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 static void rxrpc_deactivate_bundle(struct rxrpc_bundle *bundle);
44 
45 /*
46  * Get a connection ID and epoch for a client connection from the global pool.
47  * The connection struct pointer is then recorded in the idr radix tree.  The
48  * epoch doesn't change until the client is rebooted (or, at least, unless the
49  * module is unloaded).
50  */
rxrpc_get_client_connection_id(struct rxrpc_connection * conn,gfp_t gfp)51 static int rxrpc_get_client_connection_id(struct rxrpc_connection *conn,
52 					  gfp_t gfp)
53 {
54 	struct rxrpc_net *rxnet = conn->params.local->rxnet;
55 	int id;
56 
57 	_enter("");
58 
59 	idr_preload(gfp);
60 	spin_lock(&rxrpc_conn_id_lock);
61 
62 	id = idr_alloc_cyclic(&rxrpc_client_conn_ids, conn,
63 			      1, 0x40000000, GFP_NOWAIT);
64 	if (id < 0)
65 		goto error;
66 
67 	spin_unlock(&rxrpc_conn_id_lock);
68 	idr_preload_end();
69 
70 	conn->proto.epoch = rxnet->epoch;
71 	conn->proto.cid = id << RXRPC_CIDSHIFT;
72 	set_bit(RXRPC_CONN_HAS_IDR, &conn->flags);
73 	_leave(" [CID %x]", conn->proto.cid);
74 	return 0;
75 
76 error:
77 	spin_unlock(&rxrpc_conn_id_lock);
78 	idr_preload_end();
79 	_leave(" = %d", id);
80 	return id;
81 }
82 
83 /*
84  * Release a connection ID for a client connection from the global pool.
85  */
rxrpc_put_client_connection_id(struct rxrpc_connection * conn)86 static void rxrpc_put_client_connection_id(struct rxrpc_connection *conn)
87 {
88 	if (test_bit(RXRPC_CONN_HAS_IDR, &conn->flags)) {
89 		spin_lock(&rxrpc_conn_id_lock);
90 		idr_remove(&rxrpc_client_conn_ids,
91 			   conn->proto.cid >> RXRPC_CIDSHIFT);
92 		spin_unlock(&rxrpc_conn_id_lock);
93 	}
94 }
95 
96 /*
97  * Destroy the client connection ID tree.
98  */
rxrpc_destroy_client_conn_ids(void)99 void rxrpc_destroy_client_conn_ids(void)
100 {
101 	struct rxrpc_connection *conn;
102 	int id;
103 
104 	if (!idr_is_empty(&rxrpc_client_conn_ids)) {
105 		idr_for_each_entry(&rxrpc_client_conn_ids, conn, id) {
106 			pr_err("AF_RXRPC: Leaked client conn %p {%d}\n",
107 			       conn, refcount_read(&conn->ref));
108 		}
109 		BUG();
110 	}
111 
112 	idr_destroy(&rxrpc_client_conn_ids);
113 }
114 
115 /*
116  * Allocate a connection bundle.
117  */
rxrpc_alloc_bundle(struct rxrpc_conn_parameters * cp,gfp_t gfp)118 static struct rxrpc_bundle *rxrpc_alloc_bundle(struct rxrpc_conn_parameters *cp,
119 					       gfp_t gfp)
120 {
121 	struct rxrpc_bundle *bundle;
122 
123 	bundle = kzalloc(sizeof(*bundle), gfp);
124 	if (bundle) {
125 		bundle->params = *cp;
126 		rxrpc_get_peer(bundle->params.peer);
127 		refcount_set(&bundle->ref, 1);
128 		atomic_set(&bundle->active, 1);
129 		spin_lock_init(&bundle->channel_lock);
130 		INIT_LIST_HEAD(&bundle->waiting_calls);
131 	}
132 	return bundle;
133 }
134 
rxrpc_get_bundle(struct rxrpc_bundle * bundle)135 struct rxrpc_bundle *rxrpc_get_bundle(struct rxrpc_bundle *bundle)
136 {
137 	refcount_inc(&bundle->ref);
138 	return bundle;
139 }
140 
rxrpc_free_bundle(struct rxrpc_bundle * bundle)141 static void rxrpc_free_bundle(struct rxrpc_bundle *bundle)
142 {
143 	rxrpc_put_peer(bundle->params.peer);
144 	kfree(bundle);
145 }
146 
rxrpc_put_bundle(struct rxrpc_bundle * bundle)147 void rxrpc_put_bundle(struct rxrpc_bundle *bundle)
148 {
149 	unsigned int d = bundle->debug_id;
150 	bool dead;
151 	int r;
152 
153 	dead = __refcount_dec_and_test(&bundle->ref, &r);
154 
155 	_debug("PUT B=%x %d", d, r - 1);
156 	if (dead)
157 		rxrpc_free_bundle(bundle);
158 }
159 
160 /*
161  * Allocate a client connection.
