]>
Commit | Line | Data |
---|---|---|
00e0f34c 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> | |
34 | #include <linux/list.h> | |
5a0e3ad6 | 35 | #include <linux/slab.h> |
bc3b2d7f | 36 | #include <linux/export.h> |
00e0f34c AG |
37 | #include <net/inet_hashtables.h> |
38 | ||
39 | #include "rds.h" | |
40 | #include "loop.h" | |
00e0f34c AG |
41 | |
42 | #define RDS_CONNECTION_HASH_BITS 12 | |
43 | #define RDS_CONNECTION_HASH_ENTRIES (1 << RDS_CONNECTION_HASH_BITS) | |
44 | #define RDS_CONNECTION_HASH_MASK (RDS_CONNECTION_HASH_ENTRIES - 1) | |
45 | ||
46 | /* converting this to RCU is a chore for another day.. */ | |
47 | static DEFINE_SPINLOCK(rds_conn_lock); | |
48 | static unsigned long rds_conn_count; | |
49 | static struct hlist_head rds_conn_hash[RDS_CONNECTION_HASH_ENTRIES]; | |
50 | static struct kmem_cache *rds_conn_slab; | |
51 | ||
52 | static struct hlist_head *rds_conn_bucket(__be32 laddr, __be32 faddr) | |
53 | { | |
1bbdceef HFS |
54 | static u32 rds_hash_secret __read_mostly; |
55 | ||
56 | unsigned long hash; | |
57 | ||
58 | net_get_random_once(&rds_hash_secret, sizeof(rds_hash_secret)); | |
59 | ||
00e0f34c | 60 | /* Pass NULL, don't need struct net for hash */ |
1bbdceef HFS |
61 | hash = __inet_ehashfn(be32_to_cpu(laddr), 0, |
62 | be32_to_cpu(faddr), 0, | |
63 | rds_hash_secret); | |
00e0f34c AG |
64 | return &rds_conn_hash[hash & RDS_CONNECTION_HASH_MASK]; |
65 | } | |
66 | ||
67 | #define rds_conn_info_set(var, test, suffix) do { \ | |
68 | if (test) \ | |
69 | var |= RDS_INFO_CONNECTION_FLAG_##suffix; \ | |
70 | } while (0) | |
71 | ||
bcf50ef2 | 72 | /* rcu read lock must be held or the connection spinlock */ |
8f384c01 SV |
73 | static struct rds_connection *rds_conn_lookup(struct net *net, |
74 | struct hlist_head *head, | |
00e0f34c AG |
75 | __be32 laddr, __be32 faddr, |
76 | struct rds_transport *trans) | |
77 | { | |
78 | struct rds_connection *conn, *ret = NULL; | |
00e0f34c | 79 | |
b67bfe0d | 80 | hlist_for_each_entry_rcu(conn, head, c_hash_node) { |
00e0f34c | 81 | if (conn->c_faddr == faddr && conn->c_laddr == laddr && |
8f384c01 | 82 | conn->c_trans == trans && net == rds_conn_net(conn)) { |
00e0f34c AG |
83 | ret = conn; |
84 | break; | |
85 | } | |
86 | } | |
87 | rdsdebug("returning conn %p for %pI4 -> %pI4\n", ret, | |
88 | &laddr, &faddr); | |
89 | return ret; | |
90 | } | |
91 | ||
92 | /* | |
93 | * This is called by transports as they're bringing down a connection. | |
94 | * It clears partial message state so that the transport can start sending | |
95 | * and receiving over this connection again in the future. It is up to | |
96 | * the transport to have serialized this call with its send and recv. | |
97 | */ | |
d769ef81 | 98 | static void rds_conn_path_reset(struct rds_conn_path *cp) |
00e0f34c | 99 | { |
d769ef81 SV |
100 | struct rds_connection *conn = cp->cp_conn; |
101 | ||
00e0f34c AG |
102 | rdsdebug("connection %pI4 to %pI4 reset\n", |
103 | &conn->c_laddr, &conn->c_faddr); | |
104 | ||
105 | rds_stats_inc(s_conn_reset); | |
d769ef81 SV |
106 | rds_send_path_reset(cp); |
107 | cp->cp_flags = 0; | |
00e0f34c AG |
108 | |
109 | /* Do not clear next_rx_seq here, else we cannot distinguish | |
