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 static void rxrpc_call_timer_expired(struct timer_list *t)
50 struct rxrpc_call *call = from_timer(call, t, timer);
52 _enter("%d", call->debug_id);
54 if (call->state < RXRPC_CALL_COMPLETE) {
55 trace_rxrpc_timer_expired(call, jiffies);
56 __rxrpc_queue_call(call);
58 rxrpc_put_call(call, rxrpc_call_put);
62 void rxrpc_reduce_call_timer(struct rxrpc_call *call,
63 unsigned long expire_at,
65 enum rxrpc_timer_trace why)
67 if (rxrpc_try_get_call(call, rxrpc_call_got_timer)) {
68 trace_rxrpc_timer(call, why, now);
69 if (timer_reduce(&call->timer, expire_at))
70 rxrpc_put_call(call, rxrpc_call_put_notimer);
74 void rxrpc_delete_call_timer(struct rxrpc_call *call)
76 if (del_timer_sync(&call->timer))
77 rxrpc_put_call(call, rxrpc_call_put_timer);
80 static struct lock_class_key rxrpc_call_user_mutex_lock_class_key;
83 * find an extant server call
84 * - called in process context with IRQs enabled
86 struct rxrpc_call *rxrpc_find_call_by_user_ID(struct rxrpc_sock *rx,
87 unsigned long user_call_ID)
89 struct rxrpc_call *call;
92 _enter("%p,%lx", rx, user_call_ID);
94 read_lock(&rx->call_lock);
96 p = rx->calls.rb_node;
98 call = rb_entry(p, struct rxrpc_call, sock_node);
100 if (user_call_ID < call->user_call_ID)
102 else if (user_call_ID > call->user_call_ID)
105 goto found_extant_call;
108 read_unlock(&rx->call_lock);
113 rxrpc_get_call(call, rxrpc_call_got);
114 read_unlock(&rx->call_lock);
115 _leave(" = %p [%d]", call, refcount_read(&call->ref));
120 * allocate a new call
122 struct rxrpc_call *rxrpc_alloc_call(struct rxrpc_sock *rx, gfp_t gfp,
123 unsigned int debug_id)
125 struct rxrpc_call *call;
126 struct rxrpc_net *rxnet = rxrpc_net(sock_net(&rx->sk));
128 call = kmem_cache_zalloc(rxrpc_call_jar, gfp);
132 mutex_init(&call->user_mutex);
134 /* Prevent lockdep reporting a deadlock false positive between the afs
135 * filesystem and sys_sendmsg() via the mmap sem.
137 if (rx->sk.sk_kern_sock)
138 lockdep_set_class(&call->user_mutex,
139 &rxrpc_call_user_mutex_lock_class_key);
141 timer_setup(&call->timer, rxrpc_call_timer_expired, 0);
142 INIT_WORK(&call->processor, &rxrpc_process_call);
143 INIT_LIST_HEAD(&call->link);
144 INIT_LIST_HEAD(&call->chan_wait_link);
145 INIT_LIST_HEAD(&call->accept_link);
146 INIT_LIST_HEAD(&call->recvmsg_link);
147 INIT_LIST_HEAD(&call->sock_link);
148 INIT_LIST_HEAD(&call->tx_buffer);
149 skb_queue_head_init(&call->recvmsg_queue);
150 skb_queue_head_init(&call->rx_oos_queue);
151 init_waitqueue_head(&call->waitq);
152 spin_lock_init(&call->notify_lock);
153 spin_lock_init(&call->tx_lock);
154 spin_lock_init(&call->input_lock);
155 spin_lock_init(&call->acks_ack_lock);
156 rwlock_init(&call->state_lock);
157 refcount_set(&call->ref, 1);
158 call->debug_id = debug_id;
159 call->tx_total_len = -1;
160 call->next_rx_timo = 20 * HZ;
161 call->next_req_timo = 1 * HZ;
162 atomic64_set(&call->ackr_window, 0x100000001ULL);
164 memset(&call->sock_node, 0xed, sizeof(call->sock_node));
166 call->rx_winsize = rxrpc_rx_window_size;
167 call->tx_winsize = 16;
169 if (RXRPC_TX_SMSS > 2190)
171 else if (RXRPC_TX_SMSS > 1095)
175 call->cong_ssthresh = RXRPC_TX_MAX_WINDOW;
178 call->rtt_avail = RXRPC_CALL_RTT_AVAIL_MASK;
179 atomic_inc(&rxnet->nr_calls);
184 * Allocate a new client call.