162  */
163 static struct rxrpc_connection *
rxrpc_alloc_client_connection(struct rxrpc_bundle * bundle,gfp_t gfp)164 rxrpc_alloc_client_connection(struct rxrpc_bundle *bundle, gfp_t gfp)
165 {
166 	struct rxrpc_connection *conn;
167 	struct rxrpc_net *rxnet = bundle->params.local->rxnet;
168 	int ret;
169 
170 	_enter("");
171 
172 	conn = rxrpc_alloc_connection(gfp);
173 	if (!conn) {
174 		_leave(" = -ENOMEM");
175 		return ERR_PTR(-ENOMEM);
176 	}
177 
178 	refcount_set(&conn->ref, 1);
179 	conn->bundle		= bundle;
180 	conn->params		= bundle->params;
181 	conn->out_clientflag	= RXRPC_CLIENT_INITIATED;
182 	conn->state		= RXRPC_CONN_CLIENT;
183 	conn->service_id	= conn->params.service_id;
184 
185 	ret = rxrpc_get_client_connection_id(conn, gfp);
186 	if (ret < 0)
187 		goto error_0;
188 
189 	ret = rxrpc_init_client_conn_security(conn);
190 	if (ret < 0)
191 		goto error_1;
192 
193 	atomic_inc(&rxnet->nr_conns);
194 	write_lock(&rxnet->conn_lock);
195 	list_add_tail(&conn->proc_link, &rxnet->conn_proc_list);
196 	write_unlock(&rxnet->conn_lock);
197 
198 	rxrpc_get_bundle(bundle);
199 	rxrpc_get_peer(conn->params.peer);
200 	rxrpc_get_local(conn->params.local);
201 	key_get(conn->params.key);
202 
203 	trace_rxrpc_conn(conn->debug_id, rxrpc_conn_new_client,
204 			 refcount_read(&conn->ref),
205 			 __builtin_return_address(0));
206 
207 	atomic_inc(&rxnet->nr_client_conns);
208 	trace_rxrpc_client(conn, -1, rxrpc_client_alloc);
209 	_leave(" = %p", conn);
210 	return conn;
211 
212 error_1:
213 	rxrpc_put_client_connection_id(conn);
214 error_0:
215 	kfree(conn);
216 	_leave(" = %d", ret);
217 	return ERR_PTR(ret);
218 }
219 
220 /*
221  * Determine if a connection may be reused.
222  */
rxrpc_may_reuse_conn(struct rxrpc_connection * conn)223 static bool rxrpc_may_reuse_conn(struct rxrpc_connection *conn)
224 {
225 	struct rxrpc_net *rxnet;
226 	int id_cursor, id, distance, limit;
227 
228 	if (!conn)
229 		goto dont_reuse;
230 
231 	rxnet = conn->params.local->rxnet;
232 	if (test_bit(RXRPC_CONN_DONT_REUSE, &conn->flags))
233 		goto dont_reuse;
234 
235 	if (conn->state != RXRPC_CONN_CLIENT ||
236 	    conn->proto.epoch != rxnet->epoch)
237 		goto mark_dont_reuse;
238 
239 	/* The IDR tree gets very expensive on memory if the connection IDs are
240 	 * widely scattered throughout the number space, so we shall want to
241 	 * kill off connections that, say, have an ID more than about four
242 	 * times the maximum number of client conns away from the current
243 	 * allocation point to try and keep the IDs concentrated.
244 	 */
245 	id_cursor = idr_get_cursor(&rxrpc_client_conn_ids);
246 	id = conn->proto.cid >> RXRPC_CIDSHIFT;
247 	distance = id - id_cursor;
248 	if (distance < 0)
249 		distance = -distance;
250 	limit = max_t(unsigned long, atomic_read(&rxnet->nr_conns) * 4, 1024);
251 	if (distance > limit)
252 		goto mark_dont_reuse;
253 
254 	return true;
255 
256 mark_dont_reuse:
257 	set_bit(RXRPC_CONN_DONT_REUSE, &conn->flags);
258 dont_reuse:
259 	return false;
260 }
261 
262 /*
263  * Look up the conn bundle that matches the connection parameters, adding it if
264  * it doesn't yet exist.
265  */
rxrpc_look_up_bundle(struct rxrpc_conn_parameters * cp,gfp_t gfp)266 static struct rxrpc_bundle *rxrpc_look_up_bundle(struct rxrpc_conn_parameters *cp,
267 						 gfp_t gfp)
268 {
269 	static atomic_t rxrpc_bundle_id;
270 	struct rxrpc_bundle *bundle, *candidate;
271 	struct rxrpc_local *local = cp->local;
272 	struct rb_node *p, **pp, *parent;
273 	long diff;
274 
275 	_enter("{%px,%x,%u,%u}",
276 	       cp->peer, key_serial(cp->key), cp->security_level, cp->upgrade);
277 
278 	if (cp->exclusive)
279 		return rxrpc_alloc_bundle(cp, gfp);
280 
281 	/* First, see if the bundle is already there. */
282 	_debug("search 1");
283 	spin_lock(&local->client_bundles_lock);
284 	p = local->client_bundles.rb_node;
285 	while (p) {
286 		bundle = rb_entry(p, struct rxrpc_bundle, local_node);
287 
288 #define cmp(X) ((long)bundle->params.X - (long)cp->X)
289 		diff = (cmp(peer) ?:
290 			cmp(key) ?:
291 			cmp(security_level) ?:
292 			cmp(upgrade));
293 #undef cmp
294 		if (diff < 0)
295 			p = p->rb_left;
296 		else if (diff > 0)
297 			p = p->rb_right;
298 		else
299 			goto found_bundle;
300 	}
301 	spin_unlock(&local->client_bundles_lock);
302 	_debug("not found");
303 
304 	/* It wasn't.  We need to add one. */
305 	candidate = rxrpc_alloc_bundle(cp, gfp);
306 	if (!candidate)
307 		return NULL;
308 
309 	_debug("search 2");
310 	spin_lock(&local->client_bundles_lock);
311 	pp = &local->client_bundles.rb_node;
312 	parent = NULL;
313 	while (*pp) {
314 		parent = *pp;
315 		bundle = rb_entry(parent, struct rxrpc_bundle, local_node);
316 
317 #define cmp(X) ((long)bundle->params.X - (long)cp->X)
318 		diff = (cmp(peer) ?:
319 			cmp(key) ?:
320 			cmp(security_level) ?:
321 			cmp(upgrade));
322 #undef cmp
323 		if (diff < 0)
324 			pp = &(*pp)->rb_left;
325 		else if (diff > 0)
326 			pp = &(*pp)->rb_right;
327 		else
328 			goto found_bundle_free;
329 	}
330 
331 	_debug("new bundle");
332 	candidate->debug_id = atomic_inc_return(&rxrpc_bundle_id);
333 	rb_link_node(&candidate->local_node, parent, pp);
334 	rb_insert_color(&candidate->local_node, &local->client_bundles);
335 	rxrpc_get_bundle(candidate);
336 	spin_unlock(&local->client_bundles_lock);
337 	_leave(" = %u [new]", candidate->debug_id);
338 	return candidate;
339 
340 found_bundle_free:
341 	rxrpc_free_bundle(candidate);
342 found_bundle:
343 	rxrpc_get_bundle(bundle);
344 	atomic_inc(&bundle->active);
345 	spin_unlock(&local->client_bundles_lock);
346 	_leave(" = %u [found]", bundle->debug_id);
347 	return bundle;
348 }
349 
350 /*
351  * Create or find a client bundle to use for a call.
