• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* RxRPC individual remote procedure call handling
2  *
3  * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4  * Written by David Howells (dhowells@redhat.com)
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License
8  * as published by the Free Software Foundation; either version
9  * 2 of the License, or (at your option) any later version.
10  */
11 
12 #include <linux/module.h>
13 #include <linux/circ_buf.h>
14 #include <net/sock.h>
15 #include <net/af_rxrpc.h>
16 #include "ar-internal.h"
17 
18 const char *const rxrpc_call_states[] = {
19 	[RXRPC_CALL_CLIENT_SEND_REQUEST]	= "ClSndReq",
20 	[RXRPC_CALL_CLIENT_AWAIT_REPLY]		= "ClAwtRpl",
21 	[RXRPC_CALL_CLIENT_RECV_REPLY]		= "ClRcvRpl",
22 	[RXRPC_CALL_CLIENT_FINAL_ACK]		= "ClFnlACK",
23 	[RXRPC_CALL_SERVER_SECURING]		= "SvSecure",
24 	[RXRPC_CALL_SERVER_ACCEPTING]		= "SvAccept",
25 	[RXRPC_CALL_SERVER_RECV_REQUEST]	= "SvRcvReq",
26 	[RXRPC_CALL_SERVER_ACK_REQUEST]		= "SvAckReq",
27 	[RXRPC_CALL_SERVER_SEND_REPLY]		= "SvSndRpl",
28 	[RXRPC_CALL_SERVER_AWAIT_ACK]		= "SvAwtACK",
29 	[RXRPC_CALL_COMPLETE]			= "Complete",
30 	[RXRPC_CALL_SERVER_BUSY]		= "SvBusy  ",
31 	[RXRPC_CALL_REMOTELY_ABORTED]		= "RmtAbort",
32 	[RXRPC_CALL_LOCALLY_ABORTED]		= "LocAbort",
33 	[RXRPC_CALL_NETWORK_ERROR]		= "NetError",
34 	[RXRPC_CALL_DEAD]			= "Dead    ",
35 };
36 
37 struct kmem_cache *rxrpc_call_jar;
38 LIST_HEAD(rxrpc_calls);
39 DEFINE_RWLOCK(rxrpc_call_lock);
40 static unsigned rxrpc_call_max_lifetime = 60;
41 static unsigned rxrpc_dead_call_timeout = 2;
42 
43 static void rxrpc_destroy_call(struct work_struct *work);
44 static void rxrpc_call_life_expired(unsigned long _call);
45 static void rxrpc_dead_call_expired(unsigned long _call);
46 static void rxrpc_ack_time_expired(unsigned long _call);
47 static void rxrpc_resend_time_expired(unsigned long _call);
48 
49 /*
50  * allocate a new call
51  */
rxrpc_alloc_call(gfp_t gfp)52 static struct rxrpc_call *rxrpc_alloc_call(gfp_t gfp)
53 {
54 	struct rxrpc_call *call;
55 
56 	call = kmem_cache_zalloc(rxrpc_call_jar, gfp);
57 	if (!call)
58 		return NULL;
59 
60 	call->acks_winsz = 16;
61 	call->acks_window = kmalloc(call->acks_winsz * sizeof(unsigned long),
62 				    gfp);
63 	if (!call->acks_window) {
64 		kmem_cache_free(rxrpc_call_jar, call);
65 		return NULL;
66 	}
67 
68 	setup_timer(&call->lifetimer, &rxrpc_call_life_expired,
69 		    (unsigned long) call);
70 	setup_timer(&call->deadspan, &rxrpc_dead_call_expired,
71 		    (unsigned long) call);
72 	setup_timer(&call->ack_timer, &rxrpc_ack_time_expired,
73 		    (unsigned long) call);
74 	setup_timer(&call->resend_timer, &rxrpc_resend_time_expired,
75 		    (unsigned long) call);
76 	INIT_WORK(&call->destroyer, &rxrpc_destroy_call);
77 	INIT_WORK(&call->processor, &rxrpc_process_call);
78 	INIT_LIST_HEAD(&call->accept_link);
79 	skb_queue_head_init(&call->rx_queue);
80 	skb_queue_head_init(&call->rx_oos_queue);
81 	init_waitqueue_head(&call->tx_waitq);
82 	spin_lock_init(&call->lock);
83 	rwlock_init(&call->state_lock);
84 	atomic_set(&call->usage, 1);
85 	call->debug_id = atomic_inc_return(&rxrpc_debug_id);
86 	call->state = RXRPC_CALL_CLIENT_SEND_REQUEST;
87 
