From 8decab3c8dadcdf4f54ffb30df6e6f67b398b6e0 Mon Sep 17 00:00:00 2001 From: Sunil Mushran Date: Sun, 24 Jul 2011 10:23:54 -0700 Subject: ocfs2/dlm: Clean up messages in o2dlm o2dlm messages needed a facelift. Signed-off-by: Sunil Mushran --- fs/ocfs2/dlm/dlmdomain.c | 18 ++++++++++-------- 1 file changed, 10 insertions(+), 8 deletions(-) (limited to 'fs/ocfs2/dlm/dlmdomain.c') diff --git a/fs/ocfs2/dlm/dlmdomain.c b/fs/ocfs2/dlm/dlmdomain.c index 6ed6b95dcf9..ce225fb0cea 100644 --- a/fs/ocfs2/dlm/dlmdomain.c +++ b/fs/ocfs2/dlm/dlmdomain.c @@ -539,17 +539,17 @@ again: static void __dlm_print_nodes(struct dlm_ctxt *dlm) { - int node = -1; + int node = -1, num = 0; assert_spin_locked(&dlm->spinlock); - printk(KERN_NOTICE "o2dlm: Nodes in domain %s: ", dlm->name); - + printk("( "); while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES, node + 1)) < O2NM_MAX_NODES) { printk("%d ", node); + ++num; } - printk("\n"); + printk(") %u nodes\n", num); } static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data, @@ -566,11 +566,10 @@ static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data, node = exit_msg->node_idx; - printk(KERN_NOTICE "o2dlm: Node %u leaves domain %s\n", node, dlm->name); - spin_lock(&dlm->spinlock); clear_bit(node, dlm->domain_map); clear_bit(node, dlm->exit_domain_map); + printk(KERN_NOTICE "o2dlm: Node %u leaves domain %s ", node, dlm->name); __dlm_print_nodes(dlm); /* notify anything attached to the heartbeat events */ @@ -755,6 +754,7 @@ void dlm_unregister_domain(struct dlm_ctxt *dlm) dlm_mark_domain_leaving(dlm); dlm_leave_domain(dlm); + printk(KERN_NOTICE "o2dlm: Leaving domain %s\n", dlm->name); dlm_force_free_mles(dlm); dlm_complete_dlm_shutdown(dlm); } @@ -970,7 +970,7 @@ static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data, clear_bit(assert->node_idx, dlm->exit_domain_map); __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN); - printk(KERN_NOTICE "o2dlm: Node %u joins domain %s\n", + printk(KERN_NOTICE "o2dlm: Node %u joins domain %s ", assert->node_idx, dlm->name); __dlm_print_nodes(dlm); @@ -1701,8 +1701,10 @@ static int dlm_try_to_join_domain(struct dlm_ctxt *dlm) bail: spin_lock(&dlm->spinlock); __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN); - if (!status) + if (!status) { + printk(KERN_NOTICE "o2dlm: Joining domain %s ", dlm->name); __dlm_print_nodes(dlm); + } spin_unlock(&dlm->spinlock); if (ctxt) { -- cgit v1.2.3 From e9f0b6a6233105c188064328c3f15e21942d0241 Mon Sep 17 00:00:00 2001 From: Sunil Mushran Date: Sun, 24 Jul 2011 10:27:54 -0700 Subject: ocfs2/dlm: Trace insert/remove of resource to/from hash Add mlog to trace adding and removing the resource from/to the hash table. Signed-off-by: Sunil Mushran --- fs/ocfs2/dlm/dlmcommon.h | 5 ++--- fs/ocfs2/dlm/dlmdomain.c | 19 ++++++++++++------- fs/ocfs2/dlm/dlmthread.c | 2 +- 3 files changed, 15 insertions(+), 11 deletions(-) (limited to 'fs/ocfs2/dlm/dlmdomain.c') diff --git a/fs/ocfs2/dlm/dlmcommon.h b/fs/ocfs2/dlm/dlmcommon.h index 4a5d5804b13..b1748ac7475 100644 --- a/fs/ocfs2/dlm/dlmcommon.h +++ b/fs/ocfs2/dlm/dlmcommon.h @@ -877,9 +877,8 @@ static inline void dlm_lockres_get(struct dlm_lock_resource *res) kref_get(&res->refs); } void dlm_lockres_put(struct dlm_lock_resource *res); -void __dlm_unhash_lockres(struct dlm_lock_resource *res); -void __dlm_insert_lockres(struct dlm_ctxt *dlm, - struct dlm_lock_resource *res); +void __dlm_unhash_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); +void __dlm_insert_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res); struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm, const char *name, unsigned int len, diff --git a/fs/ocfs2/dlm/dlmdomain.c b/fs/ocfs2/dlm/dlmdomain.c index ce225fb0cea..200d9888ffe 100644 --- a/fs/ocfs2/dlm/dlmdomain.c +++ b/fs/ocfs2/dlm/dlmdomain.c @@ -157,16 +157,18 @@ static int dlm_protocol_compare(struct dlm_protocol_version *existing, static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm); -void __dlm_unhash_lockres(struct dlm_lock_resource *lockres) +void __dlm_unhash_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res) { - if (!hlist_unhashed(&lockres->hash_node)) { - hlist_del_init(&lockres->hash_node); - dlm_lockres_put(lockres); - } + if (hlist_unhashed(&res->hash_node)) + return; + + mlog(0, "%s: Unhash res %.*s\n", dlm->name, res->lockname.len, + res->lockname.name); + hlist_del_init(&res->hash_node); + dlm_lockres_put(res); } -void __dlm_insert_lockres(struct dlm_ctxt *dlm, - struct dlm_lock_resource *res) +void __dlm_insert_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res) { struct hlist_head *bucket; struct qstr *q; @@ -180,6 +182,9 @@ void __dlm_insert_lockres(struct dlm_ctxt *dlm, dlm_lockres_get(res); hlist_add_head(&res->hash_node, bucket); + + mlog(0, "%s: Hash res %.*s\n", dlm->name, res->lockname.len, + res->lockname.name); } struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm, diff --git a/fs/ocfs2/dlm/dlmthread.c b/fs/ocfs2/dlm/dlmthread.c index 46faac20f16..4eff65e7e59 100644 --- a/fs/ocfs2/dlm/dlmthread.c +++ b/fs/ocfs2/dlm/dlmthread.c @@ -207,7 +207,7 @@ static void dlm_purge_lockres(struct dlm_ctxt *dlm, BUG(); } - __dlm_unhash_lockres(res); + __dlm_unhash_lockres(dlm, res); /* lockres is not in the hash now. drop the flag and wake up * any processes waiting in dlm_get_lock_resource. */ -- cgit v1.2.3 From 6b27f62fc750d85bc6fc3718b3b38ec60edc2d74 Mon Sep 17 00:00:00 2001 From: Sunil Mushran Date: Sun, 24 Jul 2011 10:33:54 -0700 Subject: 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 --- fs/ocfs2/dlm/dlmdomain.c | 7 ----- fs/ocfs2/stack_o2cb.c | 67 +++++++++++++++++++++++++++++++++++++++++++----- 2 files changed, 61 insertions(+), 13 deletions(-) (limited to 'fs/ocfs2/dlm/dlmdomain.c') diff --git a/fs/ocfs2/dlm/dlmdomain.c b/fs/ocfs2/dlm/dlmdomain.c index 200d9888ffe..92f2ead0fab 100644 --- 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 be935d89a4e..94368017edb 100644 --- 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; } -- cgit v1.2.3