blob: d14445c48304cb3f4b7a946372186d8578a9b530 [file] [log] [blame]
Andy Grover00e0f342009-02-24 15:30:23 +00001/*
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>
34#include <linux/list.h>
35#include <net/inet_hashtables.h>
36
37#include "rds.h"
38#include "loop.h"
39#include "rdma.h"
40
41#define RDS_CONNECTION_HASH_BITS 12
42#define RDS_CONNECTION_HASH_ENTRIES (1 << RDS_CONNECTION_HASH_BITS)
43#define RDS_CONNECTION_HASH_MASK (RDS_CONNECTION_HASH_ENTRIES - 1)
44
45/* converting this to RCU is a chore for another day.. */
46static DEFINE_SPINLOCK(rds_conn_lock);
47static unsigned long rds_conn_count;
48static struct hlist_head rds_conn_hash[RDS_CONNECTION_HASH_ENTRIES];
49static struct kmem_cache *rds_conn_slab;
50
51static struct hlist_head *rds_conn_bucket(__be32 laddr, __be32 faddr)
52{
53 /* Pass NULL, don't need struct net for hash */
54 unsigned long hash = inet_ehashfn(NULL,
55 be32_to_cpu(laddr), 0,
56 be32_to_cpu(faddr), 0);
57 return &rds_conn_hash[hash & RDS_CONNECTION_HASH_MASK];
58}
59
60#define rds_conn_info_set(var, test, suffix) do { \
61 if (test) \
62 var |= RDS_INFO_CONNECTION_FLAG_##suffix; \
63} while (0)
64
65static inline int rds_conn_is_sending(struct rds_connection *conn)
66{
67 int ret = 0;
68
69 if (!mutex_trylock(&conn->c_send_lock))
70 ret = 1;
71 else
72 mutex_unlock(&conn->c_send_lock);
73
74 return ret;
75}
76
77static struct rds_connection *rds_conn_lookup(struct hlist_head *head,
78 __be32 laddr, __be32 faddr,
79 struct rds_transport *trans)
80{
81 struct rds_connection *conn, *ret = NULL;
82 struct hlist_node *pos;
83
84 hlist_for_each_entry(conn, pos, head, c_hash_node) {
85 if (conn->c_faddr == faddr && conn->c_laddr == laddr &&
86 conn->c_trans == trans) {
87 ret = conn;
88 break;
89 }
90 }
91 rdsdebug("returning conn %p for %pI4 -> %pI4\n", ret,
92 &laddr, &faddr);
93 return ret;
94}
95
96/*
97 * This is called by transports as they're bringing down a connection.
98 * It clears partial message state so that the transport can start sending
99 * and receiving over this connection again in the future. It is up to
100 * the transport to have serialized this call with its send and recv.
101 */
102void rds_conn_reset(struct rds_connection *conn)
103{
104 rdsdebug("connection %pI4 to %pI4 reset\n",
105 &conn->c_laddr, &conn->c_faddr);
106
107 rds_stats_inc(s_conn_reset);
108 rds_send_reset(conn);
109 conn->c_flags = 0;
110
111 /* Do not clear next_rx_seq here, else we cannot distinguish
112 * retransmitted packets from new packets, and will hand all
113 * of them to the application. That is not consistent with the
114 * reliability guarantees of RDS. */
115}
116
117/*
118 * There is only every one 'conn' for a given pair of addresses in the
119 * system at a time. They contain messages to be retransmitted and so
120 * span the lifetime of the actual underlying transport connections.
121 *
122 * For now they are not garbage collected once they're created. They
123 * are torn down as the module is removed, if ever.
124 */
125static struct rds_connection *__rds_conn_create(__be32 laddr, __be32 faddr,
126 struct rds_transport *trans, gfp_t gfp,
127 int is_outgoing)
128{
129 struct rds_connection *conn, *tmp, *parent = NULL;
130 struct hlist_head *head = rds_conn_bucket(laddr, faddr);
131 unsigned long flags;
132 int ret;
133
134 spin_lock_irqsave(&rds_conn_lock, flags);
135 conn = rds_conn_lookup(head, laddr, faddr, trans);
136 if (conn
137 && conn->c_loopback
138 && conn->c_trans != &rds_loop_transport
139 && !is_outgoing) {
140 /* This is a looped back IB connection, and we're
141 * called by the code handling the incoming connect.
