1 /* incoming call handling
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
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.
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 #include <linux/module.h>
15 #include <linux/net.h>
16 #include <linux/skbuff.h>
17 #include <linux/errqueue.h>
18 #include <linux/udp.h>
20 #include <linux/in6.h>
21 #include <linux/icmp.h>
22 #include <linux/gfp.h>
23 #include <linux/circ_buf.h>
25 #include <net/af_rxrpc.h>
27 #include "ar-internal.h"
29 static void rxrpc_dummy_notify(struct sock *sk, struct rxrpc_call *call,
30 unsigned long user_call_ID)
35 * Preallocate a single service call, connection and peer and, if possible,
36 * give them a user ID and attach the user's side of the ID to them.
38 static int rxrpc_service_prealloc_one(struct rxrpc_sock *rx,
39 struct rxrpc_backlog *b,
40 rxrpc_notify_rx_t notify_rx,
41 rxrpc_user_attach_call_t user_attach_call,
42 unsigned long user_call_ID, gfp_t gfp,
43 unsigned int debug_id)
45 const void *here = __builtin_return_address(0);
46 struct rxrpc_call *call;
47 struct rxrpc_net *rxnet = rxrpc_net(sock_net(&rx->sk));
49 unsigned int size = RXRPC_BACKLOG_MAX;
50 unsigned int head, tail, call_head, call_tail;
52 max = rx->sk.sk_max_ack_backlog;
53 tmp = rx->sk.sk_ack_backlog;
55 _leave(" = -ENOBUFS [full %u]", max);
60 /* We don't need more conns and peers than we have calls, but on the
61 * other hand, we shouldn't ever use more peers than conns or conns
64 call_head = b->call_backlog_head;
65 call_tail = READ_ONCE(b->call_backlog_tail);
66 tmp = CIRC_CNT(call_head, call_tail, size);
68 _leave(" = -ENOBUFS [enough %u]", tmp);
73 head = b->peer_backlog_head;
74 tail = READ_ONCE(b->peer_backlog_tail);
75 if (CIRC_CNT(head, tail, size) < max) {
76 struct rxrpc_peer *peer = rxrpc_alloc_peer(rx->local, gfp);
79 b->peer_backlog[head] = peer;
80 smp_store_release(&b->peer_backlog_head,
81 (head + 1) & (size - 1));
84 head = b->conn_backlog_head;
85 tail = READ_ONCE(b->conn_backlog_tail);
86 if (CIRC_CNT(head, tail, size) < max) {
87 struct rxrpc_connection *conn;
89 conn = rxrpc_prealloc_service_connection(rxnet, gfp);
92 b->conn_backlog[head] = conn;
93 smp_store_release(&b->conn_backlog_head,
94 (head + 1) & (size - 1));
96 trace_rxrpc_conn(conn->debug_id, rxrpc_conn_new_service,
97 atomic_read(&conn->usage), here);
100 /* Now it gets complicated, because calls get registered with the
101 * socket here, particularly if a user ID is preassigned by the user.
103 call = rxrpc_alloc_call(rx, gfp, debug_id);
106 call->flags |= (1 << RXRPC_CALL_IS_SERVICE);
107 call->state = RXRPC_CALL_SERVER_PREALLOC;
109 trace_rxrpc_call(call, rxrpc_call_new_service,
110 atomic_read(&call->usage),
111 here, (const void *)user_call_ID);
113 write_lock(&rx->call_lock);
114 if (user_attach_call) {
115 struct rxrpc_call *xcall;
116 struct rb_node *parent, **pp;
118 /* Check the user ID isn't already in use */
119 pp = &rx->calls.rb_node;
123 xcall = rb_entry(parent, struct rxrpc_call, sock_node);
124 if (user_call_ID < xcall->user_call_ID)
125 pp = &(*pp)->rb_left;
126 else if (user_call_ID > xcall->user_call_ID)
127 pp = &(*pp)->rb_right;
132 call->user_call_ID = user_call_ID;
133 call->notify_rx = notify_rx;
134 rxrpc_get_call(call, rxrpc_call_got_kernel);
135 user_attach_call(call, user_call_ID);
136 rxrpc_get_call(call, rxrpc_call_got_userid);
137 rb_link_node(&call->sock_node, parent, pp);
138 rb_insert_color(&call->sock_node, &rx->calls);
139 set_bit(RXRPC_CALL_HAS_USERID, &call->flags);
142 list_add(&call->sock_link, &rx->sock_calls);
144 write_unlock(&rx->call_lock);
147 write_lock(&rxnet->call_lock);
148 list_add_tail(&call->link, &rxnet->calls);
149 write_unlock(&rxnet->call_lock);
151 b->call_backlog[call_head] = call;
152 smp_store_release(&b->call_backlog_head, (call_head + 1) & (size - 1));
153 _leave(" = 0 [%d -> %lx]", call->debug_id, user_call_ID);
157 write_unlock(&rx->call_lock);
158 rxrpc_cleanup_call(call);
159 _leave(" = -EBADSLT");
164 * Preallocate sufficient service connections, calls and peers to cover the
165 * entire backlog of a socket. When a new call comes in, if we don't have
166 * sufficient of each available, the call gets rejected as busy or ignored.
