1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* RxRPC individual remote procedure call handling
4 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 #include <linux/slab.h>
11 #include <linux/module.h>
12 #include <linux/circ_buf.h>
13 #include <linux/spinlock_types.h>
15 #include <net/af_rxrpc.h>
16 #include "ar-internal.h"
18 const char *const rxrpc_call_states[NR__RXRPC_CALL_STATES] = {
19 [RXRPC_CALL_UNINITIALISED] = "Uninit ",
20 [RXRPC_CALL_CLIENT_AWAIT_CONN] = "ClWtConn",
21 [RXRPC_CALL_CLIENT_SEND_REQUEST] = "ClSndReq",
22 [RXRPC_CALL_CLIENT_AWAIT_REPLY] = "ClAwtRpl",
23 [RXRPC_CALL_CLIENT_RECV_REPLY] = "ClRcvRpl",
24 [RXRPC_CALL_SERVER_PREALLOC] = "SvPrealc",
25 [RXRPC_CALL_SERVER_SECURING] = "SvSecure",
26 [RXRPC_CALL_SERVER_RECV_REQUEST] = "SvRcvReq",
27 [RXRPC_CALL_SERVER_ACK_REQUEST] = "SvAckReq",
28 [RXRPC_CALL_SERVER_SEND_REPLY] = "SvSndRpl",
29 [RXRPC_CALL_SERVER_AWAIT_ACK] = "SvAwtACK",
30 [RXRPC_CALL_COMPLETE] = "Complete",
33 const char *const rxrpc_call_completions[NR__RXRPC_CALL_COMPLETIONS] = {
34 [RXRPC_CALL_SUCCEEDED] = "Complete",
35 [RXRPC_CALL_REMOTELY_ABORTED] = "RmtAbort",
36 [RXRPC_CALL_LOCALLY_ABORTED] = "LocAbort",
37 [RXRPC_CALL_LOCAL_ERROR] = "LocError",
38 [RXRPC_CALL_NETWORK_ERROR] = "NetError",
41 struct kmem_cache *rxrpc_call_jar;
43 static struct semaphore rxrpc_call_limiter =
44 __SEMAPHORE_INITIALIZER(rxrpc_call_limiter, 1000);
45 static struct semaphore rxrpc_kernel_call_limiter =
46 __SEMAPHORE_INITIALIZER(rxrpc_kernel_call_limiter, 1000);
48 void rxrpc_poke_call(struct rxrpc_call *call, enum rxrpc_call_poke_trace what)
50 struct rxrpc_local *local = call->local;
53 if (call->state < RXRPC_CALL_COMPLETE) {
54 spin_lock_bh(&local->lock);
55 busy = !list_empty(&call->attend_link);
56 trace_rxrpc_poke_call(call, busy, what);
58 rxrpc_get_call(call, rxrpc_call_get_poke);
59 list_add_tail(&call->attend_link, &local->call_attend_q);
61 spin_unlock_bh(&local->lock);
62 rxrpc_wake_up_io_thread(local);
66 static void rxrpc_call_timer_expired(struct timer_list *t)
68 struct rxrpc_call *call = from_timer(call, t, timer);
70 _enter("%d", call->debug_id);
72 if (call->state < RXRPC_CALL_COMPLETE) {
73 trace_rxrpc_timer_expired(call, jiffies);
74 rxrpc_queue_call(call, rxrpc_call_queue_timer);
78 void rxrpc_reduce_call_timer(struct rxrpc_call *call,
79 unsigned long expire_at,
81 enum rxrpc_timer_trace why)
83 trace_rxrpc_timer(call, why, now);
84 timer_reduce(&call->timer, expire_at);
87 static struct lock_class_key rxrpc_call_user_mutex_lock_class_key;
89 static void rxrpc_destroy_call(struct work_struct *);
92 * find an extant server call
93 * - called in process context with IRQs enabled
95 struct rxrpc_call *rxrpc_find_call_by_user_ID(struct rxrpc_sock *rx,
96 unsigned long user_call_ID)
98 struct rxrpc_call *call;
101 _enter("%p,%lx", rx, user_call_ID);
103 read_lock(&rx->call_lock);
105 p = rx->calls.rb_node;
107 call = rb_entry(p, struct rxrpc_call, sock_node);
109 if (user_call_ID < call->user_call_ID)
111 else if (user_call_ID > call->user_call_ID)
114 goto found_extant_call;
117 read_unlock(&rx->call_lock);
122 rxrpc_get_call(call, rxrpc_call_get_sendmsg);
123 read_unlock(&rx->call_lock);
124 _leave(" = %p [%d]", call, refcount_read(&call->ref));
129 * allocate a new call
131 struct rxrpc_call *rxrpc_alloc_call(struct rxrpc_sock *rx, gfp_t gfp,
132 unsigned int debug_id)
134 struct rxrpc_call *call;
135 struct rxrpc_net *rxnet = rxrpc_net(sock_net(&rx->sk));
137 call = kmem_cache_zalloc(rxrpc_call_jar, gfp);
141 mutex_init(&call->user_mutex);
143 /* Prevent lockdep reporting a deadlock false positive between the afs
144 * filesystem and sys_sendmsg() via the mmap sem.
