xref: /linux/net/rxrpc/conn_client.c (revision b5bee6ced21ca98389000b7017dd41b0cc37fa50)
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  */
40 DEFINE_IDR(rxrpc_client_conn_ids);
41 static 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  */
49 static 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 
74 error:
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  */
84 static 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  */
97 void 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, refcount_read(&conn->ref));
106 		}
107 		BUG();
108 	}
109 
110 	idr_destroy(&rxrpc_client_conn_ids);
111 }
112 
113 /*
114  * Allocate a connection bundle.
115  */
116 static 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 		refcount_set(&bundle->ref, 1);
126 		spin_lock_init(&bundle->channel_lock);
127 		INIT_LIST_HEAD(&bundle->waiting_calls);
128 	}
129 	return bundle;
130 }
131 
132 struct rxrpc_bundle *rxrpc_get_bundle(struct rxrpc_bundle *bundle)
133 {
134 	refcount_inc(&bundle->ref);
135 	return bundle;
136 }
137 
138 static void rxrpc_free_bundle(struct rxrpc_bundle *bundle)
139 {
140 	rxrpc_put_peer(bundle->params.peer);
141 	kfree(bundle);
142 }
143 
144 void rxrpc_put_bundle(struct rxrpc_bundle *bundle)
145 {
146 	unsigned int d = bundle->debug_id;
147 	bool dead;
148 	int r;
149 
150 	dead = __refcount_dec_and_test(&bundle->ref, &r);
151 
152 	_debug("PUT B=%x %d", d, r);
153 	if (dead)
154 		rxrpc_free_bundle(bundle);
155 }
156 
157 /*
158  * Allocate a client connection.
159  */
160 static struct rxrpc_connection *
161 rxrpc_alloc_client_connection(struct rxrpc_bundle *bundle, gfp_t gfp)
162 {
163 	struct rxrpc_connection *conn;
164 	struct rxrpc_net *rxnet = bundle->params.local->rxnet;
165 	int ret;
166 
167 	_enter("");
168 
169 	conn = rxrpc_alloc_connection(gfp);
170 	if (!conn) {
171 		_leave(" = -ENOMEM");
172 		return ERR_PTR(-ENOMEM);
173 	}
174 
175 	refcount_set(&conn->ref, 1);
176 	conn->bundle		= bundle;
177 	conn->params		= bundle->params;
178 	conn->out_clientflag	= RXRPC_CLIENT_INITIATED;
179 	conn->state		= RXRPC_CONN_CLIENT;
180 	conn->service_id	= conn->params.service_id;
181 
182 	ret = rxrpc_get_client_connection_id(conn, gfp);
183 	if (ret < 0)
184 		goto error_0;
185 
186 	ret = rxrpc_init_client_conn_security(conn);
187 	if (ret < 0)
188 		goto error_1;
189 
190 	atomic_inc(&rxnet->nr_conns);
191 	write_lock(&rxnet->conn_lock);
192 	list_add_tail(&conn->proc_link, &rxnet->conn_proc_list);
193 	write_unlock(&rxnet->conn_lock);
194 
195 	rxrpc_get_bundle(bundle);
196 	rxrpc_get_peer(conn->params.peer);
197 	rxrpc_get_local(conn->params.local);
198 	key_get(conn->params.key);
199 
200 	trace_rxrpc_conn(conn->debug_id, rxrpc_conn_new_client,
201 			 refcount_read(&conn->ref),
202 			 __builtin_return_address(0));
203 
204 	atomic_inc(&rxnet->nr_client_conns);
205 	trace_rxrpc_client(conn, -1, rxrpc_client_alloc);
206 	_leave(" = %p", conn);
207 	return conn;
208 
209 error_1:
210 	rxrpc_put_client_connection_id(conn);
211 error_0:
212 	kfree(conn);
213 	_leave(" = %d", ret);
214 	return ERR_PTR(ret);
215 }
216 
217 /*
218  * Determine if a connection may be reused.
219  */
220 static bool rxrpc_may_reuse_conn(struct rxrpc_connection *conn)
221 {
222 	struct rxrpc_net *rxnet;
223 	int id_cursor, id, distance, limit;
224 
225 	if (!conn)
226 		goto dont_reuse;
227 
228 	rxnet = conn->params.local->rxnet;
229 	if (test_bit(RXRPC_CONN_DONT_REUSE, &conn->flags))
230 		goto dont_reuse;
231 
232 	if (conn->state != RXRPC_CONN_CLIENT ||
233 	    conn->proto.epoch != rxnet->epoch)
234 		goto mark_dont_reuse;
235 
236 	/* The IDR tree gets very expensive on memory if the connection IDs are
237 	 * widely scattered throughout the number space, so we shall want to
238 	 * kill off connections that, say, have an ID more than about four
239 	 * times the maximum number of client conns away from the current
240 	 * allocation point to try and keep the IDs concentrated.
