[PATCH 21/32] ocfs2/dlm: Activate dlm->master_hash for master list entries

[Mark Fasheh]

From: Sunil Mushran <sunil.mushran@xxxxxxxxxx>

With this patch, the mles are stored in a hash and not a simple list.
This should improve the mle lookup time when the number of outstanding
masteries is large.

Signed-off-by: Sunil Mushran <sunil.mushran@xxxxxxxxxx>
Signed-off-by: Mark Fasheh <mfasheh@xxxxxxxx>
---
 fs/ocfs2/dlm/dlmcommon.h |    4 +-
 fs/ocfs2/dlm/dlmdebug.c  |   24 +++++++++++-------
 fs/ocfs2/dlm/dlmdomain.c |    1 -
 fs/ocfs2/dlm/dlmmaster.c |   61 ++++++++++++++++++++++++++++++++-------------
 4 files changed, 60 insertions(+), 30 deletions(-)

diff --git a/fs/ocfs2/dlm/dlmcommon.h b/fs/ocfs2/dlm/dlmcommon.h
index 425653f..aa55271 100644
--- a/fs/ocfs2/dlm/dlmcommon.h
+++ b/fs/ocfs2/dlm/dlmcommon.h
@@ -56,12 +56,13 @@ enum dlm_mle_type {
 };
 
 struct dlm_lock_name {
+	unsigned int hash;
 	unsigned int len;
 	unsigned char name[DLM_LOCKID_NAME_MAX];
 };
 
 struct dlm_master_list_entry {
-	struct list_head list;
+	struct hlist_node master_hash_node;
 	struct list_head hb_events;
 	struct dlm_ctxt *dlm;
 	spinlock_t spinlock;
@@ -152,7 +153,6 @@ struct dlm_ctxt
 	struct dlm_recovery_ctxt reco;
 	spinlock_t master_lock;
 	struct hlist_head **master_hash;
-	struct list_head master_list;
 	struct list_head mle_hb_events;
 
 	/* these give a really vague idea of the system load */
diff --git a/fs/ocfs2/dlm/dlmdebug.c b/fs/ocfs2/dlm/dlmdebug.c
index c82feb7..336a98e 100644
--- a/fs/ocfs2/dlm/dlmdebug.c
+++ b/fs/ocfs2/dlm/dlmdebug.c
@@ -501,18 +501,25 @@ static struct file_operations debug_purgelist_fops = {
 static int debug_mle_print(struct dlm_ctxt *dlm, struct debug_buffer *db)
 {
 	struct dlm_master_list_entry *mle;
-	int out = 0;
+	struct hlist_head *bucket;
+	struct hlist_node *list;
+	int i, out = 0;
 	unsigned long total = 0;
 
 	out += snprintf(db->buf + out, db->len - out,
 			"Dumping MLEs for Domain: %s\n", dlm->name);
 
 	spin_lock(&dlm->master_lock);
-	list_for_each_entry(mle, &dlm->master_list, list) {
-		++total;
-		if (db->len - out < 200)
-			continue;
-		out += dump_mle(mle, db->buf + out, db->len - out);
+	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);
+			++total;
+			if (db->len - out < 200)
+				continue;
+			out += dump_mle(mle, db->buf + out, db->len - out);
+		}
 	}
 	spin_unlock(&dlm->master_lock);
 
@@ -813,12 +820,11 @@ static int debug_state_print(struct dlm_ctxt *dlm, struct debug_buffer *db)
 	/* Lists: Dirty=Empty  Purge=InUse  PendingASTs=Empty  ... */
 	out += snprintf(db->buf + out, db->len - out,
 			"Lists: Dirty=%s  Purge=%s  PendingASTs=%s  "
-			"PendingBASTs=%s  Master=%s\n",
+			"PendingBASTs=%s\n",
 			(list_empty(&dlm->dirty_list) ? "Empty" : "InUse"),
 			(list_empty(&dlm->purge_list) ? "Empty" : "InUse"),
 			(list_empty(&dlm->pending_asts) ? "Empty" : "InUse"),
-			(list_empty(&dlm->pending_basts) ? "Empty" : "InUse"),
-			(list_empty(&dlm->master_list) ? "Empty" : "InUse"));
+			(list_empty(&dlm->pending_basts) ? "Empty" : "InUse"));
 
