1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* incoming 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/module.h>
11 #include <linux/net.h>
12 #include <linux/skbuff.h>
13 #include <linux/errqueue.h>
14 #include <linux/udp.h>
16 #include <linux/in6.h>
17 #include <linux/icmp.h>
18 #include <linux/gfp.h>
19 #include <linux/circ_buf.h>
21 #include <net/af_rxrpc.h>
23 #include "ar-internal.h"
26 * Preallocate a single service call, connection and peer and, if possible,
27 * give them a user ID and attach the user's side of the ID to them.
29 static int rxrpc_service_prealloc_one(struct rxrpc_sock *rx,
30 struct rxrpc_backlog *b,
31 rxrpc_notify_rx_t notify_rx,
32 rxrpc_user_attach_call_t user_attach_call,
33 unsigned long user_call_ID, gfp_t gfp,
34 unsigned int debug_id)
36 const void *here = __builtin_return_address(0);
37 struct rxrpc_call *call;
38 struct rxrpc_net *rxnet = rxrpc_net(sock_net(&rx->sk));
40 unsigned int size = RXRPC_BACKLOG_MAX;
41 unsigned int head, tail, call_head, call_tail;
43 max = rx->sk.sk_max_ack_backlog;
44 tmp = rx->sk.sk_ack_backlog;
46 _leave(" = -ENOBUFS [full %u]", max);
51 /* We don't need more conns and peers than we have calls, but on the
52 * other hand, we shouldn't ever use more peers than conns or conns
55 call_head = b->call_backlog_head;
56 call_tail = READ_ONCE(b->call_backlog_tail);
57 tmp = CIRC_CNT(call_head, call_tail, size);
59 _leave(" = -ENOBUFS [enough %u]", tmp);
64 head = b->peer_backlog_head;
65 tail = READ_ONCE(b->peer_backlog_tail);
66 if (CIRC_CNT(head, tail, size) < max) {
67 struct rxrpc_peer *peer = rxrpc_alloc_peer(rx->local, gfp);
70 b->peer_backlog[head] = peer;
71 smp_store_release(&b->peer_backlog_head,
72 (head + 1) & (size - 1));
75 head = b->conn_backlog_head;
76 tail = READ_ONCE(b->conn_backlog_tail);
77 if (CIRC_CNT(head, tail, size) < max) {
78 struct rxrpc_connection *conn;
80 conn = rxrpc_prealloc_service_connection(rxnet, gfp);
83 b->conn_backlog[head] = conn;
84 smp_store_release(&b->conn_backlog_head,
85 (head + 1) & (size - 1));
87 trace_rxrpc_conn(conn, rxrpc_conn_new_service,
88 atomic_read(&conn->usage), here);
91 /* Now it gets complicated, because calls get registered with the
92 * socket here, particularly if a user ID is preassigned by the user.
94 call = rxrpc_alloc_call(rx, gfp, debug_id);
97 call->flags |= (1 << RXRPC_CALL_IS_SERVICE);
98 call->state = RXRPC_CALL_SERVER_PREALLOC;
100 trace_rxrpc_call(call, rxrpc_call_new_service,
101 atomic_read(&call->usage),
102 here, (const void *)user_call_ID);
104 write_lock(&rx->call_lock);
105 if (user_attach_call) {
106 struct rxrpc_call *xcall;
107 struct rb_node *parent, **pp;
109 /* Check the user ID isn't already in use */
110 pp = &rx->calls.rb_node;
114 xcall = rb_entry(parent, struct rxrpc_call, sock_node);
115 if (user_call_ID < xcall->user_call_ID)
116 pp = &(*pp)->rb_left;
117 else if (user_call_ID > xcall->user_call_ID)
118 pp = &(*pp)->rb_right;
123 call->user_call_ID = user_call_ID;
124 call->notify_rx = notify_rx;
125 rxrpc_get_call(call, rxrpc_call_got_kernel);
126 user_attach_call(call, user_call_ID);
127 rxrpc_get_call(call, rxrpc_call_got_userid);
128 rb_link_node(&call->sock_node, parent, pp);
129 rb_insert_color(&call->sock_node, &rx->calls);
130 set_bit(RXRPC_CALL_HAS_USERID, &call->flags);
133 list_add(&call->sock_link, &rx->sock_calls);
135 write_unlock(&rx->call_lock);
138 write_lock(&rxnet->call_lock);
139 list_add_tail(&call->link, &rxnet->calls);
140 write_unlock(&rxnet->call_lock);
142 b->call_backlog[call_head] = call;
143 smp_store_release(&b->call_backlog_head, (call_head + 1) & (size - 1));
144 _leave(" = 0 [%d -> %lx]", call->debug_id, user_call_ID);
148 write_unlock(&rx->call_lock);
149 rxrpc_cleanup_call(call);
150 _leave(" = -EBADSLT");
155 * Preallocate sufficient service connections, calls and peers to cover the
156 * entire backlog of a socket. When a new call comes in, if we don't have
157 * sufficient of each available, the call gets rejected as busy or ignored.
