Revert "net/rds: Use per-namespace bind hash tables"

This reverts commit 48291e9001bc5f73a38045571ff60c608b9fc41e.

Orabug: 35811116

Signed-off-by: Greg Jumper <[email protected]>
Reviewed-by: Sharath Srinivasan <[email protected]>

Author: Greg Jumper

net/rds/af_rds.c

@@ -203,7 +203,7 @@ static int rds_release(struct socket *sock)
 	rds_clear_recv_queue(rs);
 	rds_cong_remove_socket(rs);
 
-	rds_remove_bound(rs->rs_rns, rs);
+	rds_remove_bound(rs);
 
 	rds_send_drop_to(rs, NULL);
 	rds_notify_queue_get(rs, NULL);
@@ -1484,6 +1484,8 @@ static int __init rds_init(void)
 	if (ret)
 		goto out_slab;
 
+	rds_bind_lock_init();
+
 	ret = rds_conn_init();
 	if (ret)
 		goto out_net;

net/rds/bind.c

@@ -44,11 +44,11 @@ struct bind_bucket {
 };
 
 #define BIND_HASH_SIZE 8192
+static struct bind_bucket bind_hash_table[BIND_HASH_SIZE];
 
-static struct bind_bucket *hash_to_bucket(struct rds_net *rds_ns,
-					  struct in6_addr *addr, __be16 port)
+static struct bind_bucket *hash_to_bucket(struct in6_addr *addr, __be16 port)
 {
-	return rds_ns->rns_bind_hash_table +
+	return bind_hash_table +
 		(jhash_3words((__force u32)(addr->s6_addr32[0] ^
 					    addr->s6_addr32[1]),
 			      (__force u32)(addr->s6_addr32[2] ^
@@ -99,13 +99,12 @@ static struct rds_sock *rds_bind_lookup(struct bind_bucket *bucket,
  * The rx path can race with rds_release.  We notice if rds_release() has
  * marked this socket and don't return a rs ref to the rx path.
  */
-struct rds_sock *rds_find_bound(struct rds_net *rds_ns,
-				struct in6_addr *addr, __be16 port,
+struct rds_sock *rds_find_bound(struct in6_addr *addr, __be16 port,
 				__u32 scope_id)
 {
 	struct rds_sock *rs;
 	unsigned long flags;
-	struct bind_bucket *bucket = hash_to_bucket(rds_ns, addr, port);
+	struct bind_bucket *bucket = hash_to_bucket(addr, port);
 
 	read_lock_irqsave(&bucket->lock, flags);
 	rs = rds_bind_lookup(bucket, addr, port, NULL, scope_id);
@@ -123,8 +122,8 @@ struct rds_sock *rds_find_bound(struct rds_net *rds_ns,
 }
 
 /* returns -ve errno or +ve port */
-static int rds_add_bound(struct rds_net *rds_ns, struct rds_sock *rs,
-			 struct in6_addr *addr, __be16 *port, __u32 scope_id)
+static int rds_add_bound(struct rds_sock *rs, struct in6_addr *addr,
+			 __be16 *port, __u32 scope_id)
 {
 	unsigned long flags;
 	int ret = -EADDRINUSE;
@@ -144,7 +143,7 @@ static int rds_add_bound(struct rds_net *rds_ns, struct rds_sock *rs,
 		if (rover == 0)
 			rover++;
 
-		bucket = hash_to_bucket(rds_ns, addr, cpu_to_be16(rover));
+		bucket = hash_to_bucket(addr, cpu_to_be16(rover));
 
 		write_lock_irqsave(&bucket->lock, flags);
 		rrs = rds_bind_lookup(bucket, addr, cpu_to_be16(rover), rs,
@@ -164,11 +163,11 @@ static int rds_add_bound(struct rds_net *rds_ns, struct rds_sock *rs,
 	return ret;
 }
 
