]>
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; |
840df162 | 154 | int npaths = (trans->t_mp_capable ? RDS_MPATH_WORKERS : 1); |
00e0f34c | 155 | |
bcf50ef2 | 156 | rcu_read_lock(); |
8f384c01 | 157 | conn = rds_conn_lookup(net, head, laddr, faddr, trans); |
f64f9e71 | 158 | if (conn && conn->c_loopback && conn->c_trans != &rds_loop_transport && |
1789b2c0 | 159 | laddr == faddr && !is_outgoing) { |
00e0f34c AG |
160 | /* This is a looped back IB connection, and we're |
161 | * called by the code handling the incoming connect. | |
162 | * We need a second connection object into which we | |
163 | * can stick the other QP. */ | |
164 | parent = conn; | |
165 | conn = parent->c_passive; | |
166 | } | |
bcf50ef2 | 167 | rcu_read_unlock(); |
00e0f34c AG |
168 | if (conn) |
169 | goto out; | |
170 | ||
05a178ec | 171 | conn = kmem_cache_zalloc(rds_conn_slab, gfp); |
8690bfa1 | 172 | if (!conn) { |
00e0f34c AG |
173 | conn = ERR_PTR(-ENOMEM); |
174 | goto out; | |
175 | } | |
840df162 SV |
176 | conn->c_path = kcalloc(npaths, sizeof(struct rds_conn_path), gfp); |
177 | if (!conn->c_path) { | |
178 | kmem_cache_free(rds_conn_slab, conn); | |
179 | conn = ERR_PTR(-ENOMEM); | |
180 | goto out; | |
181 | } | |
00e0f34c | 182 | |
00e0f34c | 183 | INIT_HLIST_NODE(&conn->c_hash_node); |
00e0f34c AG |
184 | conn->c_laddr = laddr; |
185 | conn->c_faddr = faddr; | |
00e0f34c | 186 | |
1c5113cf | 187 | rds_conn_net_set(conn, net); |
00e0f34c AG |
188 | |
189 | ret = rds_cong_get_maps(conn); | |
190 | if (ret) { | |
840df162 | 191 | kfree(conn->c_path); |
00e0f34c AG |
192 | kmem_cache_free(rds_conn_slab, conn); |
193 | conn = ERR_PTR(ret); | |
194 | goto out; | |
195 | } | |
196 | ||
197 | /* | |
198 | * This is where a connection becomes loopback. If *any* RDS sockets | |
199 | * can bind to the destination address then we'd rather the messages | |
200 | * flow through loopback rather than either transport. | |
201 | */ | |
d5a8ac28 | 202 | loop_trans = rds_trans_get_preferred(net, faddr); |
5adb5bc6 ZB |
203 | if (loop_trans) { |
204 | rds_trans_put(loop_trans); | |
00e0f34c AG |
205 | conn->c_loopback = 1; |
206 | if (is_outgoing && trans->t_prefer_loopback) { | |
207 | /* "outgoing" connection - and the transport | |
208 | * says it wants the connection handled by the | |
209 | * loopback transport. This is what TCP does. | |
210 | */ | |
211 | trans = &rds_loop_transport; | |
212 | } | |
213 | } | |
214 | ||
215 | conn->c_trans = trans; | |
216 | ||
5916e2c1 | 217 | init_waitqueue_head(&conn->c_hs_waitq); |
840df162 | 218 | for (i = 0; i < npaths; i++) { |
5916e2c1 SV |
219 | __rds_conn_path_init(conn, &conn->c_path[i], |
220 | is_outgoing); | |
221 | conn->c_path[i].cp_index = i; | |
222 | } | |
00e0f34c AG |
223 | ret = trans->conn_alloc(conn, gfp); |
224 | if (ret) { | |
840df162 | 225 | kfree(conn->c_path); |
00e0f34c AG |
226 | kmem_cache_free(rds_conn_slab, conn); |
227 | conn = ERR_PTR(ret); | |
228 | goto out; | |
229 | } | |
230 | ||
00e0f34c AG |
231 | rdsdebug("allocated conn %p for %pI4 -> %pI4 over %s %s\n", |
232 | conn, &laddr, &faddr, | |
233 | trans->t_name ? trans->t_name : "[unknown]", | |