186 static struct rxrpc_call *rxrpc_alloc_client_call(struct rxrpc_sock *rx,
187 struct sockaddr_rxrpc *srx,
189 unsigned int debug_id)
191 struct rxrpc_call *call;
196 call = rxrpc_alloc_call(rx, gfp, debug_id);
198 return ERR_PTR(-ENOMEM);
199 call->state = RXRPC_CALL_CLIENT_AWAIT_CONN;
200 call->service_id = srx->srx_service;
201 now = ktime_get_real();
202 call->acks_latest_ts = now;
203 call->cong_tstamp = now;
205 _leave(" = %p", call);
210 * Initiate the call ack/resend/expiry timer.
212 static void rxrpc_start_call_timer(struct rxrpc_call *call)
214 unsigned long now = jiffies;
215 unsigned long j = now + MAX_JIFFY_OFFSET;
217 call->delay_ack_at = j;
218 call->ack_lost_at = j;
221 call->expect_rx_by = j;
222 call->expect_req_by = j;
223 call->expect_term_by = j;
224 call->timer.expires = now;
228 * Wait for a call slot to become available.
230 static struct semaphore *rxrpc_get_call_slot(struct rxrpc_call_params *p, gfp_t gfp)
232 struct semaphore *limiter = &rxrpc_call_limiter;
235 limiter = &rxrpc_kernel_call_limiter;
236 if (p->interruptibility == RXRPC_UNINTERRUPTIBLE) {
240 return down_interruptible(limiter) < 0 ? NULL : limiter;
244 * Release a call slot.
246 static void rxrpc_put_call_slot(struct rxrpc_call *call)
248 struct semaphore *limiter = &rxrpc_call_limiter;
250 if (test_bit(RXRPC_CALL_KERNEL, &call->flags))
251 limiter = &rxrpc_kernel_call_limiter;
256 * Set up a call for the given parameters.
257 * - Called with the socket lock held, which it must release.
258 * - If it returns a call, the call's lock will need releasing by the caller.
260 struct rxrpc_call *rxrpc_new_client_call(struct rxrpc_sock *rx,
261 struct rxrpc_conn_parameters *cp,
262 struct sockaddr_rxrpc *srx,
263 struct rxrpc_call_params *p,
265 unsigned int debug_id)
266 __releases(&rx->sk.sk_lock.slock)
267 __acquires(&call->user_mutex)
269 struct rxrpc_call *call, *xcall;
270 struct rxrpc_net *rxnet;
271 struct semaphore *limiter;
272 struct rb_node *parent, **pp;
273 const void *here = __builtin_return_address(0);
276 _enter("%p,%lx", rx, p->user_call_ID);
278 limiter = rxrpc_get_call_slot(p, gfp);
280 release_sock(&rx->sk);
281 return ERR_PTR(-ERESTARTSYS);
284 call = rxrpc_alloc_client_call(rx, srx, gfp, debug_id);
286 release_sock(&rx->sk);
288 _leave(" = %ld", PTR_ERR(call));
292 call->interruptibility = p->interruptibility;
293 call->tx_total_len = p->tx_total_len;
294 trace_rxrpc_call(call->debug_id, rxrpc_call_new_client,
295 refcount_read(&call->ref),
296 here, (const void *)p->user_call_ID);
298 __set_bit(RXRPC_CALL_KERNEL, &call->flags);
300 /* We need to protect a partially set up call against the user as we
301 * will be acting outside the socket lock.
303 mutex_lock(&call->user_mutex);
305 /* Publish the call, even though it is incompletely set up as yet */
306 write_lock(&rx->call_lock);
308 pp = &rx->calls.rb_node;
312 xcall = rb_entry(parent, struct rxrpc_call, sock_node);
314 if (p->user_call_ID < xcall->user_call_ID)
315 pp = &(*pp)->rb_left;
316 else if (p->user_call_ID > xcall->user_call_ID)
317 pp = &(*pp)->rb_right;
319 goto error_dup_user_ID;
322 rcu_assign_pointer(call->socket, rx);
323 call->user_call_ID = p->user_call_ID;
324 __set_bit(RXRPC_CALL_HAS_USERID, &call->flags);
325 rxrpc_get_call(call, rxrpc_call_got_userid);
326 rb_link_node(&call->sock_node, parent, pp);
327 rb_insert_color(&call->sock_node, &rx->calls);
328 list_add(&call->sock_link, &rx->sock_calls);
330 write_unlock(&rx->call_lock);
333 spin_lock_bh(&rxnet->call_lock);
334 list_add_tail_rcu(&call->link, &rxnet->calls);
335 spin_unlock_bh(&rxnet->call_lock);
337 /* From this point on, the call is protected by its own lock. */
338 release_sock(&rx->sk);
340 /* Set up or get a connection record and set the protocol parameters,
341 * including channel number and call ID.
