1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* RxRPC recvmsg() implementation
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/net.h>
11 #include <linux/skbuff.h>
12 #include <linux/export.h>
13 #include <linux/sched/signal.h>
16 #include <net/af_rxrpc.h>
17 #include "ar-internal.h"
20 * Post a call for attention by the socket or kernel service. Further
21 * notifications are suppressed by putting recvmsg_link on a dummy queue.
23 void rxrpc_notify_socket(struct rxrpc_call *call)
25 struct rxrpc_sock *rx;
28 _enter("%d", call->debug_id);
30 if (!list_empty(&call->recvmsg_link))
35 rx = rcu_dereference(call->socket);
37 if (rx && sk->sk_state < RXRPC_CLOSE) {
38 if (call->notify_rx) {
39 spin_lock(&call->notify_lock);
40 call->notify_rx(sk, call, call->user_call_ID);
41 spin_unlock(&call->notify_lock);
43 write_lock(&rx->recvmsg_lock);
44 if (list_empty(&call->recvmsg_link)) {
45 rxrpc_get_call(call, rxrpc_call_get_notify_socket);
46 list_add_tail(&call->recvmsg_link, &rx->recvmsg_q);
48 write_unlock(&rx->recvmsg_lock);
50 if (!sock_flag(sk, SOCK_DEAD)) {
51 _debug("call %ps", sk->sk_data_ready);
52 sk->sk_data_ready(sk);
62 * Transition a call to the complete state.
64 bool __rxrpc_set_call_completion(struct rxrpc_call *call,
65 enum rxrpc_call_completion compl,
69 if (call->state < RXRPC_CALL_COMPLETE) {
70 call->abort_code = abort_code;
72 call->completion = compl;
73 call->state = RXRPC_CALL_COMPLETE;
74 trace_rxrpc_call_complete(call);
75 wake_up(&call->waitq);
76 rxrpc_notify_socket(call);
82 bool rxrpc_set_call_completion(struct rxrpc_call *call,
83 enum rxrpc_call_completion compl,
89 if (call->state < RXRPC_CALL_COMPLETE) {
90 write_lock(&call->state_lock);
91 ret = __rxrpc_set_call_completion(call, compl, abort_code, error);
92 write_unlock(&call->state_lock);
98 * Record that a call successfully completed.
100 bool __rxrpc_call_completed(struct rxrpc_call *call)
102 return __rxrpc_set_call_completion(call, RXRPC_CALL_SUCCEEDED, 0, 0);
105 bool rxrpc_call_completed(struct rxrpc_call *call)
109 if (call->state < RXRPC_CALL_COMPLETE) {
110 write_lock(&call->state_lock);
111 ret = __rxrpc_call_completed(call);
112 write_unlock(&call->state_lock);
118 * Record that a call is locally aborted.
120 bool __rxrpc_abort_call(const char *why, struct rxrpc_call *call,
121 rxrpc_seq_t seq, u32 abort_code, int error)
123 trace_rxrpc_abort(call->debug_id, why, call->cid, call->call_id, seq,
125 return __rxrpc_set_call_completion(call, RXRPC_CALL_LOCALLY_ABORTED,
129 bool rxrpc_abort_call(const char *why, struct rxrpc_call *call,
130 rxrpc_seq_t seq, u32 abort_code, int error)
134 write_lock(&call->state_lock);
135 ret = __rxrpc_abort_call(why, call, seq, abort_code, error);
136 write_unlock(&call->state_lock);
137 if (ret && test_bit(RXRPC_CALL_EXPOSED, &call->flags))
138 rxrpc_send_abort_packet(call);
143 * Pass a call terminating message to userspace.
145 static int rxrpc_recvmsg_term(struct rxrpc_call *call, struct msghdr *msg)
150 switch (call->completion) {
151 case RXRPC_CALL_SUCCEEDED:
153 if (rxrpc_is_service_call(call))
154 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ACK, 0, &tmp);
156 case RXRPC_CALL_REMOTELY_ABORTED:
157 tmp = call->abort_code;
158 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ABORT, 4, &tmp);
160 case RXRPC_CALL_LOCALLY_ABORTED:
161 tmp = call->abort_code;
162 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ABORT, 4, &tmp);
164 case RXRPC_CALL_NETWORK_ERROR:
166 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_NET_ERROR, 4, &tmp);
168 case RXRPC_CALL_LOCAL_ERROR:
170 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_LOCAL_ERROR, 4, &tmp);
173 pr_err("Invalid terminal call state %u\n", call->state);
178 trace_rxrpc_recvdata(call, rxrpc_recvmsg_terminal,
179 lower_32_bits(atomic64_read(&call->ackr_window)) - 1,
180 call->rx_pkt_offset, call->rx_pkt_len, ret);
185 * End the packet reception phase.