234 | is_outgoing ? "(outgoing)" : ""); | |
235 | ||
cb24405e AG |
236 | /* |
237 | * Since we ran without holding the conn lock, someone could | |
238 | * have created the same conn (either normal or passive) in the | |
239 | * interim. We check while holding the lock. If we won, we complete | |
240 | * init and return our conn. If we lost, we rollback and return the | |
241 | * other one. | |
242 | */ | |
00e0f34c | 243 | spin_lock_irqsave(&rds_conn_lock, flags); |
cb24405e AG |
244 | if (parent) { |
245 | /* Creating passive conn */ | |
246 | if (parent->c_passive) { | |
1c5113cf | 247 | trans->conn_free(conn->c_path[0].cp_transport_data); |
840df162 | 248 | kfree(conn->c_path); |
cb24405e AG |
249 | kmem_cache_free(rds_conn_slab, conn); |
250 | conn = parent->c_passive; | |
251 | } else { | |
00e0f34c | 252 | parent->c_passive = conn; |
cb24405e AG |
253 | rds_cong_add_conn(conn); |
254 | rds_conn_count++; | |
255 | } | |
00e0f34c | 256 | } else { |
cb24405e AG |
257 | /* Creating normal conn */ |
258 | struct rds_connection *found; | |
259 | ||
3b20fc38 | 260 | found = rds_conn_lookup(net, head, laddr, faddr, trans); |
cb24405e | 261 | if (found) { |
1c5113cf SV |
262 | struct rds_conn_path *cp; |
263 | int i; | |
264 | ||
840df162 | 265 | for (i = 0; i < npaths; i++) { |
1c5113cf | 266 | cp = &conn->c_path[i]; |
02105b2c SV |
267 | /* The ->conn_alloc invocation may have |
268 | * allocated resource for all paths, so all | |
269 | * of them may have to be freed here. | |
270 | */ | |
271 | if (cp->cp_transport_data) | |
272 | trans->conn_free(cp->cp_transport_data); | |
1c5113cf | 273 | } |
840df162 | 274 | kfree(conn->c_path); |
cb24405e AG |
275 | kmem_cache_free(rds_conn_slab, conn); |
276 | conn = found; | |
277 | } else { | |
905dd418 SV |
278 | conn->c_my_gen_num = rds_gen_num; |
279 | conn->c_peer_gen_num = 0; | |
3b20fc38 | 280 | hlist_add_head_rcu(&conn->c_hash_node, head); |
cb24405e AG |
281 | rds_cong_add_conn(conn); |
282 | rds_conn_count++; | |
283 | } | |
00e0f34c | 284 | } |
00e0f34c AG |
285 | spin_unlock_irqrestore(&rds_conn_lock, flags); |
286 | ||
287 | out: | |
288 | return conn; | |
289 | } | |
290 | ||
d5a8ac28 SV |
291 | struct rds_connection *rds_conn_create(struct net *net, |
292 | __be32 laddr, __be32 faddr, | |
00e0f34c AG |
293 | struct rds_transport *trans, gfp_t gfp) |
294 | { | |
d5a8ac28 | 295 | return __rds_conn_create(net, laddr, faddr, trans, gfp, 0); |
00e0f34c | 296 | } |
616b757a | 297 | EXPORT_SYMBOL_GPL(rds_conn_create); |
00e0f34c | 298 | |
d5a8ac28 SV |
299 | struct rds_connection *rds_conn_create_outgoing(struct net *net, |
300 | __be32 laddr, __be32 faddr, | |
00e0f34c AG |
301 | struct rds_transport *trans, gfp_t gfp) |
302 | { | |
d5a8ac28 | 303 | return __rds_conn_create(net, laddr, faddr, trans, gfp, 1); |
00e0f34c | 304 | } |
616b757a | 305 | EXPORT_SYMBOL_GPL(rds_conn_create_outgoing); |
00e0f34c | 306 | |
d769ef81 | 307 | void rds_conn_shutdown(struct rds_conn_path *cp) |
2dc39357 | 308 | { |
d769ef81 SV |
309 | struct rds_connection *conn = cp->cp_conn; |
310 | ||
2dc39357 | 311 | /* shut it down unless it's down already */ |
