Fix clvmd startup bug introduced in cman/gulm amalgamation. bz#145729

Improve reporting of node-specific locking errors so you'll get
somthing a little more helpfiul than "host is down" - it will now tell
you /which/ host it thinks is down.

WHATS_NEW

@@ -6,6 +6,8 @@ Version 2.01.02 -
   Split out pool sptype_names to avoid unused const.
   Always fail if random id generation fails.
   Recognise gnbd devices.
+  Fix clvmd startup bug introduced in cman/gulm amalgamation.
+  Improve reporting of node-specific locking errors.
 
 Version 2.01.01 - 19th January 2005
 ===================================

daemons/clvmd/clvm.h

@@ -43,6 +43,7 @@ struct clvm_header {
 /* Flags */
 #define CLVMD_FLAG_LOCAL        1	/* Only do this on the local node */
 #define CLVMD_FLAG_SYSTEMLV     2	/* Data in system LV under my node name */
+#define CLVMD_FLAG_NODEERRS     4       /* Reply has errors in node-specific portion */
 
 /* Name of the local socket to communicate between libclvm and clvmd */
 //static const char CLVMD_SOCKNAME[]="/var/run/clvmd";

daemons/clvmd/clvmd-cman.c

@@ -104,6 +104,11 @@ static int _init_cluster(void)
 	return 0;
 }
 
+static void _cluster_init_completed(void)
+{
+	clvmd_cluster_init_completed();
+}
+
 static int _get_main_cluster_fd()
 {
 	return cluster_sock;
@@ -507,7 +512,7 @@ static int _sync_unlock(const char *resource /* UNUSED */, int lockid)
 }
 
 static struct cluster_ops _cluster_cman_ops = {
-	.cluster_init_completed   = NULL,
+	.cluster_init_completed   = _cluster_init_completed,
 	.cluster_send_message     = _cluster_send_message,
 	.name_from_csid           = _name_from_csid,
 	.csid_from_name           = _csid_from_name,

daemons/clvmd/clvmd-gulm.c

@@ -913,11 +913,6 @@ static int get_all_cluster_nodes()
     return 0;
 }
 
-static void _cluster_init_completed(void)
-{
-	clvmd_cluster_init_completed();
-}
-
 static int _get_main_cluster_fd(void)
 {
 	return get_main_gulm_cluster_fd();
@@ -934,7 +929,7 @@ static int _cluster_send_message(void *buf, int msglen, char *csid, const char *
 }
 
 static struct cluster_ops _cluster_gulm_ops = {
-	.cluster_init_completed   = _cluster_init_completed,
+	.cluster_init_completed   = NULL,
 	.cluster_send_message     = _cluster_send_message,
 	.name_from_csid           = gulm_name_from_csid,
 	.csid_from_name           = _csid_from_name,

daemons/clvmd/clvmd.c

@@ -785,7 +785,7 @@ static int read_from_local_sock(struct local_client *thisfd)
 		if (thisfd->bits.localsock.in_progress) {
 			struct clvm_header reply;
 			reply.cmd = CLVMD_CMD_REPLY;
-			reply.status = -EBUSY;
+			reply.status = EBUSY;
 			reply.arglen = 0;
 			reply.flags = 0;
 			send_message(&reply, sizeof(reply), our_csid,
@@ -808,7 +808,7 @@ static int read_from_local_sock(struct local_client *thisfd)
 		if (!thisfd->bits.localsock.cmd) {
 			struct clvm_header reply;
 			reply.cmd = CLVMD_CMD_REPLY;
-			reply.status = -ENOMEM;
+			reply.status = ENOMEM;
 			reply.arglen = 0;
 			reply.flags = 0;
 			send_message(&reply, sizeof(reply), our_csid,
@@ -855,7 +855,7 @@ static int read_from_local_sock(struct local_client *thisfd)
 			DEBUGLOG("Unknown node: '%s'\n", inheader->node);
 
 			reply.cmd = CLVMD_CMD_REPLY;
-			reply.status = -ENOENT;
+			reply.status = ENOENT;
 			reply.flags = 0;
 			reply.arglen = 0;
 			send_message(&reply, sizeof(reply), our_csid,
@@ -886,7 +886,7 @@ static int read_from_local_sock(struct local_client *thisfd)
 			close(comms_pipe[1]);
 
 			reply.cmd = CLVMD_CMD_REPLY;
-			reply.status = -ENOMEM;
+			reply.status = ENOMEM;
 			reply.arglen = 0;
 			reply.flags = 0;
 			send_message(&reply, sizeof(reply), our_csid,
@@ -1076,7 +1076,7 @@ void process_remote_command(struct clvm_header *msg, int msglen, int fd,
 
 				/* Return a failure response */
 				head.cmd = CLVMD_CMD_REPLY;
-				head.status = -EFBIG;
+				head.status = EFBIG;
 				head.flags = 0;
 				head.clientid = msg->clientid;
 				head.arglen = 0;
@@ -1093,7 +1093,7 @@ void process_remote_command(struct clvm_header *msg, int msglen, int fd,
 			     msg->arglen);
 			/* Return a failure response */
 			head.cmd = CLVMD_CMD_REPLY;
-			head.status = -ENOMEM;
+			head.status = ENOMEM;
 			head.flags = 0;
 			head.clientid = msg->clientid;
 			head.arglen = 0;
@@ -1156,7 +1156,7 @@ void process_remote_command(struct clvm_header *msg, int msglen, int fd,
 		    do_command(NULL, msg, msglen, &replyargs, buflen,
 			       &replylen);
 	} else {
-		status = -ENOMEM;
+		status = ENOMEM;
 	}
 
