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 * Pass a call terminating message to userspace.
64 static int rxrpc_recvmsg_term(struct rxrpc_call *call, struct msghdr *msg)
69 switch (call->completion) {
70 case RXRPC_CALL_SUCCEEDED:
72 if (rxrpc_is_service_call(call))
73 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ACK, 0, &tmp);
75 case RXRPC_CALL_REMOTELY_ABORTED:
76 tmp = call->abort_code;
77 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ABORT, 4, &tmp);
79 case RXRPC_CALL_LOCALLY_ABORTED:
80 tmp = call->abort_code;
81 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ABORT, 4, &tmp);
83 case RXRPC_CALL_NETWORK_ERROR:
85 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_NET_ERROR, 4, &tmp);
87 case RXRPC_CALL_LOCAL_ERROR:
89 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_LOCAL_ERROR, 4, &tmp);
92 pr_err("Invalid terminal call state %u\n", call->state);
97 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_terminal, call->rx_hard_ack,
98 call->rx_pkt_offset, call->rx_pkt_len, ret);
103 * Pass back notification of a new call. The call is added to the
104 * to-be-accepted list. This means that the next call to be accepted might not
105 * be the last call seen awaiting acceptance, but unless we leave this on the
106 * front of the queue and block all other messages until someone gives us a
107 * user_ID for it, there's not a lot we can do.
109 static int rxrpc_recvmsg_new_call(struct rxrpc_sock *rx,
110 struct rxrpc_call *call,
111 struct msghdr *msg, int flags)
115 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_NEW_CALL, 0, &tmp);
117 if (ret == 0 && !(flags & MSG_PEEK)) {
118 _debug("to be accepted");
119 write_lock_bh(&rx->recvmsg_lock);
120 list_del_init(&call->recvmsg_link);
121 write_unlock_bh(&rx->recvmsg_lock);
123 rxrpc_get_call(call, rxrpc_call_got);
124 write_lock(&rx->call_lock);
125 list_add_tail(&call->accept_link, &rx->to_be_accepted);
126 write_unlock(&rx->call_lock);
129 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_to_be_accepted, 1, 0, 0, ret);
134 * End the packet reception phase.
136 static void rxrpc_end_rx_phase(struct rxrpc_call *call, rxrpc_serial_t serial)
138 _enter("%d,%s", call->debug_id, rxrpc_call_states[call->state]);
140 trace_rxrpc_receive(call, rxrpc_receive_end, 0, call->rx_top);
141 ASSERTCMP(call->rx_hard_ack, ==, call->rx_top);
143 if (call->state == RXRPC_CALL_CLIENT_RECV_REPLY) {
144 rxrpc_propose_ACK(call, RXRPC_ACK_IDLE, 0, serial, false, true,
145 rxrpc_propose_ack_terminal_ack);
146 //rxrpc_send_ack_packet(call, false, NULL);
149 write_lock_bh(&call->state_lock);
151 switch (call->state) {
152 case RXRPC_CALL_CLIENT_RECV_REPLY:
153 __rxrpc_call_completed(call);
154 write_unlock_bh(&call->state_lock);
157 case RXRPC_CALL_SERVER_RECV_REQUEST:
158 call->tx_phase = true;
159 call->state = RXRPC_CALL_SERVER_ACK_REQUEST;
160 call->expect_req_by = jiffies + MAX_JIFFY_OFFSET;
161 write_unlock_bh(&call->state_lock);
162 rxrpc_propose_ACK(call, RXRPC_ACK_DELAY, 0, serial, false, true,
163 rxrpc_propose_ack_processing_op);
166 write_unlock_bh(&call->state_lock);
172 * Discard a packet we've used up and advance the Rx window by one.
