struct dlm_ctxt *dlm =
container_of(work, struct dlm_ctxt, dispatched_work);
LIST_HEAD(tmp_list);
- struct list_head *iter, *iter2;
- struct dlm_work_item *item;
+ struct dlm_work_item *item, *next;
dlm_workfunc_t *workfunc;
int tot=0;
- if (!dlm_joined(dlm))
- return;
-
spin_lock(&dlm->work_lock);
list_splice_init(&dlm->work_list, &tmp_list);
spin_unlock(&dlm->work_lock);
- list_for_each_safe(iter, iter2, &tmp_list) {
+ list_for_each_entry(item, &tmp_list, list) {
tot++;
}
mlog(0, "%s: work thread has %d work items\n", dlm->name, tot);
- list_for_each_safe(iter, iter2, &tmp_list) {
- item = list_entry(iter, struct dlm_work_item, list);
+ list_for_each_entry_safe(item, next, &tmp_list, list) {
workfunc = item->func;
list_del_init(&item->list);
struct dlm_lock_resource *res;
mlog(ML_NOTICE, "%s(%d): recovery info, state=%s, dead=%u, master=%u\n",
- dlm->name, dlm->dlm_reco_thread_task->pid,
+ dlm->name, task_pid_nr(dlm->dlm_reco_thread_task),
dlm->reco.state & DLM_RECO_STATE_ACTIVE ? "ACTIVE" : "inactive",
dlm->reco.dead_node, dlm->reco.new_master);
if (dlm_in_recovery(dlm)) {
mlog(0, "%s: reco thread %d in recovery: "
"state=%d, master=%u, dead=%u\n",
- dlm->name, dlm->dlm_reco_thread_task->pid,
+ dlm->name, task_pid_nr(dlm->dlm_reco_thread_task),
dlm->reco.state, dlm->reco.new_master,
dlm->reco.dead_node);
}
return 0;
}
mlog(0, "%s(%d):recovery thread found node %u in the recovery map!\n",
- dlm->name, dlm->dlm_reco_thread_task->pid,
+ dlm->name, task_pid_nr(dlm->dlm_reco_thread_task),
dlm->reco.dead_node);
spin_unlock(&dlm->spinlock);
mlog(0, "another node will master this recovery session.\n");
}
mlog(0, "dlm=%s (%d), new_master=%u, this node=%u, dead_node=%u\n",
- dlm->name, dlm->dlm_reco_thread_task->pid, dlm->reco.new_master,
+ dlm->name, task_pid_nr(dlm->dlm_reco_thread_task), dlm->reco.new_master,
dlm->node_num, dlm->reco.dead_node);
/* it is safe to start everything back up here
return 0;
master_here:
- mlog(0, "(%d) mastering recovery of %s:%u here(this=%u)!\n",
- dlm->dlm_reco_thread_task->pid,
- dlm->name, dlm->reco.dead_node, dlm->node_num);
+ mlog(ML_NOTICE, "(%d) Node %u is the Recovery Master for the Dead Node "
+ "%u for Domain %s\n", task_pid_nr(dlm->dlm_reco_thread_task),
+ dlm->node_num, dlm->reco.dead_node, dlm->name);
status = dlm_remaster_locks(dlm, dlm->reco.dead_node);
if (status < 0) {
{
int status = 0;
struct dlm_reco_node_data *ndata;
- struct list_head *iter;
int all_nodes_done;
int destroy = 0;
int pass = 0;
/* safe to access the node data list without a lock, since this
* process is the only one to change the list */
- list_for_each(iter, &dlm->reco.node_data) {
- ndata = list_entry (iter, struct dlm_reco_node_data, list);
+ list_for_each_entry(ndata, &dlm->reco.node_data, list) {
BUG_ON(ndata->state != DLM_RECO_NODE_DATA_INIT);
ndata->state = DLM_RECO_NODE_DATA_REQUESTING;
}
} while (status != 0);
+ spin_lock(&dlm_reco_state_lock);
switch (ndata->state) {
case DLM_RECO_NODE_DATA_INIT:
case DLM_RECO_NODE_DATA_FINALIZE_SENT:
ndata->node_num, dead_node);
break;
}
+ spin_unlock(&dlm_reco_state_lock);
}
mlog(0, "done requesting all lock info\n");
