|
|
@@ -136,9 +136,9 @@ void rds_tcp_restore_callbacks(struct socket *sock,
|
|
|
* from being called while it isn't set.
|
|
|
*/
|
|
|
void rds_tcp_reset_callbacks(struct socket *sock,
|
|
|
- struct rds_connection *conn)
|
|
|
+ struct rds_conn_path *cp)
|
|
|
{
|
|
|
- struct rds_tcp_connection *tc = conn->c_transport_data;
|
|
|
+ struct rds_tcp_connection *tc = cp->cp_transport_data;
|
|
|
struct socket *osock = tc->t_sock;
|
|
|
|
|
|
if (!osock)
|
|
|
@@ -148,8 +148,8 @@ void rds_tcp_reset_callbacks(struct socket *sock,
|
|
|
* We have an outstanding SYN to this peer, which may
|
|
|
* potentially have transitioned to the RDS_CONN_UP state,
|
|
|
* so we must quiesce any send threads before resetting
|
|
|
- * c_transport_data. We quiesce these threads by setting
|
|
|
- * c_state to something other than RDS_CONN_UP, and then
|
|
|
+ * cp_transport_data. We quiesce these threads by setting
|
|
|
+ * cp_state to something other than RDS_CONN_UP, and then
|
|
|
* waiting for any existing threads in rds_send_xmit to
|
|
|
* complete release_in_xmit(). (Subsequent threads entering
|
|
|
* rds_send_xmit() will bail on !rds_conn_up().
|
|
|
@@ -164,8 +164,8 @@ void rds_tcp_reset_callbacks(struct socket *sock,
|
|
|
* RDS_CONN_RESETTTING, to ensure that rds_tcp_state_change
|
|
|
* cannot mark rds_conn_path_up() in the window before lock_sock()
|
|
|
*/
|
|
|
- atomic_set(&conn->c_state, RDS_CONN_RESETTING);
|
|
|
- wait_event(conn->c_waitq, !test_bit(RDS_IN_XMIT, &conn->c_flags));
|
|
|
+ atomic_set(&cp->cp_state, RDS_CONN_RESETTING);
|
|
|
+ wait_event(cp->cp_waitq, !test_bit(RDS_IN_XMIT, &cp->cp_flags));
|
|
|
lock_sock(osock->sk);
|
|
|
/* reset receive side state for rds_tcp_data_recv() for osock */
|
|
|
if (tc->t_tinc) {
|
|
|
@@ -186,11 +186,12 @@ void rds_tcp_reset_callbacks(struct socket *sock,
|
|
|
release_sock(osock->sk);
|
|
|
sock_release(osock);
|
|
|
newsock:
|
|
|
- rds_send_path_reset(&conn->c_path[0]);
|
|
|
+ rds_send_path_reset(cp);
|
|
|
lock_sock(sock->sk);
|
|
|
write_lock_bh(&sock->sk->sk_callback_lock);
|
|
|
tc->t_sock = sock;
|
|
|
- sock->sk->sk_user_data = conn;
|
|
|
+ tc->t_cpath = cp;
|
|
|
+ sock->sk->sk_user_data = cp;
|
|
|
sock->sk->sk_data_ready = rds_tcp_data_ready;
|
|
|
sock->sk->sk_write_space = rds_tcp_write_space;
|
|
|
sock->sk->sk_state_change = rds_tcp_state_change;
|
|
|
@@ -203,9 +204,9 @@ newsock:
|
|
|
* above rds_tcp_reset_callbacks for notes about synchronization
|
|
|
* with data path
|
|
|
*/
|
|
|