142 * We need a second connection object into which we
143 * can stick the other QP. */
144 parent = conn;
145 conn = parent->c_passive;
146 }
147 spin_unlock_irqrestore(&rds_conn_lock, flags);
148 if (conn)
149 goto out;
150
Wei Yongjun05a178e2009-04-09 14:09:44 +0000151 conn = kmem_cache_zalloc(rds_conn_slab, gfp);
Andy Grover00e0f342009-02-24 15:30:23 +0000152 if (conn == NULL) {
153 conn = ERR_PTR(-ENOMEM);
154 goto out;
155 }
156
Andy Grover00e0f342009-02-24 15:30:23 +0000157 INIT_HLIST_NODE(&conn->c_hash_node);
158 conn->c_version = RDS_PROTOCOL_3_0;
159 conn->c_laddr = laddr;
160 conn->c_faddr = faddr;
161 spin_lock_init(&conn->c_lock);
162 conn->c_next_tx_seq = 1;
163
164 mutex_init(&conn->c_send_lock);
165 INIT_LIST_HEAD(&conn->c_send_queue);
166 INIT_LIST_HEAD(&conn->c_retrans);
167
168 ret = rds_cong_get_maps(conn);
169 if (ret) {
170 kmem_cache_free(rds_conn_slab, conn);
171 conn = ERR_PTR(ret);
172 goto out;
173 }
174
175 /*
176 * This is where a connection becomes loopback. If *any* RDS sockets
177 * can bind to the destination address then we'd rather the messages
178 * flow through loopback rather than either transport.
179 */
180 if (rds_trans_get_preferred(faddr)) {
181 conn->c_loopback = 1;
182 if (is_outgoing && trans->t_prefer_loopback) {
183 /* "outgoing" connection - and the transport
184 * says it wants the connection handled by the
185 * loopback transport. This is what TCP does.
186 */
187 trans = &rds_loop_transport;
188 }
189 }
190
191 conn->c_trans = trans;
192
193 ret = trans->conn_alloc(conn, gfp);
194 if (ret) {
195 kmem_cache_free(rds_conn_slab, conn);
196 conn = ERR_PTR(ret);
197 goto out;
198 }
199
200 atomic_set(&conn->c_state, RDS_CONN_DOWN);
201 conn->c_reconnect_jiffies = 0;
202 INIT_DELAYED_WORK(&conn->c_send_w, rds_send_worker);
203 INIT_DELAYED_WORK(&conn->c_recv_w, rds_recv_worker);
204 INIT_DELAYED_WORK(&conn->c_conn_w, rds_connect_worker);
205 INIT_WORK(&conn->c_down_w, rds_shutdown_worker);
206 mutex_init(&conn->c_cm_lock);
207 conn->c_flags = 0;
208
209 rdsdebug("allocated conn %p for %pI4 -> %pI4 over %s %s\n",
210 conn, &laddr, &faddr,
211 trans->t_name ? trans->t_name : "[unknown]",
212 is_outgoing ? "(outgoing)" : "");
213
214 spin_lock_irqsave(&rds_conn_lock, flags);
215 if (parent == NULL) {
216 tmp = rds_conn_lookup(head, laddr, faddr, trans);
217 if (tmp == NULL)
218 hlist_add_head(&conn->c_hash_node, head);
219 } else {
220 tmp = parent->c_passive;
221 if (!tmp)
222 parent->c_passive = conn;
223 }
224
225 if (tmp) {
226 trans->conn_free(conn->c_transport_data);
227 kmem_cache_free(rds_conn_slab, conn);
228 conn = tmp;
229 } else {
230 rds_cong_add_conn(conn);
231 rds_conn_count++;
232 }
233
234 spin_unlock_irqrestore(&rds_conn_lock, flags);
235
236out:
237 return conn;
238}
239
240struct rds_connection *rds_conn_create(__be32 laddr, __be32 faddr,
241 struct rds_transport *trans, gfp_t gfp)
242{
243 return __rds_conn_create(laddr, faddr, trans, gfp, 0);
244}
245
246struct rds_connection *rds_conn_create_outgoing(__be32 laddr, __be32 faddr,
247 struct rds_transport *trans, gfp_t gfp)
248{
