Srinivas Eeda
2010-Sep-20 21:59 UTC
[Ocfs2-devel] [PATCH 1/1] o2dlm: free block mles during shutdown
If a node initiates shutdown after another node initiated the lock mastery
process, this node might have created block mle but will not release it if it
doesn't get the assert master from the other node. This causes block
mle's to
lie around unfreed.
This patch frees any block mles that exists on master list after the node sent
DLM_EXIT_DOMAIN_MSG to other nodes.
Signed-off-by: Srinivas Eeda <srinivas.eeda at oracle.com>
---
fs/ocfs2/dlm/dlmcommon.h | 1 +
fs/ocfs2/dlm/dlmdomain.c | 1 +
fs/ocfs2/dlm/dlmmaster.c | 33 +++++++++++++++++++++++++++++++++
3 files changed, 35 insertions(+), 0 deletions(-)
diff --git a/fs/ocfs2/dlm/dlmcommon.h b/fs/ocfs2/dlm/dlmcommon.h
index 4b6ae2c..48282dd 100644
--- a/fs/ocfs2/dlm/dlmcommon.h
+++ b/fs/ocfs2/dlm/dlmcommon.h
@@ -1030,6 +1030,7 @@ int dlm_drop_lockres_ref(struct dlm_ctxt *dlm,
struct dlm_lock_resource *res);
void dlm_clean_master_list(struct dlm_ctxt *dlm,
u8 dead_node);
+void dlm_free_block_mles(struct dlm_ctxt *dlm);
int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock);
int __dlm_lockres_has_locks(struct dlm_lock_resource *res);
int __dlm_lockres_unused(struct dlm_lock_resource *res);
diff --git a/fs/ocfs2/dlm/dlmdomain.c b/fs/ocfs2/dlm/dlmdomain.c
index 153abb5..8744fff 100644
--- a/fs/ocfs2/dlm/dlmdomain.c
+++ b/fs/ocfs2/dlm/dlmdomain.c
@@ -693,6 +693,7 @@ void dlm_unregister_domain(struct dlm_ctxt *dlm)
dlm_mark_domain_leaving(dlm);
dlm_leave_domain(dlm);
+ dlm_free_block_mles(dlm);
dlm_complete_dlm_shutdown(dlm);
}
dlm_put(dlm);
diff --git a/fs/ocfs2/dlm/dlmmaster.c b/fs/ocfs2/dlm/dlmmaster.c
index ffb4c68..5f4d6fd 100644
--- a/fs/ocfs2/dlm/dlmmaster.c
+++ b/fs/ocfs2/dlm/dlmmaster.c
@@ -3433,3 +3433,36 @@ void dlm_lockres_release_ast(struct dlm_ctxt *dlm,
wake_up(&res->wq);
wake_up(&dlm->migration_wq);
}
+
+void dlm_free_block_mles(struct dlm_ctxt *dlm)
+{
+ int i;
+ struct hlist_head *bucket;
+ struct dlm_master_list_entry *mle;
+ struct hlist_node *list;
+
+ spin_lock(&dlm->spinlock);
+ spin_lock(&dlm->master_lock);
+ 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);
+ if (mle->type != DLM_MLE_BLOCK) {
+ mlog(ML_ERROR, "mle for %.*s not destroyed, "
+ "type %d\n",
+ mle->mnamelen, mle->mname, mle->type);
+ continue;
+ }
+ spin_lock(&mle->spinlock);
+ atomic_set(&mle->woken, 1);
+ spin_unlock(&mle->spinlock);
+ wake_up(&mle->wq);
+
+ __dlm_mle_detach_hb_events(dlm, mle);
+ __dlm_put_mle(mle);
+ }
+ }
+ spin_unlock(&dlm->master_lock);
+ spin_unlock(&dlm->spinlock);
+}
--
1.5.6.5
Wengang Wang
2010-Sep-21 04:15 UTC
[Ocfs2-devel] [PATCH 1/1] o2dlm: free block mles during shutdown
