (Quorum
+ * could easily be lost between the time we wait for it here and then begin the
+ * join/leave process.)
+ */
+
+static int check_quorum(void)
+{
+	int rv = 0, i = 0;
+
+	while (!signalled) {
+		rv = cman_is_quorate(ch);
+		if (rv)
+			return TRUE;
+		else if (!quorum_wait)
+			return FALSE;
+
+		sleep(1);
+
+		if (!signalled && ++i > 9 && !(i % 10))
+			printf("%s: waiting for cluster quorum\n", prog_name);
+	}
+
+	errno = ETIMEDOUT;
+	return FALSE;
+}
+
 static int do_wait(int joining)
 {
 	int i;
@@ -156,6 +209,22 @@
 	int i, fd, rv;
 	char buf[MAXLINE];
 
+	ch = cman_init(NULL);
+
+	if (fenced_start_timeout) {
+		signal(SIGALRM, sigalarm_handler);
+		alarm(fenced_start_timeout);
+	}
+
+	if (!check_quorum()) {
+		if (errno == ETIMEDOUT)
+			printf("%s: Timed out waiting for cluster "
+			       "quorum to form.\n", prog_name);
+		cman_finish(ch);
+		return EXIT_FAILURE;
+	}
+	cman_finish(ch);
+
 	i = 0;
 	do {
 		sleep(1);
@@ -253,6 +322,7 @@
 	printf("  -V               Print program version information, then exit\n");
 	printf("  -h               Print this help, then exit\n");
 	printf("  -t               Maximum time in seconds to wait\n");
+	printf("  -Q               Fail if cluster is not quorate, don't wait\n");
 	printf("\n");
 	printf("Fenced options:\n");
 	printf("  these are passed on to fenced when it's started\n");
@@ -284,6 +354,10 @@
 			exit(EXIT_SUCCESS);
 			break;
 
+		case 'Q':
+			quorum_wait = FALSE;
+			break;
+
 		case 'w':
 			child_wait = TRUE;
 			break;

[Cluster-devel] cluster/fence/fence_tool fence_tool.c

[rpeterso]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL5
Changes by:	rpeterso at sourceware.org	2007-01-23 16:54:09

Modified files:
	fence/fence_tool: fence_tool.c 

Log message:
	Resolves: bz 222933: regression: fence_tool no longer times out
	after 300 seconds

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/fence/fence_tool/fence_tool.c.diff?cvsroot=cluster&only_with_tag=RHEL5&r1=1.23&r2=1.23.2.1

--- cluster/fence/fence_tool/fence_tool.c	2006/10/13 14:57:55	1.23
+++ cluster/fence/fence_tool/fence_tool.c	2007/01/23 16:54:09	1.23.2.1
@@ -2,7 +2,7 @@
 *******************************************************************************
 **
 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
-**  Copyright (C) 2004 Red Hat, Inc.  All rights reserved.
+**  Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
 **  
 **  This copyrighted material is made available to anyone wishing to use,
 **  modify, copy, or redistribute it subject to the terms and conditions
@@ -29,6 +29,7 @@
 
 #include "ccs.h"
 #include "copyright.cf"
+#include "libcman.h"
 #include "libgroup.h"
 
 #ifndef TRUE
@@ -36,7 +37,7 @@
 #define FALSE 0
 #endif
 
-#define OPTION_STRING			("Vhcj:f:t:w")
+#define OPTION_STRING			("Vhcj:f:t:wQ")
 #define FENCED_SOCK_PATH                "fenced_socket"
 #define MAXLINE				256
 
@@ -57,7 +58,10 @@
 char *prog_name;
 int operation;
 int child_wait = FALSE;
+int quorum_wait = TRUE;
 int fenced_start_timeout = 300; /* five minutes */
+int signalled = 0;
+cman_handle_t ch;
 
 static int get_int_arg(char argopt, char *arg)
 {
@@ -97,6 +101,11 @@
 	return 0;
 }
 
+static void sigalarm_handler(int sig)
+{
+	signalled = 1;
+}
+
 int fenced_connect(void)
 {
 	struct sockaddr_un sun;
@@ -135,6 +144,50 @@
 	return gdata.member;
 }
 
