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/slab.h> |
70041088 | 35 | #include <linux/in.h> |
3a9a231d | 36 | #include <linux/module.h> |
70041088 | 37 | #include <net/tcp.h> |
467fa153 SV |
38 | #include <net/net_namespace.h> |
39 | #include <net/netns/generic.h> | |
40 | #include <net/tcp.h> | |
70041088 AG |
41 | |
42 | #include "rds.h" | |
43 | #include "tcp.h" | |
44 | ||
45 | /* only for info exporting */ | |
46 | static DEFINE_SPINLOCK(rds_tcp_tc_list_lock); | |
47 | static LIST_HEAD(rds_tcp_tc_list); | |
ff51bf84 | 48 | static unsigned int rds_tcp_tc_count; |
70041088 AG |
49 | |
50 | /* Track rds_tcp_connection structs so they can be cleaned up */ | |
51 | static DEFINE_SPINLOCK(rds_tcp_conn_lock); | |
52 | static LIST_HEAD(rds_tcp_conn_list); | |
53 | ||
54 | static struct kmem_cache *rds_tcp_conn_slab; | |
55 | ||
56 | #define RDS_TCP_DEFAULT_BUFSIZE (128 * 1024) | |
57 | ||
58 | /* doing it this way avoids calling tcp_sk() */ | |
59 | void rds_tcp_nonagle(struct socket *sock) | |
60 | { | |
61 | mm_segment_t oldfs = get_fs(); | |
62 | int val = 1; | |
63 | ||
64 | set_fs(KERNEL_DS); | |
65 | sock->ops->setsockopt(sock, SOL_TCP, TCP_NODELAY, (char __user *)&val, | |
66 | sizeof(val)); | |
67 | set_fs(oldfs); | |
68 | } | |
69 | ||
70 | void rds_tcp_tune(struct socket *sock) | |
71 | { | |
72 | struct sock *sk = sock->sk; | |
73 | ||
74 | rds_tcp_nonagle(sock); | |
75 | ||
76 | /* | |
77 | * We're trying to saturate gigabit with the default, | |
78 | * see svc_sock_setbufsize(). | |
79 | */ | |
80 | lock_sock(sk); | |
81 | sk->sk_sndbuf = RDS_TCP_DEFAULT_BUFSIZE; | |
82 | sk->sk_rcvbuf = RDS_TCP_DEFAULT_BUFSIZE; | |
83 | sk->sk_userlocks |= SOCK_SNDBUF_LOCK|SOCK_RCVBUF_LOCK; | |
84 | release_sock(sk); | |
85 | } | |
86 | ||
87 | u32 rds_tcp_snd_nxt(struct rds_tcp_connection *tc) | |
88 | { | |
89 | return tcp_sk(tc->t_sock->sk)->snd_nxt; | |
90 | } | |
91 | ||
92 | u32 rds_tcp_snd_una(struct rds_tcp_connection *tc) | |
93 | { | |
94 | return tcp_sk(tc->t_sock->sk)->snd_una; | |
95 | } | |
96 | ||
97 | void rds_tcp_restore_callbacks(struct socket *sock, | |
98 | struct rds_tcp_connection *tc) | |
99 | { | |
100 | rdsdebug("restoring sock %p callbacks from tc %p\n", sock, tc); | |
101 | write_lock_bh(&sock->sk->sk_callback_lock); | |
102 | ||
103 | /* done under the callback_lock to serialize with write_space */ | |
104 | spin_lock(&rds_tcp_tc_list_lock); | |
105 | list_del_init(&tc->t_list_item); | |
106 | rds_tcp_tc_count--; | |
107 | spin_unlock(&rds_tcp_tc_list_lock); | |
108 | ||
109 | tc->t_sock = NULL; | |
110 | ||
111 | sock->sk->sk_write_space = tc->t_orig_write_space; | |
112 | sock->sk->sk_data_ready = tc->t_orig_data_ready; | |
113 | sock->sk->sk_state_change = tc->t_orig_state_change; | |
114 | sock->sk->sk_user_data = NULL; | |
115 | ||
116 | write_unlock_bh(&sock->sk->sk_callback_lock); | |
117 | } | |
118 | ||
119 | /* | |
120 | * This is the only path that sets tc->t_sock. Send and receive trust that | |
121 | * it is set. The RDS_CONN_CONNECTED bit protects those paths from being | |
