Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* incoming call handling
3 *
4 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
6 */
7
8#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9
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>
15#include <linux/in.h>
16#include <linux/in6.h>
17#include <linux/icmp.h>
18#include <linux/gfp.h>
19#include <linux/circ_buf.h>
20#include <net/sock.h>
21#include <net/af_rxrpc.h>
22#include <net/ip.h>
23#include "ar-internal.h"
24
25/*
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.
28 */
29static 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)
35{
36 const void *here = __builtin_return_address(0);
37 struct rxrpc_call *call;
38 struct rxrpc_net *rxnet = rxrpc_net(sock_net(&rx->sk));
39 int max, tmp;
40 unsigned int size = RXRPC_BACKLOG_MAX;
41 unsigned int head, tail, call_head, call_tail;
42
43 max = rx->sk.sk_max_ack_backlog;
44 tmp = rx->sk.sk_ack_backlog;
45 if (tmp >= max) {
46 _leave(" = -ENOBUFS [full %u]", max);
47 return -ENOBUFS;
48 }
49 max -= tmp;
50
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
53 * than calls.
54 */
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);
58 if (tmp >= max) {
59 _leave(" = -ENOBUFS [enough %u]", tmp);
60 return -ENOBUFS;
61 }
62 max = tmp + 1;
63
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);
68 if (!peer)
69 return -ENOMEM;
70 b->peer_backlog[head] = peer;
71 smp_store_release(&b->peer_backlog_head,
72 (head + 1) & (size - 1));
73 }
74
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;
79
80 conn = rxrpc_prealloc_service_connection(rxnet, gfp);
81 if (!conn)
82 return -ENOMEM;
83 b->conn_backlog[head] = conn;
84 smp_store_release(&b->conn_backlog_head,
85 (head + 1) & (size - 1));
86
87 trace_rxrpc_conn(conn->debug_id, rxrpc_conn_new_service,
88 atomic_read(&conn->usage), here);
89 }
90
91 /* Now it gets complicated, because calls get registered with the
92 * socket here, particularly if a user ID is preassigned by the user.
93 */
94 call = rxrpc_alloc_call(rx, gfp, debug_id);
95 if (!call)
96 return -ENOMEM;
97 call->flags |= (1 << RXRPC_CALL_IS_SERVICE);
98 call->state = RXRPC_CALL_SERVER_PREALLOC;
99
100 trace_rxrpc_call(call->debug_id, rxrpc_call_new_service,
101 atomic_read(&call->usage),
102 here, (const void *)user_call_ID);
103
104 write_lock(&rx->call_lock);
105 if (user_attach_call) {
106 struct rxrpc_call *xcall;
107 struct rb_node *parent, **pp;
108
109 /* Check the user ID isn't already in use */
110 pp = &rx->calls.rb_node;
111 parent = NULL;
112 while (*pp) {
113 parent = *pp;
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;
119 else
120 goto id_in_use;
121 }
122
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);
131 }
132
133 list_add(&call->sock_link, &rx->sock_calls);
134
135 write_unlock(&rx->call_lock);
136
137 rxnet = call->rxnet;
138 write_lock(&rxnet->call_lock);
139 list_add_tail(&call->link, &rxnet->calls);
140 write_unlock(&rxnet->call_lock);
141
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);
145 return 0;
146
147id_in_use:
148 write_unlock(&rx->call_lock);
149 rxrpc_cleanup_call(call);
150 _leave(" = -EBADSLT");
151 return -EBADSLT;
152}
153
154/*
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.
158 *
159 * The backlog is replenished when a connection is accepted or rejected.
160 */
161int rxrpc_service_prealloc(struct rxrpc_sock *rx, gfp_t gfp)
162{
163 struct rxrpc_backlog *b = rx->backlog;
164
165 if (!b) {
166 b = kzalloc(sizeof(struct rxrpc_backlog), gfp);
167 if (!b)
168 return -ENOMEM;
169 rx->backlog = b;
170 }
171
172 if (rx->discard_new_call)
173 return 0;
174
175 while (rxrpc_service_prealloc_one(rx, b, NULL, NULL, 0, gfp,
176 atomic_inc_return(&rxrpc_debug_id)) == 0)
177 ;
178
179 return 0;
180}
181
182/*
183 * Discard the preallocation on a service.
184 */
185void rxrpc_discard_prealloc(struct rxrpc_sock *rx)
186{
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;
190
191 if (!b)
192 return;
193 rx->backlog = NULL;
194
195 /* Make sure that there aren't any incoming calls in progress before we
196 * clear the preallocation buffers.