146 if (rx->sk.sk_kern_sock)
147 lockdep_set_class(&call->user_mutex,
148 &rxrpc_call_user_mutex_lock_class_key);
150 timer_setup(&call->timer, rxrpc_call_timer_expired, 0);
151 INIT_WORK(&call->processor, rxrpc_process_call);
152 INIT_WORK(&call->destroyer, rxrpc_destroy_call);
153 INIT_LIST_HEAD(&call->link);
154 INIT_LIST_HEAD(&call->chan_wait_link);
155 INIT_LIST_HEAD(&call->accept_link);
156 INIT_LIST_HEAD(&call->recvmsg_link);
157 INIT_LIST_HEAD(&call->sock_link);
158 INIT_LIST_HEAD(&call->attend_link);
159 INIT_LIST_HEAD(&call->tx_buffer);
160 skb_queue_head_init(&call->recvmsg_queue);
161 skb_queue_head_init(&call->rx_oos_queue);
162 init_waitqueue_head(&call->waitq);
163 spin_lock_init(&call->notify_lock);
164 spin_lock_init(&call->tx_lock);
165 spin_lock_init(&call->acks_ack_lock);
166 rwlock_init(&call->state_lock);
167 refcount_set(&call->ref, 1);
168 call->debug_id = debug_id;
169 call->tx_total_len = -1;
170 call->next_rx_timo = 20 * HZ;
171 call->next_req_timo = 1 * HZ;
172 atomic64_set(&call->ackr_window, 0x100000001ULL);
174 memset(&call->sock_node, 0xed, sizeof(call->sock_node));
176 call->rx_winsize = rxrpc_rx_window_size;
177 call->tx_winsize = 16;
179 if (RXRPC_TX_SMSS > 2190)
181 else if (RXRPC_TX_SMSS > 1095)
185 call->cong_ssthresh = RXRPC_TX_MAX_WINDOW;
188 call->rtt_avail = RXRPC_CALL_RTT_AVAIL_MASK;
189 atomic_inc(&rxnet->nr_calls);
194 * Allocate a new client call.
196 static struct rxrpc_call *rxrpc_alloc_client_call(struct rxrpc_sock *rx,
197 struct sockaddr_rxrpc *srx,
198 struct rxrpc_conn_parameters *cp,
199 struct rxrpc_call_params *p,
201 unsigned int debug_id)
203 struct rxrpc_call *call;
209 call = rxrpc_alloc_call(rx, gfp, debug_id);
211 return ERR_PTR(-ENOMEM);
212 now = ktime_get_real();
213 call->acks_latest_ts = now;
214 call->cong_tstamp = now;
215 call->state = RXRPC_CALL_CLIENT_AWAIT_CONN;
216 call->dest_srx = *srx;
217 call->interruptibility = p->interruptibility;
218 call->tx_total_len = p->tx_total_len;
219 call->key = key_get(cp->key);
220 call->local = rxrpc_get_local(cp->local, rxrpc_local_get_call);
222 __set_bit(RXRPC_CALL_KERNEL, &call->flags);
224 __set_bit(RXRPC_CALL_UPGRADE, &call->flags);
226 __set_bit(RXRPC_CALL_EXCLUSIVE, &call->flags);
228 ret = rxrpc_init_client_call_security(call);
230 __rxrpc_set_call_completion(call, RXRPC_CALL_LOCAL_ERROR, 0, ret);
231 rxrpc_put_call(call, rxrpc_call_put_discard_error);
235 trace_rxrpc_call(call->debug_id, refcount_read(&call->ref),
236 p->user_call_ID, rxrpc_call_new_client);
238 _leave(" = %p", call);
243 * Initiate the call ack/resend/expiry timer.