+/*
+ * We wait for the cluster to be quorate in this program because it's easy to
+ * kill this program if we want to quit waiting.  If we just started fenced
+ * without waiting for quorum, fenced's join would then wait for quorum in SM
+ * but we can't kill/cancel it at that point -- we have to wait for it to
+ * complete.
+ *
+ * A second reason to wait for quorum is that the unfencing step involves
+ * cluster.conf lookups through ccs, but ccsd may wait for the cluster to be
+ * quorate before responding to the lookups.  There wouldn't be a problem
+ * blocking there per se, but it's cleaner I think to just wait here first.
+ *
+ * In the case where we're leaving, we want to wait for quorum because if we go
+ * ahead and shut down fenced, the fence domain leave will block in SM where it
+ * will wait for quorum before the leave can be processed.  We can't
+ * kill/cancel the leave at that point, but we can if we're waiting here.
+ *
+ * Waiting here doesn't guarantee we won't end up blocking in SM on the join or
+ * leave, but it avoids it in some common cases which can be helpful.  (Quorum
+ * could easily be lost between the time we wait for it here and then begin the
+ * join/leave process.)
+ */
+
+static int check_quorum(void)
+{
+	int rv = 0, i = 0;
+
+	while (!signalled) {
+		rv = cman_is_quorate(ch);
+		if (rv)
+			return TRUE;
+		else if (!quorum_wait)
+			return FALSE;
+
+		sleep(1);
+
+		if (!signalled && ++i > 9 && !(i % 10))
+			printf("%s: waiting for cluster quorum\n", prog_name);
+	}
+
+	errno = ETIMEDOUT;
+	return FALSE;
+}
+
 static int do_wait(int joining)
 {
 	int i;
@@ -156,6 +209,22 @@
 	int i, fd, rv;
 	char buf[MAXLINE];
 
+	ch = cman_init(NULL);
+
+	if (fenced_start_timeout) {
+		signal(SIGALRM, sigalarm_handler);
+		alarm(fenced_start_timeout);
+	}
+
+	if (!check_quorum()) {
+		if (errno == ETIMEDOUT)
+			printf("%s: Timed out waiting for cluster "
+			       "quorum to form.\n", prog_name);
+		cman_finish(ch);
+		return EXIT_FAILURE;
+	}
+	cman_finish(ch);
+
 	i = 0;
 	do {
 		sleep(1);
@@ -253,6 +322,7 @@
 	printf("  -V               Print program version information, then exit\n");
 	printf("  -h               Print this help, then exit\n");
 	printf("  -t               Maximum time in seconds to wait\n");
+	printf("  -Q               Fail if cluster is not quorate, don't wait\n");
 	printf("\n");
 	printf("Fenced options:\n");
 	printf("  these are passed on to fenced when it's started\n");
@@ -284,6 +354,10 @@
 			exit(EXIT_SUCCESS);
 			break;
 
+		case 'Q':
+			quorum_wait = FALSE;
+			break;
+
 		case 'w':
 			child_wait = TRUE;
 			break;

[Cluster-devel] cluster/fence/fence_tool fence_tool.c

[rpeterso]

CVSROOT:	/cvs/cluster
Module name:	cluster
Changes by:	rpeterso at sourceware.org	2007-01-23 16:53:29

Modified files:
	fence/fence_tool: fence_tool.c 

Log message:
	Resolves: bz 222933: regression: fence_tool no longer times out
	after 300 seconds

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/fence/fence_tool/fence_tool.c.diff?cvsroot=cluster&r1=1.23&r2=1.24

--- cluster/fence/fence_tool/fence_tool.c	2006/10/13 14:57:55	1.23
+++ cluster/fence/fence_tool/fence_tool.c	2007/01/23 16:53:28	1.24
@@ -2,7 +2,7 @@
 *******************************************************************************
 **
 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
-**  Copyright (C) 2004 Red Hat, Inc.  All rights reserved.
+**  Copyright (C) 2004-2007 Red Hat, Inc.  All rights reserved.
 **  
 **  This copyrighted material is made available to anyone wishing to use,
 **  modify, copy, or redistribute it subject to the terms and conditions
@@ -29,6 +29,7 @@
 
 #include "ccs.h"
 #include "copyright.cf"
+#include "libcman.h"
 #include "libgroup.h"
 
 #ifndef TRUE
@@ -36,7 +37,7 @@
 #define FALSE 0
 #endif
 
-#define OPTION_STRING			("Vhcj:f:t:w")
+#define OPTION_STRING			("Vhcj:f:t:wQ")
 #define FENCED_SOCK_PATH                "fenced_socket"
 #define MAXLINE				256
 
