1 /* RxRPC recvmsg() implementation
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/net.h>
15 #include <linux/skbuff.h>
16 #include <linux/export.h>
17 #include <linux/sched/signal.h>
20 #include <net/af_rxrpc.h>
21 #include "ar-internal.h"
24 * Post a call for attention by the socket or kernel service. Further
25 * notifications are suppressed by putting recvmsg_link on a dummy queue.
27 void rxrpc_notify_socket(struct rxrpc_call *call)
29 struct rxrpc_sock *rx;
32 _enter("%d", call->debug_id);
34 if (!list_empty(&call->recvmsg_link))
39 rx = rcu_dereference(call->socket);
41 if (rx && sk->sk_state < RXRPC_CLOSE) {
42 if (call->notify_rx) {
43 spin_lock_bh(&call->notify_lock);
44 call->notify_rx(sk, call, call->user_call_ID);
45 spin_unlock_bh(&call->notify_lock);
47 write_lock_bh(&rx->recvmsg_lock);
48 if (list_empty(&call->recvmsg_link)) {
49 rxrpc_get_call(call, rxrpc_call_got);
50 list_add_tail(&call->recvmsg_link, &rx->recvmsg_q);
52 write_unlock_bh(&rx->recvmsg_lock);
54 if (!sock_flag(sk, SOCK_DEAD)) {
55 _debug("call %ps", sk->sk_data_ready);
56 sk->sk_data_ready(sk);
66 * Pass a call terminating message to userspace.
68 static int rxrpc_recvmsg_term(struct rxrpc_call *call, struct msghdr *msg)
73 switch (call->completion) {
74 case RXRPC_CALL_SUCCEEDED:
76 if (rxrpc_is_service_call(call))
77 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ACK, 0, &tmp);
79 case RXRPC_CALL_REMOTELY_ABORTED:
80 tmp = call->abort_code;
81 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ABORT, 4, &tmp);
83 case RXRPC_CALL_LOCALLY_ABORTED:
84 tmp = call->abort_code;
85 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_ABORT, 4, &tmp);
87 case RXRPC_CALL_NETWORK_ERROR:
89 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_NET_ERROR, 4, &tmp);
91 case RXRPC_CALL_LOCAL_ERROR:
93 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_LOCAL_ERROR, 4, &tmp);
96 pr_err("Invalid terminal call state %u\n", call->state);
101 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_terminal, call->rx_hard_ack,
102 call->rx_pkt_offset, call->rx_pkt_len, ret);
107 * Pass back notification of a new call. The call is added to the
108 * to-be-accepted list. This means that the next call to be accepted might not
109 * be the last call seen awaiting acceptance, but unless we leave this on the
110 * front of the queue and block all other messages until someone gives us a
111 * user_ID for it, there's not a lot we can do.
113 static int rxrpc_recvmsg_new_call(struct rxrpc_sock *rx,
114 struct rxrpc_call *call,
115 struct msghdr *msg, int flags)
119 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_NEW_CALL, 0, &tmp);
121 if (ret == 0 && !(flags & MSG_PEEK)) {
122 _debug("to be accepted");
123 write_lock_bh(&rx->recvmsg_lock);
124 list_del_init(&call->recvmsg_link);
125 write_unlock_bh(&rx->recvmsg_lock);
127 rxrpc_get_call(call, rxrpc_call_got);
128 write_lock(&rx->call_lock);
129 list_add_tail(&call->accept_link, &rx->to_be_accepted);
130 write_unlock(&rx->call_lock);
133 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_to_be_accepted, 1, 0, 0, ret);
138 * End the packet reception phase.