-void rds_tcp_set_callbacks(struct socket *sock, struct rds_connection *conn)
|
|
|
+void rds_tcp_set_callbacks(struct socket *sock, struct rds_conn_path *cp)
|
|
|
{
|
|
|
- struct rds_tcp_connection *tc = conn->c_transport_data;
|
|
|
+ struct rds_tcp_connection *tc = cp->cp_transport_data;
|
|
|
|
|
|
rdsdebug("setting sock %p callbacks to tc %p\n", sock, tc);
|
|
|
write_lock_bh(&sock->sk->sk_callback_lock);
|
|
|
@@ -221,12 +222,12 @@ void rds_tcp_set_callbacks(struct socket *sock, struct rds_connection *conn)
|
|
|
sock->sk->sk_data_ready = sock->sk->sk_user_data;
|
|
|
|
|
|
tc->t_sock = sock;
|
|
|
- tc->conn = conn;
|
|
|
+ tc->t_cpath = cp;
|
|
|
tc->t_orig_data_ready = sock->sk->sk_data_ready;
|
|
|
tc->t_orig_write_space = sock->sk->sk_write_space;
|
|
|
tc->t_orig_state_change = sock->sk->sk_state_change;
|
|
|
|
|
|
- sock->sk->sk_user_data = conn;
|
|
|
+ sock->sk->sk_user_data = cp;
|
|
|
sock->sk->sk_data_ready = rds_tcp_data_ready;
|
|
|
sock->sk->sk_write_space = rds_tcp_write_space;
|
|
|
sock->sk->sk_state_change = rds_tcp_state_change;
|
|
|
@@ -284,24 +285,29 @@ static int rds_tcp_laddr_check(struct net *net, __be32 addr)
|
|
|
static int rds_tcp_conn_alloc(struct rds_connection *conn, gfp_t gfp)
|
|
|
{
|
|
|
struct rds_tcp_connection *tc;
|
|
|
+ int i;
|
|
|
|
|
|
- tc = kmem_cache_alloc(rds_tcp_conn_slab, gfp);
|
|
|
- if (!tc)
|
|
|
- return -ENOMEM;
|
|
|
+ for (i = 0; i < RDS_MPATH_WORKERS; i++) {
|
|
|
+ tc = kmem_cache_alloc(rds_tcp_conn_slab, gfp);
|
|
|
+ if (!tc)
|
|
|
+ return -ENOMEM;
|
|
|
|
|
|
- mutex_init(&tc->t_conn_lock);
|
|
|
- tc->t_sock = NULL;
|
|
|
- tc->t_tinc = NULL;
|
|
|
- tc->t_tinc_hdr_rem = sizeof(struct rds_header);
|
|
|
- tc->t_tinc_data_rem = 0;
|
|
|
+ mutex_init(&tc->t_conn_path_lock);
|
|
|
+ tc->t_sock = NULL;
|
|
|
+ tc->t_tinc = NULL;
|
|
|
+ tc->t_tinc_hdr_rem = sizeof(struct rds_header);
|
|
|
+ tc->t_tinc_data_rem = 0;
|
|
|
|
|
|
- conn->c_transport_data = tc;
|
|
|
+ conn->c_path[i].cp_transport_data = tc;
|
|
|
+ tc->t_cpath = &conn->c_path[i];
|
|
|
|
|
|
- spin_lock_irq(&rds_tcp_conn_lock);
|
|
|
- list_add_tail(&tc->t_tcp_node, &rds_tcp_conn_list);
|
|
|
- spin_unlock_irq(&rds_tcp_conn_lock);
|
|
|
+ spin_lock_irq(&rds_tcp_conn_lock);
|
|
|
+ list_add_tail(&tc->t_tcp_node, &rds_tcp_conn_list);
|
|
|
+ spin_unlock_irq(&rds_tcp_conn_lock);
|
|
|
+ rdsdebug("rds_conn_path [%d] tc %p\n", i,
|
|
|
+ conn->c_path[i].cp_transport_data);
|
|
|
+ }
|
|
|
|
|
|
- rdsdebug("alloced tc %p\n", conn->c_transport_data);
|
|
|
return 0;
|
|
|
}