110 | * retransmitted packets from new packets, and will hand all | |
111 | * of them to the application. That is not consistent with the | |
112 | * reliability guarantees of RDS. */ | |
113 | } | |
114 | ||
1c5113cf SV |
115 | static void __rds_conn_path_init(struct rds_connection *conn, |
116 | struct rds_conn_path *cp, bool is_outgoing) | |
117 | { | |
118 | spin_lock_init(&cp->cp_lock); | |
119 | cp->cp_next_tx_seq = 1; | |
120 | init_waitqueue_head(&cp->cp_waitq); | |
121 | INIT_LIST_HEAD(&cp->cp_send_queue); | |
122 | INIT_LIST_HEAD(&cp->cp_retrans); | |
123 | ||
124 | cp->cp_conn = conn; | |
125 | atomic_set(&cp->cp_state, RDS_CONN_DOWN); | |
126 | cp->cp_send_gen = 0; | |
127 | /* cp_outgoing is per-path. So we can only set it here | |
128 | * for the single-path transports. | |
129 | */ | |
130 | if (!conn->c_trans->t_mp_capable) | |
131 | cp->cp_outgoing = (is_outgoing ? 1 : 0); | |
132 | cp->cp_reconnect_jiffies = 0; | |
133 | INIT_DELAYED_WORK(&cp->cp_send_w, rds_send_worker); | |
134 | INIT_DELAYED_WORK(&cp->cp_recv_w, rds_recv_worker); | |
135 | INIT_DELAYED_WORK(&cp->cp_conn_w, rds_connect_worker); | |
136 | INIT_WORK(&cp->cp_down_w, rds_shutdown_worker); | |
137 | mutex_init(&cp->cp_cm_lock); | |
138 | cp->cp_flags = 0; | |
139 | } | |
140 | ||
00e0f34c AG |
141 | /* |
142 | * There is only every one 'conn' for a given pair of addresses in the | |
143 | * system at a time. They contain messages to be retransmitted and so | |
144 | * span the lifetime of the actual underlying transport connections. | |
145 | * | |
146 | * For now they are not garbage collected once they're created. They | |
147 | * are torn down as the module is removed, if ever. | |
148 | */ | |
d5a8ac28 SV |
149 | static struct rds_connection *__rds_conn_create(struct net *net, |
150 | __be32 laddr, __be32 faddr, | |
00e0f34c AG |
151 | struct rds_transport *trans, gfp_t gfp, |
152 | int is_outgoing) | |
153 | { | |
cb24405e | 154 | struct rds_connection *conn, *parent = NULL; |
00e0f34c | 155 | struct hlist_head *head = rds_conn_bucket(laddr, faddr); |
5adb5bc6 | 156 | struct rds_transport *loop_trans; |
00e0f34c AG |
157 | unsigned long flags; |
158 | int ret; | |
159 | ||
bcf50ef2 | 160 | rcu_read_lock(); |
8f384c01 | 161 | conn = rds_conn_lookup(net, head, laddr, faddr, trans); |
f64f9e71 | 162 | if (conn && conn->c_loopback && conn->c_trans != &rds_loop_transport && |
1789b2c0 | 163 | laddr == faddr && !is_outgoing) { |
00e0f34c AG |
164 | /* This is a looped back IB connection, and we're |
165 | * called by the code handling the incoming connect. | |
166 | * We need a second connection object into which we | |
167 | * can stick the other QP. */ | |
168 | parent = conn; | |
169 | conn = parent->c_passive; | |
170 | } | |
bcf50ef2 | 171 | rcu_read_unlock(); |
00e0f34c AG |
172 | if (conn) |
173 | goto out; | |
174 | ||
05a178ec | 175 | conn = kmem_cache_zalloc(rds_conn_slab, gfp); |
8690bfa1 | 176 | if (!conn) { |
00e0f34c AG |
177 | conn = ERR_PTR(-ENOMEM); |
178 | goto out; | |
179 | } | |
180 | ||
00e0f34c | 181 | INIT_HLIST_NODE(&conn->c_hash_node); |
00e0f34c AG |
182 | conn->c_laddr = laddr; |
183 | conn->c_faddr = faddr; | |
00e0f34c | 184 | |
1c5113cf | 185 | rds_conn_net_set(conn, net); |
00e0f34c AG |
