Commit | Line | Data |
---|---|---|
70041088 | 1 | /* |
eee2fa6a | 2 | * Copyright (c) 2006, 2018 Oracle and/or its affiliates. All rights reserved. |
70041088 AG |
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 | */ | |
bb789763 | 82 | static |
5916e2c1 SV |
83 | struct rds_tcp_connection *rds_tcp_accept_one_path(struct rds_connection *conn) |
84 | { | |
85 | int i; | |
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 | */ | |
eee2fa6a | 91 | if (rds_addr_cmp(&conn->c_faddr, &conn->c_laddr) >= 0) { |
1a0e100f SV |
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 | ||
c14b0366 | 114 | void rds_tcp_set_linger(struct socket *sock) |
10beea7d SV |
115 | { |
116 | struct linger no_linger = { | |
117 | .l_onoff = 1, | |
118 | .l_linger = 0, | |
119 | }; | |
120 | ||
121 | kernel_setsockopt(sock, SOL_SOCKET, SO_LINGER, | |
122 | (char *)&no_linger, sizeof(no_linger)); | |
123 | } | |
124 | ||
467fa153 | 125 | int rds_tcp_accept_one(struct socket *sock) |
70041088 AG |
126 | { |
127 | struct socket *new_sock = NULL; | |
128 | struct rds_connection *conn; | |
129 | int ret; | |
130 | struct inet_sock *inet; | |
bd7c5f98 SV |
131 | struct rds_tcp_connection *rs_tcp = NULL; |
132 | int conn_state; | |
ea3b1ea5 | 133 | struct rds_conn_path *cp; |
1e2b44e7 | 134 | struct in6_addr *my_addr, *peer_addr; |
e65d4d96 KCP |
135 | #if !IS_ENABLED(CONFIG_IPV6) |
136 | struct in6_addr saddr, daddr; | |
137 | #endif | |
138 | int dev_if = 0; | |
70041088 | 139 | |
37e14f4f SV |
140 | if (!sock) /* module unload or netns delete in progress */ |
141 | return -ENETUNREACH; | |
142 | ||
0933a578 | 143 | ret = sock_create_lite(sock->sk->sk_family, |
d5a8ac28 SV |
144 | sock->sk->sk_type, sock->sk->sk_protocol, |
145 | &new_sock); | |
70041088 AG |
146 | if (ret) |
147 | goto out; | |
148 | ||
cdfbabfb | 149 | ret = sock->ops->accept(sock, new_sock, O_NONBLOCK, true); |
70041088 AG |
150 | if (ret < 0) |
151 | goto out; | |
152 | ||
84eef2b2 KCP |
153 | /* sock_create_lite() does not get a hold on the owner module so we |
154 | * need to do it here. Note that sock_release() uses sock->ops to | |
155 | * determine if it needs to decrement the reference count. So set | |
156 | * sock->ops after calling accept() in case that fails. And there's | |
157 | * no need to do try_module_get() as the listener should have a hold | |
158 | * already. | |
159 | */ | |
160 | new_sock->ops = sock->ops; | |
161 | __module_get(new_sock->ops->owner); | |
162 | ||
f711a6ae SV |
163 | ret = rds_tcp_keepalive(new_sock); |
164 | if (ret < 0) | |
165 | goto out; | |
166 | ||
70041088 AG |
167 | rds_tcp_tune(new_sock); |
168 | ||
169 | inet = inet_sk(new_sock->sk); | |
170 | ||
e65d4d96 | 171 | #if IS_ENABLED(CONFIG_IPV6) |
1e2b44e7 KCP |
172 | my_addr = &new_sock->sk->sk_v6_rcv_saddr; |
173 | peer_addr = &new_sock->sk->sk_v6_daddr; | |
e65d4d96 KCP |
174 | #else |
175 | ipv6_addr_set_v4mapped(inet->inet_saddr, &saddr); | |
176 | ipv6_addr_set_v4mapped(inet->inet_daddr, &daddr); | |
177 | my_addr = &saddr; | |
178 | peer_addr = &daddr; | |
179 | #endif | |
180 | rdsdebug("accepted family %d tcp %pI6c:%u -> %pI6c:%u\n", | |
181 | sock->sk->sk_family, | |
1e2b44e7 KCP |
182 | my_addr, ntohs(inet->inet_sport), |
183 | peer_addr, ntohs(inet->inet_dport)); | |
70041088 | 184 | |
e65d4d96 | 185 | #if IS_ENABLED(CONFIG_IPV6) |
1e2b44e7 KCP |
186 | /* sk_bound_dev_if is not set if the peer address is not link local |
187 | * address. In this case, it happens that mcast_oif is set. So | |