* done, or if anyone died */
all_nodes_done = 1;
spin_lock(&dlm_reco_state_lock);
- list_for_each(iter, &dlm->reco.node_data) {
- ndata = list_entry (iter, struct dlm_reco_node_data, list);
-
+ list_for_each_entry(ndata, &dlm->reco.node_data, list) {
mlog(0, "checking recovery state of node %u\n",
ndata->node_num);
switch (ndata->state) {
static void dlm_destroy_recovery_area(struct dlm_ctxt *dlm, u8 dead_node)
{
- struct list_head *iter, *iter2;
- struct dlm_reco_node_data *ndata;
+ struct dlm_reco_node_data *ndata, *next;
LIST_HEAD(tmplist);
spin_lock(&dlm_reco_state_lock);
list_splice_init(&dlm->reco.node_data, &tmplist);
spin_unlock(&dlm_reco_state_lock);
- list_for_each_safe(iter, iter2, &tmplist) {
- ndata = list_entry (iter, struct dlm_reco_node_data, list);
+ list_for_each_entry_safe(ndata, next, &tmplist, list) {
list_del_init(&ndata->list);
kfree(ndata);
}
}
-int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data)
+int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data,
+ void **ret_data)
{
struct dlm_ctxt *dlm = data;
struct dlm_lock_request *lr = (struct dlm_lock_request *)msg->buf;
struct dlm_lock_resource *res;
struct dlm_ctxt *dlm;
LIST_HEAD(resources);
- struct list_head *iter;
int ret;
u8 dead_node, reco_master;
int skip_all_done = 0;
/* any errors returned will be due to the new_master dying,
* the dlm_reco_thread should detect this */
- list_for_each(iter, &resources) {
- res = list_entry (iter, struct dlm_lock_resource, recovering);
+ list_for_each_entry(res, &resources, recovering) {
ret = dlm_send_one_lockres(dlm, res, mres, reco_master,
DLM_MRES_RECOVERY);
if (ret < 0) {
}
-int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data)
+int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data,
+ void **ret_data)
{
struct dlm_ctxt *dlm = data;
struct dlm_reco_data_done *done = (struct dlm_reco_data_done *)msg->buf;
- struct list_head *iter;
struct dlm_reco_node_data *ndata = NULL;
int ret = -EINVAL;
dlm->reco.dead_node, done->node_idx, dlm->node_num);
spin_lock(&dlm_reco_state_lock);
- list_for_each(iter, &dlm->reco.node_data) {
- ndata = list_entry (iter, struct dlm_reco_node_data, list);
+ list_for_each_entry(ndata, &dlm->reco.node_data, list) {
if (ndata->node_num != done->node_idx)
continue;
struct list_head *list,
u8 dead_node)
{
- struct dlm_lock_resource *res;
- struct list_head *iter, *iter2;
+ struct dlm_lock_resource *res, *next;
struct dlm_lock *lock;
spin_lock(&dlm->spinlock);
- list_for_each_safe(iter, iter2, &dlm->reco.resources) {
- res = list_entry (iter, struct dlm_lock_resource, recovering);
+ list_for_each_entry_safe(res, next, &dlm->reco.resources, recovering) {
/* always prune any $RECOVERY entries for dead nodes,
* otherwise hangs can occur during later recovery */
if (dlm_is_recovery_lock(res->lockname.name,
if (total_locks == mres_total_locks)
mres->flags |= DLM_MRES_ALL_DONE;
+ mlog(0, "%s:%.*s: sending mig lockres (%s) to %u\n",
+ dlm->name, res->lockname.len, res->lockname.name,
+ orig_flags & DLM_MRES_MIGRATION ? "migrate" : "recovery",
+ send_to);
+
/* send it */