186 | |
187 | ret = rds_cong_get_maps(conn); | |
188 | if (ret) { | |
189 | kmem_cache_free(rds_conn_slab, conn); | |
190 | conn = ERR_PTR(ret); | |
191 | goto out; | |
192 | } | |
193 | ||
194 | /* | |
195 | * This is where a connection becomes loopback. If *any* RDS sockets | |
196 | * can bind to the destination address then we'd rather the messages | |
197 | * flow through loopback rather than either transport. | |
198 | */ | |
d5a8ac28 | 199 | loop_trans = rds_trans_get_preferred(net, faddr); |
5adb5bc6 ZB |
200 | if (loop_trans) { |
201 | rds_trans_put(loop_trans); | |
00e0f34c AG |
202 | conn->c_loopback = 1; |
203 | if (is_outgoing && trans->t_prefer_loopback) { | |
204 | /* "outgoing" connection - and the transport | |
205 | * says it wants the connection handled by the | |
206 | * loopback transport. This is what TCP does. | |
207 | */ | |
208 | trans = &rds_loop_transport; | |
209 | } | |
210 | } | |
211 | ||
212 | conn->c_trans = trans; | |
213 | ||
214 | ret = trans->conn_alloc(conn, gfp); | |
215 | if (ret) { | |
216 | kmem_cache_free(rds_conn_slab, conn); | |
217 | conn = ERR_PTR(ret); | |
218 | goto out; | |
219 | } | |
220 | ||
00e0f34c AG |
221 | rdsdebug("allocated conn %p for %pI4 -> %pI4 over %s %s\n", |
222 | conn, &laddr, &faddr, | |
223 | trans->t_name ? trans->t_name : "[unknown]", | |
224 | is_outgoing ? "(outgoing)" : ""); | |
225 | ||
cb24405e AG |
226 | /* |
227 | * Since we ran without holding the conn lock, someone could | |
228 | * have created the same conn (either normal or passive) in the | |
229 | * interim. We check while holding the lock. If we won, we complete | |
230 | * init and return our conn. If we lost, we rollback and return the | |
231 | * other one. | |
232 | */ | |
00e0f34c | 233 | spin_lock_irqsave(&rds_conn_lock, flags); |
cb24405e AG |
234 | if (parent) { |
235 | /* Creating passive conn */ | |
236 | if (parent->c_passive) { | |
1c5113cf | 237 | trans->conn_free(conn->c_path[0].cp_transport_data); |
cb24405e AG |
238 | kmem_cache_free(rds_conn_slab, conn); |
239 | conn = parent->c_passive; | |
240 | } else { | |
00e0f34c | 241 | parent->c_passive = conn; |
cb24405e AG |
242 | rds_cong_add_conn(conn); |
243 | rds_conn_count++; | |
244 | } | |
00e0f34c | 245 | } else { |
cb24405e AG |
246 | /* Creating normal conn */ |
247 | struct rds_connection *found; | |
248 | ||
3b20fc38 | 249 | found = rds_conn_lookup(net, head, laddr, faddr, trans); |
cb24405e | 250 | if (found) { |
1c5113cf SV |
251 | struct rds_conn_path *cp; |
252 | int i; | |
253 | ||
254 | for (i = 0; i < RDS_MPATH_WORKERS; i++) { | |
255 | cp = &conn->c_path[i]; | |
256 | trans->conn_free(cp->cp_transport_data); | |
257 | if (!trans->t_mp_capable) | |
258 | break; | |
259 | } | |
cb24405e AG |
260 | kmem_cache_free(rds_conn_slab, conn); |
261 | conn = found; | |
262 | } else { | |
1c5113cf SV |
263 | int i; |
264 | ||
265 | for (i = 0; i < RDS_MPATH_WORKERS; i++) { | |
266 | __rds_conn_path_init(conn, &conn->c_path[i], | |
267 | is_outgoing); | |
268 | conn->c_path[i].cp_index = i; | |
269 | } | |
270 | ||
3b20fc38 | 271 | hlist_add_head_rcu(&conn->c_hash_node, head); |
cb24405e AG |
272 | rds_cong_add_conn(conn); |
273 | rds_conn_count++; | |
274 | } | |
00e0f34c | 275 | } |
