Commit | Line | Data |
---|---|---|
4a3388c8 DH |
1 | /* Client connection-specific management code. |
2 | * | |
3 | * Copyright (C) 2016 Red Hat, Inc. All Rights Reserved. | |
4 | * Written by David Howells (dhowells@redhat.com) | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or | |
7 | * modify it under the terms of the GNU General Public Licence | |
8 | * as published by the Free Software Foundation; either version | |
9 | * 2 of the Licence, or (at your option) any later version. | |
45025bce DH |
10 | * |
11 | * | |
12 | * Client connections need to be cached for a little while after they've made a | |
13 | * call so as to handle retransmitted DATA packets in case the server didn't | |
14 | * receive the final ACK or terminating ABORT we sent it. | |
15 | * | |
16 | * Client connections can be in one of a number of cache states: | |
17 | * | |
18 | * (1) INACTIVE - The connection is not held in any list and may not have been | |
19 | * exposed to the world. If it has been previously exposed, it was | |
20 | * discarded from the idle list after expiring. | |
21 | * | |
22 | * (2) WAITING - The connection is waiting for the number of client conns to | |
23 | * drop below the maximum capacity. Calls may be in progress upon it from | |
24 | * when it was active and got culled. | |
25 | * | |
26 | * The connection is on the rxrpc_waiting_client_conns list which is kept | |
27 | * in to-be-granted order. Culled conns with waiters go to the back of | |
28 | * the queue just like new conns. | |
29 | * | |
30 | * (3) ACTIVE - The connection has at least one call in progress upon it, it | |
31 | * may freely grant available channels to new calls and calls may be | |
32 | * waiting on it for channels to become available. | |
33 | * | |
34 | * The connection is on the rxrpc_active_client_conns list which is kept | |
35 | * in activation order for culling purposes. | |
36 | * | |
37 | * rxrpc_nr_active_client_conns is held incremented also. | |
38 | * | |
39 | * (4) CULLED - The connection got summarily culled to try and free up | |
40 | * capacity. Calls currently in progress on the connection are allowed to | |
41 | * continue, but new calls will have to wait. There can be no waiters in | |
42 | * this state - the conn would have to go to the WAITING state instead. | |
43 | * | |
44 | * (5) IDLE - The connection has no calls in progress upon it and must have | |
45 | * been exposed to the world (ie. the EXPOSED flag must be set). When it | |
46 | * expires, the EXPOSED flag is cleared and the connection transitions to | |
47 | * the INACTIVE state. | |
48 | * | |
49 | * The connection is on the rxrpc_idle_client_conns list which is kept in | |
50 | * order of how soon they'll expire. | |
51 | * | |
52 | * There are flags of relevance to the cache: | |
53 | * | |
54 | * (1) EXPOSED - The connection ID got exposed to the world. If this flag is | |
55 | * set, an extra ref is added to the connection preventing it from being | |
56 | * reaped when it has no calls outstanding. This flag is cleared and the | |
57 | * ref dropped when a conn is discarded from the idle list. | |
58 | * | |
59 | * This allows us to move terminal call state retransmission to the | |
60 | * connection and to discard the call immediately we think it is done | |
61 | * with. It also give us a chance to reuse the connection. | |
62 | * | |
63 | * (2) DONT_REUSE - The connection should be discarded as soon as possible and | |
64 | * should not be reused. This is set when an exclusive connection is used | |
65 | * or a call ID counter overflows. | |
66 | * | |
67 | * The caching state may only be changed if the cache lock is held. | |
68 | * | |
69 | * There are two idle client connection expiry durations. If the total number | |
70 | * of connections is below the reap threshold, we use the normal duration; if | |
71 | * it's above, we use the fast duration. | |
4a3388c8 DH |
72 | */ |
73 | ||
74 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt | |
75 | ||
76 | #include <linux/slab.h> | |
77 | #include <linux/idr.h> | |
78 | #include <linux/timer.h> | |
79 | #include "ar-internal.h" | |
80 | ||
45025bce DH |
81 | __read_mostly unsigned int rxrpc_max_client_connections = 1000; |
82 | __read_mostly unsigned int rxrpc_reap_client_connections = 900; | |
83 | __read_mostly unsigned int rxrpc_conn_idle_client_expiry = 2 * 60 * HZ; | |
84 | __read_mostly unsigned int rxrpc_conn_idle_client_fast_expiry = 2 * HZ; | |
85 | ||
86 | static unsigned int rxrpc_nr_client_conns; | |
87 | static unsigned int rxrpc_nr_active_client_conns; | |
88 | static __read_mostly bool rxrpc_kill_all_client_conns; | |
89 | ||
90 | static DEFINE_SPINLOCK(rxrpc_client_conn_cache_lock); | |
91 | static DEFINE_SPINLOCK(rxrpc_client_conn_discard_mutex); | |
92 | static LIST_HEAD(rxrpc_waiting_client_conns); | |
93 | static LIST_HEAD(rxrpc_active_client_conns); | |
94 | static LIST_HEAD(rxrpc_idle_client_conns); | |
95 | ||
4a3388c8 DH |
