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_bh(&call->notify_lock);
40 call->notify_rx(sk, call, call->user_call_ID);
41 spin_unlock_bh(&call->notify_lock);
43 write_lock_bh(&rx->recvmsg_lock);
44 if (list_empty(&call->recvmsg_link)) {
45 rxrpc_get_call(call, rxrpc_call_got);
46 list_add_tail(&call->recvmsg_link, &rx->recvmsg_q);
48 write_unlock_bh(&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_bh(&call->state_lock);
91 ret = __rxrpc_set_call_completion(call, compl, abort_code, error);
92 write_unlock_bh(&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_bh(&call->state_lock);
111 ret = __rxrpc_call_completed(call);
112 write_unlock_bh(&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_bh(&call->state_lock);
135 ret = __rxrpc_abort_call(why, call, seq, abort_code, error);
136 write_unlock_bh(&call->state_lock);
141 * Pass a call terminating message to userspace.
143 static int rxrpc_recvmsg_term(struct rxrpc_call *call, struct msghdr *msg)
148 switch (call->completion) {
149 case RXRPC_CALL_SUCCEEDED:
151 if (rxrpc_is_service_call(call))
152 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ACK, 0, &tmp);
154 case RXRPC_CALL_REMOTELY_ABORTED:
155 tmp = call->abort_code;
156 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ABORT, 4, &tmp);
158 case RXRPC_CALL_LOCALLY_ABORTED:
159 tmp = call->abort_code;
160 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ABORT, 4, &tmp);
162 case RXRPC_CALL_NETWORK_ERROR:
164 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_NET_ERROR, 4, &tmp);
166 case RXRPC_CALL_LOCAL_ERROR:
168 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_LOCAL_ERROR, 4, &tmp);
171 pr_err("Invalid terminal call state %u\n", call->state);
176 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_terminal, call->rx_hard_ack,
177 call->rx_pkt_offset, call->rx_pkt_len, ret);
182 * End the packet reception phase.
184 static void rxrpc_end_rx_phase(struct rxrpc_call *call, rxrpc_serial_t serial)
186 _enter("%d,%s", call->debug_id, rxrpc_call_states[call->state]);
188 trace_rxrpc_receive(call, rxrpc_receive_end, 0, call->rx_top);
189 ASSERTCMP(call->rx_hard_ack, ==, call->rx_top);
191 if (call->state == RXRPC_CALL_CLIENT_RECV_REPLY) {
192 rxrpc_propose_ACK(call, RXRPC_ACK_IDLE, serial, false, true,
193 rxrpc_propose_ack_terminal_ack);
194 //rxrpc_send_ack_packet(call, false, NULL);
197 write_lock_bh(&call->state_lock);
199 switch (call->state) {
200 case RXRPC_CALL_CLIENT_RECV_REPLY:
201 __rxrpc_call_completed(call);
202 write_unlock_bh(&call->state_lock);
205 case RXRPC_CALL_SERVER_RECV_REQUEST:
206 call->tx_phase = true;
207 call->state = RXRPC_CALL_SERVER_ACK_REQUEST;
208 call->expect_req_by = jiffies + MAX_JIFFY_OFFSET;
209 write_unlock_bh(&call->state_lock);
210 rxrpc_propose_ACK(call, RXRPC_ACK_DELAY, serial, false, true,
211 rxrpc_propose_ack_processing_op);
214 write_unlock_bh(&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 hard_ack, top;
232 _enter("%d", call->debug_id);
234 hard_ack = call->rx_hard_ack;
235 top = smp_load_acquire(&call->rx_top);
236 ASSERT(before(hard_ack, top));
239 ix = hard_ack & RXRPC_RXTX_BUFF_MASK;
240 skb = call->rxtx_buffer[ix];
241 rxrpc_see_skb(skb, rxrpc_skb_rotated);
244 subpacket = call->rxtx_annotations[ix] & RXRPC_RX_ANNO_SUBPACKET;
245 serial = sp->hdr.serial + subpacket;
247 if (subpacket == sp->nr_subpackets - 1 &&
248 sp->rx_flags & RXRPC_SKB_INCL_LAST)
251 call->rxtx_buffer[ix] = NULL;
252 call->rxtx_annotations[ix] = 0;
253 /* Barrier against rxrpc_input_data(). */
254 smp_store_release(&call->rx_hard_ack, hard_ack);
256 rxrpc_free_skb(skb, rxrpc_skb_freed);
258 trace_rxrpc_receive(call, rxrpc_receive_rotate, serial, hard_ack);
260 rxrpc_end_rx_phase(call, serial);
262 /* Check to see if there's an ACK that needs sending. */
263 if (atomic_inc_return(&call->ackr_nr_consumed) > 2)
264 rxrpc_propose_ACK(call, RXRPC_ACK_IDLE, serial,
266 rxrpc_propose_ack_rotate_rx);
267 if (call->ackr_reason && call->ackr_reason != RXRPC_ACK_DELAY)
268 rxrpc_send_ack_packet(call, false, NULL);
273 * Decrypt and verify a (sub)packet. The packet's length may be changed due to
274 * padding, but if this is the case, the packet length will be resident in the
275 * socket buffer. Note that we can't modify the master skb info as the skb may
276 * be the home to multiple subpackets.