d769ef81 | 312 | if (!rds_conn_path_transition(cp, RDS_CONN_DOWN, RDS_CONN_DOWN)) { |
2dc39357 AG |
313 | /* |
314 | * Quiesce the connection mgmt handlers before we start tearing | |
315 | * things down. We don't hold the mutex for the entire | |
316 | * duration of the shutdown operation, else we may be | |
317 | * deadlocking with the CM handler. Instead, the CM event | |
318 | * handler is supposed to check for state DISCONNECTING | |
319 | */ | |
d769ef81 SV |
320 | mutex_lock(&cp->cp_cm_lock); |
321 | if (!rds_conn_path_transition(cp, RDS_CONN_UP, | |
322 | RDS_CONN_DISCONNECTING) && | |
323 | !rds_conn_path_transition(cp, RDS_CONN_ERROR, | |
324 | RDS_CONN_DISCONNECTING)) { | |
325 | rds_conn_path_error(cp, | |
326 | "shutdown called in state %d\n", | |
327 | atomic_read(&cp->cp_state)); | |
328 | mutex_unlock(&cp->cp_cm_lock); | |
2dc39357 AG |
329 | return; |
330 | } | |
d769ef81 | 331 | mutex_unlock(&cp->cp_cm_lock); |
2dc39357 | 332 | |
d769ef81 SV |
333 | wait_event(cp->cp_waitq, |
334 | !test_bit(RDS_IN_XMIT, &cp->cp_flags)); | |
335 | wait_event(cp->cp_waitq, | |
336 | !test_bit(RDS_RECV_REFILL, &cp->cp_flags)); | |
7e3f2952 | 337 | |
226f7a7d | 338 | conn->c_trans->conn_path_shutdown(cp); |
d769ef81 | 339 | rds_conn_path_reset(cp); |
2dc39357 | 340 | |
d769ef81 | 341 | if (!rds_conn_path_transition(cp, RDS_CONN_DISCONNECTING, |
e97656d0 SV |
342 | RDS_CONN_DOWN) && |
343 | !rds_conn_path_transition(cp, RDS_CONN_ERROR, | |
d769ef81 | 344 | RDS_CONN_DOWN)) { |
2dc39357 AG |
345 | /* This can happen - eg when we're in the middle of tearing |
346 | * down the connection, and someone unloads the rds module. | |
e97656d0 | 347 | * Quite reproducible with loopback connections. |
2dc39357 | 348 | * Mostly harmless. |
e97656d0 SV |
349 | * |
350 | * Note that this also happens with rds-tcp because | |
351 | * we could have triggered rds_conn_path_drop in irq | |
352 | * mode from rds_tcp_state change on the receipt of | |
353 | * a FIN, thus we need to recheck for RDS_CONN_ERROR | |
354 | * here. | |
2dc39357 | 355 | */ |
d769ef81 SV |
356 | rds_conn_path_error(cp, "%s: failed to transition " |
357 | "to state DOWN, current state " | |
358 | "is %d\n", __func__, | |
359 | atomic_read(&cp->cp_state)); | |
2dc39357 AG |
360 | return; |
361 | } | |
362 | } | |
363 | ||
364 | /* Then reconnect if it's still live. | |
365 | * The passive side of an IB loopback connection is never added | |
366 | * to the conn hash, so we never trigger a reconnect on this | |
367 | * conn - the reconnect is always triggered by the active peer. */ | |
d769ef81 | 368 | cancel_delayed_work_sync(&cp->cp_conn_w); |
bcf50ef2 CM |
369 | rcu_read_lock(); |
370 | if (!hlist_unhashed(&conn->c_hash_node)) { | |
371 | rcu_read_unlock(); | |
8315011a | 372 | rds_queue_reconnect(cp); |
bcf50ef2 CM |
373 | } else { |
374 | rcu_read_unlock(); | |
375 | } | |
2dc39357 AG |
376 | } |
377 | ||
3ecc5693 SV |
378 | /* destroy a single rds_conn_path. rds_conn_destroy() iterates over |
379 | * all paths using rds_conn_path_destroy() | |
380 | */ | |
381 | static void rds_conn_path_destroy(struct rds_conn_path *cp) | |
