NFS: use a hash table for delegation lookup

nfs_delegation_find_inode currently has to walk the entire list of
delegations per inode, which can become pretty large, and can become even
larger when increasing the delegation watermark.

Add a hash table to speed up the delegation lookup, sized as a fraction
of the delegation watermark.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20250718081509.2607553-6-hch@lst.de
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>

5 files changed, 58 insertions, 2 deletions

diff --git a/fs/nfs/delegation.c b/fs/nfs/delegation.c
index ea96f77e38c2..9d3a5f29f17f 100644
--- a/fs/nfs/delegation.c
+++ b/fs/nfs/delegation.c
@@ -30,6 +30,13 @@
 static unsigned nfs_delegation_watermark = NFS_DEFAULT_DELEGATION_WATERMARK;
 module_param_named(delegation_watermark, nfs_delegation_watermark, uint, 0644);
 
+static struct hlist_head *nfs_delegation_hash(struct nfs_server *server,
+		const struct nfs_fh *fhandle)
+{
+	return server->delegation_hash_table +
+		(nfs_fhandle_hash(fhandle) & server->delegation_hash_mask);
+}
+
 static void __nfs_free_delegation(struct nfs_delegation *delegation)
 {
 	put_cred(delegation->cred);
@@ -367,6 +374,7 @@ nfs_detach_delegation_locked(struct nfs_inode *nfsi,
 		spin_unlock(&delegation->lock);
 		return NULL;
 	}
+	hlist_del_init_rcu(&delegation->hash);
 	list_del_rcu(&delegation->super_list);
 	delegation->inode = NULL;
 	rcu_assign_pointer(nfsi->delegation, NULL);
@@ -529,6 +537,8 @@ add_new:
 	spin_unlock(&inode->i_lock);
 
 	list_add_tail_rcu(&delegation->super_list, &server->delegations);
+	hlist_add_head_rcu(&delegation->hash,
+			nfs_delegation_hash(server, &NFS_I(inode)->fh));
 	rcu_assign_pointer(nfsi->delegation, delegation);
 	delegation = NULL;
 
@@ -1166,11 +1176,12 @@ static struct inode *
 nfs_delegation_find_inode_server(struct nfs_server *server,
 				 const struct nfs_fh *fhandle)
 {
+	struct hlist_head *head = nfs_delegation_hash(server, fhandle);
 	struct nfs_delegation *delegation;
 	struct super_block *freeme = NULL;
 	struct inode *res = NULL;
 
-	list_for_each_entry_rcu(delegation, &server->delegations, super_list) {
+	hlist_for_each_entry_rcu(delegation, head, hash) {
 		spin_lock(&delegation->lock);
 		if (delegation->inode != NULL &&
 		    !test_bit(NFS_DELEGATION_REVOKED, &delegation->flags) &&
@@ -1577,3 +1588,18 @@ out:
 	rcu_read_unlock();
 	return ret;
 }
+
+int nfs4_delegation_hash_alloc(struct nfs_server *server)
+{
+	int delegation_buckets, i;
+
+	delegation_buckets = roundup_pow_of_two(nfs_delegation_watermark / 16);
+	server->delegation_hash_mask = delegation_buckets - 1;
+	server->delegation_hash_table = kmalloc_array(delegation_buckets,
+			sizeof(*server->delegation_hash_table), GFP_KERNEL);
+	if (!server->delegation_hash_table)
+		return -ENOMEM;
+	for (i = 0; i < delegation_buckets; i++)
+		INIT_HLIST_HEAD(&server->delegation_hash_table[i]);
+	return 0;
+}
diff --git a/fs/nfs/delegation.h b/fs/nfs/delegation.h
index 8ff5ab9c5c25..08ec2e9c68a4 100644
--- a/fs/nfs/delegation.h
+++ b/fs/nfs/delegation.h
@@ -14,6 +14,7 @@
  * NFSv4 delegation
  */
 struct nfs_delegation {
+	struct hlist_node hash;
 	struct list_head super_list;
 	const struct cred *cred;
 	struct inode *inode;
@@ -123,4 +124,6 @@ static inline int nfs_have_delegated_mtime(struct inode *inode)
 						 NFS_DELEGATION_FLAG_TIME);
 }
 
+int nfs4_delegation_hash_alloc(struct nfs_server *server);
+
 #endif
diff --git a/fs/nfs/nfs4client.c b/fs/nfs/nfs4client.c
index 5943a192f36b..2ea98f1f116f 100644
--- a/fs/nfs/nfs4client.c
+++ b/fs/nfs/nfs4client.c
@@ -802,6 +802,7 @@ static void nfs4_destroy_server(struct nfs_server *server)
 	unset_pnfs_layoutdriver(server);
 	nfs4_purge_state_owners(server, &freeme);
 	nfs4_free_state_owners(&freeme);
+	kfree(server->delegation_hash_table);
 }
 
 /*
@@ -1096,6 +1097,10 @@ static int nfs4_server_common_setup(struct nfs_server *server,
 {
 	int error;
 
+	error = nfs4_delegation_hash_alloc(server);
+	if (error)
+		return error;
+
 	/* data servers support only a subset of NFSv4.1 */
 	if (is_ds_only_client(server->nfs_client))
 		return -EPROTONOSUPPORT;
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index ef2077e185b6..d8bebd757af3 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -10967,6 +10967,26 @@ static const struct inode_operations nfs4_file_inode_operations = {
 	.listxattr	= nfs4_listxattr,
 };
 
+static struct nfs_server *nfs4_clone_server(struct nfs_server *source,
+		struct nfs_fh *fh, struct nfs_fattr *fattr,
+		rpc_authflavor_t flavor)
+{
+	struct nfs_server *server;
+	int error;
+
+	server = nfs_clone_server(source, fh, fattr, flavor);
+	if (IS_ERR(server))
+		return server;
+
+	error = nfs4_delegation_hash_alloc(server);
+	if (error) {
+		nfs_free_server(server);
+		return ERR_PTR(error);
+	}
+
+	return server;
+}
+
 const struct nfs_rpc_ops nfs_v4_clientops = {
 	.version	= 4,			/* protocol version */
 	.dentry_ops	= &nfs4_dentry_operations,
@@ -11019,7 +11039,7 @@ const struct nfs_rpc_ops nfs_v4_clientops = {
 	.init_client	= nfs4_init_client,
 	.free_client	= nfs4_free_client,
 	.create_server	= nfs4_create_server,
-	.clone_server	= nfs_clone_server,
+	.clone_server	= nfs4_clone_server,
 	.discover_trunking = nfs4_discover_trunking,
 	.enable_swap	= nfs4_enable_swap,
 	.disable_swap	= nfs4_disable_swap,
diff --git a/include/linux/nfs_fs_sb.h b/include/linux/nfs_fs_sb.h
index a9b44f12623f..d30c0245031c 100644
--- a/include/linux/nfs_fs_sb.h
+++ b/include/linux/nfs_fs_sb.h
@@ -255,6 +255,8 @@ struct nfs_server {
 	struct list_head	layouts;
 	struct list_head	delegations;
 	atomic_long_t		nr_active_delegations;
+	unsigned int		delegation_hash_mask;
+	struct hlist_head	*delegation_hash_table;
 	struct list_head	ss_copies;
 	struct list_head	ss_src_copies;