*/
int ctdb_set_transport(struct ctdb_context *ctdb, const char *transport)
{
- int ctdb_tcp_init(struct ctdb_context *ctdb);
-#ifdef USE_INFINIBAND
- int ctdb_ibw_init(struct ctdb_context *ctdb);
-#endif /* USE_INFINIBAND */
-
- if (strcmp(transport, "tcp") == 0) {
- return ctdb_tcp_init(ctdb);
- }
-#ifdef USE_INFINIBAND
- if (strcmp(transport, "ib") == 0) {
- return ctdb_ibw_init(ctdb);
- }
-#endif /* USE_INFINIBAND */
-
- ctdb_set_error(ctdb, "Unknown transport '%s'\n", transport);
- return -1;
+ ctdb->transport = talloc_strdup(ctdb, transport);
+ return 0;
}
+
/*
set some ctdb flags
*/
will change! */
node->vnn = ctdb->num_nodes;
- if (ctdb->methods->add_node(node) != 0) {
- talloc_free(node);
- return -1;
- }
-
if (ctdb_same_address(&ctdb->address, &node->address)) {
ctdb->vnn = node->vnn;
+ node->flags |= NODE_FLAGS_CONNECTED;
}
ctdb->num_nodes++;
return 0;
}
+
+/*
+ setup the local socket name
+*/
+int ctdb_set_socketname(struct ctdb_context *ctdb, const char *socketname)
+{
+ ctdb->daemon.name = talloc_strdup(ctdb, socketname);
+ return 0;
+}
+
/*
add a node to the list of active nodes
*/
struct ctdb_req_header *hdr = (struct ctdb_req_header *)data;
TALLOC_CTX *tmp_ctx;
+ ctdb->status.node_packets_recv++;
+
/* place the packet as a child of the tmp_ctx. We then use
talloc_free() below to free it. If any of the calls want
to keep it, then they will steal it somewhere else, and the
switch (hdr->operation) {
case CTDB_REQ_CALL:
+ /* verify that the remote node that sent us the call
+ is running in the same generation instance as this node
+ */
+ if (ctdb->vnn_map->generation != hdr->generation) {
+ DEBUG(0,(__location__ " ctdb request %d of type"
+ " %d length %d from node %d to %d had an"
+ " invalid generation id:%d while our"
+ " generation id is:%d\n",
+ hdr->reqid, hdr->operation, hdr->length,
+ hdr->srcnode, hdr->destnode,
+ ctdb->vnn_map->generation,
+ hdr->generation));
+ break;
+ }
+ /* if we are in recovery mode we discard all traffic
+ until the cluster has recovered.
+ */
+ if (ctdb->recovery_mode != CTDB_RECOVERY_NORMAL) {
+ DEBUG(0,(__location__ " ctdb request %d of type"
+ " %d length %d from node %d to %d"
+ " while we are in recovery mode\n",
+ hdr->reqid, hdr->operation, hdr->length,
+ hdr->srcnode, hdr->destnode));
+ break;
+ }
+
+ ctdb->status.count.req_call++;
ctdb_request_call(ctdb, hdr);
break;
case CTDB_REPLY_CALL:
+ ctdb->status.count.reply_call++;
ctdb_reply_call(ctdb, hdr);
break;
case CTDB_REPLY_ERROR:
+ ctdb->status.count.reply_error++;
ctdb_reply_error(ctdb, hdr);
break;
- case CTDB_REPLY_REDIRECT:
- ctdb_reply_redirect(ctdb, hdr);
- break;
-
case CTDB_REQ_DMASTER:
+ ctdb->status.count.req_dmaster++;
ctdb_request_dmaster(ctdb, hdr);
break;
case CTDB_REPLY_DMASTER:
+ ctdb->status.count.reply_dmaster++;
ctdb_reply_dmaster(ctdb, hdr);
break;
case CTDB_REQ_MESSAGE:
+ ctdb->status.count.req_message++;
ctdb_request_message(ctdb, hdr);
break;
case CTDB_REQ_FINISHED:
+ ctdb->status.count.req_finished++;
ctdb_request_finished(ctdb, hdr);
break;
+ case CTDB_REQ_CONTROL:
+ ctdb->status.count.req_control++;
+ ctdb_request_control(ctdb, hdr);
+ break;
+