@@ -57,7 +58,10 @@
 char *prog_name;
 int operation;
 int child_wait = FALSE;
+int quorum_wait = TRUE;
 int fenced_start_timeout = 300; /* five minutes */
+int signalled = 0;
+cman_handle_t ch;
 
 static int get_int_arg(char argopt, char *arg)
 {
@@ -97,6 +101,11 @@
 	return 0;
 }
 
+static void sigalarm_handler(int sig)
+{
+	signalled = 1;
+}
+
 int fenced_connect(void)
 {
 	struct sockaddr_un sun;
@@ -135,6 +144,50 @@
 	return gdata.member;
 }
 
+/*
+ * We wait for the cluster to be quorate in this program because it's easy to
+ * kill this program if we want to quit waiting.  If we just started fenced
+ * without waiting for quorum, fenced's join would then wait for quorum in SM
+ * but we can't kill/cancel it at that point -- we have to wait for it to
+ * complete.
+ *
+ * A second reason to wait for quorum is that the unfencing step involves
+ * cluster.conf lookups through ccs, but ccsd may wait for the cluster to be
+ * quorate before responding to the lookups.  There wouldn't be a problem
+ * blocking there per se, but it's cleaner I think to just wait here first.
+ *
+ * In the case where we're leaving, we want to wait for quorum because if we go
+ * ahead and shut down fenced, the fence domain leave will block in SM where it
+ * will wait for quorum before the leave can be processed.  We can't
+ * kill/cancel the leave at that point, but we can if we're waiting here.
+ *
+ * Waiting here doesn't guarantee we won't end up blocking in SM on the join or
+ * leave, but it avoids it in some common cases which can be helpful.  (Quorum
+ * could easily be lost between the time we wait for it here and then begin the
+ * join/leave process.)
+ */
+
+static int check_quorum(void)
+{
+	int rv = 0, i = 0;
+
+	while (!signalled) {
+		rv = cman_is_quorate(ch);
+		if (rv)
+			return TRUE;
+		else if (!quorum_wait)
+			return FALSE;
+
+		sleep(1);
+
+		if (!signalled && ++i > 9 && !(i % 10))
+			printf("%s: waiting for cluster quorum\n", prog_name);
+	}
+
+	errno = ETIMEDOUT;
+	return FALSE;
+}
+
 static int do_wait(int joining)
 {
 	int i;
@@ -156,6 +209,22 @@
 	int i, fd, rv;
 	char buf[MAXLINE];
 
+	ch = cman_init(NULL);
+
+	if (fenced_start_timeout) {
+		signal(SIGALRM, sigalarm_handler);
+		alarm(fenced_start_timeout);
+	}
+
+	if (!check_quorum()) {
+		if (errno == ETIMEDOUT)
+			printf("%s: Timed out waiting for cluster "
+			       "quorum to form.\n", prog_name);
+		cman_finish(ch);
+		return EXIT_FAILURE;
+	}
+	cman_finish(ch);
+
 	i = 0;
 	do {
 		sleep(1);
@@ -253,6 +322,7 @@
 	printf("  -V               Print program version information, then exit\n");
 	printf("  -h               Print this help, then exit\n");
 	printf("  -t               Maximum time in seconds to wait\n");
+	printf("  -Q               Fail if cluster is not quorate, don't wait\n");
 	printf("\n");
 	printf("Fenced options:\n");
 	printf("  these are passed on to fenced when it's started\n");
@@ -284,6 +354,10 @@
 			exit(EXIT_SUCCESS);
 			break;
 
+		case 'Q':
+			quorum_wait = FALSE;
+			break;
+
 		case 'w':
 			child_wait = TRUE;
 			break;

[Cluster-devel] cluster/fence/fence_tool fence_tool.c

[rohara]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4
Changes by:	rohara at sourceware.org	2007-01-05 16:44:53

Modified files:
	fence/fence_tool: fence_tool.c 

Log message:
	Remove extra check_quorum call.