174 static void rxrpc_rotate_rx_window(struct rxrpc_call *call)
176 struct rxrpc_skb_priv *sp;
178 rxrpc_serial_t serial;
179 rxrpc_seq_t hard_ack, top;
183 _enter("%d", call->debug_id);
185 hard_ack = call->rx_hard_ack;
186 top = smp_load_acquire(&call->rx_top);
187 ASSERT(before(hard_ack, top));
190 ix = hard_ack & RXRPC_RXTX_BUFF_MASK;
191 skb = call->rxtx_buffer[ix];
192 rxrpc_see_skb(skb, rxrpc_skb_rx_rotated);
194 flags = sp->hdr.flags;
195 serial = sp->hdr.serial;
196 if (call->rxtx_annotations[ix] & RXRPC_RX_ANNO_JUMBO)
197 serial += (call->rxtx_annotations[ix] & RXRPC_RX_ANNO_JUMBO) - 1;
199 call->rxtx_buffer[ix] = NULL;
200 call->rxtx_annotations[ix] = 0;
201 /* Barrier against rxrpc_input_data(). */
202 smp_store_release(&call->rx_hard_ack, hard_ack);
204 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
206 _debug("%u,%u,%02x", hard_ack, top, flags);
207 trace_rxrpc_receive(call, rxrpc_receive_rotate, serial, hard_ack);
208 if (flags & RXRPC_LAST_PACKET) {
209 rxrpc_end_rx_phase(call, serial);
211 /* Check to see if there's an ACK that needs sending. */
212 if (after_eq(hard_ack, call->ackr_consumed + 2) ||
213 after_eq(top, call->ackr_seen + 2) ||
214 (hard_ack == top && after(hard_ack, call->ackr_consumed)))
215 rxrpc_propose_ACK(call, RXRPC_ACK_DELAY, 0, serial,
217 rxrpc_propose_ack_rotate_rx);
218 if (call->ackr_reason && call->ackr_reason != RXRPC_ACK_DELAY)
219 rxrpc_send_ack_packet(call, false, NULL);
224 * Decrypt and verify a (sub)packet. The packet's length may be changed due to
225 * padding, but if this is the case, the packet length will be resident in the
226 * socket buffer. Note that we can't modify the master skb info as the skb may
227 * be the home to multiple subpackets.
229 static int rxrpc_verify_packet(struct rxrpc_call *call, struct sk_buff *skb,
231 unsigned int offset, unsigned int len)
233 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
234 rxrpc_seq_t seq = sp->hdr.seq;
235 u16 cksum = sp->hdr.cksum;
239 /* For all but the head jumbo subpacket, the security checksum is in a
240 * jumbo header immediately prior to the data.
242 if ((annotation & RXRPC_RX_ANNO_JUMBO) > 1) {
244 if (skb_copy_bits(skb, offset - 2, &tmp, 2) < 0)
247 seq += (annotation & RXRPC_RX_ANNO_JUMBO) - 1;
250 return call->conn->security->verify_packet(call, skb, offset, len,
255 * Locate the data within a packet. This is complicated by:
257 * (1) An skb may contain a jumbo packet - so we have to find the appropriate
260 * (2) The (sub)packets may be encrypted and, if so, the encrypted portion
261 * contains an extra header which includes the true length of the data,
262 * excluding any encrypted padding.
264 static int rxrpc_locate_data(struct rxrpc_call *call, struct sk_buff *skb,
266 unsigned int *_offset, unsigned int *_len)
268 unsigned int offset = sizeof(struct rxrpc_wire_header);
271 u8 annotation = *_annotation;
273 /* Locate the subpacket */
274 len = skb->len - offset;
275 if ((annotation & RXRPC_RX_ANNO_JUMBO) > 0) {
276 offset += (((annotation & RXRPC_RX_ANNO_JUMBO) - 1) *
277 RXRPC_JUMBO_SUBPKTLEN);
278 len = (annotation & RXRPC_RX_ANNO_JLAST) ?
279 skb->len - offset : RXRPC_JUMBO_SUBPKTLEN;
282 if (!(annotation & RXRPC_RX_ANNO_VERIFIED)) {
283 ret = rxrpc_verify_packet(call, skb, annotation, offset, len);
286 *_annotation |= RXRPC_RX_ANNO_VERIFIED;
291 call->conn->security->locate_data(call, skb, _offset, _len);
296 * Deliver messages to a call. This keeps processing packets until the buffer
297 * is filled and we find either more DATA (returns 0) or the end of the DATA
298 * (returns 1). If more packets are required, it returns -EAGAIN.