241 	 */
242 	id_cursor = idr_get_cursor(&rxrpc_client_conn_ids);
243 	id = conn->proto.cid >> RXRPC_CIDSHIFT;
244 	distance = id - id_cursor;
245 	if (distance < 0)
246 		distance = -distance;
247 	limit = max_t(unsigned long, atomic_read(&rxnet->nr_conns) * 4, 1024);
248 	if (distance > limit)
249 		goto mark_dont_reuse;
250 
251 	return true;
252 
253 mark_dont_reuse:
254 	set_bit(RXRPC_CONN_DONT_REUSE, &conn->flags);
255 dont_reuse:
256 	return false;
257 }
258 
259 /*
260  * Look up the conn bundle that matches the connection parameters, adding it if
261  * it doesn't yet exist.
262  */
263 static struct rxrpc_bundle *rxrpc_look_up_bundle(struct rxrpc_conn_parameters *cp,
264 						 gfp_t gfp)
265 {
266 	static atomic_t rxrpc_bundle_id;
267 	struct rxrpc_bundle *bundle, *candidate;
268 	struct rxrpc_local *local = cp->local;
269 	struct rb_node *p, **pp, *parent;
270 	long diff;
271 
272 	_enter("{%px,%x,%u,%u}",
273 	       cp->peer, key_serial(cp->key), cp->security_level, cp->upgrade);
274 
275 	if (cp->exclusive)
276 		return rxrpc_alloc_bundle(cp, gfp);
277 
278 	/* First, see if the bundle is already there. */
279 	_debug("search 1");
280 	spin_lock(&local->client_bundles_lock);
281 	p = local->client_bundles.rb_node;
282 	while (p) {
283 		bundle = rb_entry(p, struct rxrpc_bundle, local_node);
284 
285 #define cmp(X) ((long)bundle->params.X - (long)cp->X)
286 		diff = (cmp(peer) ?:
287 			cmp(key) ?:
288 			cmp(security_level) ?:
289 			cmp(upgrade));
290 #undef cmp
291 		if (diff < 0)
292 			p = p->rb_left;
293 		else if (diff > 0)
294 			p = p->rb_right;
295 		else
296 			goto found_bundle;
297 	}
298 	spin_unlock(&local->client_bundles_lock);
299 	_debug("not found");
300 
301 	/* It wasn't.  We need to add one. */
302 	candidate = rxrpc_alloc_bundle(cp, gfp);
303 	if (!candidate)
304 		return NULL;
305 
306 	_debug("search 2");
307 	spin_lock(&local->client_bundles_lock);
308 	pp = &local->client_bundles.rb_node;
309 	parent = NULL;
310 	while (*pp) {
311 		parent = *pp;
312 		bundle = rb_entry(parent, struct rxrpc_bundle, local_node);
313 
314 #define cmp(X) ((long)bundle->params.X - (long)cp->X)
315 		diff = (cmp(peer) ?:
316 			cmp(key) ?:
317 			cmp(security_level) ?:
318 			cmp(upgrade));
319 #undef cmp
320 		if (diff < 0)
321 			pp = &(*pp)->rb_left;
322 		else if (diff > 0)
323 			pp = &(*pp)->rb_right;
324 		else
325 			goto found_bundle_free;
326 	}
327 
328 	_debug("new bundle");
329 	candidate->debug_id = atomic_inc_return(&rxrpc_bundle_id);
330 	rb_link_node(&candidate->local_node, parent, pp);
331 	rb_insert_color(&candidate->local_node, &local->client_bundles);
332 	rxrpc_get_bundle(candidate);
333 	spin_unlock(&local->client_bundles_lock);
334 	_leave(" = %u [new]", candidate->debug_id);
335 	return candidate;
336 
337 found_bundle_free:
338 	rxrpc_free_bundle(candidate);
339 found_bundle:
340 	rxrpc_get_bundle(bundle);
341 	spin_unlock(&local->client_bundles_lock);
342 	_leave(" = %u [found]", bundle->debug_id);
343 	return bundle;
344 }
345 
346 /*
347  * Create or find a client bundle to use for a call.
348  *
349  * If we return with a connection, the call will be on its waiting list.  It's
350  * left to the caller to assign a channel and wake up the call.
351  */
352 static struct rxrpc_bundle *rxrpc_prep_call(struct rxrpc_sock *rx,
353 					    struct rxrpc_call *call,
354 					    struct rxrpc_conn_parameters *cp,
355 					    struct sockaddr_rxrpc *srx,
356 					    gfp_t gfp)
357 {
358 	struct rxrpc_bundle *bundle;
359 
360 	_enter("{%d,%lx},", call->debug_id, call->user_call_ID);
361 
362 	cp->peer = rxrpc_lookup_peer(rx, cp->local, srx, gfp);
363 	if (!cp->peer)
364 		goto error;
365 
366 	call->cong_cwnd = cp->peer->cong_cwnd;
367 	if (call->cong_cwnd >= call->cong_ssthresh)
368 		call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
369 	else
370 		call->cong_mode = RXRPC_CALL_SLOW_START;
371 	if (cp->upgrade)
372 		__set_bit(RXRPC_CALL_UPGRADE, &call->flags);
373 
374 	/* Find the client connection bundle. */
375 	bundle = rxrpc_look_up_bundle(cp, gfp);
376 	if (!bundle)
377 		goto error;
378 
379 	/* Get this call queued.  Someone else may activate it whilst we're
380 	 * lining up a new connection, but that's fine.