00e0f34c AG |
276 | spin_unlock_irqrestore(&rds_conn_lock, flags); |
277 | ||
278 | out: | |
279 | return conn; | |
280 | } | |
281 | ||
d5a8ac28 SV |
282 | struct rds_connection *rds_conn_create(struct net *net, |
283 | __be32 laddr, __be32 faddr, | |
00e0f34c AG |
284 | struct rds_transport *trans, gfp_t gfp) |
285 | { | |
d5a8ac28 | 286 | return __rds_conn_create(net, laddr, faddr, trans, gfp, 0); |
00e0f34c | 287 | } |
616b757a | 288 | EXPORT_SYMBOL_GPL(rds_conn_create); |
00e0f34c | 289 | |
d5a8ac28 SV |
290 | struct rds_connection *rds_conn_create_outgoing(struct net *net, |
291 | __be32 laddr, __be32 faddr, | |
00e0f34c AG |
292 | struct rds_transport *trans, gfp_t gfp) |
293 | { | |
d5a8ac28 | 294 | return __rds_conn_create(net, laddr, faddr, trans, gfp, 1); |
00e0f34c | 295 | } |
616b757a | 296 | EXPORT_SYMBOL_GPL(rds_conn_create_outgoing); |
00e0f34c | 297 | |
d769ef81 | 298 | void rds_conn_shutdown(struct rds_conn_path *cp) |
2dc39357 | 299 | { |
d769ef81 SV |
300 | struct rds_connection *conn = cp->cp_conn; |
301 | ||
2dc39357 | 302 | /* shut it down unless it's down already */ |
d769ef81 | 303 | if (!rds_conn_path_transition(cp, RDS_CONN_DOWN, RDS_CONN_DOWN)) { |
2dc39357 AG |
304 | /* |
305 | * Quiesce the connection mgmt handlers before we start tearing | |
306 | * things down. We don't hold the mutex for the entire | |
307 | * duration of the shutdown operation, else we may be | |
308 | * deadlocking with the CM handler. Instead, the CM event | |
309 | * handler is supposed to check for state DISCONNECTING | |
310 | */ | |
d769ef81 SV |
311 | mutex_lock(&cp->cp_cm_lock); |
312 | if (!rds_conn_path_transition(cp, RDS_CONN_UP, | |
313 | RDS_CONN_DISCONNECTING) && | |
314 | !rds_conn_path_transition(cp, RDS_CONN_ERROR, | |
315 | RDS_CONN_DISCONNECTING)) { | |
316 | rds_conn_path_error(cp, | |
317 | "shutdown called in state %d\n", | |
318 | atomic_read(&cp->cp_state)); | |
319 | mutex_unlock(&cp->cp_cm_lock); | |
2dc39357 AG |
320 | return; |
321 | } | |
d769ef81 | 322 | mutex_unlock(&cp->cp_cm_lock); |
2dc39357 | 323 | |
d769ef81 SV |
324 | wait_event(cp->cp_waitq, |
325 | !test_bit(RDS_IN_XMIT, &cp->cp_flags)); | |
326 | wait_event(cp->cp_waitq, | |
327 | !test_bit(RDS_RECV_REFILL, &cp->cp_flags)); | |
7e3f2952 | 328 | |
226f7a7d | 329 | conn->c_trans->conn_path_shutdown(cp); |
d769ef81 | 330 | rds_conn_path_reset(cp); |
2dc39357 | 331 | |
d769ef81 SV |
332 | if (!rds_conn_path_transition(cp, RDS_CONN_DISCONNECTING, |
333 | RDS_CONN_DOWN)) { | |
2dc39357 AG |
334 | /* This can happen - eg when we're in the middle of tearing |
335 | * down the connection, and someone unloads the rds module. | |
336 | * Quite reproduceable with loopback connections. | |
337 | * Mostly harmless. | |
338 | */ | |
d769ef81 SV |
339 | rds_conn_path_error(cp, "%s: failed to transition " |
340 | "to state DOWN, current state " | |
341 | "is %d\n", __func__, | |
342 | atomic_read(&cp->cp_state)); | |
2dc39357 AG |
343 | return; |
344 | } | |
345 | } | |
346 | ||
347 | /* Then reconnect if it's still live. | |
348 | * The passive side of an IB loopback connection is never added | |
349 | * to the conn hash, so we never trigger a reconnect on this | |