343 ret = rxrpc_connect_call(rx, call, cp, srx, gfp);
345 goto error_attached_to_socket;
347 trace_rxrpc_call(call->debug_id, rxrpc_call_connected,
348 refcount_read(&call->ref), here, NULL);
350 rxrpc_start_call_timer(call);
352 _leave(" = %p [new]", call);
355 /* We unexpectedly found the user ID in the list after taking
356 * the call_lock. This shouldn't happen unless the user races
357 * with itself and tries to add the same user ID twice at the
358 * same time in different threads.
361 write_unlock(&rx->call_lock);
362 release_sock(&rx->sk);
363 __rxrpc_set_call_completion(call, RXRPC_CALL_LOCAL_ERROR,
364 RX_CALL_DEAD, -EEXIST);
365 trace_rxrpc_call(call->debug_id, rxrpc_call_error,
366 refcount_read(&call->ref), here, ERR_PTR(-EEXIST));
367 rxrpc_release_call(rx, call);
368 mutex_unlock(&call->user_mutex);
369 rxrpc_put_call(call, rxrpc_call_put);
370 _leave(" = -EEXIST");
371 return ERR_PTR(-EEXIST);
373 /* We got an error, but the call is attached to the socket and is in
374 * need of release. However, we might now race with recvmsg() when
375 * completing the call queues it. Return 0 from sys_sendmsg() and
376 * leave the error to recvmsg() to deal with.
378 error_attached_to_socket:
379 trace_rxrpc_call(call->debug_id, rxrpc_call_error,
380 refcount_read(&call->ref), here, ERR_PTR(ret));
381 set_bit(RXRPC_CALL_DISCONNECTED, &call->flags);
382 __rxrpc_set_call_completion(call, RXRPC_CALL_LOCAL_ERROR,
384 _leave(" = c=%08x [err]", call->debug_id);
389 * Set up an incoming call. call->conn points to the connection.
390 * This is called in BH context and isn't allowed to fail.
392 void rxrpc_incoming_call(struct rxrpc_sock *rx,
393 struct rxrpc_call *call,
396 struct rxrpc_connection *conn = call->conn;
397 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
400 _enter(",%d", call->conn->debug_id);
402 rcu_assign_pointer(call->socket, rx);
403 call->call_id = sp->hdr.callNumber;
404 call->service_id = sp->hdr.serviceId;
405 call->cid = sp->hdr.cid;
406 call->state = RXRPC_CALL_SERVER_SECURING;
407 call->cong_tstamp = skb->tstamp;
409 /* Set the channel for this call. We don't get channel_lock as we're
410 * only defending against the data_ready handler (which we're called
411 * from) and the RESPONSE packet parser (which is only really
412 * interested in call_counter and can cope with a disagreement with the
415 chan = sp->hdr.cid & RXRPC_CHANNELMASK;
416 conn->channels[chan].call_counter = call->call_id;
417 conn->channels[chan].call_id = call->call_id;
418 rcu_assign_pointer(conn->channels[chan].call, call);
420 spin_lock(&conn->params.peer->lock);
421 hlist_add_head_rcu(&call->error_link, &conn->params.peer->error_targets);
422 spin_unlock(&conn->params.peer->lock);
424 rxrpc_start_call_timer(call);
429 * Queue a call's work processor, getting a ref to pass to the work queue.
431 bool rxrpc_queue_call(struct rxrpc_call *call)
433 const void *here = __builtin_return_address(0);
436 if (!__refcount_inc_not_zero(&call->ref, &n))
438 if (rxrpc_queue_work(&call->processor))
439 trace_rxrpc_call(call->debug_id, rxrpc_call_queued, n + 1,
442 rxrpc_put_call(call, rxrpc_call_put_noqueue);
447 * Queue a call's work processor, passing the callers ref to the work queue.