187 static void rxrpc_end_rx_phase(struct rxrpc_call *call, rxrpc_serial_t serial)
189 rxrpc_seq_t whigh = READ_ONCE(call->rx_highest_seq);
191 _enter("%d,%s", call->debug_id, rxrpc_call_states[call->state]);
193 trace_rxrpc_receive(call, rxrpc_receive_end, 0, whigh);
195 if (call->state == RXRPC_CALL_CLIENT_RECV_REPLY)
196 rxrpc_propose_delay_ACK(call, serial, rxrpc_propose_ack_terminal_ack);
198 write_lock(&call->state_lock);
200 switch (call->state) {
201 case RXRPC_CALL_CLIENT_RECV_REPLY:
202 __rxrpc_call_completed(call);
203 write_unlock(&call->state_lock);
206 case RXRPC_CALL_SERVER_RECV_REQUEST:
207 call->state = RXRPC_CALL_SERVER_ACK_REQUEST;
208 call->expect_req_by = jiffies + MAX_JIFFY_OFFSET;
209 write_unlock(&call->state_lock);
210 rxrpc_propose_delay_ACK(call, serial,
211 rxrpc_propose_ack_processing_op);
214 write_unlock(&call->state_lock);
220 * Discard a packet we've used up and advance the Rx window by one.
222 static void rxrpc_rotate_rx_window(struct rxrpc_call *call)
224 struct rxrpc_skb_priv *sp;
226 rxrpc_serial_t serial;
227 rxrpc_seq_t old_consumed = call->rx_consumed, tseq;
231 _enter("%d", call->debug_id);
233 skb = skb_dequeue(&call->recvmsg_queue);
234 rxrpc_see_skb(skb, rxrpc_skb_see_rotate);
238 serial = sp->hdr.serial;
239 last = sp->hdr.flags & RXRPC_LAST_PACKET;
241 /* Barrier against rxrpc_input_data(). */
242 if (after(tseq, call->rx_consumed))
243 smp_store_release(&call->rx_consumed, tseq);
245 rxrpc_free_skb(skb, rxrpc_skb_put_rotate);
247 trace_rxrpc_receive(call, last ? rxrpc_receive_rotate_last : rxrpc_receive_rotate,
248 serial, call->rx_consumed);
250 rxrpc_end_rx_phase(call, serial);
254 /* Check to see if there's an ACK that needs sending. */
255 acked = atomic_add_return(call->rx_consumed - old_consumed,
256 &call->ackr_nr_consumed);
258 !test_and_set_bit(RXRPC_CALL_RX_IS_IDLE, &call->flags))
259 rxrpc_poke_call(call, rxrpc_call_poke_idle);
263 * Decrypt and verify a DATA packet.
265 static int rxrpc_verify_data(struct rxrpc_call *call, struct sk_buff *skb)
267 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
269 if (sp->flags & RXRPC_RX_VERIFIED)
271 return call->security->verify_packet(call, skb);
275 * Deliver messages to a call. This keeps processing packets until the buffer
276 * is filled and we find either more DATA (returns 0) or the end of the DATA
277 * (returns 1). If more packets are required, it returns -EAGAIN.
279 static int rxrpc_recvmsg_data(struct socket *sock, struct rxrpc_call *call,
280 struct msghdr *msg, struct iov_iter *iter,
281 size_t len, int flags, size_t *_offset)
283 struct rxrpc_skb_priv *sp;
287 unsigned int rx_pkt_offset, rx_pkt_len;
288 int copy, ret = -EAGAIN, ret2;
290 rx_pkt_offset = call->rx_pkt_offset;
291 rx_pkt_len = call->rx_pkt_len;
293 if (call->state >= RXRPC_CALL_SERVER_ACK_REQUEST) {
294 seq = lower_32_bits(atomic64_read(&call->ackr_window)) - 1;
299 /* No one else can be removing stuff from the queue, so we shouldn't
300 * need the Rx lock to walk it.