350 | * conn - the reconnect is always triggered by the active peer. */ | |
d769ef81 | 351 | cancel_delayed_work_sync(&cp->cp_conn_w); |
bcf50ef2 CM |
352 | rcu_read_lock(); |
353 | if (!hlist_unhashed(&conn->c_hash_node)) { | |
354 | rcu_read_unlock(); | |
3b20fc38 | 355 | if (conn->c_trans->t_type != RDS_TRANS_TCP || |
d769ef81 SV |
356 | cp->cp_outgoing == 1) |
357 | rds_queue_reconnect(cp); | |
bcf50ef2 CM |
358 | } else { |
359 | rcu_read_unlock(); | |
360 | } | |
2dc39357 AG |
361 | } |
362 | ||
3ecc5693 SV |
363 | /* destroy a single rds_conn_path. rds_conn_destroy() iterates over |
364 | * all paths using rds_conn_path_destroy() | |
365 | */ | |
366 | static void rds_conn_path_destroy(struct rds_conn_path *cp) | |
367 | { | |
368 | struct rds_message *rm, *rtmp; | |
369 | ||
370 | rds_conn_path_drop(cp); | |
371 | flush_work(&cp->cp_down_w); | |
372 | ||
373 | /* make sure lingering queued work won't try to ref the conn */ | |
374 | cancel_delayed_work_sync(&cp->cp_send_w); | |
375 | cancel_delayed_work_sync(&cp->cp_recv_w); | |
376 | ||
377 | /* tear down queued messages */ | |
378 | list_for_each_entry_safe(rm, rtmp, | |
379 | &cp->cp_send_queue, | |
380 | m_conn_item) { | |
381 | list_del_init(&rm->m_conn_item); | |
382 | BUG_ON(!list_empty(&rm->m_sock_item)); | |
383 | rds_message_put(rm); | |
384 | } | |
385 | if (cp->cp_xmit_rm) | |
386 | rds_message_put(cp->cp_xmit_rm); | |
387 | ||
388 | cp->cp_conn->c_trans->conn_free(cp->cp_transport_data); | |
389 | } | |
390 | ||
2dc39357 AG |
391 | /* |
392 | * Stop and free a connection. | |
ffcec0e1 ZB |
393 | * |
394 | * This can only be used in very limited circumstances. It assumes that once | |
395 | * the conn has been shutdown that no one else is referencing the connection. | |
396 | * We can only ensure this in the rmmod path in the current code. | |
2dc39357 | 397 | */ |
00e0f34c AG |
398 | void rds_conn_destroy(struct rds_connection *conn) |
399 | { | |
fe8ff6b5 | 400 | unsigned long flags; |
00e0f34c AG |
401 | |
402 | rdsdebug("freeing conn %p for %pI4 -> " | |
403 | "%pI4\n", conn, &conn->c_laddr, | |
404 | &conn->c_faddr); | |
405 | ||
abf45439 CM |
406 | /* Ensure conn will not be scheduled for reconnect */ |
407 | spin_lock_irq(&rds_conn_lock); | |
bcf50ef2 | 408 | hlist_del_init_rcu(&conn->c_hash_node); |
abf45439 | 409 | spin_unlock_irq(&rds_conn_lock); |
bcf50ef2 CM |
410 | synchronize_rcu(); |
411 | ||
ffcec0e1 | 412 | /* shut the connection down */ |
3ecc5693 SV |
413 | if (!conn->c_trans->t_mp_capable) { |
414 | rds_conn_path_destroy(&conn->c_path[0]); | |
415 | BUG_ON(!list_empty(&conn->c_path[0].cp_retrans)); | |
416 | } else { | |
417 | int i; | |
418 | struct rds_conn_path *cp; | |
4518071a | 419 | |
3ecc5693 SV |
420 | for (i = 0; i < RDS_MPATH_WORKERS; i++) { |
421 | cp = &conn->c_path[i]; | |
422 | rds_conn_path_destroy(cp); | |
423 | BUG_ON(!list_empty(&cp->cp_retrans)); | |
424 | } | |
00e0f34c | 425 | } |
00e0f34c AG |
426 | |
427 | /* | |
428 | * The congestion maps aren't freed up here. They're | |
429 | * freed by rds_cong_exit() after all the connections | |
430 | * have been freed. | |
431 | */ | |
432 | rds_cong_remove_conn(conn); | |
433 | ||
00e0f34c AG |
434 | kmem_cache_free(rds_conn_slab, conn); |
435 | ||
