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;
51 struct rxrpc_peer *peer = call->peer;
54 if (WARN_ON_ONCE(!peer))
58 if (call->state < RXRPC_CALL_COMPLETE) {
59 spin_lock_bh(&local->lock);
60 busy = !list_empty(&call->attend_link);
61 trace_rxrpc_poke_call(call, busy, what);
63 rxrpc_get_call(call, rxrpc_call_get_poke);
64 list_add_tail(&call->attend_link, &local->call_attend_q);
66 spin_unlock_bh(&local->lock);
67 rxrpc_wake_up_io_thread(local);
71 static void rxrpc_call_timer_expired(struct timer_list *t)
73 struct rxrpc_call *call = from_timer(call, t, timer);
75 _enter("%d", call->debug_id);
77 if (call->state < RXRPC_CALL_COMPLETE) {
78 trace_rxrpc_timer_expired(call, jiffies);
79 rxrpc_queue_call(call, rxrpc_call_queue_timer);
83 void rxrpc_reduce_call_timer(struct rxrpc_call *call,
84 unsigned long expire_at,
86 enum rxrpc_timer_trace why)
88 trace_rxrpc_timer(call, why, now);
89 timer_reduce(&call->timer, expire_at);
92 static struct lock_class_key rxrpc_call_user_mutex_lock_class_key;
94 static void rxrpc_destroy_call(struct work_struct *);
97 * find an extant server call
98 * - called in process context with IRQs enabled
100 struct rxrpc_call *rxrpc_find_call_by_user_ID(struct rxrpc_sock *rx,
101 unsigned long user_call_ID)
103 struct rxrpc_call *call;
106 _enter("%p,%lx", rx, user_call_ID);
108 read_lock(&rx->call_lock);
110 p = rx->calls.rb_node;
112 call = rb_entry(p, struct rxrpc_call, sock_node);
114 if (user_call_ID < call->user_call_ID)
116 else if (user_call_ID > call->user_call_ID)
119 goto found_extant_call;
122 read_unlock(&rx->call_lock);
127 rxrpc_get_call(call, rxrpc_call_get_sendmsg);
128 read_unlock(&rx->call_lock);
129 _leave(" = %p [%d]", call, refcount_read(&call->ref));
134 * allocate a new call
136 struct rxrpc_call *rxrpc_alloc_call(struct rxrpc_sock *rx, gfp_t gfp,
137 unsigned int debug_id)
139 struct rxrpc_call *call;
140 struct rxrpc_net *rxnet = rxrpc_net(sock_net(&rx->sk));
142 call = kmem_cache_zalloc(rxrpc_call_jar, gfp);
146 mutex_init(&call->user_mutex);
148 /* Prevent lockdep reporting a deadlock false positive between the afs
149 * filesystem and sys_sendmsg() via the mmap sem.
151 if (rx->sk.sk_kern_sock)
152 lockdep_set_class(&call->user_mutex,
153 &rxrpc_call_user_mutex_lock_class_key);
155 timer_setup(&call->timer, rxrpc_call_timer_expired, 0);
156 INIT_WORK(&call->processor, rxrpc_process_call);
157 INIT_WORK(&call->destroyer, rxrpc_destroy_call);
158 INIT_LIST_HEAD(&call->link);
159 INIT_LIST_HEAD(&call->chan_wait_link);
160 INIT_LIST_HEAD(&call->accept_link);
161 INIT_LIST_HEAD(&call->recvmsg_link);
162 INIT_LIST_HEAD(&call->sock_link);
163 INIT_LIST_HEAD(&call->attend_link);
164 INIT_LIST_HEAD(&call->tx_buffer);
165 skb_queue_head_init(&call->recvmsg_queue);
166 skb_queue_head_init(&call->rx_oos_queue);
167 init_waitqueue_head(&call->waitq);
168 spin_lock_init(&call->notify_lock);
169 spin_lock_init(&call->tx_lock);
170 spin_lock_init(&call->acks_ack_lock);
171 rwlock_init(&call->state_lock);
172 refcount_set(&call->ref, 1);
173 call->debug_id = debug_id;
174 call->tx_total_len = -1;
175 call->next_rx_timo = 20 * HZ;
176 call->next_req_timo = 1 * HZ;
177 atomic64_set(&call->ackr_window, 0x100000001ULL);
179 memset(&call->sock_node, 0xed, sizeof(call->sock_node));
181 call->rx_winsize = rxrpc_rx_window_size;
182 call->tx_winsize = 16;
184 if (RXRPC_TX_SMSS > 2190)
186 else if (RXRPC_TX_SMSS > 1095)
190 call->cong_ssthresh = RXRPC_TX_MAX_WINDOW;
193 call->rtt_avail = RXRPC_CALL_RTT_AVAIL_MASK;
194 atomic_inc(&rxnet->nr_calls);
199 * Allocate a new client call.