352  *
353  * If we return with a connection, the call will be on its waiting list.  It's
354  * left to the caller to assign a channel and wake up the call.
355  */
rxrpc_prep_call(struct rxrpc_sock * rx,struct rxrpc_call * call,struct rxrpc_conn_parameters * cp,struct sockaddr_rxrpc * srx,gfp_t gfp)356 static struct rxrpc_bundle *rxrpc_prep_call(struct rxrpc_sock *rx,
357 					    struct rxrpc_call *call,
358 					    struct rxrpc_conn_parameters *cp,
359 					    struct sockaddr_rxrpc *srx,
360 					    gfp_t gfp)
361 {
362 	struct rxrpc_bundle *bundle;
363 
364 	_enter("{%d,%lx},", call->debug_id, call->user_call_ID);
365 
366 	cp->peer = rxrpc_lookup_peer(rx, cp->local, srx, gfp);
367 	if (!cp->peer)
368 		goto error;
369 
370 	call->cong_cwnd = cp->peer->cong_cwnd;
371 	if (call->cong_cwnd >= call->cong_ssthresh)
372 		call->cong_mode = RXRPC_CALL_CONGEST_AVOIDANCE;
373 	else
374 		call->cong_mode = RXRPC_CALL_SLOW_START;
375 	if (cp->upgrade)
376 		__set_bit(RXRPC_CALL_UPGRADE, &call->flags);
377 
378 	/* Find the client connection bundle. */
379 	bundle = rxrpc_look_up_bundle(cp, gfp);
380 	if (!bundle)
381 		goto error;
382 
383 	/* Get this call queued.  Someone else may activate it whilst we're
384 	 * lining up a new connection, but that's fine.
385 	 */
386 	spin_lock(&bundle->channel_lock);
387 	list_add_tail(&call->chan_wait_link, &bundle->waiting_calls);
388 	spin_unlock(&bundle->channel_lock);
389 
390 	_leave(" = [B=%x]", bundle->debug_id);
391 	return bundle;
392 
393 error:
394 	_leave(" = -ENOMEM");
395 	return ERR_PTR(-ENOMEM);
396 }
397 
398 /*
399  * Allocate a new connection and add it into a bundle.
400  */
rxrpc_add_conn_to_bundle(struct rxrpc_bundle * bundle,gfp_t gfp)401 static void rxrpc_add_conn_to_bundle(struct rxrpc_bundle *bundle, gfp_t gfp)
402 	__releases(bundle->channel_lock)
403 {
404 	struct rxrpc_connection *candidate = NULL, *old = NULL;
405 	bool conflict;
406 	int i;
407 
408 	_enter("");
409 
410 	conflict = bundle->alloc_conn;
411 	if (!conflict)
412 		bundle->alloc_conn = true;
413 	spin_unlock(&bundle->channel_lock);
414 	if (conflict) {
415 		_leave(" [conf]");
416 		return;
417 	}
418 
419 	candidate = rxrpc_alloc_client_connection(bundle, gfp);
420 
421 	spin_lock(&bundle->channel_lock);
422 	bundle->alloc_conn = false;
423 
424 	if (IS_ERR(candidate)) {
425 		bundle->alloc_error = PTR_ERR(candidate);
426 		spin_unlock(&bundle->channel_lock);
427 		_leave(" [err %ld]", PTR_ERR(candidate));
428 		return;
429 	}
430 
431 	bundle->alloc_error = 0;
432 
433 	for (i = 0; i < ARRAY_SIZE(bundle->conns); i++) {
434 		unsigned int shift = i * RXRPC_MAXCALLS;
435 		int j;
436 
437 		old = bundle->conns[i];
438 		if (!rxrpc_may_reuse_conn(old)) {
439 			if (old)
440 				trace_rxrpc_client(old, -1, rxrpc_client_replace);
441 			candidate->bundle_shift = shift;
442 			atomic_inc(&bundle->active);
443 			bundle->conns[i] = candidate;
444 			for (j = 0; j < RXRPC_MAXCALLS; j++)
445 				set_bit(shift + j, &bundle->avail_chans);
446 			candidate = NULL;
447 			break;
448 		}
449 
450 		old = NULL;
451 	}
452 
453 	spin_unlock(&bundle->channel_lock);
454 
455 	if (candidate) {
456 		_debug("discard C=%x", candidate->debug_id);
457 		trace_rxrpc_client(candidate, -1, rxrpc_client_duplicate);
458 		rxrpc_put_connection(candidate);
459 	}
460 
461 	rxrpc_put_connection(old);
462 	_leave("");
463 }
464 
465 /*
466  * Add a connection to a bundle if there are no usable connections or we have
467  * connections waiting for extra capacity.