197 */
198 spin_lock_bh(&rx->incoming_lock);
199 spin_unlock_bh(&rx->incoming_lock);
200
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];
205 kfree(peer);
206 tail = (tail + 1) & (size - 1);
207 }
208
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);
217 kfree(conn);
218 if (atomic_dec_and_test(&rxnet->nr_conns))
219 wake_up_var(&rxnet->nr_conns);
220 tail = (tail + 1) & (size - 1);
221 }
222
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);
232 }
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);
237 }
238
239 kfree(b);
240}
241
242/*
243 * Allocate a new incoming call from the prealloc pool, along with a connection
244 * and a peer as necessary.
245 */
246static 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,
250 struct sk_buff *skb)
251{
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;
257
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), >=,
269 conn_count);
270
271 if (call_count == 0)
272 return NULL;
273
274 if (!conn) {
275 if (peer && !rxrpc_get_peer_maybe(peer))
276 peer = NULL;
277 if (!peer) {
278 peer = b->peer_backlog[peer_tail];
279 if (rxrpc_extract_addr_from_skb(&peer->srx, skb) < 0)
280 return NULL;
281 b->peer_backlog[peer_tail] = NULL;
282 smp_store_release(&b->peer_backlog_tail,
283 (peer_tail + 1) &
284 (RXRPC_BACKLOG_MAX - 1));
285
286 rxrpc_new_incoming_peer(rx, local, peer);
287 }
288
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);
298 } else {
299 rxrpc_get_connection(conn);
300 }
301
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));
307
308 rxrpc_see_call(call);
309 call->conn = conn;
310 call->security = conn->security;
311 call->peer = rxrpc_get_peer(conn->params.peer);
312 call->cong_cwnd = call->peer->cong_cwnd;
313 return call;
314}
315
316/*
317 * Set up a new incoming call. Called in BH context with the RCU read lock
318 * held.
319 *
320 * If this is for a kernel service, when we allocate the call, it will have
321 * three refs on it: (1) the kernel service, (2) the user_call_ID tree, (3) the
322 * retainer ref obtained from the backlog buffer. Prealloc calls for userspace
323 * services only have the ref from the backlog buffer. We want to pass this
324 * ref to non-BH context to dispose of.
325 *
326 * If we want to report an error, we mark the skb with the packet type and
327 * abort code and return NULL.
328 *
329 * The call is returned with the user access mutex held.
330 */
331struct rxrpc_call *rxrpc_new_incoming_call(struct rxrpc_local *local,
332 struct rxrpc_sock *rx,
333 struct sk_buff *skb)
334{
335 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
336 struct rxrpc_connection *conn;
337 struct rxrpc_peer *peer = NULL;
338 struct rxrpc_call *call;
339
340 _enter("");
341
342 spin_lock(&rx->incoming_lock);
343 if (rx->sk.sk_state == RXRPC_SERVER_LISTEN_DISABLED ||
344 rx->sk.sk_state == RXRPC_CLOSE) {
345 trace_rxrpc_abort(0, "CLS", sp->hdr.cid, sp->hdr.callNumber,
346 sp->hdr.seq, RX_INVALID_OPERATION, ESHUTDOWN);
347 skb->mark = RXRPC_SKB_MARK_REJECT_ABORT;
348 skb->priority = RX_INVALID_OPERATION;
349 _leave(" = NULL [close]");
350 call = NULL;
351 goto out;
352 }
353
354 /* The peer, connection and call may all have sprung into existence due
355 * to a duplicate packet being handled on another CPU in parallel, so
356 * we have to recheck the routing. However, we're now holding
357 * rx->incoming_lock, so the values should remain stable.
358 */
359 conn = rxrpc_find_connection_rcu(local, skb, &peer);
360
361 call = rxrpc_alloc_incoming_call(rx, local, peer, conn, skb);
362 if (!call) {
363 skb->mark = RXRPC_SKB_MARK_REJECT_BUSY;
364 _leave(" = NULL [busy]");
365 call = NULL;
366 goto out;
367 }
368
369 trace_rxrpc_receive(call, rxrpc_receive_incoming,
370 sp->hdr.serial, sp->hdr.seq);
371
372 /* Lock the call to prevent rxrpc_kernel_send/recv_data() and
373 * sendmsg()/recvmsg() inconveniently stealing the mutex once the
374 * notification is generated.
375 *
376 * The BUG should never happen because the kernel should be well
377 * behaved enough not to access the call before the first notification
378 * event and userspace is prevented from doing so until the state is
379 * appropriate.