96 | /* |
97 | * We use machine-unique IDs for our client connections. | |
98 | */ | |
99 | DEFINE_IDR(rxrpc_client_conn_ids); | |
100 | static DEFINE_SPINLOCK(rxrpc_conn_id_lock); | |
101 | ||
45025bce DH |
102 | static void rxrpc_cull_active_client_conns(void); |
103 | static void rxrpc_discard_expired_client_conns(struct work_struct *); | |
104 | ||
105 | static DECLARE_DELAYED_WORK(rxrpc_client_conn_reap, | |
106 | rxrpc_discard_expired_client_conns); | |
107 | ||
4a3388c8 DH |
108 | /* |
109 | * Get a connection ID and epoch for a client connection from the global pool. | |
110 | * The connection struct pointer is then recorded in the idr radix tree. The | |
090f85de DH |
111 | * epoch doesn't change until the client is rebooted (or, at least, unless the |
112 | * module is unloaded). | |
4a3388c8 | 113 | */ |
c6d2b8d7 DH |
114 | static int rxrpc_get_client_connection_id(struct rxrpc_connection *conn, |
115 | gfp_t gfp) | |
4a3388c8 | 116 | { |
4a3388c8 DH |
117 | int id; |
118 | ||
119 | _enter(""); | |
120 | ||
121 | idr_preload(gfp); | |
4a3388c8 DH |
122 | spin_lock(&rxrpc_conn_id_lock); |
123 | ||
090f85de DH |
124 | id = idr_alloc_cyclic(&rxrpc_client_conn_ids, conn, |
125 | 1, 0x40000000, GFP_NOWAIT); | |
126 | if (id < 0) | |
127 | goto error; | |
4a3388c8 DH |
128 | |
129 | spin_unlock(&rxrpc_conn_id_lock); | |
4a3388c8 DH |
130 | idr_preload_end(); |
131 | ||
090f85de | 132 | conn->proto.epoch = rxrpc_epoch; |
4a3388c8 DH |
133 | conn->proto.cid = id << RXRPC_CIDSHIFT; |
134 | set_bit(RXRPC_CONN_HAS_IDR, &conn->flags); | |
090f85de | 135 | _leave(" [CID %x]", conn->proto.cid); |
4a3388c8 DH |
136 | return 0; |
137 | ||
138 | error: | |
139 | spin_unlock(&rxrpc_conn_id_lock); | |
4a3388c8 DH |
140 | idr_preload_end(); |
141 | _leave(" = %d", id); | |
142 | return id; | |
143 | } | |
144 | ||
145 | /* | |
146 | * Release a connection ID for a client connection from the global pool. | |
147 | */ | |
001c1122 | 148 | static void rxrpc_put_client_connection_id(struct rxrpc_connection *conn) |
4a3388c8 DH |
149 | { |
150 | if (test_bit(RXRPC_CONN_HAS_IDR, &conn->flags)) { | |
151 | spin_lock(&rxrpc_conn_id_lock); | |
152 | idr_remove(&rxrpc_client_conn_ids, | |
153 | conn->proto.cid >> RXRPC_CIDSHIFT); | |
154 | spin_unlock(&rxrpc_conn_id_lock); | |
155 | } | |
156 | } | |
eb9b9d22 DH |
157 | |
158 | /* | |
159 | * Destroy the client connection ID tree. | |
160 | */ | |
161 | void rxrpc_destroy_client_conn_ids(void) | |
162 | { | |
163 | struct rxrpc_connection *conn; | |
164 | int id; | |
165 | ||
166 | if (!idr_is_empty(&rxrpc_client_conn_ids)) { | |
167 | idr_for_each_entry(&rxrpc_client_conn_ids, conn, id) { | |
168 | pr_err("AF_RXRPC: Leaked client conn %p {%d}\n", | |
169 | conn, atomic_read(&conn->usage)); | |
170 | } | |
171 | BUG(); | |
172 | } | |
173 | ||
174 | idr_destroy(&rxrpc_client_conn_ids); | |
175 | } | |
c6d2b8d7 DH |
176 | |
177 | /* | |
45025bce | 178 | * Allocate a client connection. |
c6d2b8d7 DH |
179 | */ |
180 | static struct rxrpc_connection * | |
181 | rxrpc_alloc_client_connection(struct rxrpc_conn_parameters *cp, gfp_t gfp) | |
182 | { | |
183 | struct rxrpc_connection *conn; | |
184 | int ret; | |
185 | ||
186 | _enter(""); | |
187 | ||
188 | conn = rxrpc_alloc_connection(gfp); | |
189 | if (!conn) { | |
190 | _leave(" = -ENOMEM"); | |
191 | return ERR_PTR(-ENOMEM); | |
192 | } | |
193 | ||
45025bce DH |
194 | atomic_set(&conn->usage, 1); |
195 | if (conn->params.exclusive) | |
196 | __set_bit(RXRPC_CONN_DONT_REUSE, &conn->flags); | |
197 | ||
c6d2b8d7 | 198 | conn->params = *cp; |
c6d2b8d7 DH |
199 | conn->out_clientflag = RXRPC_CLIENT_INITIATED; |
200 | conn->state = RXRPC_CONN_CLIENT; | |
201 | ||
c6d2b8d7 DH |
202 | ret = rxrpc_get_client_connection_id(conn, gfp); |
203 | if (ret < 0) | |
204 | goto error_0; | |
205 | ||
206 | ret = rxrpc_init_client_conn_security(conn); | |
207 | if (ret < 0) | |
208 | goto error_1; | |
209 | ||
210 | ret = conn->security->prime_packet_security(conn); | |
211 | if (ret < 0) | |
212 | goto error_2; | |
213 | ||
214 | write_lock(&rxrpc_connection_lock); | |
4d028b2c | 215 | list_add_tail(&conn->proc_link, &rxrpc_connection_proc_list); |
c6d2b8d7 DH |
216 | write_unlock(&rxrpc_connection_lock); |
217 | ||
218 | /* We steal the caller's peer ref. */ | |
219 | cp->peer = NULL; | |
220 | rxrpc_get_local(conn->params.local); | |
221 | key_get(conn->params.key); | |
222 | ||
223 | _leave(" = %p", conn); | |
224 | return conn; | |
225 | ||
226 | error_2: | |
227 | conn->security->clear(conn); | |
228 | error_1: | |
229 | rxrpc_put_client_connection_id(conn); | |
230 | error_0: | |
231 | kfree(conn); | |
232 | _leave(" = %d", ret); | |
233 | return ERR_PTR(ret); | |
234 | } | |
235 | ||
236 | /* | |
45025bce | 237 | * Determine if a connection may be reused. |
c6d2b8d7 | 238 | */ |
45025bce DH |
239 | static bool rxrpc_may_reuse_conn(struct rxrpc_connection *conn) |
240 | { | |
241 | int id_cursor, id, distance, limit; | |