201 static struct rxrpc_call *rxrpc_alloc_client_call(struct rxrpc_sock *rx,
202 struct sockaddr_rxrpc *srx,
204 unsigned int debug_id)
206 struct rxrpc_call *call;
211 call = rxrpc_alloc_call(rx, gfp, debug_id);
213 return ERR_PTR(-ENOMEM);
214 call->state = RXRPC_CALL_CLIENT_AWAIT_CONN;
215 call->service_id = srx->srx_service;
216 now = ktime_get_real();
217 call->acks_latest_ts = now;
218 call->cong_tstamp = now;
220 _leave(" = %p", call);
225 * Initiate the call ack/resend/expiry timer.
227 static void rxrpc_start_call_timer(struct rxrpc_call *call)
229 unsigned long now = jiffies;
230 unsigned long j = now + MAX_JIFFY_OFFSET;
232 call->delay_ack_at = j;
233 call->ack_lost_at = j;
236 call->expect_rx_by = j;
237 call->expect_req_by = j;
238 call->expect_term_by = j;
239 call->timer.expires = now;
243 * Wait for a call slot to become available.
245 static struct semaphore *rxrpc_get_call_slot(struct rxrpc_call_params *p, gfp_t gfp)
247 struct semaphore *limiter = &rxrpc_call_limiter;
250 limiter = &rxrpc_kernel_call_limiter;
251 if (p->interruptibility == RXRPC_UNINTERRUPTIBLE) {
255 return down_interruptible(limiter) < 0 ? NULL : limiter;
259 * Release a call slot.
261 static void rxrpc_put_call_slot(struct rxrpc_call *call)
263 struct semaphore *limiter = &rxrpc_call_limiter;
265 if (test_bit(RXRPC_CALL_KERNEL, &call->flags))
266 limiter = &rxrpc_kernel_call_limiter;
271 * Set up a call for the given parameters.
272 * - Called with the socket lock held, which it must release.
273 * - If it returns a call, the call's lock will need releasing by the caller.
275 struct rxrpc_call *rxrpc_new_client_call(struct rxrpc_sock *rx,
276 struct rxrpc_conn_parameters *cp,
277 struct sockaddr_rxrpc *srx,
278 struct rxrpc_call_params *p,
280 unsigned int debug_id)
281 __releases(&rx->sk.sk_lock.slock)
282 __acquires(&call->user_mutex)
284 struct rxrpc_call *call, *xcall;
285 struct rxrpc_net *rxnet;
286 struct semaphore *limiter;
287 struct rb_node *parent, **pp;
290 _enter("%p,%lx", rx, p->user_call_ID);
292 limiter = rxrpc_get_call_slot(p, gfp);
294 release_sock(&rx->sk);
295 return ERR_PTR(-ERESTARTSYS);
298 call = rxrpc_alloc_client_call(rx, srx, gfp, debug_id);
300 release_sock(&rx->sk);
302 _leave(" = %ld", PTR_ERR(call));
306 call->interruptibility = p->interruptibility;
307 call->tx_total_len = p->tx_total_len;
308 trace_rxrpc_call(call->debug_id, refcount_read(&call->ref),
309 p->user_call_ID, rxrpc_call_new_client);
311 __set_bit(RXRPC_CALL_KERNEL, &call->flags);
313 /* We need to protect a partially set up call against the user as we
314 * will be acting outside the socket lock.