380 */
381 if (!mutex_trylock(&call->user_mutex))
382 BUG();
383
384 /* Make the call live. */
385 rxrpc_incoming_call(rx, call, skb);
386 conn = call->conn;
387
388 if (rx->notify_new_call)
389 rx->notify_new_call(&rx->sk, call, call->user_call_ID);
390 else
391 sk_acceptq_added(&rx->sk);
392
393 spin_lock(&conn->state_lock);
394 switch (conn->state) {
395 case RXRPC_CONN_SERVICE_UNSECURED:
396 conn->state = RXRPC_CONN_SERVICE_CHALLENGING;
397 set_bit(RXRPC_CONN_EV_CHALLENGE, &call->conn->events);
398 rxrpc_queue_conn(call->conn);
399 break;
400
401 case RXRPC_CONN_SERVICE:
402 write_lock(&call->state_lock);
403 if (call->state < RXRPC_CALL_COMPLETE) {
404 if (rx->discard_new_call)
405 call->state = RXRPC_CALL_SERVER_RECV_REQUEST;
406 else
407 call->state = RXRPC_CALL_SERVER_ACCEPTING;
408 }
409 write_unlock(&call->state_lock);
410 break;
411
412 case RXRPC_CONN_REMOTELY_ABORTED:
413 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
414 conn->abort_code, conn->error);
415 break;
416 case RXRPC_CONN_LOCALLY_ABORTED:
417 rxrpc_abort_call("CON", call, sp->hdr.seq,
418 conn->abort_code, conn->error);
419 break;
420 default:
421 BUG();
422 }
423 spin_unlock(&conn->state_lock);
424
425 if (call->state == RXRPC_CALL_SERVER_ACCEPTING)
426 rxrpc_notify_socket(call);
427
428 /* We have to discard the prealloc queue's ref here and rely on a
429 * combination of the RCU read lock and refs held either by the socket
430 * (recvmsg queue, to-be-accepted queue or user ID tree) or the kernel
431 * service to prevent the call from being deallocated too early.
432 */
433 rxrpc_put_call(call, rxrpc_call_put);
434
435 _leave(" = %p{%d}", call, call->debug_id);
436out:
437 spin_unlock(&rx->incoming_lock);
438 return call;
439}
440
441/*
442 * handle acceptance of a call by userspace
443 * - assign the user call ID to the call at the front of the queue
444 * - called with the socket locked.
445 */
446struct rxrpc_call *rxrpc_accept_call(struct rxrpc_sock *rx,
447 unsigned long user_call_ID,
448 rxrpc_notify_rx_t notify_rx)
449 __releases(&rx->sk.sk_lock.slock)
450 __acquires(call->user_mutex)
451{
452 struct rxrpc_call *call;
453 struct rb_node *parent, **pp;
454 int ret;
455
456 _enter(",%lx", user_call_ID);
457
458 ASSERT(!irqs_disabled());
459
460 write_lock(&rx->call_lock);
461
462 if (list_empty(&rx->to_be_accepted)) {
463 write_unlock(&rx->call_lock);
464 release_sock(&rx->sk);
465 kleave(" = -ENODATA [empty]");
466 return ERR_PTR(-ENODATA);
467 }
468
469 /* check the user ID isn't already in use */
470 pp = &rx->calls.rb_node;
471 parent = NULL;
472 while (*pp) {
473 parent = *pp;
474 call = rb_entry(parent, struct rxrpc_call, sock_node);
475
476 if (user_call_ID < call->user_call_ID)
477 pp = &(*pp)->rb_left;
478 else if (user_call_ID > call->user_call_ID)
479 pp = &(*pp)->rb_right;
480 else
481 goto id_in_use;
482 }
483
484 /* Dequeue the first call and check it's still valid. We gain
485 * responsibility for the queue's reference.
486 */
487 call = list_entry(rx->to_be_accepted.next,
488 struct rxrpc_call, accept_link);
489 write_unlock(&rx->call_lock);
490
491 /* We need to gain the mutex from the interrupt handler without
492 * upsetting lockdep, so we have to release it there and take it here.
493 * We are, however, still holding the socket lock, so other accepts
494 * must wait for us and no one can add the user ID behind our backs.