242 | ||
243 | if (test_bit(RXRPC_CONN_DONT_REUSE, &conn->flags)) | |
244 | goto dont_reuse; | |
245 | ||
246 | if (conn->proto.epoch != rxrpc_epoch) | |
247 | goto mark_dont_reuse; | |
248 | ||
249 | /* The IDR tree gets very expensive on memory if the connection IDs are | |
250 | * widely scattered throughout the number space, so we shall want to | |
251 | * kill off connections that, say, have an ID more than about four | |
252 | * times the maximum number of client conns away from the current | |
253 | * allocation point to try and keep the IDs concentrated. | |
254 | */ | |
255 | id_cursor = READ_ONCE(rxrpc_client_conn_ids.cur); | |
256 | id = conn->proto.cid >> RXRPC_CIDSHIFT; | |
257 | distance = id - id_cursor; | |
258 | if (distance < 0) | |
259 | distance = -distance; | |
260 | limit = round_up(rxrpc_max_client_connections, IDR_SIZE) * 4; | |
261 | if (distance > limit) | |
262 | goto mark_dont_reuse; | |
263 | ||
264 | return true; | |
265 | ||
266 | mark_dont_reuse: | |
267 | set_bit(RXRPC_CONN_DONT_REUSE, &conn->flags); | |
268 | dont_reuse: | |
269 | return false; | |
270 | } | |
271 | ||
272 | /* | |
273 | * Create or find a client connection to use for a call. | |
274 | * | |
275 | * If we return with a connection, the call will be on its waiting list. It's | |
276 | * left to the caller to assign a channel and wake up the call. | |
277 | */ | |
278 | static int rxrpc_get_client_conn(struct rxrpc_call *call, | |
279 | struct rxrpc_conn_parameters *cp, | |
280 | struct sockaddr_rxrpc *srx, | |
281 | gfp_t gfp) | |
c6d2b8d7 DH |
282 | { |
283 | struct rxrpc_connection *conn, *candidate = NULL; | |
284 | struct rxrpc_local *local = cp->local; | |
285 | struct rb_node *p, **pp, *parent; | |
286 | long diff; | |
45025bce | 287 | int ret = -ENOMEM; |
c6d2b8d7 DH |
288 | |
289 | _enter("{%d,%lx},", call->debug_id, call->user_call_ID); | |
290 | ||
291 | cp->peer = rxrpc_lookup_peer(cp->local, srx, gfp); | |
292 | if (!cp->peer) | |
45025bce | 293 | goto error; |
c6d2b8d7 | 294 | |
45025bce DH |
295 | /* If the connection is not meant to be exclusive, search the available |
296 | * connections to see if the connection we want to use already exists. | |
297 | */ | |
c6d2b8d7 | 298 | if (!cp->exclusive) { |
c6d2b8d7 DH |
299 | _debug("search 1"); |
300 | spin_lock(&local->client_conns_lock); | |
301 | p = local->client_conns.rb_node; | |
302 | while (p) { | |
303 | conn = rb_entry(p, struct rxrpc_connection, client_node); | |
304 | ||
305 | #define cmp(X) ((long)conn->params.X - (long)cp->X) | |
306 | diff = (cmp(peer) ?: | |
307 | cmp(key) ?: | |
308 | cmp(security_level)); | |
45025bce DH |
309 | #undef cmp |
310 | if (diff < 0) { | |
c6d2b8d7 | 311 | p = p->rb_left; |
45025bce | 312 | } else if (diff > 0) { |
c6d2b8d7 | 313 | p = p->rb_right; |
45025bce DH |
314 | } else { |
315 | if (rxrpc_may_reuse_conn(conn) && | |
316 | rxrpc_get_connection_maybe(conn)) | |
317 | goto found_extant_conn; | |
318 | /* The connection needs replacing. It's better | |
319 | * to effect that when we have something to | |
320 | * replace it with so that we don't have to | |
321 | * rebalance the tree twice. | |
322 | */ | |
323 | break; | |
324 | } | |
c6d2b8d7 DH |
325 | } |
326 | spin_unlock(&local->client_conns_lock); | |
327 | } | |
328 | ||
45025bce DH |
329 | /* There wasn't a connection yet or we need an exclusive connection. |
330 | * We need to create a candidate and then potentially redo the search | |
331 | * in case we're racing with another thread also trying to connect on a | |
332 | * shareable connection. | |
333 | */ | |
334 | _debug("new conn"); | |
c6d2b8d7 | 335 | candidate = rxrpc_alloc_client_connection(cp, gfp); |
45025bce DH |
336 | if (IS_ERR(candidate)) { |
337 | ret = PTR_ERR(candidate); | |
338 | goto error_peer; | |
c6d2b8d7 DH |
339 | } |
340 | ||
45025bce DH |
341 | /* Add the call to the new connection's waiting list in case we're |
342 | * going to have to wait for the connection to come live. It's our | |
343 | * connection, so we want first dibs on the channel slots. We would | |
344 | * normally have to take channel_lock but we do this before anyone else | |
345 | * can see the connection. | |
346 | */ | |
347 | list_add_tail(&call->chan_wait_link, &candidate->waiting_calls); | |
348 | ||
c6d2b8d7 | 349 | if (cp->exclusive) { |
45025bce | 350 | call->conn = candidate; |
278ac0cd | 351 | call->security_ix = candidate->security_ix; |
45025bce DH |
352 | _leave(" = 0 [exclusive %d]", candidate->debug_id); |
353 | return 0; | |
c6d2b8d7 DH |
354 | } |
355 | ||
45025bce DH |
356 | /* Publish the new connection for userspace to find. We need to redo |
357 | * the search before doing this lest we race with someone else adding a | |
358 | * conflicting instance. | |
c6d2b8d7 DH |
359 | */ |
360 | _debug("search 2"); | |
361 | spin_lock(&local->client_conns_lock); | |
362 | ||
363 | pp = &local->client_conns.rb_node; | |
364 | parent = NULL; | |