168 * The backlog is replenished when a connection is accepted or rejected.
170 int rxrpc_service_prealloc(struct rxrpc_sock *rx, gfp_t gfp)
172 struct rxrpc_backlog *b = rx->backlog;
175 b = kzalloc(sizeof(struct rxrpc_backlog), gfp);
181 if (rx->discard_new_call)
184 while (rxrpc_service_prealloc_one(rx, b, NULL, NULL, 0, gfp,
185 atomic_inc_return(&rxrpc_debug_id)) == 0)
192 * Discard the preallocation on a service.
194 void rxrpc_discard_prealloc(struct rxrpc_sock *rx)
196 struct rxrpc_backlog *b = rx->backlog;
197 struct rxrpc_net *rxnet = rxrpc_net(sock_net(&rx->sk));
198 unsigned int size = RXRPC_BACKLOG_MAX, head, tail;
204 /* Make sure that there aren't any incoming calls in progress before we
205 * clear the preallocation buffers.
207 spin_lock_bh(&rx->incoming_lock);
208 spin_unlock_bh(&rx->incoming_lock);
210 head = b->peer_backlog_head;
211 tail = b->peer_backlog_tail;
212 while (CIRC_CNT(head, tail, size) > 0) {
213 struct rxrpc_peer *peer = b->peer_backlog[tail];
214 rxrpc_put_local(peer->local);
216 tail = (tail + 1) & (size - 1);
219 head = b->conn_backlog_head;
220 tail = b->conn_backlog_tail;
221 while (CIRC_CNT(head, tail, size) > 0) {
222 struct rxrpc_connection *conn = b->conn_backlog[tail];
223 write_lock(&rxnet->conn_lock);
224 list_del(&conn->link);
225 list_del(&conn->proc_link);
226 write_unlock(&rxnet->conn_lock);
228 if (atomic_dec_and_test(&rxnet->nr_conns))
229 wake_up_var(&rxnet->nr_conns);
230 tail = (tail + 1) & (size - 1);
233 head = b->call_backlog_head;
234 tail = b->call_backlog_tail;
235 while (CIRC_CNT(head, tail, size) > 0) {
236 struct rxrpc_call *call = b->call_backlog[tail];
237 rcu_assign_pointer(call->socket, rx);
238 if (rx->discard_new_call) {
239 _debug("discard %lx", call->user_call_ID);
240 rx->discard_new_call(call, call->user_call_ID);
242 call->notify_rx = rxrpc_dummy_notify;
243 rxrpc_put_call(call, rxrpc_call_put_kernel);
245 rxrpc_call_completed(call);
246 rxrpc_release_call(rx, call);
247 rxrpc_put_call(call, rxrpc_call_put);
248 tail = (tail + 1) & (size - 1);
255 * Allocate a new incoming call from the prealloc pool, along with a connection
256 * and a peer as necessary.