188 | * just use it. | |
189 | */ | |
190 | if ((ipv6_addr_type(my_addr) & IPV6_ADDR_LINKLOCAL) && | |
191 | !(ipv6_addr_type(peer_addr) & IPV6_ADDR_LINKLOCAL)) { | |
192 | struct ipv6_pinfo *inet6; | |
193 | ||
194 | inet6 = inet6_sk(new_sock->sk); | |
195 | dev_if = inet6->mcast_oif; | |
196 | } else { | |
197 | dev_if = new_sock->sk->sk_bound_dev_if; | |
198 | } | |
e65d4d96 KCP |
199 | #endif |
200 | ||
d5a8ac28 | 201 | conn = rds_conn_create(sock_net(sock->sk), |
e65d4d96 | 202 | my_addr, peer_addr, |
1e2b44e7 | 203 | &rds_tcp_transport, GFP_KERNEL, dev_if); |
eee2fa6a | 204 | |
70041088 AG |
205 | if (IS_ERR(conn)) { |
206 | ret = PTR_ERR(conn); | |
207 | goto out; | |
208 | } | |
f711a6ae | 209 | /* An incoming SYN request came in, and TCP just accepted it. |
f711a6ae SV |
210 | * |
211 | * If the client reboots, this conn will need to be cleaned up. | |
212 | * rds_tcp_state_change() will do that cleanup | |
213 | */ | |
5916e2c1 SV |
214 | rs_tcp = rds_tcp_accept_one_path(conn); |
215 | if (!rs_tcp) | |
216 | goto rst_nsk; | |
02105b2c | 217 | mutex_lock(&rs_tcp->t_conn_path_lock); |
5916e2c1 SV |
218 | cp = rs_tcp->t_cpath; |
219 | conn_state = rds_conn_path_state(cp); | |
1a0e100f SV |
220 | WARN_ON(conn_state == RDS_CONN_UP); |
221 | if (conn_state != RDS_CONN_CONNECTING && conn_state != RDS_CONN_ERROR) | |
bd7c5f98 | 222 | goto rst_nsk; |
eb192840 | 223 | if (rs_tcp->t_sock) { |
41500c3e SV |
224 | /* Duelling SYN has been handled in rds_tcp_accept_one() */ |
225 | rds_tcp_reset_callbacks(new_sock, cp); | |
226 | /* rds_connect_path_complete() marks RDS_CONN_UP */ | |
227 | rds_connect_path_complete(cp, RDS_CONN_RESETTING); | |
335b48d9 | 228 | } else { |
ea3b1ea5 SV |
229 | rds_tcp_set_callbacks(new_sock, cp); |
230 | rds_connect_path_complete(cp, RDS_CONN_CONNECTING); | |
eb192840 | 231 | } |
70041088 AG |
232 | new_sock = NULL; |
233 | ret = 0; | |
69b92b5b SV |
234 | if (conn->c_npaths == 0) |
235 | rds_send_ping(cp->cp_conn, cp->cp_index); | |
bd7c5f98 SV |
236 | goto out; |
237 | rst_nsk: | |
10beea7d SV |
238 | /* reset the newly returned accept sock and bail. |
239 | * It is safe to set linger on new_sock because the RDS connection | |
240 | * has not been brought up on new_sock, so no RDS-level data could | |
241 | * be pending on it. By setting linger, we achieve the side-effect | |
242 | * of avoiding TIME_WAIT state on new_sock. | |
243 | */ | |
244 | rds_tcp_set_linger(new_sock); | |
335b48d9 | 245 | kernel_sock_shutdown(new_sock, SHUT_RDWR); |
bd7c5f98 | 246 | ret = 0; |
70041088 | 247 | out: |
bd7c5f98 | 248 | if (rs_tcp) |
02105b2c | 249 | mutex_unlock(&rs_tcp->t_conn_path_lock); |
70041088 AG |
250 | if (new_sock) |
251 | sock_release(new_sock); | |
252 | return ret; | |
253 | } | |
254 | ||
676d2369 | 255 | void rds_tcp_listen_data_ready(struct sock *sk) |
70041088 | 256 | { |
676d2369 | 257 | void (*ready)(struct sock *sk); |
70041088 AG |
258 | |
259 | rdsdebug("listen data ready sk %p\n", sk); | |
260 | ||
38036629 | 261 | read_lock_bh(&sk->sk_callback_lock); |
70041088 | 262 | ready = sk->sk_user_data; |
8690bfa1 | 263 | if (!ready) { /* check for teardown race */ |
70041088 AG |
264 | ready = sk->sk_data_ready; |
265 | goto out; | |
266 | } | |
267 | ||
268 | /* | |
269 | * ->sk_data_ready is also called for a newly established child socket | |
270 | * before it has been accepted and the accepter has set up their | |
271 | * data_ready.. we only want to queue listen work for our listening | |
272 | * socket | |