245 static void rxrpc_start_call_timer(struct rxrpc_call *call)
247 unsigned long now = jiffies;
248 unsigned long j = now + MAX_JIFFY_OFFSET;
250 call->delay_ack_at = j;
251 call->ack_lost_at = j;
254 call->expect_rx_by = j;
255 call->expect_req_by = j;
256 call->expect_term_by = j;
257 call->timer.expires = now;
261 * Wait for a call slot to become available.
263 static struct semaphore *rxrpc_get_call_slot(struct rxrpc_call_params *p, gfp_t gfp)
265 struct semaphore *limiter = &rxrpc_call_limiter;
268 limiter = &rxrpc_kernel_call_limiter;
269 if (p->interruptibility == RXRPC_UNINTERRUPTIBLE) {
273 return down_interruptible(limiter) < 0 ? NULL : limiter;
277 * Release a call slot.
279 static void rxrpc_put_call_slot(struct rxrpc_call *call)
281 struct semaphore *limiter = &rxrpc_call_limiter;
283 if (test_bit(RXRPC_CALL_KERNEL, &call->flags))
284 limiter = &rxrpc_kernel_call_limiter;
289 * Set up a call for the given parameters.
290 * - Called with the socket lock held, which it must release.
291 * - If it returns a call, the call's lock will need releasing by the caller.
293 struct rxrpc_call *rxrpc_new_client_call(struct rxrpc_sock *rx,
294 struct rxrpc_conn_parameters *cp,
295 struct sockaddr_rxrpc *srx,
296 struct rxrpc_call_params *p,
298 unsigned int debug_id)
299 __releases(&rx->sk.sk_lock.slock)
300 __acquires(&call->user_mutex)
302 struct rxrpc_call *call, *xcall;
303 struct rxrpc_net *rxnet;
304 struct semaphore *limiter;
305 struct rb_node *parent, **pp;
308 _enter("%p,%lx", rx, p->user_call_ID);
310 limiter = rxrpc_get_call_slot(p, gfp);
312 release_sock(&rx->sk);
313 return ERR_PTR(-ERESTARTSYS);
316 call = rxrpc_alloc_client_call(rx, srx, cp, p, gfp, debug_id);
318 release_sock(&rx->sk);
320 _leave(" = %ld", PTR_ERR(call));
324 /* We need to protect a partially set up call against the user as we
325 * will be acting outside the socket lock.
327 mutex_lock(&call->user_mutex);
329 /* Publish the call, even though it is incompletely set up as yet */
330 write_lock(&rx->call_lock);
332 pp = &rx->calls.rb_node;
336 xcall = rb_entry(parent, struct rxrpc_call, sock_node);
338 if (p->user_call_ID < xcall->user_call_ID)
339 pp = &(*pp)->rb_left;
340 else if (p->user_call_ID > xcall->user_call_ID)
341 pp = &(*pp)->rb_right;
343 goto error_dup_user_ID;
346 rcu_assign_pointer(call->socket, rx);
347 call->user_call_ID = p->user_call_ID;
348 __set_bit(RXRPC_CALL_HAS_USERID, &call->flags);
349 rxrpc_get_call(call, rxrpc_call_get_userid);
350 rb_link_node(&call->sock_node, parent, pp);
351 rb_insert_color(&call->sock_node, &rx->calls);
352 list_add(&call->sock_link, &rx->sock_calls);
354 write_unlock(&rx->call_lock);
357 spin_lock_bh(&rxnet->call_lock);
358 list_add_tail_rcu(&call->link, &rxnet->calls);
359 spin_unlock_bh(&rxnet->call_lock);
361 /* From this point on, the call is protected by its own lock. */
362 release_sock(&rx->sk);
364 /* Set up or get a connection record and set the protocol parameters,
365 * including channel number and call ID.
