]>
Commit | Line | Data |
---|---|---|
70041088 AG |
1 | /* |
2 | * Copyright (c) 2006 Oracle. 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> | |
5a0e3ad6 | 34 | #include <linux/gfp.h> |
70041088 AG |
35 | #include <linux/in.h> |
36 | #include <net/tcp.h> | |
37 | ||
38 | #include "rds.h" | |
39 | #include "tcp.h" | |
40 | ||
467fa153 | 41 | int rds_tcp_keepalive(struct socket *sock) |
f711a6ae SV |
42 | { |
43 | /* values below based on xs_udp_default_timeout */ | |
44 | int keepidle = 5; /* send a probe 'keepidle' secs after last data */ | |
45 | int keepcnt = 5; /* number of unack'ed probes before declaring dead */ | |
46 | int keepalive = 1; | |
47 | int ret = 0; | |
48 | ||
49 | ret = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE, | |
50 | (char *)&keepalive, sizeof(keepalive)); | |
51 | if (ret < 0) | |
52 | goto bail; | |
53 | ||
54 | ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPCNT, | |
55 | (char *)&keepcnt, sizeof(keepcnt)); | |
56 | if (ret < 0) | |
57 | goto bail; | |
58 | ||
59 | ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPIDLE, | |
60 | (char *)&keepidle, sizeof(keepidle)); | |
61 | if (ret < 0) | |
62 | goto bail; | |
63 | ||
64 | /* KEEPINTVL is the interval between successive probes. We follow | |
65 | * the model in xs_tcp_finish_connecting() and re-use keepidle. | |
66 | */ | |
67 | ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPINTVL, | |
68 | (char *)&keepidle, sizeof(keepidle)); | |
69 | bail: | |
70 | return ret; | |
71 | } | |
72 | ||
5916e2c1 SV |
73 | /* rds_tcp_accept_one_path(): if accepting on cp_index > 0, make sure the |
74 | * client's ipaddr < server's ipaddr. Otherwise, close the accepted | |
75 | * socket and force a reconneect from smaller -> larger ip addr. The reason | |
76 | * we special case cp_index 0 is to allow the rds probe ping itself to itself | |
77 | * get through efficiently. | |
78 | * Since reconnects are only initiated from the node with the numerically | |
79 | * smaller ip address, we recycle conns in RDS_CONN_ERROR on the passive side | |
80 | * by moving them to CONNECTING in this function. | |
81 | */ | |
82 | struct rds_tcp_connection *rds_tcp_accept_one_path(struct rds_connection *conn) | |
83 | { | |
84 | int i; | |
85 | bool peer_is_smaller = (conn->c_faddr < conn->c_laddr); | |
1a0e100f | 86 | int npaths = max_t(int, 1, conn->c_npaths); |
5916e2c1 | 87 | |
1a0e100f | 88 | /* for mprds, all paths MUST be initiated by the peer |
5916e2c1 SV |
89 | * with the smaller address. |
90 | */ | |
1a0e100f SV |
91 | if (!peer_is_smaller) { |
92 | /* Make sure we initiate at least one path if this | |
93 | * has not already been done; rds_start_mprds() will | |
94 | * take care of additional paths, if necessary. | |
95 | */ | |
96 | if (npaths == 1) | |
97 | rds_conn_path_connect_if_down(&conn->c_path[0]); | |
5916e2c1 | 98 | return NULL; |
1a0e100f | 99 | } |
5916e2c1 | 100 | |
117d15bb | 101 | for (i = 0; i < npaths; i++) { |
5916e2c1 SV |
102 | struct rds_conn_path *cp = &conn->c_path[i]; |
103 | ||
104 | if (rds_conn_path_transition(cp, RDS_CONN_DOWN, | |
105 | RDS_CONN_CONNECTING) || | |
106 | rds_conn_path_transition(cp, RDS_CONN_ERROR, | |
107 | RDS_CONN_CONNECTING)) { | |
108 | return cp->cp_transport_data; | |
109 | } | |
110 | } | |
111 | return NULL; | |
112 | } | |
113 | ||
467fa153 | 114 | int rds_tcp_accept_one(struct socket *sock) |