140 static void rxrpc_end_rx_phase(struct rxrpc_call *call, rxrpc_serial_t serial)
142 _enter("%d,%s", call->debug_id, rxrpc_call_states[call->state]);
144 trace_rxrpc_receive(call, rxrpc_receive_end, 0, call->rx_top);
145 ASSERTCMP(call->rx_hard_ack, ==, call->rx_top);
147 if (call->state == RXRPC_CALL_CLIENT_RECV_REPLY) {
148 rxrpc_propose_ACK(call, RXRPC_ACK_IDLE, 0, serial, false, true,
149 rxrpc_propose_ack_terminal_ack);
150 //rxrpc_send_ack_packet(call, false, NULL);
153 write_lock_bh(&call->state_lock);
155 switch (call->state) {
156 case RXRPC_CALL_CLIENT_RECV_REPLY:
157 __rxrpc_call_completed(call);
158 write_unlock_bh(&call->state_lock);
161 case RXRPC_CALL_SERVER_RECV_REQUEST:
162 call->tx_phase = true;
163 call->state = RXRPC_CALL_SERVER_ACK_REQUEST;
164 call->expect_req_by = jiffies + MAX_JIFFY_OFFSET;
165 write_unlock_bh(&call->state_lock);
166 rxrpc_propose_ACK(call, RXRPC_ACK_DELAY, 0, serial, false, true,
167 rxrpc_propose_ack_processing_op);
170 write_unlock_bh(&call->state_lock);
176 * Discard a packet we've used up and advance the Rx window by one.
178 static void rxrpc_rotate_rx_window(struct rxrpc_call *call)
180 struct rxrpc_skb_priv *sp;
182 rxrpc_serial_t serial;
183 rxrpc_seq_t hard_ack, top;
187 _enter("%d", call->debug_id);
189 hard_ack = call->rx_hard_ack;
190 top = smp_load_acquire(&call->rx_top);
191 ASSERT(before(hard_ack, top));
194 ix = hard_ack & RXRPC_RXTX_BUFF_MASK;
195 skb = call->rxtx_buffer[ix];
196 rxrpc_see_skb(skb, rxrpc_skb_rx_rotated);
198 flags = sp->hdr.flags;
199 serial = sp->hdr.serial;
200 if (call->rxtx_annotations[ix] & RXRPC_RX_ANNO_JUMBO)
201 serial += (call->rxtx_annotations[ix] & RXRPC_RX_ANNO_JUMBO) - 1;
203 call->rxtx_buffer[ix] = NULL;
204 call->rxtx_annotations[ix] = 0;
205 /* Barrier against rxrpc_input_data(). */
206 smp_store_release(&call->rx_hard_ack, hard_ack);
208 rxrpc_free_skb(skb, rxrpc_skb_rx_freed);
210 _debug("%u,%u,%02x", hard_ack, top, flags);
211 trace_rxrpc_receive(call, rxrpc_receive_rotate, serial, hard_ack);
212 if (flags & RXRPC_LAST_PACKET) {
213 rxrpc_end_rx_phase(call, serial);
215 /* Check to see if there's an ACK that needs sending. */
216 if (after_eq(hard_ack, call->ackr_consumed + 2) ||
217 after_eq(top, call->ackr_seen + 2) ||
218 (hard_ack == top && after(hard_ack, call->ackr_consumed)))
219 rxrpc_propose_ACK(call, RXRPC_ACK_DELAY, 0, serial,
221 rxrpc_propose_ack_rotate_rx);
222 if (call->ackr_reason && call->ackr_reason != RXRPC_ACK_DELAY)
223 rxrpc_send_ack_packet(call, false, NULL);
228 * Decrypt and verify a (sub)packet. The packet's length may be changed due to
229 * padding, but if this is the case, the packet length will be resident in the
230 * socket buffer. Note that we can't modify the master skb info as the skb may
231 * be the home to multiple subpackets.
233 static int rxrpc_verify_packet(struct rxrpc_call *call, struct sk_buff *skb,
235 unsigned int offset, unsigned int len)
237 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
238 rxrpc_seq_t seq = sp->hdr.seq;
239 u16 cksum = sp->hdr.cksum;
243 /* For all but the head jumbo subpacket, the security checksum is in a
244 * jumbo header immediately prior to the data.
