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; | |
1c5113cf SV |
127 | cp->cp_reconnect_jiffies = 0; |
128 | INIT_DELAYED_WORK(&cp->cp_send_w, rds_send_worker); | |
129 | INIT_DELAYED_WORK(&cp->cp_recv_w, rds_recv_worker); | |
130 | INIT_DELAYED_WORK(&cp->cp_conn_w, rds_connect_worker); | |
131 | INIT_WORK(&cp->cp_down_w, rds_shutdown_worker); | |
132 | mutex_init(&cp->cp_cm_lock); | |
133 | cp->cp_flags = 0; | |
134 | } | |
135 | ||
00e0f34c AG |
136 | /* |
137 | * There is only every one 'conn' for a given pair of addresses in the | |
138 | * system at a time. They contain messages to be retransmitted and so | |
139 | * span the lifetime of the actual underlying transport connections. | |
140 | * | |
141 | * For now they are not garbage collected once they're created. They | |
142 | * are torn down as the module is removed, if ever. | |
143 | */ | |
d5a8ac28 SV |
144 | static struct rds_connection *__rds_conn_create(struct net *net, |
145 | __be32 laddr, __be32 faddr, | |
00e0f34c AG |
146 | struct rds_transport *trans, gfp_t gfp, |
147 | int is_outgoing) | |
148 | { | |
cb24405e | 149 | struct rds_connection *conn, *parent = NULL; |
00e0f34c | 150 | struct hlist_head *head = rds_conn_bucket(laddr, faddr); |
5adb5bc6 | 151 | struct rds_transport *loop_trans; |
00e0f34c | 152 | unsigned long flags; |
5916e2c1 | 153 | int ret, i; |
00e0f34c | 154 | |
bcf50ef2 | 155 | rcu_read_lock(); |
8f384c01 | 156 | conn = rds_conn_lookup(net, head, laddr, faddr, trans); |
f64f9e71 | 157 | if (conn && conn->c_loopback && conn->c_trans != &rds_loop_transport && |
1789b2c0 | 158 | laddr == faddr && !is_outgoing) { |
00e0f34c AG |
159 | /* This is a looped back IB connection, and we're |
160 | * called by the code handling the incoming connect. | |
161 | * We need a second connection object into which we | |
162 | * can stick the other QP. */ | |
163 | parent = conn; | |
164 | conn = parent->c_passive; | |
165 | } | |
bcf50ef2 | 166 | rcu_read_unlock(); |
00e0f34c AG |
167 | if (conn) |
168 | goto out; | |
169 | ||
05a178ec | 170 | conn = kmem_cache_zalloc(rds_conn_slab, gfp); |
8690bfa1 | 171 | if (!conn) { |
00e0f34c AG |
172 | conn = ERR_PTR(-ENOMEM); |
173 | goto out; | |
174 | } | |
175 | ||
00e0f34c | 176 | INIT_HLIST_NODE(&conn->c_hash_node); |
00e0f34c AG |
177 | conn->c_laddr = laddr; |
178 | conn->c_faddr = faddr; | |
00e0f34c | 179 | |
1c5113cf | 180 | rds_conn_net_set(conn, net); |
00e0f34c AG |
181 | |
182 | ret = rds_cong_get_maps(conn); | |
183 | if (ret) { | |
184 | kmem_cache_free(rds_conn_slab, conn); | |
185 | conn = ERR_PTR(ret); | |
186 | goto out; | |
187 | } | |
188 | ||
189 | /* | |
190 | * This is where a connection becomes loopback. If *any* RDS sockets | |
191 | * can bind to the destination address then we'd rather the messages | |
192 | * flow through loopback rather than either transport. | |
193 | */ | |
d5a8ac28 | 194 | loop_trans = rds_trans_get_preferred(net, faddr); |
5adb5bc6 ZB |
195 | if (loop_trans) { |
196 | rds_trans_put(loop_trans); | |
00e0f34c AG |
197 | conn->c_loopback = 1; |
198 | if (is_outgoing && trans->t_prefer_loopback) { | |
199 | /* "outgoing" connection - and the transport | |