70041088 AG |
115 | { |
116 | struct socket *new_sock = NULL; | |
117 | struct rds_connection *conn; | |
118 | int ret; | |
119 | struct inet_sock *inet; | |
bd7c5f98 SV |
120 | struct rds_tcp_connection *rs_tcp = NULL; |
121 | int conn_state; | |
ea3b1ea5 | 122 | struct rds_conn_path *cp; |
70041088 | 123 | |
37e14f4f SV |
124 | if (!sock) /* module unload or netns delete in progress */ |
125 | return -ENETUNREACH; | |
126 | ||
d5a8ac28 SV |
127 | ret = sock_create_kern(sock_net(sock->sk), sock->sk->sk_family, |
128 | sock->sk->sk_type, sock->sk->sk_protocol, | |
129 | &new_sock); | |
70041088 AG |
130 | if (ret) |
131 | goto out; | |
132 | ||
133 | new_sock->type = sock->type; | |
134 | new_sock->ops = sock->ops; | |
135 | ret = sock->ops->accept(sock, new_sock, O_NONBLOCK); | |
136 | if (ret < 0) | |
137 | goto out; | |
138 | ||
f711a6ae SV |
139 | ret = rds_tcp_keepalive(new_sock); |
140 | if (ret < 0) | |
141 | goto out; | |
142 | ||
70041088 AG |
143 | rds_tcp_tune(new_sock); |
144 | ||
145 | inet = inet_sk(new_sock->sk); | |
146 | ||
6884b348 JP |
147 | rdsdebug("accepted tcp %pI4:%u -> %pI4:%u\n", |
148 | &inet->inet_saddr, ntohs(inet->inet_sport), | |
149 | &inet->inet_daddr, ntohs(inet->inet_dport)); | |
70041088 | 150 | |
d5a8ac28 SV |
151 | conn = rds_conn_create(sock_net(sock->sk), |
152 | inet->inet_saddr, inet->inet_daddr, | |
c720c7e8 | 153 | &rds_tcp_transport, GFP_KERNEL); |
70041088 AG |
154 | if (IS_ERR(conn)) { |
155 | ret = PTR_ERR(conn); | |
156 | goto out; | |
157 | } | |
f711a6ae | 158 | /* An incoming SYN request came in, and TCP just accepted it. |
f711a6ae SV |
159 | * |
160 | * If the client reboots, this conn will need to be cleaned up. | |
161 | * rds_tcp_state_change() will do that cleanup | |
162 | */ | |
5916e2c1 SV |
163 | rs_tcp = rds_tcp_accept_one_path(conn); |
164 | if (!rs_tcp) | |
165 | goto rst_nsk; | |
02105b2c | 166 | mutex_lock(&rs_tcp->t_conn_path_lock); |
5916e2c1 SV |
167 | cp = rs_tcp->t_cpath; |
168 | conn_state = rds_conn_path_state(cp); | |
1a0e100f SV |
169 | WARN_ON(conn_state == RDS_CONN_UP); |
170 | if (conn_state != RDS_CONN_CONNECTING && conn_state != RDS_CONN_ERROR) | |
bd7c5f98 | 171 | goto rst_nsk; |
eb192840 SV |
172 | if (rs_tcp->t_sock) { |
173 | /* Need to resolve a duelling SYN between peers. | |
174 | * We have an outstanding SYN to this peer, which may | |
175 | * potentially have transitioned to the RDS_CONN_UP state, | |
176 | * so we must quiesce any send threads before resetting | |
177 | * c_transport_data. | |
178 | */ | |
c948bb5c | 179 | if (ntohl(inet->inet_saddr) < ntohl(inet->inet_daddr) || |
5916e2c1 | 180 | !cp->cp_outgoing) { |
bd7c5f98 | 181 | goto rst_nsk; |
c948bb5c | 182 | } else { |
ea3b1ea5 | 183 | rds_tcp_reset_callbacks(new_sock, cp); |
5916e2c1 | 184 | cp->cp_outgoing = 0; |
9c79440e | 185 | /* rds_connect_path_complete() marks RDS_CONN_UP */ |
ea3b1ea5 | 186 | rds_connect_path_complete(cp, RDS_CONN_RESETTING); |
eb192840 | 187 | } |
335b48d9 | 188 | } else { |
ea3b1ea5 SV |
189 | rds_tcp_set_callbacks(new_sock, cp); |
190 | rds_connect_path_complete(cp, RDS_CONN_CONNECTING); | |
eb192840 | 191 | } |
70041088 AG |
192 | new_sock = NULL; |
193 | ret = 0; | |
bd7c5f98 SV |
194 | goto out; |
195 | rst_nsk: | |
196 | /* reset the newly returned accept sock and bail */ | |
335b48d9 | 197 | kernel_sock_shutdown(new_sock, SHUT_RDWR); |