|
|
|
|
|
|
@@ -318,6 +324,17 @@ static void rds_tcp_conn_free(void *arg)
|
|
|
kmem_cache_free(rds_tcp_conn_slab, tc);
|
|
|
}
|
|
|
|
|
|
+static bool list_has_conn(struct list_head *list, struct rds_connection *conn)
|
|
|
+{
|
|
|
+ struct rds_tcp_connection *tc, *_tc;
|
|
|
+
|
|
|
+ list_for_each_entry_safe(tc, _tc, list, t_tcp_node) {
|
|
|
+ if (tc->t_cpath->cp_conn == conn)
|
|
|
+ return true;
|
|
|
+ }
|
|
|
+ return false;
|
|
|
+}
|
|
|
+
|
|
|
static void rds_tcp_destroy_conns(void)
|
|
|
{
|
|
|
struct rds_tcp_connection *tc, *_tc;
|
|
|
@@ -325,29 +342,28 @@ static void rds_tcp_destroy_conns(void)
|
|
|
|
|
|
/* avoid calling conn_destroy with irqs off */
|
|
|
spin_lock_irq(&rds_tcp_conn_lock);
|
|
|
- list_splice(&rds_tcp_conn_list, &tmp_list);
|
|
|
- INIT_LIST_HEAD(&rds_tcp_conn_list);
|
|
|
+ list_for_each_entry_safe(tc, _tc, &rds_tcp_conn_list, t_tcp_node) {
|
|
|
+ if (!list_has_conn(&tmp_list, tc->t_cpath->cp_conn))
|
|
|
+ list_move_tail(&tc->t_tcp_node, &tmp_list);
|
|
|
+ }
|
|
|
spin_unlock_irq(&rds_tcp_conn_lock);
|
|
|
|
|
|
- list_for_each_entry_safe(tc, _tc, &tmp_list, t_tcp_node) {
|
|
|
- if (tc->conn->c_passive)
|
|
|
- rds_conn_destroy(tc->conn->c_passive);
|
|
|
- rds_conn_destroy(tc->conn);
|
|
|
- }
|
|
|
+ list_for_each_entry_safe(tc, _tc, &tmp_list, t_tcp_node)
|
|
|
+ rds_conn_destroy(tc->t_cpath->cp_conn);
|
|
|
}
|
|
|
|
|
|
static void rds_tcp_exit(void);
|
|
|
|
|
|
struct rds_transport rds_tcp_transport = {
|
|
|
.laddr_check = rds_tcp_laddr_check,
|
|
|
- .xmit_prepare = rds_tcp_xmit_prepare,
|
|
|
- .xmit_complete = rds_tcp_xmit_complete,
|
|
|
+ .xmit_path_prepare = rds_tcp_xmit_path_prepare,
|
|
|
+ .xmit_path_complete = rds_tcp_xmit_path_complete,
|
|
|
.xmit = rds_tcp_xmit,
|
|
|
- .recv = rds_tcp_recv,
|
|
|
+ .recv_path = rds_tcp_recv_path,
|
|
|
.conn_alloc = rds_tcp_conn_alloc,
|
|
|
.conn_free = rds_tcp_conn_free,
|
|
|
- .conn_connect = rds_tcp_conn_connect,
|
|
|
- .conn_shutdown = rds_tcp_conn_shutdown,
|
|
|
+ .conn_path_connect = rds_tcp_conn_path_connect,
|
|
|
+ .conn_path_shutdown = rds_tcp_conn_path_shutdown,
|
|
|
.inc_copy_to_user = rds_tcp_inc_copy_to_user,
|
|
|
.inc_free = rds_tcp_inc_free,
|
|
|
.stats_info_copy = rds_tcp_stats_info_copy,
|
|
|
@@ -489,10 +505,30 @@ static struct pernet_operations rds_tcp_net_ops = {
|
|
|
.size = sizeof(struct rds_tcp_net),
|
|
|
};
|
|
|
|
|
|
+/* explicitly send a RST on each socket, thereby releasing any socket refcnts
|
|
|
+ * that may otherwise hold up netns deletion.