382 | { | |
383 | struct rds_message *rm, *rtmp; | |
384 | ||
02105b2c SV |
385 | if (!cp->cp_transport_data) |
386 | return; | |
387 | ||
3ecc5693 SV |
388 | /* make sure lingering queued work won't try to ref the conn */ |
389 | cancel_delayed_work_sync(&cp->cp_send_w); | |
390 | cancel_delayed_work_sync(&cp->cp_recv_w); | |
391 | ||
aed20a53 SV |
392 | rds_conn_path_drop(cp, true); |
393 | flush_work(&cp->cp_down_w); | |
394 | ||
3ecc5693 SV |
395 | /* tear down queued messages */ |
396 | list_for_each_entry_safe(rm, rtmp, | |
397 | &cp->cp_send_queue, | |
398 | m_conn_item) { | |
399 | list_del_init(&rm->m_conn_item); | |
400 | BUG_ON(!list_empty(&rm->m_sock_item)); | |
401 | rds_message_put(rm); | |
402 | } | |
403 | if (cp->cp_xmit_rm) | |
404 | rds_message_put(cp->cp_xmit_rm); | |
405 | ||
406 | cp->cp_conn->c_trans->conn_free(cp->cp_transport_data); | |
407 | } | |
408 | ||
2dc39357 AG |
409 | /* |
410 | * Stop and free a connection. | |
ffcec0e1 ZB |
411 | * |
412 | * This can only be used in very limited circumstances. It assumes that once | |
413 | * the conn has been shutdown that no one else is referencing the connection. | |
414 | * We can only ensure this in the rmmod path in the current code. | |
2dc39357 | 415 | */ |
00e0f34c AG |
416 | void rds_conn_destroy(struct rds_connection *conn) |
417 | { | |
fe8ff6b5 | 418 | unsigned long flags; |
02105b2c SV |
419 | int i; |
420 | struct rds_conn_path *cp; | |
840df162 | 421 | int npaths = (conn->c_trans->t_mp_capable ? RDS_MPATH_WORKERS : 1); |
00e0f34c AG |
422 | |
423 | rdsdebug("freeing conn %p for %pI4 -> " | |
424 | "%pI4\n", conn, &conn->c_laddr, | |
425 | &conn->c_faddr); | |
426 | ||
c14b0366 | 427 | conn->c_destroy_in_prog = 1; |
abf45439 CM |
428 | /* Ensure conn will not be scheduled for reconnect */ |
429 | spin_lock_irq(&rds_conn_lock); | |
bcf50ef2 | 430 | hlist_del_init_rcu(&conn->c_hash_node); |
abf45439 | 431 | spin_unlock_irq(&rds_conn_lock); |
bcf50ef2 CM |
432 | synchronize_rcu(); |
433 | ||
ffcec0e1 | 434 | /* shut the connection down */ |
840df162 | 435 | for (i = 0; i < npaths; i++) { |
02105b2c SV |
436 | cp = &conn->c_path[i]; |
437 | rds_conn_path_destroy(cp); | |
438 | BUG_ON(!list_empty(&cp->cp_retrans)); | |
00e0f34c | 439 | } |
00e0f34c AG |
440 | |
441 | /* | |
442 | * The congestion maps aren't freed up here. They're | |
443 | * freed by rds_cong_exit() after all the connections | |
444 | * have been freed. | |
445 | */ | |
446 | rds_cong_remove_conn(conn); | |
447 | ||
8edc3aff | 448 | put_net(conn->c_net); |
840df162 | 449 | kfree(conn->c_path); |
00e0f34c AG |
450 | kmem_cache_free(rds_conn_slab, conn); |
451 | ||
fe8ff6b5 | 452 | spin_lock_irqsave(&rds_conn_lock, flags); |
00e0f34c | 453 | rds_conn_count--; |
fe8ff6b5 | 454 | spin_unlock_irqrestore(&rds_conn_lock, flags); |
00e0f34c | 455 | } |
616b757a | 456 | EXPORT_SYMBOL_GPL(rds_conn_destroy); |
00e0f34c AG |
457 | |
458 | static void rds_conn_message_info(struct socket *sock, unsigned int len, | |
459 | struct rds_info_iterator *iter, | |
460 | struct rds_info_lengths *lens, | |
461 | int want_send) | |
