void *nodemap, u32 flags);
static void dlm_deref_lockres_worker(struct dlm_work_item *item, void *data);
+static inline void __dlm_mle_name(struct dlm_master_list_entry *mle,
+ unsigned char **name, unsigned int *namelen,
+ unsigned int *namehash)
+{
+ BUG_ON(mle->type != DLM_MLE_BLOCK &&
+ mle->type != DLM_MLE_MASTER &&
+ mle->type != DLM_MLE_MIGRATION);
+
+ if (mle->type != DLM_MLE_MASTER) {
+ *name = mle->u.mlename.name;
+ *namelen = mle->u.mlename.len;
+ if (namehash)
+ *namehash = mle->u.mlename.hash;
+ } else {
+ *name = (unsigned char *)mle->u.mleres->lockname.name;
+ *namelen = mle->u.mleres->lockname.len;
+ if (namehash)
+ *namehash = mle->u.mleres->lockname.hash;
+ }
+}
+
static inline int dlm_mle_equal(struct dlm_ctxt *dlm,
struct dlm_master_list_entry *mle,
const char *name,
unsigned int namelen)
{
- struct dlm_lock_resource *res;
+ unsigned char *mlename;
+ unsigned int mlelen;
if (dlm != mle->dlm)
return 0;
- if (mle->type == DLM_MLE_BLOCK ||
- mle->type == DLM_MLE_MIGRATION) {
- if (namelen != mle->u.name.len ||
- memcmp(name, mle->u.name.name, namelen)!=0)
- return 0;
- } else {
- res = mle->u.res;
- if (namelen != res->lockname.len ||
- memcmp(res->lockname.name, name, namelen) != 0)
- return 0;
- }
+ __dlm_mle_name(mle, &mlename, &mlelen, NULL);
+
+ if (namelen != mlelen || memcmp(name, mlename, namelen) != 0)
+ return 0;
+
return 1;
}
mle->dlm = dlm;
mle->type = type;
- INIT_LIST_HEAD(&mle->list);
+ INIT_HLIST_NODE(&mle->master_hash_node);
INIT_LIST_HEAD(&mle->hb_events);
memset(mle->maybe_map, 0, sizeof(mle->maybe_map));
spin_lock_init(&mle->spinlock);
mle->new_master = O2NM_MAX_NODES;
mle->inuse = 0;
+ BUG_ON(mle->type != DLM_MLE_BLOCK &&
+ mle->type != DLM_MLE_MASTER &&
+ mle->type != DLM_MLE_MIGRATION);
+
if (mle->type == DLM_MLE_MASTER) {
BUG_ON(!res);
- mle->u.res = res;
- } else if (mle->type == DLM_MLE_BLOCK) {
- BUG_ON(!name);
- memcpy(mle->u.name.name, name, namelen);
- mle->u.name.len = namelen;
- } else /* DLM_MLE_MIGRATION */ {
+ mle->u.mleres = res;
+ } else {
BUG_ON(!name);
- memcpy(mle->u.name.name, name, namelen);
- mle->u.name.len = namelen;
+ memcpy(mle->u.mlename.name, name, namelen);
+ mle->u.mlename.len = namelen;
+ mle->u.mlename.hash = dlm_lockid_hash(name, namelen);
}
/* copy off the node_map and register hb callbacks on our copy */
__dlm_mle_attach_hb_events(dlm, mle);
}
+void __dlm_unlink_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle)
+{
+ assert_spin_locked(&dlm->spinlock);
+ assert_spin_locked(&dlm->master_lock);
+
+ if (!hlist_unhashed(&mle->master_hash_node))
+ hlist_del_init(&mle->master_hash_node);
+}
+
+void __dlm_insert_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle)
+{
+ struct hlist_head *bucket;
+ unsigned char *mname;
+ unsigned int mlen, hash;
+
+ assert_spin_locked(&dlm->master_lock);
+
+ __dlm_mle_name(mle, &mname, &mlen, &hash);
+ bucket = dlm_master_hash(dlm, hash);
+ hlist_add_head(&mle->master_hash_node, bucket);
+}
/* returns 1 if found, 0 if not */
static int dlm_find_mle(struct dlm_ctxt *dlm,
char *name, unsigned int namelen)
{
struct dlm_master_list_entry *tmpmle;
+ struct hlist_head *bucket;
