Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* Client connection-specific management code.
3 *
4 * Copyright (C) 2016, 2020 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
6 *
7 * Client connections need to be cached for a little while after they've made a
8 * call so as to handle retransmitted DATA packets in case the server didn't
9 * receive the final ACK or terminating ABORT we sent it.
10 *
11 * There are flags of relevance to the cache:
12 *
13 * (2) DONT_REUSE - The connection should be discarded as soon as possible and
14 * should not be reused. This is set when an exclusive connection is used
15 * or a call ID counter overflows.
16 *
17 * The caching state may only be changed if the cache lock is held.
18 *
19 * There are two idle client connection expiry durations. If the total number
20 * of connections is below the reap threshold, we use the normal duration; if
21 * it's above, we use the fast duration.
22 */
23
24#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
25
26#include <linux/slab.h>
27#include <linux/idr.h>
28#include <linux/timer.h>
29#include <linux/sched/signal.h>
30
31#include "ar-internal.h"
32
33__read_mostly unsigned int rxrpc_reap_client_connections = 900;
34__read_mostly unsigned long rxrpc_conn_idle_client_expiry = 2 * 60 * HZ;
35__read_mostly unsigned long rxrpc_conn_idle_client_fast_expiry = 2 * HZ;
36
37static void rxrpc_activate_bundle(struct rxrpc_bundle *bundle)
38{
39 atomic_inc(&bundle->active);
40}
41
42/*
43 * Release a connection ID for a client connection.
44 */
45static void rxrpc_put_client_connection_id(struct rxrpc_local *local,
46 struct rxrpc_connection *conn)
47{
48 idr_remove(&local->conn_ids, conn->proto.cid >> RXRPC_CIDSHIFT);
49}
50
51/*
52 * Destroy the client connection ID tree.
53 */
54static void rxrpc_destroy_client_conn_ids(struct rxrpc_local *local)
55{
56 struct rxrpc_connection *conn;
57 int id;
58
59 if (!idr_is_empty(&local->conn_ids)) {
60 idr_for_each_entry(&local->conn_ids, conn, id) {
61 pr_err("AF_RXRPC: Leaked client conn %p {%d}\n",
62 conn, refcount_read(&conn->ref));
63 }
64 BUG();
65 }
66
67 idr_destroy(&local->conn_ids);
68}
69
70/*
71 * Allocate a connection bundle.
72 */
73static struct rxrpc_bundle *rxrpc_alloc_bundle(struct rxrpc_call *call,
74 gfp_t gfp)
75{
76 static atomic_t rxrpc_bundle_id;
77 struct rxrpc_bundle *bundle;
78
79 bundle = kzalloc(sizeof(*bundle), gfp);
80 if (bundle) {
81 bundle->local = call->local;
82 bundle->peer = rxrpc_get_peer(call->peer, rxrpc_peer_get_bundle);
83 bundle->key = key_get(call->key);
84 bundle->security = call->security;
85 bundle->exclusive = test_bit(RXRPC_CALL_EXCLUSIVE, &call->flags);
86 bundle->upgrade = test_bit(RXRPC_CALL_UPGRADE, &call->flags);
87 bundle->service_id = call->dest_srx.srx_service;
88 bundle->security_level = call->security_level;
89 bundle->debug_id = atomic_inc_return(&rxrpc_bundle_id);
90 refcount_set(&bundle->ref, 1);
91 atomic_set(&bundle->active, 1);
92 INIT_LIST_HEAD(&bundle->waiting_calls);
93 trace_rxrpc_bundle(bundle->debug_id, 1, rxrpc_bundle_new);
94
95 write_lock(&bundle->local->rxnet->conn_lock);
96 list_add_tail(&bundle->proc_link, &bundle->local->rxnet->bundle_proc_list);
97 write_unlock(&bundle->local->rxnet->conn_lock);
98 }
99 return bundle;
100}
101
102struct rxrpc_bundle *rxrpc_get_bundle(struct rxrpc_bundle *bundle,
103 enum rxrpc_bundle_trace why)
104{
105 int r;
106
107 __refcount_inc(&bundle->ref, &r);
108 trace_rxrpc_bundle(bundle->debug_id, r + 1, why);
109 return bundle;
110}
111
112static void rxrpc_free_bundle(struct rxrpc_bundle *bundle)
113{
114 trace_rxrpc_bundle(bundle->debug_id, refcount_read(&bundle->ref),
115 rxrpc_bundle_free);
116 write_lock(&bundle->local->rxnet->conn_lock);
117 list_del(&bundle->proc_link);
118 write_unlock(&bundle->local->rxnet->conn_lock);
119 rxrpc_put_peer(bundle->peer, rxrpc_peer_put_bundle);
120 key_put(bundle->key);
121 kfree(bundle);
122}
123
124void rxrpc_put_bundle(struct rxrpc_bundle *bundle, enum rxrpc_bundle_trace why)
125{
126 unsigned int id;
127 bool dead;
128 int r;
129
130 if (bundle) {
131 id = bundle->debug_id;
132 dead = __refcount_dec_and_test(&bundle->ref, &r);
133 trace_rxrpc_bundle(id, r - 1, why);
134 if (dead)
135 rxrpc_free_bundle(bundle);
136 }
137}
138
139/*
140 * Get rid of outstanding client connection preallocations when a local
141 * endpoint is destroyed.
142 */
143void rxrpc_purge_client_connections(struct rxrpc_local *local)
144{
145 rxrpc_destroy_client_conn_ids(local);
146}
147
148/*
149 * Allocate a client connection.
150 */
151static struct rxrpc_connection *
152rxrpc_alloc_client_connection(struct rxrpc_bundle *bundle)
153{
154 struct rxrpc_connection *conn;
155 struct rxrpc_local *local = bundle->local;
156 struct rxrpc_net *rxnet = local->rxnet;
157 int id;
158
159 _enter("");
160
161 conn = rxrpc_alloc_connection(rxnet, GFP_ATOMIC | __GFP_NOWARN);
162 if (!conn)
163 return ERR_PTR(-ENOMEM);
164
165 id = idr_alloc_cyclic(&local->conn_ids, conn, 1, 0x40000000,
166 GFP_ATOMIC | __GFP_NOWARN);
167 if (id < 0) {
168 kfree(conn);
169 return ERR_PTR(id);
170 }
171
172 refcount_set(&conn->ref, 1);
173 conn->proto.cid = id << RXRPC_CIDSHIFT;
174 conn->proto.epoch = local->rxnet->epoch;
175 conn->out_clientflag = RXRPC_CLIENT_INITIATED;
176 conn->bundle = rxrpc_get_bundle(bundle, rxrpc_bundle_get_client_conn);
177 conn->local = rxrpc_get_local(bundle->local, rxrpc_local_get_client_conn);
178 conn->peer = rxrpc_get_peer(bundle->peer, rxrpc_peer_get_client_conn);
179 conn->key = key_get(bundle->key);
180 conn->security = bundle->security;
181 conn->exclusive = bundle->exclusive;
182 conn->upgrade = bundle->upgrade;
183 conn->orig_service_id = bundle->service_id;
184 conn->security_level = bundle->security_level;
185 conn->state = RXRPC_CONN_CLIENT_UNSECURED;
186 conn->service_id = conn->orig_service_id;
187
188 if (conn->security == &rxrpc_no_security)
189 conn->state = RXRPC_CONN_CLIENT;
190
191 atomic_inc(&rxnet->nr_conns);
192 write_lock(&rxnet->conn_lock);
193 list_add_tail(&conn->proc_link, &rxnet->conn_proc_list);
194 write_unlock(&rxnet->conn_lock);
195
196 rxrpc_see_connection(conn, rxrpc_conn_new_client);
197
198 atomic_inc(&rxnet->nr_client_conns);
199 trace_rxrpc_client(conn, -1, rxrpc_client_alloc);
200 return conn;
201}
202
203/*
204 * Determine if a connection may be reused.
205 */
206static bool rxrpc_may_reuse_conn(struct rxrpc_connection *conn)
207{
208 struct rxrpc_net *rxnet;
209 int id_cursor, id, distance, limit;
210
211 if (!conn)
212 goto dont_reuse;
213
214 rxnet = conn->rxnet;
215 if (test_bit(RXRPC_CONN_DONT_REUSE, &conn->flags))
216 goto dont_reuse;
217
218 if ((conn->state != RXRPC_CONN_CLIENT_UNSECURED &&
219 conn->state != RXRPC_CONN_CLIENT) ||
220 conn->proto.epoch != rxnet->epoch)
221 goto mark_dont_reuse;
222
223 /* The IDR tree gets very expensive on memory if the connection IDs are
224 * widely scattered throughout the number space, so we shall want to
225 * kill off connections that, say, have an ID more than about four
226 * times the maximum number of client conns away from the current
227 * allocation point to try and keep the IDs concentrated.