495 */
496 if (mutex_lock_interruptible(&call->user_mutex) < 0) {
497 release_sock(&rx->sk);
498 kleave(" = -ERESTARTSYS");
499 return ERR_PTR(-ERESTARTSYS);
500 }
501
502 write_lock(&rx->call_lock);
503 list_del_init(&call->accept_link);
504 sk_acceptq_removed(&rx->sk);
505 rxrpc_see_call(call);
506
507 /* Find the user ID insertion point. */
508 pp = &rx->calls.rb_node;
509 parent = NULL;
510 while (*pp) {
511 parent = *pp;
512 call = rb_entry(parent, struct rxrpc_call, sock_node);
513
514 if (user_call_ID < call->user_call_ID)
515 pp = &(*pp)->rb_left;
516 else if (user_call_ID > call->user_call_ID)
517 pp = &(*pp)->rb_right;
518 else
519 BUG();
520 }
521
522 write_lock_bh(&call->state_lock);
523 switch (call->state) {
524 case RXRPC_CALL_SERVER_ACCEPTING:
525 call->state = RXRPC_CALL_SERVER_RECV_REQUEST;
526 break;
527 case RXRPC_CALL_COMPLETE:
528 ret = call->error;
529 goto out_release;
530 default:
531 BUG();
532 }
533
534 /* formalise the acceptance */
535 call->notify_rx = notify_rx;
536 call->user_call_ID = user_call_ID;
537 rxrpc_get_call(call, rxrpc_call_got_userid);
538 rb_link_node(&call->sock_node, parent, pp);
539 rb_insert_color(&call->sock_node, &rx->calls);
540 if (test_and_set_bit(RXRPC_CALL_HAS_USERID, &call->flags))
541 BUG();
542
543 write_unlock_bh(&call->state_lock);
544 write_unlock(&rx->call_lock);
545 rxrpc_notify_socket(call);
546 rxrpc_service_prealloc(rx, GFP_KERNEL);
547 release_sock(&rx->sk);
548 _leave(" = %p{%d}", call, call->debug_id);
549 return call;
550
551out_release:
552 _debug("release %p", call);
553 write_unlock_bh(&call->state_lock);
554 write_unlock(&rx->call_lock);
555 rxrpc_release_call(rx, call);
556 rxrpc_put_call(call, rxrpc_call_put);
557 goto out;
558
559id_in_use:
560 ret = -EBADSLT;
561 write_unlock(&rx->call_lock);
562out:
563 rxrpc_service_prealloc(rx, GFP_KERNEL);
564 release_sock(&rx->sk);
565 _leave(" = %d", ret);
566 return ERR_PTR(ret);
567}
568
569/*
570 * Handle rejection of a call by userspace
571 * - reject the call at the front of the queue
572 */
573int rxrpc_reject_call(struct rxrpc_sock *rx)
574{
575 struct rxrpc_call *call;
576 bool abort = false;
577 int ret;
578
579 _enter("");
580
581 ASSERT(!irqs_disabled());
582
583 write_lock(&rx->call_lock);
584
585 if (list_empty(&rx->to_be_accepted)) {
586 write_unlock(&rx->call_lock);
587 return -ENODATA;
588 }
589
590 /* Dequeue the first call and check it's still valid. We gain
591 * responsibility for the queue's reference.
592 */
593 call = list_entry(rx->to_be_accepted.next,
594 struct rxrpc_call, accept_link);
595 list_del_init(&call->accept_link);
596 sk_acceptq_removed(&rx->sk);
597 rxrpc_see_call(call);
598
599 write_lock_bh(&call->state_lock);
600 switch (call->state) {
601 case RXRPC_CALL_SERVER_ACCEPTING:
602 __rxrpc_abort_call("REJ", call, 1, RX_USER_ABORT, -ECONNABORTED);
603 abort = true;
604 /* fall through */
605 case RXRPC_CALL_COMPLETE:
606 ret = call->error;
607 goto out_discard;
608 default:
609 BUG();
610 }
611
612out_discard:
613 write_unlock_bh(&call->state_lock);
614 write_unlock(&rx->call_lock);
615 if (abort) {
616 rxrpc_send_abort_packet(call);
617 rxrpc_release_call(rx, call);
618 rxrpc_put_call(call, rxrpc_call_put);
619 }
620 rxrpc_service_prealloc(rx, GFP_KERNEL);
621 _leave(" = %d", ret);
622 return ret;
623}
624
625/*
626 * rxrpc_kernel_charge_accept - Charge up socket with preallocated calls
627 * @sock: The socket on which to preallocate
628 * @notify_rx: Event notification function for the call
629 * @user_attach_call: Func to attach call to user_call_ID
630 * @user_call_ID: The tag to attach to the preallocated call
631 * @gfp: The allocation conditions.
632 * @debug_id: The tracing debug ID.
633 *
634 * Charge up the socket with preallocated calls, each with a user ID. A
635 * function should be provided to effect the attachment from the user's side.
636 * The user is given a ref to hold on the call.
637 *
638 * Note that the call may be come connected before this function returns.
639 */
640int rxrpc_kernel_charge_accept(struct socket *sock,
641 rxrpc_notify_rx_t notify_rx,
642 rxrpc_user_attach_call_t user_attach_call,
643 unsigned long user_call_ID, gfp_t gfp,
644 unsigned int debug_id)
645{
646 struct rxrpc_sock *rx = rxrpc_sk(sock->sk);
647 struct rxrpc_backlog *b = rx->backlog;
648
649 if (sock->sk->sk_state == RXRPC_CLOSE)
650 return -ESHUTDOWN;
651
652 return rxrpc_service_prealloc_one(rx, b, notify_rx,
653 user_attach_call, user_call_ID,
654 gfp, debug_id);
655}
656EXPORT_SYMBOL(rxrpc_kernel_charge_accept);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/* incoming call handling
3 *
4 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
6 */
7
8#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9
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>
15#include <linux/in.h>
16#include <linux/in6.h>
17#include <linux/icmp.h>
18#include <linux/gfp.h>
19#include <linux/circ_buf.h>
20#include <net/sock.h>
21#include <net/af_rxrpc.h>
22#include <net/ip.h>
23#include "ar-internal.h"
24
25static void rxrpc_dummy_notify(struct sock *sk, struct rxrpc_call *call,
26 unsigned long user_call_ID)
27{
28}
29
30/*
31 * Preallocate a single service call, connection and peer and, if possible,
32 * give them a user ID and attach the user's side of the ID to them.