122 | * called while it isn't set. | |
123 | */ | |
124 | void rds_tcp_set_callbacks(struct socket *sock, struct rds_connection *conn) | |
125 | { | |
126 | struct rds_tcp_connection *tc = conn->c_transport_data; | |
127 | ||
128 | rdsdebug("setting sock %p callbacks to tc %p\n", sock, tc); | |
129 | write_lock_bh(&sock->sk->sk_callback_lock); | |
130 | ||
131 | /* done under the callback_lock to serialize with write_space */ | |
132 | spin_lock(&rds_tcp_tc_list_lock); | |
133 | list_add_tail(&tc->t_list_item, &rds_tcp_tc_list); | |
134 | rds_tcp_tc_count++; | |
135 | spin_unlock(&rds_tcp_tc_list_lock); | |
136 | ||
137 | /* accepted sockets need our listen data ready undone */ | |
138 | if (sock->sk->sk_data_ready == rds_tcp_listen_data_ready) | |
139 | sock->sk->sk_data_ready = sock->sk->sk_user_data; | |
140 | ||
141 | tc->t_sock = sock; | |
142 | tc->conn = conn; | |
143 | tc->t_orig_data_ready = sock->sk->sk_data_ready; | |
144 | tc->t_orig_write_space = sock->sk->sk_write_space; | |
145 | tc->t_orig_state_change = sock->sk->sk_state_change; | |
146 | ||
147 | sock->sk->sk_user_data = conn; | |
148 | sock->sk->sk_data_ready = rds_tcp_data_ready; | |
149 | sock->sk->sk_write_space = rds_tcp_write_space; | |
150 | sock->sk->sk_state_change = rds_tcp_state_change; | |
151 | ||
152 | write_unlock_bh(&sock->sk->sk_callback_lock); | |
153 | } | |
154 | ||
155 | static void rds_tcp_tc_info(struct socket *sock, unsigned int len, | |
156 | struct rds_info_iterator *iter, | |
157 | struct rds_info_lengths *lens) | |
158 | { | |
159 | struct rds_info_tcp_socket tsinfo; | |
160 | struct rds_tcp_connection *tc; | |
161 | unsigned long flags; | |
162 | struct sockaddr_in sin; | |
163 | int sinlen; | |
164 | ||
165 | spin_lock_irqsave(&rds_tcp_tc_list_lock, flags); | |
166 | ||
167 | if (len / sizeof(tsinfo) < rds_tcp_tc_count) | |
168 | goto out; | |
169 | ||
170 | list_for_each_entry(tc, &rds_tcp_tc_list, t_list_item) { | |
171 | ||
172 | sock->ops->getname(sock, (struct sockaddr *)&sin, &sinlen, 0); | |
173 | tsinfo.local_addr = sin.sin_addr.s_addr; | |
174 | tsinfo.local_port = sin.sin_port; | |
175 | sock->ops->getname(sock, (struct sockaddr *)&sin, &sinlen, 1); | |
176 | tsinfo.peer_addr = sin.sin_addr.s_addr; | |
177 | tsinfo.peer_port = sin.sin_port; | |
178 | ||
179 | tsinfo.hdr_rem = tc->t_tinc_hdr_rem; | |
180 | tsinfo.data_rem = tc->t_tinc_data_rem; | |
181 | tsinfo.last_sent_nxt = tc->t_last_sent_nxt; | |
182 | tsinfo.last_expected_una = tc->t_last_expected_una; | |
183 | tsinfo.last_seen_una = tc->t_last_seen_una; | |
184 | ||
185 | rds_info_copy(iter, &tsinfo, sizeof(tsinfo)); | |
186 | } | |
187 | ||
188 | out: | |
189 | lens->nr = rds_tcp_tc_count; | |
190 | lens->each = sizeof(tsinfo); | |
191 | ||
192 | spin_unlock_irqrestore(&rds_tcp_tc_list_lock, flags); | |
193 | } | |
194 | ||
d5a8ac28 | 195 | static int rds_tcp_laddr_check(struct net *net, __be32 addr) |
70041088 | 196 | { |
d5a8ac28 | 197 | if (inet_addr_type(net, addr) == RTN_LOCAL) |
70041088 AG |
198 | return 0; |
199 | return -EADDRNOTAVAIL; | |