316 mutex_lock(&call->user_mutex);
318 /* Publish the call, even though it is incompletely set up as yet */
319 write_lock(&rx->call_lock);
321 pp = &rx->calls.rb_node;
325 xcall = rb_entry(parent, struct rxrpc_call, sock_node);
327 if (p->user_call_ID < xcall->user_call_ID)
328 pp = &(*pp)->rb_left;
329 else if (p->user_call_ID > xcall->user_call_ID)
330 pp = &(*pp)->rb_right;
332 goto error_dup_user_ID;
335 rcu_assign_pointer(call->socket, rx);
336 call->user_call_ID = p->user_call_ID;
337 __set_bit(RXRPC_CALL_HAS_USERID, &call->flags);
338 rxrpc_get_call(call, rxrpc_call_get_userid);
339 rb_link_node(&call->sock_node, parent, pp);
340 rb_insert_color(&call->sock_node, &rx->calls);
341 list_add(&call->sock_link, &rx->sock_calls);
343 write_unlock(&rx->call_lock);
346 spin_lock_bh(&rxnet->call_lock);
347 list_add_tail_rcu(&call->link, &rxnet->calls);
348 spin_unlock_bh(&rxnet->call_lock);
350 /* From this point on, the call is protected by its own lock. */
351 release_sock(&rx->sk);
353 /* Set up or get a connection record and set the protocol parameters,
354 * including channel number and call ID.
356 ret = rxrpc_connect_call(rx, call, cp, srx, gfp);
358 goto error_attached_to_socket;
360 rxrpc_see_call(call, rxrpc_call_see_connected);
362 rxrpc_start_call_timer(call);
364 _leave(" = %p [new]", call);
367 /* We unexpectedly found the user ID in the list after taking
368 * the call_lock. This shouldn't happen unless the user races
369 * with itself and tries to add the same user ID twice at the
370 * same time in different threads.
373 write_unlock(&rx->call_lock);
374 release_sock(&rx->sk);
375 __rxrpc_set_call_completion(call, RXRPC_CALL_LOCAL_ERROR,
376 RX_CALL_DEAD, -EEXIST);
377 trace_rxrpc_call(call->debug_id, refcount_read(&call->ref), 0,
378 rxrpc_call_see_userid_exists);
379 rxrpc_release_call(rx, call);
380 mutex_unlock(&call->user_mutex);
381 rxrpc_put_call(call, rxrpc_call_put_userid_exists);
382 _leave(" = -EEXIST");
383 return ERR_PTR(-EEXIST);
385 /* We got an error, but the call is attached to the socket and is in
386 * need of release. However, we might now race with recvmsg() when
387 * completing the call queues it. Return 0 from sys_sendmsg() and
388 * leave the error to recvmsg() to deal with.
390 error_attached_to_socket:
391 trace_rxrpc_call(call->debug_id, refcount_read(&call->ref), ret,
392 rxrpc_call_see_connect_failed);
393 set_bit(RXRPC_CALL_DISCONNECTED, &call->flags);
394 __rxrpc_set_call_completion(call, RXRPC_CALL_LOCAL_ERROR,
396 _leave(" = c=%08x [err]", call->debug_id);
401 * Set up an incoming call. call->conn points to the connection.
402 * This is called in BH context and isn't allowed to fail.
404 void rxrpc_incoming_call(struct rxrpc_sock *rx,
405 struct rxrpc_call *call,
408 struct rxrpc_connection *conn = call->conn;
409 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
412 _enter(",%d", call->conn->debug_id);
414 rcu_assign_pointer(call->socket, rx);
415 call->call_id = sp->hdr.callNumber;
416 call->service_id = sp->hdr.serviceId;
417 call->cid = sp->hdr.cid;
418 call->state = RXRPC_CALL_SERVER_SECURING;
419 call->cong_tstamp = skb->tstamp;
421 /* Set the channel for this call. We don't get channel_lock as we're
422 * only defending against the data_ready handler (which we're called
423 * from) and the RESPONSE packet parser (which is only really
424 * interested in call_counter and can cope with a disagreement with the
427 chan = sp->hdr.cid & RXRPC_CHANNELMASK;
428 conn->channels[chan].call_counter = call->call_id;
429 conn->channels[chan].call_id = call->call_id;
430 rcu_assign_pointer(conn->channels[chan].call, call);
432 spin_lock(&conn->peer->lock);
433 hlist_add_head_rcu(&call->error_link, &conn->peer->error_targets);
434 spin_unlock(&conn->peer->lock);
436 rxrpc_start_call_timer(call);
441 * Queue a call's work processor.