228 */
229 id_cursor = idr_get_cursor(&conn->local->conn_ids);
230 id = conn->proto.cid >> RXRPC_CIDSHIFT;
231 distance = id - id_cursor;
232 if (distance < 0)
233 distance = -distance;
234 limit = max_t(unsigned long, atomic_read(&rxnet->nr_conns) * 4, 1024);
235 if (distance > limit)
236 goto mark_dont_reuse;
237
238 return true;
239
240mark_dont_reuse:
241 set_bit(RXRPC_CONN_DONT_REUSE, &conn->flags);
242dont_reuse:
243 return false;
244}
245
246/*
247 * Look up the conn bundle that matches the connection parameters, adding it if
248 * it doesn't yet exist.
249 */
250int rxrpc_look_up_bundle(struct rxrpc_call *call, gfp_t gfp)
251{
252 struct rxrpc_bundle *bundle, *candidate;
253 struct rxrpc_local *local = call->local;
254 struct rb_node *p, **pp, *parent;
255 long diff;
256 bool upgrade = test_bit(RXRPC_CALL_UPGRADE, &call->flags);
257
258 _enter("{%px,%x,%u,%u}",
259 call->peer, key_serial(call->key), call->security_level,
260 upgrade);
261
262 if (test_bit(RXRPC_CALL_EXCLUSIVE, &call->flags)) {
263 call->bundle = rxrpc_alloc_bundle(call, gfp);
264 return call->bundle ? 0 : -ENOMEM;
265 }
266
267 /* First, see if the bundle is already there. */
268 _debug("search 1");
269 spin_lock(&local->client_bundles_lock);
270 p = local->client_bundles.rb_node;
271 while (p) {
272 bundle = rb_entry(p, struct rxrpc_bundle, local_node);
273
274#define cmp(X, Y) ((long)(X) - (long)(Y))
275 diff = (cmp(bundle->peer, call->peer) ?:
276 cmp(bundle->key, call->key) ?:
277 cmp(bundle->security_level, call->security_level) ?:
278 cmp(bundle->upgrade, upgrade));
279#undef cmp
280 if (diff < 0)
281 p = p->rb_left;
282 else if (diff > 0)
283 p = p->rb_right;
284 else
285 goto found_bundle;
286 }
287 spin_unlock(&local->client_bundles_lock);
288 _debug("not found");
289
290 /* It wasn't. We need to add one. */
291 candidate = rxrpc_alloc_bundle(call, gfp);
292 if (!candidate)
293 return -ENOMEM;
294
295 _debug("search 2");
296 spin_lock(&local->client_bundles_lock);
297 pp = &local->client_bundles.rb_node;
298 parent = NULL;
299 while (*pp) {
300 parent = *pp;
301 bundle = rb_entry(parent, struct rxrpc_bundle, local_node);
302
303#define cmp(X, Y) ((long)(X) - (long)(Y))
304 diff = (cmp(bundle->peer, call->peer) ?:
305 cmp(bundle->key, call->key) ?:
306 cmp(bundle->security_level, call->security_level) ?:
307 cmp(bundle->upgrade, upgrade));
308#undef cmp
309 if (diff < 0)
310 pp = &(*pp)->rb_left;
311 else if (diff > 0)
312 pp = &(*pp)->rb_right;
313 else
314 goto found_bundle_free;
315 }
316
317 _debug("new bundle");
318 rb_link_node(&candidate->local_node, parent, pp);
319 rb_insert_color(&candidate->local_node, &local->client_bundles);
320 call->bundle = rxrpc_get_bundle(candidate, rxrpc_bundle_get_client_call);
321 spin_unlock(&local->client_bundles_lock);
322 _leave(" = B=%u [new]", call->bundle->debug_id);
323 return 0;
324
325found_bundle_free:
326 rxrpc_free_bundle(candidate);
327found_bundle:
328 call->bundle = rxrpc_get_bundle(bundle, rxrpc_bundle_get_client_call);
329 rxrpc_activate_bundle(bundle);
330 spin_unlock(&local->client_bundles_lock);
331 _leave(" = B=%u [found]", call->bundle->debug_id);
332 return 0;
333}
334
335/*
336 * Allocate a new connection and add it into a bundle.
337 */
338static bool rxrpc_add_conn_to_bundle(struct rxrpc_bundle *bundle,
339 unsigned int slot)
340{
341 struct rxrpc_connection *conn, *old;
342 unsigned int shift = slot * RXRPC_MAXCALLS;
343 unsigned int i;
344
345 old = bundle->conns[slot];
346 if (old) {
347 bundle->conns[slot] = NULL;
348 bundle->conn_ids[slot] = 0;
349 trace_rxrpc_client(old, -1, rxrpc_client_replace);
350 rxrpc_put_connection(old, rxrpc_conn_put_noreuse);
351 }
352
353 conn = rxrpc_alloc_client_connection(bundle);
354 if (IS_ERR(conn)) {
355 bundle->alloc_error = PTR_ERR(conn);
356 return false;
357 }
358
359 rxrpc_activate_bundle(bundle);
360 conn->bundle_shift = shift;
361 bundle->conns[slot] = conn;
362 bundle->conn_ids[slot] = conn->debug_id;
363 for (i = 0; i < RXRPC_MAXCALLS; i++)
364 set_bit(shift + i, &bundle->avail_chans);
365 return true;
366}
367
368/*
369 * Add a connection to a bundle if there are no usable connections or we have
370 * connections waiting for extra capacity.
371 */
372static bool rxrpc_bundle_has_space(struct rxrpc_bundle *bundle)
373{
374 int slot = -1, i, usable;
375
376 _enter("");
377
378 bundle->alloc_error = 0;
379
380 /* See if there are any usable connections. */
381 usable = 0;
382 for (i = 0; i < ARRAY_SIZE(bundle->conns); i++) {
383 if (rxrpc_may_reuse_conn(bundle->conns[i]))
384 usable++;
385 else if (slot == -1)
386 slot = i;
387 }
388
389 if (!usable && bundle->upgrade)
390 bundle->try_upgrade = true;
391
392 if (!usable)
393 goto alloc_conn;
394
395 if (!bundle->avail_chans &&
396 !bundle->try_upgrade &&
397 usable < ARRAY_SIZE(bundle->conns))
398 goto alloc_conn;
399
400 _leave("");
401 return usable;
402
403alloc_conn:
404 return slot >= 0 ? rxrpc_add_conn_to_bundle(bundle, slot) : false;
405}
406
407/*
408 * Assign a channel to the call at the front of the queue and wake the call up.
409 * We don't increment the callNumber counter until this number has been exposed
410 * to the world.
411 */
412static void rxrpc_activate_one_channel(struct rxrpc_connection *conn,
413 unsigned int channel)
414{
415 struct rxrpc_channel *chan = &conn->channels[channel];
416 struct rxrpc_bundle *bundle = conn->bundle;
417 struct rxrpc_call *call = list_entry(bundle->waiting_calls.next,
418 struct rxrpc_call, wait_link);
419 u32 call_id = chan->call_counter + 1;
420
421 _enter("C=%x,%u", conn->debug_id, channel);
422
423 list_del_init(&call->wait_link);
424
425 trace_rxrpc_client(conn, channel, rxrpc_client_chan_activate);
426
427 /* Cancel the final ACK on the previous call if it hasn't been sent yet
428 * as the DATA packet will implicitly ACK it.
429 */
430 clear_bit(RXRPC_CONN_FINAL_ACK_0 + channel, &conn->flags);
431 clear_bit(conn->bundle_shift + channel, &bundle->avail_chans);
432
433 rxrpc_see_call(call, rxrpc_call_see_activate_client);
434 call->conn = rxrpc_get_connection(conn, rxrpc_conn_get_activate_call);
435 call->cid = conn->proto.cid | channel;
436 call->call_id = call_id;
437 call->dest_srx.srx_service = conn->service_id;
438 call->cong_ssthresh = call->peer->cong_ssthresh;
439 if (call->cong_cwnd >= call->cong_ssthresh)
440 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
441 else
442 call->cong_mode = RXRPC_CALL_SLOW_START;
443
444 chan->call_id = call_id;
445 chan->call_debug_id = call->debug_id;
446 chan->call = call;
447
448 rxrpc_see_call(call, rxrpc_call_see_connected);
449 trace_rxrpc_connect_call(call);
450 call->tx_last_sent = ktime_get_real();
451 rxrpc_start_call_timer(call);
452 rxrpc_set_call_state(call, RXRPC_CALL_CLIENT_SEND_REQUEST);
453 wake_up(&call->waitq);
454}
455
456/*
457 * Remove a connection from the idle list if it's on it.
458 */
459static void rxrpc_unidle_conn(struct rxrpc_connection *conn)
460{
461 if (!list_empty(&conn->cache_link)) {
462 list_del_init(&conn->cache_link);
463 rxrpc_put_connection(conn, rxrpc_conn_put_unidle);
464 }
465}
466
467/*
468 * Assign channels and callNumbers to waiting calls.