365 | while (*pp) { | |
366 | parent = *pp; | |
367 | conn = rb_entry(parent, struct rxrpc_connection, client_node); | |
368 | ||
45025bce | 369 | #define cmp(X) ((long)conn->params.X - (long)candidate->params.X) |
c6d2b8d7 DH |
370 | diff = (cmp(peer) ?: |
371 | cmp(key) ?: | |
372 | cmp(security_level)); | |
45025bce DH |
373 | #undef cmp |
374 | if (diff < 0) { | |
c6d2b8d7 | 375 | pp = &(*pp)->rb_left; |
45025bce | 376 | } else if (diff > 0) { |
c6d2b8d7 | 377 | pp = &(*pp)->rb_right; |
45025bce DH |
378 | } else { |
379 | if (rxrpc_may_reuse_conn(conn) && | |
380 | rxrpc_get_connection_maybe(conn)) | |
381 | goto found_extant_conn; | |
382 | /* The old connection is from an outdated epoch. */ | |
383 | _debug("replace conn"); | |
384 | clear_bit(RXRPC_CONN_IN_CLIENT_CONNS, &conn->flags); | |
385 | rb_replace_node(&conn->client_node, | |
386 | &candidate->client_node, | |
387 | &local->client_conns); | |
388 | goto candidate_published; | |
389 | } | |
c6d2b8d7 DH |
390 | } |
391 | ||
c6d2b8d7 | 392 | _debug("new conn"); |
001c1122 DH |
393 | rb_link_node(&candidate->client_node, parent, pp); |
394 | rb_insert_color(&candidate->client_node, &local->client_conns); | |
c6d2b8d7 | 395 | |
45025bce DH |
396 | candidate_published: |
397 | set_bit(RXRPC_CONN_IN_CLIENT_CONNS, &candidate->flags); | |
398 | call->conn = candidate; | |
278ac0cd | 399 | call->security_ix = candidate->security_ix; |
c6d2b8d7 | 400 | spin_unlock(&local->client_conns_lock); |
45025bce DH |
401 | _leave(" = 0 [new %d]", candidate->debug_id); |
402 | return 0; | |
c6d2b8d7 | 403 | |
45025bce DH |
404 | /* We come here if we found a suitable connection already in existence. |
405 | * Discard any candidate we may have allocated, and try to get a | |
406 | * channel on this one. | |
407 | */ | |
408 | found_extant_conn: | |
409 | _debug("found conn"); | |
410 | spin_unlock(&local->client_conns_lock); | |
c6d2b8d7 | 411 | |
45025bce DH |
412 | rxrpc_put_connection(candidate); |
413 | candidate = NULL; | |
c6d2b8d7 | 414 | |
45025bce DH |
415 | spin_lock(&conn->channel_lock); |
416 | call->conn = conn; | |
278ac0cd | 417 | call->security_ix = conn->security_ix; |
45025bce | 418 | list_add(&call->chan_wait_link, &conn->waiting_calls); |
c6d2b8d7 | 419 | spin_unlock(&conn->channel_lock); |
45025bce DH |
420 | _leave(" = 0 [extant %d]", conn->debug_id); |
421 | return 0; | |
422 | ||
423 | error_peer: | |
c6d2b8d7 DH |
424 | rxrpc_put_peer(cp->peer); |
425 | cp->peer = NULL; | |
45025bce DH |
426 | error: |
427 | _leave(" = %d", ret); | |
428 | return ret; | |
429 | } | |
c6d2b8d7 | 430 | |
45025bce DH |
431 | /* |
432 | * Activate a connection. | |
433 | */ | |
434 | static void rxrpc_activate_conn(struct rxrpc_connection *conn) | |
435 | { | |
436 | conn->cache_state = RXRPC_CONN_CLIENT_ACTIVE; | |
437 | rxrpc_nr_active_client_conns++; | |
438 | list_move_tail(&conn->cache_link, &rxrpc_active_client_conns); | |
439 | } | |
440 | ||
441 | /* | |
442 | * Attempt to animate a connection for a new call. | |
443 | * | |
444 | * If it's not exclusive, the connection is in the endpoint tree, and we're in | |
445 | * the conn's list of those waiting to grab a channel. There is, however, a | |
446 | * limit on the number of live connections allowed at any one time, so we may | |
447 | * have to wait for capacity to become available. | |
448 | * | |
449 | * Note that a connection on the waiting queue might *also* have active | |
450 | * channels if it has been culled to make space and then re-requested by a new | |
451 | * call. | |
452 | */ | |
453 | static void rxrpc_animate_client_conn(struct rxrpc_connection *conn) | |
454 | { | |
455 | unsigned int nr_conns; | |
456 | ||
457 | _enter("%d,%d", conn->debug_id, conn->cache_state); | |
458 | ||
459 | if (conn->cache_state == RXRPC_CONN_CLIENT_ACTIVE) | |
460 | goto out; | |
461 | ||
462 | spin_lock(&rxrpc_client_conn_cache_lock); | |
463 | ||
464 | nr_conns = rxrpc_nr_client_conns; | |
465 | if (!test_and_set_bit(RXRPC_CONN_COUNTED, &conn->flags)) | |
466 | rxrpc_nr_client_conns = nr_conns + 1; | |
467 | ||
468 | switch (conn->cache_state) { | |
469 | case RXRPC_CONN_CLIENT_ACTIVE: | |
470 | case RXRPC_CONN_CLIENT_WAITING: | |
471 | break; | |
472 | ||
473 | case RXRPC_CONN_CLIENT_INACTIVE: | |
474 | case RXRPC_CONN_CLIENT_CULLED: | |
475 | case RXRPC_CONN_CLIENT_IDLE: | |
476 | if (nr_conns >= rxrpc_max_client_connections) | |
477 | goto wait_for_capacity; | |
478 | goto activate_conn; | |
479 | ||
480 | default: | |
481 | BUG(); | |
001c1122 DH |
482 | } |
483 | ||
45025bce DH |
484 | out_unlock: |
485 | spin_unlock(&rxrpc_client_conn_cache_lock); | |
486 | out: | |
487 | _leave(" [%d]", conn->cache_state); | |
488 | return; | |
c6d2b8d7 | 489 | |
45025bce DH |
490 | activate_conn: |
491 | _debug("activate"); | |
492 | rxrpc_activate_conn(conn); | |
493 | goto out_unlock; | |
494 | ||
495 | wait_for_capacity: | |
496 | _debug("wait"); | |