443 void rxrpc_queue_call(struct rxrpc_call *call, enum rxrpc_call_trace why)
445 if (rxrpc_queue_work(&call->processor))
446 trace_rxrpc_call(call->debug_id, refcount_read(&call->ref), 0, why);
450 * Note the re-emergence of a call.
452 void rxrpc_see_call(struct rxrpc_call *call, enum rxrpc_call_trace why)
455 int r = refcount_read(&call->ref);
457 trace_rxrpc_call(call->debug_id, r, 0, why);
461 bool rxrpc_try_get_call(struct rxrpc_call *call, enum rxrpc_call_trace why)
465 if (!__refcount_inc_not_zero(&call->ref, &r))
467 trace_rxrpc_call(call->debug_id, r + 1, 0, why);
472 * Note the addition of a ref on a call.
474 void rxrpc_get_call(struct rxrpc_call *call, enum rxrpc_call_trace why)
478 __refcount_inc(&call->ref, &r);
479 trace_rxrpc_call(call->debug_id, r + 1, 0, why);
483 * Clean up the Rx skb ring.
485 static void rxrpc_cleanup_ring(struct rxrpc_call *call)
487 skb_queue_purge(&call->recvmsg_queue);
488 skb_queue_purge(&call->rx_oos_queue);
492 * Detach a call from its owning socket.
494 void rxrpc_release_call(struct rxrpc_sock *rx, struct rxrpc_call *call)
496 struct rxrpc_connection *conn = call->conn;
499 _enter("{%d,%d}", call->debug_id, refcount_read(&call->ref));
501 trace_rxrpc_call(call->debug_id, refcount_read(&call->ref),
502 call->flags, rxrpc_call_see_release);
504 ASSERTCMP(call->state, ==, RXRPC_CALL_COMPLETE);
506 if (test_and_set_bit(RXRPC_CALL_RELEASED, &call->flags))
509 rxrpc_put_call_slot(call);
510 del_timer_sync(&call->timer);
512 /* Make sure we don't get any more notifications */
513 write_lock_bh(&rx->recvmsg_lock);
515 if (!list_empty(&call->recvmsg_link)) {
516 _debug("unlinking once-pending call %p { e=%lx f=%lx }",
517 call, call->events, call->flags);
518 list_del(&call->recvmsg_link);
522 /* list_empty() must return false in rxrpc_notify_socket() */
523 call->recvmsg_link.next = NULL;
524 call->recvmsg_link.prev = NULL;
526 write_unlock_bh(&rx->recvmsg_lock);
528 rxrpc_put_call(call, rxrpc_call_put_unnotify);
530 write_lock(&rx->call_lock);
532 if (test_and_clear_bit(RXRPC_CALL_HAS_USERID, &call->flags)) {
533 rb_erase(&call->sock_node, &rx->calls);
534 memset(&call->sock_node, 0xdd, sizeof(call->sock_node));
535 rxrpc_put_call(call, rxrpc_call_put_userid_exists);
538 list_del(&call->sock_link);
539 write_unlock(&rx->call_lock);
541 _debug("RELEASE CALL %p (%d CONN %p)", call, call->debug_id, conn);
543 if (conn && !test_bit(RXRPC_CALL_DISCONNECTED, &call->flags))
544 rxrpc_disconnect_call(call);
546 call->security->free_call_crypto(call);
551 * release all the calls associated with a socket
553 void rxrpc_release_calls_on_socket(struct rxrpc_sock *rx)
555 struct rxrpc_call *call;
559 while (!list_empty(&rx->to_be_accepted)) {
560 call = list_entry(rx->to_be_accepted.next,
561 struct rxrpc_call, accept_link);
562 list_del(&call->accept_link);
563 rxrpc_abort_call("SKR", call, 0, RX_CALL_DEAD, -ECONNRESET);
564 rxrpc_put_call(call, rxrpc_call_put_release_sock_tba);
567 while (!list_empty(&rx->sock_calls)) {
568 call = list_entry(rx->sock_calls.next,
569 struct rxrpc_call, sock_link);
570 rxrpc_get_call(call, rxrpc_call_get_release_sock);
571 rxrpc_abort_call("SKT", call, 0, RX_CALL_DEAD, -ECONNRESET);
572 rxrpc_send_abort_packet(call);
573 rxrpc_release_call(rx, call);
574 rxrpc_put_call(call, rxrpc_call_put_release_sock);
583 void rxrpc_put_call(struct rxrpc_call *call, enum rxrpc_call_trace why)
585 struct rxrpc_net *rxnet = call->rxnet;
586 unsigned int debug_id = call->debug_id;
590 ASSERT(call != NULL);
592 dead = __refcount_dec_and_test(&call->ref, &r);
593 trace_rxrpc_call(debug_id, r - 1, 0, why);
595 ASSERTCMP(call->state, ==, RXRPC_CALL_COMPLETE);
597 if (!list_empty(&call->link)) {
598 spin_lock_bh(&rxnet->call_lock);
599 list_del_init(&call->link);
600 spin_unlock_bh(&rxnet->call_lock);
603 rxrpc_cleanup_call(call);
608 * Free up the call under RCU.
