Merge branch 'x86-paravirt-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[linux-2.6-block.git] / net / rxrpc / conn_object.c
1 /* RxRPC virtual connection handler, common bits.
2  *
3  * Copyright (C) 2007, 2016 Red Hat, Inc. All Rights Reserved.
4  * Written by David Howells (dhowells@redhat.com)
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public 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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13
14 #include <linux/module.h>
15 #include <linux/slab.h>
16 #include <linux/net.h>
17 #include <linux/skbuff.h>
18 #include "ar-internal.h"
19
20 /*
21  * Time till a connection expires after last use (in seconds).
22  */
23 unsigned int __read_mostly rxrpc_connection_expiry = 10 * 60;
24 unsigned int __read_mostly rxrpc_closed_conn_expiry = 10;
25
26 static void rxrpc_destroy_connection(struct rcu_head *);
27
28 static void rxrpc_connection_timer(struct timer_list *timer)
29 {
30         struct rxrpc_connection *conn =
31                 container_of(timer, struct rxrpc_connection, timer);
32
33         rxrpc_queue_conn(conn);
34 }
35
36 /*
37  * allocate a new connection
38  */
39 struct rxrpc_connection *rxrpc_alloc_connection(gfp_t gfp)
40 {
41         struct rxrpc_connection *conn;
42
43         _enter("");
44
45         conn = kzalloc(sizeof(struct rxrpc_connection), gfp);
46         if (conn) {
47                 INIT_LIST_HEAD(&conn->cache_link);
48                 spin_lock_init(&conn->channel_lock);
49                 INIT_LIST_HEAD(&conn->waiting_calls);
50                 timer_setup(&conn->timer, &rxrpc_connection_timer, 0);
51                 INIT_WORK(&conn->processor, &rxrpc_process_connection);
52                 INIT_LIST_HEAD(&conn->proc_link);
53                 INIT_LIST_HEAD(&conn->link);
54                 skb_queue_head_init(&conn->rx_queue);
55                 conn->security = &rxrpc_no_security;
56                 spin_lock_init(&conn->state_lock);
57                 conn->debug_id = atomic_inc_return(&rxrpc_debug_id);
58                 conn->size_align = 4;
59                 conn->idle_timestamp = jiffies;
60         }
61
62         _leave(" = %p{%d}", conn, conn ? conn->debug_id : 0);
63         return conn;
64 }
65
66 /*
67  * Look up a connection in the cache by protocol parameters.
68  *
69  * If successful, a pointer to the connection is returned, but no ref is taken.
70  * NULL is returned if there is no match.
71  *
72  * When searching for a service call, if we find a peer but no connection, we
73  * return that through *_peer in case we need to create a new service call.
74  *
75  * The caller must be holding the RCU read lock.
76  */
77 struct rxrpc_connection *rxrpc_find_connection_rcu(struct rxrpc_local *local,
78                                                    struct sk_buff *skb,
79                                                    struct rxrpc_peer **_peer)
80 {
81         struct rxrpc_connection *conn;
82         struct rxrpc_conn_proto k;
83         struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
84         struct sockaddr_rxrpc srx;
85         struct rxrpc_peer *peer;
86
87         _enter(",%x", sp->hdr.cid & RXRPC_CIDMASK);
88
89         if (rxrpc_extract_addr_from_skb(local, &srx, skb) < 0)
90                 goto not_found;
91
92         /* We may have to handle mixing IPv4 and IPv6 */
93         if (srx.transport.family != local->srx.transport.family) {
94                 pr_warn_ratelimited("AF_RXRPC: Protocol mismatch %u not %u\n",
95                                     srx.transport.family,
96                                     local->srx.transport.family);
97                 goto not_found;
98         }
99
100         k.epoch = sp->hdr.epoch;
101         k.cid   = sp->hdr.cid & RXRPC_CIDMASK;
102
103         if (rxrpc_to_server(sp)) {
104                 /* We need to look up service connections by the full protocol
105                  * parameter set.  We look up the peer first as an intermediate
106                  * step and then the connection from the peer's tree.
107                  */
108                 peer = rxrpc_lookup_peer_rcu(local, &srx);
109                 if (!peer)
110                         goto not_found;
111                 *_peer = peer;
112                 conn = rxrpc_find_service_conn_rcu(peer, skb);
113                 if (!conn || atomic_read(&conn->usage) == 0)
114                         goto not_found;
115                 _leave(" = %p", conn);
116                 return conn;
117         } else {
118                 /* Look up client connections by connection ID alone as their
119                  * IDs are unique for this machine.