88 	memset(&call->sock_node, 0xed, sizeof(call->sock_node));
89 
90 	call->rx_data_expect = 1;
91 	call->rx_data_eaten = 0;
92 	call->rx_first_oos = 0;
93 	call->ackr_win_top = call->rx_data_eaten + 1 + RXRPC_MAXACKS;
94 	call->creation_jif = jiffies;
95 	return call;
96 }
97 
98 /*
99  * allocate a new client call and attempt to to get a connection slot for it
100  */
rxrpc_alloc_client_call(struct rxrpc_sock * rx,struct rxrpc_transport * trans,struct rxrpc_conn_bundle * bundle,gfp_t gfp)101 static struct rxrpc_call *rxrpc_alloc_client_call(
102 	struct rxrpc_sock *rx,
103 	struct rxrpc_transport *trans,
104 	struct rxrpc_conn_bundle *bundle,
105 	gfp_t gfp)
106 {
107 	struct rxrpc_call *call;
108 	int ret;
109 
110 	_enter("");
111 
112 	ASSERT(rx != NULL);
113 	ASSERT(trans != NULL);
114 	ASSERT(bundle != NULL);
115 
116 	call = rxrpc_alloc_call(gfp);
117 	if (!call)
118 		return ERR_PTR(-ENOMEM);
119 
120 	sock_hold(&rx->sk);
121 	call->socket = rx;
122 	call->rx_data_post = 1;
123 
124 	ret = rxrpc_connect_call(rx, trans, bundle, call, gfp);
125 	if (ret < 0) {
126 		kmem_cache_free(rxrpc_call_jar, call);
127 		return ERR_PTR(ret);
128 	}
129 
130 	spin_lock(&call->conn->trans->peer->lock);
131 	list_add(&call->error_link, &call->conn->trans->peer->error_targets);
132 	spin_unlock(&call->conn->trans->peer->lock);
133 
134 	call->lifetimer.expires = jiffies + rxrpc_call_max_lifetime * HZ;
135 	add_timer(&call->lifetimer);
136 
137 	_leave(" = %p", call);
138 	return call;
139 }
140 
141 /*
142  * set up a call for the given data
143  * - called in process context with IRQs enabled
144  */
rxrpc_get_client_call(struct rxrpc_sock * rx,struct rxrpc_transport * trans,struct rxrpc_conn_bundle * bundle,unsigned long user_call_ID,int create,gfp_t gfp)145 struct rxrpc_call *rxrpc_get_client_call(struct rxrpc_sock *rx,
146 					 struct rxrpc_transport *trans,
147 					 struct rxrpc_conn_bundle *bundle,
148 					 unsigned long user_call_ID,
149 					 int create,
150 					 gfp_t gfp)
151 {
152 	struct rxrpc_call *call, *candidate;
153 	struct rb_node *p, *parent, **pp;
154 
155 	_enter("%p,%d,%d,%lx,%d",
156 	       rx, trans ? trans->debug_id : -1, bundle ? bundle->debug_id : -1,
157 	       user_call_ID, create);
158 
159 	/* search the extant calls first for one that matches the specified
160 	 * user ID */
161 	read_lock(&rx->call_lock);
162 
163 	p = rx->calls.rb_node;
164 	while (p) {
165 		call = rb_entry(p, struct rxrpc_call, sock_node);
166 
167 		if (user_call_ID < call->user_call_ID)
168 			p = p->rb_left;
169 		else if (user_call_ID > call->user_call_ID)
170 			p = p->rb_right;
171 		else
172 			goto found_extant_call;
173 	}
174 
175 	read_unlock(&rx->call_lock);
176 
177 	if (!create || !trans)
178 		return ERR_PTR(-EBADSLT);
179 
180 	/* not yet present - create a candidate for a new record and then
181 	 * redo the search */
182 	candidate = rxrpc_alloc_client_call(rx, trans, bundle, gfp);
183 	if (IS_ERR(candidate)) {
184 		_leave(" = %ld", PTR_ERR(candidate));
185 		return candidate;
186 	}
187 
188 	candidate->user_call_ID = user_call_ID;
189 	__set_bit(RXRPC_CALL_HAS_USERID, &candidate->flags);
190 
191 	write_lock(&rx->call_lock);
192 
193 	pp = &rx->calls.rb_node;