-void rds_remove_bound(struct rds_net *rds_ns, struct rds_sock *rs)
+void rds_remove_bound(struct rds_sock *rs)
 {
 	unsigned long flags;
 	struct bind_bucket *bucket =
-		hash_to_bucket(rds_ns, &rs->rs_bound_addr, rs->rs_bound_port);
+		hash_to_bucket(&rs->rs_bound_addr, rs->rs_bound_port);
 
 	write_lock_irqsave(&bucket->lock, flags);
 
@@ -191,9 +190,7 @@ int rds_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 	struct rds_sock *rs = rds_sk_to_rs(sk);
 	struct in6_addr v6addr, *binding_addr;
 	struct rds_transport *trans;
-	struct rds_net *rds_ns;
 	__u32 scope_id = 0;
-	struct net *net;
 	int ret = 0;
 	__be16 port;
 	bool release_trans_on_error;
@@ -268,20 +265,21 @@ int rds_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 		goto out;
 	}
 
-	net = sock_net(sock->sk);
 	/* The transport can be set using SO_RDS_TRANSPORT option before the
 	 * socket is bound.
 	 */
 	if (rs->rs_transport) {
 		trans = rs->rs_transport;
 		if (!trans->laddr_check ||
-		    trans->laddr_check(net, binding_addr, scope_id) != 0) {
+		    trans->laddr_check(sock_net(sock->sk),
+				       binding_addr, scope_id) != 0) {
 			ret = -ENOPROTOOPT;
 			goto out;
 		}
 		release_trans_on_error = false;
 	} else {
-		trans = rds_trans_get_preferred(net, binding_addr, scope_id);
+		trans = rds_trans_get_preferred(sock_net(sock->sk),
+						binding_addr, scope_id);
 		if (!trans) {
 			ret = -EADDRNOTAVAIL;
 			pr_info_ratelimited("RDS: %s could not find a transport for %pI6c, load rds_tcp or rds_rdma?\n",
@@ -292,8 +290,7 @@ int rds_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 		release_trans_on_error = true;
 	}
 
-	rds_ns = rs->rs_rns;
-	ret = rds_add_bound(rds_ns, rs, binding_addr, &port, scope_id);
+	ret = rds_add_bound(rs, binding_addr, &port, scope_id);
 	if (ret && release_trans_on_error) {
 		rds_trans_put(rs->rs_transport);
 		rs->rs_transport = NULL;
@@ -304,33 +301,9 @@ int rds_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
 	return ret;
 }
 
-int rds_bind_tbl_net_init(struct rds_net *rds_ns)
+void rds_bind_lock_init(void)
 {
-	struct bind_bucket *bind_hash_table;
 	int i;
-
-	bind_hash_table = kzalloc(sizeof(*bind_hash_table) *
-				  BIND_HASH_SIZE, GFP_KERNEL);
-	if (!bind_hash_table)
-		return -ENOMEM;
-
-	for (i = 0; i < BIND_HASH_SIZE; i++) {
-		rwlock_init(&bind_hash_table[i].lock);
-		INIT_HLIST_HEAD(&bind_hash_table[i].head);
-	}
-	rds_ns->rns_bind_hash_table = bind_hash_table;
-
-	return 0;
-}
-
-void rds_bind_tbl_net_exit(struct rds_net *rds_ns)
-{
-	struct bind_bucket *bind_hash_table;
-	int i;
-
-	bind_hash_table = rds_ns->rns_bind_hash_table;
 	for (i = 0; i < BIND_HASH_SIZE; i++)
-		WARN_ON(!hlist_empty(&bind_hash_table[i].head));
-	kfree(bind_hash_table);
-	rds_ns->rns_bind_hash_table = NULL;
+		rwlock_init(&bind_hash_table[i].lock);
 }

net/rds/rds.h

@@ -245,15 +245,11 @@ enum rds_hb_state {
 	HB_PONG_RCVD,
 };
 