 	/* If it wasn't a reply, then reply */
@@ -1191,7 +1191,7 @@ void process_remote_command(struct clvm_header *msg, int msglen, int fd,
 				    (aggreply,
 				     replylen + sizeof(struct clvm_header)) < 0
 				    && replylen > 0)
-					agghead->status = -EFBIG;
+					agghead->status = EFBIG;
 
 				send_message(agghead,
 					     sizeof(struct clvm_header), csid,
@@ -1216,7 +1216,7 @@ void process_remote_command(struct clvm_header *msg, int msglen, int fd,
 			DEBUGLOG("Error attempting to realloc return buffer\n");
 			/* Return a failure response */
 			head.cmd = CLVMD_CMD_REPLY;
-			head.status = -ENOMEM;
+			head.status = ENOMEM;
 			head.flags = 0;
 			head.clientid = msg->clientid;
 			head.arglen = 0;
@@ -1254,7 +1254,7 @@ static void add_reply_to_list(struct local_client *client, int status,
 		if (len > 0) {
 			reply->replymsg = malloc(len);
 			if (!reply->replymsg) {
-				reply->status = -ENOMEM;
+				reply->status = ENOMEM;
 			} else {
 				memcpy(reply->replymsg, buf, len);
 			}
@@ -1445,9 +1445,10 @@ static void send_local_reply(struct local_client *client, int status, int fd)
 	replybuf = malloc(message_len);
 
 	clientreply = (struct clvm_header *) replybuf;
-	clientreply->status = -status;
+	clientreply->status = status;
 	clientreply->cmd = CLVMD_CMD_REPLY;
 	clientreply->node[0] = '\0';
+	clientreply->flags = 0;
 
 	ptr = clientreply->args;
 
@@ -1459,6 +1460,9 @@ static void send_local_reply(struct local_client *client, int status, int fd)
 		strcpy(ptr, thisreply->node);
 		ptr += strlen(thisreply->node) + 1;
 
+		if (thisreply->status)
+			clientreply->flags |= CLVMD_FLAG_NODEERRS;
+
 		*(int *) ptr = thisreply->status;
 		ptr += sizeof(int);
 
@@ -1627,7 +1631,7 @@ static int add_to_lvmqueue(struct local_client *client, struct clvm_header *msg,
 
 	cmd = malloc(sizeof(struct lvm_thread_cmd));
 	if (!cmd)
-		return -ENOMEM;
+		return ENOMEM;
 
 	cmd->msg = malloc(msglen);
 	if (!cmd->msg) {
@@ -1709,7 +1713,7 @@ static void check_all_callback(struct local_client *client, char *csid,
 			       int node_up)
 {
 	if (!node_up)
-		add_reply_to_list(client, -EHOSTDOWN, csid, "CLVMD not running",
+		add_reply_to_list(client, EHOSTDOWN, csid, "CLVMD not running",
 				  18);
 }
 

lib/locking/cluster_locking.c

@@ -147,10 +147,16 @@ static int _send_request(char *inbuf, int inlen, char **retbuf)
 	}
 
 	/* Was it an error ? */
-	if (outheader->status < 0) {
+	if (outheader->status != 0) {
-		errno = -outheader->status;
+		errno = outheader->status;
-		log_error("cluster request failed: %s", strerror(errno));
+
-		return 0;
+		/* Only return an error here if there are no node-specific
+		   errors present in the message that might have more detail */
+		if (!(outheader->flags & CLVMD_FLAG_NODEERRS)) {
+			log_error("cluster request failed: %s", strerror(errno));
+			return 0;
+		}
+
 	}
 
 	return 1;
@@ -328,7 +334,7 @@ static int _lock_for_cluster(unsigned char cmd, unsigned int flags, char *name)
 	 * VG locks are just that: locks, and have no side effects
 	 * so we only need to do them on the local node because all
 	 * locks are cluster-wide.
-	 * Also, if the lock is exclusive it makes no sense to try to 
+	 * Also, if the lock is exclusive it makes no sense to try to
 	 * acquire it on all nodes, so just do that on the local node too.
 	 */
 	if (cmd == CLVMD_CMD_LOCK_VG ||
@@ -341,10 +347,11 @@ static int _lock_for_cluster(unsigned char cmd, unsigned int flags, char *name)
 
 	/* If any nodes were down then display them and return an error */
 	for (i = 0; i < num_responses; i++) {
-		if (response[i].status == -EHOSTDOWN) {
+		if (response[i].status == EHOSTDOWN) {
 			log_error("clvmd not running on node %s",
 				  response[i].node);
 			status = 0;
+			errno = response[i].status;
 		} else if (response[i].status) {
 			log_error("Error locking on node %s: %s",
 				  response[i].node,
@@ -352,6 +359,7 @@ static int _lock_for_cluster(unsigned char cmd, unsigned int flags, char *name)
 				  	response[i].response :
 				  	strerror(response[i].status));
 			status = 0;
+			errno = response[i].status;
 		}
 	}