fe8ff6b5 | 436 | spin_lock_irqsave(&rds_conn_lock, flags); |
00e0f34c | 437 | rds_conn_count--; |
fe8ff6b5 | 438 | spin_unlock_irqrestore(&rds_conn_lock, flags); |
00e0f34c | 439 | } |
616b757a | 440 | EXPORT_SYMBOL_GPL(rds_conn_destroy); |
00e0f34c AG |
441 | |
442 | static void rds_conn_message_info(struct socket *sock, unsigned int len, | |
443 | struct rds_info_iterator *iter, | |
444 | struct rds_info_lengths *lens, | |
445 | int want_send) | |
446 | { | |
447 | struct hlist_head *head; | |
00e0f34c AG |
448 | struct list_head *list; |
449 | struct rds_connection *conn; | |
450 | struct rds_message *rm; | |
00e0f34c | 451 | unsigned int total = 0; |
501dcccd | 452 | unsigned long flags; |
00e0f34c | 453 | size_t i; |
992c9ec5 | 454 | int j; |
00e0f34c AG |
455 | |
456 | len /= sizeof(struct rds_info_message); | |
457 | ||
bcf50ef2 | 458 | rcu_read_lock(); |
00e0f34c AG |
459 | |
460 | for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash); | |
461 | i++, head++) { | |
b67bfe0d | 462 | hlist_for_each_entry_rcu(conn, head, c_hash_node) { |
992c9ec5 SV |
463 | struct rds_conn_path *cp; |
464 | ||
465 | for (j = 0; j < RDS_MPATH_WORKERS; j++) { | |
466 | cp = &conn->c_path[j]; | |
467 | if (want_send) | |
468 | list = &cp->cp_send_queue; | |
469 | else | |
470 | list = &cp->cp_retrans; | |
471 | ||
472 | spin_lock_irqsave(&cp->cp_lock, flags); | |
473 | ||
474 | /* XXX too lazy to maintain counts.. */ | |
475 | list_for_each_entry(rm, list, m_conn_item) { | |
476 | total++; | |
477 | if (total <= len) | |
478 | rds_inc_info_copy(&rm->m_inc, | |
479 | iter, | |
480 | conn->c_laddr, | |
481 | conn->c_faddr, | |
482 | 0); | |
483 | } | |
484 | ||
485 | spin_unlock_irqrestore(&cp->cp_lock, flags); | |
486 | if (!conn->c_trans->t_mp_capable) | |
487 | break; | |
00e0f34c | 488 | } |
00e0f34c AG |
489 | } |
490 | } | |
bcf50ef2 | 491 | rcu_read_unlock(); |
00e0f34c AG |
492 | |
493 | lens->nr = total; | |
494 | lens->each = sizeof(struct rds_info_message); | |
495 | } | |
496 | ||
497 | static void rds_conn_message_info_send(struct socket *sock, unsigned int len, | |
498 | struct rds_info_iterator *iter, | |
499 | struct rds_info_lengths *lens) | |
500 | { | |
501 | rds_conn_message_info(sock, len, iter, lens, 1); | |
502 | } | |
503 | ||
504 | static void rds_conn_message_info_retrans(struct socket *sock, | |
505 | unsigned int len, | |
506 | struct rds_info_iterator *iter, | |
507 | struct rds_info_lengths *lens) | |
508 | { | |
509 | rds_conn_message_info(sock, len, iter, lens, 0); | |
510 | } | |
511 | ||
512 | void rds_for_each_conn_info(struct socket *sock, unsigned int len, | |
513 | struct rds_info_iterator *iter, | |
514 | struct rds_info_lengths *lens, | |
515 | int (*visitor)(struct rds_connection *, void *), | |
516 | size_t item_len) | |
517 | { | |
518 | uint64_t buffer[(item_len + 7) / 8]; | |
519 | struct hlist_head *head; | |
00e0f34c | 520 | struct rds_connection *conn; |
00e0f34c AG |
521 | size_t i; |
522 | ||
bcf50ef2 | 523 | rcu_read_lock(); |
00e0f34c AG |
524 | |
525 | lens->nr = 0; | |
526 | lens->each = item_len; | |
527 | ||
528 | for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash); | |
529 | i++, head++) { | |
b67bfe0d | 530 | hlist_for_each_entry_rcu(conn, head, c_hash_node) { |