449 bool __rxrpc_queue_call(struct rxrpc_call *call)
451 const void *here = __builtin_return_address(0);
452 int n = refcount_read(&call->ref);
454 if (rxrpc_queue_work(&call->processor))
455 trace_rxrpc_call(call->debug_id, rxrpc_call_queued_ref, n,
458 rxrpc_put_call(call, rxrpc_call_put_noqueue);
463 * Note the re-emergence of a call.
465 void rxrpc_see_call(struct rxrpc_call *call)
467 const void *here = __builtin_return_address(0);
469 int n = refcount_read(&call->ref);
471 trace_rxrpc_call(call->debug_id, rxrpc_call_seen, n,
476 bool rxrpc_try_get_call(struct rxrpc_call *call, enum rxrpc_call_trace op)
478 const void *here = __builtin_return_address(0);
481 if (!__refcount_inc_not_zero(&call->ref, &n))
483 trace_rxrpc_call(call->debug_id, op, n + 1, here, NULL);
488 * Note the addition of a ref on a call.
490 void rxrpc_get_call(struct rxrpc_call *call, enum rxrpc_call_trace op)
492 const void *here = __builtin_return_address(0);
495 __refcount_inc(&call->ref, &n);
496 trace_rxrpc_call(call->debug_id, op, n + 1, here, NULL);
500 * Clean up the Rx skb ring.
502 static void rxrpc_cleanup_ring(struct rxrpc_call *call)
504 skb_queue_purge(&call->recvmsg_queue);
505 skb_queue_purge(&call->rx_oos_queue);
509 * Detach a call from its owning socket.
511 void rxrpc_release_call(struct rxrpc_sock *rx, struct rxrpc_call *call)
513 const void *here = __builtin_return_address(0);
514 struct rxrpc_connection *conn = call->conn;
517 _enter("{%d,%d}", call->debug_id, refcount_read(&call->ref));
519 trace_rxrpc_call(call->debug_id, rxrpc_call_release,
520 refcount_read(&call->ref),
521 here, (const void *)call->flags);
523 ASSERTCMP(call->state, ==, RXRPC_CALL_COMPLETE);
525 if (test_and_set_bit(RXRPC_CALL_RELEASED, &call->flags))
528 rxrpc_put_call_slot(call);
529 rxrpc_delete_call_timer(call);
531 /* Make sure we don't get any more notifications */
532 write_lock_bh(&rx->recvmsg_lock);
534 if (!list_empty(&call->recvmsg_link)) {
535 _debug("unlinking once-pending call %p { e=%lx f=%lx }",
536 call, call->events, call->flags);
537 list_del(&call->recvmsg_link);
541 /* list_empty() must return false in rxrpc_notify_socket() */
542 call->recvmsg_link.next = NULL;
543 call->recvmsg_link.prev = NULL;
545 write_unlock_bh(&rx->recvmsg_lock);
547 rxrpc_put_call(call, rxrpc_call_put);
549 write_lock(&rx->call_lock);
551 if (test_and_clear_bit(RXRPC_CALL_HAS_USERID, &call->flags)) {
552 rb_erase(&call->sock_node, &rx->calls);
553 memset(&call->sock_node, 0xdd, sizeof(call->sock_node));
554 rxrpc_put_call(call, rxrpc_call_put_userid);
557 list_del(&call->sock_link);
558 write_unlock(&rx->call_lock);
560 _debug("RELEASE CALL %p (%d CONN %p)", call, call->debug_id, conn);
562 if (conn && !test_bit(RXRPC_CALL_DISCONNECTED, &call->flags))
563 rxrpc_disconnect_call(call);
565 call->security->free_call_crypto(call);
570 * release all the calls associated with a socket
572 void rxrpc_release_calls_on_socket(struct rxrpc_sock *rx)
574 struct rxrpc_call *call;
578 while (!list_empty(&rx->to_be_accepted)) {
579 call = list_entry(rx->to_be_accepted.next,
580 struct rxrpc_call, accept_link);
581 list_del(&call->accept_link);
582 rxrpc_abort_call("SKR", call, 0, RX_CALL_DEAD, -ECONNRESET);
583 rxrpc_put_call(call, rxrpc_call_put);
586 while (!list_empty(&rx->sock_calls)) {
587 call = list_entry(rx->sock_calls.next,
588 struct rxrpc_call, sock_link);
589 rxrpc_get_call(call, rxrpc_call_got);
590 rxrpc_abort_call("SKT", call, 0, RX_CALL_DEAD, -ECONNRESET);
591 rxrpc_send_abort_packet(call);
592 rxrpc_release_call(rx, call);
593 rxrpc_put_call(call, rxrpc_call_put);
602 void rxrpc_put_call(struct rxrpc_call *call, enum rxrpc_call_trace op)
604 struct rxrpc_net *rxnet = call->rxnet;
605 const void *here = __builtin_return_address(0);
606 unsigned int debug_id = call->debug_id;
610 ASSERT(call != NULL);
612 dead = __refcount_dec_and_test(&call->ref, &n);
613 trace_rxrpc_call(debug_id, op, n, here, NULL);
615 _debug("call %d dead", call->debug_id);
616 ASSERTCMP(call->state, ==, RXRPC_CALL_COMPLETE);
618 if (!list_empty(&call->link)) {
619 spin_lock_bh(&rxnet->call_lock);
620 list_del_init(&call->link);
621 spin_unlock_bh(&rxnet->call_lock);
624 rxrpc_cleanup_call(call);
629 * Final call destruction - but must be done in process context.