278 static int rxrpc_verify_packet(struct rxrpc_call *call, struct sk_buff *skb,
280 unsigned int offset, unsigned int len)
282 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
283 rxrpc_seq_t seq = sp->hdr.seq;
284 u16 cksum = sp->hdr.cksum;
285 u8 subpacket = annotation & RXRPC_RX_ANNO_SUBPACKET;
289 /* For all but the head jumbo subpacket, the security checksum is in a
290 * jumbo header immediately prior to the data.
294 if (skb_copy_bits(skb, offset - 2, &tmp, 2) < 0)
300 return call->security->verify_packet(call, skb, offset, len,
305 * Locate the data within a packet. This is complicated by:
307 * (1) An skb may contain a jumbo packet - so we have to find the appropriate
310 * (2) The (sub)packets may be encrypted and, if so, the encrypted portion
311 * contains an extra header which includes the true length of the data,
312 * excluding any encrypted padding.
314 static int rxrpc_locate_data(struct rxrpc_call *call, struct sk_buff *skb,
316 unsigned int *_offset, unsigned int *_len,
319 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
320 unsigned int offset = sizeof(struct rxrpc_wire_header);
324 u8 annotation = *_annotation;
325 u8 subpacket = annotation & RXRPC_RX_ANNO_SUBPACKET;
327 /* Locate the subpacket */
328 offset += subpacket * RXRPC_JUMBO_SUBPKTLEN;
329 len = skb->len - offset;
330 if (subpacket < sp->nr_subpackets - 1)
331 len = RXRPC_JUMBO_DATALEN;
332 else if (sp->rx_flags & RXRPC_SKB_INCL_LAST)
335 if (!(annotation & RXRPC_RX_ANNO_VERIFIED)) {
336 ret = rxrpc_verify_packet(call, skb, annotation, offset, len);
339 *_annotation |= RXRPC_RX_ANNO_VERIFIED;
345 call->security->locate_data(call, skb, _offset, _len);
350 * Deliver messages to a call. This keeps processing packets until the buffer
351 * is filled and we find either more DATA (returns 0) or the end of the DATA
352 * (returns 1). If more packets are required, it returns -EAGAIN.
354 static int rxrpc_recvmsg_data(struct socket *sock, struct rxrpc_call *call,
355 struct msghdr *msg, struct iov_iter *iter,
356 size_t len, int flags, size_t *_offset)
358 struct rxrpc_skb_priv *sp;
360 rxrpc_serial_t serial;
361 rxrpc_seq_t hard_ack, top, seq;
364 unsigned int rx_pkt_offset, rx_pkt_len;
365 int ix, copy, ret = -EAGAIN, ret2;
367 if (test_and_clear_bit(RXRPC_CALL_RX_UNDERRUN, &call->flags) &&
369 rxrpc_send_ack_packet(call, false, NULL);
371 rx_pkt_offset = call->rx_pkt_offset;
372 rx_pkt_len = call->rx_pkt_len;
373 rx_pkt_last = call->rx_pkt_last;
375 if (call->state >= RXRPC_CALL_SERVER_ACK_REQUEST) {
376 seq = call->rx_hard_ack;
381 /* Barriers against rxrpc_input_data(). */
382 hard_ack = call->rx_hard_ack;
385 while (top = smp_load_acquire(&call->rx_top),
388 ix = seq & RXRPC_RXTX_BUFF_MASK;
389 skb = call->rxtx_buffer[ix];
391 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_hole, seq,
392 rx_pkt_offset, rx_pkt_len, 0);
396 rxrpc_see_skb(skb, rxrpc_skb_seen);
399 if (!(flags & MSG_PEEK)) {
400 serial = sp->hdr.serial;
401 serial += call->rxtx_annotations[ix] & RXRPC_RX_ANNO_SUBPACKET;