469 */
470static void rxrpc_activate_channels(struct rxrpc_bundle *bundle)
471{
472 struct rxrpc_connection *conn;
473 unsigned long avail, mask;
474 unsigned int channel, slot;
475
476 trace_rxrpc_client(NULL, -1, rxrpc_client_activate_chans);
477
478 if (bundle->try_upgrade)
479 mask = 1;
480 else
481 mask = ULONG_MAX;
482
483 while (!list_empty(&bundle->waiting_calls)) {
484 avail = bundle->avail_chans & mask;
485 if (!avail)
486 break;
487 channel = __ffs(avail);
488 clear_bit(channel, &bundle->avail_chans);
489
490 slot = channel / RXRPC_MAXCALLS;
491 conn = bundle->conns[slot];
492 if (!conn)
493 break;
494
495 if (bundle->try_upgrade)
496 set_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags);
497 rxrpc_unidle_conn(conn);
498
499 channel &= (RXRPC_MAXCALLS - 1);
500 conn->act_chans |= 1 << channel;
501 rxrpc_activate_one_channel(conn, channel);
502 }
503}
504
505/*
506 * Connect waiting channels (called from the I/O thread).
507 */
508void rxrpc_connect_client_calls(struct rxrpc_local *local)
509{
510 struct rxrpc_call *call;
511
512 while ((call = list_first_entry_or_null(&local->new_client_calls,
513 struct rxrpc_call, wait_link))
514 ) {
515 struct rxrpc_bundle *bundle = call->bundle;
516
517 spin_lock(&local->client_call_lock);
518 list_move_tail(&call->wait_link, &bundle->waiting_calls);
519 rxrpc_see_call(call, rxrpc_call_see_waiting_call);
520 spin_unlock(&local->client_call_lock);
521
522 if (rxrpc_bundle_has_space(bundle))
523 rxrpc_activate_channels(bundle);
524 }
525}
526
527/*
528 * Note that a call, and thus a connection, is about to be exposed to the
529 * world.
530 */
531void rxrpc_expose_client_call(struct rxrpc_call *call)
532{
533 unsigned int channel = call->cid & RXRPC_CHANNELMASK;
534 struct rxrpc_connection *conn = call->conn;
535 struct rxrpc_channel *chan = &conn->channels[channel];
536
537 if (!test_and_set_bit(RXRPC_CALL_EXPOSED, &call->flags)) {
538 /* Mark the call ID as being used. If the callNumber counter
539 * exceeds ~2 billion, we kill the connection after its
540 * outstanding calls have finished so that the counter doesn't
541 * wrap.
542 */
543 chan->call_counter++;
544 if (chan->call_counter >= INT_MAX)
545 set_bit(RXRPC_CONN_DONT_REUSE, &conn->flags);
546 trace_rxrpc_client(conn, channel, rxrpc_client_exposed);
547
548 spin_lock(&call->peer->lock);
549 hlist_add_head(&call->error_link, &call->peer->error_targets);
550 spin_unlock(&call->peer->lock);
551 }
552}
553
554/*
555 * Set the reap timer.
556 */
557static void rxrpc_set_client_reap_timer(struct rxrpc_local *local)
558{
559 if (!local->kill_all_client_conns) {
560 unsigned long now = jiffies;
561 unsigned long reap_at = now + rxrpc_conn_idle_client_expiry;
562
563 if (local->rxnet->live)
564 timer_reduce(&local->client_conn_reap_timer, reap_at);
565 }
566}
567
568/*
569 * Disconnect a client call.
570 */
571void rxrpc_disconnect_client_call(struct rxrpc_bundle *bundle, struct rxrpc_call *call)
572{
573 struct rxrpc_connection *conn;
574 struct rxrpc_channel *chan = NULL;
575 struct rxrpc_local *local = bundle->local;
576 unsigned int channel;
577 bool may_reuse;
578 u32 cid;
579
580 _enter("c=%x", call->debug_id);
581
582 /* Calls that have never actually been assigned a channel can simply be
583 * discarded.
584 */
585 conn = call->conn;
586 if (!conn) {
587 _debug("call is waiting");
588 ASSERTCMP(call->call_id, ==, 0);
589 ASSERT(!test_bit(RXRPC_CALL_EXPOSED, &call->flags));
590 /* May still be on ->new_client_calls. */
591 spin_lock(&local->client_call_lock);
592 list_del_init(&call->wait_link);
593 spin_unlock(&local->client_call_lock);
594 return;
595 }
596
597 cid = call->cid;
598 channel = cid & RXRPC_CHANNELMASK;
599 chan = &conn->channels[channel];
600 trace_rxrpc_client(conn, channel, rxrpc_client_chan_disconnect);
601
602 if (WARN_ON(chan->call != call))
603 return;
604
605 may_reuse = rxrpc_may_reuse_conn(conn);
606
607 /* If a client call was exposed to the world, we save the result for
608 * retransmission.
609 *
610 * We use a barrier here so that the call number and abort code can be
611 * read without needing to take a lock.
612 *
613 * TODO: Make the incoming packet handler check this and handle
614 * terminal retransmission without requiring access to the call.
615 */
616 if (test_bit(RXRPC_CALL_EXPOSED, &call->flags)) {
617 _debug("exposed %u,%u", call->call_id, call->abort_code);
618 __rxrpc_disconnect_call(conn, call);
619
620 if (test_and_clear_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags)) {
621 trace_rxrpc_client(conn, channel, rxrpc_client_to_active);
622 bundle->try_upgrade = false;
623 if (may_reuse)
624 rxrpc_activate_channels(bundle);
625 }
626 }
627
628 /* See if we can pass the channel directly to another call. */
629 if (may_reuse && !list_empty(&bundle->waiting_calls)) {
630 trace_rxrpc_client(conn, channel, rxrpc_client_chan_pass);
631 rxrpc_activate_one_channel(conn, channel);
632 return;
633 }
634
635 /* Schedule the final ACK to be transmitted in a short while so that it
636 * can be skipped if we find a follow-on call. The first DATA packet
637 * of the follow on call will implicitly ACK this call.
638 */
639 if (call->completion == RXRPC_CALL_SUCCEEDED &&
640 test_bit(RXRPC_CALL_EXPOSED, &call->flags)) {
641 unsigned long final_ack_at = jiffies + 2;
642
643 chan->final_ack_at = final_ack_at;
644 smp_wmb(); /* vs rxrpc_process_delayed_final_acks() */
645 set_bit(RXRPC_CONN_FINAL_ACK_0 + channel, &conn->flags);
646 rxrpc_reduce_conn_timer(conn, final_ack_at);
647 }
648
649 /* Deactivate the channel. */
650 chan->call = NULL;
651 set_bit(conn->bundle_shift + channel, &conn->bundle->avail_chans);
652 conn->act_chans &= ~(1 << channel);
653
654 /* If no channels remain active, then put the connection on the idle
655 * list for a short while. Give it a ref to stop it going away if it
656 * becomes unbundled.
657 */
658 if (!conn->act_chans) {
659 trace_rxrpc_client(conn, channel, rxrpc_client_to_idle);
660 conn->idle_timestamp = jiffies;
661
662 rxrpc_get_connection(conn, rxrpc_conn_get_idle);
663 list_move_tail(&conn->cache_link, &local->idle_client_conns);
664
665 rxrpc_set_client_reap_timer(local);
666 }
667}
668
669/*
670 * Remove a connection from a bundle.
671 */
672static void rxrpc_unbundle_conn(struct rxrpc_connection *conn)
673{
674 struct rxrpc_bundle *bundle = conn->bundle;
675 unsigned int bindex;
676 int i;
677
678 _enter("C=%x", conn->debug_id);
679
680 if (conn->flags & RXRPC_CONN_FINAL_ACK_MASK)
681 rxrpc_process_delayed_final_acks(conn, true);
682
683 bindex = conn->bundle_shift / RXRPC_MAXCALLS;
684 if (bundle->conns[bindex] == conn) {
685 _debug("clear slot %u", bindex);
686 bundle->conns[bindex] = NULL;
687 bundle->conn_ids[bindex] = 0;
688 for (i = 0; i < RXRPC_MAXCALLS; i++)
689 clear_bit(conn->bundle_shift + i, &bundle->avail_chans);
690 rxrpc_put_client_connection_id(bundle->local, conn);
691 rxrpc_deactivate_bundle(bundle);
692 rxrpc_put_connection(conn, rxrpc_conn_put_unbundle);
693 }
694}
695
696/*
697 * Drop the active count on a bundle.