200 | } | |
201 | ||
202 | static int rds_tcp_conn_alloc(struct rds_connection *conn, gfp_t gfp) | |
203 | { | |
204 | struct rds_tcp_connection *tc; | |
205 | ||
206 | tc = kmem_cache_alloc(rds_tcp_conn_slab, gfp); | |
8690bfa1 | 207 | if (!tc) |
70041088 AG |
208 | return -ENOMEM; |
209 | ||
210 | tc->t_sock = NULL; | |
211 | tc->t_tinc = NULL; | |
212 | tc->t_tinc_hdr_rem = sizeof(struct rds_header); | |
213 | tc->t_tinc_data_rem = 0; | |
214 | ||
215 | conn->c_transport_data = tc; | |
216 | ||
217 | spin_lock_irq(&rds_tcp_conn_lock); | |
218 | list_add_tail(&tc->t_tcp_node, &rds_tcp_conn_list); | |
219 | spin_unlock_irq(&rds_tcp_conn_lock); | |
220 | ||
221 | rdsdebug("alloced tc %p\n", conn->c_transport_data); | |
222 | return 0; | |
223 | } | |
224 | ||
225 | static void rds_tcp_conn_free(void *arg) | |
226 | { | |
227 | struct rds_tcp_connection *tc = arg; | |
8200a59f | 228 | unsigned long flags; |
70041088 | 229 | rdsdebug("freeing tc %p\n", tc); |
8200a59f PE |
230 | |
231 | spin_lock_irqsave(&rds_tcp_conn_lock, flags); | |
232 | list_del(&tc->t_tcp_node); | |
233 | spin_unlock_irqrestore(&rds_tcp_conn_lock, flags); | |
234 | ||
70041088 AG |
235 | kmem_cache_free(rds_tcp_conn_slab, tc); |
236 | } | |
237 | ||
238 | static void rds_tcp_destroy_conns(void) | |
239 | { | |
240 | struct rds_tcp_connection *tc, *_tc; | |
241 | LIST_HEAD(tmp_list); | |
242 | ||
243 | /* avoid calling conn_destroy with irqs off */ | |
244 | spin_lock_irq(&rds_tcp_conn_lock); | |
245 | list_splice(&rds_tcp_conn_list, &tmp_list); | |
246 | INIT_LIST_HEAD(&rds_tcp_conn_list); | |
247 | spin_unlock_irq(&rds_tcp_conn_lock); | |
248 | ||
249 | list_for_each_entry_safe(tc, _tc, &tmp_list, t_tcp_node) { | |
250 | if (tc->conn->c_passive) | |
251 | rds_conn_destroy(tc->conn->c_passive); | |
252 | rds_conn_destroy(tc->conn); | |
253 | } | |
254 | } | |
255 | ||
467fa153 | 256 | static void rds_tcp_exit(void); |
70041088 AG |
257 | |
258 | struct rds_transport rds_tcp_transport = { | |
259 | .laddr_check = rds_tcp_laddr_check, | |
260 | .xmit_prepare = rds_tcp_xmit_prepare, | |
261 | .xmit_complete = rds_tcp_xmit_complete, | |
70041088 AG |
262 | .xmit = rds_tcp_xmit, |
263 | .recv = rds_tcp_recv, | |
264 | .conn_alloc = rds_tcp_conn_alloc, | |
265 | .conn_free = rds_tcp_conn_free, | |
266 | .conn_connect = rds_tcp_conn_connect, | |
267 | .conn_shutdown = rds_tcp_conn_shutdown, | |
268 | .inc_copy_to_user = rds_tcp_inc_copy_to_user, | |
70041088 AG |
269 | .inc_free = rds_tcp_inc_free, |
270 | .stats_info_copy = rds_tcp_stats_info_copy, | |
271 | .exit = rds_tcp_exit, | |
272 | .t_owner = THIS_MODULE, | |
273 | .t_name = "tcp", | |
335776bd | 274 | .t_type = RDS_TRANS_TCP, |
70041088 AG |
275 | .t_prefer_loopback = 1, |
276 | }; | |
277 | ||
467fa153 SV |
278 | static int rds_tcp_netid; |
279 | ||
280 | /* per-network namespace private data for this module */ | |
281 | struct rds_tcp_net { | |
282 | struct socket *rds_tcp_listen_sock; | |
283 | struct work_struct rds_tcp_accept_w; | |
284 | }; | |
285 | ||
286 | static void rds_tcp_accept_worker(struct work_struct *work) | |