ret = o2net_send_message(DLM_MIG_LOCKRES_MSG, dlm->key, mres,
sz, send_to, &status);
u8 flags, u8 master)
{
/* mres here is one full page */
- memset(mres, 0, PAGE_SIZE);
+ clear_page(mres);
mres->lockname_len = namelen;
memcpy(mres->lockname, lockname, namelen);
mres->num_locks = 0;
(ml->type == LKM_EXMODE ||
memcmp(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN))) {
mlog(ML_ERROR, "mismatched lvbs!\n");
- __dlm_print_one_lock_resource(lock->lockres);
+ dlm_print_one_lock_resource(lock->lockres);
BUG();
}
memcpy(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN);
return 0;
}
+static void dlm_add_dummy_lock(struct dlm_ctxt *dlm,
+ struct dlm_migratable_lockres *mres)
+{
+ struct dlm_lock dummy;
+ memset(&dummy, 0, sizeof(dummy));
+ dummy.ml.cookie = 0;
+ dummy.ml.type = LKM_IVMODE;
+ dummy.ml.convert_type = LKM_IVMODE;
+ dummy.ml.highest_blocked = LKM_IVMODE;
+ dummy.lksb = NULL;
+ dummy.ml.node = dlm->node_num;
+ dlm_add_lock_to_array(&dummy, mres, DLM_BLOCKED_LIST);
+}
+
+static inline int dlm_is_dummy_lock(struct dlm_ctxt *dlm,
+ struct dlm_migratable_lock *ml,
+ u8 *nodenum)
+{
+ if (unlikely(ml->cookie == 0 &&
+ ml->type == LKM_IVMODE &&
+ ml->convert_type == LKM_IVMODE &&
+ ml->highest_blocked == LKM_IVMODE &&
+ ml->list == DLM_BLOCKED_LIST)) {
+ *nodenum = ml->node;
+ return 1;
+ }
+ return 0;
+}
int dlm_send_one_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
struct dlm_migratable_lockres *mres,
u8 send_to, u8 flags)
{
- struct list_head *queue, *iter;
+ struct list_head *queue;
int total_locks, i;
u64 mig_cookie = 0;
struct dlm_lock *lock;
total_locks = 0;
for (i=DLM_GRANTED_LIST; i<=DLM_BLOCKED_LIST; i++) {
queue = dlm_list_idx_to_ptr(res, i);
- list_for_each(iter, queue) {
- lock = list_entry (iter, struct dlm_lock, list);
-
+ list_for_each_entry(lock, queue, list) {
/* add another lock. */
total_locks++;
if (!dlm_add_lock_to_array(lock, mres, i))
goto error;
}
}
+ if (total_locks == 0) {
+ /* send a dummy lock to indicate a mastery reference only */
+ mlog(0, "%s:%.*s: sending dummy lock to %u, %s\n",
+ dlm->name, res->lockname.len, res->lockname.name,
+ send_to, flags & DLM_MRES_RECOVERY ? "recovery" :
+ "migration");
+ dlm_add_dummy_lock(dlm, mres);
+ }
/* flush any remaining locks */
ret = dlm_send_mig_lockres_msg(dlm, mres, send_to, res, total_locks);
if (ret < 0)
* do we spin? returning an error only delays the problem really
*/
-int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data)
+int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
+ void **ret_data)
{
struct dlm_ctxt *dlm = data;
struct dlm_migratable_lockres *mres =
(struct dlm_migratable_lockres *)msg->buf;
int ret = 0;
u8 real_master;
+ u8 extra_refs = 0;
char *buf = NULL;
struct dlm_work_item *item = NULL;
struct dlm_lock_resource *res = NULL;
__dlm_insert_lockres(dlm, res);
spin_unlock(&dlm->spinlock);
+ /* Add an extra ref for this lock-less lockres lest the
+ * dlm_thread purges it before we get the chance to add
+ * locks to it */
+ dlm_lockres_get(res);
+
+ /* There are three refs that need to be put.
+ * 1. Taken above.
+ * 2. kref_init in dlm_new_lockres()->dlm_init_lockres().