698 */
699void rxrpc_deactivate_bundle(struct rxrpc_bundle *bundle)
700{
701 struct rxrpc_local *local;
702 bool need_put = false;
703
704 if (!bundle)
705 return;
706
707 local = bundle->local;
708 if (atomic_dec_and_lock(&bundle->active, &local->client_bundles_lock)) {
709 if (!bundle->exclusive) {
710 _debug("erase bundle");
711 rb_erase(&bundle->local_node, &local->client_bundles);
712 need_put = true;
713 }
714
715 spin_unlock(&local->client_bundles_lock);
716 if (need_put)
717 rxrpc_put_bundle(bundle, rxrpc_bundle_put_discard);
718 }
719}
720
721/*
722 * Clean up a dead client connection.
723 */
724void rxrpc_kill_client_conn(struct rxrpc_connection *conn)
725{
726 struct rxrpc_local *local = conn->local;
727 struct rxrpc_net *rxnet = local->rxnet;
728
729 _enter("C=%x", conn->debug_id);
730
731 trace_rxrpc_client(conn, -1, rxrpc_client_cleanup);
732 atomic_dec(&rxnet->nr_client_conns);
733
734 rxrpc_put_client_connection_id(local, conn);
735}
736
737/*
738 * Discard expired client connections from the idle list. Each conn in the
739 * idle list has been exposed and holds an extra ref because of that.
740 *
741 * This may be called from conn setup or from a work item so cannot be
742 * considered non-reentrant.
743 */
744void rxrpc_discard_expired_client_conns(struct rxrpc_local *local)
745{
746 struct rxrpc_connection *conn;
747 unsigned long expiry, conn_expires_at, now;
748 unsigned int nr_conns;
749
750 _enter("");
751
752 /* We keep an estimate of what the number of conns ought to be after
753 * we've discarded some so that we don't overdo the discarding.
754 */
755 nr_conns = atomic_read(&local->rxnet->nr_client_conns);
756
757next:
758 conn = list_first_entry_or_null(&local->idle_client_conns,
759 struct rxrpc_connection, cache_link);
760 if (!conn)
761 return;
762
763 if (!local->kill_all_client_conns) {
764 /* If the number of connections is over the reap limit, we
765 * expedite discard by reducing the expiry timeout. We must,
766 * however, have at least a short grace period to be able to do
767 * final-ACK or ABORT retransmission.
768 */
769 expiry = rxrpc_conn_idle_client_expiry;
770 if (nr_conns > rxrpc_reap_client_connections)
771 expiry = rxrpc_conn_idle_client_fast_expiry;
772 if (conn->local->service_closed)
773 expiry = rxrpc_closed_conn_expiry * HZ;
774
775 conn_expires_at = conn->idle_timestamp + expiry;
776
777 now = jiffies;
778 if (time_after(conn_expires_at, now))
779 goto not_yet_expired;
780 }
781
782 atomic_dec(&conn->active);
783 trace_rxrpc_client(conn, -1, rxrpc_client_discard);
784 list_del_init(&conn->cache_link);
785
786 rxrpc_unbundle_conn(conn);
787 /* Drop the ->cache_link ref */
788 rxrpc_put_connection(conn, rxrpc_conn_put_discard_idle);
789
790 nr_conns--;
791 goto next;
792
793not_yet_expired:
794 /* The connection at the front of the queue hasn't yet expired, so
795 * schedule the work item for that point if we discarded something.
796 *
797 * We don't worry if the work item is already scheduled - it can look
798 * after rescheduling itself at a later time. We could cancel it, but
799 * then things get messier.
800 */
801 _debug("not yet");
802 if (!local->kill_all_client_conns)
803 timer_reduce(&local->client_conn_reap_timer, conn_expires_at);
804
805 _leave("");
806}
807
808/*
809 * Clean up the client connections on a local endpoint.
810 */
811void rxrpc_clean_up_local_conns(struct rxrpc_local *local)
812{
813 struct rxrpc_connection *conn;
814
815 _enter("");
816
817 local->kill_all_client_conns = true;
818
819 del_timer_sync(&local->client_conn_reap_timer);
820
821 while ((conn = list_first_entry_or_null(&local->idle_client_conns,
822 struct rxrpc_connection, cache_link))) {
823 list_del_init(&conn->cache_link);
824 atomic_dec(&conn->active);
825 trace_rxrpc_client(conn, -1, rxrpc_client_discard);
826 rxrpc_unbundle_conn(conn);
827 rxrpc_put_connection(conn, rxrpc_conn_put_local_dead);
828 }
829
830 _leave(" [culled]");
831}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* Client connection-specific management code.
3 *
4 * Copyright (C) 2016, 2020 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
6 *
7 * Client connections need to be cached for a little while after they've made a
8 * call so as to handle retransmitted DATA packets in case the server didn't
9 * receive the final ACK or terminating ABORT we sent it.
10 *
11 * There are flags of relevance to the cache:
12 *
13 * (2) DONT_REUSE - The connection should be discarded as soon as possible and
14 * should not be reused. This is set when an exclusive connection is used
15 * or a call ID counter overflows.
16 *
17 * The caching state may only be changed if the cache lock is held.
18 *
19 * There are two idle client connection expiry durations. If the total number
20 * of connections is below the reap threshold, we use the normal duration; if
21 * it's above, we use the fast duration.
22 */
23
24#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
25
26#include <linux/slab.h>
27#include <linux/idr.h>
28#include <linux/timer.h>
29#include <linux/sched/signal.h>
30
31#include "ar-internal.h"
32
33__read_mostly unsigned int rxrpc_reap_client_connections = 900;
34__read_mostly unsigned long rxrpc_conn_idle_client_expiry = 2 * 60 * HZ;
35__read_mostly unsigned long rxrpc_conn_idle_client_fast_expiry = 2 * HZ;
36
37/*
38 * We use machine-unique IDs for our client connections.
39 */
40DEFINE_IDR(rxrpc_client_conn_ids);
41static DEFINE_SPINLOCK(rxrpc_conn_id_lock);
42
43/*
44 * Get a connection ID and epoch for a client connection from the global pool.
45 * The connection struct pointer is then recorded in the idr radix tree. The
46 * epoch doesn't change until the client is rebooted (or, at least, unless the
47 * module is unloaded).
48 */
49static int rxrpc_get_client_connection_id(struct rxrpc_connection *conn,
50 gfp_t gfp)
51{
52 struct rxrpc_net *rxnet = conn->params.local->rxnet;
53 int id;
54
55 _enter("");
56
57 idr_preload(gfp);
58 spin_lock(&rxrpc_conn_id_lock);
59
60 id = idr_alloc_cyclic(&rxrpc_client_conn_ids, conn,
61 1, 0x40000000, GFP_NOWAIT);
62 if (id < 0)
63 goto error;
64
65 spin_unlock(&rxrpc_conn_id_lock);
66 idr_preload_end();
67
68 conn->proto.epoch = rxnet->epoch;
69 conn->proto.cid = id << RXRPC_CIDSHIFT;
70 set_bit(RXRPC_CONN_HAS_IDR, &conn->flags);
71 _leave(" [CID %x]", conn->proto.cid);
72 return 0;
73
74error:
75 spin_unlock(&rxrpc_conn_id_lock);
76 idr_preload_end();
77 _leave(" = %d", id);
78 return id;
79}
80
81/*
82 * Release a connection ID for a client connection from the global pool.
83 */
84static void rxrpc_put_client_connection_id(struct rxrpc_connection *conn)
85{
86 if (test_bit(RXRPC_CONN_HAS_IDR, &conn->flags)) {
87 spin_lock(&rxrpc_conn_id_lock);
88 idr_remove(&rxrpc_client_conn_ids,
89 conn->proto.cid >> RXRPC_CIDSHIFT);
90 spin_unlock(&rxrpc_conn_id_lock);
91 }
92}
93
94/*
95 * Destroy the client connection ID tree.
96 */
97void rxrpc_destroy_client_conn_ids(void)
98{
99 struct rxrpc_connection *conn;
100 int id;
101
102 if (!idr_is_empty(&rxrpc_client_conn_ids)) {
103 idr_for_each_entry(&rxrpc_client_conn_ids, conn, id) {
104 pr_err("AF_RXRPC: Leaked client conn %p {%d}\n",
105 conn, atomic_read(&conn->usage));
106 }
107 BUG();
108 }
109
110 idr_destroy(&rxrpc_client_conn_ids);
111}
112
113/*
114 * Allocate a connection bundle.