159 * The backlog is replenished when a connection is accepted or rejected.
161 int rxrpc_service_prealloc(struct rxrpc_sock *rx, gfp_t gfp)
163 struct rxrpc_backlog *b = rx->backlog;
166 b = kzalloc(sizeof(struct rxrpc_backlog), gfp);
172 if (rx->discard_new_call)
175 while (rxrpc_service_prealloc_one(rx, b, NULL, NULL, 0, gfp,
176 atomic_inc_return(&rxrpc_debug_id)) == 0)
183 * Discard the preallocation on a service.
185 void rxrpc_discard_prealloc(struct rxrpc_sock *rx)
187 struct rxrpc_backlog *b = rx->backlog;
188 struct rxrpc_net *rxnet = rxrpc_net(sock_net(&rx->sk));
189 unsigned int size = RXRPC_BACKLOG_MAX, head, tail;
195 /* Make sure that there aren't any incoming calls in progress before we
196 * clear the preallocation buffers.
198 spin_lock_bh(&rx->incoming_lock);
199 spin_unlock_bh(&rx->incoming_lock);
201 head = b->peer_backlog_head;
202 tail = b->peer_backlog_tail;
203 while (CIRC_CNT(head, tail, size) > 0) {
204 struct rxrpc_peer *peer = b->peer_backlog[tail];
206 tail = (tail + 1) & (size - 1);
209 head = b->conn_backlog_head;
210 tail = b->conn_backlog_tail;
211 while (CIRC_CNT(head, tail, size) > 0) {
212 struct rxrpc_connection *conn = b->conn_backlog[tail];
213 write_lock(&rxnet->conn_lock);
214 list_del(&conn->link);
215 list_del(&conn->proc_link);
216 write_unlock(&rxnet->conn_lock);
218 if (atomic_dec_and_test(&rxnet->nr_conns))
219 wake_up_var(&rxnet->nr_conns);
220 tail = (tail + 1) & (size - 1);
223 head = b->call_backlog_head;
224 tail = b->call_backlog_tail;
225 while (CIRC_CNT(head, tail, size) > 0) {
226 struct rxrpc_call *call = b->call_backlog[tail];
227 rcu_assign_pointer(call->socket, rx);
228 if (rx->discard_new_call) {
229 _debug("discard %lx", call->user_call_ID);
230 rx->discard_new_call(call, call->user_call_ID);
231 rxrpc_put_call(call, rxrpc_call_put_kernel);
233 rxrpc_call_completed(call);
234 rxrpc_release_call(rx, call);
235 rxrpc_put_call(call, rxrpc_call_put);
236 tail = (tail + 1) & (size - 1);
243 * Allocate a new incoming call from the prealloc pool, along with a connection
244 * and a peer as necessary.