402 trace_rxrpc_receive(call, rxrpc_receive_front,
407 sock_recv_timestamp(msg, sock->sk, skb);
409 if (rx_pkt_offset == 0) {
410 ret2 = rxrpc_locate_data(call, skb,
411 &call->rxtx_annotations[ix],
412 &rx_pkt_offset, &rx_pkt_len,
414 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_next, seq,
415 rx_pkt_offset, rx_pkt_len, ret2);
421 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_cont, seq,
422 rx_pkt_offset, rx_pkt_len, 0);
425 /* We have to handle short, empty and used-up DATA packets. */
426 remain = len - *_offset;
431 ret2 = skb_copy_datagram_iter(skb, rx_pkt_offset, iter,
438 /* handle piecemeal consumption of data packets */
439 rx_pkt_offset += copy;
444 if (rx_pkt_len > 0) {
445 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_full, seq,
446 rx_pkt_offset, rx_pkt_len, 0);
447 ASSERTCMP(*_offset, ==, len);
452 /* The whole packet has been transferred. */
453 if (!(flags & MSG_PEEK))
454 rxrpc_rotate_rx_window(call);
459 ASSERTCMP(seq, ==, READ_ONCE(call->rx_top));
468 if (!(flags & MSG_PEEK)) {
469 call->rx_pkt_offset = rx_pkt_offset;
470 call->rx_pkt_len = rx_pkt_len;
471 call->rx_pkt_last = rx_pkt_last;
474 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_data_return, seq,
475 rx_pkt_offset, rx_pkt_len, ret);
477 set_bit(RXRPC_CALL_RX_UNDERRUN, &call->flags);
482 * Receive a message from an RxRPC socket
483 * - we need to be careful about two or more threads calling recvmsg
486 int rxrpc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
489 struct rxrpc_call *call;
490 struct rxrpc_sock *rx = rxrpc_sk(sock->sk);
498 trace_rxrpc_recvmsg(NULL, rxrpc_recvmsg_enter, 0, 0, 0, 0);
500 if (flags & (MSG_OOB | MSG_TRUNC))
503 timeo = sock_rcvtimeo(&rx->sk, flags & MSG_DONTWAIT);
508 /* Return immediately if a client socket has no outstanding calls */
509 if (RB_EMPTY_ROOT(&rx->calls) &&
510 list_empty(&rx->recvmsg_q) &&
511 rx->sk.sk_state != RXRPC_SERVER_LISTENING) {
512 release_sock(&rx->sk);
516 if (list_empty(&rx->recvmsg_q)) {
523 release_sock(&rx->sk);
525 /* Wait for something to happen */
526 prepare_to_wait_exclusive(sk_sleep(&rx->sk), &wait,
528 ret = sock_error(&rx->sk);
532 if (list_empty(&rx->recvmsg_q)) {
533 if (signal_pending(current))
534 goto wait_interrupted;
535 trace_rxrpc_recvmsg(NULL, rxrpc_recvmsg_wait,
537 timeo = schedule_timeout(timeo);
539 finish_wait(sk_sleep(&rx->sk), &wait);
543 /* Find the next call and dequeue it if we're not just peeking. If we
544 * do dequeue it, that comes with a ref that we will need to release.
546 write_lock_bh(&rx->recvmsg_lock);
547 l = rx->recvmsg_q.next;
548 call = list_entry(l, struct rxrpc_call, recvmsg_link);
549 if (!(flags & MSG_PEEK))
550 list_del_init(&call->recvmsg_link);
552 rxrpc_get_call(call, rxrpc_call_got);
553 write_unlock_bh(&rx->recvmsg_lock);
555 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_dequeue, 0, 0, 0, 0);
557 /* We're going to drop the socket lock, so we need to lock the call
558 * against interference by sendmsg.