115 */
116static struct rxrpc_bundle *rxrpc_alloc_bundle(struct rxrpc_conn_parameters *cp,
117 gfp_t gfp)
118{
119 struct rxrpc_bundle *bundle;
120
121 bundle = kzalloc(sizeof(*bundle), gfp);
122 if (bundle) {
123 bundle->params = *cp;
124 rxrpc_get_peer(bundle->params.peer);
125 atomic_set(&bundle->usage, 1);
126 spin_lock_init(&bundle->channel_lock);
127 INIT_LIST_HEAD(&bundle->waiting_calls);
128 }
129 return bundle;
130}
131
132struct rxrpc_bundle *rxrpc_get_bundle(struct rxrpc_bundle *bundle)
133{
134 atomic_inc(&bundle->usage);
135 return bundle;
136}
137
138void rxrpc_put_bundle(struct rxrpc_bundle *bundle)
139{
140 unsigned int d = bundle->debug_id;
141 unsigned int u = atomic_dec_return(&bundle->usage);
142
143 _debug("PUT B=%x %u", d, u);
144 if (u == 0) {
145 rxrpc_put_peer(bundle->params.peer);
146 kfree(bundle);
147 }
148}
149
150/*
151 * Allocate a client connection.
152 */
153static struct rxrpc_connection *
154rxrpc_alloc_client_connection(struct rxrpc_bundle *bundle, gfp_t gfp)
155{
156 struct rxrpc_connection *conn;
157 struct rxrpc_net *rxnet = bundle->params.local->rxnet;
158 int ret;
159
160 _enter("");
161
162 conn = rxrpc_alloc_connection(gfp);
163 if (!conn) {
164 _leave(" = -ENOMEM");
165 return ERR_PTR(-ENOMEM);
166 }
167
168 atomic_set(&conn->usage, 1);
169 conn->bundle = bundle;
170 conn->params = bundle->params;
171 conn->out_clientflag = RXRPC_CLIENT_INITIATED;
172 conn->state = RXRPC_CONN_CLIENT;
173 conn->service_id = conn->params.service_id;
174
175 ret = rxrpc_get_client_connection_id(conn, gfp);
176 if (ret < 0)
177 goto error_0;
178
179 ret = rxrpc_init_client_conn_security(conn);
180 if (ret < 0)
181 goto error_1;
182
183 atomic_inc(&rxnet->nr_conns);
184 write_lock(&rxnet->conn_lock);
185 list_add_tail(&conn->proc_link, &rxnet->conn_proc_list);
186 write_unlock(&rxnet->conn_lock);
187
188 rxrpc_get_bundle(bundle);
189 rxrpc_get_peer(conn->params.peer);
190 rxrpc_get_local(conn->params.local);
191 key_get(conn->params.key);
192
193 trace_rxrpc_conn(conn->debug_id, rxrpc_conn_new_client,
194 atomic_read(&conn->usage),
195 __builtin_return_address(0));
196
197 atomic_inc(&rxnet->nr_client_conns);
198 trace_rxrpc_client(conn, -1, rxrpc_client_alloc);
199 _leave(" = %p", conn);
200 return conn;
201
202error_1:
203 rxrpc_put_client_connection_id(conn);
204error_0:
205 kfree(conn);
206 _leave(" = %d", ret);
207 return ERR_PTR(ret);
208}
209
210/*
211 * Determine if a connection may be reused.
212 */
213static bool rxrpc_may_reuse_conn(struct rxrpc_connection *conn)
214{
215 struct rxrpc_net *rxnet;
216 int id_cursor, id, distance, limit;
217
218 if (!conn)
219 goto dont_reuse;
220
221 rxnet = conn->params.local->rxnet;
222 if (test_bit(RXRPC_CONN_DONT_REUSE, &conn->flags))
223 goto dont_reuse;
224
225 if (conn->state != RXRPC_CONN_CLIENT ||
226 conn->proto.epoch != rxnet->epoch)
227 goto mark_dont_reuse;
228
229 /* The IDR tree gets very expensive on memory if the connection IDs are
230 * widely scattered throughout the number space, so we shall want to
231 * kill off connections that, say, have an ID more than about four
232 * times the maximum number of client conns away from the current
233 * allocation point to try and keep the IDs concentrated.
234 */
235 id_cursor = idr_get_cursor(&rxrpc_client_conn_ids);
236 id = conn->proto.cid >> RXRPC_CIDSHIFT;
237 distance = id - id_cursor;
238 if (distance < 0)
239 distance = -distance;
240 limit = max_t(unsigned long, atomic_read(&rxnet->nr_conns) * 4, 1024);
241 if (distance > limit)
242 goto mark_dont_reuse;
243
244 return true;
245
246mark_dont_reuse:
247 set_bit(RXRPC_CONN_DONT_REUSE, &conn->flags);
248dont_reuse:
249 return false;
250}
251
252/*
253 * Look up the conn bundle that matches the connection parameters, adding it if
254 * it doesn't yet exist.
255 */
256static struct rxrpc_bundle *rxrpc_look_up_bundle(struct rxrpc_conn_parameters *cp,
257 gfp_t gfp)
258{
259 static atomic_t rxrpc_bundle_id;
260 struct rxrpc_bundle *bundle, *candidate;
261 struct rxrpc_local *local = cp->local;
262 struct rb_node *p, **pp, *parent;
263 long diff;
264
265 _enter("{%px,%x,%u,%u}",
266 cp->peer, key_serial(cp->key), cp->security_level, cp->upgrade);
267
268 if (cp->exclusive)
269 return rxrpc_alloc_bundle(cp, gfp);
270
271 /* First, see if the bundle is already there. */
272 _debug("search 1");
273 spin_lock(&local->client_bundles_lock);
274 p = local->client_bundles.rb_node;
275 while (p) {
276 bundle = rb_entry(p, struct rxrpc_bundle, local_node);
277
278#define cmp(X) ((long)bundle->params.X - (long)cp->X)
279 diff = (cmp(peer) ?:
280 cmp(key) ?:
281 cmp(security_level) ?:
282 cmp(upgrade));
283#undef cmp
284 if (diff < 0)
285 p = p->rb_left;
286 else if (diff > 0)
287 p = p->rb_right;
288 else
289 goto found_bundle;
290 }
291 spin_unlock(&local->client_bundles_lock);
292 _debug("not found");
293
294 /* It wasn't. We need to add one. */
295 candidate = rxrpc_alloc_bundle(cp, gfp);
296 if (!candidate)
297 return NULL;
298
299 _debug("search 2");
300 spin_lock(&local->client_bundles_lock);
301 pp = &local->client_bundles.rb_node;
302 parent = NULL;
303 while (*pp) {
304 parent = *pp;
305 bundle = rb_entry(parent, struct rxrpc_bundle, local_node);
306
307#define cmp(X) ((long)bundle->params.X - (long)cp->X)
308 diff = (cmp(peer) ?:
309 cmp(key) ?:
310 cmp(security_level) ?:
311 cmp(upgrade));
312#undef cmp
313 if (diff < 0)
314 pp = &(*pp)->rb_left;
315 else if (diff > 0)
316 pp = &(*pp)->rb_right;
317 else
318 goto found_bundle_free;
319 }
320
321 _debug("new bundle");
322 candidate->debug_id = atomic_inc_return(&rxrpc_bundle_id);
323 rb_link_node(&candidate->local_node, parent, pp);
324 rb_insert_color(&candidate->local_node, &local->client_bundles);
325 rxrpc_get_bundle(candidate);
326 spin_unlock(&local->client_bundles_lock);
327 _leave(" = %u [new]", candidate->debug_id);
328 return candidate;
329
330found_bundle_free:
331 kfree(candidate);
332found_bundle:
333 rxrpc_get_bundle(bundle);
334 spin_unlock(&local->client_bundles_lock);
335 _leave(" = %u [found]", bundle->debug_id);
336 return bundle;
337}
338
339/*
340 * Create or find a client bundle to use for a call.
341 *
342 * If we return with a connection, the call will be on its waiting list. It's
343 * left to the caller to assign a channel and wake up the call.
344 */
345static struct rxrpc_bundle *rxrpc_prep_call(struct rxrpc_sock *rx,
346 struct rxrpc_call *call,
347 struct rxrpc_conn_parameters *cp,
348 struct sockaddr_rxrpc *srx,
349 gfp_t gfp)
350{
351 struct rxrpc_bundle *bundle;
352
353 _enter("{%d,%lx},", call->debug_id, call->user_call_ID);
354
355 cp->peer = rxrpc_lookup_peer(rx, cp->local, srx, gfp);
356 if (!cp->peer)
357 goto error;
358
359 call->cong_cwnd = cp->peer->cong_cwnd;
360 if (call->cong_cwnd >= call->cong_ssthresh)
361 call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
362 else
363 call->cong_mode = RXRPC_CALL_SLOW_START;
364 if (cp->upgrade)
365 __set_bit(RXRPC_CALL_UPGRADE, &call->flags);
366
367 /* Find the client connection bundle. */
368 bundle = rxrpc_look_up_bundle(cp, gfp);
369 if (!bundle)
370 goto error;
371
372 /* Get this call queued. Someone else may activate it whilst we're
373 * lining up a new connection, but that's fine.
