ocfs2/cluster: Cluster up now includes network connections too

The cluster up check only checks to see if the node is heartbeating or not.
If yes it continues assuming that the node is connected to all the nodes. But
if that is not the case, the cluster join aborts with a stack of errors that
are not easy to comprehend.

This patch adds the network connect check upfront and prints the nodes that
the node is not yet connected to, before aborting.

Signed-off-by: Sunil Mushran <sunil.mushran@oracle.com>

diff --git a/fs/ocfs2/dlm/dlmdomain.c b/fs/ocfs2/dlm/dlmdomain.c
index 200d9888ffee66bed9dc84043f34858fcb2a7bc5..92f2ead0fab6de22fa138cc4410dee6e1544216c 100644 (file)
--- a/fs/ocfs2/dlm/dlmdomain.c
+++ b/fs/ocfs2/dlm/dlmdomain.c
@@ -2138,13 +2138,6 @@ struct dlm_ctxt * dlm_register_domain(const char *domain,
                goto leave;
        }
 
-       if (!o2hb_check_local_node_heartbeating()) {
-               mlog(ML_ERROR, "the local node has not been configured, or is "
-                    "not heartbeating\n");
-               ret = -EPROTO;
-               goto leave;
-       }
-
        mlog(0, "register called for domain \"%s\"\n", domain);
 
 retry:

diff --git a/fs/ocfs2/stack_o2cb.c b/fs/ocfs2/stack_o2cb.c
index be935d89a4e8867dac2551a0feb3c593b09668f9..94368017edb378ce1e3961d1976408c954677c9a 100644 (file)
--- a/fs/ocfs2/stack_o2cb.c
+++ b/fs/ocfs2/stack_o2cb.c
@@ -28,6 +28,7 @@
 #include "cluster/masklog.h"
 #include "cluster/nodemanager.h"
 #include "cluster/heartbeat.h"
+#include "cluster/tcp.h"
 
 #include "stackglue.h"
 
@@ -255,6 +256,61 @@ static void o2cb_dump_lksb(struct ocfs2_dlm_lksb *lksb)
        dlm_print_one_lock(lksb->lksb_o2dlm.lockid);
 }
 
+/*
+ * Check if this node is heartbeating and is connected to all other
+ * heartbeating nodes.
+ */
+static int o2cb_cluster_check(void)
+{
+       u8 node_num;
+       int i;
+       unsigned long hbmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
+       unsigned long netmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
+
+       node_num = o2nm_this_node();
+       if (node_num == O2NM_MAX_NODES) {
+               printk(KERN_ERR "o2cb: This node has not been configured.\n");
+               return -EINVAL;
+       }
+
+       /*
+        * o2dlm expects o2net sockets to be created. If not, then
+        * dlm_join_domain() fails with a stack of errors which are both cryptic
+        * and incomplete. The idea here is to detect upfront whether we have
+        * managed to connect to all nodes or not. If not, then list the nodes
+        * to allow the user to check the configuration (incorrect IP, firewall,
+        * etc.) Yes, this is racy. But its not the end of the world.
+        */
+#define        O2CB_MAP_STABILIZE_COUNT        60
+       for (i = 0; i < O2CB_MAP_STABILIZE_COUNT; ++i) {
+               o2hb_fill_node_map(hbmap, sizeof(hbmap));
+               if (!test_bit(node_num, hbmap)) {
+                       printk(KERN_ERR "o2cb: %s heartbeat has not been "
+                              "started.\n", (o2hb_global_heartbeat_active() ?
+                                             "Global" : "Local"));
+                       return -EINVAL;
+               }
+               o2net_fill_node_map(netmap, sizeof(netmap));
+               /* Force set the current node to allow easy compare */
+               set_bit(node_num, netmap);
+               if (!memcmp(hbmap, netmap, sizeof(hbmap)))
+                       return 0;
+               if (i < O2CB_MAP_STABILIZE_COUNT)
+                       msleep(1000);
+       }
+
+       printk(KERN_ERR "o2cb: This node could not connect to nodes:");
+       i = -1;
+       while ((i = find_next_bit(hbmap, O2NM_MAX_NODES,
+                                 i + 1)) < O2NM_MAX_NODES) {
+               if (!test_bit(i, netmap))
+                       printk(" %u", i);
+       }
+       printk(".\n");
+
+       return -ENOTCONN;
+}
+
 /*
  * Called from the dlm when it's about to evict a node. This is how the
  * classic stack signals node death.
@@ -280,12 +336,11 @@ static int o2cb_cluster_connect(struct ocfs2_cluster_connection *conn)
        BUG_ON(conn == NULL);
        BUG_ON(conn->cc_proto == NULL);
 
-       /* for now we only have one cluster/node, make sure we see it
-        * in the heartbeat universe */
-       if (!o2hb_check_local_node_heartbeating()) {
-               if (o2hb_global_heartbeat_active())
-                       mlog(ML_ERROR, "Global heartbeat not started\n");
-               rc = -EINVAL;
+       /* Ensure cluster stack is up and all nodes are connected */
+       rc = o2cb_cluster_check();
+       if (rc) {
+               printk(KERN_ERR "o2cb: Cluster check failed. Fix errors "
+                      "before retrying.\n");
                goto out;
        }