367 ret = rxrpc_connect_call(rx, call, cp, srx, gfp);
369 goto error_attached_to_socket;
371 rxrpc_see_call(call, rxrpc_call_see_connected);
373 rxrpc_start_call_timer(call);
375 _leave(" = %p [new]", call);
378 /* We unexpectedly found the user ID in the list after taking
379 * the call_lock. This shouldn't happen unless the user races
380 * with itself and tries to add the same user ID twice at the
381 * same time in different threads.
384 write_unlock(&rx->call_lock);
385 release_sock(&rx->sk);
386 __rxrpc_set_call_completion(call, RXRPC_CALL_LOCAL_ERROR,
387 RX_CALL_DEAD, -EEXIST);
388 trace_rxrpc_call(call->debug_id, refcount_read(&call->ref), 0,
389 rxrpc_call_see_userid_exists);
390 rxrpc_release_call(rx, call);
391 mutex_unlock(&call->user_mutex);
392 rxrpc_put_call(call, rxrpc_call_put_userid_exists);
393 _leave(" = -EEXIST");
394 return ERR_PTR(-EEXIST);
396 /* We got an error, but the call is attached to the socket and is in
397 * need of release. However, we might now race with recvmsg() when
398 * completing the call queues it. Return 0 from sys_sendmsg() and
399 * leave the error to recvmsg() to deal with.
401 error_attached_to_socket:
402 trace_rxrpc_call(call->debug_id, refcount_read(&call->ref), ret,
403 rxrpc_call_see_connect_failed);
404 set_bit(RXRPC_CALL_DISCONNECTED, &call->flags);
405 __rxrpc_set_call_completion(call, RXRPC_CALL_LOCAL_ERROR,
407 _leave(" = c=%08x [err]", call->debug_id);
412 * Set up an incoming call. call->conn points to the connection.
413 * This is called in BH context and isn't allowed to fail.
415 void rxrpc_incoming_call(struct rxrpc_sock *rx,
416 struct rxrpc_call *call,
419 struct rxrpc_connection *conn = call->conn;
420 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
423 _enter(",%d", call->conn->debug_id);
425 rcu_assign_pointer(call->socket, rx);
426 call->call_id = sp->hdr.callNumber;
427 call->dest_srx.srx_service = sp->hdr.serviceId;
428 call->cid = sp->hdr.cid;
429 call->state = RXRPC_CALL_SERVER_SECURING;
430 call->cong_tstamp = skb->tstamp;
432 /* Set the channel for this call. We don't get channel_lock as we're
433 * only defending against the data_ready handler (which we're called
434 * from) and the RESPONSE packet parser (which is only really
435 * interested in call_counter and can cope with a disagreement with the
438 chan = sp->hdr.cid & RXRPC_CHANNELMASK;
439 conn->channels[chan].call_counter = call->call_id;
440 conn->channels[chan].call_id = call->call_id;
441 rcu_assign_pointer(conn->channels[chan].call, call);
443 spin_lock(&conn->peer->lock);
444 hlist_add_head_rcu(&call->error_link, &conn->peer->error_targets);
445 spin_unlock(&conn->peer->lock);
447 rxrpc_start_call_timer(call);
452 * Queue a call's work processor.
454 void rxrpc_queue_call(struct rxrpc_call *call, enum rxrpc_call_trace why)
456 if (rxrpc_queue_work(&call->processor))
457 trace_rxrpc_call(call->debug_id, refcount_read(&call->ref), 0, why);
461 * Note the re-emergence of a call.
463 void rxrpc_see_call(struct rxrpc_call *call, enum rxrpc_call_trace why)
466 int r = refcount_read(&call->ref);
468 trace_rxrpc_call(call->debug_id, r, 0, why);
472 bool rxrpc_try_get_call(struct rxrpc_call *call, enum rxrpc_call_trace why)
476 if (!__refcount_inc_not_zero(&call->ref, &r))
478 trace_rxrpc_call(call->debug_id, r + 1, 0, why);
483 * Note the addition of a ref on a call.