33 */
34static int rxrpc_service_prealloc_one(struct rxrpc_sock *rx,
35 struct rxrpc_backlog *b,
36 rxrpc_notify_rx_t notify_rx,
37 rxrpc_user_attach_call_t user_attach_call,
38 unsigned long user_call_ID, gfp_t gfp,
39 unsigned int debug_id)
40{
41 const void *here = __builtin_return_address(0);
42 struct rxrpc_call *call, *xcall;
43 struct rxrpc_net *rxnet = rxrpc_net(sock_net(&rx->sk));
44 struct rb_node *parent, **pp;
45 int max, tmp;
46 unsigned int size = RXRPC_BACKLOG_MAX;
47 unsigned int head, tail, call_head, call_tail;
48
49 max = rx->sk.sk_max_ack_backlog;
50 tmp = rx->sk.sk_ack_backlog;
51 if (tmp >= max) {
52 _leave(" = -ENOBUFS [full %u]", max);
53 return -ENOBUFS;
54 }
55 max -= tmp;
56
57 /* We don't need more conns and peers than we have calls, but on the
58 * other hand, we shouldn't ever use more peers than conns or conns
59 * than calls.
60 */
61 call_head = b->call_backlog_head;
62 call_tail = READ_ONCE(b->call_backlog_tail);
63 tmp = CIRC_CNT(call_head, call_tail, size);
64 if (tmp >= max) {
65 _leave(" = -ENOBUFS [enough %u]", tmp);
66 return -ENOBUFS;
67 }
68 max = tmp + 1;
69
70 head = b->peer_backlog_head;
71 tail = READ_ONCE(b->peer_backlog_tail);
72 if (CIRC_CNT(head, tail, size) < max) {
73 struct rxrpc_peer *peer = rxrpc_alloc_peer(rx->local, gfp);
74 if (!peer)
75 return -ENOMEM;
76 b->peer_backlog[head] = peer;
77 smp_store_release(&b->peer_backlog_head,
78 (head + 1) & (size - 1));
79 }
80
81 head = b->conn_backlog_head;
82 tail = READ_ONCE(b->conn_backlog_tail);
83 if (CIRC_CNT(head, tail, size) < max) {
84 struct rxrpc_connection *conn;
85
86 conn = rxrpc_prealloc_service_connection(rxnet, gfp);
87 if (!conn)
88 return -ENOMEM;
89 b->conn_backlog[head] = conn;
90 smp_store_release(&b->conn_backlog_head,
91 (head + 1) & (size - 1));
92
93 trace_rxrpc_conn(conn->debug_id, rxrpc_conn_new_service,
94 atomic_read(&conn->usage), here);
95 }
96
97 /* Now it gets complicated, because calls get registered with the
98 * socket here, with a user ID preassigned by the user.
99 */
100 call = rxrpc_alloc_call(rx, gfp, debug_id);
101 if (!call)
102 return -ENOMEM;
103 call->flags |= (1 << RXRPC_CALL_IS_SERVICE);
104 call->state = RXRPC_CALL_SERVER_PREALLOC;
105
106 trace_rxrpc_call(call->debug_id, rxrpc_call_new_service,
107 atomic_read(&call->usage),
108 here, (const void *)user_call_ID);
109
110 write_lock(&rx->call_lock);
111
112 /* Check the user ID isn't already in use */
113 pp = &rx->calls.rb_node;
114 parent = NULL;
115 while (*pp) {
116 parent = *pp;
117 xcall = rb_entry(parent, struct rxrpc_call, sock_node);
118 if (user_call_ID < xcall->user_call_ID)
119 pp = &(*pp)->rb_left;
120 else if (user_call_ID > xcall->user_call_ID)
121 pp = &(*pp)->rb_right;
122 else
123 goto id_in_use;
124 }
125
126 call->user_call_ID = user_call_ID;
127 call->notify_rx = notify_rx;
128 if (user_attach_call) {
129 rxrpc_get_call(call, rxrpc_call_got_kernel);
130 user_attach_call(call, user_call_ID);
131 }
132
133 rxrpc_get_call(call, rxrpc_call_got_userid);
134 rb_link_node(&call->sock_node, parent, pp);
135 rb_insert_color(&call->sock_node, &rx->calls);
136 set_bit(RXRPC_CALL_HAS_USERID, &call->flags);
137
138 list_add(&call->sock_link, &rx->sock_calls);
139
140 write_unlock(&rx->call_lock);
141
142 rxnet = call->rxnet;
143 write_lock(&rxnet->call_lock);
144 list_add_tail(&call->link, &rxnet->calls);
145 write_unlock(&rxnet->call_lock);
146
147 b->call_backlog[call_head] = call;
148 smp_store_release(&b->call_backlog_head, (call_head + 1) & (size - 1));
149 _leave(" = 0 [%d -> %lx]", call->debug_id, user_call_ID);
150 return 0;
151
152id_in_use:
153 write_unlock(&rx->call_lock);
154 rxrpc_cleanup_call(call);
155 _leave(" = -EBADSLT");
156 return -EBADSLT;
157}
158
159/*
160 * Allocate the preallocation buffers for incoming service calls. These must
161 * be charged manually.