120                  */
121                 conn = idr_find(&rxrpc_client_conn_ids,
122                                 sp->hdr.cid >> RXRPC_CIDSHIFT);
123                 if (!conn || atomic_read(&conn->usage) == 0) {
124                         _debug("no conn");
125                         goto not_found;
126                 }
127
128                 if (conn->proto.epoch != k.epoch ||
129                     conn->params.local != local)
130                         goto not_found;
131
132                 peer = conn->params.peer;
133                 switch (srx.transport.family) {
134                 case AF_INET:
135                         if (peer->srx.transport.sin.sin_port !=
136                             srx.transport.sin.sin_port ||
137                             peer->srx.transport.sin.sin_addr.s_addr !=
138                             srx.transport.sin.sin_addr.s_addr)
139                                 goto not_found;
140                         break;
141 #ifdef CONFIG_AF_RXRPC_IPV6
142                 case AF_INET6:
143                         if (peer->srx.transport.sin6.sin6_port !=
144                             srx.transport.sin6.sin6_port ||
145                             memcmp(&peer->srx.transport.sin6.sin6_addr,
146                                    &srx.transport.sin6.sin6_addr,
147                                    sizeof(struct in6_addr)) != 0)
148                                 goto not_found;
149                         break;
150 #endif
151                 default:
152                         BUG();
153                 }
154
155                 _leave(" = %p", conn);
156                 return conn;
157         }
158
159 not_found:
160         _leave(" = NULL");
161         return NULL;
162 }
163
164 /*
165  * Disconnect a call and clear any channel it occupies when that call
166  * terminates.  The caller must hold the channel_lock and must release the
167  * call's ref on the connection.
168  */
169 void __rxrpc_disconnect_call(struct rxrpc_connection *conn,
170                              struct rxrpc_call *call)
171 {
172         struct rxrpc_channel *chan =
173                 &conn->channels[call->cid & RXRPC_CHANNELMASK];
174
175         _enter("%d,%x", conn->debug_id, call->cid);
176
177         if (rcu_access_pointer(chan->call) == call) {
178                 /* Save the result of the call so that we can repeat it if necessary
179                  * through the channel, whilst disposing of the actual call record.
180                  */
181                 trace_rxrpc_disconnect_call(call);
182                 switch (call->completion) {
183                 case RXRPC_CALL_SUCCEEDED:
184                         chan->last_seq = call->rx_hard_ack;
185                         chan->last_type = RXRPC_PACKET_TYPE_ACK;
186                         break;
187                 case RXRPC_CALL_LOCALLY_ABORTED:
188                         chan->last_abort = call->abort_code;
189                         chan->last_type = RXRPC_PACKET_TYPE_ABORT;
190                         break;
191                 default:
192                         chan->last_abort = RX_USER_ABORT;
193                         chan->last_type = RXRPC_PACKET_TYPE_ABORT;
194                         break;
195                 }
196
197                 /* Sync with rxrpc_conn_retransmit(). */
198                 smp_wmb();
199                 chan->last_call = chan->call_id;
200                 chan->call_id = chan->call_counter;
201
202                 rcu_assign_pointer(chan->call, NULL);
203         }
204
205         _leave("");
206 }
207
208 /*
209  * Disconnect a call and clear any channel it occupies when that call
210  * terminates.
211  */
212 void rxrpc_disconnect_call(struct rxrpc_call *call)
213 {
214         struct rxrpc_connection *conn = call->conn;
215
216         call->peer->cong_cwnd = call->cong_cwnd;
217
218         spin_lock_bh(&conn->params.peer->lock);
219         hlist_del_rcu(&call->error_link);
220         spin_unlock_bh(&conn->params.peer->lock);
221
222         if (rxrpc_is_client_call(call))
223                 return rxrpc_disconnect_client_call(call);
224
225         spin_lock(&conn->channel_lock);
226         __rxrpc_disconnect_call(conn, call);
227         spin_unlock(&conn->channel_lock);
228
229         call->conn = NULL;
230         conn->idle_timestamp = jiffies;
231         rxrpc_put_connection(conn);
232 }
233
234 /*
235  * Kill off a connection.