200 | * says it wants the connection handled by the | |
201 | * loopback transport. This is what TCP does. | |
202 | */ | |
203 | trans = &rds_loop_transport; | |
204 | } | |
205 | } | |
206 | ||
207 | conn->c_trans = trans; | |
208 | ||
5916e2c1 SV |
209 | init_waitqueue_head(&conn->c_hs_waitq); |
210 | for (i = 0; i < RDS_MPATH_WORKERS; i++) { | |
211 | __rds_conn_path_init(conn, &conn->c_path[i], | |
212 | is_outgoing); | |
213 | conn->c_path[i].cp_index = i; | |
214 | } | |
00e0f34c AG |
215 | ret = trans->conn_alloc(conn, gfp); |
216 | if (ret) { | |
217 | kmem_cache_free(rds_conn_slab, conn); | |
218 | conn = ERR_PTR(ret); | |
219 | goto out; | |
220 | } | |
221 | ||
00e0f34c AG |
222 | rdsdebug("allocated conn %p for %pI4 -> %pI4 over %s %s\n", |
223 | conn, &laddr, &faddr, | |
224 | trans->t_name ? trans->t_name : "[unknown]", | |
225 | is_outgoing ? "(outgoing)" : ""); | |
226 | ||
cb24405e AG |
227 | /* |
228 | * Since we ran without holding the conn lock, someone could | |
229 | * have created the same conn (either normal or passive) in the | |
230 | * interim. We check while holding the lock. If we won, we complete | |
231 | * init and return our conn. If we lost, we rollback and return the | |
232 | * other one. | |
233 | */ | |
00e0f34c | 234 | spin_lock_irqsave(&rds_conn_lock, flags); |
cb24405e AG |
235 | if (parent) { |
236 | /* Creating passive conn */ | |
237 | if (parent->c_passive) { | |
1c5113cf | 238 | trans->conn_free(conn->c_path[0].cp_transport_data); |
cb24405e AG |
239 | kmem_cache_free(rds_conn_slab, conn); |
240 | conn = parent->c_passive; | |
241 | } else { | |
00e0f34c | 242 | parent->c_passive = conn; |
cb24405e AG |
243 | rds_cong_add_conn(conn); |
244 | rds_conn_count++; | |
245 | } | |
00e0f34c | 246 | } else { |
cb24405e AG |
247 | /* Creating normal conn */ |
248 | struct rds_connection *found; | |
249 | ||
3b20fc38 | 250 | found = rds_conn_lookup(net, head, laddr, faddr, trans); |
cb24405e | 251 | if (found) { |
1c5113cf SV |
252 | struct rds_conn_path *cp; |
253 | int i; | |
254 | ||
255 | for (i = 0; i < RDS_MPATH_WORKERS; i++) { | |
256 | cp = &conn->c_path[i]; | |
02105b2c SV |
257 | /* The ->conn_alloc invocation may have |
258 | * allocated resource for all paths, so all | |
259 | * of them may have to be freed here. | |
260 | */ | |
261 | if (cp->cp_transport_data) | |
262 | trans->conn_free(cp->cp_transport_data); | |
1c5113cf | 263 | } |
cb24405e AG |
264 | kmem_cache_free(rds_conn_slab, conn); |
265 | conn = found; | |
266 | } else { | |
905dd418 SV |
267 | conn->c_my_gen_num = rds_gen_num; |
268 | conn->c_peer_gen_num = 0; | |
3b20fc38 | 269 | hlist_add_head_rcu(&conn->c_hash_node, head); |
cb24405e AG |
270 | rds_cong_add_conn(conn); |
271 | rds_conn_count++; | |
272 | } | |
00e0f34c | 273 | } |
00e0f34c AG |
274 | spin_unlock_irqrestore(&rds_conn_lock, flags); |
275 | ||
276 | out: | |
277 | return conn; | |
278 | } | |
279 | ||
d5a8ac28 SV |
280 | struct rds_connection *rds_conn_create(struct net *net, |
281 | __be32 laddr, __be32 faddr, | |
00e0f34c AG |
282 | struct rds_transport *trans, gfp_t gfp) |
283 | { | |