162 */
163int rxrpc_service_prealloc(struct rxrpc_sock *rx, gfp_t gfp)
164{
165 struct rxrpc_backlog *b = rx->backlog;
166
167 if (!b) {
168 b = kzalloc(sizeof(struct rxrpc_backlog), gfp);
169 if (!b)
170 return -ENOMEM;
171 rx->backlog = b;
172 }
173
174 return 0;
175}
176
177/*
178 * Discard the preallocation on a service.
179 */
180void rxrpc_discard_prealloc(struct rxrpc_sock *rx)
181{
182 struct rxrpc_backlog *b = rx->backlog;
183 struct rxrpc_net *rxnet = rxrpc_net(sock_net(&rx->sk));
184 unsigned int size = RXRPC_BACKLOG_MAX, head, tail;
185
186 if (!b)
187 return;
188 rx->backlog = NULL;
189
190 /* Make sure that there aren't any incoming calls in progress before we
191 * clear the preallocation buffers.
192 */
193 spin_lock_bh(&rx->incoming_lock);
194 spin_unlock_bh(&rx->incoming_lock);
195
196 head = b->peer_backlog_head;
197 tail = b->peer_backlog_tail;
198 while (CIRC_CNT(head, tail, size) > 0) {
199 struct rxrpc_peer *peer = b->peer_backlog[tail];
200 kfree(peer);
201 tail = (tail + 1) & (size - 1);
202 }
203
204 head = b->conn_backlog_head;
205 tail = b->conn_backlog_tail;
206 while (CIRC_CNT(head, tail, size) > 0) {
207 struct rxrpc_connection *conn = b->conn_backlog[tail];
208 write_lock(&rxnet->conn_lock);
209 list_del(&conn->link);
210 list_del(&conn->proc_link);
211 write_unlock(&rxnet->conn_lock);
212 kfree(conn);
213 if (atomic_dec_and_test(&rxnet->nr_conns))
214 wake_up_var(&rxnet->nr_conns);
215 tail = (tail + 1) & (size - 1);
216 }
217
218 head = b->call_backlog_head;
219 tail = b->call_backlog_tail;
220 while (CIRC_CNT(head, tail, size) > 0) {
221 struct rxrpc_call *call = b->call_backlog[tail];
222 rcu_assign_pointer(call->socket, rx);
223 if (rx->discard_new_call) {
224 _debug("discard %lx", call->user_call_ID);
225 rx->discard_new_call(call, call->user_call_ID);
226 if (call->notify_rx)
227 call->notify_rx = rxrpc_dummy_notify;
228 rxrpc_put_call(call, rxrpc_call_put_kernel);
229 }
230 rxrpc_call_completed(call);
231 rxrpc_release_call(rx, call);
232 rxrpc_put_call(call, rxrpc_call_put);
233 tail = (tail + 1) & (size - 1);
234 }
235
236 kfree(b);
237}
238
239/*
240 * Ping the other end to fill our RTT cache and to retrieve the rwind
241 * and MTU parameters.
242 */
243static void rxrpc_send_ping(struct rxrpc_call *call, struct sk_buff *skb)
244{
245 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
246 ktime_t now = skb->tstamp;
247
248 if (call->peer->rtt_count < 3 ||
249 ktime_before(ktime_add_ms(call->peer->rtt_last_req, 1000), now))
250 rxrpc_propose_ACK(call, RXRPC_ACK_PING, sp->hdr.serial,
251 true, true,
252 rxrpc_propose_ack_ping_for_params);
253}
254
255/*
256 * Allocate a new incoming call from the prealloc pool, along with a connection
257 * and a peer as necessary.