246 if ((annotation & RXRPC_RX_ANNO_JUMBO) > 1) {
248 if (skb_copy_bits(skb, offset - 2, &tmp, 2) < 0)
251 seq += (annotation & RXRPC_RX_ANNO_JUMBO) - 1;
254 return call->conn->security->verify_packet(call, skb, offset, len,
259 * Locate the data within a packet. This is complicated by:
261 * (1) An skb may contain a jumbo packet - so we have to find the appropriate
264 * (2) The (sub)packets may be encrypted and, if so, the encrypted portion
265 * contains an extra header which includes the true length of the data,
266 * excluding any encrypted padding.
268 static int rxrpc_locate_data(struct rxrpc_call *call, struct sk_buff *skb,
270 unsigned int *_offset, unsigned int *_len)
272 unsigned int offset = sizeof(struct rxrpc_wire_header);
275 u8 annotation = *_annotation;
277 /* Locate the subpacket */
278 len = skb->len - offset;
279 if ((annotation & RXRPC_RX_ANNO_JUMBO) > 0) {
280 offset += (((annotation & RXRPC_RX_ANNO_JUMBO) - 1) *
281 RXRPC_JUMBO_SUBPKTLEN);
282 len = (annotation & RXRPC_RX_ANNO_JLAST) ?
283 skb->len - offset : RXRPC_JUMBO_SUBPKTLEN;
286 if (!(annotation & RXRPC_RX_ANNO_VERIFIED)) {
287 ret = rxrpc_verify_packet(call, skb, annotation, offset, len);
290 *_annotation |= RXRPC_RX_ANNO_VERIFIED;
295 call->conn->security->locate_data(call, skb, _offset, _len);
300 * Deliver messages to a call. This keeps processing packets until the buffer
301 * is filled and we find either more DATA (returns 0) or the end of the DATA
302 * (returns 1). If more packets are required, it returns -EAGAIN.
304 static int rxrpc_recvmsg_data(struct socket *sock, struct rxrpc_call *call,
305 struct msghdr *msg, struct iov_iter *iter,
306 size_t len, int flags, size_t *_offset)
308 struct rxrpc_skb_priv *sp;
310 rxrpc_seq_t hard_ack, top, seq;
313 unsigned int rx_pkt_offset, rx_pkt_len;
314 int ix, copy, ret = -EAGAIN, ret2;
316 if (test_and_clear_bit(RXRPC_CALL_RX_UNDERRUN, &call->flags) &&
318 rxrpc_send_ack_packet(call, false, NULL);
320 rx_pkt_offset = call->rx_pkt_offset;
321 rx_pkt_len = call->rx_pkt_len;
323 if (call->state >= RXRPC_CALL_SERVER_ACK_REQUEST) {
324 seq = call->rx_hard_ack;
329 /* Barriers against rxrpc_input_data(). */
330 hard_ack = call->rx_hard_ack;
332 while (top = smp_load_acquire(&call->rx_top),
335 ix = seq & RXRPC_RXTX_BUFF_MASK;
336 skb = call->rxtx_buffer[ix];
338 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_hole, seq,
339 rx_pkt_offset, rx_pkt_len, 0);
343 rxrpc_see_skb(skb, rxrpc_skb_rx_seen);
346 if (!(flags & MSG_PEEK))
347 trace_rxrpc_receive(call, rxrpc_receive_front,
348 sp->hdr.serial, seq);
351 sock_recv_timestamp(msg, sock->sk, skb);
353 if (rx_pkt_offset == 0) {
354 ret2 = rxrpc_locate_data(call, skb,
355 &call->rxtx_annotations[ix],
356 &rx_pkt_offset, &rx_pkt_len);
357 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_next, seq,
358 rx_pkt_offset, rx_pkt_len, ret2);
364 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_cont, seq,
365 rx_pkt_offset, rx_pkt_len, 0);
368 /* We have to handle short, empty and used-up DATA packets. */
369 remain = len - *_offset;
374 ret2 = skb_copy_datagram_iter(skb, rx_pkt_offset, iter,
381 /* handle piecemeal consumption of data packets */
382 rx_pkt_offset += copy;
387 if (rx_pkt_len > 0) {