Hi Srini, On 10-09-20 14:59, Srinivas Eeda wrote:> If a node initiates shutdown after another node initiated the lock mastery > process, this node might have created block mle but will not release it if it > doesn't get the assert master from the other node. This causes block mle's to > lie around unfreed. > > This patch frees any block mles that exists on master list after the node sent > DLM_EXIT_DOMAIN_MSG to other nodes. > > Signed-off-by: Srinivas Eeda <srinivas.eeda at oracle.com> > --- > fs/ocfs2/dlm/dlmcommon.h | 1 + > fs/ocfs2/dlm/dlmdomain.c | 1 + > fs/ocfs2/dlm/dlmmaster.c | 33 +++++++++++++++++++++++++++++++++ > 3 files changed, 35 insertions(+), 0 deletions(-) > > diff --git a/fs/ocfs2/dlm/dlmcommon.h b/fs/ocfs2/dlm/dlmcommon.h > index 4b6ae2c..48282dd 100644 > --- a/fs/ocfs2/dlm/dlmcommon.h > +++ b/fs/ocfs2/dlm/dlmcommon.h > @@ -1030,6 +1030,7 @@ int dlm_drop_lockres_ref(struct dlm_ctxt *dlm, > struct dlm_lock_resource *res); > void dlm_clean_master_list(struct dlm_ctxt *dlm, > u8 dead_node); > +void dlm_free_block_mles(struct dlm_ctxt *dlm); > int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock); > int __dlm_lockres_has_locks(struct dlm_lock_resource *res); > int __dlm_lockres_unused(struct dlm_lock_resource *res); > diff --git a/fs/ocfs2/dlm/dlmdomain.c b/fs/ocfs2/dlm/dlmdomain.c > index 153abb5..8744fff 100644 > --- a/fs/ocfs2/dlm/dlmdomain.c > +++ b/fs/ocfs2/dlm/dlmdomain.c > @@ -693,6 +693,7 @@ void dlm_unregister_domain(struct dlm_ctxt *dlm) > > dlm_mark_domain_leaving(dlm); > dlm_leave_domain(dlm); > + dlm_free_block_mles(dlm); > dlm_complete_dlm_shutdown(dlm); > } > dlm_put(dlm); > diff --git a/fs/ocfs2/dlm/dlmmaster.c b/fs/ocfs2/dlm/dlmmaster.c > index ffb4c68..5f4d6fd 100644 > --- a/fs/ocfs2/dlm/dlmmaster.c > +++ b/fs/ocfs2/dlm/dlmmaster.c > @@ -3433,3 +3433,36 @@ void dlm_lockres_release_ast(struct dlm_ctxt *dlm, > wake_up(&res->wq); > wake_up(&dlm->migration_wq); > } > + > +void dlm_free_block_mles(struct dlm_ctxt *dlm) > +{ > + int i; > + struct hlist_head *bucket; > + struct dlm_master_list_entry *mle; > + struct hlist_node *list; > + > + spin_lock(&dlm->spinlock); > + spin_lock(&dlm->master_lock); > + for (i = 0; i < DLM_HASH_BUCKETS; i++) { > + bucket = dlm_master_hash(dlm, i); > + hlist_for_each(list, bucket) {I guess it can be the last reference on mles? If so, don't we need "hlist_for_each_safe" here since we are removing entries? regards, wengang.> + mle = hlist_entry(list, struct dlm_master_list_entry, > + master_hash_node); > + if (mle->type != DLM_MLE_BLOCK) { > + mlog(ML_ERROR, "mle for %.*s not destroyed, " > + "type %d\n", > + mle->mnamelen, mle->mname, mle->type); > + continue; > + } > + spin_lock(&mle->spinlock); > + atomic_set(&mle->woken, 1); > + spin_unlock(&mle->spinlock); > + wake_up(&mle->wq); > + > + __dlm_mle_detach_hb_events(dlm, mle); > + __dlm_put_mle(mle); > + } > + } > + spin_unlock(&dlm->master_lock); > + spin_unlock(&dlm->spinlock); > +} > -- > 1.5.6.5