300 static int rxrpc_recvmsg_data(struct socket *sock, struct rxrpc_call *call,
301 struct msghdr *msg, struct iov_iter *iter,
302 size_t len, int flags, size_t *_offset)
304 struct rxrpc_skb_priv *sp;
306 rxrpc_seq_t hard_ack, top, seq;
309 unsigned int rx_pkt_offset, rx_pkt_len;
310 int ix, copy, ret = -EAGAIN, ret2;
312 if (test_and_clear_bit(RXRPC_CALL_RX_UNDERRUN, &call->flags) &&
314 rxrpc_send_ack_packet(call, false, NULL);
316 rx_pkt_offset = call->rx_pkt_offset;
317 rx_pkt_len = call->rx_pkt_len;
319 if (call->state >= RXRPC_CALL_SERVER_ACK_REQUEST) {
320 seq = call->rx_hard_ack;
325 /* Barriers against rxrpc_input_data(). */
326 hard_ack = call->rx_hard_ack;
328 while (top = smp_load_acquire(&call->rx_top),
331 ix = seq & RXRPC_RXTX_BUFF_MASK;
332 skb = call->rxtx_buffer[ix];
334 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_hole, seq,
335 rx_pkt_offset, rx_pkt_len, 0);
339 rxrpc_see_skb(skb, rxrpc_skb_rx_seen);
342 if (!(flags & MSG_PEEK))
343 trace_rxrpc_receive(call, rxrpc_receive_front,
344 sp->hdr.serial, seq);
347 sock_recv_timestamp(msg, sock->sk, skb);
349 if (rx_pkt_offset == 0) {
350 ret2 = rxrpc_locate_data(call, skb,
351 &call->rxtx_annotations[ix],
352 &rx_pkt_offset, &rx_pkt_len);
353 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_next, seq,
354 rx_pkt_offset, rx_pkt_len, ret2);
360 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_cont, seq,
361 rx_pkt_offset, rx_pkt_len, 0);
364 /* We have to handle short, empty and used-up DATA packets. */
365 remain = len - *_offset;
370 ret2 = skb_copy_datagram_iter(skb, rx_pkt_offset, iter,
377 /* handle piecemeal consumption of data packets */
378 rx_pkt_offset += copy;
383 if (rx_pkt_len > 0) {
384 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_full, seq,
385 rx_pkt_offset, rx_pkt_len, 0);
386 ASSERTCMP(*_offset, ==, len);
391 /* The whole packet has been transferred. */
392 last = sp->hdr.flags & RXRPC_LAST_PACKET;
393 if (!(flags & MSG_PEEK))
394 rxrpc_rotate_rx_window(call);
399 ASSERTCMP(seq, ==, READ_ONCE(call->rx_top));
408 if (!(flags & MSG_PEEK)) {
409 call->rx_pkt_offset = rx_pkt_offset;
410 call->rx_pkt_len = rx_pkt_len;
413 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_data_return, seq,
414 rx_pkt_offset, rx_pkt_len, ret);
416 set_bit(RXRPC_CALL_RX_UNDERRUN, &call->flags);
421 * Receive a message from an RxRPC socket
422 * - we need to be careful about two or more threads calling recvmsg
425 int rxrpc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
428 struct rxrpc_call *call;
429 struct rxrpc_sock *rx = rxrpc_sk(sock->sk);
437 trace_rxrpc_recvmsg(NULL, rxrpc_recvmsg_enter, 0, 0, 0, 0);
439 if (flags & (MSG_OOB | MSG_TRUNC))
442 timeo = sock_rcvtimeo(&rx->sk, flags & MSG_DONTWAIT);
447 /* Return immediately if a client socket has no outstanding calls */
448 if (RB_EMPTY_ROOT(&rx->calls) &&
449 list_empty(&rx->recvmsg_q) &&
450 rx->sk.sk_state != RXRPC_SERVER_LISTENING) {
451 release_sock(&rx->sk);
455 if (list_empty(&rx->recvmsg_q)) {
462 release_sock(&rx->sk);
464 /* Wait for something to happen */
465 prepare_to_wait_exclusive(sk_sleep(&rx->sk), &wait,
467 ret = sock_error(&rx->sk);
471 if (list_empty(&rx->recvmsg_q)) {
472 if (signal_pending(current))
473 goto wait_interrupted;
474 trace_rxrpc_recvmsg(NULL, rxrpc_recvmsg_wait,
476 timeo = schedule_timeout(timeo);
478 finish_wait(sk_sleep(&rx->sk), &wait);
482 /* Find the next call and dequeue it if we're not just peeking. If we
483 * do dequeue it, that comes with a ref that we will need to release.