258 static struct rxrpc_call *rxrpc_alloc_incoming_call(struct rxrpc_sock *rx,
259 struct rxrpc_local *local,
260 struct rxrpc_peer *peer,
261 struct rxrpc_connection *conn,
264 struct rxrpc_backlog *b = rx->backlog;
265 struct rxrpc_call *call;
266 unsigned short call_head, conn_head, peer_head;
267 unsigned short call_tail, conn_tail, peer_tail;
268 unsigned short call_count, conn_count;
270 /* #calls >= #conns >= #peers must hold true. */
271 call_head = smp_load_acquire(&b->call_backlog_head);
272 call_tail = b->call_backlog_tail;
273 call_count = CIRC_CNT(call_head, call_tail, RXRPC_BACKLOG_MAX);
274 conn_head = smp_load_acquire(&b->conn_backlog_head);
275 conn_tail = b->conn_backlog_tail;
276 conn_count = CIRC_CNT(conn_head, conn_tail, RXRPC_BACKLOG_MAX);
277 ASSERTCMP(conn_count, >=, call_count);
278 peer_head = smp_load_acquire(&b->peer_backlog_head);
279 peer_tail = b->peer_backlog_tail;
280 ASSERTCMP(CIRC_CNT(peer_head, peer_tail, RXRPC_BACKLOG_MAX), >=,
287 if (peer && !rxrpc_get_peer_maybe(peer))
290 peer = b->peer_backlog[peer_tail];
291 if (rxrpc_extract_addr_from_skb(local, &peer->srx, skb) < 0)
293 b->peer_backlog[peer_tail] = NULL;
294 smp_store_release(&b->peer_backlog_tail,
296 (RXRPC_BACKLOG_MAX - 1));
298 rxrpc_new_incoming_peer(rx, local, peer);
301 /* Now allocate and set up the connection */
302 conn = b->conn_backlog[conn_tail];
303 b->conn_backlog[conn_tail] = NULL;
304 smp_store_release(&b->conn_backlog_tail,
305 (conn_tail + 1) & (RXRPC_BACKLOG_MAX - 1));
306 conn->params.local = rxrpc_get_local(local);
307 conn->params.peer = peer;
308 rxrpc_see_connection(conn);
309 rxrpc_new_incoming_connection(rx, conn, skb);
311 rxrpc_get_connection(conn);
314 /* And now we can allocate and set up a new call */
315 call = b->call_backlog[call_tail];
316 b->call_backlog[call_tail] = NULL;
317 smp_store_release(&b->call_backlog_tail,
318 (call_tail + 1) & (RXRPC_BACKLOG_MAX - 1));
320 rxrpc_see_call(call);
322 call->peer = rxrpc_get_peer(conn->params.peer);
323 call->cong_cwnd = call->peer->cong_cwnd;
328 * Set up a new incoming call. Called in BH context with the RCU read lock
331 * If this is for a kernel service, when we allocate the call, it will have
332 * three refs on it: (1) the kernel service, (2) the user_call_ID tree, (3) the
333 * retainer ref obtained from the backlog buffer. Prealloc calls for userspace
334 * services only have the ref from the backlog buffer. We want to pass this
335 * ref to non-BH context to dispose of.
337 * If we want to report an error, we mark the skb with the packet type and
338 * abort code and return NULL.
340 * The call is returned with the user access mutex held.
342 struct rxrpc_call *rxrpc_new_incoming_call(struct rxrpc_local *local,
343 struct rxrpc_sock *rx,
346 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
347 struct rxrpc_connection *conn;
348 struct rxrpc_peer *peer = NULL;
349 struct rxrpc_call *call;
353 spin_lock(&rx->incoming_lock);
354 if (rx->sk.sk_state == RXRPC_SERVER_LISTEN_DISABLED ||
355 rx->sk.sk_state == RXRPC_CLOSE) {
356 trace_rxrpc_abort(0, "CLS", sp->hdr.cid, sp->hdr.callNumber,
357 sp->hdr.seq, RX_INVALID_OPERATION, ESHUTDOWN);
358 skb->mark = RXRPC_SKB_MARK_REJECT_ABORT;
359 skb->priority = RX_INVALID_OPERATION;
360 _leave(" = NULL [close]");
365 /* The peer, connection and call may all have sprung into existence due
366 * to a duplicate packet being handled on another CPU in parallel, so
367 * we have to recheck the routing. However, we're now holding
368 * rx->incoming_lock, so the values should remain stable.
370 conn = rxrpc_find_connection_rcu(local, skb, &peer);
372 call = rxrpc_alloc_incoming_call(rx, local, peer, conn, skb);
374 skb->mark = RXRPC_SKB_MARK_REJECT_BUSY;
375 _leave(" = NULL [busy]");
380 trace_rxrpc_receive(call, rxrpc_receive_incoming,
381 sp->hdr.serial, sp->hdr.seq);
383 /* Lock the call to prevent rxrpc_kernel_send/recv_data() and
384 * sendmsg()/recvmsg() inconveniently stealing the mutex once the
385 * notification is generated.