236  */
237 void rxrpc_kill_connection(struct rxrpc_connection *conn)
238 {
239         struct rxrpc_net *rxnet = conn->params.local->rxnet;
240
241         ASSERT(!rcu_access_pointer(conn->channels[0].call) &&
242                !rcu_access_pointer(conn->channels[1].call) &&
243                !rcu_access_pointer(conn->channels[2].call) &&
244                !rcu_access_pointer(conn->channels[3].call));
245         ASSERT(list_empty(&conn->cache_link));
246
247         write_lock(&rxnet->conn_lock);
248         list_del_init(&conn->proc_link);
249         write_unlock(&rxnet->conn_lock);
250
251         /* Drain the Rx queue.  Note that even though we've unpublished, an
252          * incoming packet could still be being added to our Rx queue, so we
253          * will need to drain it again in the RCU cleanup handler.
254          */
255         rxrpc_purge_queue(&conn->rx_queue);
256
257         /* Leave final destruction to RCU.  The connection processor work item
258          * must carry a ref on the connection to prevent us getting here whilst
259          * it is queued or running.
260          */
261         call_rcu(&conn->rcu, rxrpc_destroy_connection);
262 }
263
264 /*
265  * Queue a connection's work processor, getting a ref to pass to the work
266  * queue.
267  */
268 bool rxrpc_queue_conn(struct rxrpc_connection *conn)
269 {
270         const void *here = __builtin_return_address(0);
271         int n = atomic_fetch_add_unless(&conn->usage, 1, 0);
272         if (n == 0)
273                 return false;
274         if (rxrpc_queue_work(&conn->processor))
275                 trace_rxrpc_conn(conn, rxrpc_conn_queued, n + 1, here);
276         else
277                 rxrpc_put_connection(conn);
278         return true;
279 }
280
281 /*
282  * Note the re-emergence of a connection.
283  */
284 void rxrpc_see_connection(struct rxrpc_connection *conn)
285 {
286         const void *here = __builtin_return_address(0);
287         if (conn) {
288                 int n = atomic_read(&conn->usage);
289
290                 trace_rxrpc_conn(conn, rxrpc_conn_seen, n, here);
291         }
292 }
293
294 /*
295  * Get a ref on a connection.
296  */
297 void rxrpc_get_connection(struct rxrpc_connection *conn)
298 {
299         const void *here = __builtin_return_address(0);
300         int n = atomic_inc_return(&conn->usage);
301
302         trace_rxrpc_conn(conn, rxrpc_conn_got, n, here);
303 }
304
305 /*
306  * Try to get a ref on a connection.
307  */
308 struct rxrpc_connection *
309 rxrpc_get_connection_maybe(struct rxrpc_connection *conn)
310 {
311         const void *here = __builtin_return_address(0);
312
313         if (conn) {
314                 int n = atomic_fetch_add_unless(&conn->usage, 1, 0);
315                 if (n > 0)
316                         trace_rxrpc_conn(conn, rxrpc_conn_got, n + 1, here);
317                 else
318                         conn = NULL;
319         }
320         return conn;
321 }
322
323 /*
324  * Set the service connection reap timer.
325  */
326 static void rxrpc_set_service_reap_timer(struct rxrpc_net *rxnet,
327                                          unsigned long reap_at)
328 {
329         if (rxnet->live)
330                 timer_reduce(&rxnet->service_conn_reap_timer, reap_at);
331 }
332
333 /*
334  * Release a service connection
335  */
336 void rxrpc_put_service_conn(struct rxrpc_connection *conn)
337 {
338         const void *here = __builtin_return_address(0);
339         int n;
340
341         n = atomic_dec_return(&conn->usage);
342         trace_rxrpc_conn(conn, rxrpc_conn_put_service, n, here);
343         ASSERTCMP(n, >=, 0);
344         if (n == 1)
345                 rxrpc_set_service_reap_timer(conn->params.local->rxnet,
346                                              jiffies + rxrpc_connection_expiry);
347 }
348
349 /*
350  * destroy a virtual connection
351  */
352 static void rxrpc_destroy_connection(struct rcu_head *rcu)
353 {
354         struct rxrpc_connection *conn =
355                 container_of(rcu, struct rxrpc_connection, rcu);
356
357         _enter("{%d,u=%d}", conn->debug_id, atomic_read(&conn->usage));
358
359         ASSERTCMP(atomic_read(&conn->usage), ==, 0);
360
361         _net("DESTROY CONN %d", conn->debug_id);
362
363         del_timer_sync(&conn->timer);
364         rxrpc_purge_queue(&conn->rx_queue);
365
366         conn->security->clear(conn);
367         key_put(conn->params.key);
368         key_put(conn->server_key);
369         rxrpc_put_peer(conn->params.peer);
370
371         if (atomic_dec_and_test(&conn->params.local->rxnet->nr_conns))