d5a8ac28 | 284 | return __rds_conn_create(net, laddr, faddr, trans, gfp, 0); |
00e0f34c | 285 | } |
616b757a | 286 | EXPORT_SYMBOL_GPL(rds_conn_create); |
00e0f34c | 287 | |
d5a8ac28 SV |
288 | struct rds_connection *rds_conn_create_outgoing(struct net *net, |
289 | __be32 laddr, __be32 faddr, | |
00e0f34c AG |
290 | struct rds_transport *trans, gfp_t gfp) |
291 | { | |
d5a8ac28 | 292 | return __rds_conn_create(net, laddr, faddr, trans, gfp, 1); |
00e0f34c | 293 | } |
616b757a | 294 | EXPORT_SYMBOL_GPL(rds_conn_create_outgoing); |
00e0f34c | 295 | |
d769ef81 | 296 | void rds_conn_shutdown(struct rds_conn_path *cp) |
2dc39357 | 297 | { |
d769ef81 SV |
298 | struct rds_connection *conn = cp->cp_conn; |
299 | ||
2dc39357 | 300 | /* shut it down unless it's down already */ |
d769ef81 | 301 | if (!rds_conn_path_transition(cp, RDS_CONN_DOWN, RDS_CONN_DOWN)) { |
2dc39357 AG |
302 | /* |
303 | * Quiesce the connection mgmt handlers before we start tearing | |
304 | * things down. We don't hold the mutex for the entire | |
305 | * duration of the shutdown operation, else we may be | |
306 | * deadlocking with the CM handler. Instead, the CM event | |
307 | * handler is supposed to check for state DISCONNECTING | |
308 | */ | |
d769ef81 SV |
309 | mutex_lock(&cp->cp_cm_lock); |
310 | if (!rds_conn_path_transition(cp, RDS_CONN_UP, | |
311 | RDS_CONN_DISCONNECTING) && | |
312 | !rds_conn_path_transition(cp, RDS_CONN_ERROR, | |
313 | RDS_CONN_DISCONNECTING)) { | |
314 | rds_conn_path_error(cp, | |
315 | "shutdown called in state %d\n", | |
316 | atomic_read(&cp->cp_state)); | |
317 | mutex_unlock(&cp->cp_cm_lock); | |
2dc39357 AG |
318 | return; |
319 | } | |
d769ef81 | 320 | mutex_unlock(&cp->cp_cm_lock); |
2dc39357 | 321 | |
d769ef81 SV |
322 | wait_event(cp->cp_waitq, |
323 | !test_bit(RDS_IN_XMIT, &cp->cp_flags)); | |
324 | wait_event(cp->cp_waitq, | |
325 | !test_bit(RDS_RECV_REFILL, &cp->cp_flags)); | |
7e3f2952 | 326 | |
226f7a7d | 327 | conn->c_trans->conn_path_shutdown(cp); |
d769ef81 | 328 | rds_conn_path_reset(cp); |
2dc39357 | 329 | |
d769ef81 | 330 | if (!rds_conn_path_transition(cp, RDS_CONN_DISCONNECTING, |
e97656d0 SV |
331 | RDS_CONN_DOWN) && |
332 | !rds_conn_path_transition(cp, RDS_CONN_ERROR, | |
d769ef81 | 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. | |
e97656d0 | 336 | * Quite reproducible with loopback connections. |
2dc39357 | 337 | * Mostly harmless. |
e97656d0 SV |
338 | * |
339 | * Note that this also happens with rds-tcp because | |
340 | * we could have triggered rds_conn_path_drop in irq | |
341 | * mode from rds_tcp_state change on the receipt of | |
342 | * a FIN, thus we need to recheck for RDS_CONN_ERROR | |
343 | * here. | |
2dc39357 | 344 | */ |
d769ef81 SV |
345 | rds_conn_path_error(cp, "%s: failed to transition " |
346 | "to state DOWN, current state " | |
347 | "is %d\n", __func__, | |
348 | atomic_read(&cp->cp_state)); | |
2dc39357 AG |
349 | return; |
350 | } | |
351 | } | |
352 | ||
353 | /* Then reconnect if it's still live. | |
354 | * The passive side of an IB loopback connection is never added | |
355 | * to the conn hash, so we never trigger a reconnect on this | |