387 * The BUG should never happen because the kernel should be well
388 * behaved enough not to access the call before the first notification
389 * event and userspace is prevented from doing so until the state is
392 if (!mutex_trylock(&call->user_mutex))
395 /* Make the call live. */
396 rxrpc_incoming_call(rx, call, skb);
399 if (rx->notify_new_call)
400 rx->notify_new_call(&rx->sk, call, call->user_call_ID);
402 sk_acceptq_added(&rx->sk);
404 spin_lock(&conn->state_lock);
405 switch (conn->state) {
406 case RXRPC_CONN_SERVICE_UNSECURED:
407 conn->state = RXRPC_CONN_SERVICE_CHALLENGING;
408 set_bit(RXRPC_CONN_EV_CHALLENGE, &call->conn->events);
409 rxrpc_queue_conn(call->conn);
412 case RXRPC_CONN_SERVICE:
413 write_lock(&call->state_lock);
414 if (call->state < RXRPC_CALL_COMPLETE) {
415 if (rx->discard_new_call)
416 call->state = RXRPC_CALL_SERVER_RECV_REQUEST;
418 call->state = RXRPC_CALL_SERVER_ACCEPTING;
420 write_unlock(&call->state_lock);
423 case RXRPC_CONN_REMOTELY_ABORTED:
424 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
425 conn->abort_code, conn->error);
427 case RXRPC_CONN_LOCALLY_ABORTED:
428 rxrpc_abort_call("CON", call, sp->hdr.seq,
429 conn->abort_code, conn->error);
434 spin_unlock(&conn->state_lock);
436 if (call->state == RXRPC_CALL_SERVER_ACCEPTING)
437 rxrpc_notify_socket(call);
439 /* We have to discard the prealloc queue's ref here and rely on a
440 * combination of the RCU read lock and refs held either by the socket
441 * (recvmsg queue, to-be-accepted queue or user ID tree) or the kernel
442 * service to prevent the call from being deallocated too early.
444 rxrpc_put_call(call, rxrpc_call_put);
446 _leave(" = %p{%d}", call, call->debug_id);
448 spin_unlock(&rx->incoming_lock);
453 * handle acceptance of a call by userspace
454 * - assign the user call ID to the call at the front of the queue
455 * - called with the socket locked.
457 struct rxrpc_call *rxrpc_accept_call(struct rxrpc_sock *rx,
458 unsigned long user_call_ID,
459 rxrpc_notify_rx_t notify_rx)
460 __releases(&rx->sk.sk_lock.slock)
461 __acquires(call->user_mutex)
463 struct rxrpc_call *call;
464 struct rb_node *parent, **pp;
467 _enter(",%lx", user_call_ID);
469 ASSERT(!irqs_disabled());
471 write_lock(&rx->call_lock);
473 if (list_empty(&rx->to_be_accepted)) {
474 write_unlock(&rx->call_lock);
475 release_sock(&rx->sk);
476 kleave(" = -ENODATA [empty]");
477 return ERR_PTR(-ENODATA);
480 /* check the user ID isn't already in use */
481 pp = &rx->calls.rb_node;
485 call = rb_entry(parent, struct rxrpc_call, sock_node);
487 if (user_call_ID < call->user_call_ID)
488 pp = &(*pp)->rb_left;
489 else if (user_call_ID > call->user_call_ID)
490 pp = &(*pp)->rb_right;
495 /* Dequeue the first call and check it's still valid. We gain
496 * responsibility for the queue's reference.
498 call = list_entry(rx->to_be_accepted.next,
499 struct rxrpc_call, accept_link);
500 write_unlock(&rx->call_lock);
502 /* We need to gain the mutex from the interrupt handler without
503 * upsetting lockdep, so we have to release it there and take it here.
504 * We are, however, still holding the socket lock, so other accepts
505 * must wait for us and no one can add the user ID behind our backs.