374 */
375 spin_lock(&bundle->channel_lock);
376 list_add_tail(&call->chan_wait_link, &bundle->waiting_calls);
377 spin_unlock(&bundle->channel_lock);
378
379 _leave(" = [B=%x]", bundle->debug_id);
380 return bundle;
381
382error:
383 _leave(" = -ENOMEM");
384 return ERR_PTR(-ENOMEM);
385}
386
387/*
388 * Allocate a new connection and add it into a bundle.
389 */
390static void rxrpc_add_conn_to_bundle(struct rxrpc_bundle *bundle, gfp_t gfp)
391 __releases(bundle->channel_lock)
392{
393 struct rxrpc_connection *candidate = NULL, *old = NULL;
394 bool conflict;
395 int i;
396
397 _enter("");
398
399 conflict = bundle->alloc_conn;
400 if (!conflict)
401 bundle->alloc_conn = true;
402 spin_unlock(&bundle->channel_lock);
403 if (conflict) {
404 _leave(" [conf]");
405 return;
406 }
407
408 candidate = rxrpc_alloc_client_connection(bundle, gfp);
409
410 spin_lock(&bundle->channel_lock);
411 bundle->alloc_conn = false;
412
413 if (IS_ERR(candidate)) {
414 bundle->alloc_error = PTR_ERR(candidate);
415 spin_unlock(&bundle->channel_lock);
416 _leave(" [err %ld]", PTR_ERR(candidate));
417 return;
418 }
419
420 bundle->alloc_error = 0;
421
422 for (i = 0; i < ARRAY_SIZE(bundle->conns); i++) {
423 unsigned int shift = i * RXRPC_MAXCALLS;
424 int j;
425
426 old = bundle->conns[i];
427 if (!rxrpc_may_reuse_conn(old)) {
428 if (old)
429 trace_rxrpc_client(old, -1, rxrpc_client_replace);
430 candidate->bundle_shift = shift;
431 bundle->conns[i] = candidate;
432 for (j = 0; j < RXRPC_MAXCALLS; j++)
433 set_bit(shift + j, &bundle->avail_chans);
434 candidate = NULL;
435 break;
436 }
437
438 old = NULL;
439 }
440
441 spin_unlock(&bundle->channel_lock);
442
443 if (candidate) {
444 _debug("discard C=%x", candidate->debug_id);
445 trace_rxrpc_client(candidate, -1, rxrpc_client_duplicate);
446 rxrpc_put_connection(candidate);
447 }
448
449 rxrpc_put_connection(old);
450 _leave("");
451}
452
453/*
454 * Add a connection to a bundle if there are no usable connections or we have
455 * connections waiting for extra capacity.
456 */
457static void rxrpc_maybe_add_conn(struct rxrpc_bundle *bundle, gfp_t gfp)
458{
459 struct rxrpc_call *call;
460 int i, usable;
461
462 _enter("");
463
464 spin_lock(&bundle->channel_lock);
465
466 /* See if there are any usable connections. */
467 usable = 0;
468 for (i = 0; i < ARRAY_SIZE(bundle->conns); i++)
469 if (rxrpc_may_reuse_conn(bundle->conns[i]))
470 usable++;
471
472 if (!usable && !list_empty(&bundle->waiting_calls)) {
473 call = list_first_entry(&bundle->waiting_calls,
474 struct rxrpc_call, chan_wait_link);
475 if (test_bit(RXRPC_CALL_UPGRADE, &call->flags))
476 bundle->try_upgrade = true;
477 }
478
479 if (!usable)
480 goto alloc_conn;
481
482 if (!bundle->avail_chans &&
483 !bundle->try_upgrade &&
484 !list_empty(&bundle->waiting_calls) &&
485 usable < ARRAY_SIZE(bundle->conns))
486 goto alloc_conn;
487
488 spin_unlock(&bundle->channel_lock);
489 _leave("");
490 return;
491
492alloc_conn:
493 return rxrpc_add_conn_to_bundle(bundle, gfp);
494}
495
496/*
497 * Assign a channel to the call at the front of the queue and wake the call up.
498 * We don't increment the callNumber counter until this number has been exposed
499 * to the world.
500 */
501static void rxrpc_activate_one_channel(struct rxrpc_connection *conn,
502 unsigned int channel)
503{
504 struct rxrpc_channel *chan = &conn->channels[channel];
505 struct rxrpc_bundle *bundle = conn->bundle;
506 struct rxrpc_call *call = list_entry(bundle->waiting_calls.next,
507 struct rxrpc_call, chan_wait_link);
508 u32 call_id = chan->call_counter + 1;
509
510 _enter("C=%x,%u", conn->debug_id, channel);
511
512 trace_rxrpc_client(conn, channel, rxrpc_client_chan_activate);
513
514 /* Cancel the final ACK on the previous call if it hasn't been sent yet
515 * as the DATA packet will implicitly ACK it.
516 */
517 clear_bit(RXRPC_CONN_FINAL_ACK_0 + channel, &conn->flags);
518 clear_bit(conn->bundle_shift + channel, &bundle->avail_chans);
519
520 rxrpc_see_call(call);
521 list_del_init(&call->chan_wait_link);
522 call->peer = rxrpc_get_peer(conn->params.peer);
523 call->conn = rxrpc_get_connection(conn);
524 call->cid = conn->proto.cid | channel;
525 call->call_id = call_id;
526 call->security = conn->security;
527 call->security_ix = conn->security_ix;
528 call->service_id = conn->service_id;
529
530 trace_rxrpc_connect_call(call);
531 _net("CONNECT call %08x:%08x as call %d on conn %d",
532 call->cid, call->call_id, call->debug_id, conn->debug_id);
533
534 write_lock_bh(&call->state_lock);
535 call->state = RXRPC_CALL_CLIENT_SEND_REQUEST;
536 write_unlock_bh(&call->state_lock);
537
538 /* Paired with the read barrier in rxrpc_connect_call(). This orders
539 * cid and epoch in the connection wrt to call_id without the need to
540 * take the channel_lock.
541 *
542 * We provisionally assign a callNumber at this point, but we don't
543 * confirm it until the call is about to be exposed.
544 *
545 * TODO: Pair with a barrier in the data_ready handler when that looks
546 * at the call ID through a connection channel.
547 */
548 smp_wmb();
549
550 chan->call_id = call_id;
551 chan->call_debug_id = call->debug_id;
552 rcu_assign_pointer(chan->call, call);
553 wake_up(&call->waitq);
554}
555
556/*
557 * Remove a connection from the idle list if it's on it.
558 */
559static void rxrpc_unidle_conn(struct rxrpc_bundle *bundle, struct rxrpc_connection *conn)
560{
561 struct rxrpc_net *rxnet = bundle->params.local->rxnet;
562 bool drop_ref;
563
564 if (!list_empty(&conn->cache_link)) {
565 drop_ref = false;
566 spin_lock(&rxnet->client_conn_cache_lock);
567 if (!list_empty(&conn->cache_link)) {
568 list_del_init(&conn->cache_link);
569 drop_ref = true;
570 }
571 spin_unlock(&rxnet->client_conn_cache_lock);
572 if (drop_ref)
573 rxrpc_put_connection(conn);
574 }
575}
576
577/*
578 * Assign channels and callNumbers to waiting calls with channel_lock
579 * held by caller.
580 */
581static void rxrpc_activate_channels_locked(struct rxrpc_bundle *bundle)
582{
583 struct rxrpc_connection *conn;
584 unsigned long avail, mask;
585 unsigned int channel, slot;
586
587 if (bundle->try_upgrade)
588 mask = 1;
589 else
590 mask = ULONG_MAX;
591
592 while (!list_empty(&bundle->waiting_calls)) {
593 avail = bundle->avail_chans & mask;
594 if (!avail)
595 break;
596 channel = __ffs(avail);
597 clear_bit(channel, &bundle->avail_chans);
598
599 slot = channel / RXRPC_MAXCALLS;
600 conn = bundle->conns[slot];
601 if (!conn)
602 break;
603
604 if (bundle->try_upgrade)
605 set_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags);
606 rxrpc_unidle_conn(bundle, conn);
607
608 channel &= (RXRPC_MAXCALLS - 1);
609 conn->act_chans |= 1 << channel;
610 rxrpc_activate_one_channel(conn, channel);
611 }
612}
613
614/*
615 * Assign channels and callNumbers to waiting calls.
616 */
617static void rxrpc_activate_channels(struct rxrpc_bundle *bundle)
618{
619 _enter("B=%x", bundle->debug_id);
620
621 trace_rxrpc_client(NULL, -1, rxrpc_client_activate_chans);
622
623 if (!bundle->avail_chans)
624 return;
625
626 spin_lock(&bundle->channel_lock);
627 rxrpc_activate_channels_locked(bundle);
628 spin_unlock(&bundle->channel_lock);
629 _leave("");
630}
631
632/*
633 * Wait for a callNumber and a channel to be granted to a call.