356 | * conn - the reconnect is always triggered by the active peer. */ | |
d769ef81 | 357 | cancel_delayed_work_sync(&cp->cp_conn_w); |
bcf50ef2 CM |
358 | rcu_read_lock(); |
359 | if (!hlist_unhashed(&conn->c_hash_node)) { | |
360 | rcu_read_unlock(); | |
8315011a | 361 | rds_queue_reconnect(cp); |
bcf50ef2 CM |
362 | } else { |
363 | rcu_read_unlock(); | |
364 | } | |
2dc39357 AG |
365 | } |
366 | ||
3ecc5693 SV |
367 | /* destroy a single rds_conn_path. rds_conn_destroy() iterates over |
368 | * all paths using rds_conn_path_destroy() | |
369 | */ | |
370 | static void rds_conn_path_destroy(struct rds_conn_path *cp) | |
371 | { | |
372 | struct rds_message *rm, *rtmp; | |
373 | ||
02105b2c SV |
374 | if (!cp->cp_transport_data) |
375 | return; | |
376 | ||
3ecc5693 SV |
377 | rds_conn_path_drop(cp); |
378 | flush_work(&cp->cp_down_w); | |
379 | ||
380 | /* make sure lingering queued work won't try to ref the conn */ | |
381 | cancel_delayed_work_sync(&cp->cp_send_w); | |
382 | cancel_delayed_work_sync(&cp->cp_recv_w); | |
383 | ||
384 | /* tear down queued messages */ | |
385 | list_for_each_entry_safe(rm, rtmp, | |
386 | &cp->cp_send_queue, | |
387 | m_conn_item) { | |
388 | list_del_init(&rm->m_conn_item); | |
389 | BUG_ON(!list_empty(&rm->m_sock_item)); | |
390 | rds_message_put(rm); | |
391 | } | |
392 | if (cp->cp_xmit_rm) | |
393 | rds_message_put(cp->cp_xmit_rm); | |
394 | ||
395 | cp->cp_conn->c_trans->conn_free(cp->cp_transport_data); | |
396 | } | |
397 | ||
2dc39357 AG |
398 | /* |
399 | * Stop and free a connection. | |
ffcec0e1 ZB |
400 | * |
401 | * This can only be used in very limited circumstances. It assumes that once | |
402 | * the conn has been shutdown that no one else is referencing the connection. | |
403 | * We can only ensure this in the rmmod path in the current code. | |
2dc39357 | 404 | */ |
00e0f34c AG |
405 | void rds_conn_destroy(struct rds_connection *conn) |
406 | { | |
fe8ff6b5 | 407 | unsigned long flags; |
02105b2c SV |
408 | int i; |
409 | struct rds_conn_path *cp; | |
00e0f34c AG |
410 | |
411 | rdsdebug("freeing conn %p for %pI4 -> " | |
412 | "%pI4\n", conn, &conn->c_laddr, | |
413 | &conn->c_faddr); | |
414 | ||
c14b0366 | 415 | conn->c_destroy_in_prog = 1; |
abf45439 CM |
416 | /* Ensure conn will not be scheduled for reconnect */ |
417 | spin_lock_irq(&rds_conn_lock); | |
bcf50ef2 | 418 | hlist_del_init_rcu(&conn->c_hash_node); |
abf45439 | 419 | spin_unlock_irq(&rds_conn_lock); |
bcf50ef2 CM |
420 | synchronize_rcu(); |
421 | ||
ffcec0e1 | 422 | /* shut the connection down */ |
02105b2c SV |
423 | for (i = 0; i < RDS_MPATH_WORKERS; i++) { |
424 | cp = &conn->c_path[i]; | |
425 | rds_conn_path_destroy(cp); | |
426 | BUG_ON(!list_empty(&cp->cp_retrans)); | |
00e0f34c | 427 | } |
00e0f34c AG |
428 | |
429 | /* | |
430 | * The congestion maps aren't freed up here. They're | |
431 | * freed by rds_cong_exit() after all the connections | |
432 | * have been freed. | |
433 | */ | |
434 | rds_cong_remove_conn(conn); | |
435 | ||
8edc3aff | 436 | put_net(conn->c_net); |
00e0f34c AG |
437 | kmem_cache_free(rds_conn_slab, conn); |
438 | ||
fe8ff6b5 | 439 | spin_lock_irqsave(&rds_conn_lock, flags); |
