RDS: Connection handling
[linux-2.6.git] / net / rds / connection.c
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>
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.. */
46 static DEFINE_SPINLOCK(rds_conn_lock);
47 static unsigned long rds_conn_count;
48 static struct hlist_head rds_conn_hash[RDS_CONNECTION_HASH_ENTRIES];
49 static struct kmem_cache *rds_conn_slab;
50
51 static 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
65 static 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
77 static 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  */
102 void 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  */
125 static 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
151         conn = kmem_cache_alloc(rds_conn_slab, gfp);
152         if (conn == NULL) {
153                 conn = ERR_PTR(-ENOMEM);
154                 goto out;
155         }
156
157         memset(conn, 0, sizeof(*conn));
158
159         INIT_HLIST_NODE(&conn->c_hash_node);
160         conn->c_version = RDS_PROTOCOL_3_0;
161         conn->c_laddr = laddr;
162         conn->c_faddr = faddr;
163         spin_lock_init(&conn->c_lock);
164         conn->c_next_tx_seq = 1;
165
166         mutex_init(&conn->c_send_lock);
167         INIT_LIST_HEAD(&conn->c_send_queue);
168         INIT_LIST_HEAD(&conn->c_retrans);
169
170         ret = rds_cong_get_maps(conn);
171         if (ret) {
172                 kmem_cache_free(rds_conn_slab, conn);
173                 conn = ERR_PTR(ret);
174                 goto out;
175         }
176
177         /*
178          * This is where a connection becomes loopback.  If *any* RDS sockets
179          * can bind to the destination address then we'd rather the messages
180          * flow through loopback rather than either transport.
181          */
182         if (rds_trans_get_preferred(faddr)) {
183                 conn->c_loopback = 1;
184                 if (is_outgoing && trans->t_prefer_loopback) {
185                         /* "outgoing" connection - and the transport
186                          * says it wants the connection handled by the
187                          * loopback transport. This is what TCP does.
188                          */
189                         trans = &rds_loop_transport;
190                 }
191         }
192
193         conn->c_trans = trans;
194
195         ret = trans->conn_alloc(conn, gfp);
196         if (ret) {
197                 kmem_cache_free(rds_conn_slab, conn);
198                 conn = ERR_PTR(ret);
199                 goto out;
200         }
201
202         atomic_set(&conn->c_state, RDS_CONN_DOWN);
203         conn->c_reconnect_jiffies = 0;
204         INIT_DELAYED_WORK(&conn->c_send_w, rds_send_worker);
205         INIT_DELAYED_WORK(&conn->c_recv_w, rds_recv_worker);
206         INIT_DELAYED_WORK(&conn->c_conn_w, rds_connect_worker);
207         INIT_WORK(&conn->c_down_w, rds_shutdown_worker);
208         mutex_init(&conn->c_cm_lock);
209         conn->c_flags = 0;
210
211         rdsdebug("allocated conn %p for %pI4 -> %pI4 over %s %s\n",
212           conn, &laddr, &faddr,
213           trans->t_name ? trans->t_name : "[unknown]",
214           is_outgoing ? "(outgoing)" : "");
215
216         spin_lock_irqsave(&rds_conn_lock, flags);
217         if (parent == NULL) {
218                 tmp = rds_conn_lookup(head, laddr, faddr, trans);
219                 if (tmp == NULL)
220                         hlist_add_head(&conn->c_hash_node, head);
221         } else {
222                 tmp = parent->c_passive;
223                 if (!tmp)
224                         parent->c_passive = conn;
225         }
226
227         if (tmp) {
228                 trans->conn_free(conn->c_transport_data);
229                 kmem_cache_free(rds_conn_slab, conn);
230                 conn = tmp;
231         } else {
232                 rds_cong_add_conn(conn);
233                 rds_conn_count++;
234         }
235
236         spin_unlock_irqrestore(&rds_conn_lock, flags);
237
238 out:
239         return conn;
240 }
241
242 struct rds_connection *rds_conn_create(__be32 laddr, __be32 faddr,
243                                        struct rds_transport *trans, gfp_t gfp)
244 {
245         return __rds_conn_create(laddr, faddr, trans, gfp, 0);
246 }
247
248 struct rds_connection *rds_conn_create_outgoing(__be32 laddr, __be32 faddr,
249                                        struct rds_transport *trans, gfp_t gfp)