497 | conn->cache_state = RXRPC_CONN_CLIENT_WAITING; | |
498 | list_move_tail(&conn->cache_link, &rxrpc_waiting_client_conns); | |
499 | goto out_unlock; | |
500 | } | |
501 | ||
502 | /* | |
503 | * Deactivate a channel. | |
504 | */ | |
505 | static void rxrpc_deactivate_one_channel(struct rxrpc_connection *conn, | |
506 | unsigned int channel) | |
507 | { | |
508 | struct rxrpc_channel *chan = &conn->channels[channel]; | |
509 | ||
510 | rcu_assign_pointer(chan->call, NULL); | |
511 | conn->active_chans &= ~(1 << channel); | |
512 | } | |
513 | ||
514 | /* | |
515 | * Assign a channel to the call at the front of the queue and wake the call up. | |
516 | * We don't increment the callNumber counter until this number has been exposed | |
517 | * to the world. | |
518 | */ | |
519 | static void rxrpc_activate_one_channel(struct rxrpc_connection *conn, | |
520 | unsigned int channel) | |
521 | { | |
522 | struct rxrpc_channel *chan = &conn->channels[channel]; | |
523 | struct rxrpc_call *call = list_entry(conn->waiting_calls.next, | |
524 | struct rxrpc_call, chan_wait_link); | |
525 | u32 call_id = chan->call_counter + 1; | |
526 | ||
af338a9e DH |
527 | write_lock_bh(&call->state_lock); |
528 | call->state = RXRPC_CALL_CLIENT_SEND_REQUEST; | |
529 | write_unlock_bh(&call->state_lock); | |
530 | ||
e34d4234 | 531 | rxrpc_see_call(call); |
45025bce DH |
532 | list_del_init(&call->chan_wait_link); |
533 | conn->active_chans |= 1 << channel; | |
534 | call->peer = rxrpc_get_peer(conn->params.peer); | |
535 | call->cid = conn->proto.cid | channel; | |
536 | call->call_id = call_id; | |
537 | ||
538 | _net("CONNECT call %08x:%08x as call %d on conn %d", | |
539 | call->cid, call->call_id, call->debug_id, conn->debug_id); | |
540 | ||
541 | /* Paired with the read barrier in rxrpc_wait_for_channel(). This | |
542 | * orders cid and epoch in the connection wrt to call_id without the | |
543 | * need to take the channel_lock. | |
544 | * | |
545 | * We provisionally assign a callNumber at this point, but we don't | |
546 | * confirm it until the call is about to be exposed. | |
547 | * | |
548 | * TODO: Pair with a barrier in the data_ready handler when that looks | |
549 | * at the call ID through a connection channel. | |
550 | */ | |
551 | smp_wmb(); | |
552 | chan->call_id = call_id; | |
553 | rcu_assign_pointer(chan->call, call); | |
554 | wake_up(&call->waitq); | |
555 | } | |
556 | ||
557 | /* | |
558 | * Assign channels and callNumbers to waiting calls. | |
559 | */ | |
560 | static void rxrpc_activate_channels(struct rxrpc_connection *conn) | |
561 | { | |
562 | unsigned char mask; | |
563 | ||
564 | _enter("%d", conn->debug_id); | |
565 | ||
566 | if (conn->cache_state != RXRPC_CONN_CLIENT_ACTIVE || | |
567 | conn->active_chans == RXRPC_ACTIVE_CHANS_MASK) | |
568 | return; | |
569 | ||
570 | spin_lock(&conn->channel_lock); | |
571 | ||
572 | while (!list_empty(&conn->waiting_calls) && | |
573 | (mask = ~conn->active_chans, | |
574 | mask &= RXRPC_ACTIVE_CHANS_MASK, | |
575 | mask != 0)) | |
576 | rxrpc_activate_one_channel(conn, __ffs(mask)); | |
577 | ||
578 | spin_unlock(&conn->channel_lock); | |
579 | _leave(""); | |
580 | } | |
581 | ||
582 | /* | |
583 | * Wait for a callNumber and a channel to be granted to a call. | |
584 | */ | |
585 | static int rxrpc_wait_for_channel(struct rxrpc_call *call, gfp_t gfp) | |
586 | { | |
587 | int ret = 0; | |
588 | ||
589 | _enter("%d", call->debug_id); | |
590 | ||
591 | if (!call->call_id) { | |
592 | DECLARE_WAITQUEUE(myself, current); | |
c6d2b8d7 | 593 | |
c6d2b8d7 | 594 | if (!gfpflags_allow_blocking(gfp)) { |
45025bce DH |
595 | ret = -EAGAIN; |
596 | goto out; | |
c6d2b8d7 DH |
597 | } |
598 | ||
45025bce | 599 | add_wait_queue_exclusive(&call->waitq, &myself); |
c6d2b8d7 DH |
600 | for (;;) { |
601 | set_current_state(TASK_INTERRUPTIBLE); | |
45025bce DH |
602 | if (call->call_id) |
603 | break; | |
604 | if (signal_pending(current)) { | |
605 | ret = -ERESTARTSYS; | |
c6d2b8d7 | 606 | break; |
45025bce | 607 | } |
c6d2b8d7 DH |
608 | schedule(); |
609 | } | |
45025bce | 610 | remove_wait_queue(&call->waitq, &myself); |
c6d2b8d7 DH |
611 | __set_current_state(TASK_RUNNING); |
612 | } | |
613 | ||
45025bce DH |
614 | /* Paired with the write barrier in rxrpc_activate_one_channel(). */ |
615 | smp_rmb(); | |
616 | ||
617 | out: | |
618 | _leave(" = %d", ret); | |
619 | return ret; | |
620 | } | |
621 | ||
622 | /* | |
623 | * find a connection for a call | |
624 | * - called in process context with IRQs enabled | |
625 | */ | |
626 | int rxrpc_connect_call(struct rxrpc_call *call, | |
627 | struct rxrpc_conn_parameters *cp, | |
628 | struct sockaddr_rxrpc *srx, | |
629 | gfp_t gfp) | |
630 | { | |
631 | int ret; | |
632 | ||
633 | _enter("{%d,%lx},", call->debug_id, call->user_call_ID); | |
634 | ||
635 | rxrpc_discard_expired_client_conns(NULL); | |
636 | rxrpc_cull_active_client_conns(); | |
637 | ||
638 | ret = rxrpc_get_client_conn(call, cp, srx, gfp); | |