00e0f34c | 440 | rds_conn_count--; |
fe8ff6b5 | 441 | spin_unlock_irqrestore(&rds_conn_lock, flags); |
00e0f34c | 442 | } |
616b757a | 443 | EXPORT_SYMBOL_GPL(rds_conn_destroy); |
00e0f34c AG |
444 | |
445 | static void rds_conn_message_info(struct socket *sock, unsigned int len, | |
446 | struct rds_info_iterator *iter, | |
447 | struct rds_info_lengths *lens, | |
448 | int want_send) | |
449 | { | |
450 | struct hlist_head *head; | |
00e0f34c AG |
451 | struct list_head *list; |
452 | struct rds_connection *conn; | |
453 | struct rds_message *rm; | |
00e0f34c | 454 | unsigned int total = 0; |
501dcccd | 455 | unsigned long flags; |
00e0f34c | 456 | size_t i; |
992c9ec5 | 457 | int j; |
00e0f34c AG |
458 | |
459 | len /= sizeof(struct rds_info_message); | |
460 | ||
bcf50ef2 | 461 | rcu_read_lock(); |
00e0f34c AG |
462 | |
463 | for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash); | |
464 | i++, head++) { | |
b67bfe0d | 465 | hlist_for_each_entry_rcu(conn, head, c_hash_node) { |
992c9ec5 SV |
466 | struct rds_conn_path *cp; |
467 | ||
468 | for (j = 0; j < RDS_MPATH_WORKERS; j++) { | |
469 | cp = &conn->c_path[j]; | |
470 | if (want_send) | |
471 | list = &cp->cp_send_queue; | |
472 | else | |
473 | list = &cp->cp_retrans; | |
474 | ||
475 | spin_lock_irqsave(&cp->cp_lock, flags); | |
476 | ||
477 | /* XXX too lazy to maintain counts.. */ | |
478 | list_for_each_entry(rm, list, m_conn_item) { | |
479 | total++; | |
480 | if (total <= len) | |
481 | rds_inc_info_copy(&rm->m_inc, | |
482 | iter, | |
483 | conn->c_laddr, | |
484 | conn->c_faddr, | |
485 | 0); | |
486 | } | |
487 | ||
488 | spin_unlock_irqrestore(&cp->cp_lock, flags); | |
489 | if (!conn->c_trans->t_mp_capable) | |
490 | break; | |
00e0f34c | 491 | } |
00e0f34c AG |
492 | } |
493 | } | |
bcf50ef2 | 494 | rcu_read_unlock(); |
00e0f34c AG |
495 | |
496 | lens->nr = total; | |
497 | lens->each = sizeof(struct rds_info_message); | |
498 | } | |
499 | ||
500 | static void rds_conn_message_info_send(struct socket *sock, unsigned int len, | |
501 | struct rds_info_iterator *iter, | |
502 | struct rds_info_lengths *lens) | |
503 | { | |
504 | rds_conn_message_info(sock, len, iter, lens, 1); | |
505 | } | |
506 | ||
507 | static void rds_conn_message_info_retrans(struct socket *sock, | |
508 | unsigned int len, | |
509 | struct rds_info_iterator *iter, | |
510 | struct rds_info_lengths *lens) | |
511 | { | |
512 | rds_conn_message_info(sock, len, iter, lens, 0); | |
513 | } | |
514 | ||
515 | void rds_for_each_conn_info(struct socket *sock, unsigned int len, | |
516 | struct rds_info_iterator *iter, | |
517 | struct rds_info_lengths *lens, | |
518 | int (*visitor)(struct rds_connection *, void *), | |
519 | size_t item_len) | |
520 | { | |
521 | uint64_t buffer[(item_len + 7) / 8]; | |
522 | struct hlist_head *head; | |
00e0f34c | 523 | struct rds_connection *conn; |
00e0f34c AG |
524 | size_t i; |
525 | ||
bcf50ef2 | 526 | rcu_read_lock(); |
00e0f34c AG |
527 | |
528 | lens->nr = 0; | |
529 | lens->each = item_len; | |
530 | ||
531 | for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash); | |
532 | i++, head++) { | |