b21dd450 SV |
273 | * |
274 | * (*ready)() may be null if we are racing with netns delete, and | |
275 | * the listen socket is being torn down. | |
70041088 AG |
276 | */ |
277 | if (sk->sk_state == TCP_LISTEN) | |
467fa153 | 278 | rds_tcp_accept_work(sk); |
a93d01f5 SV |
279 | else |
280 | ready = rds_tcp_listen_sock_def_readable(sock_net(sk)); | |
70041088 AG |
281 | |
282 | out: | |
38036629 | 283 | read_unlock_bh(&sk->sk_callback_lock); |
b21dd450 SV |
284 | if (ready) |
285 | ready(sk); | |
70041088 AG |
286 | } |
287 | ||
1e2b44e7 | 288 | struct socket *rds_tcp_listen_init(struct net *net, bool isv6) |
70041088 | 289 | { |
70041088 | 290 | struct socket *sock = NULL; |
1e2b44e7 KCP |
291 | struct sockaddr_storage ss; |
292 | struct sockaddr_in6 *sin6; | |
293 | struct sockaddr_in *sin; | |
294 | int addr_len; | |
70041088 AG |
295 | int ret; |
296 | ||
1e2b44e7 KCP |
297 | ret = sock_create_kern(net, isv6 ? PF_INET6 : PF_INET, SOCK_STREAM, |
298 | IPPROTO_TCP, &sock); | |
299 | if (ret < 0) { | |
300 | rdsdebug("could not create %s listener socket: %d\n", | |
301 | isv6 ? "IPv6" : "IPv4", ret); | |
70041088 | 302 | goto out; |
1e2b44e7 | 303 | } |
70041088 | 304 | |
4a17fd52 | 305 | sock->sk->sk_reuse = SK_CAN_REUSE; |
70041088 AG |
306 | rds_tcp_nonagle(sock); |
307 | ||
308 | write_lock_bh(&sock->sk->sk_callback_lock); | |
309 | sock->sk->sk_user_data = sock->sk->sk_data_ready; | |
310 | sock->sk->sk_data_ready = rds_tcp_listen_data_ready; | |
311 | write_unlock_bh(&sock->sk->sk_callback_lock); | |
312 | ||
1e2b44e7 KCP |
313 | if (isv6) { |
314 | sin6 = (struct sockaddr_in6 *)&ss; | |
315 | sin6->sin6_family = PF_INET6; | |
316 | sin6->sin6_addr = in6addr_any; | |
317 | sin6->sin6_port = (__force u16)htons(RDS_TCP_PORT); | |
318 | sin6->sin6_scope_id = 0; | |
319 | sin6->sin6_flowinfo = 0; | |
320 | addr_len = sizeof(*sin6); | |
321 | } else { | |
322 | sin = (struct sockaddr_in *)&ss; | |
323 | sin->sin_family = PF_INET; | |
324 | sin->sin_addr.s_addr = INADDR_ANY; | |
325 | sin->sin_port = (__force u16)htons(RDS_TCP_PORT); | |
326 | addr_len = sizeof(*sin); | |
327 | } | |
70041088 | 328 | |
1e2b44e7 KCP |
329 | ret = sock->ops->bind(sock, (struct sockaddr *)&ss, addr_len); |
330 | if (ret < 0) { | |
331 | rdsdebug("could not bind %s listener socket: %d\n", | |
332 | isv6 ? "IPv6" : "IPv4", ret); | |
70041088 | 333 | goto out; |
1e2b44e7 | 334 | } |
70041088 AG |
335 | |
336 | ret = sock->ops->listen(sock, 64); | |
337 | if (ret < 0) | |
338 | goto out; | |
339 | ||
467fa153 | 340 | return sock; |
70041088 AG |
341 | out: |
342 | if (sock) | |
343 | sock_release(sock); | |
467fa153 | 344 | return NULL; |
70041088 AG |
345 | } |
346 | ||
b21dd450 | 347 | void rds_tcp_listen_stop(struct socket *sock, struct work_struct *acceptor) |
70041088 | 348 | { |
70041088 AG |
349 | struct sock *sk; |
350 | ||
8690bfa1 | 351 | if (!sock) |
70041088 AG |
352 | return; |
353 | ||
354 | sk = sock->sk; | |
355 | ||
356 | /* serialize with and prevent further callbacks */ | |
357 | lock_sock(sk); | |
358 | write_lock_bh(&sk->sk_callback_lock); | |
359 | if (sk->sk_user_data) { | |
360 | sk->sk_data_ready = sk->sk_user_data; | |
361 | sk->sk_user_data = NULL; | |
362 | } | |
363 | write_unlock_bh(&sk->sk_callback_lock); | |
364 | release_sock(sk); | |
365 | ||
366 | /* wait for accepts to stop and close the socket */ | |
367 | flush_workqueue(rds_wq); | |
b21dd450 | 368 | flush_work(acceptor); |
70041088 | 369 | sock_release(sock); |
70041088 | 370 | } |