00e0f34c AG |
531 | |
532 | /* XXX no c_lock usage.. */ | |
533 | if (!visitor(conn, buffer)) | |
534 | continue; | |
535 | ||
536 | /* We copy as much as we can fit in the buffer, | |
537 | * but we count all items so that the caller | |
538 | * can resize the buffer. */ | |
539 | if (len >= item_len) { | |
540 | rds_info_copy(iter, buffer, item_len); | |
541 | len -= item_len; | |
542 | } | |
543 | lens->nr++; | |
544 | } | |
545 | } | |
bcf50ef2 | 546 | rcu_read_unlock(); |
00e0f34c | 547 | } |
616b757a | 548 | EXPORT_SYMBOL_GPL(rds_for_each_conn_info); |
00e0f34c | 549 | |
992c9ec5 SV |
550 | void rds_walk_conn_path_info(struct socket *sock, unsigned int len, |
551 | struct rds_info_iterator *iter, | |
552 | struct rds_info_lengths *lens, | |
553 | int (*visitor)(struct rds_conn_path *, void *), | |
554 | size_t item_len) | |
555 | { | |
556 | u64 buffer[(item_len + 7) / 8]; | |
557 | struct hlist_head *head; | |
558 | struct rds_connection *conn; | |
559 | size_t i; | |
560 | int j; | |
561 | ||
562 | rcu_read_lock(); | |
563 | ||
564 | lens->nr = 0; | |
565 | lens->each = item_len; | |
566 | ||
567 | for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash); | |
568 | i++, head++) { | |
569 | hlist_for_each_entry_rcu(conn, head, c_hash_node) { | |
570 | struct rds_conn_path *cp; | |
571 | ||
572 | for (j = 0; j < RDS_MPATH_WORKERS; j++) { | |
573 | cp = &conn->c_path[j]; | |
574 | ||
575 | /* XXX no cp_lock usage.. */ | |
576 | if (!visitor(cp, buffer)) | |
577 | continue; | |
578 | if (!conn->c_trans->t_mp_capable) | |
579 | break; | |
580 | } | |
581 | ||
582 | /* We copy as much as we can fit in the buffer, | |
583 | * but we count all items so that the caller | |
584 | * can resize the buffer. | |
585 | */ | |
586 | if (len >= item_len) { | |
587 | rds_info_copy(iter, buffer, item_len); | |
588 | len -= item_len; | |
589 | } | |
590 | lens->nr++; | |
591 | } | |
592 | } | |
593 | rcu_read_unlock(); | |
594 | } | |
595 | ||
596 | static int rds_conn_info_visitor(struct rds_conn_path *cp, void *buffer) | |
00e0f34c AG |
597 | { |
598 | struct rds_info_connection *cinfo = buffer; | |
599 | ||
992c9ec5 SV |
600 | cinfo->next_tx_seq = cp->cp_next_tx_seq; |
601 | cinfo->next_rx_seq = cp->cp_next_rx_seq; | |
602 | cinfo->laddr = cp->cp_conn->c_laddr; | |
603 | cinfo->faddr = cp->cp_conn->c_faddr; | |
604 | strncpy(cinfo->transport, cp->cp_conn->c_trans->t_name, | |
00e0f34c AG |
605 | sizeof(cinfo->transport)); |
606 | cinfo->flags = 0; | |
607 | ||
992c9ec5 | 608 | rds_conn_info_set(cinfo->flags, test_bit(RDS_IN_XMIT, &cp->cp_flags), |
0f4b1c7e | 609 | SENDING); |
00e0f34c AG |
610 | /* XXX Future: return the state rather than these funky bits */ |
611 | rds_conn_info_set(cinfo->flags, | |
992c9ec5 | 612 | atomic_read(&cp->cp_state) == RDS_CONN_CONNECTING, |
00e0f34c AG |
613 | CONNECTING); |
614 | rds_conn_info_set(cinfo->flags, | |
992c9ec5 | 615 | atomic_read(&cp->cp_state) == RDS_CONN_UP, |
00e0f34c AG |
616 | CONNECTED); |
617 | return 1; | |
618 | } | |
619 | ||
620 | static void rds_conn_info(struct socket *sock, unsigned int len, | |
621 | struct rds_info_iterator *iter, | |
622 | struct rds_info_lengths *lens) | |
623 | { | |