b67bfe0d | 533 | hlist_for_each_entry_rcu(conn, head, c_hash_node) { |
00e0f34c AG |
534 | |
535 | /* XXX no c_lock usage.. */ | |
536 | if (!visitor(conn, buffer)) | |
537 | continue; | |
538 | ||
539 | /* We copy as much as we can fit in the buffer, | |
540 | * but we count all items so that the caller | |
541 | * can resize the buffer. */ | |
542 | if (len >= item_len) { | |
543 | rds_info_copy(iter, buffer, item_len); | |
544 | len -= item_len; | |
545 | } | |
546 | lens->nr++; | |
547 | } | |
548 | } | |
bcf50ef2 | 549 | rcu_read_unlock(); |
00e0f34c | 550 | } |
616b757a | 551 | EXPORT_SYMBOL_GPL(rds_for_each_conn_info); |
00e0f34c | 552 | |
bb789763 SS |
553 | static void rds_walk_conn_path_info(struct socket *sock, unsigned int len, |
554 | struct rds_info_iterator *iter, | |
555 | struct rds_info_lengths *lens, | |
556 | int (*visitor)(struct rds_conn_path *, void *), | |
557 | size_t item_len) | |
992c9ec5 SV |
558 | { |
559 | u64 buffer[(item_len + 7) / 8]; | |
560 | struct hlist_head *head; | |
561 | struct rds_connection *conn; | |
562 | size_t i; | |
563 | int j; | |
564 | ||
565 | rcu_read_lock(); | |
566 | ||
567 | lens->nr = 0; | |
568 | lens->each = item_len; | |
569 | ||
570 | for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash); | |
571 | i++, head++) { | |
572 | hlist_for_each_entry_rcu(conn, head, c_hash_node) { | |
573 | struct rds_conn_path *cp; | |
574 | ||
575 | for (j = 0; j < RDS_MPATH_WORKERS; j++) { | |
576 | cp = &conn->c_path[j]; | |
577 | ||
578 | /* XXX no cp_lock usage.. */ | |
579 | if (!visitor(cp, buffer)) | |
580 | continue; | |
581 | if (!conn->c_trans->t_mp_capable) | |
582 | break; | |
583 | } | |
584 | ||
585 | /* We copy as much as we can fit in the buffer, | |
586 | * but we count all items so that the caller | |
587 | * can resize the buffer. | |
588 | */ | |
589 | if (len >= item_len) { | |
590 | rds_info_copy(iter, buffer, item_len); | |
591 | len -= item_len; | |
592 | } | |
593 | lens->nr++; | |
594 | } | |
595 | } | |
596 | rcu_read_unlock(); | |
597 | } | |
598 | ||
599 | static int rds_conn_info_visitor(struct rds_conn_path *cp, void *buffer) | |
00e0f34c AG |
600 | { |
601 | struct rds_info_connection *cinfo = buffer; | |
602 | ||
992c9ec5 SV |
603 | cinfo->next_tx_seq = cp->cp_next_tx_seq; |
604 | cinfo->next_rx_seq = cp->cp_next_rx_seq; | |
605 | cinfo->laddr = cp->cp_conn->c_laddr; | |
606 | cinfo->faddr = cp->cp_conn->c_faddr; | |
607 | strncpy(cinfo->transport, cp->cp_conn->c_trans->t_name, | |
00e0f34c AG |
608 | sizeof(cinfo->transport)); |
609 | cinfo->flags = 0; | |
610 | ||
992c9ec5 | 611 | rds_conn_info_set(cinfo->flags, test_bit(RDS_IN_XMIT, &cp->cp_flags), |
0f4b1c7e | 612 | SENDING); |
00e0f34c AG |
613 | /* XXX Future: return the state rather than these funky bits */ |
614 | rds_conn_info_set(cinfo->flags, | |
992c9ec5 | 615 | atomic_read(&cp->cp_state) == RDS_CONN_CONNECTING, |
00e0f34c AG |
616 | CONNECTING); |
617 | rds_conn_info_set(cinfo->flags, | |
992c9ec5 | 618 | atomic_read(&cp->cp_state) == RDS_CONN_UP, |
00e0f34c AG |
619 | CONNECTED); |
620 | return 1; | |
621 | } | |
622 | ||