+ * 3. dlm_lookup_lockres()
+ * The first one is handled at the end of this function. The
+ * other two are handled in the worker thread after locks have
+ * been attached. Yes, we don't wait for purge time to match
+ * kref_init. The lockres will still have atleast one ref
+ * added because it is in the hash __dlm_insert_lockres() */
+ extra_refs++;
+
/* now that the new lockres is inserted,
* make it usable by other processes */
spin_lock(&res->spinlock);
res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
spin_unlock(&res->spinlock);
-
- /* add an extra ref for just-allocated lockres
- * otherwise the lockres will be purged immediately */
- dlm_lockres_get(res);
-
+ wake_up(&res->wq);
}
/* at this point we have allocated everything we need,
* and we have a hashed lockres with an extra ref and
* the proper res->state flags. */
ret = 0;
+ spin_lock(&res->spinlock);
+ /* drop this either when master requery finds a different master
+ * or when a lock is added by the recovery worker */
+ dlm_lockres_grab_inflight_ref(dlm, res);
if (mres->master == DLM_LOCK_RES_OWNER_UNKNOWN) {
/* migration cannot have an unknown master */
BUG_ON(!(mres->flags & DLM_MRES_RECOVERY));
"unknown owner.. will need to requery: "
"%.*s\n", mres->lockname_len, mres->lockname);
} else {
- spin_lock(&res->spinlock);
+ /* take a reference now to pin the lockres, drop it
+ * when locks are added in the worker */
dlm_change_lockres_owner(dlm, res, dlm->node_num);
- spin_unlock(&res->spinlock);
}
+ spin_unlock(&res->spinlock);
/* queue up work for dlm_mig_lockres_worker */
dlm_grab(dlm); /* get an extra ref for the work item */
dlm_init_work_item(dlm, item, dlm_mig_lockres_worker, buf);
item->u.ml.lockres = res; /* already have a ref */
item->u.ml.real_master = real_master;
+ item->u.ml.extra_ref = extra_refs;
spin_lock(&dlm->work_lock);
list_add_tail(&item->list, &dlm->work_list);
spin_unlock(&dlm->work_lock);
queue_work(dlm->dlm_worker, &dlm->dispatched_work);
leave:
+ /* One extra ref taken needs to be put here */
+ if (extra_refs)
+ dlm_lockres_put(res);
+
dlm_put(dlm);
if (ret < 0) {
if (buf)
static void dlm_mig_lockres_worker(struct dlm_work_item *item, void *data)
{
- struct dlm_ctxt *dlm = data;
+ struct dlm_ctxt *dlm;
struct dlm_migratable_lockres *mres;
int ret = 0;
struct dlm_lock_resource *res;
u8 real_master;
+ u8 extra_ref;
dlm = item->dlm;
mres = (struct dlm_migratable_lockres *)data;
res = item->u.ml.lockres;
real_master = item->u.ml.real_master;
+ extra_ref = item->u.ml.extra_ref;
if (real_master == DLM_LOCK_RES_OWNER_UNKNOWN) {
/* this case is super-rare. only occurs if
"this node will take it.\n",
res->lockname.len, res->lockname.name);
} else {
+ spin_lock(&res->spinlock);
+ dlm_lockres_drop_inflight_ref(dlm, res);
+ spin_unlock(&res->spinlock);
mlog(0, "master needs to respond to sender "
"that node %u still owns %.*s\n",
real_master, res->lockname.len,
}
leave:
+ /* See comment in dlm_mig_lockres_handler() */
+ if (res) {
+ if (extra_ref)
+ dlm_lockres_put(res);
+ dlm_lockres_put(res);
+ }
kfree(data);
mlog_exit(ret);
}
/* this function cannot error, so unless the sending
* or receiving of the message failed, the owner can
* be trusted */
-int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data)
+int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data,
+ void **ret_data)
{
struct dlm_ctxt *dlm = data;
struct dlm_master_requery *req = (struct dlm_master_requery *)msg->buf;
/* retry!? */
BUG();
}
- }
+ } else /* put.. incase we are not the master */
+ dlm_lockres_put(res);
spin_unlock(&res->spinlock);
}
spin_unlock(&dlm->spinlock);
{
struct dlm_migratable_lock *ml;
struct list_head *queue;
+ struct list_head *tmpq = NULL;
struct dlm_lock *newlock = NULL;
struct dlm_lockstatus *lksb = NULL;
int ret = 0;
- int i, bad;
- struct list_head *iter;
+ int i, j, bad;
struct dlm_lock *lock = NULL;
+ u8 from = O2NM_MAX_NODES;
+ unsigned int added = 0;
mlog(0, "running %d locks for this lockres\n", mres->num_locks);
for (i=0; i<mres->num_locks; i++) {
ml = &(mres->ml[i]);
+
+ if (dlm_is_dummy_lock(dlm, ml, &from)) {
+ /* placeholder, just need to set the refmap bit */
+ BUG_ON(mres->num_locks != 1);
+ mlog(0, "%s:%.*s: dummy lock for %u\n",
+ dlm->name, mres->lockname_len, mres->lockname,
+ from);
+ spin_lock(&res->spinlock);
+ dlm_lockres_set_refmap_bit(from, res);
+ spin_unlock(&res->spinlock);
+ added++;
+ break;
+ }
BUG_ON(ml->highest_blocked != LKM_IVMODE);
newlock = NULL;
lksb = NULL;
queue = dlm_list_num_to_pointer(res, ml->list);
+ tmpq = NULL;
/* if the lock is for the local node it needs to
* be moved to the proper location within the queue.