639 | if (ret < 0) | |
640 | return ret; | |
641 | ||
642 | rxrpc_animate_client_conn(call->conn); | |
643 | rxrpc_activate_channels(call->conn); | |
644 | ||
645 | ret = rxrpc_wait_for_channel(call, gfp); | |
646 | if (ret < 0) | |
647 | rxrpc_disconnect_client_call(call); | |
648 | ||
649 | _leave(" = %d", ret); | |
650 | return ret; | |
651 | } | |
652 | ||
653 | /* | |
654 | * Note that a connection is about to be exposed to the world. Once it is | |
655 | * exposed, we maintain an extra ref on it that stops it from being summarily | |
656 | * discarded before it's (a) had a chance to deal with retransmission and (b) | |
657 | * had a chance at re-use (the per-connection security negotiation is | |
658 | * expensive). | |
659 | */ | |
660 | static void rxrpc_expose_client_conn(struct rxrpc_connection *conn) | |
661 | { | |
662 | if (!test_and_set_bit(RXRPC_CONN_EXPOSED, &conn->flags)) | |
663 | rxrpc_get_connection(conn); | |
664 | } | |
665 | ||
666 | /* | |
667 | * Note that a call, and thus a connection, is about to be exposed to the | |
668 | * world. | |
669 | */ | |
670 | void rxrpc_expose_client_call(struct rxrpc_call *call) | |
671 | { | |
672 | struct rxrpc_connection *conn = call->conn; | |
673 | struct rxrpc_channel *chan = | |
674 | &conn->channels[call->cid & RXRPC_CHANNELMASK]; | |
675 | ||
676 | if (!test_and_set_bit(RXRPC_CALL_EXPOSED, &call->flags)) { | |
677 | /* Mark the call ID as being used. If the callNumber counter | |
678 | * exceeds ~2 billion, we kill the connection after its | |
679 | * outstanding calls have finished so that the counter doesn't | |
680 | * wrap. | |
681 | */ | |
682 | chan->call_counter++; | |
683 | if (chan->call_counter >= INT_MAX) | |
684 | set_bit(RXRPC_CONN_DONT_REUSE, &conn->flags); | |
685 | rxrpc_expose_client_conn(conn); | |
686 | } | |
687 | } | |
688 | ||
689 | /* | |
690 | * Disconnect a client call. | |
691 | */ | |
692 | void rxrpc_disconnect_client_call(struct rxrpc_call *call) | |
693 | { | |
694 | unsigned int channel = call->cid & RXRPC_CHANNELMASK; | |
695 | struct rxrpc_connection *conn = call->conn; | |
696 | struct rxrpc_channel *chan = &conn->channels[channel]; | |
697 | ||
698 | call->conn = NULL; | |
699 | ||
c6d2b8d7 DH |
700 | spin_lock(&conn->channel_lock); |
701 | ||
45025bce DH |
702 | /* Calls that have never actually been assigned a channel can simply be |
703 | * discarded. If the conn didn't get used either, it will follow | |
704 | * immediately unless someone else grabs it in the meantime. | |
705 | */ | |
706 | if (!list_empty(&call->chan_wait_link)) { | |
707 | _debug("call is waiting"); | |
708 | ASSERTCMP(call->call_id, ==, 0); | |
709 | ASSERT(!test_bit(RXRPC_CALL_EXPOSED, &call->flags)); | |
710 | list_del_init(&call->chan_wait_link); | |
711 | ||
712 | /* We must deactivate or idle the connection if it's now | |
713 | * waiting for nothing. | |
714 | */ | |
715 | spin_lock(&rxrpc_client_conn_cache_lock); | |
716 | if (conn->cache_state == RXRPC_CONN_CLIENT_WAITING && | |
717 | list_empty(&conn->waiting_calls) && | |
718 | !conn->active_chans) | |
719 | goto idle_connection; | |
720 | goto out; | |
721 | } | |
722 | ||
723 | ASSERTCMP(rcu_access_pointer(chan->call), ==, call); | |
724 | ASSERTCMP(atomic_read(&conn->usage), >=, 2); | |
725 | ||
726 | /* If a client call was exposed to the world, we save the result for | |
727 | * retransmission. | |
728 | * | |
729 | * We use a barrier here so that the call number and abort code can be | |
730 | * read without needing to take a lock. | |
731 | * | |
732 | * TODO: Make the incoming packet handler check this and handle | |
733 | * terminal retransmission without requiring access to the call. | |
734 | */ | |
735 | if (test_bit(RXRPC_CALL_EXPOSED, &call->flags)) { | |
f5c17aae | 736 | _debug("exposed %u,%u", call->call_id, call->abort_code); |
45025bce DH |
737 | __rxrpc_disconnect_call(conn, call); |
738 | } | |
739 | ||
740 | /* See if we can pass the channel directly to another call. */ | |
741 | if (conn->cache_state == RXRPC_CONN_CLIENT_ACTIVE && | |
742 | !list_empty(&conn->waiting_calls)) { | |
743 | _debug("pass chan"); | |
744 | rxrpc_activate_one_channel(conn, channel); | |
745 | goto out_2; | |
746 | } | |
747 | ||
748 | /* Things are more complex and we need the cache lock. We might be | |
749 | * able to simply idle the conn or it might now be lurking on the wait | |
750 | * list. It might even get moved back to the active list whilst we're | |
751 | * waiting for the lock. | |
752 | */ | |
753 | spin_lock(&rxrpc_client_conn_cache_lock); | |
754 | ||
755 | switch (conn->cache_state) { | |
756 | case RXRPC_CONN_CLIENT_ACTIVE: | |
757 | if (list_empty(&conn->waiting_calls)) { | |
758 | rxrpc_deactivate_one_channel(conn, channel); | |
759 | if (!conn->active_chans) { | |
760 | rxrpc_nr_active_client_conns--; | |
761 | goto idle_connection; | |
762 | } | |
763 | goto out; | |
764 | } | |
765 | ||
766 | _debug("pass chan 2"); | |