560 if (!mutex_trylock(&call->user_mutex)) {
562 if (flags & MSG_DONTWAIT)
563 goto error_requeue_call;
565 if (mutex_lock_interruptible(&call->user_mutex) < 0)
566 goto error_requeue_call;
569 release_sock(&rx->sk);
571 if (test_bit(RXRPC_CALL_RELEASED, &call->flags))
574 if (test_bit(RXRPC_CALL_HAS_USERID, &call->flags)) {
575 if (flags & MSG_CMSG_COMPAT) {
576 unsigned int id32 = call->user_call_ID;
578 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_USER_CALL_ID,
579 sizeof(unsigned int), &id32);
581 unsigned long idl = call->user_call_ID;
583 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_USER_CALL_ID,
584 sizeof(unsigned long), &idl);
587 goto error_unlock_call;
590 if (msg->msg_name && call->peer) {
591 struct sockaddr_rxrpc *srx = msg->msg_name;
592 size_t len = sizeof(call->peer->srx);
594 memcpy(msg->msg_name, &call->peer->srx, len);
595 srx->srx_service = call->service_id;
596 msg->msg_namelen = len;
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, msg, &msg->msg_iter, len,
608 if (after(call->rx_top, call->rx_hard_ack) &&
609 call->rxtx_buffer[(call->rx_hard_ack + 1) & RXRPC_RXTX_BUFF_MASK])
610 rxrpc_notify_socket(call);
618 goto error_unlock_call;
620 if (call->state == RXRPC_CALL_COMPLETE) {
621 ret = rxrpc_recvmsg_term(call, msg);
623 goto error_unlock_call;
624 if (!(flags & MSG_PEEK))
625 rxrpc_release_call(rx, call);
626 msg->msg_flags |= MSG_EOR;
631 msg->msg_flags |= MSG_MORE;
633 msg->msg_flags &= ~MSG_MORE;
637 mutex_unlock(&call->user_mutex);
638 rxrpc_put_call(call, rxrpc_call_put);
639 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_return, 0, 0, 0, ret);
643 if (!(flags & MSG_PEEK)) {
644 write_lock_bh(&rx->recvmsg_lock);
645 list_add(&call->recvmsg_link, &rx->recvmsg_q);
646 write_unlock_bh(&rx->recvmsg_lock);
647 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_requeue, 0, 0, 0, 0);
649 rxrpc_put_call(call, rxrpc_call_put);
652 release_sock(&rx->sk);
654 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_return, 0, 0, 0, ret);
658 ret = sock_intr_errno(timeo);
660 finish_wait(sk_sleep(&rx->sk), &wait);
666 * rxrpc_kernel_recv_data - Allow a kernel service to receive data/info
667 * @sock: The socket that the call exists on
668 * @call: The call to send data through
669 * @iter: The buffer to receive into
670 * @_len: The amount of data we want to receive (decreased on return)
671 * @want_more: True if more data is expected to be read
672 * @_abort: Where the abort code is stored if -ECONNABORTED is returned
673 * @_service: Where to store the actual service ID (may be upgraded)
675 * Allow a kernel service to receive data and pick up information about the
676 * state of a call. Returns 0 if got what was asked for and there's more
677 * available, 1 if we got what was asked for and we're at the end of the data
678 * and -EAGAIN if we need more data.
680 * Note that we may return -EAGAIN to drain empty packets at the end of the
681 * data, even if we've already copied over the requested data.
683 * *_abort should also be initialised to 0.
685 int rxrpc_kernel_recv_data(struct socket *sock, struct rxrpc_call *call,
686 struct iov_iter *iter, size_t *_len,
687 bool want_more, u32 *_abort, u16 *_service)
692 _enter("{%d,%s},%zu,%d",
693 call->debug_id, rxrpc_call_states[call->state],
696 ASSERTCMP(call->state, !=, RXRPC_CALL_SERVER_SECURING);
698 mutex_lock(&call->user_mutex);
700 switch (READ_ONCE(call->state)) {
701 case RXRPC_CALL_CLIENT_RECV_REPLY:
702 case RXRPC_CALL_SERVER_RECV_REQUEST:
703 case RXRPC_CALL_SERVER_ACK_REQUEST:
704 ret = rxrpc_recvmsg_data(sock, call, NULL, iter,
710 /* We can only reach here with a partially full buffer if we
711 * have reached the end of the data. We must otherwise have a
712 * full buffer or have been given -EAGAIN.
715 if (iov_iter_count(iter) > 0)
718 goto read_phase_complete;
727 case RXRPC_CALL_COMPLETE:
738 switch (call->ackr_reason) {
741 case RXRPC_ACK_DELAY:
746 rxrpc_send_ack_packet(call, false, NULL);
750 *_service = call->service_id;
751 mutex_unlock(&call->user_mutex);
752 _leave(" = %d [%zu,%d]", ret, iov_iter_count(iter), *_abort);
756 trace_rxrpc_rx_eproto(call, 0, tracepoint_string("short_data"));
760 trace_rxrpc_rx_eproto(call, 0, tracepoint_string("excess_data"));
764 *_abort = call->abort_code;
766 if (call->completion == RXRPC_CALL_SUCCEEDED) {
768 if (iov_iter_count(iter) > 0)
773 EXPORT_SYMBOL(rxrpc_kernel_recv_data);