634 */
635static int rxrpc_wait_for_channel(struct rxrpc_bundle *bundle,
636 struct rxrpc_call *call, gfp_t gfp)
637{
638 DECLARE_WAITQUEUE(myself, current);
639 int ret = 0;
640
641 _enter("%d", call->debug_id);
642
643 if (!gfpflags_allow_blocking(gfp)) {
644 rxrpc_maybe_add_conn(bundle, gfp);
645 rxrpc_activate_channels(bundle);
646 ret = bundle->alloc_error ?: -EAGAIN;
647 goto out;
648 }
649
650 add_wait_queue_exclusive(&call->waitq, &myself);
651 for (;;) {
652 rxrpc_maybe_add_conn(bundle, gfp);
653 rxrpc_activate_channels(bundle);
654 ret = bundle->alloc_error;
655 if (ret < 0)
656 break;
657
658 switch (call->interruptibility) {
659 case RXRPC_INTERRUPTIBLE:
660 case RXRPC_PREINTERRUPTIBLE:
661 set_current_state(TASK_INTERRUPTIBLE);
662 break;
663 case RXRPC_UNINTERRUPTIBLE:
664 default:
665 set_current_state(TASK_UNINTERRUPTIBLE);
666 break;
667 }
668 if (READ_ONCE(call->state) != RXRPC_CALL_CLIENT_AWAIT_CONN)
669 break;
670 if ((call->interruptibility == RXRPC_INTERRUPTIBLE ||
671 call->interruptibility == RXRPC_PREINTERRUPTIBLE) &&
672 signal_pending(current)) {
673 ret = -ERESTARTSYS;
674 break;
675 }
676 schedule();
677 }
678 remove_wait_queue(&call->waitq, &myself);
679 __set_current_state(TASK_RUNNING);
680
681out:
682 _leave(" = %d", ret);
683 return ret;
684}
685
686/*
687 * find a connection for a call
688 * - called in process context with IRQs enabled
689 */
690int rxrpc_connect_call(struct rxrpc_sock *rx,
691 struct rxrpc_call *call,
692 struct rxrpc_conn_parameters *cp,
693 struct sockaddr_rxrpc *srx,
694 gfp_t gfp)
695{
696 struct rxrpc_bundle *bundle;
697 struct rxrpc_net *rxnet = cp->local->rxnet;
698 int ret = 0;
699
700 _enter("{%d,%lx},", call->debug_id, call->user_call_ID);
701
702 rxrpc_discard_expired_client_conns(&rxnet->client_conn_reaper);
703
704 bundle = rxrpc_prep_call(rx, call, cp, srx, gfp);
705 if (IS_ERR(bundle)) {
706 ret = PTR_ERR(bundle);
707 goto out;
708 }
709
710 if (call->state == RXRPC_CALL_CLIENT_AWAIT_CONN) {
711 ret = rxrpc_wait_for_channel(bundle, call, gfp);
712 if (ret < 0)
713 goto wait_failed;
714 }
715
716granted_channel:
717 /* Paired with the write barrier in rxrpc_activate_one_channel(). */
718 smp_rmb();
719
720out_put_bundle:
721 rxrpc_put_bundle(bundle);
722out:
723 _leave(" = %d", ret);
724 return ret;
725
726wait_failed:
727 spin_lock(&bundle->channel_lock);
728 list_del_init(&call->chan_wait_link);
729 spin_unlock(&bundle->channel_lock);
730
731 if (call->state != RXRPC_CALL_CLIENT_AWAIT_CONN) {
732 ret = 0;
733 goto granted_channel;
734 }
735
736 trace_rxrpc_client(call->conn, ret, rxrpc_client_chan_wait_failed);
737 rxrpc_set_call_completion(call, RXRPC_CALL_LOCAL_ERROR, 0, ret);
738 rxrpc_disconnect_client_call(bundle, call);
739 goto out_put_bundle;
740}
741
742/*
743 * Note that a call, and thus a connection, is about to be exposed to the
744 * world.
745 */
746void rxrpc_expose_client_call(struct rxrpc_call *call)
747{
748 unsigned int channel = call->cid & RXRPC_CHANNELMASK;
749 struct rxrpc_connection *conn = call->conn;
750 struct rxrpc_channel *chan = &conn->channels[channel];
751
752 if (!test_and_set_bit(RXRPC_CALL_EXPOSED, &call->flags)) {
753 /* Mark the call ID as being used. If the callNumber counter
754 * exceeds ~2 billion, we kill the connection after its
755 * outstanding calls have finished so that the counter doesn't
756 * wrap.
757 */
758 chan->call_counter++;
759 if (chan->call_counter >= INT_MAX)
760 set_bit(RXRPC_CONN_DONT_REUSE, &conn->flags);
761 trace_rxrpc_client(conn, channel, rxrpc_client_exposed);
762 }
763}
764
765/*
766 * Set the reap timer.
767 */
768static void rxrpc_set_client_reap_timer(struct rxrpc_net *rxnet)
769{
770 if (!rxnet->kill_all_client_conns) {
771 unsigned long now = jiffies;
772 unsigned long reap_at = now + rxrpc_conn_idle_client_expiry;
773
774 if (rxnet->live)
775 timer_reduce(&rxnet->client_conn_reap_timer, reap_at);
776 }
777}
778
779/*
780 * Disconnect a client call.
781 */
782void rxrpc_disconnect_client_call(struct rxrpc_bundle *bundle, struct rxrpc_call *call)
783{
784 struct rxrpc_connection *conn;
785 struct rxrpc_channel *chan = NULL;
786 struct rxrpc_net *rxnet = bundle->params.local->rxnet;
787 unsigned int channel;
788 bool may_reuse;
789 u32 cid;
790
791 _enter("c=%x", call->debug_id);
792
793 spin_lock(&bundle->channel_lock);
794 set_bit(RXRPC_CALL_DISCONNECTED, &call->flags);
795
796 /* Calls that have never actually been assigned a channel can simply be
797 * discarded.
798 */
799 conn = call->conn;
800 if (!conn) {
801 _debug("call is waiting");
802 ASSERTCMP(call->call_id, ==, 0);
803 ASSERT(!test_bit(RXRPC_CALL_EXPOSED, &call->flags));
804 list_del_init(&call->chan_wait_link);
805 goto out;
806 }
807
808 cid = call->cid;
809 channel = cid & RXRPC_CHANNELMASK;
810 chan = &conn->channels[channel];
811 trace_rxrpc_client(conn, channel, rxrpc_client_chan_disconnect);
812
813 if (rcu_access_pointer(chan->call) != call) {
814 spin_unlock(&bundle->channel_lock);
815 BUG();
816 }
817
818 may_reuse = rxrpc_may_reuse_conn(conn);
819
820 /* If a client call was exposed to the world, we save the result for
821 * retransmission.
822 *
823 * We use a barrier here so that the call number and abort code can be
824 * read without needing to take a lock.
825 *
826 * TODO: Make the incoming packet handler check this and handle
827 * terminal retransmission without requiring access to the call.
828 */
829 if (test_bit(RXRPC_CALL_EXPOSED, &call->flags)) {
830 _debug("exposed %u,%u", call->call_id, call->abort_code);
831 __rxrpc_disconnect_call(conn, call);
832
833 if (test_and_clear_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags)) {
834 trace_rxrpc_client(conn, channel, rxrpc_client_to_active);
835 bundle->try_upgrade = false;
836 if (may_reuse)
837 rxrpc_activate_channels_locked(bundle);
838 }
839
840 }
841
842 /* See if we can pass the channel directly to another call. */
843 if (may_reuse && !list_empty(&bundle->waiting_calls)) {
844 trace_rxrpc_client(conn, channel, rxrpc_client_chan_pass);
845 rxrpc_activate_one_channel(conn, channel);
846 goto out;
847 }
848
849 /* Schedule the final ACK to be transmitted in a short while so that it
850 * can be skipped if we find a follow-on call. The first DATA packet
851 * of the follow on call will implicitly ACK this call.
852 */
853 if (call->completion == RXRPC_CALL_SUCCEEDED &&
854 test_bit(RXRPC_CALL_EXPOSED, &call->flags)) {
855 unsigned long final_ack_at = jiffies + 2;
856
857 WRITE_ONCE(chan->final_ack_at, final_ack_at);
858 smp_wmb(); /* vs rxrpc_process_delayed_final_acks() */
859 set_bit(RXRPC_CONN_FINAL_ACK_0 + channel, &conn->flags);
860 rxrpc_reduce_conn_timer(conn, final_ack_at);
861 }
862
863 /* Deactivate the channel. */
864 rcu_assign_pointer(chan->call, NULL);
865 set_bit(conn->bundle_shift + channel, &conn->bundle->avail_chans);
866 conn->act_chans &= ~(1 << channel);
867
868 /* If no channels remain active, then put the connection on the idle
869 * list for a short while. Give it a ref to stop it going away if it
870 * becomes unbundled.