302 skb = skb_peek(&call->recvmsg_queue);
304 rxrpc_see_skb(skb, rxrpc_skb_see_recvmsg);
308 if (!(flags & MSG_PEEK))
309 trace_rxrpc_receive(call, rxrpc_receive_front,
310 sp->hdr.serial, seq);
313 sock_recv_timestamp(msg, sock->sk, skb);
315 if (rx_pkt_offset == 0) {
316 ret2 = rxrpc_verify_data(call, skb);
317 rx_pkt_offset = sp->offset;
318 rx_pkt_len = sp->len;
319 trace_rxrpc_recvdata(call, rxrpc_recvmsg_next, seq,
320 rx_pkt_offset, rx_pkt_len, ret2);
326 trace_rxrpc_recvdata(call, rxrpc_recvmsg_cont, seq,
327 rx_pkt_offset, rx_pkt_len, 0);
330 /* We have to handle short, empty and used-up DATA packets. */
331 remain = len - *_offset;
336 ret2 = skb_copy_datagram_iter(skb, rx_pkt_offset, iter,
343 /* handle piecemeal consumption of data packets */
344 rx_pkt_offset += copy;
349 if (rx_pkt_len > 0) {
350 trace_rxrpc_recvdata(call, rxrpc_recvmsg_full, seq,
351 rx_pkt_offset, rx_pkt_len, 0);
352 ASSERTCMP(*_offset, ==, len);
357 /* The whole packet has been transferred. */
358 if (sp->hdr.flags & RXRPC_LAST_PACKET)
363 skb = skb_peek_next(skb, &call->recvmsg_queue);
365 if (!(flags & MSG_PEEK))
366 rxrpc_rotate_rx_window(call);
370 if (!(flags & MSG_PEEK)) {
371 call->rx_pkt_offset = rx_pkt_offset;
372 call->rx_pkt_len = rx_pkt_len;
375 trace_rxrpc_recvdata(call, rxrpc_recvmsg_data_return, seq,
376 rx_pkt_offset, rx_pkt_len, ret);
378 set_bit(RXRPC_CALL_RX_IS_IDLE, &call->flags);
383 * Receive a message from an RxRPC socket
384 * - we need to be careful about two or more threads calling recvmsg
387 int rxrpc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
390 struct rxrpc_call *call;
391 struct rxrpc_sock *rx = rxrpc_sk(sock->sk);
393 unsigned int call_debug_id = 0;
400 trace_rxrpc_recvmsg(0, rxrpc_recvmsg_enter, 0);
402 if (flags & (MSG_OOB | MSG_TRUNC))
405 timeo = sock_rcvtimeo(&rx->sk, flags & MSG_DONTWAIT);
410 /* Return immediately if a client socket has no outstanding calls */
411 if (RB_EMPTY_ROOT(&rx->calls) &&
412 list_empty(&rx->recvmsg_q) &&
413 rx->sk.sk_state != RXRPC_SERVER_LISTENING) {
414 release_sock(&rx->sk);
418 if (list_empty(&rx->recvmsg_q)) {
425 release_sock(&rx->sk);
427 /* Wait for something to happen */
428 prepare_to_wait_exclusive(sk_sleep(&rx->sk), &wait,
430 ret = sock_error(&rx->sk);
434 if (list_empty(&rx->recvmsg_q)) {
435 if (signal_pending(current))
436 goto wait_interrupted;
437 trace_rxrpc_recvmsg(0, rxrpc_recvmsg_wait, 0);
438 timeo = schedule_timeout(timeo);
440 finish_wait(sk_sleep(&rx->sk), &wait);
444 /* Find the next call and dequeue it if we're not just peeking. If we
445 * do dequeue it, that comes with a ref that we will need to release.
447 write_lock(&rx->recvmsg_lock);
448 l = rx->recvmsg_q.next;
449 call = list_entry(l, struct rxrpc_call, recvmsg_link);
450 if (!(flags & MSG_PEEK))
451 list_del_init(&call->recvmsg_link);
453 rxrpc_get_call(call, rxrpc_call_get_recvmsg);
454 write_unlock(&rx->recvmsg_lock);
456 call_debug_id = call->debug_id;
457 trace_rxrpc_recvmsg(call_debug_id, rxrpc_recvmsg_dequeue, 0);
459 /* We're going to drop the socket lock, so we need to lock the call
460 * against interference by sendmsg.