258 */
259static struct rxrpc_call *rxrpc_alloc_incoming_call(struct rxrpc_sock *rx,
260 struct rxrpc_local *local,
261 struct rxrpc_peer *peer,
262 struct rxrpc_connection *conn,
263 const struct rxrpc_security *sec,
264 struct key *key,
265 struct sk_buff *skb)
266{
267 struct rxrpc_backlog *b = rx->backlog;
268 struct rxrpc_call *call;
269 unsigned short call_head, conn_head, peer_head;
270 unsigned short call_tail, conn_tail, peer_tail;
271 unsigned short call_count, conn_count;
272
273 /* #calls >= #conns >= #peers must hold true. */
274 call_head = smp_load_acquire(&b->call_backlog_head);
275 call_tail = b->call_backlog_tail;
276 call_count = CIRC_CNT(call_head, call_tail, RXRPC_BACKLOG_MAX);
277 conn_head = smp_load_acquire(&b->conn_backlog_head);
278 conn_tail = b->conn_backlog_tail;
279 conn_count = CIRC_CNT(conn_head, conn_tail, RXRPC_BACKLOG_MAX);
280 ASSERTCMP(conn_count, >=, call_count);
281 peer_head = smp_load_acquire(&b->peer_backlog_head);
282 peer_tail = b->peer_backlog_tail;
283 ASSERTCMP(CIRC_CNT(peer_head, peer_tail, RXRPC_BACKLOG_MAX), >=,
284 conn_count);
285
286 if (call_count == 0)
287 return NULL;
288
289 if (!conn) {
290 if (peer && !rxrpc_get_peer_maybe(peer))
291 peer = NULL;
292 if (!peer) {
293 peer = b->peer_backlog[peer_tail];
294 if (rxrpc_extract_addr_from_skb(&peer->srx, skb) < 0)
295 return NULL;
296 b->peer_backlog[peer_tail] = NULL;
297 smp_store_release(&b->peer_backlog_tail,
298 (peer_tail + 1) &
299 (RXRPC_BACKLOG_MAX - 1));
300
301 rxrpc_new_incoming_peer(rx, local, peer);
302 }
303
304 /* Now allocate and set up the connection */
305 conn = b->conn_backlog[conn_tail];
306 b->conn_backlog[conn_tail] = NULL;
307 smp_store_release(&b->conn_backlog_tail,
308 (conn_tail + 1) & (RXRPC_BACKLOG_MAX - 1));
309 conn->params.local = rxrpc_get_local(local);
310 conn->params.peer = peer;
311 rxrpc_see_connection(conn);
312 rxrpc_new_incoming_connection(rx, conn, sec, key, skb);
313 } else {
314 rxrpc_get_connection(conn);
315 }
316
317 /* And now we can allocate and set up a new call */
318 call = b->call_backlog[call_tail];
319 b->call_backlog[call_tail] = NULL;
320 smp_store_release(&b->call_backlog_tail,
321 (call_tail + 1) & (RXRPC_BACKLOG_MAX - 1));
322
323 rxrpc_see_call(call);
324 call->conn = conn;
325 call->security = conn->security;
326 call->security_ix = conn->security_ix;
327 call->peer = rxrpc_get_peer(conn->params.peer);
328 call->cong_cwnd = call->peer->cong_cwnd;
329 return call;
330}
331
332/*
333 * Set up a new incoming call. Called in BH context with the RCU read lock
334 * held.
335 *
336 * If this is for a kernel service, when we allocate the call, it will have
337 * three refs on it: (1) the kernel service, (2) the user_call_ID tree, (3) the
338 * retainer ref obtained from the backlog buffer. Prealloc calls for userspace
339 * services only have the ref from the backlog buffer. We want to pass this
340 * ref to non-BH context to dispose of.
341 *
342 * If we want to report an error, we mark the skb with the packet type and
343 * abort code and return NULL.
344 *
345 * The call is returned with the user access mutex held.
346 */
347struct rxrpc_call *rxrpc_new_incoming_call(struct rxrpc_local *local,
348 struct rxrpc_sock *rx,
349 struct sk_buff *skb)
350{
351 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
352 const struct rxrpc_security *sec = NULL;
353 struct rxrpc_connection *conn;
354 struct rxrpc_peer *peer = NULL;
355 struct rxrpc_call *call = NULL;
356 struct key *key = NULL;
357
358 _enter("");
359
360 spin_lock(&rx->incoming_lock);
361 if (rx->sk.sk_state == RXRPC_SERVER_LISTEN_DISABLED ||
362 rx->sk.sk_state == RXRPC_CLOSE) {
363 trace_rxrpc_abort(0, "CLS", sp->hdr.cid, sp->hdr.callNumber,
364 sp->hdr.seq, RX_INVALID_OPERATION, ESHUTDOWN);
365 skb->mark = RXRPC_SKB_MARK_REJECT_ABORT;
366 skb->priority = RX_INVALID_OPERATION;
367 goto no_call;
368 }
369
370 /* The peer, connection and call may all have sprung into existence due
371 * to a duplicate packet being handled on another CPU in parallel, so
372 * we have to recheck the routing. However, we're now holding
373 * rx->incoming_lock, so the values should remain stable.