468  */
rxrpc_maybe_add_conn(struct rxrpc_bundle * bundle,gfp_t gfp)469 static void rxrpc_maybe_add_conn(struct rxrpc_bundle *bundle, gfp_t gfp)
470 {
471 	struct rxrpc_call *call;
472 	int i, usable;
473 
474 	_enter("");
475 
476 	spin_lock(&bundle->channel_lock);
477 
478 	/* See if there are any usable connections. */
479 	usable = 0;
480 	for (i = 0; i < ARRAY_SIZE(bundle->conns); i++)
481 		if (rxrpc_may_reuse_conn(bundle->conns[i]))
482 			usable++;
483 
484 	if (!usable && !list_empty(&bundle->waiting_calls)) {
485 		call = list_first_entry(&bundle->waiting_calls,
486 					struct rxrpc_call, chan_wait_link);
487 		if (test_bit(RXRPC_CALL_UPGRADE, &call->flags))
488 			bundle->try_upgrade = true;
489 	}
490 
491 	if (!usable)
492 		goto alloc_conn;
493 
494 	if (!bundle->avail_chans &&
495 	    !bundle->try_upgrade &&
496 	    !list_empty(&bundle->waiting_calls) &&
497 	    usable < ARRAY_SIZE(bundle->conns))
498 		goto alloc_conn;
499 
500 	spin_unlock(&bundle->channel_lock);
501 	_leave("");
502 	return;
503 
504 alloc_conn:
505 	return rxrpc_add_conn_to_bundle(bundle, gfp);
506 }
507 
508 /*
509  * Assign a channel to the call at the front of the queue and wake the call up.
510  * We don't increment the callNumber counter until this number has been exposed
511  * to the world.
512  */
rxrpc_activate_one_channel(struct rxrpc_connection * conn,unsigned int channel)513 static void rxrpc_activate_one_channel(struct rxrpc_connection *conn,
514 				       unsigned int channel)
515 {
516 	struct rxrpc_channel *chan = &conn->channels[channel];
517 	struct rxrpc_bundle *bundle = conn->bundle;
518 	struct rxrpc_call *call = list_entry(bundle->waiting_calls.next,
519 					     struct rxrpc_call, chan_wait_link);
520 	u32 call_id = chan->call_counter + 1;
521 
522 	_enter("C=%x,%u", conn->debug_id, channel);
523 
524 	trace_rxrpc_client(conn, channel, rxrpc_client_chan_activate);
525 
526 	/* Cancel the final ACK on the previous call if it hasn't been sent yet
527 	 * as the DATA packet will implicitly ACK it.
528 	 */
529 	clear_bit(RXRPC_CONN_FINAL_ACK_0 + channel, &conn->flags);
530 	clear_bit(conn->bundle_shift + channel, &bundle->avail_chans);
531 
532 	rxrpc_see_call(call);
533 	list_del_init(&call->chan_wait_link);
534 	call->peer	= rxrpc_get_peer(conn->params.peer);
535 	call->conn	= rxrpc_get_connection(conn);
536 	call->cid	= conn->proto.cid | channel;
537 	call->call_id	= call_id;
538 	call->security	= conn->security;
539 	call->security_ix = conn->security_ix;
540 	call->service_id = conn->service_id;
541 
542 	trace_rxrpc_connect_call(call);
543 	_net("CONNECT call %08x:%08x as call %d on conn %d",
544 	     call->cid, call->call_id, call->debug_id, conn->debug_id);
545 
546 	write_lock_bh(&call->state_lock);
547 	call->state = RXRPC_CALL_CLIENT_SEND_REQUEST;
548 	write_unlock_bh(&call->state_lock);
549 
550 	/* Paired with the read barrier in rxrpc_connect_call().  This orders
551 	 * cid and epoch in the connection wrt to call_id without the need to
552 	 * take the channel_lock.
553 	 *
554 	 * We provisionally assign a callNumber at this point, but we don't
555 	 * confirm it until the call is about to be exposed.
556 	 *
557 	 * TODO: Pair with a barrier in the data_ready handler when that looks
558 	 * at the call ID through a connection channel.
559 	 */
560 	smp_wmb();
561 
562 	chan->call_id		= call_id;
563 	chan->call_debug_id	= call->debug_id;
564 	rcu_assign_pointer(chan->call, call);
565 	wake_up(&call->waitq);
566 }
567 
568 /*
569  * Remove a connection from the idle list if it's on it.
570  */
rxrpc_unidle_conn(struct rxrpc_bundle * bundle,struct rxrpc_connection * conn)571 static void rxrpc_unidle_conn(struct rxrpc_bundle *bundle, struct rxrpc_connection *conn)
572 {
573 	struct rxrpc_net *rxnet = bundle->params.local->rxnet;
574 	bool drop_ref;
575 
576 	if (!list_empty(&conn->cache_link)) {
577 		drop_ref = false;
578 		spin_lock(&rxnet->client_conn_cache_lock);
579 		if (!list_empty(&conn->cache_link)) {
580 			list_del_init(&conn->cache_link);
581 			drop_ref = true;
582 		}
583 		spin_unlock(&rxnet->client_conn_cache_lock);
584 		if (drop_ref)
585 			rxrpc_put_connection(conn);
586 	}
587 }
588 
589 /*
590  * Assign channels and callNumbers to waiting calls with channel_lock
591  * held by caller.