246 static struct rxrpc_call *rxrpc_alloc_incoming_call(struct rxrpc_sock *rx,
247 struct rxrpc_local *local,
248 struct rxrpc_peer *peer,
249 struct rxrpc_connection *conn,
252 struct rxrpc_backlog *b = rx->backlog;
253 struct rxrpc_call *call;
254 unsigned short call_head, conn_head, peer_head;
255 unsigned short call_tail, conn_tail, peer_tail;
256 unsigned short call_count, conn_count;
258 /* #calls >= #conns >= #peers must hold true. */
259 call_head = smp_load_acquire(&b->call_backlog_head);
260 call_tail = b->call_backlog_tail;
261 call_count = CIRC_CNT(call_head, call_tail, RXRPC_BACKLOG_MAX);
262 conn_head = smp_load_acquire(&b->conn_backlog_head);
263 conn_tail = b->conn_backlog_tail;
264 conn_count = CIRC_CNT(conn_head, conn_tail, RXRPC_BACKLOG_MAX);
265 ASSERTCMP(conn_count, >=, call_count);
266 peer_head = smp_load_acquire(&b->peer_backlog_head);
267 peer_tail = b->peer_backlog_tail;
268 ASSERTCMP(CIRC_CNT(peer_head, peer_tail, RXRPC_BACKLOG_MAX), >=,
275 if (peer && !rxrpc_get_peer_maybe(peer))
278 peer = b->peer_backlog[peer_tail];
279 if (rxrpc_extract_addr_from_skb(&peer->srx, skb) < 0)
281 b->peer_backlog[peer_tail] = NULL;
282 smp_store_release(&b->peer_backlog_tail,
284 (RXRPC_BACKLOG_MAX - 1));
286 rxrpc_new_incoming_peer(rx, local, peer);
289 /* Now allocate and set up the connection */
290 conn = b->conn_backlog[conn_tail];
291 b->conn_backlog[conn_tail] = NULL;
292 smp_store_release(&b->conn_backlog_tail,
293 (conn_tail + 1) & (RXRPC_BACKLOG_MAX - 1));
294 conn->params.local = rxrpc_get_local(local);
295 conn->params.peer = peer;
296 rxrpc_see_connection(conn);
297 rxrpc_new_incoming_connection(rx, conn, skb);
299 rxrpc_get_connection(conn);
302 /* And now we can allocate and set up a new call */
303 call = b->call_backlog[call_tail];
304 b->call_backlog[call_tail] = NULL;
305 smp_store_release(&b->call_backlog_tail,
306 (call_tail + 1) & (RXRPC_BACKLOG_MAX - 1));
308 rxrpc_see_call(call);
310 call->peer = rxrpc_get_peer(conn->params.peer);
311 call->cong_cwnd = call->peer->cong_cwnd;
316 * Set up a new incoming call. Called in BH context with the RCU read lock
319 * If this is for a kernel service, when we allocate the call, it will have
320 * three refs on it: (1) the kernel service, (2) the user_call_ID tree, (3) the
321 * retainer ref obtained from the backlog buffer. Prealloc calls for userspace
322 * services only have the ref from the backlog buffer. We want to pass this
323 * ref to non-BH context to dispose of.
325 * If we want to report an error, we mark the skb with the packet type and
326 * abort code and return NULL.
328 * The call is returned with the user access mutex held.
330 struct rxrpc_call *rxrpc_new_incoming_call(struct rxrpc_local *local,
331 struct rxrpc_sock *rx,
334 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
335 struct rxrpc_connection *conn;
336 struct rxrpc_peer *peer = NULL;
337 struct rxrpc_call *call;
341 spin_lock(&rx->incoming_lock);
342 if (rx->sk.sk_state == RXRPC_SERVER_LISTEN_DISABLED ||
343 rx->sk.sk_state == RXRPC_CLOSE) {
344 trace_rxrpc_abort(0, "CLS", sp->hdr.cid, sp->hdr.callNumber,
345 sp->hdr.seq, RX_INVALID_OPERATION, ESHUTDOWN);
346 skb->mark = RXRPC_SKB_MARK_REJECT_ABORT;
347 skb->priority = RX_INVALID_OPERATION;
348 _leave(" = NULL [close]");
353 /* The peer, connection and call may all have sprung into existence due
354 * to a duplicate packet being handled on another CPU in parallel, so
355 * we have to recheck the routing. However, we're now holding
356 * rx->incoming_lock, so the values should remain stable.
358 conn = rxrpc_find_connection_rcu(local, skb, &peer);
360 call = rxrpc_alloc_incoming_call(rx, local, peer, conn, skb);
362 skb->mark = RXRPC_SKB_MARK_REJECT_BUSY;
363 _leave(" = NULL [busy]");
368 trace_rxrpc_receive(call, rxrpc_receive_incoming,
369 sp->hdr.serial, sp->hdr.seq);
371 /* Lock the call to prevent rxrpc_kernel_send/recv_data() and
372 * sendmsg()/recvmsg() inconveniently stealing the mutex once the
373 * notification is generated.