+ case CTDB_REPLY_CONTROL:
+ ctdb->status.count.reply_control++;
+ ctdb_reply_control(ctdb, hdr);
+ break;
+
default:
DEBUG(0,("%s: Packet with unknown operation %d\n",
__location__, hdr->operation));
talloc_free(tmp_ctx);
}
+/*
+ called by the transport layer when a packet comes in
+*/
+void ctdb_recv_raw_pkt(void *p, uint8_t *data, uint32_t length)
+{
+ struct ctdb_context *ctdb = talloc_get_type(p, struct ctdb_context);
+ ctdb_recv_pkt(ctdb, data, length);
+}
+
/*
called by the transport layer when a node is dead
*/
static void ctdb_node_dead(struct ctdb_node *node)
{
node->ctdb->num_connected--;
+ node->flags &= ~NODE_FLAGS_CONNECTED;
DEBUG(1,("%s: node %s is dead: %d connected\n",
node->ctdb->name, node->name, node->ctdb->num_connected));
}
static void ctdb_node_connected(struct ctdb_node *node)
{
node->ctdb->num_connected++;
+ node->flags |= NODE_FLAGS_CONNECTED;
DEBUG(1,("%s: connected to %s - %d connected\n",
node->ctdb->name, node->name, node->ctdb->num_connected));
}
DEBUG(3,("ctdb_connect_wait: got all %d nodes\n", expected));
}
+struct queue_next {
+ struct ctdb_context *ctdb;
+ struct ctdb_req_header *hdr;
+};
+
+
+/*
+ trigered when a deferred packet is due
+ */
+static void queue_next_trigger(struct event_context *ev, struct timed_event *te,
+ struct timeval t, void *private_data)
+{
+ struct queue_next *q = talloc_get_type(private_data, struct queue_next);
+ ctdb_recv_pkt(q->ctdb, (uint8_t *)q->hdr, q->hdr->length);
+ talloc_free(q);
+}
+
+/*
+ defer a packet, so it is processed on the next event loop
+ this is used for sending packets to ourselves
+ */
+static void ctdb_defer_packet(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
+{
+ struct queue_next *q;
+ q = talloc(ctdb, struct queue_next);
+ if (q == NULL) {
+ DEBUG(0,(__location__ " Failed to allocate deferred packet\n"));
+ return;
+ }
+ q->ctdb = ctdb;
+ q->hdr = talloc_memdup(ctdb, hdr, hdr->length);
+ if (q->hdr == NULL) {
+ DEBUG(0,("Error copying deferred packet to self\n"));
+ return;
+ }
+#if 0
+ /* use this to put packets directly into our recv function */
+ ctdb_recv_pkt(q->ctdb, (uint8_t *)q->hdr, q->hdr->length);
+ talloc_free(q);
+#else
+ event_add_timed(ctdb->ev, q, timeval_zero(), queue_next_trigger, q);
+#endif
+}
+
/*
queue a packet or die
*/
void ctdb_queue_packet(struct ctdb_context *ctdb, struct ctdb_req_header *hdr)
{
struct ctdb_node *node;
+ ctdb->status.node_packets_sent++;
+
+ if (!ctdb_validate_vnn(ctdb, hdr->destnode)) {
+ DEBUG(0,(__location__ " cant send to node %u that does not exist\n",
+ hdr->destnode));
+ return;
+ }
+
node = ctdb->nodes[hdr->destnode];
- if (ctdb->methods->queue_pkt(node, (uint8_t *)hdr, hdr->length) != 0) {
+
+ if (hdr->destnode == ctdb->vnn && !(ctdb->flags & CTDB_FLAG_SELF_CONNECT)) {
+ ctdb_defer_packet(ctdb, hdr);
+ } else if (ctdb->methods->queue_pkt(node, (uint8_t *)hdr, hdr->length) != 0) {
ctdb_fatal(ctdb, "Unable to queue packet\n");
}
}
ctdb = talloc_zero(ev, struct ctdb_context);
ctdb->ev = ev;
+ ctdb->recovery_mode = CTDB_RECOVERY_NORMAL;
ctdb->upcalls = &ctdb_upcalls;
ctdb->idr = idr_init(ctdb);
ctdb->max_lacount = CTDB_DEFAULT_MAX_LACOUNT;