767 | rxrpc_activate_one_channel(conn, channel); | |
768 | goto out; | |
769 | ||
770 | case RXRPC_CONN_CLIENT_CULLED: | |
771 | rxrpc_deactivate_one_channel(conn, channel); | |
772 | ASSERT(list_empty(&conn->waiting_calls)); | |
773 | if (!conn->active_chans) | |
774 | goto idle_connection; | |
775 | goto out; | |
776 | ||
777 | case RXRPC_CONN_CLIENT_WAITING: | |
778 | rxrpc_deactivate_one_channel(conn, channel); | |
779 | goto out; | |
780 | ||
781 | default: | |
782 | BUG(); | |
783 | } | |
c6d2b8d7 | 784 | |
45025bce DH |
785 | out: |
786 | spin_unlock(&rxrpc_client_conn_cache_lock); | |
787 | out_2: | |
788 | spin_unlock(&conn->channel_lock); | |
c6d2b8d7 | 789 | rxrpc_put_connection(conn); |
45025bce DH |
790 | _leave(""); |
791 | return; | |
792 | ||
793 | idle_connection: | |
794 | /* As no channels remain active, the connection gets deactivated | |
795 | * immediately or moved to the idle list for a short while. | |
796 | */ | |
797 | if (test_bit(RXRPC_CONN_EXPOSED, &conn->flags)) { | |
798 | _debug("make idle"); | |
799 | conn->idle_timestamp = jiffies; | |
800 | conn->cache_state = RXRPC_CONN_CLIENT_IDLE; | |
801 | list_move_tail(&conn->cache_link, &rxrpc_idle_client_conns); | |
802 | if (rxrpc_idle_client_conns.next == &conn->cache_link && | |
803 | !rxrpc_kill_all_client_conns) | |
804 | queue_delayed_work(rxrpc_workqueue, | |
805 | &rxrpc_client_conn_reap, | |
806 | rxrpc_conn_idle_client_expiry); | |
807 | } else { | |
808 | _debug("make inactive"); | |
809 | conn->cache_state = RXRPC_CONN_CLIENT_INACTIVE; | |
810 | list_del_init(&conn->cache_link); | |
811 | } | |
812 | goto out; | |
c6d2b8d7 | 813 | } |
001c1122 DH |
814 | |
815 | /* | |
45025bce | 816 | * Clean up a dead client connection. |
001c1122 | 817 | */ |
45025bce DH |
818 | static struct rxrpc_connection * |
819 | rxrpc_put_one_client_conn(struct rxrpc_connection *conn) | |
001c1122 | 820 | { |
45025bce | 821 | struct rxrpc_connection *next; |
001c1122 | 822 | struct rxrpc_local *local = conn->params.local; |
45025bce | 823 | unsigned int nr_conns; |
001c1122 | 824 | |
45025bce DH |
825 | if (test_bit(RXRPC_CONN_IN_CLIENT_CONNS, &conn->flags)) { |
826 | spin_lock(&local->client_conns_lock); | |
827 | if (test_and_clear_bit(RXRPC_CONN_IN_CLIENT_CONNS, | |
828 | &conn->flags)) | |
829 | rb_erase(&conn->client_node, &local->client_conns); | |
830 | spin_unlock(&local->client_conns_lock); | |
831 | } | |
001c1122 DH |
832 | |
833 | rxrpc_put_client_connection_id(conn); | |
45025bce DH |
834 | |
835 | ASSERTCMP(conn->cache_state, ==, RXRPC_CONN_CLIENT_INACTIVE); | |
836 | ||
837 | if (!test_bit(RXRPC_CONN_COUNTED, &conn->flags)) | |
838 | return NULL; | |
839 | ||
840 | spin_lock(&rxrpc_client_conn_cache_lock); | |
841 | nr_conns = --rxrpc_nr_client_conns; | |
842 | ||
843 | next = NULL; | |
844 | if (nr_conns < rxrpc_max_client_connections && | |
845 | !list_empty(&rxrpc_waiting_client_conns)) { | |
846 | next = list_entry(rxrpc_waiting_client_conns.next, | |
847 | struct rxrpc_connection, cache_link); | |
848 | rxrpc_get_connection(next); | |
849 | rxrpc_activate_conn(next); | |
850 | } | |
851 | ||
852 | spin_unlock(&rxrpc_client_conn_cache_lock); | |
853 | rxrpc_kill_connection(conn); | |
854 | ||
855 | if (next) | |
856 | rxrpc_activate_channels(next); | |
857 | ||
858 | /* We need to get rid of the temporary ref we took upon next, but we | |
859 | * can't call rxrpc_put_connection() recursively. | |
860 | */ | |
861 | return next; | |
862 | } | |
863 | ||
864 | /* | |
865 | * Clean up a dead client connections. | |
866 | */ | |
867 | void rxrpc_put_client_conn(struct rxrpc_connection *conn) | |
868 | { | |
869 | struct rxrpc_connection *next; | |
870 | ||
871 | do { | |
872 | _enter("%p{u=%d,d=%d}", | |
873 | conn, atomic_read(&conn->usage), conn->debug_id); | |
874 | ||
875 | next = rxrpc_put_one_client_conn(conn); | |
876 | ||
877 | if (!next) | |
878 | break; | |
879 | conn = next; | |
880 | } while (atomic_dec_and_test(&conn->usage)); | |
881 | ||
882 | _leave(""); | |
883 | } | |
884 | ||
885 | /* | |
886 | * Kill the longest-active client connections to make room for new ones. | |
887 | */ | |
888 | static void rxrpc_cull_active_client_conns(void) | |
889 | { | |
890 | struct rxrpc_connection *conn; | |
891 | unsigned int nr_conns = rxrpc_nr_client_conns; | |
892 | unsigned int nr_active, limit; | |
893 | ||
894 | _enter(""); | |
895 | ||
896 | ASSERTCMP(nr_conns, >=, 0); | |
897 | if (nr_conns < rxrpc_max_client_connections) { | |
898 | _leave(" [ok]"); | |
899 | return; | |
900 | } | |
901 | limit = rxrpc_reap_client_connections; | |
902 | ||
903 | spin_lock(&rxrpc_client_conn_cache_lock); | |
904 | nr_active = rxrpc_nr_active_client_conns; | |
905 | ||
906 | while (nr_active > limit) { | |
907 | ASSERT(!list_empty(&rxrpc_active_client_conns)); | |
908 | conn = list_entry(rxrpc_active_client_conns.next, | |
909 | struct rxrpc_connection, cache_link); | |