374 */
375 conn = rxrpc_find_connection_rcu(local, skb, &peer);
376
377 if (!conn && !rxrpc_look_up_server_security(local, rx, &sec, &key, skb))
378 goto no_call;
379
380 call = rxrpc_alloc_incoming_call(rx, local, peer, conn, sec, key, skb);
381 key_put(key);
382 if (!call) {
383 skb->mark = RXRPC_SKB_MARK_REJECT_BUSY;
384 goto no_call;
385 }
386
387 trace_rxrpc_receive(call, rxrpc_receive_incoming,
388 sp->hdr.serial, sp->hdr.seq);
389
390 /* Make the call live. */
391 rxrpc_incoming_call(rx, call, skb);
392 conn = call->conn;
393
394 if (rx->notify_new_call)
395 rx->notify_new_call(&rx->sk, call, call->user_call_ID);
396
397 spin_lock(&conn->state_lock);
398 switch (conn->state) {
399 case RXRPC_CONN_SERVICE_UNSECURED:
400 conn->state = RXRPC_CONN_SERVICE_CHALLENGING;
401 set_bit(RXRPC_CONN_EV_CHALLENGE, &call->conn->events);
402 rxrpc_queue_conn(call->conn);
403 break;
404
405 case RXRPC_CONN_SERVICE:
406 write_lock(&call->state_lock);
407 if (call->state < RXRPC_CALL_COMPLETE)
408 call->state = RXRPC_CALL_SERVER_RECV_REQUEST;
409 write_unlock(&call->state_lock);
410 break;
411
412 case RXRPC_CONN_REMOTELY_ABORTED:
413 rxrpc_set_call_completion(call, RXRPC_CALL_REMOTELY_ABORTED,
414 conn->abort_code, conn->error);
415 break;
416 case RXRPC_CONN_LOCALLY_ABORTED:
417 rxrpc_abort_call("CON", call, sp->hdr.seq,
418 conn->abort_code, conn->error);
419 break;
420 default:
421 BUG();
422 }
423 spin_unlock(&conn->state_lock);
424 spin_unlock(&rx->incoming_lock);
425
426 rxrpc_send_ping(call, skb);
427
428 /* We have to discard the prealloc queue's ref here and rely on a
429 * combination of the RCU read lock and refs held either by the socket
430 * (recvmsg queue, to-be-accepted queue or user ID tree) or the kernel
431 * service to prevent the call from being deallocated too early.
432 */
433 rxrpc_put_call(call, rxrpc_call_put);
434
435 _leave(" = %p{%d}", call, call->debug_id);
436 return call;
437
438no_call:
439 spin_unlock(&rx->incoming_lock);
440 _leave(" = NULL [%u]", skb->mark);
441 return NULL;
442}
443
444/*
445 * Charge up socket with preallocated calls, attaching user call IDs.
446 */
447int rxrpc_user_charge_accept(struct rxrpc_sock *rx, unsigned long user_call_ID)
448{
449 struct rxrpc_backlog *b = rx->backlog;
450
451 if (rx->sk.sk_state == RXRPC_CLOSE)
452 return -ESHUTDOWN;
453
454 return rxrpc_service_prealloc_one(rx, b, NULL, NULL, user_call_ID,
455 GFP_KERNEL,
456 atomic_inc_return(&rxrpc_debug_id));
457}
458
459/*
460 * rxrpc_kernel_charge_accept - Charge up socket with preallocated calls
461 * @sock: The socket on which to preallocate
462 * @notify_rx: Event notification function for the call
463 * @user_attach_call: Func to attach call to user_call_ID
464 * @user_call_ID: The tag to attach to the preallocated call
465 * @gfp: The allocation conditions.
466 * @debug_id: The tracing debug ID.
467 *
468 * Charge up the socket with preallocated calls, each with a user ID. A
469 * function should be provided to effect the attachment from the user's side.
470 * The user is given a ref to hold on the call.
471 *
472 * Note that the call may be come connected before this function returns.
473 */
474int rxrpc_kernel_charge_accept(struct socket *sock,
475 rxrpc_notify_rx_t notify_rx,
476 rxrpc_user_attach_call_t user_attach_call,
477 unsigned long user_call_ID, gfp_t gfp,
478 unsigned int debug_id)
479{
480 struct rxrpc_sock *rx = rxrpc_sk(sock->sk);
481 struct rxrpc_backlog *b = rx->backlog;
482
483 if (sock->sk->sk_state == RXRPC_CLOSE)
484 return -ESHUTDOWN;
485
486 return rxrpc_service_prealloc_one(rx, b, notify_rx,
487 user_attach_call, user_call_ID,
488 gfp, debug_id);
489}
490EXPORT_SYMBOL(rxrpc_kernel_charge_accept);