#define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_DOMAIN)
#include "cluster/masklog.h"
+/*
+ * ocfs2 node maps are array of long int, which limits to send them freely
+ * across the wire due to endianness issues. To workaround this, we convert
+ * long ints to byte arrays. Following 3 routines are helper functions to
+ * set/test/copy bits within those array of bytes
+ */
+static inline void byte_set_bit(u8 nr, u8 map[])
+{
+ map[nr >> 3] |= (1UL << (nr & 7));
+}
+
+static inline int byte_test_bit(u8 nr, u8 map[])
+{
+ return ((1UL << (nr & 7)) & (map[nr >> 3])) != 0;
+}
+
+static inline void byte_copymap(u8 dmap[], unsigned long smap[],
+ unsigned int sz)
+{
+ unsigned int nn;
+
+ if (!sz)
+ return;
+
+ memset(dmap, 0, ((sz + 7) >> 3));
+ for (nn = 0 ; nn < sz; nn++)
+ if (test_bit(nn, smap))
+ byte_set_bit(nn, dmap);
+}
+
static void dlm_free_pagevec(void **vec, int pages)
{
while (pages--)
void __dlm_unhash_lockres(struct dlm_lock_resource *lockres)
{
- hlist_del_init(&lockres->hash_node);
- dlm_lockres_put(lockres);
+ if (!hlist_unhashed(&lockres->hash_node)) {
+ hlist_del_init(&lockres->hash_node);
+ dlm_lockres_put(lockres);
+ }
}
void __dlm_insert_lockres(struct dlm_ctxt *dlm,
dlm_lockres_put(res);
- cond_resched_lock(&dlm->spinlock);
-
if (dropped)
goto redo_bucket;
}
+ cond_resched_lock(&dlm->spinlock);
num += n;
mlog(0, "%s: touched %d lockreses in bucket %d "
"(tot=%d)\n", dlm->name, n, i, num);
dlm_kick_thread(dlm, NULL);
while (dlm_migrate_all_locks(dlm)) {
+ /* Give dlm_thread time to purge the lockres' */
+ msleep(500);
mlog(0, "%s: more migration to do\n", dlm->name);
}
dlm_mark_domain_leaving(dlm);
struct dlm_query_join_request *query;
enum dlm_query_join_response response;
struct dlm_ctxt *dlm = NULL;
+ u8 nodenum;
query = (struct dlm_query_join_request *) msg->buf;
spin_lock(&dlm_domain_lock);
dlm = __dlm_lookup_domain_full(query->domain, query->name_len);
+ if (!dlm)
+ goto unlock_respond;
+
+ /*
+ * There is a small window where the joining node may not see the
+ * node(s) that just left but still part of the cluster. DISALLOW
+ * join request if joining node has different node map.
+ */
+ nodenum=0;
+ while (nodenum < O2NM_MAX_NODES) {
+ if (test_bit(nodenum, dlm->domain_map)) {
+ if (!byte_test_bit(nodenum, query->node_map)) {
+ mlog(0, "disallow join as node %u does not "
+ "have node %u in its nodemap\n",
+ query->node_idx, nodenum);
+ response = JOIN_DISALLOW;
+ goto unlock_respond;
+ }
+ }
+ nodenum++;
+ }
+
/* Once the dlm ctxt is marked as leaving then we don't want
* to be put in someone's domain map.
* Also, explicitly disallow joining at certain troublesome
/* Disallow parallel joins. */
response = JOIN_DISALLOW;
} else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
- mlog(ML_NOTICE, "node %u trying to join, but recovery "
+ mlog(0, "node %u trying to join, but recovery "
"is ongoing.\n", bit);
response = JOIN_DISALLOW;
} else if (test_bit(bit, dlm->recovery_map)) {
- mlog(ML_NOTICE, "node %u trying to join, but it "
+ mlog(0, "node %u trying to join, but it "
"still needs recovery.\n", bit);
response = JOIN_DISALLOW;
} else if (test_bit(bit, dlm->domain_map)) {
- mlog(ML_NOTICE, "node %u trying to join, but it "
+ mlog(0, "node %u trying to join, but it "
"is still in the domain! needs recovery?\n",
bit);
response = JOIN_DISALLOW;
spin_unlock(&dlm->spinlock);
}
+unlock_respond:
spin_unlock(&dlm_domain_lock);
respond:
join_msg.name_len = strlen(dlm->name);
memcpy(join_msg.domain, dlm->name, join_msg.name_len);
+ /* copy live node map to join message */
+ byte_copymap(join_msg.node_map, dlm->live_nodes_map, O2NM_MAX_NODES);
+
status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg,
sizeof(join_msg), node, &retval);
if (status < 0 && status != -ENOPROTOOPT) {
{
int status = 0, tmpstat, node;
struct domain_join_ctxt *ctxt;
- enum dlm_query_join_response response;
+ enum dlm_query_join_response response = JOIN_DISALLOW;
mlog_entry("%p", dlm);
status = o2net_register_handler(DLM_ASSERT_MASTER_MSG, dlm->key,
sizeof(struct dlm_assert_master),
dlm_assert_master_handler,
- dlm, NULL, &dlm->dlm_domain_handlers);
+ dlm, dlm_assert_master_post_handler,
+ &dlm->dlm_domain_handlers);
if (status)
goto bail;
static int dlm_join_domain(struct dlm_ctxt *dlm)
{
int status;
+ unsigned int backoff;
+ unsigned int total_backoff = 0;
BUG_ON(!dlm);
}
do {
- unsigned int backoff;
status = dlm_try_to_join_domain(dlm);
/* If we're racing another node to the join, then we
* need to back off temporarily and let them
* complete. */
+#define DLM_JOIN_TIMEOUT_MSECS 90000
if (status == -EAGAIN) {
if (signal_pending(current)) {
status = -ERESTARTSYS;
goto bail;
}
+ if (total_backoff >
+ msecs_to_jiffies(DLM_JOIN_TIMEOUT_MSECS)) {
+ status = -ERESTARTSYS;
+ mlog(ML_NOTICE, "Timed out joining dlm domain "
+ "%s after %u msecs\n", dlm->name,
+ jiffies_to_msecs(total_backoff));
+ goto bail;
+ }
+
/*
* <chip> After you!
* <dale> No, after you!
*/
backoff = (unsigned int)(jiffies & 0x3);
backoff *= DLM_DOMAIN_BACKOFF_MS;
+ total_backoff += backoff;
mlog(0, "backoff %d\n", backoff);
msleep(backoff);
}