485 void rxrpc_get_call(struct rxrpc_call *call, enum rxrpc_call_trace why)
489 __refcount_inc(&call->ref, &r);
490 trace_rxrpc_call(call->debug_id, r + 1, 0, why);
494 * Clean up the Rx skb ring.
496 static void rxrpc_cleanup_ring(struct rxrpc_call *call)
498 skb_queue_purge(&call->recvmsg_queue);
499 skb_queue_purge(&call->rx_oos_queue);
503 * Detach a call from its owning socket.
505 void rxrpc_release_call(struct rxrpc_sock *rx, struct rxrpc_call *call)
507 struct rxrpc_connection *conn = call->conn;
510 _enter("{%d,%d}", call->debug_id, refcount_read(&call->ref));
512 trace_rxrpc_call(call->debug_id, refcount_read(&call->ref),
513 call->flags, rxrpc_call_see_release);
515 ASSERTCMP(call->state, ==, RXRPC_CALL_COMPLETE);
517 if (test_and_set_bit(RXRPC_CALL_RELEASED, &call->flags))
520 rxrpc_put_call_slot(call);
521 del_timer_sync(&call->timer);
523 /* Make sure we don't get any more notifications */
524 write_lock_bh(&rx->recvmsg_lock);
526 if (!list_empty(&call->recvmsg_link)) {
527 _debug("unlinking once-pending call %p { e=%lx f=%lx }",
528 call, call->events, call->flags);
529 list_del(&call->recvmsg_link);
533 /* list_empty() must return false in rxrpc_notify_socket() */
534 call->recvmsg_link.next = NULL;
535 call->recvmsg_link.prev = NULL;
537 write_unlock_bh(&rx->recvmsg_lock);
539 rxrpc_put_call(call, rxrpc_call_put_unnotify);
541 write_lock(&rx->call_lock);
543 if (test_and_clear_bit(RXRPC_CALL_HAS_USERID, &call->flags)) {
544 rb_erase(&call->sock_node, &rx->calls);
545 memset(&call->sock_node, 0xdd, sizeof(call->sock_node));
546 rxrpc_put_call(call, rxrpc_call_put_userid_exists);
549 list_del(&call->sock_link);
550 write_unlock(&rx->call_lock);
552 _debug("RELEASE CALL %p (%d CONN %p)", call, call->debug_id, conn);
554 if (conn && !test_bit(RXRPC_CALL_DISCONNECTED, &call->flags))
555 rxrpc_disconnect_call(call);
557 call->security->free_call_crypto(call);
562 * release all the calls associated with a socket
564 void rxrpc_release_calls_on_socket(struct rxrpc_sock *rx)
566 struct rxrpc_call *call;
570 while (!list_empty(&rx->to_be_accepted)) {
571 call = list_entry(rx->to_be_accepted.next,
572 struct rxrpc_call, accept_link);
573 list_del(&call->accept_link);
574 rxrpc_abort_call("SKR", call, 0, RX_CALL_DEAD, -ECONNRESET);
575 rxrpc_put_call(call, rxrpc_call_put_release_sock_tba);
578 while (!list_empty(&rx->sock_calls)) {
579 call = list_entry(rx->sock_calls.next,
580 struct rxrpc_call, sock_link);
581 rxrpc_get_call(call, rxrpc_call_get_release_sock);
582 rxrpc_abort_call("SKT", call, 0, RX_CALL_DEAD, -ECONNRESET);
583 rxrpc_send_abort_packet(call);
584 rxrpc_release_call(rx, call);
585 rxrpc_put_call(call, rxrpc_call_put_release_sock);
594 void rxrpc_put_call(struct rxrpc_call *call, enum rxrpc_call_trace why)
596 struct rxrpc_net *rxnet = call->rxnet;
597 unsigned int debug_id = call->debug_id;
601 ASSERT(call != NULL);
603 dead = __refcount_dec_and_test(&call->ref, &r);
604 trace_rxrpc_call(debug_id, r - 1, 0, why);
606 ASSERTCMP(call->state, ==, RXRPC_CALL_COMPLETE);
608 if (!list_empty(&call->link)) {
609 spin_lock_bh(&rxnet->call_lock);
610 list_del_init(&call->link);
611 spin_unlock_bh(&rxnet->call_lock);
614 rxrpc_cleanup_call(call);
619 * Free up the call under RCU.