381 	 */
382 	spin_lock(&bundle->channel_lock);
383 	list_add_tail(&call->chan_wait_link, &bundle->waiting_calls);
384 	spin_unlock(&bundle->channel_lock);
385 
386 	_leave(" = [B=%x]", bundle->debug_id);
387 	return bundle;
388 
389 error:
390 	_leave(" = -ENOMEM");
391 	return ERR_PTR(-ENOMEM);
392 }
393 
394 /*
395  * Allocate a new connection and add it into a bundle.
396  */
397 static void rxrpc_add_conn_to_bundle(struct rxrpc_bundle *bundle, gfp_t gfp)
398 	__releases(bundle->channel_lock)
399 {
400 	struct rxrpc_connection *candidate = NULL, *old = NULL;
401 	bool conflict;
402 	int i;
403 
404 	_enter("");
405 
406 	conflict = bundle->alloc_conn;
407 	if (!conflict)
408 		bundle->alloc_conn = true;
409 	spin_unlock(&bundle->channel_lock);
410 	if (conflict) {
411 		_leave(" [conf]");
412 		return;
413 	}
414 
415 	candidate = rxrpc_alloc_client_connection(bundle, gfp);
416 
417 	spin_lock(&bundle->channel_lock);
418 	bundle->alloc_conn = false;
419 
420 	if (IS_ERR(candidate)) {
421 		bundle->alloc_error = PTR_ERR(candidate);
422 		spin_unlock(&bundle->channel_lock);
423 		_leave(" [err %ld]", PTR_ERR(candidate));
424 		return;
425 	}
426 
427 	bundle->alloc_error = 0;
428 
429 	for (i = 0; i < ARRAY_SIZE(bundle->conns); i++) {
430 		unsigned int shift = i * RXRPC_MAXCALLS;
431 		int j;
432 
433 		old = bundle->conns[i];
434 		if (!rxrpc_may_reuse_conn(old)) {
435 			if (old)
436 				trace_rxrpc_client(old, -1, rxrpc_client_replace);
437 			candidate->bundle_shift = shift;
438 			bundle->conns[i] = candidate;
439 			for (j = 0; j < RXRPC_MAXCALLS; j++)
440 				set_bit(shift + j, &bundle->avail_chans);
441 			candidate = NULL;
442 			break;
443 		}
444 
445 		old = NULL;
446 	}
447 
448 	spin_unlock(&bundle->channel_lock);
449 
450 	if (candidate) {
451 		_debug("discard C=%x", candidate->debug_id);
452 		trace_rxrpc_client(candidate, -1, rxrpc_client_duplicate);
453 		rxrpc_put_connection(candidate);
454 	}
455 
456 	rxrpc_put_connection(old);
457 	_leave("");
458 }
459 
460 /*
461  * Add a connection to a bundle if there are no usable connections or we have
462  * connections waiting for extra capacity.
463  */
464 static void rxrpc_maybe_add_conn(struct rxrpc_bundle *bundle, gfp_t gfp)
465 {
466 	struct rxrpc_call *call;
467 	int i, usable;
468 
469 	_enter("");
470 
471 	spin_lock(&bundle->channel_lock);
472 
473 	/* See if there are any usable connections. */
474 	usable = 0;
475 	for (i = 0; i < ARRAY_SIZE(bundle->conns); i++)
476 		if (rxrpc_may_reuse_conn(bundle->conns[i]))
477 			usable++;
478 
479 	if (!usable && !list_empty(&bundle->waiting_calls)) {
480 		call = list_first_entry(&bundle->waiting_calls,
481 					struct rxrpc_call, chan_wait_link);
482 		if (test_bit(RXRPC_CALL_UPGRADE, &call->flags))
483 			bundle->try_upgrade = true;
484 	}
485 
486 	if (!usable)
487 		goto alloc_conn;
488 
489 	if (!bundle->avail_chans &&
490 	    !bundle->try_upgrade &&
491 	    !list_empty(&bundle->waiting_calls) &&
492 	    usable < ARRAY_SIZE(bundle->conns))
493 		goto alloc_conn;
494 
495 	spin_unlock(&bundle->channel_lock);
496 	_leave("");
497 	return;
498 
499 alloc_conn:
500 	return rxrpc_add_conn_to_bundle(bundle, gfp);
501 }
502 
503 /*
504  * Assign a channel to the call at the front of the queue and wake the call up.