592  */
rxrpc_activate_channels_locked(struct rxrpc_bundle * bundle)593 static void rxrpc_activate_channels_locked(struct rxrpc_bundle *bundle)
594 {
595 	struct rxrpc_connection *conn;
596 	unsigned long avail, mask;
597 	unsigned int channel, slot;
598 
599 	if (bundle->try_upgrade)
600 		mask = 1;
601 	else
602 		mask = ULONG_MAX;
603 
604 	while (!list_empty(&bundle->waiting_calls)) {
605 		avail = bundle->avail_chans & mask;
606 		if (!avail)
607 			break;
608 		channel = __ffs(avail);
609 		clear_bit(channel, &bundle->avail_chans);
610 
611 		slot = channel / RXRPC_MAXCALLS;
612 		conn = bundle->conns[slot];
613 		if (!conn)
614 			break;
615 
616 		if (bundle->try_upgrade)
617 			set_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags);
618 		rxrpc_unidle_conn(bundle, conn);
619 
620 		channel &= (RXRPC_MAXCALLS - 1);
621 		conn->act_chans	|= 1 << channel;
622 		rxrpc_activate_one_channel(conn, channel);
623 	}
624 }
625 
626 /*
627  * Assign channels and callNumbers to waiting calls.
628  */
rxrpc_activate_channels(struct rxrpc_bundle * bundle)629 static void rxrpc_activate_channels(struct rxrpc_bundle *bundle)
630 {
631 	_enter("B=%x", bundle->debug_id);
632 
633 	trace_rxrpc_client(NULL, -1, rxrpc_client_activate_chans);
634 
635 	if (!bundle->avail_chans)
636 		return;
637 
638 	spin_lock(&bundle->channel_lock);
639 	rxrpc_activate_channels_locked(bundle);
640 	spin_unlock(&bundle->channel_lock);
641 	_leave("");
642 }
643 
644 /*
645  * Wait for a callNumber and a channel to be granted to a call.
646  */
rxrpc_wait_for_channel(struct rxrpc_bundle * bundle,struct rxrpc_call * call,gfp_t gfp)647 static int rxrpc_wait_for_channel(struct rxrpc_bundle *bundle,
648 				  struct rxrpc_call *call, gfp_t gfp)
649 {
650 	DECLARE_WAITQUEUE(myself, current);
651 	int ret = 0;
652 
653 	_enter("%d", call->debug_id);
654 
655 	if (!gfpflags_allow_blocking(gfp)) {
656 		rxrpc_maybe_add_conn(bundle, gfp);
657 		rxrpc_activate_channels(bundle);
658 		ret = bundle->alloc_error ?: -EAGAIN;
659 		goto out;
660 	}
661 
662 	add_wait_queue_exclusive(&call->waitq, &myself);
663 	for (;;) {
664 		rxrpc_maybe_add_conn(bundle, gfp);
665 		rxrpc_activate_channels(bundle);
666 		ret = bundle->alloc_error;
667 		if (ret < 0)
668 			break;
669 
670 		switch (call->interruptibility) {
671 		case RXRPC_INTERRUPTIBLE:
672 		case RXRPC_PREINTERRUPTIBLE:
673 			set_current_state(TASK_INTERRUPTIBLE);
674 			break;
675 		case RXRPC_UNINTERRUPTIBLE:
676 		default:
677 			set_current_state(TASK_UNINTERRUPTIBLE);
678 			break;
679 		}
680 		if (READ_ONCE(call->state) != RXRPC_CALL_CLIENT_AWAIT_CONN)
681 			break;
682 		if ((call->interruptibility == RXRPC_INTERRUPTIBLE ||
683 		     call->interruptibility == RXRPC_PREINTERRUPTIBLE) &&
684 		    signal_pending(current)) {
685 			ret = -ERESTARTSYS;
686 			break;
687 		}
688 		schedule();
689 	}
690 	remove_wait_queue(&call->waitq, &myself);
691 	__set_current_state(TASK_RUNNING);
692 
693 out:
694 	_leave(" = %d", ret);
695 	return ret;
696 }
697 
698 /*
699  * find a connection for a call
700  * - called in process context with IRQs enabled
701  */
rxrpc_connect_call(struct rxrpc_sock * rx,struct rxrpc_call * call,struct rxrpc_conn_parameters * cp,struct sockaddr_rxrpc * srx,gfp_t gfp)702 int rxrpc_connect_call(struct rxrpc_sock *rx,
703 		       struct rxrpc_call *call,
704 		       struct rxrpc_conn_parameters *cp,
705 		       struct sockaddr_rxrpc *srx,
706 		       gfp_t gfp)
707 {
708 	struct rxrpc_bundle *bundle;
709 	struct rxrpc_net *rxnet = cp->local->rxnet;
710 	int ret = 0;
711 
712 	_enter("{%d,%lx},", call->debug_id, call->user_call_ID);
713 
714 	rxrpc_discard_expired_client_conns(&rxnet->client_conn_reaper);
715 
716 	bundle = rxrpc_prep_call(rx, call, cp, srx, gfp);
717 	if (IS_ERR(bundle)) {
718 		ret = PTR_ERR(bundle);
719 		goto out;
720 	}
721 
722 	if (call->state == RXRPC_CALL_CLIENT_AWAIT_CONN) {
723 		ret = rxrpc_wait_for_channel(bundle, call, gfp);
724 		if (ret < 0)
725 			goto wait_failed;
726 	}
727 
728 granted_channel:
729 	/* Paired with the write barrier in rxrpc_activate_one_channel(). */
730 	smp_rmb();
731 
732 out_put_bundle:
733 	rxrpc_deactivate_bundle(bundle);
734 	rxrpc_put_bundle(bundle);
735 out:
736 	_leave(" = %d", ret);
737 	return ret;
738 
739 wait_failed:
740 	spin_lock(&bundle->channel_lock);
741 	list_del_init(&call->chan_wait_link);
742 	spin_unlock(&bundle->channel_lock);
743 
744 	if (call->state != RXRPC_CALL_CLIENT_AWAIT_CONN) {
745 		ret = 0;
746 		goto granted_channel;
747 	}
748 
749 	trace_rxrpc_client(call->conn, ret, rxrpc_client_chan_wait_failed);
750 	rxrpc_set_call_completion(call, RXRPC_CALL_LOCAL_ERROR, 0, ret);
751 	rxrpc_disconnect_client_call(bundle, call);
752 	goto out_put_bundle;
753 }
754 
755 /*
756  * Note that a call, and thus a connection, is about to be exposed to the
757  * world.