BUG_ON(!(mres->flags & DLM_MRES_MIGRATION));
spin_lock(&res->spinlock);
- list_for_each(iter, queue) {
- lock = list_entry (iter, struct dlm_lock, list);
- if (lock->ml.cookie != ml->cookie)
- lock = NULL;
- else
+ for (j = DLM_GRANTED_LIST; j <= DLM_BLOCKED_LIST; j++) {
+ tmpq = dlm_list_idx_to_ptr(res, j);
+ list_for_each_entry(lock, tmpq, list) {
+ if (lock->ml.cookie != ml->cookie)
+ lock = NULL;
+ else
+ break;
+ }
+ if (lock)
break;
}
/* lock is always created locally first, and
* destroyed locally last. it must be on the list */
if (!lock) {
- u64 c = ml->cookie;
+ __be64 c = ml->cookie;
mlog(ML_ERROR, "could not find local lock "
"with cookie %u:%llu!\n",
- dlm_get_lock_cookie_node(c),
- dlm_get_lock_cookie_seq(c));
+ dlm_get_lock_cookie_node(be64_to_cpu(c)),
+ dlm_get_lock_cookie_seq(be64_to_cpu(c)));
+ __dlm_print_one_lock_resource(res);
BUG();
}
BUG_ON(lock->ml.node != ml->node);
+ if (tmpq != queue) {
+ mlog(0, "lock was on %u instead of %u for %.*s\n",
+ j, ml->list, res->lockname.len, res->lockname.name);
+ spin_unlock(&res->spinlock);
+ continue;
+ }
+
/* see NOTE above about why we do not update
* to match the master here */
/* do not alter lock refcount. switching lists. */
list_move_tail(&lock->list, queue);
spin_unlock(&res->spinlock);
+ added++;
mlog(0, "just reordered a local lock!\n");
continue;
spin_lock(&res->spinlock);
list_for_each_entry(lock, queue, list) {
if (lock->ml.cookie == ml->cookie) {
- u64 c = lock->ml.cookie;
+ __be64 c = lock->ml.cookie;
mlog(ML_ERROR, "%s:%.*s: %u:%llu: lock already "
"exists on this lockres!\n", dlm->name,
res->lockname.len, res->lockname.name,
- dlm_get_lock_cookie_node(c),
- dlm_get_lock_cookie_seq(c));
+ dlm_get_lock_cookie_node(be64_to_cpu(c)),
+ dlm_get_lock_cookie_seq(be64_to_cpu(c)));
mlog(ML_NOTICE, "sent lock: type=%d, conv=%d, "
"node=%u, cookie=%u:%llu, queue=%d\n",
ml->type, ml->convert_type, ml->node,
- dlm_get_lock_cookie_node(ml->cookie),
- dlm_get_lock_cookie_seq(ml->cookie),
+ dlm_get_lock_cookie_node(be64_to_cpu(ml->cookie)),
+ dlm_get_lock_cookie_seq(be64_to_cpu(ml->cookie)),
ml->list);
__dlm_print_one_lock_resource(res);
if (!bad) {
dlm_lock_get(newlock);
list_add_tail(&newlock->list, queue);
+ mlog(0, "%s:%.*s: added lock for node %u, "
+ "setting refmap bit\n", dlm->name,
+ res->lockname.len, res->lockname.name, ml->node);
+ dlm_lockres_set_refmap_bit(ml->node, res);
+ added++;
}
spin_unlock(&res->spinlock);
}
mlog(0, "done running all the locks\n");
leave:
+ /* balance the ref taken when the work was queued */
+ spin_lock(&res->spinlock);
+ dlm_lockres_drop_inflight_ref(dlm, res);
+ spin_unlock(&res->spinlock);
+
if (ret < 0) {
mlog_errno(ret);
if (newlock)
struct dlm_lock_resource *res)
{
int i;
- struct list_head *queue, *iter, *iter2;
- struct dlm_lock *lock;
+ struct list_head *queue;
+ struct dlm_lock *lock, *next;
res->state |= DLM_LOCK_RES_RECOVERING;
if (!list_empty(&res->recovering)) {
"Recovering res %s:%.*s, is already on recovery list!\n",
dlm->name, res->lockname.len, res->lockname.name);
list_del_init(&res->recovering);
+ dlm_lockres_put(res);
}
/* We need to hold a reference while on the recovery list */
dlm_lockres_get(res);
/* find any pending locks and put them back on proper list */
for (i=DLM_BLOCKED_LIST; i>=DLM_GRANTED_LIST; i--) {