462 | { | |
463 | struct hlist_head *head; | |
00e0f34c AG |
464 | struct list_head *list; |
465 | struct rds_connection *conn; | |
466 | struct rds_message *rm; | |
00e0f34c | 467 | unsigned int total = 0; |
501dcccd | 468 | unsigned long flags; |
00e0f34c | 469 | size_t i; |
992c9ec5 | 470 | int j; |
00e0f34c AG |
471 | |
472 | len /= sizeof(struct rds_info_message); | |
473 | ||
bcf50ef2 | 474 | rcu_read_lock(); |
00e0f34c AG |
475 | |
476 | for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash); | |
477 | i++, head++) { | |
b67bfe0d | 478 | hlist_for_each_entry_rcu(conn, head, c_hash_node) { |
992c9ec5 | 479 | struct rds_conn_path *cp; |
840df162 SV |
480 | int npaths; |
481 | ||
482 | npaths = (conn->c_trans->t_mp_capable ? | |
483 | RDS_MPATH_WORKERS : 1); | |
992c9ec5 | 484 | |
840df162 | 485 | for (j = 0; j < npaths; j++) { |
992c9ec5 SV |
486 | cp = &conn->c_path[j]; |
487 | if (want_send) | |
488 | list = &cp->cp_send_queue; | |
489 | else | |
490 | list = &cp->cp_retrans; | |
491 | ||
492 | spin_lock_irqsave(&cp->cp_lock, flags); | |
493 | ||
494 | /* XXX too lazy to maintain counts.. */ | |
495 | list_for_each_entry(rm, list, m_conn_item) { | |
496 | total++; | |
497 | if (total <= len) | |
498 | rds_inc_info_copy(&rm->m_inc, | |
499 | iter, | |
500 | conn->c_laddr, | |
501 | conn->c_faddr, | |
502 | 0); | |
503 | } | |
504 | ||
505 | spin_unlock_irqrestore(&cp->cp_lock, flags); | |
00e0f34c | 506 | } |
00e0f34c AG |
507 | } |
508 | } | |
bcf50ef2 | 509 | rcu_read_unlock(); |
00e0f34c AG |
510 | |
511 | lens->nr = total; | |
512 | lens->each = sizeof(struct rds_info_message); | |
513 | } | |
514 | ||
515 | static void rds_conn_message_info_send(struct socket *sock, unsigned int len, | |
516 | struct rds_info_iterator *iter, | |
517 | struct rds_info_lengths *lens) | |
518 | { | |
519 | rds_conn_message_info(sock, len, iter, lens, 1); | |
520 | } | |
521 | ||
522 | static void rds_conn_message_info_retrans(struct socket *sock, | |
523 | unsigned int len, | |
524 | struct rds_info_iterator *iter, | |
525 | struct rds_info_lengths *lens) | |
526 | { | |
527 | rds_conn_message_info(sock, len, iter, lens, 0); | |
528 | } | |
529 | ||
530 | void rds_for_each_conn_info(struct socket *sock, unsigned int len, | |
531 | struct rds_info_iterator *iter, | |
532 | struct rds_info_lengths *lens, | |
533 | int (*visitor)(struct rds_connection *, void *), | |
534 | size_t item_len) | |
535 | { | |
536 | uint64_t buffer[(item_len + 7) / 8]; | |
537 | struct hlist_head *head; | |
00e0f34c | 538 | struct rds_connection *conn; |
00e0f34c AG |
539 | size_t i; |
540 | ||
bcf50ef2 | 541 | rcu_read_lock(); |
00e0f34c AG |
542 | |
543 | lens->nr = 0; | |
544 | lens->each = item_len; | |
545 | ||
546 | for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash); | |
547 | i++, head++) { | |
b67bfe0d | 548 | hlist_for_each_entry_rcu(conn, head, c_hash_node) { |
00e0f34c AG |
549 | |
550 | /* XXX no c_lock usage.. */ | |
551 | if (!visitor(conn, buffer)) | |
552 | continue; | |
553 | ||
554 | /* We copy as much as we can fit in the buffer, | |
555 | * but we count all items so that the caller | |