758  */
rxrpc_expose_client_call(struct rxrpc_call * call)759 void rxrpc_expose_client_call(struct rxrpc_call *call)
760 {
761 	unsigned int channel = call->cid & RXRPC_CHANNELMASK;
762 	struct rxrpc_connection *conn = call->conn;
763 	struct rxrpc_channel *chan = &conn->channels[channel];
764 
765 	if (!test_and_set_bit(RXRPC_CALL_EXPOSED, &call->flags)) {
766 		/* Mark the call ID as being used.  If the callNumber counter
767 		 * exceeds ~2 billion, we kill the connection after its
768 		 * outstanding calls have finished so that the counter doesn't
769 		 * wrap.
770 		 */
771 		chan->call_counter++;
772 		if (chan->call_counter >= INT_MAX)
773 			set_bit(RXRPC_CONN_DONT_REUSE, &conn->flags);
774 		trace_rxrpc_client(conn, channel, rxrpc_client_exposed);
775 	}
776 }
777 
778 /*
779  * Set the reap timer.
780  */
rxrpc_set_client_reap_timer(struct rxrpc_net * rxnet)781 static void rxrpc_set_client_reap_timer(struct rxrpc_net *rxnet)
782 {
783 	if (!rxnet->kill_all_client_conns) {
784 		unsigned long now = jiffies;
785 		unsigned long reap_at = now + rxrpc_conn_idle_client_expiry;
786 
787 		if (rxnet->live)
788 			timer_reduce(&rxnet->client_conn_reap_timer, reap_at);
789 	}
790 }
791 
792 /*
793  * Disconnect a client call.
794  */
rxrpc_disconnect_client_call(struct rxrpc_bundle * bundle,struct rxrpc_call * call)795 void rxrpc_disconnect_client_call(struct rxrpc_bundle *bundle, struct rxrpc_call *call)
796 {
797 	struct rxrpc_connection *conn;
798 	struct rxrpc_channel *chan = NULL;
799 	struct rxrpc_net *rxnet = bundle->params.local->rxnet;
800 	unsigned int channel;
801 	bool may_reuse;
802 	u32 cid;
803 
804 	_enter("c=%x", call->debug_id);
805 
806 	spin_lock(&bundle->channel_lock);
807 	set_bit(RXRPC_CALL_DISCONNECTED, &call->flags);
808 
809 	/* Calls that have never actually been assigned a channel can simply be
810 	 * discarded.
811 	 */
812 	conn = call->conn;
813 	if (!conn) {
814 		_debug("call is waiting");
815 		ASSERTCMP(call->call_id, ==, 0);
816 		ASSERT(!test_bit(RXRPC_CALL_EXPOSED, &call->flags));
817 		list_del_init(&call->chan_wait_link);
818 		goto out;
819 	}
820 
821 	cid = call->cid;
822 	channel = cid & RXRPC_CHANNELMASK;
823 	chan = &conn->channels[channel];
824 	trace_rxrpc_client(conn, channel, rxrpc_client_chan_disconnect);
825 
826 	if (rcu_access_pointer(chan->call) != call) {
827 		spin_unlock(&bundle->channel_lock);
828 		BUG();
829 	}
830 
831 	may_reuse = rxrpc_may_reuse_conn(conn);
832 
833 	/* If a client call was exposed to the world, we save the result for
834 	 * retransmission.
835 	 *
836 	 * We use a barrier here so that the call number and abort code can be
837 	 * read without needing to take a lock.
838 	 *
839 	 * TODO: Make the incoming packet handler check this and handle
840 	 * terminal retransmission without requiring access to the call.
841 	 */
842 	if (test_bit(RXRPC_CALL_EXPOSED, &call->flags)) {
843 		_debug("exposed %u,%u", call->call_id, call->abort_code);
844 		__rxrpc_disconnect_call(conn, call);
845 
846 		if (test_and_clear_bit(RXRPC_CONN_PROBING_FOR_UPGRADE, &conn->flags)) {
847 			trace_rxrpc_client(conn, channel, rxrpc_client_to_active);
848 			bundle->try_upgrade = false;
849 			if (may_reuse)
850 				rxrpc_activate_channels_locked(bundle);
851 		}
852 
853 	}
854 
855 	/* See if we can pass the channel directly to another call. */
856 	if (may_reuse && !list_empty(&bundle->waiting_calls)) {
857 		trace_rxrpc_client(conn, channel, rxrpc_client_chan_pass);
858 		rxrpc_activate_one_channel(conn, channel);
859 		goto out;
860 	}
861 
862 	/* Schedule the final ACK to be transmitted in a short while so that it
863 	 * can be skipped if we find a follow-on call.  The first DATA packet
864 	 * of the follow on call will implicitly ACK this call.