388 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_full, seq,
389 rx_pkt_offset, rx_pkt_len, 0);
390 ASSERTCMP(*_offset, ==, len);
395 /* The whole packet has been transferred. */
396 last = sp->hdr.flags & RXRPC_LAST_PACKET;
397 if (!(flags & MSG_PEEK))
398 rxrpc_rotate_rx_window(call);
403 ASSERTCMP(seq, ==, READ_ONCE(call->rx_top));
412 if (!(flags & MSG_PEEK)) {
413 call->rx_pkt_offset = rx_pkt_offset;
414 call->rx_pkt_len = rx_pkt_len;
417 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_data_return, seq,
418 rx_pkt_offset, rx_pkt_len, ret);
420 set_bit(RXRPC_CALL_RX_UNDERRUN, &call->flags);
425 * Receive a message from an RxRPC socket
426 * - we need to be careful about two or more threads calling recvmsg
429 int rxrpc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
432 struct rxrpc_call *call;
433 struct rxrpc_sock *rx = rxrpc_sk(sock->sk);
441 trace_rxrpc_recvmsg(NULL, rxrpc_recvmsg_enter, 0, 0, 0, 0);
443 if (flags & (MSG_OOB | MSG_TRUNC))
446 timeo = sock_rcvtimeo(&rx->sk, flags & MSG_DONTWAIT);
451 /* Return immediately if a client socket has no outstanding calls */
452 if (RB_EMPTY_ROOT(&rx->calls) &&
453 list_empty(&rx->recvmsg_q) &&
454 rx->sk.sk_state != RXRPC_SERVER_LISTENING) {
455 release_sock(&rx->sk);
459 if (list_empty(&rx->recvmsg_q)) {
466 release_sock(&rx->sk);
468 /* Wait for something to happen */
469 prepare_to_wait_exclusive(sk_sleep(&rx->sk), &wait,
471 ret = sock_error(&rx->sk);
475 if (list_empty(&rx->recvmsg_q)) {
476 if (signal_pending(current))
477 goto wait_interrupted;
478 trace_rxrpc_recvmsg(NULL, rxrpc_recvmsg_wait,
480 timeo = schedule_timeout(timeo);
482 finish_wait(sk_sleep(&rx->sk), &wait);
486 /* Find the next call and dequeue it if we're not just peeking. If we
487 * do dequeue it, that comes with a ref that we will need to release.
489 write_lock_bh(&rx->recvmsg_lock);
490 l = rx->recvmsg_q.next;
491 call = list_entry(l, struct rxrpc_call, recvmsg_link);
492 if (!(flags & MSG_PEEK))
493 list_del_init(&call->recvmsg_link);
495 rxrpc_get_call(call, rxrpc_call_got);
496 write_unlock_bh(&rx->recvmsg_lock);
498 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_dequeue, 0, 0, 0, 0);
500 /* We're going to drop the socket lock, so we need to lock the call
501 * against interference by sendmsg.
503 if (!mutex_trylock(&call->user_mutex)) {
505 if (flags & MSG_DONTWAIT)
506 goto error_requeue_call;
508 if (mutex_lock_interruptible(&call->user_mutex) < 0)
509 goto error_requeue_call;
512 release_sock(&rx->sk);
514 if (test_bit(RXRPC_CALL_RELEASED, &call->flags))
517 if (test_bit(RXRPC_CALL_HAS_USERID, &call->flags)) {
518 if (flags & MSG_CMSG_COMPAT) {
519 unsigned int id32 = call->user_call_ID;
521 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_USER_CALL_ID,
522 sizeof(unsigned int), &id32);
524 unsigned long idl = call->user_call_ID;
526 ret = put_cmsg(msg, SOL_RXRPC, RXRPC_USER_CALL_ID,
527 sizeof(unsigned long), &idl);
530 goto error_unlock_call;
534 struct sockaddr_rxrpc *srx = msg->msg_name;
535 size_t len = sizeof(call->peer->srx);
537 memcpy(msg->msg_name, &call->peer->srx, len);
538 srx->srx_service = call->service_id;
539 msg->msg_namelen = len;
542 switch (READ_ONCE(call->state)) {