+ struct hlist_node *list;
+ unsigned int hash;
assert_spin_locked(&dlm->master_lock);
- list_for_each_entry(tmpmle, &dlm->master_list, list) {
+ hash = dlm_lockid_hash(name, namelen);
+ bucket = dlm_master_hash(dlm, hash);
+ hlist_for_each(list, bucket) {
+ tmpmle = hlist_entry(list, struct dlm_master_list_entry,
+ master_hash_node);
if (!dlm_mle_equal(dlm, tmpmle, name, namelen))
continue;
dlm_get_mle(tmpmle);
{
struct dlm_master_list_entry *mle;
struct dlm_ctxt *dlm;
+ unsigned char *mname;
+ unsigned int mlen;
mlog_entry_void();
mle = container_of(kref, struct dlm_master_list_entry, mle_refs);
dlm = mle->dlm;
- if (mle->type != DLM_MLE_MASTER) {
- mlog(0, "calling mle_release for %.*s, type %d\n",
- mle->u.name.len, mle->u.name.name, mle->type);
- } else {
- mlog(0, "calling mle_release for %.*s, type %d\n",
- mle->u.res->lockname.len,
- mle->u.res->lockname.name, mle->type);
- }
assert_spin_locked(&dlm->spinlock);
assert_spin_locked(&dlm->master_lock);
+ __dlm_mle_name(mle, &mname, &mlen, NULL);
+ mlog(0, "Releasing mle for %.*s, type %d\n", mlen, mname, mle->type);
+
/* remove from list if not already */
- if (!list_empty(&mle->list))
- list_del_init(&mle->list);
+ __dlm_unlink_mle(dlm, mle);
/* detach the mle from the domain node up/down events */
__dlm_mle_detach_hb_events(dlm, mle);
alloc_mle = NULL;
dlm_init_mle(mle, DLM_MLE_MASTER, dlm, res, NULL, 0);
set_bit(dlm->node_num, mle->maybe_map);
- list_add(&mle->list, &dlm->master_list);
+ __dlm_insert_mle(dlm, mle);
/* still holding the dlm spinlock, check the recovery map
* to see if there are any nodes that still need to be
res->lockname.len,
res->lockname.name);
mle->type = DLM_MLE_MASTER;
- mle->u.res = res;
+ mle->u.mleres = res;
}
}
}
struct dlm_ctxt *dlm = mle->dlm;
struct dlm_master_request request;
int ret, response=0, resend;
+ unsigned char *mlename;
+ unsigned int mlenamelen;
memset(&request, 0, sizeof(request));
request.node_idx = dlm->node_num;
BUG_ON(mle->type == DLM_MLE_MIGRATION);
- if (mle->type != DLM_MLE_MASTER) {
- request.namelen = mle->u.name.len;
- memcpy(request.name, mle->u.name.name, request.namelen);
- } else {
- request.namelen = mle->u.res->lockname.len;
- memcpy(request.name, mle->u.res->lockname.name,
- request.namelen);
- }
+ __dlm_mle_name(mle, &mlename, &mlenamelen, NULL);
+
+ request.namelen = (u8)mlenamelen;
+ memcpy(request.name, mlename, request.namelen);
again:
ret = o2net_send_message(DLM_MASTER_REQUEST_MSG, dlm->key, &request,
// "add the block.\n");
dlm_init_mle(mle, DLM_MLE_BLOCK, dlm, NULL, name, namelen);
set_bit(request->node_idx, mle->maybe_map);
- list_add(&mle->list, &dlm->master_list);
+ __dlm_insert_mle(dlm, mle);
response = DLM_MASTER_RESP_NO;
} else {
// mlog(0, "mle was found\n");
assert->node_idx, rr, extra_ref, mle->inuse);
dlm_print_one_mle(mle);
}
- list_del_init(&mle->list);
+ __dlm_unlink_mle(dlm, mle);
__dlm_mle_detach_hb_events(dlm, mle);
__dlm_put_mle(mle);
if (extra_ref) {
tmp->master = master;
atomic_set(&tmp->woken, 1);
wake_up(&tmp->wq);
- /* remove it from the list so that only one
- * mle will be found */
- list_del_init(&tmp->list);