287 | { | |
288 | struct rds_tcp_net *rtn = container_of(work, | |
289 | struct rds_tcp_net, | |
290 | rds_tcp_accept_w); | |
291 | ||
292 | while (rds_tcp_accept_one(rtn->rds_tcp_listen_sock) == 0) | |
293 | cond_resched(); | |
294 | } | |
295 | ||
296 | void rds_tcp_accept_work(struct sock *sk) | |
297 | { | |
298 | struct net *net = sock_net(sk); | |
299 | struct rds_tcp_net *rtn = net_generic(net, rds_tcp_netid); | |
300 | ||
301 | queue_work(rds_wq, &rtn->rds_tcp_accept_w); | |
302 | } | |
303 | ||
304 | static __net_init int rds_tcp_init_net(struct net *net) | |
305 | { | |
306 | struct rds_tcp_net *rtn = net_generic(net, rds_tcp_netid); | |
307 | ||
308 | rtn->rds_tcp_listen_sock = rds_tcp_listen_init(net); | |
309 | if (!rtn->rds_tcp_listen_sock) { | |
310 | pr_warn("could not set up listen sock\n"); | |
311 | return -EAFNOSUPPORT; | |
312 | } | |
313 | INIT_WORK(&rtn->rds_tcp_accept_w, rds_tcp_accept_worker); | |
314 | return 0; | |
315 | } | |
316 | ||
317 | static void __net_exit rds_tcp_exit_net(struct net *net) | |
318 | { | |
319 | struct rds_tcp_net *rtn = net_generic(net, rds_tcp_netid); | |
320 | ||
321 | /* If rds_tcp_exit_net() is called as a result of netns deletion, | |
322 | * the rds_tcp_kill_sock() device notifier would already have cleaned | |
323 | * up the listen socket, thus there is no work to do in this function. | |
324 | * | |
325 | * If rds_tcp_exit_net() is called as a result of module unload, | |
326 | * i.e., due to rds_tcp_exit() -> unregister_pernet_subsys(), then | |
327 | * we do need to clean up the listen socket here. | |
328 | */ | |
329 | if (rtn->rds_tcp_listen_sock) { | |
330 | rds_tcp_listen_stop(rtn->rds_tcp_listen_sock); | |
331 | rtn->rds_tcp_listen_sock = NULL; | |
332 | flush_work(&rtn->rds_tcp_accept_w); | |
333 | } | |
334 | } | |
335 | ||
336 | static struct pernet_operations rds_tcp_net_ops = { | |
337 | .init = rds_tcp_init_net, | |
338 | .exit = rds_tcp_exit_net, | |
339 | .id = &rds_tcp_netid, | |
340 | .size = sizeof(struct rds_tcp_net), | |
341 | }; | |
342 | ||
343 | static void rds_tcp_kill_sock(struct net *net) | |
344 | { | |
345 | struct rds_tcp_connection *tc, *_tc; | |
346 | struct sock *sk; | |
347 | LIST_HEAD(tmp_list); | |
348 | struct rds_tcp_net *rtn = net_generic(net, rds_tcp_netid); | |
349 | ||
350 | rds_tcp_listen_stop(rtn->rds_tcp_listen_sock); | |
351 | rtn->rds_tcp_listen_sock = NULL; | |
352 | flush_work(&rtn->rds_tcp_accept_w); | |
353 | spin_lock_irq(&rds_tcp_conn_lock); | |
354 | list_for_each_entry_safe(tc, _tc, &rds_tcp_conn_list, t_tcp_node) { | |
355 | struct net *c_net = read_pnet(&tc->conn->c_net); | |
356 | ||
357 | if (net != c_net || !tc->t_sock) | |
358 | continue; | |
359 | list_move_tail(&tc->t_tcp_node, &tmp_list); | |
360 | } | |
361 | spin_unlock_irq(&rds_tcp_conn_lock); | |
362 | list_for_each_entry_safe(tc, _tc, &tmp_list, t_tcp_node) { | |
363 | sk = tc->t_sock->sk; | |
364 | sk->sk_prot->disconnect(sk, 0); | |
365 | tcp_done(sk); | |
366 | if (tc->conn->c_passive) | |