865 	 */
866 	if (call->completion == RXRPC_CALL_SUCCEEDED &&
867 	    test_bit(RXRPC_CALL_EXPOSED, &call->flags)) {
868 		unsigned long final_ack_at = jiffies + 2;
869 
870 		WRITE_ONCE(chan->final_ack_at, final_ack_at);
871 		smp_wmb(); /* vs rxrpc_process_delayed_final_acks() */
872 		set_bit(RXRPC_CONN_FINAL_ACK_0 + channel, &conn->flags);
873 		rxrpc_reduce_conn_timer(conn, final_ack_at);
874 	}
875 
876 	/* Deactivate the channel. */
877 	rcu_assign_pointer(chan->call, NULL);
878 	set_bit(conn->bundle_shift + channel, &conn->bundle->avail_chans);
879 	conn->act_chans	&= ~(1 << channel);
880 
881 	/* If no channels remain active, then put the connection on the idle
882 	 * list for a short while.  Give it a ref to stop it going away if it
883 	 * becomes unbundled.
884 	 */
885 	if (!conn->act_chans) {
886 		trace_rxrpc_client(conn, channel, rxrpc_client_to_idle);
887 		conn->idle_timestamp = jiffies;
888 
889 		rxrpc_get_connection(conn);
890 		spin_lock(&rxnet->client_conn_cache_lock);
891 		list_move_tail(&conn->cache_link, &rxnet->idle_client_conns);
892 		spin_unlock(&rxnet->client_conn_cache_lock);
893 
894 		rxrpc_set_client_reap_timer(rxnet);
895 	}
896 
897 out:
898 	spin_unlock(&bundle->channel_lock);
899 	_leave("");
900 	return;
901 }
902 
903 /*
904  * Remove a connection from a bundle.
905  */
rxrpc_unbundle_conn(struct rxrpc_connection * conn)906 static void rxrpc_unbundle_conn(struct rxrpc_connection *conn)
907 {
908 	struct rxrpc_bundle *bundle = conn->bundle;
909 	unsigned int bindex;
910 	bool need_drop = false;
911 	int i;
912 
913 	_enter("C=%x", conn->debug_id);
914 
915 	if (conn->flags & RXRPC_CONN_FINAL_ACK_MASK)
916 		rxrpc_process_delayed_final_acks(conn, true);
917 
918 	spin_lock(&bundle->channel_lock);
919 	bindex = conn->bundle_shift / RXRPC_MAXCALLS;
920 	if (bundle->conns[bindex] == conn) {
921 		_debug("clear slot %u", bindex);
922 		bundle->conns[bindex] = NULL;
923 		for (i = 0; i < RXRPC_MAXCALLS; i++)
924 			clear_bit(conn->bundle_shift + i, &bundle->avail_chans);
925 		need_drop = true;
926 	}
927 	spin_unlock(&bundle->channel_lock);
928 
929 	if (need_drop) {
930 		rxrpc_deactivate_bundle(bundle);
931 		rxrpc_put_connection(conn);
932 	}
933 }
934 
935 /*
936  * Drop the active count on a bundle.
937  */
rxrpc_deactivate_bundle(struct rxrpc_bundle * bundle)938 static void rxrpc_deactivate_bundle(struct rxrpc_bundle *bundle)
939 {
940 	struct rxrpc_local *local = bundle->params.local;
941 	bool need_put = false;
942 
943 	if (atomic_dec_and_lock(&bundle->active, &local->client_bundles_lock)) {
944 		if (!bundle->params.exclusive) {
945 			_debug("erase bundle");
946 			rb_erase(&bundle->local_node, &local->client_bundles);
947 			need_put = true;
948 		}
949 
950 		spin_unlock(&local->client_bundles_lock);
951 		if (need_put)
952 			rxrpc_put_bundle(bundle);
953 	}
954 }
955 
956 /*
957  * Clean up a dead client connection.
958  */
rxrpc_kill_client_conn(struct rxrpc_connection * conn)959 static void rxrpc_kill_client_conn(struct rxrpc_connection *conn)
960 {
961 	struct rxrpc_local *local = conn->params.local;
962 	struct rxrpc_net *rxnet = local->rxnet;
963 
964 	_enter("C=%x", conn->debug_id);
965 
966 	trace_rxrpc_client(conn, -1, rxrpc_client_cleanup);
967 	atomic_dec(&rxnet->nr_client_conns);
968 
969 	rxrpc_put_client_connection_id(conn);
970 	rxrpc_kill_connection(conn);
971 }
972 
973 /*
974  * Clean up a dead client connections.
975  */
rxrpc_put_client_conn(struct rxrpc_connection * conn)976 void rxrpc_put_client_conn(struct rxrpc_connection *conn)
977 {
978 	const void *here = __builtin_return_address(0);
979 	unsigned int debug_id = conn->debug_id;
980 	bool dead;
981 	int r;
982 
983 	dead = __refcount_dec_and_test(&conn->ref, &r);
984 	trace_rxrpc_conn(debug_id, rxrpc_conn_put_client, r - 1, here);
985 	if (dead)
986 		rxrpc_kill_client_conn(conn);
987 }
988 
989 /*
990  * Discard expired client connections from the idle list.  Each conn in the
991  * idle list has been exposed and holds an extra ref because of that.
992  *
993  * This may be called from conn setup or from a work item so cannot be
994  * considered non-reentrant.