485 write_lock_bh(&rx->recvmsg_lock);
486 l = rx->recvmsg_q.next;
487 call = list_entry(l, struct rxrpc_call, recvmsg_link);
488 if (!(flags & MSG_PEEK))
489 list_del_init(&call->recvmsg_link);
491 rxrpc_get_call(call, rxrpc_call_got);
492 write_unlock_bh(&rx->recvmsg_lock);
494 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_dequeue, 0, 0, 0, 0);
496 /* We're going to drop the socket lock, so we need to lock the call
497 * against interference by sendmsg.
499 if (!mutex_trylock(&call->user_mutex)) {
501 if (flags & MSG_DONTWAIT)
502 goto error_requeue_call;
504 if (mutex_lock_interruptible(&call->user_mutex) < 0)
505 goto error_requeue_call;
508 release_sock(&rx->sk);
510 if (test_bit(RXRPC_CALL_RELEASED, &call->flags))
513 if (test_bit(RXRPC_CALL_HAS_USERID, &call->flags)) {
514 if (flags & MSG_CMSG_COMPAT) {
515 unsigned int id32 = call->user_call_ID;
517 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_USER_CALL_ID,
518 sizeof(unsigned int), &id32);
520 unsigned long idl = call->user_call_ID;
522 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_USER_CALL_ID,
523 sizeof(unsigned long), &idl);
526 goto error_unlock_call;
530 struct sockaddr_rxrpc *srx = msg->msg_name;
531 size_t len = sizeof(call->peer->srx);
533 memcpy(msg->msg_name, &call->peer->srx, len);
534 srx->srx_service = call->service_id;
535 msg->msg_namelen = len;
538 switch (READ_ONCE(call->state)) {
539 case RXRPC_CALL_SERVER_ACCEPTING:
540 ret = rxrpc_recvmsg_new_call(rx, call, msg, flags);
542 case RXRPC_CALL_CLIENT_RECV_REPLY:
543 case RXRPC_CALL_SERVER_RECV_REQUEST:
544 case RXRPC_CALL_SERVER_ACK_REQUEST:
545 ret = rxrpc_recvmsg_data(sock, call, msg, &msg->msg_iter, len,
550 if (after(call->rx_top, call->rx_hard_ack) &&
551 call->rxtx_buffer[(call->rx_hard_ack + 1) & RXRPC_RXTX_BUFF_MASK])
552 rxrpc_notify_socket(call);
560 goto error_unlock_call;
562 if (call->state == RXRPC_CALL_COMPLETE) {
563 ret = rxrpc_recvmsg_term(call, msg);
565 goto error_unlock_call;
566 if (!(flags & MSG_PEEK))
567 rxrpc_release_call(rx, call);
568 msg->msg_flags |= MSG_EOR;
573 msg->msg_flags |= MSG_MORE;
575 msg->msg_flags &= ~MSG_MORE;
579 mutex_unlock(&call->user_mutex);
580 rxrpc_put_call(call, rxrpc_call_put);
581 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_return, 0, 0, 0, ret);
585 if (!(flags & MSG_PEEK)) {
586 write_lock_bh(&rx->recvmsg_lock);
587 list_add(&call->recvmsg_link, &rx->recvmsg_q);
588 write_unlock_bh(&rx->recvmsg_lock);
589 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_requeue, 0, 0, 0, 0);
591 rxrpc_put_call(call, rxrpc_call_put);
594 release_sock(&rx->sk);
596 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_return, 0, 0, 0, ret);
600 ret = sock_intr_errno(timeo);
602 finish_wait(sk_sleep(&rx->sk), &wait);
608 * rxrpc_kernel_recv_data - Allow a kernel service to receive data/info
609 * @sock: The socket that the call exists on
610 * @call: The call to send data through
611 * @iter: The buffer to receive into
612 * @want_more: True if more data is expected to be read
613 * @_abort: Where the abort code is stored if -ECONNABORTED is returned
614 * @_service: Where to store the actual service ID (may be upgraded)
616 * Allow a kernel service to receive data and pick up information about the
617 * state of a call. Returns 0 if got what was asked for and there's more
618 * available, 1 if we got what was asked for and we're at the end of the data
619 * and -EAGAIN if we need more data.