367 | rds_conn_destroy(tc->conn->c_passive); | |
368 | rds_conn_destroy(tc->conn); | |
369 | } | |
370 | } | |
371 | ||
372 | static int rds_tcp_dev_event(struct notifier_block *this, | |
373 | unsigned long event, void *ptr) | |
374 | { | |
375 | struct net_device *dev = netdev_notifier_info_to_dev(ptr); | |
376 | ||
377 | /* rds-tcp registers as a pernet subys, so the ->exit will only | |
378 | * get invoked after network acitivity has quiesced. We need to | |
379 | * clean up all sockets to quiesce network activity, and use | |
380 | * the unregistration of the per-net loopback device as a trigger | |
381 | * to start that cleanup. | |
382 | */ | |
383 | if (event == NETDEV_UNREGISTER_FINAL && | |
384 | dev->ifindex == LOOPBACK_IFINDEX) | |
385 | rds_tcp_kill_sock(dev_net(dev)); | |
386 | ||
387 | return NOTIFY_DONE; | |
388 | } | |
389 | ||
390 | static struct notifier_block rds_tcp_dev_notifier = { | |
391 | .notifier_call = rds_tcp_dev_event, | |
392 | .priority = -10, /* must be called after other network notifiers */ | |
393 | }; | |
394 | ||
395 | static void rds_tcp_exit(void) | |
396 | { | |
397 | rds_info_deregister_func(RDS_INFO_TCP_SOCKETS, rds_tcp_tc_info); | |
398 | unregister_pernet_subsys(&rds_tcp_net_ops); | |
399 | if (unregister_netdevice_notifier(&rds_tcp_dev_notifier)) | |
400 | pr_warn("could not unregister rds_tcp_dev_notifier\n"); | |
401 | rds_tcp_destroy_conns(); | |
402 | rds_trans_unregister(&rds_tcp_transport); | |
403 | rds_tcp_recv_exit(); | |
404 | kmem_cache_destroy(rds_tcp_conn_slab); | |
405 | } | |
406 | module_exit(rds_tcp_exit); | |
407 | ||
ff51bf84 | 408 | static int rds_tcp_init(void) |
70041088 AG |
409 | { |
410 | int ret; | |
411 | ||
412 | rds_tcp_conn_slab = kmem_cache_create("rds_tcp_connection", | |
413 | sizeof(struct rds_tcp_connection), | |
414 | 0, 0, NULL); | |
8690bfa1 | 415 | if (!rds_tcp_conn_slab) { |
70041088 AG |
416 | ret = -ENOMEM; |
417 | goto out; | |
418 | } | |
419 | ||
467fa153 SV |
420 | ret = register_netdevice_notifier(&rds_tcp_dev_notifier); |
421 | if (ret) { | |
422 | pr_warn("could not register rds_tcp_dev_notifier\n"); | |
423 | goto out; | |
424 | } | |
425 | ||
426 | ret = register_pernet_subsys(&rds_tcp_net_ops); | |
427 | if (ret) | |
428 | goto out_slab; | |
429 | ||
70041088 AG |
430 | ret = rds_tcp_recv_init(); |
431 | if (ret) | |
432 | goto out_slab; | |
433 | ||
434 | ret = rds_trans_register(&rds_tcp_transport); | |
435 | if (ret) | |
436 | goto out_recv; | |
437 | ||
70041088 AG |
438 | rds_info_register_func(RDS_INFO_TCP_SOCKETS, rds_tcp_tc_info); |
439 | ||
440 | goto out; | |
441 | ||
70041088 AG |
442 | out_recv: |
443 | rds_tcp_recv_exit(); | |
444 | out_slab: | |
467fa153 | 445 | unregister_pernet_subsys(&rds_tcp_net_ops); |
70041088 AG |
446 | kmem_cache_destroy(rds_tcp_conn_slab); |
447 | out: | |
448 | return ret; | |
449 | } | |
450 | module_init(rds_tcp_init); | |
451 | ||
452 | MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>"); | |
453 | MODULE_DESCRIPTION("RDS: TCP transport"); | |
454 | MODULE_LICENSE("Dual BSD/GPL"); | |
455 |