249 return __rds_conn_create(laddr, faddr, trans, gfp, 1);
250}
251
252void rds_conn_destroy(struct rds_connection *conn)
253{
254 struct rds_message *rm, *rtmp;
255
256 rdsdebug("freeing conn %p for %pI4 -> "
257 "%pI4\n", conn, &conn->c_laddr,
258 &conn->c_faddr);
259
260 hlist_del_init(&conn->c_hash_node);
261
262 /* wait for the rds thread to shut it down */
263 atomic_set(&conn->c_state, RDS_CONN_ERROR);
264 cancel_delayed_work(&conn->c_conn_w);
265 queue_work(rds_wq, &conn->c_down_w);
266 flush_workqueue(rds_wq);
267
268 /* tear down queued messages */
269 list_for_each_entry_safe(rm, rtmp,
270 &conn->c_send_queue,
271 m_conn_item) {
272 list_del_init(&rm->m_conn_item);
273 BUG_ON(!list_empty(&rm->m_sock_item));
274 rds_message_put(rm);
275 }
276 if (conn->c_xmit_rm)
277 rds_message_put(conn->c_xmit_rm);
278
279 conn->c_trans->conn_free(conn->c_transport_data);
280
281 /*
282 * The congestion maps aren't freed up here. They're
283 * freed by rds_cong_exit() after all the connections
284 * have been freed.
285 */
286 rds_cong_remove_conn(conn);
287
288 BUG_ON(!list_empty(&conn->c_retrans));
289 kmem_cache_free(rds_conn_slab, conn);
290
291 rds_conn_count--;
292}
293
294static void rds_conn_message_info(struct socket *sock, unsigned int len,
295 struct rds_info_iterator *iter,
296 struct rds_info_lengths *lens,
297 int want_send)
298{
299 struct hlist_head *head;
300 struct hlist_node *pos;
301 struct list_head *list;
302 struct rds_connection *conn;
303 struct rds_message *rm;
304 unsigned long flags;
305 unsigned int total = 0;
306 size_t i;
307
308 len /= sizeof(struct rds_info_message);
309
310 spin_lock_irqsave(&rds_conn_lock, flags);
311
312 for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash);
313 i++, head++) {
314 hlist_for_each_entry(conn, pos, head, c_hash_node) {
315 if (want_send)
316 list = &conn->c_send_queue;
317 else
318 list = &conn->c_retrans;
319
320 spin_lock(&conn->c_lock);
321
322 /* XXX too lazy to maintain counts.. */
323 list_for_each_entry(rm, list, m_conn_item) {
324 total++;
325 if (total <= len)
326 rds_inc_info_copy(&rm->m_inc, iter,
327 conn->c_laddr,
328 conn->c_faddr, 0);
329 }
330
331 spin_unlock(&conn->c_lock);
332 }
333 }
334
335 spin_unlock_irqrestore(&rds_conn_lock, flags);
336
337 lens->nr = total;
338 lens->each = sizeof(struct rds_info_message);
339}
340
341static void rds_conn_message_info_send(struct socket *sock, unsigned int len,
342 struct rds_info_iterator *iter,
343 struct rds_info_lengths *lens)
344{
345 rds_conn_message_info(sock, len, iter, lens, 1);
346}
347
348static void rds_conn_message_info_retrans(struct socket *sock,
349 unsigned int len,
350 struct rds_info_iterator *iter,
351 struct rds_info_lengths *lens)
352{
353 rds_conn_message_info(sock, len, iter, lens, 0);
354}
355
356void rds_for_each_conn_info(struct socket *sock, unsigned int len,
357 struct rds_info_iterator *iter,
358 struct rds_info_lengths *lens,
359 int (*visitor)(struct rds_connection *, void *),
360 size_t item_len)
361{
362 uint64_t buffer[(item_len + 7) / 8];
363 struct hlist_head *head;
364 struct hlist_node *pos;
365 struct hlist_node *tmp;
