| 1 | |
| 2 | * Copyright (c) 2006, 2017 Oracle and/or its affiliates. All rights reserved.
|
| 3 | *
|
| 4 | * This software is available to you under a choice of one of two
|
| 5 | * licenses. You may choose to be licensed under the terms of the GNU
|
| 6 | * General Public License (GPL) Version 2, available from the file
|
| 7 | * COPYING in the main directory of this source tree, or the
|
| 8 | * OpenIB.org BSD license below:
|
| 9 | *
|
| 10 | * Redistribution and use in source and binary forms, with or
|
| 11 | * without modification, are permitted provided that the following
|
| 12 | * conditions are met:
|
| 13 | *
|
| 14 | * - Redistributions of source code must retain the above
|
| 15 | * copyright notice, this list of conditions and the following
|
| 16 | * disclaimer.
|
| 17 | *
|
| 18 | * - Redistributions in binary form must reproduce the above
|
| 19 | * copyright notice, this list of conditions and the following
|
| 20 | * disclaimer in the documentation and/or other materials
|
| 21 | * provided with the distribution.
|
| 22 | *
|
| 23 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
|
| 24 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
|
| 25 | * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
|
| 26 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
|
| 27 | * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
|
| 28 | * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
|
| 29 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
| 30 | * SOFTWARE.
|
| 31 | *
|
| 32 |
|
| 33 | #include <linux/kernel.h>
|
| 34 | #include <linux/in.h>
|
| 35 | #include <net/tcp.h>
|
| 36 |
|
| 37 | #include "rds.h"
|
| 38 | #include "tcp.h"
|
| 39 |
|
| 40 | void rds_tcp_state_change(struct sock *sk)
|
| 41 | {
|
| 42 | void (*state_change)(struct sock *sk);
|
| 43 | struct rds_conn_path *cp;
|
| 44 | struct rds_tcp_connection *tc;
|
| 45 |
|
| 46 | read_lock_bh(&sk->sk_callback_lock);
|
| 47 | cp = sk->sk_user_data;
|
| 48 | if (!cp) {
|
| 49 | state_change = sk->sk_state_change;
|
| 50 | goto out;
|
| 51 | }
|
| 52 | tc = cp->cp_transport_data;
|
| 53 | state_change = tc->t_orig_state_change;
|
| 54 |
|
| 55 | rdsdebug("sock %p state_change to %d\n", tc->t_sock, sk->sk_state);
|
| 56 |
|
| 57 | switch (sk->sk_state) {
|
| 58 |
|
| 59 | case TCP_SYN_SENT:
|
| 60 | case TCP_SYN_RECV:
|
| 61 | break;
|
| 62 | case TCP_ESTABLISHED:
|
| 63 | |
| 64 | * TCP ports going from <smaller-ip>.<transient> to
|
| 65 | * <larger-ip>.<RDS_TCP_PORT>. We avoid marking the
|
| 66 | * RDS connection as RDS_CONN_UP until the reconnect,
|
| 67 | * to avoid RDS datagram loss.
|
| 68 |
|
| 69 | if (rds_addr_cmp(&cp->cp_conn->c_laddr,
|
| 70 | &cp->cp_conn->c_faddr) >= 0 &&
|
| 71 | rds_conn_path_transition(cp, RDS_CONN_CONNECTING,
|
| 72 | RDS_CONN_ERROR)) {
|
| 73 | rds_conn_path_drop(cp, false);
|
| 74 | } else {
|
| 75 | rds_connect_path_complete(cp, RDS_CONN_CONNECTING);
|
| 76 | }
|
| 77 | break;
|
| 78 | case TCP_CLOSING:
|
| 79 | case TCP_TIME_WAIT:
|
| 80 | if (wq_has_sleeper(&tc->t_recv_done_waitq))
|
| 81 | wake_up(&tc->t_recv_done_waitq);
|
| 82 | break;
|
| 83 | case TCP_CLOSE_WAIT:
|
| 84 | case TCP_LAST_ACK:
|
| 85 | case TCP_CLOSE:
|
| 86 | if (wq_has_sleeper(&tc->t_recv_done_waitq))
|
| 87 | wake_up(&tc->t_recv_done_waitq);
|
| 88 | rds_conn_path_drop(cp, false);
|
| 89 | break;
|
| 90 | default:
|
| 91 | break;
|
| 92 | }
|
| 93 | out:
|
| 94 | read_unlock_bh(&sk->sk_callback_lock);
|
| 95 | state_change(sk);
|
| 96 | }
|
| 97 |
|
| 98 | int rds_tcp_conn_path_connect(struct rds_conn_path *cp)
|
| 99 | {
|
| 100 | struct socket *sock = NULL;
|
| 101 | struct sockaddr_in6 sin6;
|
| 102 | struct sockaddr_in sin;
|
| 103 | struct sockaddr *addr;
|
| 104 | int port_low, port_high, port;
|
| 105 | int port_groups, groups_left;
|
| 106 | int addrlen;
|
| 107 | bool isv6;
|
| 108 | int ret;
|
| 109 | struct rds_connection *conn = cp->cp_conn;
|
| 110 | struct rds_tcp_connection *tc = cp->cp_transport_data;
|
| 111 |
|
| 112 | |
| 113 | * the handshake probe has determined the number of paths.