910 | ASSERTCMP(conn->cache_state, ==, RXRPC_CONN_CLIENT_ACTIVE); | |
911 | ||
912 | if (list_empty(&conn->waiting_calls)) { | |
913 | conn->cache_state = RXRPC_CONN_CLIENT_CULLED; | |
914 | list_del_init(&conn->cache_link); | |
915 | } else { | |
916 | conn->cache_state = RXRPC_CONN_CLIENT_WAITING; | |
917 | list_move_tail(&conn->cache_link, | |
918 | &rxrpc_waiting_client_conns); | |
919 | } | |
920 | ||
921 | nr_active--; | |
922 | } | |
923 | ||
924 | rxrpc_nr_active_client_conns = nr_active; | |
925 | spin_unlock(&rxrpc_client_conn_cache_lock); | |
926 | ASSERTCMP(nr_active, >=, 0); | |
927 | _leave(" [culled]"); | |
928 | } | |
929 | ||
930 | /* | |
931 | * Discard expired client connections from the idle list. Each conn in the | |
932 | * idle list has been exposed and holds an extra ref because of that. | |
933 | * | |
934 | * This may be called from conn setup or from a work item so cannot be | |
935 | * considered non-reentrant. | |
936 | */ | |
937 | static void rxrpc_discard_expired_client_conns(struct work_struct *work) | |
938 | { | |
939 | struct rxrpc_connection *conn; | |
940 | unsigned long expiry, conn_expires_at, now; | |
941 | unsigned int nr_conns; | |
942 | bool did_discard = false; | |
943 | ||
944 | _enter("%c", work ? 'w' : 'n'); | |
945 | ||
946 | if (list_empty(&rxrpc_idle_client_conns)) { | |
947 | _leave(" [empty]"); | |
948 | return; | |
949 | } | |
950 | ||
951 | /* Don't double up on the discarding */ | |
952 | if (!spin_trylock(&rxrpc_client_conn_discard_mutex)) { | |
953 | _leave(" [already]"); | |
954 | return; | |
955 | } | |
956 | ||
957 | /* We keep an estimate of what the number of conns ought to be after | |
958 | * we've discarded some so that we don't overdo the discarding. | |
959 | */ | |
960 | nr_conns = rxrpc_nr_client_conns; | |
961 | ||
962 | next: | |
963 | spin_lock(&rxrpc_client_conn_cache_lock); | |
964 | ||
965 | if (list_empty(&rxrpc_idle_client_conns)) | |
966 | goto out; | |
967 | ||
968 | conn = list_entry(rxrpc_idle_client_conns.next, | |
969 | struct rxrpc_connection, cache_link); | |
970 | ASSERT(test_bit(RXRPC_CONN_EXPOSED, &conn->flags)); | |
971 | ||
972 | if (!rxrpc_kill_all_client_conns) { | |
973 | /* If the number of connections is over the reap limit, we | |
974 | * expedite discard by reducing the expiry timeout. We must, | |
975 | * however, have at least a short grace period to be able to do | |
976 | * final-ACK or ABORT retransmission. | |
977 | */ | |
978 | expiry = rxrpc_conn_idle_client_expiry; | |
979 | if (nr_conns > rxrpc_reap_client_connections) | |
980 | expiry = rxrpc_conn_idle_client_fast_expiry; | |
981 | ||
982 | conn_expires_at = conn->idle_timestamp + expiry; | |
983 | ||
984 | now = READ_ONCE(jiffies); | |
985 | if (time_after(conn_expires_at, now)) | |
986 | goto not_yet_expired; | |
987 | } | |
988 | ||
989 | _debug("discard conn %d", conn->debug_id); | |
990 | if (!test_and_clear_bit(RXRPC_CONN_EXPOSED, &conn->flags)) | |
991 | BUG(); | |
992 | conn->cache_state = RXRPC_CONN_CLIENT_INACTIVE; | |
993 | list_del_init(&conn->cache_link); | |
994 | ||
995 | spin_unlock(&rxrpc_client_conn_cache_lock); | |
996 | ||
997 | /* When we cleared the EXPOSED flag, we took on responsibility for the | |
998 | * reference that that had on the usage count. We deal with that here. | |
999 | * If someone re-sets the flag and re-gets the ref, that's fine. | |
1000 | */ | |
1001 | rxrpc_put_connection(conn); | |
1002 | did_discard = true; | |
1003 | nr_conns--; | |
1004 | goto next; | |
1005 | ||
1006 | not_yet_expired: | |
1007 | /* The connection at the front of the queue hasn't yet expired, so | |
1008 | * schedule the work item for that point if we discarded something. | |
1009 | * | |
1010 | * We don't worry if the work item is already scheduled - it can look | |
1011 | * after rescheduling itself at a later time. We could cancel it, but | |
1012 | * then things get messier. | |
1013 | */ | |
1014 | _debug("not yet"); | |
1015 | if (!rxrpc_kill_all_client_conns) | |
1016 | queue_delayed_work(rxrpc_workqueue, | |
1017 | &rxrpc_client_conn_reap, | |
1018 | conn_expires_at - now); | |
1019 | ||
1020 | out: | |
1021 | spin_unlock(&rxrpc_client_conn_cache_lock); | |
1022 | spin_unlock(&rxrpc_client_conn_discard_mutex); | |
1023 | _leave(""); | |
1024 | } | |
1025 | ||
1026 | /* | |
1027 | * Preemptively destroy all the client connection records rather than waiting | |
1028 | * for them to time out | |
1029 | */ | |
1030 | void __exit rxrpc_destroy_all_client_connections(void) | |
1031 | { | |
1032 | _enter(""); | |
1033 | ||
1034 | spin_lock(&rxrpc_client_conn_cache_lock); | |
1035 | rxrpc_kill_all_client_conns = true; | |
1036 | spin_unlock(&rxrpc_client_conn_cache_lock); | |
1037 | ||
1038 | cancel_delayed_work(&rxrpc_client_conn_reap); | |
1039 | ||
1040 | if (!queue_delayed_work(rxrpc_workqueue, &rxrpc_client_conn_reap, 0)) | |
1041 | _debug("destroy: queue failed"); | |
1042 | ||
1043 | _leave(""); | |
001c1122 | 1044 | } |