505  * We don't increment the callNumber counter until this number has been exposed
506  * to the world.
507  */
508 static void rxrpc_activate_one_channel(struct rxrpc_connection *conn,
509 				       unsigned int channel)
510 {
511 	struct rxrpc_channel *chan = &conn->channels[channel];
512 	struct rxrpc_bundle *bundle = conn->bundle;
513 	struct rxrpc_call *call = list_entry(bundle->waiting_calls.next,
514 					     struct rxrpc_call, chan_wait_link);
515 	u32 call_id = chan->call_counter + 1;
516 
517 	_enter("C=%x,%u", conn->debug_id, channel);
518 
519 	trace_rxrpc_client(conn, channel, rxrpc_client_chan_activate);
520 
521 	/* Cancel the final ACK on the previous call if it hasn't been sent yet
522 	 * as the DATA packet will implicitly ACK it.
523 	 */
524 	clear_bit(RXRPC_CONN_FINAL_ACK_0 + channel, &conn->flags);
525 	clear_bit(conn->bundle_shift + channel, &bundle->avail_chans);
526 
527 	rxrpc_see_call(call);
528 	list_del_init(&call->chan_wait_link);
529 	call->peer	= rxrpc_get_peer(conn->params.peer);
530 	call->conn	= rxrpc_get_connection(conn);
531 	call->cid	= conn->proto.cid | channel;
532 	call->call_id	= call_id;
533 	call->security	= conn->security;
534 	call->security_ix = conn->security_ix;
535 	call->service_id = conn->service_id;
536 
537 	trace_rxrpc_connect_call(call);
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 	write_lock_bh(&call->state_lock);
542 	call->state = RXRPC_CALL_CLIENT_SEND_REQUEST;
543 	write_unlock_bh(&call->state_lock);
544 
545 	/* Paired with the read barrier in rxrpc_connect_call().  This orders
546 	 * cid and epoch in the connection wrt to call_id without the need to
547 	 * take the channel_lock.
548 	 *
549 	 * We provisionally assign a callNumber at this point, but we don't
550 	 * confirm it until the call is about to be exposed.
551 	 *
552 	 * TODO: Pair with a barrier in the data_ready handler when that looks
553 	 * at the call ID through a connection channel.
554 	 */
555 	smp_wmb();
556 
557 	chan->call_id		= call_id;
558 	chan->call_debug_id	= call->debug_id;
559 	rcu_assign_pointer(chan->call, call);
560 	wake_up(&call->waitq);
561 }
562 
563 /*
564  * Remove a connection from the idle list if it's on it.
565  */
566 static void rxrpc_unidle_conn(struct rxrpc_bundle *bundle, struct rxrpc_connection *conn)
567 {
568 	struct rxrpc_net *rxnet = bundle->params.local->rxnet;
569 	bool drop_ref;
570 
571 	if (!list_empty(&conn->cache_link)) {
572 		drop_ref = false;
573 		spin_lock(&rxnet->client_conn_cache_lock);
574 		if (!list_empty(&conn->cache_link)) {
575 			list_del_init(&conn->cache_link);
576 			drop_ref = true;
577 		}
578 		spin_unlock(&rxnet->client_conn_cache_lock);
579 		if (drop_ref)
580 			rxrpc_put_connection(conn);
581 	}
582 }
583 
584 /*
585  * Assign channels and callNumbers to waiting calls with channel_lock
586  * held by caller.
587  */
588 static void rxrpc_activate_channels_locked(struct rxrpc_bundle *bundle)
589 {
590 	struct rxrpc_connection *conn;
591 	unsigned long avail, mask;
592 	unsigned int channel, slot;
593 
594 	if (bundle->try_upgrade)
595 		mask = 1;
596 	else
597 		mask = ULONG_MAX;
598 
599 	while (!list_empty(&bundle->waiting_calls)) {
600 		avail = bundle->avail_chans & mask;
601 		if (!avail)
602 			break;
603 		channel = __ffs(avail);
604 		clear_bit(channel, &bundle->avail_chans);
605 
606 		slot = channel / RXRPC_MAXCALLS;
607 		conn = bundle->conns[slot];
608 		if (!conn)
609 			break;
610 
611 		if (bundle->try_upgrade)
612 			set_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags);
613 		rxrpc_unidle_conn(bundle, conn);
614 
615 		channel &= (RXRPC_MAXCALLS - 1);
616 		conn->act_chans	|= 1 << channel;
617 		rxrpc_activate_one_channel(conn, channel);
618 	}
619 }
620 
621 /*
622  * Assign channels and callNumbers to waiting calls.