- /* this was obviously WRONG. mle is uninited here. should be tmp. */
+ /* remove it so that only one mle will be found */
+ __dlm_unlink_mle(dlm, tmp);
__dlm_mle_detach_hb_events(dlm, tmp);
ret = DLM_MIGRATE_RESPONSE_MASTERY_REF;
mlog(0, "%s:%.*s: master=%u, newmaster=%u, "
mle->master = master;
/* do this for consistency with other mle types */
set_bit(new_master, mle->maybe_map);
- list_add(&mle->list, &dlm->master_list);
+ __dlm_insert_mle(dlm, mle);
return ret;
}
+/*
+ * Sets the owner of the lockres, associated to the mle, to UNKNOWN
+ */
+static struct dlm_lock_resource *dlm_reset_mleres_owner(struct dlm_ctxt *dlm,
+ struct dlm_master_list_entry *mle)
+{
+ struct dlm_lock_resource *res;
+ unsigned int hash;
+
+ /* Find the lockres associated to the mle and set its owner to UNK */
+ hash = dlm_lockid_hash(mle->u.mlename.name, mle->u.mlename.len);
+ res = __dlm_lookup_lockres(dlm, mle->u.mlename.name, mle->u.mlename.len,
+ hash);
+ if (res) {
+ spin_unlock(&dlm->master_lock);
+
+ /* move lockres onto recovery list */
+ spin_lock(&res->spinlock);
+ dlm_set_lockres_owner(dlm, res, DLM_LOCK_RES_OWNER_UNKNOWN);
+ dlm_move_lockres_to_recovery_list(dlm, res);
+ spin_unlock(&res->spinlock);
+ dlm_lockres_put(res);
+
+ /* about to get rid of mle, detach from heartbeat */
+ __dlm_mle_detach_hb_events(dlm, mle);
+
+ /* dump the mle */
+ spin_lock(&dlm->master_lock);
+ __dlm_put_mle(mle);
+ spin_unlock(&dlm->master_lock);
+ }
+
+ return res;
+}
+
+static void dlm_clean_migration_mle(struct dlm_ctxt *dlm,
+ struct dlm_master_list_entry *mle)
+{
+ __dlm_mle_detach_hb_events(dlm, mle);
+
+ spin_lock(&mle->spinlock);
+ __dlm_unlink_mle(dlm, mle);
+ atomic_set(&mle->woken, 1);
+ spin_unlock(&mle->spinlock);
+
+ wake_up(&mle->wq);
+}
+
+static void dlm_clean_block_mle(struct dlm_ctxt *dlm,
+ struct dlm_master_list_entry *mle, u8 dead_node)
+{
+ int bit;
+
+ BUG_ON(mle->type != DLM_MLE_BLOCK);
+
+ spin_lock(&mle->spinlock);
+ bit = find_next_bit(mle->maybe_map, O2NM_MAX_NODES, 0);
+ if (bit != dead_node) {
+ mlog(0, "mle found, but dead node %u would not have been "
+ "master\n", dead_node);
+ spin_unlock(&mle->spinlock);
+ } else {
+ /* Must drop the refcount by one since the assert_master will
+ * never arrive. This may result in the mle being unlinked and
+ * freed, but there may still be a process waiting in the
+ * dlmlock path which is fine. */
+ mlog(0, "node %u was expected master\n", dead_node);
+ atomic_set(&mle->woken, 1);
+ spin_unlock(&mle->spinlock);
+ wake_up(&mle->wq);
+
+ /* Do not need events any longer, so detach from heartbeat */
+ __dlm_mle_detach_hb_events(dlm, mle);
+ __dlm_put_mle(mle);
+ }
+}
void dlm_clean_master_list(struct dlm_ctxt *dlm, u8 dead_node)
{
- struct dlm_master_list_entry *mle, *next;
+ struct dlm_master_list_entry *mle;
struct dlm_lock_resource *res;
- unsigned int hash;
+ struct hlist_head *bucket;
+ struct hlist_node *list;
+ unsigned int i;
mlog_entry("dlm=%s, dead node=%u\n", dlm->name, dead_node);
top:
/* clean the master list */
spin_lock(&dlm->master_lock);
- list_for_each_entry_safe(mle, next, &dlm->master_list, list) {