bd7c5f98 | 198 | ret = 0; |
70041088 | 199 | out: |
bd7c5f98 | 200 | if (rs_tcp) |
02105b2c | 201 | mutex_unlock(&rs_tcp->t_conn_path_lock); |
70041088 AG |
202 | if (new_sock) |
203 | sock_release(new_sock); | |
204 | return ret; | |
205 | } | |
206 | ||
676d2369 | 207 | void rds_tcp_listen_data_ready(struct sock *sk) |
70041088 | 208 | { |
676d2369 | 209 | void (*ready)(struct sock *sk); |
70041088 AG |
210 | |
211 | rdsdebug("listen data ready sk %p\n", sk); | |
212 | ||
38036629 | 213 | read_lock_bh(&sk->sk_callback_lock); |
70041088 | 214 | ready = sk->sk_user_data; |
8690bfa1 | 215 | if (!ready) { /* check for teardown race */ |
70041088 AG |
216 | ready = sk->sk_data_ready; |
217 | goto out; | |
218 | } | |
219 | ||
220 | /* | |
221 | * ->sk_data_ready is also called for a newly established child socket | |
222 | * before it has been accepted and the accepter has set up their | |
223 | * data_ready.. we only want to queue listen work for our listening | |
224 | * socket | |
225 | */ | |
226 | if (sk->sk_state == TCP_LISTEN) | |
467fa153 | 227 | rds_tcp_accept_work(sk); |
a93d01f5 SV |
228 | else |
229 | ready = rds_tcp_listen_sock_def_readable(sock_net(sk)); | |
70041088 AG |
230 | |
231 | out: | |
38036629 | 232 | read_unlock_bh(&sk->sk_callback_lock); |
676d2369 | 233 | ready(sk); |
70041088 AG |
234 | } |
235 | ||
467fa153 | 236 | struct socket *rds_tcp_listen_init(struct net *net) |
70041088 AG |
237 | { |
238 | struct sockaddr_in sin; | |
239 | struct socket *sock = NULL; | |
240 | int ret; | |
241 | ||
467fa153 | 242 | ret = sock_create_kern(net, PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock); |
70041088 AG |
243 | if (ret < 0) |
244 | goto out; | |
245 | ||
4a17fd52 | 246 | sock->sk->sk_reuse = SK_CAN_REUSE; |
70041088 AG |
247 | rds_tcp_nonagle(sock); |
248 | ||
249 | write_lock_bh(&sock->sk->sk_callback_lock); | |
250 | sock->sk->sk_user_data = sock->sk->sk_data_ready; | |
251 | sock->sk->sk_data_ready = rds_tcp_listen_data_ready; | |
252 | write_unlock_bh(&sock->sk->sk_callback_lock); | |
253 | ||
01728371 | 254 | sin.sin_family = PF_INET; |
70041088 AG |
255 | sin.sin_addr.s_addr = (__force u32)htonl(INADDR_ANY); |
256 | sin.sin_port = (__force u16)htons(RDS_TCP_PORT); | |
257 | ||
258 | ret = sock->ops->bind(sock, (struct sockaddr *)&sin, sizeof(sin)); | |
259 | if (ret < 0) | |
260 | goto out; | |
261 | ||
262 | ret = sock->ops->listen(sock, 64); | |
263 | if (ret < 0) | |
264 | goto out; | |
265 | ||
467fa153 | 266 | return sock; |
70041088 AG |
267 | out: |
268 | if (sock) | |
269 | sock_release(sock); | |
467fa153 | 270 | return NULL; |
70041088 AG |
271 | } |
272 | ||
467fa153 | 273 | void rds_tcp_listen_stop(struct socket *sock) |
70041088 | 274 | { |
70041088 AG |
275 | struct sock *sk; |
276 | ||
8690bfa1 | 277 | if (!sock) |
70041088 AG |
278 | return; |
279 | ||
280 | sk = sock->sk; | |
281 | ||
282 | /* serialize with and prevent further callbacks */ | |
283 | lock_sock(sk); | |
284 | write_lock_bh(&sk->sk_callback_lock); | |
285 | if (sk->sk_user_data) { | |
286 | sk->sk_data_ready = sk->sk_user_data; | |
287 | sk->sk_user_data = NULL; | |
288 | } | |
289 | write_unlock_bh(&sk->sk_callback_lock); | |
290 | release_sock(sk); | |
291 | ||
292 | /* wait for accepts to stop and close the socket */ | |
293 | flush_workqueue(rds_wq); | |
294 | sock_release(sock); | |
70041088 | 295 | } |