623  */
624 static void rxrpc_activate_channels(struct rxrpc_bundle *bundle)
625 {
626 	_enter("B=%x", bundle->debug_id);
627 
628 	trace_rxrpc_client(NULL, -1, rxrpc_client_activate_chans);
629 
630 	if (!bundle->avail_chans)
631 		return;
632 
633 	spin_lock(&bundle->channel_lock);
634 	rxrpc_activate_channels_locked(bundle);
635 	spin_unlock(&bundle->channel_lock);
636 	_leave("");
637 }
638 
639 /*
640  * Wait for a callNumber and a channel to be granted to a call.
641  */
642 static int rxrpc_wait_for_channel(struct rxrpc_bundle *bundle,
643 				  struct rxrpc_call *call, gfp_t gfp)
644 {
645 	DECLARE_WAITQUEUE(myself, current);
646 	int ret = 0;
647 
648 	_enter("%d", call->debug_id);
649 
650 	if (!gfpflags_allow_blocking(gfp)) {
651 		rxrpc_maybe_add_conn(bundle, gfp);
652 		rxrpc_activate_channels(bundle);
653 		ret = bundle->alloc_error ?: -EAGAIN;
654 		goto out;
655 	}
656 
657 	add_wait_queue_exclusive(&call->waitq, &myself);
658 	for (;;) {
659 		rxrpc_maybe_add_conn(bundle, gfp);
660 		rxrpc_activate_channels(bundle);
661 		ret = bundle->alloc_error;
662 		if (ret < 0)
663 			break;
664 
665 		switch (call->interruptibility) {
666 		case RXRPC_INTERRUPTIBLE:
667 		case RXRPC_PREINTERRUPTIBLE:
668 			set_current_state(TASK_INTERRUPTIBLE);
669 			break;
670 		case RXRPC_UNINTERRUPTIBLE:
671 		default:
672 			set_current_state(TASK_UNINTERRUPTIBLE);
673 			break;
674 		}
675 		if (READ_ONCE(call->state) != RXRPC_CALL_CLIENT_AWAIT_CONN)
676 			break;
677 		if ((call->interruptibility == RXRPC_INTERRUPTIBLE ||
678 		     call->interruptibility == RXRPC_PREINTERRUPTIBLE) &&
679 		    signal_pending(current)) {
680 			ret = -ERESTARTSYS;
681 			break;
682 		}
683 		schedule();
684 	}
685 	remove_wait_queue(&call->waitq, &myself);
686 	__set_current_state(TASK_RUNNING);
687 
688 out:
689 	_leave(" = %d", ret);
690 	return ret;
691 }
692 
693 /*
694  * find a connection for a call
695  * - called in process context with IRQs enabled
696  */
697 int rxrpc_connect_call(struct rxrpc_sock *rx,
698 		       struct rxrpc_call *call,
699 		       struct rxrpc_conn_parameters *cp,
700 		       struct sockaddr_rxrpc *srx,
701 		       gfp_t gfp)
702 {
703 	struct rxrpc_bundle *bundle;
704 	struct rxrpc_net *rxnet = cp->local->rxnet;
705 	int ret = 0;
706 
707 	_enter("{%d,%lx},", call->debug_id, call->user_call_ID);
708 
709 	rxrpc_discard_expired_client_conns(&rxnet->client_conn_reaper);
710 
711 	bundle = rxrpc_prep_call(rx, call, cp, srx, gfp);
712 	if (IS_ERR(bundle)) {
713 		ret = PTR_ERR(bundle);
714 		goto out;
715 	}
716 
717 	if (call->state == RXRPC_CALL_CLIENT_AWAIT_CONN) {
718 		ret = rxrpc_wait_for_channel(bundle, call, gfp);
719 		if (ret < 0)
720 			goto wait_failed;
721 	}
722 
723 granted_channel:
724 	/* Paired with the write barrier in rxrpc_activate_one_channel(). */
725 	smp_rmb();
726 
727 out_put_bundle:
728 	rxrpc_put_bundle(bundle);
729 out:
730 	_leave(" = %d", ret);
731 	return ret;
732 
733 wait_failed:
734 	spin_lock(&bundle->channel_lock);
735 	list_del_init(&call->chan_wait_link);
736 	spin_unlock(&bundle->channel_lock);
737 
738 	if (call->state != RXRPC_CALL_CLIENT_AWAIT_CONN) {
739 		ret = 0;
740 		goto granted_channel;
741 	}
742 
743 	trace_rxrpc_client(call->conn, ret, rxrpc_client_chan_wait_failed);
744 	rxrpc_set_call_completion(call, RXRPC_CALL_LOCAL_ERROR, 0, ret);
745 	rxrpc_disconnect_client_call(bundle, call);
746 	goto out_put_bundle;
747 }
748 
749 /*
750  * Note that a call, and thus a connection, is about to be exposed to the
751  * world.