366 struct rds_connection *conn;
367 unsigned long flags;
368 size_t i;
369
370 spin_lock_irqsave(&rds_conn_lock, flags);
371
372 lens->nr = 0;
373 lens->each = item_len;
374
375 for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash);
376 i++, head++) {
377 hlist_for_each_entry_safe(conn, pos, tmp, head, c_hash_node) {
378
379 /* XXX no c_lock usage.. */
380 if (!visitor(conn, buffer))
381 continue;
382
383 /* We copy as much as we can fit in the buffer,
384 * but we count all items so that the caller
385 * can resize the buffer. */
386 if (len >= item_len) {
387 rds_info_copy(iter, buffer, item_len);
388 len -= item_len;
389 }
390 lens->nr++;
391 }
392 }
393
394 spin_unlock_irqrestore(&rds_conn_lock, flags);
395}
396
397static int rds_conn_info_visitor(struct rds_connection *conn,
398 void *buffer)
399{
400 struct rds_info_connection *cinfo = buffer;
401
402 cinfo->next_tx_seq = conn->c_next_tx_seq;
403 cinfo->next_rx_seq = conn->c_next_rx_seq;
404 cinfo->laddr = conn->c_laddr;
405 cinfo->faddr = conn->c_faddr;
406 strncpy(cinfo->transport, conn->c_trans->t_name,
407 sizeof(cinfo->transport));
408 cinfo->flags = 0;
409
410 rds_conn_info_set(cinfo->flags,
411 rds_conn_is_sending(conn), SENDING);
412 /* XXX Future: return the state rather than these funky bits */
413 rds_conn_info_set(cinfo->flags,
414 atomic_read(&conn->c_state) == RDS_CONN_CONNECTING,
415 CONNECTING);
416 rds_conn_info_set(cinfo->flags,
417 atomic_read(&conn->c_state) == RDS_CONN_UP,
418 CONNECTED);
419 return 1;
420}
421
422static void rds_conn_info(struct socket *sock, unsigned int len,
423 struct rds_info_iterator *iter,
424 struct rds_info_lengths *lens)
425{
426 rds_for_each_conn_info(sock, len, iter, lens,
427 rds_conn_info_visitor,
428 sizeof(struct rds_info_connection));
429}
430
431int __init rds_conn_init(void)
432{
433 rds_conn_slab = kmem_cache_create("rds_connection",
434 sizeof(struct rds_connection),
435 0, 0, NULL);
436 if (rds_conn_slab == NULL)
437 return -ENOMEM;
438
439 rds_info_register_func(RDS_INFO_CONNECTIONS, rds_conn_info);
440 rds_info_register_func(RDS_INFO_SEND_MESSAGES,
441 rds_conn_message_info_send);
442 rds_info_register_func(RDS_INFO_RETRANS_MESSAGES,
443 rds_conn_message_info_retrans);
444
445 return 0;
446}
447
448void rds_conn_exit(void)
449{
450 rds_loop_exit();
451
452 WARN_ON(!hlist_empty(rds_conn_hash));
453
454 kmem_cache_destroy(rds_conn_slab);
455
456 rds_info_deregister_func(RDS_INFO_CONNECTIONS, rds_conn_info);
457 rds_info_deregister_func(RDS_INFO_SEND_MESSAGES,
458 rds_conn_message_info_send);
459 rds_info_deregister_func(RDS_INFO_RETRANS_MESSAGES,
460 rds_conn_message_info_retrans);
461}
462
463/*
464 * Force a disconnect
465 */
466void rds_conn_drop(struct rds_connection *conn)
467{
468 atomic_set(&conn->c_state, RDS_CONN_ERROR);
469 queue_work(rds_wq, &conn->c_down_w);
470}
471
472/*
473 * An error occurred on the connection
474 */
475void
476__rds_conn_error(struct rds_connection *conn, const char *fmt, ...)
477{
478 va_list ap;
479
480 va_start(ap, fmt);
481 vprintk(fmt, ap);
482 va_end(ap);
483
484 rds_conn_drop(conn);
485}