250 {
251         return __rds_conn_create(laddr, faddr, trans, gfp, 1);
252 }
253
254 void rds_conn_destroy(struct rds_connection *conn)
255 {
256         struct rds_message *rm, *rtmp;
257
258         rdsdebug("freeing conn %p for %pI4 -> "
259                  "%pI4\n", conn, &conn->c_laddr,
260                  &conn->c_faddr);
261
262         hlist_del_init(&conn->c_hash_node);
263
264         /* wait for the rds thread to shut it down */
265         atomic_set(&conn->c_state, RDS_CONN_ERROR);
266         cancel_delayed_work(&conn->c_conn_w);
267         queue_work(rds_wq, &conn->c_down_w);
268         flush_workqueue(rds_wq);
269
270         /* tear down queued messages */
271         list_for_each_entry_safe(rm, rtmp,
272                                  &conn->c_send_queue,
273                                  m_conn_item) {
274                 list_del_init(&rm->m_conn_item);
275                 BUG_ON(!list_empty(&rm->m_sock_item));
276                 rds_message_put(rm);
277         }
278         if (conn->c_xmit_rm)
279                 rds_message_put(conn->c_xmit_rm);
280
281         conn->c_trans->conn_free(conn->c_transport_data);
282
283         /*
284          * The congestion maps aren't freed up here.  They're
285          * freed by rds_cong_exit() after all the connections
286          * have been freed.
287          */
288         rds_cong_remove_conn(conn);
289
290         BUG_ON(!list_empty(&conn->c_retrans));
291         kmem_cache_free(rds_conn_slab, conn);
292
293         rds_conn_count--;
294 }
295
296 static void rds_conn_message_info(struct socket *sock, unsigned int len,
297                                   struct rds_info_iterator *iter,
298                                   struct rds_info_lengths *lens,
299                                   int want_send)
300 {
301         struct hlist_head *head;
302         struct hlist_node *pos;
303         struct list_head *list;
304         struct rds_connection *conn;
305         struct rds_message *rm;
306         unsigned long flags;
307         unsigned int total = 0;
308         size_t i;
309
310         len /= sizeof(struct rds_info_message);
311
312         spin_lock_irqsave(&rds_conn_lock, flags);
313
314         for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash);
315              i++, head++) {
316                 hlist_for_each_entry(conn, pos, head, c_hash_node) {
317                         if (want_send)
318                                 list = &conn->c_send_queue;
319                         else
320                                 list = &conn->c_retrans;
321
322                         spin_lock(&conn->c_lock);
323
324                         /* XXX too lazy to maintain counts.. */
325                         list_for_each_entry(rm, list, m_conn_item) {
326                                 total++;
327                                 if (total <= len)
328                                         rds_inc_info_copy(&rm->m_inc, iter,
329                                                           conn->c_laddr,
330                                                           conn->c_faddr, 0);
331                         }
332
333                         spin_unlock(&conn->c_lock);
334                 }
335         }
336
337         spin_unlock_irqrestore(&rds_conn_lock, flags);
338
339         lens->nr = total;
340         lens->each = sizeof(struct rds_info_message);
341 }
342
343 static void rds_conn_message_info_send(struct socket *sock, unsigned int len,
344                                        struct rds_info_iterator *iter,
345                                        struct rds_info_lengths *lens)
346 {
347         rds_conn_message_info(sock, len, iter, lens, 1);
348 }
349
350 static void rds_conn_message_info_retrans(struct socket *sock,
351                                           unsigned int len,
352                                           struct rds_info_iterator *iter,
353                                           struct rds_info_lengths *lens)
354 {
355         rds_conn_message_info(sock, len, iter, lens, 0);
356 }
357
358 void rds_for_each_conn_info(struct socket *sock, unsigned int len,
359                           struct rds_info_iterator *iter,
360                           struct rds_info_lengths *lens,
361                           int (*visitor)(struct rds_connection *, void *),
362                           size_t item_len)
363 {