752  */
753 void rxrpc_expose_client_call(struct rxrpc_call *call)
754 {
755 	unsigned int channel = call->cid & RXRPC_CHANNELMASK;
756 	struct rxrpc_connection *conn = call->conn;
757 	struct rxrpc_channel *chan = &conn->channels[channel];
758 
759 	if (!test_and_set_bit(RXRPC_CALL_EXPOSED, &call->flags)) {
760 		/* Mark the call ID as being used.  If the callNumber counter
761 		 * exceeds ~2 billion, we kill the connection after its
762 		 * outstanding calls have finished so that the counter doesn't
763 		 * wrap.
764 		 */
765 		chan->call_counter++;
766 		if (chan->call_counter >= INT_MAX)
767 			set_bit(RXRPC_CONN_DONT_REUSE, &conn->flags);
768 		trace_rxrpc_client(conn, channel, rxrpc_client_exposed);
769 	}
770 }
771 
772 /*
773  * Set the reap timer.
774  */
775 static void rxrpc_set_client_reap_timer(struct rxrpc_net *rxnet)
776 {
777 	if (!rxnet->kill_all_client_conns) {
778 		unsigned long now = jiffies;
779 		unsigned long reap_at = now + rxrpc_conn_idle_client_expiry;
780 
781 		if (rxnet->live)
782 			timer_reduce(&rxnet->client_conn_reap_timer, reap_at);
783 	}
784 }
785 
786 /*
787  * Disconnect a client call.
788  */
789 void rxrpc_disconnect_client_call(struct rxrpc_bundle *bundle, struct rxrpc_call *call)
790 {
791 	struct rxrpc_connection *conn;
792 	struct rxrpc_channel *chan = NULL;
793 	struct rxrpc_net *rxnet = bundle->params.local->rxnet;
794 	unsigned int channel;
795 	bool may_reuse;
796 	u32 cid;
797 
798 	_enter("c=%x", call->debug_id);
799 
800 	spin_lock(&bundle->channel_lock);
801 	set_bit(RXRPC_CALL_DISCONNECTED, &call->flags);
802 
803 	/* Calls that have never actually been assigned a channel can simply be
804 	 * discarded.
805 	 */
806 	conn = call->conn;
807 	if (!conn) {
808 		_debug("call is waiting");
809 		ASSERTCMP(call->call_id, ==, 0);
810 		ASSERT(!test_bit(RXRPC_CALL_EXPOSED, &call->flags));
811 		list_del_init(&call->chan_wait_link);
812 		goto out;
813 	}
814 
815 	cid = call->cid;
816 	channel = cid & RXRPC_CHANNELMASK;
817 	chan = &conn->channels[channel];
818 	trace_rxrpc_client(conn, channel, rxrpc_client_chan_disconnect);
819 
820 	if (rcu_access_pointer(chan->call) != call) {
821 		spin_unlock(&bundle->channel_lock);
822 		BUG();
823 	}
824 
825 	may_reuse = rxrpc_may_reuse_conn(conn);
826 
827 	/* If a client call was exposed to the world, we save the result for
828 	 * retransmission.
829 	 *
830 	 * We use a barrier here so that the call number and abort code can be
831 	 * read without needing to take a lock.
832 	 *
833 	 * TODO: Make the incoming packet handler check this and handle
834 	 * terminal retransmission without requiring access to the call.
835 	 */
836 	if (test_bit(RXRPC_CALL_EXPOSED, &call->flags)) {
837 		_debug("exposed %u,%u", call->call_id, call->abort_code);
838 		__rxrpc_disconnect_call(conn, call);
839 
840 		if (test_and_clear_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags)) {
841 			trace_rxrpc_client(conn, channel, rxrpc_client_to_active);
842 			bundle->try_upgrade = false;
843 			if (may_reuse)
844 				rxrpc_activate_channels_locked(bundle);
845 		}
846 
847 	}
848 
849 	/* See if we can pass the channel directly to another call. */
850 	if (may_reuse && !list_empty(&bundle->waiting_calls)) {
851 		trace_rxrpc_client(conn, channel, rxrpc_client_chan_pass);
852 		rxrpc_activate_one_channel(conn, channel);
853 		goto out;
854 	}
855 
856 	/* Schedule the final ACK to be transmitted in a short while so that it
857 	 * can be skipped if we find a follow-on call.  The first DATA packet
858 	 * of the follow on call will implicitly ACK this call.
859 	 */
860 	if (call->completion == RXRPC_CALL_SUCCEEDED &&
861 	    test_bit(RXRPC_CALL_EXPOSED, &call->flags)) {
862 		unsigned long final_ack_at = jiffies + 2;
863 
864 		WRITE_ONCE(chan->final_ack_at, final_ack_at);
865 		smp_wmb(); /* vs rxrpc_process_delayed_final_acks() */
866 		set_bit(RXRPC_CONN_FINAL_ACK_0 + channel, &conn->flags);
867 		rxrpc_reduce_conn_timer(conn, final_ack_at);
868 	}
869 
870 	/* Deactivate the channel. */
871 	rcu_assign_pointer(chan->call, NULL);
872 	set_bit(conn->bundle_shift + channel, &conn->bundle->avail_chans);
873 	conn->act_chans	&= ~(1 << channel);
874 
875 	/* If no channels remain active, then put the connection on the idle
876 	 * list for a short while.  Give it a ref to stop it going away if it
877 	 * becomes unbundled.