621 * Note that we may return -EAGAIN to drain empty packets at the end of the
622 * data, even if we've already copied over the requested data.
624 * *_abort should also be initialised to 0.
626 int rxrpc_kernel_recv_data(struct socket *sock, struct rxrpc_call *call,
627 struct iov_iter *iter,
628 bool want_more, u32 *_abort, u16 *_service)
633 _enter("{%d,%s},%zu,%d",
634 call->debug_id, rxrpc_call_states[call->state],
635 iov_iter_count(iter), want_more);
637 ASSERTCMP(call->state, !=, RXRPC_CALL_SERVER_ACCEPTING);
639 mutex_lock(&call->user_mutex);
641 switch (READ_ONCE(call->state)) {
642 case RXRPC_CALL_CLIENT_RECV_REPLY:
643 case RXRPC_CALL_SERVER_RECV_REQUEST:
644 case RXRPC_CALL_SERVER_ACK_REQUEST:
645 ret = rxrpc_recvmsg_data(sock, call, NULL, iter,
646 iov_iter_count(iter), 0,
651 /* We can only reach here with a partially full buffer if we
652 * have reached the end of the data. We must otherwise have a
653 * full buffer or have been given -EAGAIN.
656 if (iov_iter_count(iter) > 0)
659 goto read_phase_complete;
668 case RXRPC_CALL_COMPLETE:
679 switch (call->ackr_reason) {
682 case RXRPC_ACK_DELAY:
687 rxrpc_send_ack_packet(call, false, NULL);
691 *_service = call->service_id;
692 mutex_unlock(&call->user_mutex);
693 _leave(" = %d [%zu,%d]", ret, iov_iter_count(iter), *_abort);
697 trace_rxrpc_rx_eproto(call, 0, tracepoint_string("short_data"));
701 trace_rxrpc_rx_eproto(call, 0, tracepoint_string("excess_data"));
705 *_abort = call->abort_code;
707 if (call->completion == RXRPC_CALL_SUCCEEDED) {
709 if (iov_iter_count(iter) > 0)
714 EXPORT_SYMBOL(rxrpc_kernel_recv_data);
717 * rxrpc_kernel_get_reply_time - Get timestamp on first reply packet
718 * @sock: The socket that the call exists on
719 * @call: The call to query
720 * @_ts: Where to put the timestamp
722 * Retrieve the timestamp from the first DATA packet of the reply if it is
723 * in the ring. Returns true if successful, false if not.
725 bool rxrpc_kernel_get_reply_time(struct socket *sock, struct rxrpc_call *call,
729 rxrpc_seq_t hard_ack, top, seq;
730 bool success = false;
732 mutex_lock(&call->user_mutex);
734 if (READ_ONCE(call->state) != RXRPC_CALL_CLIENT_RECV_REPLY)
737 hard_ack = call->rx_hard_ack;
742 top = smp_load_acquire(&call->rx_top);
746 skb = call->rxtx_buffer[seq & RXRPC_RXTX_BUFF_MASK];
750 *_ts = skb_get_ktime(skb);
754 mutex_unlock(&call->user_mutex);
757 EXPORT_SYMBOL(rxrpc_kernel_get_reply_time);