194 	parent = NULL;
195 	while (*pp) {
196 		parent = *pp;
197 		call = rb_entry(parent, struct rxrpc_call, sock_node);
198 
199 		if (user_call_ID < call->user_call_ID)
200 			pp = &(*pp)->rb_left;
201 		else if (user_call_ID > call->user_call_ID)
202 			pp = &(*pp)->rb_right;
203 		else
204 			goto found_extant_second;
205 	}
206 
207 	/* second search also failed; add the new call */
208 	call = candidate;
209 	candidate = NULL;
210 	rxrpc_get_call(call);
211 
212 	rb_link_node(&call->sock_node, parent, pp);
213 	rb_insert_color(&call->sock_node, &rx->calls);
214 	write_unlock(&rx->call_lock);
215 
216 	write_lock_bh(&rxrpc_call_lock);
217 	list_add_tail(&call->link, &rxrpc_calls);
218 	write_unlock_bh(&rxrpc_call_lock);
219 
220 	_net("CALL new %d on CONN %d", call->debug_id, call->conn->debug_id);
221 
222 	_leave(" = %p [new]", call);
223 	return call;
224 
225 	/* we found the call in the list immediately */
226 found_extant_call:
227 	rxrpc_get_call(call);
228 	read_unlock(&rx->call_lock);
229 	_leave(" = %p [extant %d]", call, atomic_read(&call->usage));
230 	return call;
231 
232 	/* we found the call on the second time through the list */
233 found_extant_second:
234 	rxrpc_get_call(call);
235 	write_unlock(&rx->call_lock);
236 	rxrpc_put_call(candidate);
237 	_leave(" = %p [second %d]", call, atomic_read(&call->usage));
238 	return call;
239 }
240 
241 /*
242  * set up an incoming call
243  * - called in process context with IRQs enabled
244  */
rxrpc_incoming_call(struct rxrpc_sock * rx,struct rxrpc_connection * conn,struct rxrpc_header * hdr,gfp_t gfp)245 struct rxrpc_call *rxrpc_incoming_call(struct rxrpc_sock *rx,
246 				       struct rxrpc_connection *conn,
247 				       struct rxrpc_header *hdr,
248 				       gfp_t gfp)
249 {
250 	struct rxrpc_call *call, *candidate;
251 	struct rb_node **p, *parent;
252 	__be32 call_id;
253 
254 	_enter(",%d,,%x", conn->debug_id, gfp);
255 
256 	ASSERT(rx != NULL);
257 
258 	candidate = rxrpc_alloc_call(gfp);
259 	if (!candidate)
260 		return ERR_PTR(-EBUSY);
261 
262 	candidate->socket = rx;
263 	candidate->conn = conn;
264 	candidate->cid = hdr->cid;
265 	candidate->call_id = hdr->callNumber;
266 	candidate->channel = ntohl(hdr->cid) & RXRPC_CHANNELMASK;
267 	candidate->rx_data_post = 0;
268 	candidate->state = RXRPC_CALL_SERVER_ACCEPTING;
269 	if (conn->security_ix > 0)
270 		candidate->state = RXRPC_CALL_SERVER_SECURING;
271 
272 	write_lock_bh(&conn->lock);
273 
274 	/* set the channel for this call */
275 	call = conn->channels[candidate->channel];
276 	_debug("channel[%u] is %p", candidate->channel, call);
277 	if (call && call->call_id == hdr->callNumber) {
278 		/* already set; must've been a duplicate packet */
279 		_debug("extant call [%d]", call->state);
280 		ASSERTCMP(call->conn, ==, conn);
281 
282 		read_lock(&call->state_lock);
283 		switch (call->state) {
284 		case RXRPC_CALL_LOCALLY_ABORTED:
285 			if (!test_and_set_bit(RXRPC_CALL_ABORT, &call->events))
286 				rxrpc_queue_call(call);
287 		case RXRPC_CALL_REMOTELY_ABORTED:
288 			read_unlock(&call->state_lock);
289 			goto aborted_call;
290 		default:
291 			rxrpc_get_call(call);
292 			read_unlock(&call->state_lock);
293 			goto extant_call;