878 	 */
879 	if (!conn->act_chans) {
880 		trace_rxrpc_client(conn, channel, rxrpc_client_to_idle);
881 		conn->idle_timestamp = jiffies;
882 
883 		rxrpc_get_connection(conn);
884 		spin_lock(&rxnet->client_conn_cache_lock);
885 		list_move_tail(&conn->cache_link, &rxnet->idle_client_conns);
886 		spin_unlock(&rxnet->client_conn_cache_lock);
887 
888 		rxrpc_set_client_reap_timer(rxnet);
889 	}
890 
891 out:
892 	spin_unlock(&bundle->channel_lock);
893 	_leave("");
894 	return;
895 }
896 
897 /*
898  * Remove a connection from a bundle.
899  */
900 static void rxrpc_unbundle_conn(struct rxrpc_connection *conn)
901 {
902 	struct rxrpc_bundle *bundle = conn->bundle;
903 	struct rxrpc_local *local = bundle->params.local;
904 	unsigned int bindex;
905 	bool need_drop = false, need_put = false;
906 	int i;
907 
908 	_enter("C=%x", conn->debug_id);
909 
910 	if (conn->flags & RXRPC_CONN_FINAL_ACK_MASK)
911 		rxrpc_process_delayed_final_acks(conn, true);
912 
913 	spin_lock(&bundle->channel_lock);
914 	bindex = conn->bundle_shift / RXRPC_MAXCALLS;
915 	if (bundle->conns[bindex] == conn) {
916 		_debug("clear slot %u", bindex);
917 		bundle->conns[bindex] = NULL;
918 		for (i = 0; i < RXRPC_MAXCALLS; i++)
919 			clear_bit(conn->bundle_shift + i, &bundle->avail_chans);
920 		need_drop = true;
921 	}
922 	spin_unlock(&bundle->channel_lock);
923 
924 	/* If there are no more connections, remove the bundle */
925 	if (!bundle->avail_chans) {
926 		_debug("maybe unbundle");
927 		spin_lock(&local->client_bundles_lock);
928 
929 		for (i = 0; i < ARRAY_SIZE(bundle->conns); i++)
930 			if (bundle->conns[i])
931 				break;
932 		if (i == ARRAY_SIZE(bundle->conns) && !bundle->params.exclusive) {
933 			_debug("erase bundle");
934 			rb_erase(&bundle->local_node, &local->client_bundles);
935 			need_put = true;
936 		}
937 
938 		spin_unlock(&local->client_bundles_lock);
939 		if (need_put)
940 			rxrpc_put_bundle(bundle);
941 	}
942 
943 	if (need_drop)
944 		rxrpc_put_connection(conn);
945 	_leave("");
946 }
947 
948 /*
949  * Clean up a dead client connection.
950  */
951 static void rxrpc_kill_client_conn(struct rxrpc_connection *conn)
952 {
953 	struct rxrpc_local *local = conn->params.local;
954 	struct rxrpc_net *rxnet = local->rxnet;
955 
956 	_enter("C=%x", conn->debug_id);
957 
958 	trace_rxrpc_client(conn, -1, rxrpc_client_cleanup);
959 	atomic_dec(&rxnet->nr_client_conns);
960 
961 	rxrpc_put_client_connection_id(conn);
962 	rxrpc_kill_connection(conn);
963 }
964 
965 /*
966  * Clean up a dead client connections.
967  */
968 void rxrpc_put_client_conn(struct rxrpc_connection *conn)
969 {
970 	const void *here = __builtin_return_address(0);
971 	unsigned int debug_id = conn->debug_id;
972 	bool dead;
973 	int r;
974 
975 	dead = __refcount_dec_and_test(&conn->ref, &r);
976 	trace_rxrpc_conn(debug_id, rxrpc_conn_put_client, r - 1, here);
977 	if (dead)
978 		rxrpc_kill_client_conn(conn);
979 }
980 
981 /*
982  * Discard expired client connections from the idle list.  Each conn in the
983  * idle list has been exposed and holds an extra ref because of that.
984  *
985  * This may be called from conn setup or from a work item so cannot be
986  * considered non-reentrant.
987  */
988 void rxrpc_discard_expired_client_conns(struct work_struct *work)
989 {
990 	struct rxrpc_connection *conn;
991 	struct rxrpc_net *rxnet =
992 		container_of(work, struct rxrpc_net, client_conn_reaper);
993 	unsigned long expiry, conn_expires_at, now;
994 	unsigned int nr_conns;
995 
996 	_enter("");
997 
998 	if (list_empty(&rxnet->idle_client_conns)) {
999 		_leave(" [empty]");
1000 		return;
1001 	}
1002 
1003 	/* Don't double up on the discarding */
1004 	if (!spin_trylock(&rxnet->client_conn_discard_lock)) {
1005 		_leave(" [already]");
1006 		return;
1007 	}
1008 
1009 	/* We keep an estimate of what the number of conns ought to be after
1010 	 * we've discarded some so that we don't overdo the discarding.