462 if (!mutex_trylock(&call->user_mutex)) {
464 if (flags & MSG_DONTWAIT)
465 goto error_requeue_call;
467 if (mutex_lock_interruptible(&call->user_mutex) < 0)
468 goto error_requeue_call;
471 release_sock(&rx->sk);
473 if (test_bit(RXRPC_CALL_RELEASED, &call->flags))
476 if (test_bit(RXRPC_CALL_HAS_USERID, &call->flags)) {
477 if (flags & MSG_CMSG_COMPAT) {
478 unsigned int id32 = call->user_call_ID;
480 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_USER_CALL_ID,
481 sizeof(unsigned int), &id32);
483 unsigned long idl = call->user_call_ID;
485 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_USER_CALL_ID,
486 sizeof(unsigned long), &idl);
489 goto error_unlock_call;
492 if (msg->msg_name && call->peer) {
493 size_t len = sizeof(call->dest_srx);
495 memcpy(msg->msg_name, &call->dest_srx, len);
496 msg->msg_namelen = len;
499 switch (READ_ONCE(call->state)) {
500 case RXRPC_CALL_CLIENT_RECV_REPLY:
501 case RXRPC_CALL_SERVER_RECV_REQUEST:
502 case RXRPC_CALL_SERVER_ACK_REQUEST:
503 ret = rxrpc_recvmsg_data(sock, call, msg, &msg->msg_iter, len,
508 if (!skb_queue_empty(&call->recvmsg_queue))
509 rxrpc_notify_socket(call);
517 goto error_unlock_call;
519 if (call->state == RXRPC_CALL_COMPLETE) {
520 ret = rxrpc_recvmsg_term(call, msg);
522 goto error_unlock_call;
523 if (!(flags & MSG_PEEK))
524 rxrpc_release_call(rx, call);
525 msg->msg_flags |= MSG_EOR;
530 msg->msg_flags |= MSG_MORE;
532 msg->msg_flags &= ~MSG_MORE;
536 mutex_unlock(&call->user_mutex);
537 rxrpc_put_call(call, rxrpc_call_put_recvmsg);
538 trace_rxrpc_recvmsg(call_debug_id, rxrpc_recvmsg_return, ret);
542 if (!(flags & MSG_PEEK)) {
543 write_lock(&rx->recvmsg_lock);
544 list_add(&call->recvmsg_link, &rx->recvmsg_q);
545 write_unlock(&rx->recvmsg_lock);
546 trace_rxrpc_recvmsg(call_debug_id, rxrpc_recvmsg_requeue, 0);
548 rxrpc_put_call(call, rxrpc_call_put_recvmsg);
551 release_sock(&rx->sk);
553 trace_rxrpc_recvmsg(call_debug_id, rxrpc_recvmsg_return, ret);
557 ret = sock_intr_errno(timeo);
559 finish_wait(sk_sleep(&rx->sk), &wait);
565 * rxrpc_kernel_recv_data - Allow a kernel service to receive data/info
566 * @sock: The socket that the call exists on
567 * @call: The call to send data through
568 * @iter: The buffer to receive into
569 * @_len: The amount of data we want to receive (decreased on return)
570 * @want_more: True if more data is expected to be read
571 * @_abort: Where the abort code is stored if -ECONNABORTED is returned
572 * @_service: Where to store the actual service ID (may be upgraded)
574 * Allow a kernel service to receive data and pick up information about the
575 * state of a call. Returns 0 if got what was asked for and there's more
576 * available, 1 if we got what was asked for and we're at the end of the data
577 * and -EAGAIN if we need more data.
579 * Note that we may return -EAGAIN to drain empty packets at the end of the
580 * data, even if we've already copied over the requested data.
582 * *_abort should also be initialised to 0.
584 int rxrpc_kernel_recv_data(struct socket *sock, struct rxrpc_call *call,
585 struct iov_iter *iter, size_t *_len,
586 bool want_more, u32 *_abort, u16 *_service)
591 _enter("{%d,%s},%zu,%d",
592 call->debug_id, rxrpc_call_states[call->state],
595 ASSERTCMP(call->state, !=, RXRPC_CALL_SERVER_SECURING);
597 mutex_lock(&call->user_mutex);
599 switch (READ_ONCE(call->state)) {
600 case RXRPC_CALL_CLIENT_RECV_REPLY:
601 case RXRPC_CALL_SERVER_RECV_REQUEST:
602 case RXRPC_CALL_SERVER_ACK_REQUEST:
603 ret = rxrpc_recvmsg_data(sock, call, NULL, iter,
609 /* We can only reach here with a partially full buffer if we
610 * have reached the end of the data. We must otherwise have a
611 * full buffer or have been given -EAGAIN.
614 if (iov_iter_count(iter) > 0)
617 goto read_phase_complete;
626 case RXRPC_CALL_COMPLETE:
638 *_service = call->dest_srx.srx_service;
639 mutex_unlock(&call->user_mutex);
640 _leave(" = %d [%zu,%d]", ret, iov_iter_count(iter), *_abort);
644 trace_rxrpc_rx_eproto(call, 0, tracepoint_string("short_data"));
648 trace_rxrpc_rx_eproto(call, 0, tracepoint_string("excess_data"));
652 *_abort = call->abort_code;
654 if (call->completion == RXRPC_CALL_SUCCEEDED) {
656 if (iov_iter_count(iter) > 0)
661 EXPORT_SYMBOL(rxrpc_kernel_recv_data);