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/fence/fence_tool/fence_tool.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.5.2.15&r2=1.5.2.16

--- cluster/fence/fence_tool/fence_tool.c	2007/01/05 16:24:29	1.5.2.15
+++ cluster/fence/fence_tool/fence_tool.c	2007/01/05 16:44:53	1.5.2.16
@@ -385,9 +385,6 @@
 		return EXIT_FAILURE;
 	}
 
-	if (!check_quorum())
-		return EXIT_FAILURE;
-
 	close(cl_sock);
 
 	kill(pid, SIGTERM);

[Cluster-devel] cluster/fence/fence_tool fence_tool.c

[rohara]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4
Changes by:	rohara at sourceware.org	2007-01-05 16:24:30

Modified files:
	fence/fence_tool: fence_tool.c 

Log message:
	Add timeout option for fence_tool leave.

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/fence/fence_tool/fence_tool.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.5.2.14&r2=1.5.2.15

--- cluster/fence/fence_tool/fence_tool.c	2006/12/20 18:14:29	1.5.2.14
+++ cluster/fence/fence_tool/fence_tool.c	2007/01/05 16:24:29	1.5.2.15
@@ -62,7 +62,7 @@
 int operation;
 int child_wait;
 int quorum_wait = TRUE;
-int fenced_start_timeout = 0;
+int fenced_timeout = 0;
 int signalled = 0;
 int cl_sock;
 char our_name[MAX_CLUSTER_MEMBER_NAME_LEN+1];
@@ -282,9 +282,9 @@
 
 	setup_sock();
 
-       	if (fenced_start_timeout) {
+       	if (fenced_timeout) {
                	signal(SIGALRM, sigalarm_handler);
-               	alarm(fenced_start_timeout);
+               	alarm(fenced_timeout);
        	}
 
 	if (!check_quorum()) {
@@ -373,6 +373,18 @@
 	check_mounted();
 	setup_sock();
 
+	if (fenced_timeout) {
+               	signal(SIGALRM, sigalarm_handler);
+               	alarm(fenced_timeout);
+       	}
+
+	if (!check_quorum()) {
+		if (errno == ETIMEDOUT)
+			printf("%s: Timed out waiting for cluster "
+			       "quorum to form.\n", prog_name);
+		return EXIT_FAILURE;
+	}
+
 	if (!check_quorum())
 		return EXIT_FAILURE;
 
@@ -490,7 +502,7 @@
 			break;
 
                 case 't':
-                        fenced_start_timeout = get_int_arg(optchar, optarg);
+                        fenced_timeout = get_int_arg(optchar, optarg);
                         break;
 
 		case 'c':

[Cluster-devel] cluster/fence/fence_tool fence_tool.c

[jparsons]

CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4
Changes by:	jparsons at sourceware.org	2006-10-23 16:23:56

Modified files:
	fence/fence_tool: fence_tool.c 

Log message:
	This fixes bz 203804, wherein pid files are left around. Thx jlayton.

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/fence/fence_tool/fence_tool.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.5.2.12&r2=1.5.2.13

--- cluster/fence/fence_tool/fence_tool.c	2006/02/09 15:25:11	1.5.2.12
+++ cluster/fence/fence_tool/fence_tool.c	2006/10/23 16:23:56	1.5.2.13
@@ -357,7 +357,7 @@
 {
 	FILE *f;
 	char buf[33] = "";
-	int pid = 0;
+	int pid = 0, error;
 
 	lockfile();
 
@@ -379,7 +379,10 @@
 	close(cl_sock);
 
 	kill(pid, SIGTERM);
-
+	error = unlink(LOCKFILE_NAME);
+	if (error)
+		die("signal sent, but unable to unlink %s: %s", LOCKFILE_NAME,
+			strerror(error)); 
 	return EXIT_SUCCESS;
 }
 

[Cluster-devel] cluster/fence/fence_tool fence_tool.c

[rohara]

CVSROOT:	/cvs/cluster
Module name:	cluster
Changes by:	rohara at sourceware.org	2006-07-10 17:04:08

Modified files:
	fence/fence_tool: fence_tool.c 

Log message:
	Added "self" parament to dispatch_fence_agent.
	
	Needed for SCSI persistent reservation (fence_scsi).

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/fence/fence_tool/fence_tool.c.diff?cvsroot=cluster&r1=1.20&r2=1.21

--- cluster/fence/fence_tool/fence_tool.c	2006/02/15 22:00:41	1.20
+++ cluster/fence/fence_tool/fence_tool.c	2006/07/10 17:04:08	1.21
@@ -56,7 +56,7 @@
 char *prog_name;
 int operation;
 
-int dispatch_fence_agent(int cd, char *victim);
+int dispatch_fence_agent(int cd, char *victim, char *self);
 
 static int check_mounted(void)
 {