631 static void rxrpc_destroy_call(struct work_struct *work)
633 struct rxrpc_call *call = container_of(work, struct rxrpc_call, processor);
634 struct rxrpc_net *rxnet = call->rxnet;
636 rxrpc_delete_call_timer(call);
638 rxrpc_put_connection(call->conn);
639 rxrpc_put_peer(call->peer);
640 kmem_cache_free(rxrpc_call_jar, call);
641 if (atomic_dec_and_test(&rxnet->nr_calls))
642 wake_up_var(&rxnet->nr_calls);
646 * Final call destruction under RCU.
648 static void rxrpc_rcu_destroy_call(struct rcu_head *rcu)
650 struct rxrpc_call *call = container_of(rcu, struct rxrpc_call, rcu);
653 INIT_WORK(&call->processor, rxrpc_destroy_call);
654 if (!rxrpc_queue_work(&call->processor))
657 rxrpc_destroy_call(&call->processor);
664 void rxrpc_cleanup_call(struct rxrpc_call *call)
666 struct rxrpc_txbuf *txb;
668 memset(&call->sock_node, 0xcd, sizeof(call->sock_node));
670 ASSERTCMP(call->state, ==, RXRPC_CALL_COMPLETE);
671 ASSERT(test_bit(RXRPC_CALL_RELEASED, &call->flags));
673 rxrpc_cleanup_ring(call);
674 while ((txb = list_first_entry_or_null(&call->tx_buffer,
675 struct rxrpc_txbuf, call_link))) {
676 list_del(&txb->call_link);
677 rxrpc_put_txbuf(txb, rxrpc_txbuf_put_cleaned);
679 rxrpc_put_txbuf(call->tx_pending, rxrpc_txbuf_put_cleaned);
680 rxrpc_free_skb(call->acks_soft_tbl, rxrpc_skb_cleaned);
682 call_rcu(&call->rcu, rxrpc_rcu_destroy_call);
686 * Make sure that all calls are gone from a network namespace. To reach this
687 * point, any open UDP sockets in that namespace must have been closed, so any
688 * outstanding calls cannot be doing I/O.
690 void rxrpc_destroy_all_calls(struct rxrpc_net *rxnet)
692 struct rxrpc_call *call;
696 if (!list_empty(&rxnet->calls)) {
697 spin_lock_bh(&rxnet->call_lock);
699 while (!list_empty(&rxnet->calls)) {
700 call = list_entry(rxnet->calls.next,
701 struct rxrpc_call, link);
702 _debug("Zapping call %p", call);
704 rxrpc_see_call(call);
705 list_del_init(&call->link);
707 pr_err("Call %p still in use (%d,%s,%lx,%lx)!\n",
708 call, refcount_read(&call->ref),
709 rxrpc_call_states[call->state],
710 call->flags, call->events);
712 spin_unlock_bh(&rxnet->call_lock);
714 spin_lock_bh(&rxnet->call_lock);
717 spin_unlock_bh(&rxnet->call_lock);
720 atomic_dec(&rxnet->nr_calls);
721 wait_var_event(&rxnet->nr_calls, !atomic_read(&rxnet->nr_calls));