|
|
|
+ */
|
|
|
+static void rds_tcp_conn_paths_destroy(struct rds_connection *conn)
|
|
|
+{
|
|
|
+ struct rds_conn_path *cp;
|
|
|
+ struct rds_tcp_connection *tc;
|
|
|
+ int i;
|
|
|
+ struct sock *sk;
|
|
|
+
|
|
|
+ for (i = 0; i < RDS_MPATH_WORKERS; i++) {
|
|
|
+ cp = &conn->c_path[i];
|
|
|
+ tc = cp->cp_transport_data;
|
|
|
+ if (!tc->t_sock)
|
|
|
+ continue;
|
|
|
+ sk = tc->t_sock->sk;
|
|
|
+ sk->sk_prot->disconnect(sk, 0);
|
|
|
+ tcp_done(sk);
|
|
|
+ }
|
|
|
+}
|
|
|
+
|
|
|
static void rds_tcp_kill_sock(struct net *net)
|
|
|
{
|
|
|
struct rds_tcp_connection *tc, *_tc;
|
|
|
- struct sock *sk;
|
|
|
LIST_HEAD(tmp_list);
|
|
|
struct rds_tcp_net *rtn = net_generic(net, rds_tcp_netid);
|
|
|
|
|
|
@@ -501,20 +537,17 @@ static void rds_tcp_kill_sock(struct net *net)
|
|
|
flush_work(&rtn->rds_tcp_accept_w);
|
|
|
spin_lock_irq(&rds_tcp_conn_lock);
|
|
|
list_for_each_entry_safe(tc, _tc, &rds_tcp_conn_list, t_tcp_node) {
|
|
|
- struct net *c_net = read_pnet(&tc->conn->c_net);
|
|
|
+ struct net *c_net = read_pnet(&tc->t_cpath->cp_conn->c_net);
|
|
|
|
|
|
if (net != c_net || !tc->t_sock)
|
|
|
continue;
|
|
|
- list_move_tail(&tc->t_tcp_node, &tmp_list);
|
|
|
+ if (!list_has_conn(&tmp_list, tc->t_cpath->cp_conn))
|
|
|
+ list_move_tail(&tc->t_tcp_node, &tmp_list);
|
|
|
}
|
|
|
spin_unlock_irq(&rds_tcp_conn_lock);
|
|
|
list_for_each_entry_safe(tc, _tc, &tmp_list, t_tcp_node) {
|
|
|
- sk = tc->t_sock->sk;
|
|
|
- sk->sk_prot->disconnect(sk, 0);
|
|
|
- tcp_done(sk);
|
|
|
- if (tc->conn->c_passive)
|
|
|
- rds_conn_destroy(tc->conn->c_passive);
|
|
|
- rds_conn_destroy(tc->conn);
|
|
|
+ rds_tcp_conn_paths_destroy(tc->t_cpath->cp_conn);
|
|
|
+ rds_conn_destroy(tc->t_cpath->cp_conn);
|
|
|
}
|
|
|
}
|
|
|
|
|
|
@@ -552,12 +585,13 @@ static void rds_tcp_sysctl_reset(struct net *net)
|
|
|
|
|
|
spin_lock_irq(&rds_tcp_conn_lock);
|
|
|
list_for_each_entry_safe(tc, _tc, &rds_tcp_conn_list, t_tcp_node) {
|
|
|
- struct net *c_net = read_pnet(&tc->conn->c_net);
|
|
|
+ struct net *c_net = read_pnet(&tc->t_cpath->cp_conn->c_net);
|
|
|
|
|
|
if (net != c_net || !tc->t_sock)
|
|
|
continue;
|
|
|
|
|
|
- rds_conn_drop(tc->conn); /* reconnect with new parameters */
|
|
|
+ /* reconnect with new parameters */
|
|
|
+ rds_conn_path_drop(tc->t_cpath);
|
|
|
}
|
|
|
spin_unlock_irq(&rds_tcp_conn_lock);
|
|
|
}
|
David S. Miller