-struct bind_bucket;
-
 struct rds_net {
 	/* The following socket info is used for stats gathering */
 	struct mutex		rns_sock_lock;
 	u32			rns_sock_count;
 	struct list_head	rns_sock_list;
-
-	struct bind_bucket	*rns_bind_hash_table;
 };
 
 #define IS_CANONICAL(laddr, faddr) (htonl(laddr) < htonl(faddr))
@@ -381,7 +377,6 @@ struct rds_connection {
 	unsigned int		c_proposed_version;
 	unsigned int		c_version;
 	possible_net_t		c_net;
-	struct rds_net		*c_rns;
 
 	/* Re-connect stall diagnostics */
 	unsigned long           c_reconnect_start;
@@ -419,15 +414,11 @@ struct net *rds_conn_net(struct rds_connection *conn)
 	return read_pnet(&conn->c_net);
 }
 
-struct rds_net *rds_ns(struct net *);
-
 static inline
 void rds_conn_net_set(struct rds_connection *conn, struct net *net)
 {
 	/* Once set, never changed. */
 	write_pnet(&conn->c_net, net);
-
-	conn->c_rns = rds_ns(net);
 }
 
 #define RDS_FLAG_CONG_BITMAP		0x01
@@ -1102,12 +1093,10 @@ void debug_sock_put(struct sock *sock);
 
 /* bind.c */
 int rds_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len);
-void rds_remove_bound(struct rds_net *rds_ns, struct rds_sock *rs);
-struct rds_sock *rds_find_bound(struct rds_net *rds_ns,
-				struct in6_addr *addr, __be16 port,
+void rds_remove_bound(struct rds_sock *rs);
+struct rds_sock *rds_find_bound(struct in6_addr *addr, __be16 port,
 				__u32 scope_id);
-int rds_bind_tbl_net_init(struct rds_net *rds_ns);
-void rds_bind_tbl_net_exit(struct rds_net *rds_ns);
+void rds_bind_lock_init(void);
 
 /* cong.c */
 int rds_cong_monitor_init(void);
@@ -1540,6 +1529,7 @@ int rds_trans_init(void);
 void rds_trans_exit(void);
 
 /* rds_ns.c */
+struct rds_net *rds_ns(struct net *net);
 int rds_reg_pernet(void);
 void rds_unreg_pernet(void);
 

net/rds/rds_ns.c

@@ -47,7 +47,6 @@ EXPORT_SYMBOL(rds_ns);
 static __net_init int rds_init_net(struct net *net)
 {
 	struct rds_net *rns;
-	int ret;
 
 	rns = rds_ns(net);
 
@@ -56,23 +55,13 @@ static __net_init int rds_init_net(struct net *net)
 	mutex_init(&rns->rns_sock_lock);
 	INIT_LIST_HEAD(&rns->rns_sock_list);
 
-	ret = rds_bind_tbl_net_init(rns);
-	if (ret)
-		goto err;
-
 	return 0;
-
-err:
-	mutex_destroy(&rns->rns_sock_lock);
-	return ret;
 }
 
 static void rds_exit_net(struct net *net)
 {
 	struct rds_net *rns = rds_ns(net);
 
-	rds_bind_tbl_net_exit(rns);
-
 	mutex_destroy(&rns->rns_sock_lock);
 }
 

net/rds/recv.c

@@ -455,7 +455,7 @@ void rds_recv_incoming(struct rds_connection *conn, struct in6_addr *saddr,
 	if (!inc->i_hdr.h_sport)
 		rds_stats_inc(s_recv_pong);
 
-	rs = rds_find_bound(conn->c_rns, daddr, inc->i_hdr.h_dport, conn->c_bound_if);
+	rs = rds_find_bound(daddr, inc->i_hdr.h_dport, conn->c_bound_if);
 	if (!rs) {
 		rds_stats_inc(s_recv_drop_no_sock);
 		dropreason = "no socket";