1011 	 */
1012 	nr_conns = atomic_read(&rxnet->nr_client_conns);
1013 
1014 next:
1015 	spin_lock(&rxnet->client_conn_cache_lock);
1016 
1017 	if (list_empty(&rxnet->idle_client_conns))
1018 		goto out;
1019 
1020 	conn = list_entry(rxnet->idle_client_conns.next,
1021 			  struct rxrpc_connection, cache_link);
1022 
1023 	if (!rxnet->kill_all_client_conns) {
1024 		/* If the number of connections is over the reap limit, we
1025 		 * expedite discard by reducing the expiry timeout.  We must,
1026 		 * however, have at least a short grace period to be able to do
1027 		 * final-ACK or ABORT retransmission.
1028 		 */
1029 		expiry = rxrpc_conn_idle_client_expiry;
1030 		if (nr_conns > rxrpc_reap_client_connections)
1031 			expiry = rxrpc_conn_idle_client_fast_expiry;
1032 		if (conn->params.local->service_closed)
1033 			expiry = rxrpc_closed_conn_expiry * HZ;
1034 
1035 		conn_expires_at = conn->idle_timestamp + expiry;
1036 
1037 		now = READ_ONCE(jiffies);
1038 		if (time_after(conn_expires_at, now))
1039 			goto not_yet_expired;
1040 	}
1041 
1042 	trace_rxrpc_client(conn, -1, rxrpc_client_discard);
1043 	list_del_init(&conn->cache_link);
1044 
1045 	spin_unlock(&rxnet->client_conn_cache_lock);
1046 
1047 	rxrpc_unbundle_conn(conn);
1048 	rxrpc_put_connection(conn); /* Drop the ->cache_link ref */
1049 
1050 	nr_conns--;
1051 	goto next;
1052 
1053 not_yet_expired:
1054 	/* The connection at the front of the queue hasn't yet expired, so
1055 	 * schedule the work item for that point if we discarded something.
1056 	 *
1057 	 * We don't worry if the work item is already scheduled - it can look
1058 	 * after rescheduling itself at a later time.  We could cancel it, but
1059 	 * then things get messier.
1060 	 */
1061 	_debug("not yet");
1062 	if (!rxnet->kill_all_client_conns)
1063 		timer_reduce(&rxnet->client_conn_reap_timer, conn_expires_at);
1064 
1065 out:
1066 	spin_unlock(&rxnet->client_conn_cache_lock);
1067 	spin_unlock(&rxnet->client_conn_discard_lock);
1068 	_leave("");
1069 }
1070 
1071 /*
1072  * Preemptively destroy all the client connection records rather than waiting
1073  * for them to time out
1074  */
1075 void rxrpc_destroy_all_client_connections(struct rxrpc_net *rxnet)
1076 {
1077 	_enter("");
1078 
1079 	spin_lock(&rxnet->client_conn_cache_lock);
1080 	rxnet->kill_all_client_conns = true;
1081 	spin_unlock(&rxnet->client_conn_cache_lock);
1082 
1083 	del_timer_sync(&rxnet->client_conn_reap_timer);
1084 
1085 	if (!rxrpc_queue_work(&rxnet->client_conn_reaper))
1086 		_debug("destroy: queue failed");
1087 
1088 	_leave("");
1089 }
1090 
1091 /*
1092  * Clean up the client connections on a local endpoint.
1093  */
1094 void rxrpc_clean_up_local_conns(struct rxrpc_local *local)
1095 {
1096 	struct rxrpc_connection *conn, *tmp;
1097 	struct rxrpc_net *rxnet = local->rxnet;
1098 	LIST_HEAD(graveyard);
1099 
1100 	_enter("");
1101 
1102 	spin_lock(&rxnet->client_conn_cache_lock);
1103 
1104 	list_for_each_entry_safe(conn, tmp, &rxnet->idle_client_conns,
1105 				 cache_link) {
1106 		if (conn->params.local == local) {
1107 			trace_rxrpc_client(conn, -1, rxrpc_client_discard);
1108 			list_move(&conn->cache_link, &graveyard);
1109 		}
1110 	}
1111 
1112 	spin_unlock(&rxnet->client_conn_cache_lock);
1113 
1114 	while (!list_empty(&graveyard)) {
1115 		conn = list_entry(graveyard.next,
1116 				  struct rxrpc_connection, cache_link);
1117 		list_del_init(&conn->cache_link);
1118 		rxrpc_unbundle_conn(conn);
1119 		rxrpc_put_connection(conn);
1120 	}
1121 
1122 	_leave(" [culled]");
1123 }
1124