556 | * can resize the buffer. */ | |
557 | if (len >= item_len) { | |
558 | rds_info_copy(iter, buffer, item_len); | |
559 | len -= item_len; | |
560 | } | |
561 | lens->nr++; | |
562 | } | |
563 | } | |
bcf50ef2 | 564 | rcu_read_unlock(); |
00e0f34c | 565 | } |
616b757a | 566 | EXPORT_SYMBOL_GPL(rds_for_each_conn_info); |
00e0f34c | 567 | |
bb789763 SS |
568 | static void rds_walk_conn_path_info(struct socket *sock, unsigned int len, |
569 | struct rds_info_iterator *iter, | |
570 | struct rds_info_lengths *lens, | |
571 | int (*visitor)(struct rds_conn_path *, void *), | |
572 | size_t item_len) | |
992c9ec5 SV |
573 | { |
574 | u64 buffer[(item_len + 7) / 8]; | |
575 | struct hlist_head *head; | |
576 | struct rds_connection *conn; | |
577 | size_t i; | |
578 | int j; | |
579 | ||
580 | rcu_read_lock(); | |
581 | ||
582 | lens->nr = 0; | |
583 | lens->each = item_len; | |
584 | ||
585 | for (i = 0, head = rds_conn_hash; i < ARRAY_SIZE(rds_conn_hash); | |
586 | i++, head++) { | |
587 | hlist_for_each_entry_rcu(conn, head, c_hash_node) { | |
588 | struct rds_conn_path *cp; | |
840df162 | 589 | int npaths; |
992c9ec5 | 590 | |
840df162 SV |
591 | npaths = (conn->c_trans->t_mp_capable ? |
592 | RDS_MPATH_WORKERS : 1); | |
593 | for (j = 0; j < npaths; j++) { | |
992c9ec5 SV |
594 | cp = &conn->c_path[j]; |
595 | ||
596 | /* XXX no cp_lock usage.. */ | |
597 | if (!visitor(cp, buffer)) | |
598 | continue; | |
992c9ec5 SV |
599 | } |
600 | ||
601 | /* We copy as much as we can fit in the buffer, | |
602 | * but we count all items so that the caller | |
603 | * can resize the buffer. | |
604 | */ | |
605 | if (len >= item_len) { | |
606 | rds_info_copy(iter, buffer, item_len); | |
607 | len -= item_len; | |
608 | } | |
609 | lens->nr++; | |
610 | } | |
611 | } | |
612 | rcu_read_unlock(); | |
613 | } | |
614 | ||
615 | static int rds_conn_info_visitor(struct rds_conn_path *cp, void *buffer) | |
00e0f34c AG |
616 | { |
617 | struct rds_info_connection *cinfo = buffer; | |
618 | ||
992c9ec5 SV |
619 | cinfo->next_tx_seq = cp->cp_next_tx_seq; |
620 | cinfo->next_rx_seq = cp->cp_next_rx_seq; | |
621 | cinfo->laddr = cp->cp_conn->c_laddr; | |
622 | cinfo->faddr = cp->cp_conn->c_faddr; | |
623 | strncpy(cinfo->transport, cp->cp_conn->c_trans->t_name, | |
00e0f34c AG |
624 | sizeof(cinfo->transport)); |
625 | cinfo->flags = 0; | |
626 | ||
992c9ec5 | 627 | rds_conn_info_set(cinfo->flags, test_bit(RDS_IN_XMIT, &cp->cp_flags), |
0f4b1c7e | 628 | SENDING); |
00e0f34c AG |
629 | /* XXX Future: return the state rather than these funky bits */ |
630 | rds_conn_info_set(cinfo->flags, | |
992c9ec5 | 631 | atomic_read(&cp->cp_state) == RDS_CONN_CONNECTING, |
00e0f34c AG |
632 | CONNECTING); |
633 | rds_conn_info_set(cinfo->flags, | |
992c9ec5 | 634 | atomic_read(&cp->cp_state) == RDS_CONN_UP, |
00e0f34c AG |
635 | CONNECTED); |
636 | return 1; | |
637 | } | |
638 | ||
639 | static void rds_conn_info(struct socket *sock, unsigned int len, | |
640 | struct rds_info_iterator *iter, | |
641 | struct rds_info_lengths *lens) | |
642 | { | |
992c9ec5 | 643 | rds_walk_conn_path_info(sock, len, iter, lens, |