294 		}
295 	}
296 
297 	if (call) {
298 		/* it seems the channel is still in use from the previous call
299 		 * - ditch the old binding if its call is now complete */
300 		_debug("CALL: %u { %s }",
301 		       call->debug_id, rxrpc_call_states[call->state]);
302 
303 		if (call->state >= RXRPC_CALL_COMPLETE) {
304 			conn->channels[call->channel] = NULL;
305 		} else {
306 			write_unlock_bh(&conn->lock);
307 			kmem_cache_free(rxrpc_call_jar, candidate);
308 			_leave(" = -EBUSY");
309 			return ERR_PTR(-EBUSY);
310 		}
311 	}
312 
313 	/* check the call number isn't duplicate */
314 	_debug("check dup");
315 	call_id = hdr->callNumber;
316 	p = &conn->calls.rb_node;
317 	parent = NULL;
318 	while (*p) {
319 		parent = *p;
320 		call = rb_entry(parent, struct rxrpc_call, conn_node);
321 
322 		if (call_id < call->call_id)
323 			p = &(*p)->rb_left;
324 		else if (call_id > call->call_id)
325 			p = &(*p)->rb_right;
326 		else
327 			goto old_call;
328 	}
329 
330 	/* make the call available */
331 	_debug("new call");
332 	call = candidate;
333 	candidate = NULL;
334 	rb_link_node(&call->conn_node, parent, p);
335 	rb_insert_color(&call->conn_node, &conn->calls);
336 	conn->channels[call->channel] = call;
337 	sock_hold(&rx->sk);
338 	atomic_inc(&conn->usage);
339 	write_unlock_bh(&conn->lock);
340 
341 	spin_lock(&conn->trans->peer->lock);
342 	list_add(&call->error_link, &conn->trans->peer->error_targets);
343 	spin_unlock(&conn->trans->peer->lock);
344 
345 	write_lock_bh(&rxrpc_call_lock);
346 	list_add_tail(&call->link, &rxrpc_calls);
347 	write_unlock_bh(&rxrpc_call_lock);
348 
349 	_net("CALL incoming %d on CONN %d", call->debug_id, call->conn->debug_id);
350 
351 	call->lifetimer.expires = jiffies + rxrpc_call_max_lifetime * HZ;
352 	add_timer(&call->lifetimer);
353 	_leave(" = %p {%d} [new]", call, call->debug_id);
354 	return call;
355 
356 extant_call:
357 	write_unlock_bh(&conn->lock);
358 	kmem_cache_free(rxrpc_call_jar, candidate);
359 	_leave(" = %p {%d} [extant]", call, call ? call->debug_id : -1);
360 	return call;
361 
362 aborted_call:
363 	write_unlock_bh(&conn->lock);
364 	kmem_cache_free(rxrpc_call_jar, candidate);
365 	_leave(" = -ECONNABORTED");
366 	return ERR_PTR(-ECONNABORTED);
367 
368 old_call:
369 	write_unlock_bh(&conn->lock);
370 	kmem_cache_free(rxrpc_call_jar, candidate);
371 	_leave(" = -ECONNRESET [old]");
372 	return ERR_PTR(-ECONNRESET);
373 }
374 
375 /*
376  * find an extant server call
377  * - called in process context with IRQs enabled
378  */
rxrpc_find_server_call(struct rxrpc_sock * rx,unsigned long user_call_ID)379 struct rxrpc_call *rxrpc_find_server_call(struct rxrpc_sock *rx,
380 					  unsigned long user_call_ID)
381 {
382 	struct rxrpc_call *call;
383 	struct rb_node *p;
384 
385 	_enter("%p,%lx", rx, user_call_ID);
386 
387 	/* search the extant calls for one that matches the specified user
388 	 * ID */
389 	read_lock(&rx->call_lock);
390 
391 	p = rx->calls.rb_node;
392 	while (p) {
393 		call = rb_entry(p, struct rxrpc_call, sock_node);
394 
395 		if (user_call_ID < call->user_call_ID)
396 			p = p->rb_left;
397 		else if (user_call_ID > call->user_call_ID)
398 			p = p->rb_right;
399 		else
400 			goto found_extant_call;