|
| 114 |
|
| 115 | if (cp->cp_index > 0 && cp->cp_conn->c_npaths < 2)
|
| 116 | return -EAGAIN;
|
| 117 |
|
| 118 | mutex_lock(&tc->t_conn_path_lock);
|
| 119 |
|
| 120 | if (rds_conn_path_up(cp)) {
|
| 121 | mutex_unlock(&tc->t_conn_path_lock);
|
| 122 | return 0;
|
| 123 | }
|
| 124 | if (ipv6_addr_v4mapped(&conn->c_laddr)) {
|
| 125 | ret = sock_create_kern(rds_conn_net(conn), PF_INET,
|
| 126 | SOCK_STREAM, IPPROTO_TCP, &sock);
|
| 127 | isv6 = false;
|
| 128 | } else {
|
| 129 | ret = sock_create_kern(rds_conn_net(conn), PF_INET6,
|
| 130 | SOCK_STREAM, IPPROTO_TCP, &sock);
|
| 131 | isv6 = true;
|
| 132 | }
|
| 133 |
|
| 134 | if (ret < 0)
|
| 135 | goto out;
|
| 136 |
|
| 137 | if (!rds_tcp_tune(sock)) {
|
| 138 | ret = -EINVAL;
|
| 139 | goto out;
|
| 140 | }
|
| 141 |
|
| 142 | if (isv6) {
|
| 143 | sin6.sin6_family = AF_INET6;
|
| 144 | sin6.sin6_addr = conn->c_laddr;
|
| 145 | sin6.sin6_port = 0;
|
| 146 | sin6.sin6_flowinfo = 0;
|
| 147 | sin6.sin6_scope_id = conn->c_dev_if;
|
| 148 | addr = (struct sockaddr *)&sin6;
|
| 149 | addrlen = sizeof(sin6);
|
| 150 | } else {
|
| 151 | sin.sin_family = AF_INET;
|
| 152 | sin.sin_addr.s_addr = conn->c_laddr.s6_addr32[3];
|
| 153 | sin.sin_port = 0;
|
| 154 | addr = (struct sockaddr *)&sin;
|
| 155 | addrlen = sizeof(sin);
|
| 156 | }
|
| 157 |
|
| 158 |
|
| 159 | inet_get_local_port_range(rds_conn_net(conn), &port_low, &port_high);
|
| 160 | port_low = ALIGN(port_low, RDS_MPATH_WORKERS);
|
| 161 | port_groups = (port_high - port_low + 1) / RDS_MPATH_WORKERS;
|
| 162 | ret = -EADDRINUSE;
|
| 163 | groups_left = port_groups;
|
| 164 | while (groups_left-- > 0 && ret) {
|
| 165 | if (++tc->t_client_port_group >= port_groups)
|
| 166 | tc->t_client_port_group = 0;
|
| 167 | port = port_low +
|
| 168 | tc->t_client_port_group * RDS_MPATH_WORKERS +
|
| 169 | cp->cp_index;
|
| 170 |
|
| 171 | if (isv6)
|
| 172 | sin6.sin6_port = htons(port);
|
| 173 | else
|
| 174 | sin.sin_port = htons(port);
|
| 175 | ret = kernel_bind(sock, (struct sockaddr_unsized *)addr,
|
| 176 | addrlen);
|
| 177 | }
|
| 178 | if (ret) {
|
| 179 | rdsdebug("bind failed with %d at address %pI6c\n",
|
| 180 | ret, &conn->c_laddr);
|
| 181 | goto out;
|
| 182 | }
|
| 183 |
|
| 184 | if (isv6) {
|
| 185 | sin6.sin6_family = AF_INET6;
|
| 186 | sin6.sin6_addr = conn->c_faddr;
|
| 187 | sin6.sin6_port = htons(RDS_TCP_PORT);
|
| 188 | sin6.sin6_flowinfo = 0;
|
| 189 | sin6.sin6_scope_id = conn->c_dev_if;
|
| 190 | addr = (struct sockaddr *)&sin6;
|
| 191 | addrlen = sizeof(sin6);
|
| 192 | } else {
|
| 193 | sin.sin_family = AF_INET;
|
| 194 | sin.sin_addr.s_addr = conn->c_faddr.s6_addr32[3];
|
| 195 | sin.sin_port = htons(RDS_TCP_PORT);
|
| 196 | addr = (struct sockaddr *)&sin;
|
| 197 | addrlen = sizeof(sin);
|
| 198 | }
|
| 199 |
|
| 200 | |
| 201 | * once we call connect() we can start getting callbacks and they
|
| 202 | * own the socket
|
| 203 |
|
| 204 | rds_tcp_set_callbacks(sock, cp);
|
| 205 | ret = kernel_connect(sock, (struct sockaddr_unsized *)addr, addrlen, O_NONBLOCK);
|
| 206 |
|
| 207 | rdsdebug("connect to address %pI6c returned %d\n", &conn->c_faddr, ret);
|
| 208 | if (ret == -EINPROGRESS)
|
| 209 | ret = 0;
|
| 210 | if (ret == 0) {
|
| 211 | rds_tcp_keepalive(sock);
|
| 212 | sock = NULL;
|
| 213 | } else {
|
| 214 | rds_tcp_restore_callbacks(sock, cp->cp_transport_data);
|
| 215 | }
|
| 216 |
|
| 217 | out:
|
| 218 | mutex_unlock(&tc->t_conn_path_lock);
|
| 219 | if (sock)
|
| 220 | sock_release(sock);
|
| 221 | return ret;
|
| 222 | }
|
| 223 |
|
| 224 | |
| 225 | * Before killing the tcp socket this needs to serialize with callbacks. The
|
| 226 | * caller has already grabbed the sending sem so we're serialized with other
|
| 227 | * senders.