364         uint64_t buffer[(item_len + 7) / 8];
365         struct hlist_head *head;
366         struct hlist_node *pos;
367         struct hlist_node *tmp;
368         struct rds_connection *conn;
369         unsigned long flags;
370         size_t i;
371
372         spin_lock_irqsave(&rds_conn_lock, flags);
373
374         lens->nr = 0;
375         lens->each = item_len;
376
377         for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash);
378              i++, head++) {
379                 hlist_for_each_entry_safe(conn, pos, tmp, head, c_hash_node) {
380
381                         /* XXX no c_lock usage.. */
382                         if (!visitor(conn, buffer))
383                                 continue;
384
385                         /* We copy as much as we can fit in the buffer,
386                          * but we count all items so that the caller
387                          * can resize the buffer. */
388                         if (len >= item_len) {
389                                 rds_info_copy(iter, buffer, item_len);
390                                 len -= item_len;
391                         }
392                         lens->nr++;
393                 }
394         }
395
396         spin_unlock_irqrestore(&rds_conn_lock, flags);
397 }
398
399 static int rds_conn_info_visitor(struct rds_connection *conn,
400                                   void *buffer)
401 {
402         struct rds_info_connection *cinfo = buffer;
403
404         cinfo->next_tx_seq = conn->c_next_tx_seq;
405         cinfo->next_rx_seq = conn->c_next_rx_seq;
406         cinfo->laddr = conn->c_laddr;
407         cinfo->faddr = conn->c_faddr;
408         strncpy(cinfo->transport, conn->c_trans->t_name,
409                 sizeof(cinfo->transport));
410         cinfo->flags = 0;
411
412         rds_conn_info_set(cinfo->flags,
413                           rds_conn_is_sending(conn), SENDING);
414         /* XXX Future: return the state rather than these funky bits */
415         rds_conn_info_set(cinfo->flags,
416                           atomic_read(&conn->c_state) == RDS_CONN_CONNECTING,
417                           CONNECTING);
418         rds_conn_info_set(cinfo->flags,
419                           atomic_read(&conn->c_state) == RDS_CONN_UP,
420                           CONNECTED);
421         return 1;
422 }
423
424 static void rds_conn_info(struct socket *sock, unsigned int len,
425                           struct rds_info_iterator *iter,
426                           struct rds_info_lengths *lens)
427 {
428         rds_for_each_conn_info(sock, len, iter, lens,
429                                 rds_conn_info_visitor,
430                                 sizeof(struct rds_info_connection));
431 }
432
433 int __init rds_conn_init(void)
434 {
435         rds_conn_slab = kmem_cache_create("rds_connection",
436                                           sizeof(struct rds_connection),
437                                           0, 0, NULL);
438         if (rds_conn_slab == NULL)
439                 return -ENOMEM;
440
441         rds_info_register_func(RDS_INFO_CONNECTIONS, rds_conn_info);
442         rds_info_register_func(RDS_INFO_SEND_MESSAGES,
443                                rds_conn_message_info_send);
444         rds_info_register_func(RDS_INFO_RETRANS_MESSAGES,
445                                rds_conn_message_info_retrans);
446
447         return 0;
448 }
449
450 void rds_conn_exit(void)
451 {
452         rds_loop_exit();
453
454         WARN_ON(!hlist_empty(rds_conn_hash));
455
456         kmem_cache_destroy(rds_conn_slab);
457
458         rds_info_deregister_func(RDS_INFO_CONNECTIONS, rds_conn_info);
459         rds_info_deregister_func(RDS_INFO_SEND_MESSAGES,
460                                  rds_conn_message_info_send);
461         rds_info_deregister_func(RDS_INFO_RETRANS_MESSAGES,
462                                  rds_conn_message_info_retrans);
463 }
464
465 /*
466  * Force a disconnect
467  */
468 void rds_conn_drop(struct rds_connection *conn)
469 {
470         atomic_set(&conn->c_state, RDS_CONN_ERROR);
471         queue_work(rds_wq, &conn->c_down_w);
472 }
473
474 /*
475  * An error occurred on the connection
476  */
477 void
478 __rds_conn_error(struct rds_connection *conn, const char *fmt, ...)
479 {
480         va_list ap;
481
482         va_start(ap, fmt);
483         vprintk(fmt, ap);
484         va_end(ap);
485
486         rds_conn_drop(conn);
487 }