401 	}
402 
403 	read_unlock(&rx->call_lock);
404 	_leave(" = NULL");
405 	return NULL;
406 
407 	/* we found the call in the list immediately */
408 found_extant_call:
409 	rxrpc_get_call(call);
410 	read_unlock(&rx->call_lock);
411 	_leave(" = %p [%d]", call, atomic_read(&call->usage));
412 	return call;
413 }
414 
415 /*
416  * detach a call from a socket and set up for release
417  */
rxrpc_release_call(struct rxrpc_call * call)418 void rxrpc_release_call(struct rxrpc_call *call)
419 {
420 	struct rxrpc_connection *conn = call->conn;
421 	struct rxrpc_sock *rx = call->socket;
422 
423 	_enter("{%d,%d,%d,%d}",
424 	       call->debug_id, atomic_read(&call->usage),
425 	       atomic_read(&call->ackr_not_idle),
426 	       call->rx_first_oos);
427 
428 	spin_lock_bh(&call->lock);
429 	if (test_and_set_bit(RXRPC_CALL_RELEASED, &call->flags))
430 		BUG();
431 	spin_unlock_bh(&call->lock);
432 
433 	/* dissociate from the socket
434 	 * - the socket's ref on the call is passed to the death timer
435 	 */
436 	_debug("RELEASE CALL %p (%d CONN %p)", call, call->debug_id, conn);
437 
438 	write_lock_bh(&rx->call_lock);
439 	if (!list_empty(&call->accept_link)) {
440 		_debug("unlinking once-pending call %p { e=%lx f=%lx }",
441 		       call, call->events, call->flags);
442 		ASSERT(!test_bit(RXRPC_CALL_HAS_USERID, &call->flags));
443 		list_del_init(&call->accept_link);
444 		sk_acceptq_removed(&rx->sk);
445 	} else if (test_bit(RXRPC_CALL_HAS_USERID, &call->flags)) {
446 		rb_erase(&call->sock_node, &rx->calls);
447 		memset(&call->sock_node, 0xdd, sizeof(call->sock_node));
448 		clear_bit(RXRPC_CALL_HAS_USERID, &call->flags);
449 	}
450 	write_unlock_bh(&rx->call_lock);
451 
452 	/* free up the channel for reuse */
453 	spin_lock(&conn->trans->client_lock);
454 	write_lock_bh(&conn->lock);
455 	write_lock(&call->state_lock);
456 
457 	if (conn->channels[call->channel] == call)
458 		conn->channels[call->channel] = NULL;
459 
460 	if (conn->out_clientflag && conn->bundle) {
461 		conn->avail_calls++;
462 		switch (conn->avail_calls) {
463 		case 1:
464 			list_move_tail(&conn->bundle_link,
465 				       &conn->bundle->avail_conns);
466 		case 2 ... RXRPC_MAXCALLS - 1:
467 			ASSERT(conn->channels[0] == NULL ||
468 			       conn->channels[1] == NULL ||
469 			       conn->channels[2] == NULL ||
470 			       conn->channels[3] == NULL);
471 			break;
472 		case RXRPC_MAXCALLS:
473 			list_move_tail(&conn->bundle_link,
474 				       &conn->bundle->unused_conns);
475 			ASSERT(conn->channels[0] == NULL &&
476 			       conn->channels[1] == NULL &&
477 			       conn->channels[2] == NULL &&
478 			       conn->channels[3] == NULL);
479 			break;
480 		default:
481 			printk(KERN_ERR "RxRPC: conn->avail_calls=%d\n",
482 			       conn->avail_calls);
483 			BUG();
484 		}
485 	}
486 
487 	spin_unlock(&conn->trans->client_lock);
488 
489 	if (call->state < RXRPC_CALL_COMPLETE &&
490 	    call->state != RXRPC_CALL_CLIENT_FINAL_ACK) {
491 		_debug("+++ ABORTING STATE %d +++\n", call->state);
492 		call->state = RXRPC_CALL_LOCALLY_ABORTED;
493 		call->abort_code = RX_CALL_DEAD;
494 		set_bit(RXRPC_CALL_ABORT, &call->events);
495 		rxrpc_queue_call(call);
496 	}