375 * The BUG should never happen because the kernel should be well
376 * behaved enough not to access the call before the first notification
377 * event and userspace is prevented from doing so until the state is
380 if (!mutex_trylock(&call->user_mutex))
383 /* Make the call live. */
384 rxrpc_incoming_call(rx, call, skb);
387 if (rx->notify_new_call)
388 rx->notify_new_call(&rx->sk, call, call->user_call_ID);
390 sk_acceptq_added(&rx->sk);
392 spin_lock(&conn->state_lock);
393 switch (conn->state) {
394 case RXRPC_CONN_SERVICE_UNSECURED:
395 conn->state = RXRPC_CONN_SERVICE_CHALLENGING;
396 set_bit(RXRPC_CONN_EV_CHALLENGE, &call->conn->events);
397 rxrpc_queue_conn(call->conn);
400 case RXRPC_CONN_SERVICE:
401 write_lock(&call->state_lock);
402 if (call->state < RXRPC_CALL_COMPLETE) {
403 if (rx->discard_new_call)
404 call->state = RXRPC_CALL_SERVER_RECV_REQUEST;
406 call->state = RXRPC_CALL_SERVER_ACCEPTING;
408 write_unlock(&call->state_lock);
411 case RXRPC_CONN_REMOTELY_ABORTED:
412 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
413 conn->abort_code, conn->error);
415 case RXRPC_CONN_LOCALLY_ABORTED:
416 rxrpc_abort_call("CON", call, sp->hdr.seq,
417 conn->abort_code, conn->error);
422 spin_unlock(&conn->state_lock);
424 if (call->state == RXRPC_CALL_SERVER_ACCEPTING)
425 rxrpc_notify_socket(call);
427 /* We have to discard the prealloc queue's ref here and rely on a
428 * combination of the RCU read lock and refs held either by the socket
429 * (recvmsg queue, to-be-accepted queue or user ID tree) or the kernel
430 * service to prevent the call from being deallocated too early.
432 rxrpc_put_call(call, rxrpc_call_put);
434 _leave(" = %p{%d}", call, call->debug_id);
436 spin_unlock(&rx->incoming_lock);
441 * handle acceptance of a call by userspace
442 * - assign the user call ID to the call at the front of the queue
443 * - called with the socket locked.
445 struct rxrpc_call *rxrpc_accept_call(struct rxrpc_sock *rx,
446 unsigned long user_call_ID,
447 rxrpc_notify_rx_t notify_rx)
448 __releases(&rx->sk.sk_lock.slock)
449 __acquires(call->user_mutex)
451 struct rxrpc_call *call;
452 struct rb_node *parent, **pp;
455 _enter(",%lx", user_call_ID);
457 ASSERT(!irqs_disabled());
459 write_lock(&rx->call_lock);
461 if (list_empty(&rx->to_be_accepted)) {
462 write_unlock(&rx->call_lock);
463 release_sock(&rx->sk);
464 kleave(" = -ENODATA [empty]");
465 return ERR_PTR(-ENODATA);
468 /* check the user ID isn't already in use */
469 pp = &rx->calls.rb_node;
473 call = rb_entry(parent, struct rxrpc_call, sock_node);
475 if (user_call_ID < call->user_call_ID)
476 pp = &(*pp)->rb_left;
477 else if (user_call_ID > call->user_call_ID)
478 pp = &(*pp)->rb_right;
483 /* Dequeue the first call and check it's still valid. We gain
484 * responsibility for the queue's reference.
486 call = list_entry(rx->to_be_accepted.next,
487 struct rxrpc_call, accept_link);
488 write_unlock(&rx->call_lock);
490 /* We need to gain the mutex from the interrupt handler without
491 * upsetting lockdep, so we have to release it there and take it here.
492 * We are, however, still holding the socket lock, so other accepts
493 * must wait for us and no one can add the user ID behind our backs.