507 if (mutex_lock_interruptible(&call->user_mutex) < 0) {
508 release_sock(&rx->sk);
509 kleave(" = -ERESTARTSYS");
510 return ERR_PTR(-ERESTARTSYS);
513 write_lock(&rx->call_lock);
514 list_del_init(&call->accept_link);
515 sk_acceptq_removed(&rx->sk);
516 rxrpc_see_call(call);
518 /* Find the user ID insertion point. */
519 pp = &rx->calls.rb_node;
523 call = rb_entry(parent, struct rxrpc_call, sock_node);
525 if (user_call_ID < call->user_call_ID)
526 pp = &(*pp)->rb_left;
527 else if (user_call_ID > call->user_call_ID)
528 pp = &(*pp)->rb_right;
533 write_lock_bh(&call->state_lock);
534 switch (call->state) {
535 case RXRPC_CALL_SERVER_ACCEPTING:
536 call->state = RXRPC_CALL_SERVER_RECV_REQUEST;
538 case RXRPC_CALL_COMPLETE:
545 /* formalise the acceptance */
546 call->notify_rx = notify_rx;
547 call->user_call_ID = user_call_ID;
548 rxrpc_get_call(call, rxrpc_call_got_userid);
549 rb_link_node(&call->sock_node, parent, pp);
550 rb_insert_color(&call->sock_node, &rx->calls);
551 if (test_and_set_bit(RXRPC_CALL_HAS_USERID, &call->flags))
554 write_unlock_bh(&call->state_lock);
555 write_unlock(&rx->call_lock);
556 rxrpc_notify_socket(call);
557 rxrpc_service_prealloc(rx, GFP_KERNEL);
558 release_sock(&rx->sk);
559 _leave(" = %p{%d}", call, call->debug_id);
563 _debug("release %p", call);
564 write_unlock_bh(&call->state_lock);
565 write_unlock(&rx->call_lock);
566 rxrpc_release_call(rx, call);
567 rxrpc_put_call(call, rxrpc_call_put);
572 write_unlock(&rx->call_lock);
574 rxrpc_service_prealloc(rx, GFP_KERNEL);
575 release_sock(&rx->sk);
576 _leave(" = %d", ret);
581 * Handle rejection of a call by userspace
582 * - reject the call at the front of the queue
584 int rxrpc_reject_call(struct rxrpc_sock *rx)
586 struct rxrpc_call *call;
592 ASSERT(!irqs_disabled());
594 write_lock(&rx->call_lock);
596 if (list_empty(&rx->to_be_accepted)) {
597 write_unlock(&rx->call_lock);
601 /* Dequeue the first call and check it's still valid. We gain
602 * responsibility for the queue's reference.
604 call = list_entry(rx->to_be_accepted.next,
605 struct rxrpc_call, accept_link);
606 list_del_init(&call->accept_link);
607 sk_acceptq_removed(&rx->sk);
608 rxrpc_see_call(call);
610 write_lock_bh(&call->state_lock);
611 switch (call->state) {
612 case RXRPC_CALL_SERVER_ACCEPTING:
613 __rxrpc_abort_call("REJ", call, 1, RX_USER_ABORT, -ECONNABORTED);
616 case RXRPC_CALL_COMPLETE:
624 write_unlock_bh(&call->state_lock);
625 write_unlock(&rx->call_lock);
627 rxrpc_send_abort_packet(call);
628 rxrpc_release_call(rx, call);
629 rxrpc_put_call(call, rxrpc_call_put);
631 rxrpc_service_prealloc(rx, GFP_KERNEL);
632 _leave(" = %d", ret);
637 * rxrpc_kernel_charge_accept - Charge up socket with preallocated calls
638 * @sock: The socket on which to preallocate
639 * @notify_rx: Event notification function for the call
640 * @user_attach_call: Func to attach call to user_call_ID
641 * @user_call_ID: The tag to attach to the preallocated call
642 * @gfp: The allocation conditions.
643 * @debug_id: The tracing debug ID.
645 * Charge up the socket with preallocated calls, each with a user ID. A
646 * function should be provided to effect the attachment from the user's side.
647 * The user is given a ref to hold on the call.
649 * Note that the call may be come connected before this function returns.
651 int rxrpc_kernel_charge_accept(struct socket *sock,
652 rxrpc_notify_rx_t notify_rx,
653 rxrpc_user_attach_call_t user_attach_call,
654 unsigned long user_call_ID, gfp_t gfp,
655 unsigned int debug_id)
657 struct rxrpc_sock *rx = rxrpc_sk(sock->sk);
658 struct rxrpc_backlog *b = rx->backlog;
660 if (sock->sk->sk_state == RXRPC_CLOSE)
663 return rxrpc_service_prealloc_one(rx, b, notify_rx,
664 user_attach_call, user_call_ID,
667 EXPORT_SYMBOL(rxrpc_kernel_charge_accept);