497 	write_unlock(&call->state_lock);
498 	write_unlock_bh(&conn->lock);
499 
500 	/* clean up the Rx queue */
501 	if (!skb_queue_empty(&call->rx_queue) ||
502 	    !skb_queue_empty(&call->rx_oos_queue)) {
503 		struct rxrpc_skb_priv *sp;
504 		struct sk_buff *skb;
505 
506 		_debug("purge Rx queues");
507 
508 		spin_lock_bh(&call->lock);
509 		while ((skb = skb_dequeue(&call->rx_queue)) ||
510 		       (skb = skb_dequeue(&call->rx_oos_queue))) {
511 			sp = rxrpc_skb(skb);
512 			if (sp->call) {
513 				ASSERTCMP(sp->call, ==, call);
514 				rxrpc_put_call(call);
515 				sp->call = NULL;
516 			}
517 			skb->destructor = NULL;
518 			spin_unlock_bh(&call->lock);
519 
520 			_debug("- zap %s %%%u #%u",
521 			       rxrpc_pkts[sp->hdr.type],
522 			       ntohl(sp->hdr.serial),
523 			       ntohl(sp->hdr.seq));
524 			rxrpc_free_skb(skb);
525 			spin_lock_bh(&call->lock);
526 		}
527 		spin_unlock_bh(&call->lock);
528 
529 		ASSERTCMP(call->state, !=, RXRPC_CALL_COMPLETE);
530 	}
531 
532 	del_timer_sync(&call->resend_timer);
533 	del_timer_sync(&call->ack_timer);
534 	del_timer_sync(&call->lifetimer);
535 	call->deadspan.expires = jiffies + rxrpc_dead_call_timeout * HZ;
536 	add_timer(&call->deadspan);
537 
538 	_leave("");
539 }
540 
541 /*
542  * handle a dead call being ready for reaping
543  */
rxrpc_dead_call_expired(unsigned long _call)544 static void rxrpc_dead_call_expired(unsigned long _call)
545 {
546 	struct rxrpc_call *call = (struct rxrpc_call *) _call;
547 
548 	_enter("{%d}", call->debug_id);
549 
550 	write_lock_bh(&call->state_lock);
551 	call->state = RXRPC_CALL_DEAD;
552 	write_unlock_bh(&call->state_lock);
553 	rxrpc_put_call(call);
554 }
555 
556 /*
557  * mark a call as to be released, aborting it if it's still in progress
558  * - called with softirqs disabled
559  */
rxrpc_mark_call_released(struct rxrpc_call * call)560 static void rxrpc_mark_call_released(struct rxrpc_call *call)
561 {
562 	bool sched;
563 
564 	write_lock(&call->state_lock);
565 	if (call->state < RXRPC_CALL_DEAD) {
566 		sched = false;
567 		if (call->state < RXRPC_CALL_COMPLETE) {
568 			_debug("abort call %p", call);
569 			call->state = RXRPC_CALL_LOCALLY_ABORTED;
570 			call->abort_code = RX_CALL_DEAD;
571 			if (!test_and_set_bit(RXRPC_CALL_ABORT, &call->events))
572 				sched = true;
573 		}
574 		if (!test_and_set_bit(RXRPC_CALL_RELEASE, &call->events))
575 			sched = true;
576 		if (sched)
577 			rxrpc_queue_call(call);
578 	}
579 	write_unlock(&call->state_lock);
580 }
581 
582 /*
583  * release all the calls associated with a socket
584  */
rxrpc_release_calls_on_socket(struct rxrpc_sock * rx)585 void rxrpc_release_calls_on_socket(struct rxrpc_sock *rx)
586 {
587 	struct rxrpc_call *call;
588 	struct rb_node *p;
589 
590 	_enter("%p", rx);
591 
592 	read_lock_bh(&rx->call_lock);
593 
594 	/* mark all the calls as no longer wanting incoming packets */
595 	for (p = rb_first(&rx->calls); p; p = rb_next(p)) {
596 		call = rb_entry(p, struct rxrpc_call, sock_node);
597 		rxrpc_mark_call_released(call);
598 	}
599 
600 	/* kill the not-yet-accepted incoming calls */
601 	list_for_each_entry(call, &rx->secureq, accept_link) {
602 		rxrpc_mark_call_released(call);
603 	}
604 