372                 wake_up_var(&conn->params.local->rxnet->nr_conns);
373         rxrpc_put_local(conn->params.local);
374
375         kfree(conn);
376         _leave("");
377 }
378
379 /*
380  * reap dead service connections
381  */
382 void rxrpc_service_connection_reaper(struct work_struct *work)
383 {
384         struct rxrpc_connection *conn, *_p;
385         struct rxrpc_net *rxnet =
386                 container_of(work, struct rxrpc_net, service_conn_reaper);
387         unsigned long expire_at, earliest, idle_timestamp, now;
388
389         LIST_HEAD(graveyard);
390
391         _enter("");
392
393         now = jiffies;
394         earliest = now + MAX_JIFFY_OFFSET;
395
396         write_lock(&rxnet->conn_lock);
397         list_for_each_entry_safe(conn, _p, &rxnet->service_conns, link) {
398                 ASSERTCMP(atomic_read(&conn->usage), >, 0);
399                 if (likely(atomic_read(&conn->usage) > 1))
400                         continue;
401                 if (conn->state == RXRPC_CONN_SERVICE_PREALLOC)
402                         continue;
403
404                 if (rxnet->live) {
405                         idle_timestamp = READ_ONCE(conn->idle_timestamp);
406                         expire_at = idle_timestamp + rxrpc_connection_expiry * HZ;
407                         if (conn->params.local->service_closed)
408                                 expire_at = idle_timestamp + rxrpc_closed_conn_expiry * HZ;
409
410                         _debug("reap CONN %d { u=%d,t=%ld }",
411                                conn->debug_id, atomic_read(&conn->usage),
412                                (long)expire_at - (long)now);
413
414                         if (time_before(now, expire_at)) {
415                                 if (time_before(expire_at, earliest))
416                                         earliest = expire_at;
417                                 continue;
418                         }
419                 }
420
421                 /* The usage count sits at 1 whilst the object is unused on the
422                  * list; we reduce that to 0 to make the object unavailable.
423                  */
424                 if (atomic_cmpxchg(&conn->usage, 1, 0) != 1)
425                         continue;
426                 trace_rxrpc_conn(conn, rxrpc_conn_reap_service, 0, NULL);
427
428                 if (rxrpc_conn_is_client(conn))
429                         BUG();
430                 else
431                         rxrpc_unpublish_service_conn(conn);
432
433                 list_move_tail(&conn->link, &graveyard);
434         }
435         write_unlock(&rxnet->conn_lock);
436
437         if (earliest != now + MAX_JIFFY_OFFSET) {
438                 _debug("reschedule reaper %ld", (long)earliest - (long)now);
439                 ASSERT(time_after(earliest, now));
440                 rxrpc_set_service_reap_timer(rxnet, earliest);
441         }
442
443         while (!list_empty(&graveyard)) {
444                 conn = list_entry(graveyard.next, struct rxrpc_connection,
445                                   link);
446                 list_del_init(&conn->link);
447
448                 ASSERTCMP(atomic_read(&conn->usage), ==, 0);
449                 rxrpc_kill_connection(conn);
450         }
451
452         _leave("");
453 }
454
455 /*
456  * preemptively destroy all the service connection records rather than
457  * waiting for them to time out
458  */
459 void rxrpc_destroy_all_connections(struct rxrpc_net *rxnet)
460 {
461         struct rxrpc_connection *conn, *_p;
462         bool leak = false;
463
464         _enter("");
465
466         atomic_dec(&rxnet->nr_conns);
467         rxrpc_destroy_all_client_connections(rxnet);
468
469         del_timer_sync(&rxnet->service_conn_reap_timer);
470         rxrpc_queue_work(&rxnet->service_conn_reaper);
471         flush_workqueue(rxrpc_workqueue);
472
473         write_lock(&rxnet->conn_lock);
474         list_for_each_entry_safe(conn, _p, &rxnet->service_conns, link) {
475                 pr_err("AF_RXRPC: Leaked conn %p {%d}\n",
476                        conn, atomic_read(&conn->usage));
477                 leak = true;
478         }
479         write_unlock(&rxnet->conn_lock);
480         BUG_ON(leak);
481
482         ASSERT(list_empty(&rxnet->conn_proc_list));
483
484         /* We need to wait for the connections to be destroyed by RCU as they
485          * pin things that we still need to get rid of.
486          */
487         wait_var_event(&rxnet->nr_conns, !atomic_read(&rxnet->nr_conns));
488         _leave("");
489 }