 	/* Purge Count: xxx  Refs: xxx */
 	out += snprintf(db->buf + out, db->len - out,
diff --git a/fs/ocfs2/dlm/dlmdomain.c b/fs/ocfs2/dlm/dlmdomain.c
index 4531504..869648c 100644
--- a/fs/ocfs2/dlm/dlmdomain.c
+++ b/fs/ocfs2/dlm/dlmdomain.c
@@ -1597,7 +1597,6 @@ static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain,
 	init_waitqueue_head(&dlm->reco.event);
 	init_waitqueue_head(&dlm->ast_wq);
 	init_waitqueue_head(&dlm->migration_wq);
-	INIT_LIST_HEAD(&dlm->master_list);
 	INIT_LIST_HEAD(&dlm->mle_hb_events);
 
 	dlm->joining_node = DLM_LOCK_RES_OWNER_UNKNOWN;
diff --git a/fs/ocfs2/dlm/dlmmaster.c b/fs/ocfs2/dlm/dlmmaster.c
index ec6da3c..8045581 100644
--- a/fs/ocfs2/dlm/dlmmaster.c
+++ b/fs/ocfs2/dlm/dlmmaster.c
@@ -69,7 +69,8 @@ static int dlm_do_assert_master(struct dlm_ctxt *dlm,
 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 char **name, unsigned int *namelen,
+				  unsigned int *namehash)
 {
 	BUG_ON(mle->type != DLM_MLE_BLOCK &&
 	       mle->type != DLM_MLE_MASTER &&
@@ -78,9 +79,13 @@ static inline void __dlm_mle_name(struct dlm_master_list_entry *mle,
 	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;
 	}
 }
 
@@ -95,7 +100,7 @@ static inline int dlm_mle_equal(struct dlm_ctxt *dlm,
 	if (dlm != mle->dlm)
 		return 0;
 
-	__dlm_mle_name(mle, &mlename, &mlelen);
+	__dlm_mle_name(mle, &mlename, &mlelen, NULL);
 
 	if (namelen != mlelen || memcmp(name, mlename, namelen) != 0)
 		return 0;
@@ -294,7 +299,7 @@ static void dlm_init_mle(struct dlm_master_list_entry *mle,
 
 	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);
@@ -317,6 +322,7 @@ static void dlm_init_mle(struct dlm_master_list_entry *mle,
 		BUG_ON(!name);
 		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 */
@@ -334,15 +340,21 @@ 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 (!list_empty(&mle->list))
-		list_del_init(&mle->list);
+	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);
 
-	list_add(&mle->list, &dlm->master_list);
+	__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 */
@@ -351,10 +363,17 @@ 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);
@@ -428,23 +447,20 @@ static void dlm_mle_release(struct kref *kref)
 {
 	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.mlename.len, mle->u.mlename.name, mle->type);
-	} else {
-		mlog(0, "calling mle_release for %.*s, type %d\n",
-		     mle->u.mleres->lockname.len,
-		     mle->u.mleres->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 */
 	__dlm_unlink_mle(dlm, mle);
 
@@ -1342,7 +1358,7 @@ static int dlm_do_master_request(struct dlm_lock_resource *res,
 
 	BUG_ON(mle->type == DLM_MLE_MIGRATION);
 
-	__dlm_mle_name(mle, &mlename, &mlenamelen);
+	__dlm_mle_name(mle, &mlename, &mlenamelen, NULL);
 
 	request.namelen = (u8)mlenamelen;
 	memcpy(request.name, mlename, request.namelen);
@@ -3286,8 +3302,11 @@ static void dlm_clean_block_mle(struct dlm_ctxt *dlm,
 
 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;
+	struct hlist_head *bucket;
+	struct hlist_node *list;
+	unsigned int i;
 
 	mlog_entry("dlm=%s, dead node=%u\n", dlm->name, dead_node);
 top:
@@ -3295,7 +3314,12 @@ 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);
@@ -3351,6 +3375,7 @@ top:
 		/* this may be the last reference */
 		__dlm_put_mle(mle);
 	}
+	}
 	spin_unlock(&dlm->master_lock);
 }
 
-- 
1.5.6

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/