queue = dlm_list_idx_to_ptr(res, i);
- list_for_each_safe(iter, iter2, queue) {
- lock = list_entry (iter, struct dlm_lock, list);
+ list_for_each_entry_safe(lock, next, queue, list) {
dlm_lock_get(lock);
if (lock->convert_pending) {
/* move converting lock back to granted */
u8 dead_node, u8 new_master)
{
int i;
- struct list_head *iter, *iter2;
struct hlist_node *hash_iter;
struct hlist_head *bucket;
-
- struct dlm_lock_resource *res;
+ struct dlm_lock_resource *res, *next;
mlog_entry_void();
assert_spin_locked(&dlm->spinlock);
- list_for_each_safe(iter, iter2, &dlm->reco.resources) {
- res = list_entry (iter, struct dlm_lock_resource, recovering);
+ list_for_each_entry_safe(res, next, &dlm->reco.resources, recovering) {
if (res->owner == dead_node) {
list_del_init(&res->recovering);
spin_lock(&res->spinlock);
+ /* new_master has our reference from
+ * the lock state sent during recovery */
dlm_change_lockres_owner(dlm, res, new_master);
res->state &= ~DLM_LOCK_RES_RECOVERING;
- if (!__dlm_lockres_unused(res))
+ if (__dlm_lockres_has_locks(res))
__dlm_dirty_lockres(dlm, res);
spin_unlock(&res->spinlock);
wake_up(&res->wq);
dlm_lockres_put(res);
}
spin_lock(&res->spinlock);
+ /* new_master has our reference from
+ * the lock state sent during recovery */
dlm_change_lockres_owner(dlm, res, new_master);
res->state &= ~DLM_LOCK_RES_RECOVERING;
- if (!__dlm_lockres_unused(res))
+ if (__dlm_lockres_has_locks(res))
__dlm_dirty_lockres(dlm, res);
spin_unlock(&res->spinlock);
wake_up(&res->wq);
static void dlm_revalidate_lvb(struct dlm_ctxt *dlm,
struct dlm_lock_resource *res, u8 dead_node)
{
- struct list_head *iter, *queue;
+ struct list_head *queue;
struct dlm_lock *lock;
int blank_lvb = 0, local = 0;
int i;
for (i=DLM_GRANTED_LIST; i<=DLM_CONVERTING_LIST; i++) {
queue = dlm_list_idx_to_ptr(res, i);
- list_for_each(iter, queue) {
- lock = list_entry (iter, struct dlm_lock, list);
+ list_for_each_entry(lock, queue, list) {
if (lock->ml.node == search_node) {
if (dlm_lvb_needs_invalidation(lock, local)) {
/* zero the lksb lvb and lockres lvb */
static void dlm_free_dead_locks(struct dlm_ctxt *dlm,
struct dlm_lock_resource *res, u8 dead_node)
{
- struct list_head *iter, *tmpiter;
- struct dlm_lock *lock;
+ struct dlm_lock *lock, *next;
+ unsigned int freed = 0;
/* this node is the lockres master:
* 1) remove any stale locks for the dead node
assert_spin_locked(&dlm->spinlock);
assert_spin_locked(&res->spinlock);
+ /* We do two dlm_lock_put(). One for removing from list and the other is
+ * to force the DLM_UNLOCK_FREE_LOCK action so as to free the locks */
+
/* TODO: check pending_asts, pending_basts here */
- list_for_each_safe(iter, tmpiter, &res->granted) {
- lock = list_entry (iter, struct dlm_lock, list);
+ list_for_each_entry_safe(lock, next, &res->granted, list) {
if (lock->ml.node == dead_node) {
list_del_init(&lock->list);
dlm_lock_put(lock);
+ /* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */
+ dlm_lock_put(lock);
+ freed++;
}
}
- list_for_each_safe(iter, tmpiter, &res->converting) {
- lock = list_entry (iter, struct dlm_lock, list);
+ list_for_each_entry_safe(lock, next, &res->converting, list) {
if (lock->ml.node == dead_node) {
list_del_init(&lock->list);
dlm_lock_put(lock);
+ /* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */
+ dlm_lock_put(lock);
+ freed++;
}
}
- list_for_each_safe(iter, tmpiter, &res->blocked) {
- lock = list_entry (iter, struct dlm_lock, list);
+ list_for_each_entry_safe(lock, next, &res->blocked, list) {
if (lock->ml.node == dead_node) {
list_del_init(&lock->list);
dlm_lock_put(lock);
+ /* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */
+ dlm_lock_put(lock);
+ freed++;
}
}
+ if (freed) {
+ mlog(0, "%s:%.*s: freed %u locks for dead node %u, "
+ "dropping ref from lockres\n", dlm->name,
+ res->lockname.len, res->lockname.name, freed, dead_node);
+ BUG_ON(!test_bit(dead_node, res->refmap));
+ dlm_lockres_clear_refmap_bit(dead_node, res);
+ } else if (test_bit(dead_node, res->refmap)) {
+ mlog(0, "%s:%.*s: dead node %u had a ref, but had "
+ "no locks and had not purged before dying\n", dlm->name,
+ res->lockname.len, res->lockname.name, dead_node);
+ dlm_lockres_clear_refmap_bit(dead_node, res);
+ }
+
/* do not kick thread yet */
__dlm_dirty_lockres(dlm, res);
}
spin_lock(&res->spinlock);
/* zero the lvb if necessary */
dlm_revalidate_lvb(dlm, res, dead_node);
- if (res->owner == dead_node)
+ if (res->owner == dead_node) {
+ if (res->state & DLM_LOCK_RES_DROPPING_REF)
+ mlog(0, "%s:%.*s: owned by "
+ "dead node %u, this node was "
+ "dropping its ref when it died. "
+ "continue, dropping the flag.\n",
+ dlm->name, res->lockname.len,
+ res->lockname.name, dead_node);
+
+ /* the wake_up for this will happen when the
+ * RECOVERING flag is dropped later */
+ res->state &= ~DLM_LOCK_RES_DROPPING_REF;
+
dlm_move_lockres_to_recovery_list(dlm, res);
- else if (res->owner == dlm->node_num) {
+ } else if (res->owner == dlm->node_num) {
dlm_free_dead_locks(dlm, res, dead_node);
__dlm_lockres_calc_usage(dlm, res);
}
}
}
+ /* Clean up join state on node death. */
+ if (dlm->joining_node == idx) {
+ mlog(0, "Clearing join state for node %u\n", idx);
+ __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
+ }
+
/* check to see if the node is already considered dead */
if (!test_bit(idx, dlm->live_nodes_map)) {
mlog(0, "for domain %s, node %d is already dead. "
clear_bit(idx, dlm->live_nodes_map);
- /* Clean up join state on node death. */
- if (dlm->joining_node == idx) {
- mlog(0, "Clearing join state for node %u\n", idx);
- __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
- }
-
/* make sure local cleanup occurs before the heartbeat events */
if (!test_bit(idx, dlm->recovery_map))
dlm_do_local_recovery_cleanup(dlm, idx);
if (!dlm_grab(dlm))
return;
+ /*
+ * This will notify any dlm users that a node in our domain
+ * went away without notifying us first.
+ */
+ if (test_bit(idx, dlm->domain_map))
+ dlm_fire_domain_eviction_callbacks(dlm, idx);
+
spin_lock(&dlm->spinlock);
__dlm_hb_node_down(dlm, idx);
spin_unlock(&dlm->spinlock);
return ret;
}
-int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data)
+int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data,
+ void **ret_data)
{
struct dlm_ctxt *dlm = data;
struct dlm_begin_reco *br = (struct dlm_begin_reco *)msg->buf;
return ret;
}
-int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data)
+int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data,
+ void **ret_data)
{
struct dlm_ctxt *dlm = data;
struct dlm_finalize_reco *fr = (struct dlm_finalize_reco *)msg->buf;