992c9ec5 | 624 | rds_walk_conn_path_info(sock, len, iter, lens, |
00e0f34c AG |
625 | rds_conn_info_visitor, |
626 | sizeof(struct rds_info_connection)); | |
627 | } | |
628 | ||
ef87b7ea | 629 | int rds_conn_init(void) |
00e0f34c AG |
630 | { |
631 | rds_conn_slab = kmem_cache_create("rds_connection", | |
632 | sizeof(struct rds_connection), | |
633 | 0, 0, NULL); | |
8690bfa1 | 634 | if (!rds_conn_slab) |
00e0f34c AG |
635 | return -ENOMEM; |
636 | ||
637 | rds_info_register_func(RDS_INFO_CONNECTIONS, rds_conn_info); | |
638 | rds_info_register_func(RDS_INFO_SEND_MESSAGES, | |
639 | rds_conn_message_info_send); | |
640 | rds_info_register_func(RDS_INFO_RETRANS_MESSAGES, | |
641 | rds_conn_message_info_retrans); | |
642 | ||
643 | return 0; | |
644 | } | |
645 | ||
646 | void rds_conn_exit(void) | |
647 | { | |
648 | rds_loop_exit(); | |
649 | ||
650 | WARN_ON(!hlist_empty(rds_conn_hash)); | |
651 | ||
652 | kmem_cache_destroy(rds_conn_slab); | |
653 | ||
654 | rds_info_deregister_func(RDS_INFO_CONNECTIONS, rds_conn_info); | |
655 | rds_info_deregister_func(RDS_INFO_SEND_MESSAGES, | |
656 | rds_conn_message_info_send); | |
657 | rds_info_deregister_func(RDS_INFO_RETRANS_MESSAGES, | |
658 | rds_conn_message_info_retrans); | |
659 | } | |
660 | ||
661 | /* | |
662 | * Force a disconnect | |
663 | */ | |
0cb43965 SV |
664 | void rds_conn_path_drop(struct rds_conn_path *cp) |
665 | { | |
666 | atomic_set(&cp->cp_state, RDS_CONN_ERROR); | |
667 | queue_work(rds_wq, &cp->cp_down_w); | |
668 | } | |
669 | EXPORT_SYMBOL_GPL(rds_conn_path_drop); | |
670 | ||
00e0f34c AG |
671 | void rds_conn_drop(struct rds_connection *conn) |
672 | { | |
0cb43965 | 673 | rds_conn_path_drop(&conn->c_path[0]); |
00e0f34c | 674 | } |
616b757a | 675 | EXPORT_SYMBOL_GPL(rds_conn_drop); |
00e0f34c | 676 | |
f3c6808d ZB |
677 | /* |
678 | * If the connection is down, trigger a connect. We may have scheduled a | |
679 | * delayed reconnect however - in this case we should not interfere. | |
680 | */ | |
3c0a5900 SV |
681 | void rds_conn_path_connect_if_down(struct rds_conn_path *cp) |
682 | { | |
683 | if (rds_conn_path_state(cp) == RDS_CONN_DOWN && | |
684 | !test_and_set_bit(RDS_RECONNECT_PENDING, &cp->cp_flags)) | |
685 | queue_delayed_work(rds_wq, &cp->cp_conn_w, 0); | |
686 | } | |
687 | ||
f3c6808d ZB |
688 | void rds_conn_connect_if_down(struct rds_connection *conn) |
689 | { | |
3c0a5900 SV |
690 | WARN_ON(conn->c_trans->t_mp_capable); |
691 | rds_conn_path_connect_if_down(&conn->c_path[0]); | |
f3c6808d ZB |
692 | } |
693 | EXPORT_SYMBOL_GPL(rds_conn_connect_if_down); | |
694 | ||
00e0f34c AG |
695 | /* |
696 | * An error occurred on the connection | |
697 | */ | |
698 | void | |
699 | __rds_conn_error(struct rds_connection *conn, const char *fmt, ...) | |
700 | { | |
701 | va_list ap; | |
702 | ||
703 | va_start(ap, fmt); | |
704 | vprintk(fmt, ap); | |
705 | va_end(ap); | |
706 | ||
707 | rds_conn_drop(conn); | |
708 | } | |
fb1b3dc4 SV |
709 | |
710 | void | |
711 | __rds_conn_path_error(struct rds_conn_path *cp, const char *fmt, ...) | |
712 | { | |
713 | va_list ap; | |
714 | ||
715 | va_start(ap, fmt); | |
716 | vprintk(fmt, ap); | |
717 | va_end(ap); | |
718 | ||
719 | rds_conn_path_drop(cp); | |
720 | } |