623 | static void rds_conn_info(struct socket *sock, unsigned int len, | |
624 | struct rds_info_iterator *iter, | |
625 | struct rds_info_lengths *lens) | |
626 | { | |
992c9ec5 | 627 | rds_walk_conn_path_info(sock, len, iter, lens, |
00e0f34c AG |
628 | rds_conn_info_visitor, |
629 | sizeof(struct rds_info_connection)); | |
630 | } | |
631 | ||
ef87b7ea | 632 | int rds_conn_init(void) |
00e0f34c AG |
633 | { |
634 | rds_conn_slab = kmem_cache_create("rds_connection", | |
635 | sizeof(struct rds_connection), | |
636 | 0, 0, NULL); | |
8690bfa1 | 637 | if (!rds_conn_slab) |
00e0f34c AG |
638 | return -ENOMEM; |
639 | ||
640 | rds_info_register_func(RDS_INFO_CONNECTIONS, rds_conn_info); | |
641 | rds_info_register_func(RDS_INFO_SEND_MESSAGES, | |
642 | rds_conn_message_info_send); | |
643 | rds_info_register_func(RDS_INFO_RETRANS_MESSAGES, | |
644 | rds_conn_message_info_retrans); | |
645 | ||
646 | return 0; | |
647 | } | |
648 | ||
649 | void rds_conn_exit(void) | |
650 | { | |
651 | rds_loop_exit(); | |
652 | ||
653 | WARN_ON(!hlist_empty(rds_conn_hash)); | |
654 | ||
655 | kmem_cache_destroy(rds_conn_slab); | |
656 | ||
657 | rds_info_deregister_func(RDS_INFO_CONNECTIONS, rds_conn_info); | |
658 | rds_info_deregister_func(RDS_INFO_SEND_MESSAGES, | |
659 | rds_conn_message_info_send); | |
660 | rds_info_deregister_func(RDS_INFO_RETRANS_MESSAGES, | |
661 | rds_conn_message_info_retrans); | |
662 | } | |
663 | ||
664 | /* | |
665 | * Force a disconnect | |
666 | */ | |
0cb43965 SV |
667 | void rds_conn_path_drop(struct rds_conn_path *cp) |
668 | { | |
669 | atomic_set(&cp->cp_state, RDS_CONN_ERROR); | |
670 | queue_work(rds_wq, &cp->cp_down_w); | |
671 | } | |
672 | EXPORT_SYMBOL_GPL(rds_conn_path_drop); | |
673 | ||
00e0f34c AG |
674 | void rds_conn_drop(struct rds_connection *conn) |
675 | { | |
5916e2c1 | 676 | WARN_ON(conn->c_trans->t_mp_capable); |
0cb43965 | 677 | rds_conn_path_drop(&conn->c_path[0]); |
00e0f34c | 678 | } |
616b757a | 679 | EXPORT_SYMBOL_GPL(rds_conn_drop); |
00e0f34c | 680 | |
f3c6808d ZB |
681 | /* |
682 | * If the connection is down, trigger a connect. We may have scheduled a | |
683 | * delayed reconnect however - in this case we should not interfere. | |
684 | */ | |
3c0a5900 SV |
685 | void rds_conn_path_connect_if_down(struct rds_conn_path *cp) |
686 | { | |
687 | if (rds_conn_path_state(cp) == RDS_CONN_DOWN && | |
688 | !test_and_set_bit(RDS_RECONNECT_PENDING, &cp->cp_flags)) | |
689 | queue_delayed_work(rds_wq, &cp->cp_conn_w, 0); | |
690 | } | |
1a0e100f | 691 | EXPORT_SYMBOL_GPL(rds_conn_path_connect_if_down); |
3c0a5900 | 692 | |
f3c6808d ZB |
693 | void rds_conn_connect_if_down(struct rds_connection *conn) |
694 | { | |
3c0a5900 SV |
695 | WARN_ON(conn->c_trans->t_mp_capable); |
696 | rds_conn_path_connect_if_down(&conn->c_path[0]); | |
f3c6808d ZB |
697 | } |
698 | EXPORT_SYMBOL_GPL(rds_conn_connect_if_down); | |
699 | ||
fb1b3dc4 SV |
700 | void |
701 | __rds_conn_path_error(struct rds_conn_path *cp, const char *fmt, ...) | |
702 | { | |
703 | va_list ap; | |
704 | ||
705 | va_start(ap, fmt); | |
706 | vprintk(fmt, ap); | |
707 | va_end(ap); | |
708 | ||
709 | rds_conn_path_drop(cp); | |
710 | } |