871 */
872 if (!conn->act_chans) {
873 trace_rxrpc_client(conn, channel, rxrpc_client_to_idle);
874 conn->idle_timestamp = jiffies;
875
876 rxrpc_get_connection(conn);
877 spin_lock(&rxnet->client_conn_cache_lock);
878 list_move_tail(&conn->cache_link, &rxnet->idle_client_conns);
879 spin_unlock(&rxnet->client_conn_cache_lock);
880
881 rxrpc_set_client_reap_timer(rxnet);
882 }
883
884out:
885 spin_unlock(&bundle->channel_lock);
886 _leave("");
887 return;
888}
889
890/*
891 * Remove a connection from a bundle.
892 */
893static void rxrpc_unbundle_conn(struct rxrpc_connection *conn)
894{
895 struct rxrpc_bundle *bundle = conn->bundle;
896 struct rxrpc_local *local = bundle->params.local;
897 unsigned int bindex;
898 bool need_drop = false, need_put = false;
899 int i;
900
901 _enter("C=%x", conn->debug_id);
902
903 if (conn->flags & RXRPC_CONN_FINAL_ACK_MASK)
904 rxrpc_process_delayed_final_acks(conn, true);
905
906 spin_lock(&bundle->channel_lock);
907 bindex = conn->bundle_shift / RXRPC_MAXCALLS;
908 if (bundle->conns[bindex] == conn) {
909 _debug("clear slot %u", bindex);
910 bundle->conns[bindex] = NULL;
911 for (i = 0; i < RXRPC_MAXCALLS; i++)
912 clear_bit(conn->bundle_shift + i, &bundle->avail_chans);
913 need_drop = true;
914 }
915 spin_unlock(&bundle->channel_lock);
916
917 /* If there are no more connections, remove the bundle */
918 if (!bundle->avail_chans) {
919 _debug("maybe unbundle");
920 spin_lock(&local->client_bundles_lock);
921
922 for (i = 0; i < ARRAY_SIZE(bundle->conns); i++)
923 if (bundle->conns[i])
924 break;
925 if (i == ARRAY_SIZE(bundle->conns) && !bundle->params.exclusive) {
926 _debug("erase bundle");
927 rb_erase(&bundle->local_node, &local->client_bundles);
928 need_put = true;
929 }
930
931 spin_unlock(&local->client_bundles_lock);
932 if (need_put)
933 rxrpc_put_bundle(bundle);
934 }
935
936 if (need_drop)
937 rxrpc_put_connection(conn);
938 _leave("");
939}
940
941/*
942 * Clean up a dead client connection.
943 */
944static void rxrpc_kill_client_conn(struct rxrpc_connection *conn)
945{
946 struct rxrpc_local *local = conn->params.local;
947 struct rxrpc_net *rxnet = local->rxnet;
948
949 _enter("C=%x", conn->debug_id);
950
951 trace_rxrpc_client(conn, -1, rxrpc_client_cleanup);
952 atomic_dec(&rxnet->nr_client_conns);
953
954 rxrpc_put_client_connection_id(conn);
955 rxrpc_kill_connection(conn);
956}
957
958/*
959 * Clean up a dead client connections.
960 */
961void rxrpc_put_client_conn(struct rxrpc_connection *conn)
962{
963 const void *here = __builtin_return_address(0);
964 unsigned int debug_id = conn->debug_id;
965 int n;
966
967 n = atomic_dec_return(&conn->usage);
968 trace_rxrpc_conn(debug_id, rxrpc_conn_put_client, n, here);
969 if (n <= 0) {
970 ASSERTCMP(n, >=, 0);
971 rxrpc_kill_client_conn(conn);
972 }
973}
974
975/*
976 * Discard expired client connections from the idle list. Each conn in the
977 * idle list has been exposed and holds an extra ref because of that.
978 *
979 * This may be called from conn setup or from a work item so cannot be
980 * considered non-reentrant.
981 */
982void rxrpc_discard_expired_client_conns(struct work_struct *work)
983{
984 struct rxrpc_connection *conn;
985 struct rxrpc_net *rxnet =
986 container_of(work, struct rxrpc_net, client_conn_reaper);
987 unsigned long expiry, conn_expires_at, now;
988 unsigned int nr_conns;
989
990 _enter("");
991
992 if (list_empty(&rxnet->idle_client_conns)) {
993 _leave(" [empty]");
994 return;
995 }
996
997 /* Don't double up on the discarding */
998 if (!spin_trylock(&rxnet->client_conn_discard_lock)) {
999 _leave(" [already]");
1000 return;
1001 }
1002
1003 /* We keep an estimate of what the number of conns ought to be after
1004 * we've discarded some so that we don't overdo the discarding.
1005 */
1006 nr_conns = atomic_read(&rxnet->nr_client_conns);
1007
1008next:
1009 spin_lock(&rxnet->client_conn_cache_lock);
1010
1011 if (list_empty(&rxnet->idle_client_conns))
1012 goto out;
1013
1014 conn = list_entry(rxnet->idle_client_conns.next,
1015 struct rxrpc_connection, cache_link);
1016
1017 if (!rxnet->kill_all_client_conns) {
1018 /* If the number of connections is over the reap limit, we
1019 * expedite discard by reducing the expiry timeout. We must,
1020 * however, have at least a short grace period to be able to do
1021 * final-ACK or ABORT retransmission.
1022 */
1023 expiry = rxrpc_conn_idle_client_expiry;
1024 if (nr_conns > rxrpc_reap_client_connections)
1025 expiry = rxrpc_conn_idle_client_fast_expiry;
1026 if (conn->params.local->service_closed)
1027 expiry = rxrpc_closed_conn_expiry * HZ;
1028
1029 conn_expires_at = conn->idle_timestamp + expiry;
1030
1031 now = READ_ONCE(jiffies);
1032 if (time_after(conn_expires_at, now))
1033 goto not_yet_expired;
1034 }
1035
1036 trace_rxrpc_client(conn, -1, rxrpc_client_discard);
1037 list_del_init(&conn->cache_link);
1038
1039 spin_unlock(&rxnet->client_conn_cache_lock);
1040
1041 rxrpc_unbundle_conn(conn);
1042 rxrpc_put_connection(conn); /* Drop the ->cache_link ref */
1043
1044 nr_conns--;
1045 goto next;
1046
1047not_yet_expired:
1048 /* The connection at the front of the queue hasn't yet expired, so
1049 * schedule the work item for that point if we discarded something.
1050 *
1051 * We don't worry if the work item is already scheduled - it can look
1052 * after rescheduling itself at a later time. We could cancel it, but
1053 * then things get messier.
1054 */
1055 _debug("not yet");
1056 if (!rxnet->kill_all_client_conns)
1057 timer_reduce(&rxnet->client_conn_reap_timer, conn_expires_at);
1058
1059out:
1060 spin_unlock(&rxnet->client_conn_cache_lock);
1061 spin_unlock(&rxnet->client_conn_discard_lock);
1062 _leave("");
1063}
1064
1065/*
1066 * Preemptively destroy all the client connection records rather than waiting
1067 * for them to time out
1068 */
1069void rxrpc_destroy_all_client_connections(struct rxrpc_net *rxnet)
1070{
1071 _enter("");
1072
1073 spin_lock(&rxnet->client_conn_cache_lock);
1074 rxnet->kill_all_client_conns = true;
1075 spin_unlock(&rxnet->client_conn_cache_lock);
1076
1077 del_timer_sync(&rxnet->client_conn_reap_timer);
1078
1079 if (!rxrpc_queue_work(&rxnet->client_conn_reaper))
1080 _debug("destroy: queue failed");
1081
1082 _leave("");
1083}
1084
1085/*
1086 * Clean up the client connections on a local endpoint.
1087 */
1088void rxrpc_clean_up_local_conns(struct rxrpc_local *local)
1089{
1090 struct rxrpc_connection *conn, *tmp;
1091 struct rxrpc_net *rxnet = local->rxnet;
1092 LIST_HEAD(graveyard);
1093
1094 _enter("");
1095
1096 spin_lock(&rxnet->client_conn_cache_lock);
1097
1098 list_for_each_entry_safe(conn, tmp, &rxnet->idle_client_conns,
1099 cache_link) {
1100 if (conn->params.local == local) {
1101 trace_rxrpc_client(conn, -1, rxrpc_client_discard);
1102 list_move(&conn->cache_link, &graveyard);
1103 }
1104 }
1105
1106 spin_unlock(&rxnet->client_conn_cache_lock);
1107
1108 while (!list_empty(&graveyard)) {
1109 conn = list_entry(graveyard.next,
1110 struct rxrpc_connection, cache_link);
1111 list_del_init(&conn->cache_link);
1112 rxrpc_unbundle_conn(conn);
1113 rxrpc_put_connection(conn);
1114 }
1115
1116 _leave(" [culled]");
1117}