605 	list_for_each_entry(call, &rx->acceptq, accept_link) {
606 		rxrpc_mark_call_released(call);
607 	}
608 
609 	read_unlock_bh(&rx->call_lock);
610 	_leave("");
611 }
612 
613 /*
614  * release a call
615  */
__rxrpc_put_call(struct rxrpc_call * call)616 void __rxrpc_put_call(struct rxrpc_call *call)
617 {
618 	ASSERT(call != NULL);
619 
620 	_enter("%p{u=%d}", call, atomic_read(&call->usage));
621 
622 	ASSERTCMP(atomic_read(&call->usage), >, 0);
623 
624 	if (atomic_dec_and_test(&call->usage)) {
625 		_debug("call %d dead", call->debug_id);
626 		ASSERTCMP(call->state, ==, RXRPC_CALL_DEAD);
627 		rxrpc_queue_work(&call->destroyer);
628 	}
629 	_leave("");
630 }
631 
632 /*
633  * clean up a call
634  */
rxrpc_cleanup_call(struct rxrpc_call * call)635 static void rxrpc_cleanup_call(struct rxrpc_call *call)
636 {
637 	_net("DESTROY CALL %d", call->debug_id);
638 
639 	ASSERT(call->socket);
640 
641 	memset(&call->sock_node, 0xcd, sizeof(call->sock_node));
642 
643 	del_timer_sync(&call->lifetimer);
644 	del_timer_sync(&call->deadspan);
645 	del_timer_sync(&call->ack_timer);
646 	del_timer_sync(&call->resend_timer);
647 
648 	ASSERT(test_bit(RXRPC_CALL_RELEASED, &call->flags));
649 	ASSERTCMP(call->events, ==, 0);
650 	if (work_pending(&call->processor)) {
651 		_debug("defer destroy");
652 		rxrpc_queue_work(&call->destroyer);
653 		return;
654 	}
655 
656 	if (call->conn) {
657 		spin_lock(&call->conn->trans->peer->lock);
658 		list_del(&call->error_link);
659 		spin_unlock(&call->conn->trans->peer->lock);
660 
661 		write_lock_bh(&call->conn->lock);
662 		rb_erase(&call->conn_node, &call->conn->calls);
663 		write_unlock_bh(&call->conn->lock);
664 		rxrpc_put_connection(call->conn);
665 	}
666 
667 	if (call->acks_window) {
668 		_debug("kill Tx window %d",
669 		       CIRC_CNT(call->acks_head, call->acks_tail,
670 				call->acks_winsz));
671 		smp_mb();
672 		while (CIRC_CNT(call->acks_head, call->acks_tail,
673 				call->acks_winsz) > 0) {
674 			struct rxrpc_skb_priv *sp;
675 			unsigned long _skb;
676 
677 			_skb = call->acks_window[call->acks_tail] & ~1;
678 			sp = rxrpc_skb((struct sk_buff *) _skb);
679 			_debug("+++ clear Tx %u", ntohl(sp->hdr.seq));
680 			rxrpc_free_skb((struct sk_buff *) _skb);
681 			call->acks_tail =
682 				(call->acks_tail + 1) & (call->acks_winsz - 1);
683 		}
684 
685 		kfree(call->acks_window);
686 	}
687 
688 	rxrpc_free_skb(call->tx_pending);
689 
690 	rxrpc_purge_queue(&call->rx_queue);
691 	ASSERT(skb_queue_empty(&call->rx_oos_queue));
692 	sock_put(&call->socket->sk);
693 	kmem_cache_free(rxrpc_call_jar, call);
694 }
695 
696 /*
697  * destroy a call
698  */
rxrpc_destroy_call(struct work_struct * work)699 static void rxrpc_destroy_call(struct work_struct *work)
700 {
701 	struct rxrpc_call *call =
702 		container_of(work, struct rxrpc_call, destroyer);
703 
704 	_enter("%p{%d,%d,%p}",
705 	       call, atomic_read(&call->usage), call->channel, call->conn);
706 
707 	ASSERTCMP(call->state, ==, RXRPC_CALL_DEAD);
708 
709 	write_lock_bh(&rxrpc_call_lock);
710 	list_del_init(&call->link);
711 	write_unlock_bh(&rxrpc_call_lock);
712 
713 	rxrpc_cleanup_call(call);
714 	_leave("");
715 }
716 
717 /*
718  * preemptively destroy all the call records from a transport endpoint rather