495 if (mutex_lock_interruptible(&call->user_mutex) < 0) {
496 release_sock(&rx->sk);
497 kleave(" = -ERESTARTSYS");
498 return ERR_PTR(-ERESTARTSYS);
501 write_lock(&rx->call_lock);
502 list_del_init(&call->accept_link);
503 sk_acceptq_removed(&rx->sk);
504 rxrpc_see_call(call);
506 /* Find the user ID insertion point. */
507 pp = &rx->calls.rb_node;
511 call = rb_entry(parent, struct rxrpc_call, sock_node);
513 if (user_call_ID < call->user_call_ID)
514 pp = &(*pp)->rb_left;
515 else if (user_call_ID > call->user_call_ID)
516 pp = &(*pp)->rb_right;
521 write_lock_bh(&call->state_lock);
522 switch (call->state) {
523 case RXRPC_CALL_SERVER_ACCEPTING:
524 call->state = RXRPC_CALL_SERVER_RECV_REQUEST;
526 case RXRPC_CALL_COMPLETE:
533 /* formalise the acceptance */
534 call->notify_rx = notify_rx;
535 call->user_call_ID = user_call_ID;
536 rxrpc_get_call(call, rxrpc_call_got_userid);
537 rb_link_node(&call->sock_node, parent, pp);
538 rb_insert_color(&call->sock_node, &rx->calls);
539 if (test_and_set_bit(RXRPC_CALL_HAS_USERID, &call->flags))
542 write_unlock_bh(&call->state_lock);
543 write_unlock(&rx->call_lock);
544 rxrpc_notify_socket(call);
545 rxrpc_service_prealloc(rx, GFP_KERNEL);
546 release_sock(&rx->sk);
547 _leave(" = %p{%d}", call, call->debug_id);
551 _debug("release %p", call);
552 write_unlock_bh(&call->state_lock);
553 write_unlock(&rx->call_lock);
554 rxrpc_release_call(rx, call);
555 rxrpc_put_call(call, rxrpc_call_put);
560 write_unlock(&rx->call_lock);
562 rxrpc_service_prealloc(rx, GFP_KERNEL);
563 release_sock(&rx->sk);
564 _leave(" = %d", ret);
569 * Handle rejection of a call by userspace
570 * - reject the call at the front of the queue
572 int rxrpc_reject_call(struct rxrpc_sock *rx)
574 struct rxrpc_call *call;
580 ASSERT(!irqs_disabled());
582 write_lock(&rx->call_lock);
584 if (list_empty(&rx->to_be_accepted)) {
585 write_unlock(&rx->call_lock);
589 /* Dequeue the first call and check it's still valid. We gain
590 * responsibility for the queue's reference.
592 call = list_entry(rx->to_be_accepted.next,
593 struct rxrpc_call, accept_link);
594 list_del_init(&call->accept_link);
595 sk_acceptq_removed(&rx->sk);
596 rxrpc_see_call(call);
598 write_lock_bh(&call->state_lock);
599 switch (call->state) {
600 case RXRPC_CALL_SERVER_ACCEPTING:
601 __rxrpc_abort_call("REJ", call, 1, RX_USER_ABORT, -ECONNABORTED);
604 case RXRPC_CALL_COMPLETE:
612 write_unlock_bh(&call->state_lock);
613 write_unlock(&rx->call_lock);
615 rxrpc_send_abort_packet(call);
616 rxrpc_release_call(rx, call);
617 rxrpc_put_call(call, rxrpc_call_put);
619 rxrpc_service_prealloc(rx, GFP_KERNEL);
620 _leave(" = %d", ret);
625 * rxrpc_kernel_charge_accept - Charge up socket with preallocated calls
626 * @sock: The socket on which to preallocate
627 * @notify_rx: Event notification function for the call
628 * @user_attach_call: Func to attach call to user_call_ID
629 * @user_call_ID: The tag to attach to the preallocated call
630 * @gfp: The allocation conditions.
631 * @debug_id: The tracing debug ID.
633 * Charge up the socket with preallocated calls, each with a user ID. A
634 * function should be provided to effect the attachment from the user's side.
635 * The user is given a ref to hold on the call.
637 * Note that the call may be come connected before this function returns.
639 int rxrpc_kernel_charge_accept(struct socket *sock,
640 rxrpc_notify_rx_t notify_rx,
641 rxrpc_user_attach_call_t user_attach_call,
642 unsigned long user_call_ID, gfp_t gfp,
643 unsigned int debug_id)
645 struct rxrpc_sock *rx = rxrpc_sk(sock->sk);
646 struct rxrpc_backlog *b = rx->backlog;
648 if (sock->sk->sk_state == RXRPC_CLOSE)
651 return rxrpc_service_prealloc_one(rx, b, notify_rx,
652 user_attach_call, user_call_ID,
655 EXPORT_SYMBOL(rxrpc_kernel_charge_accept);