621 static void rxrpc_rcu_free_call(struct rcu_head *rcu)
623 struct rxrpc_call *call = container_of(rcu, struct rxrpc_call, rcu);
624 struct rxrpc_net *rxnet = READ_ONCE(call->rxnet);
626 kmem_cache_free(rxrpc_call_jar, call);
627 if (atomic_dec_and_test(&rxnet->nr_calls))
628 wake_up_var(&rxnet->nr_calls);
632 * Final call destruction - but must be done in process context.
634 static void rxrpc_destroy_call(struct work_struct *work)
636 struct rxrpc_call *call = container_of(work, struct rxrpc_call, destroyer);
637 struct rxrpc_txbuf *txb;
639 del_timer_sync(&call->timer);
640 cancel_work_sync(&call->processor); /* The processor may restart the timer */
641 del_timer_sync(&call->timer);
643 rxrpc_cleanup_ring(call);
644 while ((txb = list_first_entry_or_null(&call->tx_buffer,
645 struct rxrpc_txbuf, call_link))) {
646 list_del(&txb->call_link);
647 rxrpc_put_txbuf(txb, rxrpc_txbuf_put_cleaned);
649 rxrpc_put_txbuf(call->tx_pending, rxrpc_txbuf_put_cleaned);
650 rxrpc_free_skb(call->acks_soft_tbl, rxrpc_skb_put_ack);
651 rxrpc_put_connection(call->conn, rxrpc_conn_put_call);
652 rxrpc_put_peer(call->peer, rxrpc_peer_put_call);
653 rxrpc_put_local(call->local, rxrpc_local_put_call);
654 call_rcu(&call->rcu, rxrpc_rcu_free_call);
660 void rxrpc_cleanup_call(struct rxrpc_call *call)
662 memset(&call->sock_node, 0xcd, sizeof(call->sock_node));
664 ASSERTCMP(call->state, ==, RXRPC_CALL_COMPLETE);
665 ASSERT(test_bit(RXRPC_CALL_RELEASED, &call->flags));
667 del_timer_sync(&call->timer);
668 cancel_work(&call->processor);
670 if (rcu_read_lock_held() || work_busy(&call->processor))
671 /* Can't use the rxrpc workqueue as we need to cancel/flush
672 * something that may be running/waiting there.
674 schedule_work(&call->destroyer);
676 rxrpc_destroy_call(&call->destroyer);
680 * Make sure that all calls are gone from a network namespace. To reach this
681 * point, any open UDP sockets in that namespace must have been closed, so any
682 * outstanding calls cannot be doing I/O.
684 void rxrpc_destroy_all_calls(struct rxrpc_net *rxnet)
686 struct rxrpc_call *call;
690 if (!list_empty(&rxnet->calls)) {
691 spin_lock_bh(&rxnet->call_lock);
693 while (!list_empty(&rxnet->calls)) {
694 call = list_entry(rxnet->calls.next,
695 struct rxrpc_call, link);
696 _debug("Zapping call %p", call);
698 rxrpc_see_call(call, rxrpc_call_see_zap);
699 list_del_init(&call->link);
701 pr_err("Call %p still in use (%d,%s,%lx,%lx)!\n",
702 call, refcount_read(&call->ref),
703 rxrpc_call_states[call->state],
704 call->flags, call->events);
706 spin_unlock_bh(&rxnet->call_lock);
708 spin_lock_bh(&rxnet->call_lock);
711 spin_unlock_bh(&rxnet->call_lock);
714 atomic_dec(&rxnet->nr_calls);
715 wait_var_event(&rxnet->nr_calls, !atomic_read(&rxnet->nr_calls));