719  * than waiting for them to time out
720  */
rxrpc_destroy_all_calls(void)721 void __exit rxrpc_destroy_all_calls(void)
722 {
723 	struct rxrpc_call *call;
724 
725 	_enter("");
726 	write_lock_bh(&rxrpc_call_lock);
727 
728 	while (!list_empty(&rxrpc_calls)) {
729 		call = list_entry(rxrpc_calls.next, struct rxrpc_call, link);
730 		_debug("Zapping call %p", call);
731 
732 		list_del_init(&call->link);
733 
734 		switch (atomic_read(&call->usage)) {
735 		case 0:
736 			ASSERTCMP(call->state, ==, RXRPC_CALL_DEAD);
737 			break;
738 		case 1:
739 			if (del_timer_sync(&call->deadspan) != 0 &&
740 			    call->state != RXRPC_CALL_DEAD)
741 				rxrpc_dead_call_expired((unsigned long) call);
742 			if (call->state != RXRPC_CALL_DEAD)
743 				break;
744 		default:
745 			printk(KERN_ERR "RXRPC:"
746 			       " Call %p still in use (%d,%d,%s,%lx,%lx)!\n",
747 			       call, atomic_read(&call->usage),
748 			       atomic_read(&call->ackr_not_idle),
749 			       rxrpc_call_states[call->state],
750 			       call->flags, call->events);
751 			if (!skb_queue_empty(&call->rx_queue))
752 				printk(KERN_ERR"RXRPC: Rx queue occupied\n");
753 			if (!skb_queue_empty(&call->rx_oos_queue))
754 				printk(KERN_ERR"RXRPC: OOS queue occupied\n");
755 			break;
756 		}
757 
758 		write_unlock_bh(&rxrpc_call_lock);
759 		cond_resched();
760 		write_lock_bh(&rxrpc_call_lock);
761 	}
762 
763 	write_unlock_bh(&rxrpc_call_lock);
764 	_leave("");
765 }
766 
767 /*
768  * handle call lifetime being exceeded
769  */
rxrpc_call_life_expired(unsigned long _call)770 static void rxrpc_call_life_expired(unsigned long _call)
771 {
772 	struct rxrpc_call *call = (struct rxrpc_call *) _call;
773 
774 	if (call->state >= RXRPC_CALL_COMPLETE)
775 		return;
776 
777 	_enter("{%d}", call->debug_id);
778 	read_lock_bh(&call->state_lock);
779 	if (call->state < RXRPC_CALL_COMPLETE) {
780 		set_bit(RXRPC_CALL_LIFE_TIMER, &call->events);
781 		rxrpc_queue_call(call);
782 	}
783 	read_unlock_bh(&call->state_lock);
784 }
785 
786 /*
787  * handle resend timer expiry
788  */
rxrpc_resend_time_expired(unsigned long _call)789 static void rxrpc_resend_time_expired(unsigned long _call)
790 {
791 	struct rxrpc_call *call = (struct rxrpc_call *) _call;
792 
793 	_enter("{%d}", call->debug_id);
794 
795 	if (call->state >= RXRPC_CALL_COMPLETE)
796 		return;
797 
798 	read_lock_bh(&call->state_lock);
799 	clear_bit(RXRPC_CALL_RUN_RTIMER, &call->flags);
800 	if (call->state < RXRPC_CALL_COMPLETE &&
801 	    !test_and_set_bit(RXRPC_CALL_RESEND_TIMER, &call->events))
802 		rxrpc_queue_call(call);
803 	read_unlock_bh(&call->state_lock);
804 }
805 
806 /*
807  * handle ACK timer expiry
808  */
rxrpc_ack_time_expired(unsigned long _call)809 static void rxrpc_ack_time_expired(unsigned long _call)
810 {
811 	struct rxrpc_call *call = (struct rxrpc_call *) _call;
812 
813 	_enter("{%d}", call->debug_id);
814 
815 	if (call->state >= RXRPC_CALL_COMPLETE)
816 		return;
817 
818 	read_lock_bh(&call->state_lock);
819 	if (call->state < RXRPC_CALL_COMPLETE &&
820 	    !test_and_set_bit(RXRPC_CALL_ACK, &call->events))
821 		rxrpc_queue_call(call);
822 	read_unlock_bh(&call->state_lock);
823 }
824