995  */
rxrpc_discard_expired_client_conns(struct work_struct * work)996 void rxrpc_discard_expired_client_conns(struct work_struct *work)
997 {
998 	struct rxrpc_connection *conn;
999 	struct rxrpc_net *rxnet =
1000 		container_of(work, struct rxrpc_net, client_conn_reaper);
1001 	unsigned long expiry, conn_expires_at, now;
1002 	unsigned int nr_conns;
1003 
1004 	_enter("");
1005 
1006 	if (list_empty(&rxnet->idle_client_conns)) {
1007 		_leave(" [empty]");
1008 		return;
1009 	}
1010 
1011 	/* Don't double up on the discarding */
1012 	if (!spin_trylock(&rxnet->client_conn_discard_lock)) {
1013 		_leave(" [already]");
1014 		return;
1015 	}
1016 
1017 	/* We keep an estimate of what the number of conns ought to be after
1018 	 * we've discarded some so that we don't overdo the discarding.
1019 	 */
1020 	nr_conns = atomic_read(&rxnet->nr_client_conns);
1021 
1022 next:
1023 	spin_lock(&rxnet->client_conn_cache_lock);
1024 
1025 	if (list_empty(&rxnet->idle_client_conns))
1026 		goto out;
1027 
1028 	conn = list_entry(rxnet->idle_client_conns.next,
1029 			  struct rxrpc_connection, cache_link);
1030 
1031 	if (!rxnet->kill_all_client_conns) {
1032 		/* If the number of connections is over the reap limit, we
1033 		 * expedite discard by reducing the expiry timeout.  We must,
1034 		 * however, have at least a short grace period to be able to do
1035 		 * final-ACK or ABORT retransmission.
1036 		 */
1037 		expiry = rxrpc_conn_idle_client_expiry;
1038 		if (nr_conns > rxrpc_reap_client_connections)
1039 			expiry = rxrpc_conn_idle_client_fast_expiry;
1040 		if (conn->params.local->service_closed)
1041 			expiry = rxrpc_closed_conn_expiry * HZ;
1042 
1043 		conn_expires_at = conn->idle_timestamp + expiry;
1044 
1045 		now = READ_ONCE(jiffies);
1046 		if (time_after(conn_expires_at, now))
1047 			goto not_yet_expired;
1048 	}
1049 
1050 	trace_rxrpc_client(conn, -1, rxrpc_client_discard);
1051 	list_del_init(&conn->cache_link);
1052 
1053 	spin_unlock(&rxnet->client_conn_cache_lock);
1054 
1055 	rxrpc_unbundle_conn(conn);
1056 	rxrpc_put_connection(conn); /* Drop the ->cache_link ref */
1057 
1058 	nr_conns--;
1059 	goto next;
1060 
1061 not_yet_expired:
1062 	/* The connection at the front of the queue hasn't yet expired, so
1063 	 * schedule the work item for that point if we discarded something.
1064 	 *
1065 	 * We don't worry if the work item is already scheduled - it can look
1066 	 * after rescheduling itself at a later time.  We could cancel it, but
1067 	 * then things get messier.
1068 	 */
1069 	_debug("not yet");
1070 	if (!rxnet->kill_all_client_conns)
1071 		timer_reduce(&rxnet->client_conn_reap_timer, conn_expires_at);
1072 
1073 out:
1074 	spin_unlock(&rxnet->client_conn_cache_lock);
1075 	spin_unlock(&rxnet->client_conn_discard_lock);
1076 	_leave("");
1077 }
1078 
1079 /*
1080  * Preemptively destroy all the client connection records rather than waiting
1081  * for them to time out
1082  */
rxrpc_destroy_all_client_connections(struct rxrpc_net * rxnet)1083 void rxrpc_destroy_all_client_connections(struct rxrpc_net *rxnet)
1084 {
1085 	_enter("");
1086 
1087 	spin_lock(&rxnet->client_conn_cache_lock);
1088 	rxnet->kill_all_client_conns = true;
1089 	spin_unlock(&rxnet->client_conn_cache_lock);
1090 
1091 	del_timer_sync(&rxnet->client_conn_reap_timer);
1092 
1093 	if (!rxrpc_queue_work(&rxnet->client_conn_reaper))
1094 		_debug("destroy: queue failed");
1095 
1096 	_leave("");
1097 }
1098 
1099 /*
1100  * Clean up the client connections on a local endpoint.
1101  */
rxrpc_clean_up_local_conns(struct rxrpc_local * local)1102 void rxrpc_clean_up_local_conns(struct rxrpc_local *local)
1103 {
1104 	struct rxrpc_connection *conn, *tmp;
1105 	struct rxrpc_net *rxnet = local->rxnet;
1106 	LIST_HEAD(graveyard);
1107 
1108 	_enter("");
1109 
1110 	spin_lock(&rxnet->client_conn_cache_lock);
1111 
1112 	list_for_each_entry_safe(conn, tmp, &rxnet->idle_client_conns,
1113 				 cache_link) {
1114 		if (conn->params.local == local) {
1115 			trace_rxrpc_client(conn, -1, rxrpc_client_discard);
1116 			list_move(&conn->cache_link, &graveyard);
1117 		}
1118 	}
1119 
1120 	spin_unlock(&rxnet->client_conn_cache_lock);
1121 
1122 	while (!list_empty(&graveyard)) {
1123 		conn = list_entry(graveyard.next,
1124 				  struct rxrpc_connection, cache_link);
1125 		list_del_init(&conn->cache_link);
1126 		rxrpc_unbundle_conn(conn);
1127 		rxrpc_put_connection(conn);
1128 	}
1129 
1130 	_leave(" [culled]");
1131 }
1132