|
| 228 | *
|
| 229 | * TCP calls the callbacks with the sock lock so we hold it while we reset the
|
| 230 | * callbacks to those set by TCP. Our callbacks won't execute again once we
|
| 231 | * hold the sock lock.
|
| 232 |
|
| 233 | void rds_tcp_conn_path_shutdown(struct rds_conn_path *cp)
|
| 234 | {
|
| 235 | struct rds_tcp_connection *tc = cp->cp_transport_data;
|
| 236 | struct socket *sock = tc->t_sock;
|
| 237 | struct sock *sk;
|
| 238 | unsigned int rounds;
|
| 239 |
|
| 240 | rdsdebug("shutting down conn %p tc %p sock %p\n",
|
| 241 | cp->cp_conn, tc, sock);
|
| 242 |
|
| 243 | if (sock) {
|
| 244 | sk = sock->sk;
|
| 245 | if (rds_destroy_pending(cp->cp_conn))
|
| 246 | sock_no_linger(sk);
|
| 247 |
|
| 248 | sock->ops->shutdown(sock, SHUT_WR);
|
| 249 |
|
| 250 | |
| 251 | * wait until we processed all incoming messages
|
| 252 | * and we're sure that there won't be any more:
|
| 253 | * i.e. state CLOSING, TIME_WAIT, CLOSE_WAIT,
|
| 254 | * LAST_ACK, or CLOSE (RFC 793).
|
| 255 | *
|
| 256 | * Give up waiting after 5 seconds and allow messages
|
| 257 | * to theoretically get dropped, if the TCP transition
|
| 258 | * didn't happen.
|
| 259 |
|
| 260 | rounds = 0;
|
| 261 | do {
|
| 262 | |
| 263 | * since "rds_recv_worker" only dispatches messages
|
| 264 | * while the connection is still in RDS_CONN_UP
|
| 265 | * and there is no guarantee that "rds_tcp_data_ready"
|
| 266 | * was called nor that "sk_data_ready" still points to
|
| 267 | * it.
|
| 268 |
|
| 269 | rds_tcp_recv_path(cp);
|
| 270 | } while (!wait_event_timeout(tc->t_recv_done_waitq,
|
| 271 | (sk->sk_state == TCP_CLOSING ||
|
| 272 | sk->sk_state == TCP_TIME_WAIT ||
|
| 273 | sk->sk_state == TCP_CLOSE_WAIT ||
|
| 274 | sk->sk_state == TCP_LAST_ACK ||
|
| 275 | sk->sk_state == TCP_CLOSE) &&
|
| 276 | skb_queue_empty_lockless(&sk->sk_receive_queue),
|
| 277 | msecs_to_jiffies(100)) &&
|
| 278 | ++rounds < 50);
|
| 279 | lock_sock(sk);
|
| 280 |
|
| 281 |
|
| 282 | tc->t_last_seen_una = rds_tcp_snd_una(tc);
|
| 283 | rds_send_path_drop_acked(cp, rds_tcp_snd_una(tc),
|
| 284 | rds_tcp_is_acked);
|
| 285 |
|
| 286 | rds_tcp_restore_callbacks(sock, tc);
|
| 287 |
|
| 288 | release_sock(sk);
|
| 289 | sock_release(sock);
|
| 290 | }
|
| 291 |
|
| 292 | if (tc->t_tinc) {
|
| 293 | rds_inc_put(&tc->t_tinc->ti_inc);
|
| 294 | tc->t_tinc = NULL;
|
| 295 | }
|
| 296 | tc->t_tinc_hdr_rem = sizeof(struct rds_header);
|
| 297 | tc->t_tinc_data_rem = 0;
|
| 298 | }
|