+ for (i = 0; i < DLM_HASH_BUCKETS; i++) {
+ bucket = dlm_master_hash(dlm, i);
+ hlist_for_each(list, bucket) {
+ mle = hlist_entry(list, struct dlm_master_list_entry,
+ master_hash_node);
+
BUG_ON(mle->type != DLM_MLE_BLOCK &&
mle->type != DLM_MLE_MASTER &&
mle->type != DLM_MLE_MIGRATION);
* need to clean up if the dead node would have
* been the master. */
if (mle->type == DLM_MLE_BLOCK) {
- int bit;
-
- spin_lock(&mle->spinlock);
- bit = find_next_bit(mle->maybe_map, O2NM_MAX_NODES, 0);
- if (bit != dead_node) {
- mlog(0, "mle found, but dead node %u would "
- "not have been master\n", dead_node);
- spin_unlock(&mle->spinlock);
- } else {
- /* must drop the refcount by one since the
- * assert_master will never arrive. this
- * may result in the mle being unlinked and
- * freed, but there may still be a process
- * waiting in the dlmlock path which is fine. */
- mlog(0, "node %u was expected master\n",
- dead_node);
- atomic_set(&mle->woken, 1);
- spin_unlock(&mle->spinlock);
- wake_up(&mle->wq);
- /* do not need events any longer, so detach
- * from heartbeat */
- __dlm_mle_detach_hb_events(dlm, mle);
- __dlm_put_mle(mle);
- }
+ dlm_clean_block_mle(dlm, mle, dead_node);
continue;
}
/* if we have reached this point, this mle needs to
* be removed from the list and freed. */
-
- /* remove from the list early. NOTE: unlinking
- * list_head while in list_for_each_safe */
- __dlm_mle_detach_hb_events(dlm, mle);
- spin_lock(&mle->spinlock);
- list_del_init(&mle->list);
- atomic_set(&mle->woken, 1);
- spin_unlock(&mle->spinlock);
- wake_up(&mle->wq);
+ dlm_clean_migration_mle(dlm, mle);
mlog(0, "%s: node %u died during migration from "
"%u to %u!\n", dlm->name, dead_node,
mle->master, mle->new_master);
- /* if there is a lockres associated with this
- * mle, find it and set its owner to UNKNOWN */
- hash = dlm_lockid_hash(mle->u.name.name, mle->u.name.len);
- res = __dlm_lookup_lockres(dlm, mle->u.name.name,
- mle->u.name.len, hash);
- if (res) {
- /* unfortunately if we hit this rare case, our
- * lock ordering is messed. we need to drop
- * the master lock so that we can take the
- * lockres lock, meaning that we will have to
- * restart from the head of list. */
- spin_unlock(&dlm->master_lock);
-
- /* move lockres onto recovery list */
- spin_lock(&res->spinlock);
- dlm_set_lockres_owner(dlm, res,
- DLM_LOCK_RES_OWNER_UNKNOWN);
- dlm_move_lockres_to_recovery_list(dlm, res);
- spin_unlock(&res->spinlock);
- dlm_lockres_put(res);
-
- /* about to get rid of mle, detach from heartbeat */
- __dlm_mle_detach_hb_events(dlm, mle);
-
- /* dump the mle */
- spin_lock(&dlm->master_lock);
- __dlm_put_mle(mle);
- spin_unlock(&dlm->master_lock);
+ /* If we find a lockres associated with the mle, we've
+ * hit this rare case that messes up our lock ordering.
+ * If so, we need to drop the master lock so that we can
+ * take the lockres lock, meaning that we will have to
+ * restart from the head of list. */
+ res = dlm_reset_mleres_owner(dlm, mle);
+ if (res)
/* restart */
goto top;
- }
/* this may be the last reference */
__dlm_put_mle(mle);
}
+ }
spin_unlock(&dlm->master_lock);
}