00e0f34c AG |
644 | rds_conn_info_visitor, |
645 | sizeof(struct rds_info_connection)); | |
646 | } | |
647 | ||
ef87b7ea | 648 | int rds_conn_init(void) |
00e0f34c AG |
649 | { |
650 | rds_conn_slab = kmem_cache_create("rds_connection", | |
651 | sizeof(struct rds_connection), | |
652 | 0, 0, NULL); | |
8690bfa1 | 653 | if (!rds_conn_slab) |
00e0f34c AG |
654 | return -ENOMEM; |
655 | ||
656 | rds_info_register_func(RDS_INFO_CONNECTIONS, rds_conn_info); | |
657 | rds_info_register_func(RDS_INFO_SEND_MESSAGES, | |
658 | rds_conn_message_info_send); | |
659 | rds_info_register_func(RDS_INFO_RETRANS_MESSAGES, | |
660 | rds_conn_message_info_retrans); | |
661 | ||
662 | return 0; | |
663 | } | |
664 | ||
665 | void rds_conn_exit(void) | |
666 | { | |
667 | rds_loop_exit(); | |
668 | ||
669 | WARN_ON(!hlist_empty(rds_conn_hash)); | |
670 | ||
671 | kmem_cache_destroy(rds_conn_slab); | |
672 | ||
673 | rds_info_deregister_func(RDS_INFO_CONNECTIONS, rds_conn_info); | |
674 | rds_info_deregister_func(RDS_INFO_SEND_MESSAGES, | |
675 | rds_conn_message_info_send); | |
676 | rds_info_deregister_func(RDS_INFO_RETRANS_MESSAGES, | |
677 | rds_conn_message_info_retrans); | |
678 | } | |
679 | ||
680 | /* | |
681 | * Force a disconnect | |
682 | */ | |
aed20a53 | 683 | void rds_conn_path_drop(struct rds_conn_path *cp, bool destroy) |
0cb43965 SV |
684 | { |
685 | atomic_set(&cp->cp_state, RDS_CONN_ERROR); | |
aed20a53 SV |
686 | |
687 | if (!destroy && cp->cp_conn->c_destroy_in_prog) | |
688 | return; | |
689 | ||
0cb43965 SV |
690 | queue_work(rds_wq, &cp->cp_down_w); |
691 | } | |
692 | EXPORT_SYMBOL_GPL(rds_conn_path_drop); | |
693 | ||
00e0f34c AG |
694 | void rds_conn_drop(struct rds_connection *conn) |
695 | { | |
5916e2c1 | 696 | WARN_ON(conn->c_trans->t_mp_capable); |
aed20a53 | 697 | rds_conn_path_drop(&conn->c_path[0], false); |
00e0f34c | 698 | } |
616b757a | 699 | EXPORT_SYMBOL_GPL(rds_conn_drop); |
00e0f34c | 700 | |
f3c6808d ZB |
701 | /* |
702 | * If the connection is down, trigger a connect. We may have scheduled a | |
703 | * delayed reconnect however - in this case we should not interfere. | |
704 | */ | |
3c0a5900 SV |
705 | void rds_conn_path_connect_if_down(struct rds_conn_path *cp) |
706 | { | |
707 | if (rds_conn_path_state(cp) == RDS_CONN_DOWN && | |
708 | !test_and_set_bit(RDS_RECONNECT_PENDING, &cp->cp_flags)) | |
709 | queue_delayed_work(rds_wq, &cp->cp_conn_w, 0); | |
710 | } | |
1a0e100f | 711 | EXPORT_SYMBOL_GPL(rds_conn_path_connect_if_down); |
3c0a5900 | 712 | |
f3c6808d ZB |
713 | void rds_conn_connect_if_down(struct rds_connection *conn) |
714 | { | |
3c0a5900 SV |
715 | WARN_ON(conn->c_trans->t_mp_capable); |
716 | rds_conn_path_connect_if_down(&conn->c_path[0]); | |
f3c6808d ZB |
717 | } |
718 | EXPORT_SYMBOL_GPL(rds_conn_connect_if_down); | |
719 | ||
fb1b3dc4 SV |
720 | void |
721 | __rds_conn_path_error(struct rds_conn_path *cp, const char *fmt, ...) | |
722 | { | |
723 | va_list ap; | |
724 | ||
725 | va_start(ap, fmt); | |
726 | vprintk(fmt, ap); | |
727 | va_end(ap); | |
728 | ||
aed20a53 | 729 | rds_conn_path_drop(cp, false); |
fb1b3dc4 | 730 | } |