610 static void rxrpc_rcu_free_call(struct rcu_head *rcu)
612 struct rxrpc_call *call = container_of(rcu, struct rxrpc_call, rcu);
613 struct rxrpc_net *rxnet = READ_ONCE(call->rxnet);
615 kmem_cache_free(rxrpc_call_jar, call);
616 if (atomic_dec_and_test(&rxnet->nr_calls))
617 wake_up_var(&rxnet->nr_calls);
621 * Final call destruction - but must be done in process context.
623 static void rxrpc_destroy_call(struct work_struct *work)
625 struct rxrpc_call *call = container_of(work, struct rxrpc_call, destroyer);
626 struct rxrpc_txbuf *txb;
628 del_timer_sync(&call->timer);
629 cancel_work_sync(&call->processor); /* The processor may restart the timer */
630 del_timer_sync(&call->timer);
632 rxrpc_cleanup_ring(call);
633 while ((txb = list_first_entry_or_null(&call->tx_buffer,
634 struct rxrpc_txbuf, call_link))) {
635 list_del(&txb->call_link);
636 rxrpc_put_txbuf(txb, rxrpc_txbuf_put_cleaned);
638 rxrpc_put_txbuf(call->tx_pending, rxrpc_txbuf_put_cleaned);
639 rxrpc_free_skb(call->acks_soft_tbl, rxrpc_skb_put_ack);
640 rxrpc_put_connection(call->conn, rxrpc_conn_put_call);
641 rxrpc_put_peer(call->peer, rxrpc_peer_put_call);
642 call_rcu(&call->rcu, rxrpc_rcu_free_call);
648 void rxrpc_cleanup_call(struct rxrpc_call *call)
650 memset(&call->sock_node, 0xcd, sizeof(call->sock_node));
652 ASSERTCMP(call->state, ==, RXRPC_CALL_COMPLETE);
653 ASSERT(test_bit(RXRPC_CALL_RELEASED, &call->flags));
655 del_timer_sync(&call->timer);
656 cancel_work(&call->processor);
658 if (rcu_read_lock_held() || work_busy(&call->processor))
659 /* Can't use the rxrpc workqueue as we need to cancel/flush
660 * something that may be running/waiting there.
662 schedule_work(&call->destroyer);
664 rxrpc_destroy_call(&call->destroyer);
668 * Make sure that all calls are gone from a network namespace. To reach this
669 * point, any open UDP sockets in that namespace must have been closed, so any
670 * outstanding calls cannot be doing I/O.
672 void rxrpc_destroy_all_calls(struct rxrpc_net *rxnet)
674 struct rxrpc_call *call;
678 if (!list_empty(&rxnet->calls)) {
679 spin_lock_bh(&rxnet->call_lock);
681 while (!list_empty(&rxnet->calls)) {
682 call = list_entry(rxnet->calls.next,
683 struct rxrpc_call, link);
684 _debug("Zapping call %p", call);
686 rxrpc_see_call(call, rxrpc_call_see_zap);
687 list_del_init(&call->link);
689 pr_err("Call %p still in use (%d,%s,%lx,%lx)!\n",
690 call, refcount_read(&call->ref),
691 rxrpc_call_states[call->state],
692 call->flags, call->events);
694 spin_unlock_bh(&rxnet->call_lock);
696 spin_lock_bh(&rxnet->call_lock);
699 spin_unlock_bh(&rxnet->call_lock);
702 atomic_dec(&rxnet->nr_calls);
703 wait_var_event(&rxnet->nr_calls, !atomic_read(&rxnet->nr_calls));