543 case RXRPC_CALL_SERVER_ACCEPTING:
544 ret = rxrpc_recvmsg_new_call(rx, call, msg, flags);
546 case RXRPC_CALL_CLIENT_RECV_REPLY:
547 case RXRPC_CALL_SERVER_RECV_REQUEST:
548 case RXRPC_CALL_SERVER_ACK_REQUEST:
549 ret = rxrpc_recvmsg_data(sock, call, msg, &msg->msg_iter, len,
554 if (after(call->rx_top, call->rx_hard_ack) &&
555 call->rxtx_buffer[(call->rx_hard_ack + 1) & RXRPC_RXTX_BUFF_MASK])
556 rxrpc_notify_socket(call);
564 goto error_unlock_call;
566 if (call->state == RXRPC_CALL_COMPLETE) {
567 ret = rxrpc_recvmsg_term(call, msg);
569 goto error_unlock_call;
570 if (!(flags & MSG_PEEK))
571 rxrpc_release_call(rx, call);
572 msg->msg_flags |= MSG_EOR;
577 msg->msg_flags |= MSG_MORE;
579 msg->msg_flags &= ~MSG_MORE;
583 mutex_unlock(&call->user_mutex);
584 rxrpc_put_call(call, rxrpc_call_put);
585 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_return, 0, 0, 0, ret);
589 if (!(flags & MSG_PEEK)) {
590 write_lock_bh(&rx->recvmsg_lock);
591 list_add(&call->recvmsg_link, &rx->recvmsg_q);
592 write_unlock_bh(&rx->recvmsg_lock);
593 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_requeue, 0, 0, 0, 0);
595 rxrpc_put_call(call, rxrpc_call_put);
598 release_sock(&rx->sk);
600 trace_rxrpc_recvmsg(call, rxrpc_recvmsg_return, 0, 0, 0, ret);
604 ret = sock_intr_errno(timeo);
606 finish_wait(sk_sleep(&rx->sk), &wait);
612 * rxrpc_kernel_recv_data - Allow a kernel service to receive data/info
613 * @sock: The socket that the call exists on
614 * @call: The call to send data through
615 * @iter: The buffer to receive into
616 * @want_more: True if more data is expected to be read
617 * @_abort: Where the abort code is stored if -ECONNABORTED is returned
618 * @_service: Where to store the actual service ID (may be upgraded)
620 * Allow a kernel service to receive data and pick up information about the
621 * state of a call. Returns 0 if got what was asked for and there's more
622 * available, 1 if we got what was asked for and we're at the end of the data
623 * and -EAGAIN if we need more data.
625 * Note that we may return -EAGAIN to drain empty packets at the end of the
626 * data, even if we've already copied over the requested data.
628 * *_abort should also be initialised to 0.
630 int rxrpc_kernel_recv_data(struct socket *sock, struct rxrpc_call *call,
631 struct iov_iter *iter,
632 bool want_more, u32 *_abort, u16 *_service)
637 _enter("{%d,%s},%zu,%d",
638 call->debug_id, rxrpc_call_states[call->state],
639 iov_iter_count(iter), want_more);
641 ASSERTCMP(call->state, !=, RXRPC_CALL_SERVER_ACCEPTING);
643 mutex_lock(&call->user_mutex);
645 switch (READ_ONCE(call->state)) {
646 case RXRPC_CALL_CLIENT_RECV_REPLY:
647 case RXRPC_CALL_SERVER_RECV_REQUEST:
648 case RXRPC_CALL_SERVER_ACK_REQUEST:
649 ret = rxrpc_recvmsg_data(sock, call, NULL, iter,
650 iov_iter_count(iter), 0,
655 /* We can only reach here with a partially full buffer if we
656 * have reached the end of the data. We must otherwise have a
657 * full buffer or have been given -EAGAIN.
660 if (iov_iter_count(iter) > 0)
663 goto read_phase_complete;
672 case RXRPC_CALL_COMPLETE:
683 switch (call->ackr_reason) {
686 case RXRPC_ACK_DELAY:
691 rxrpc_send_ack_packet(call, false, NULL);
695 *_service = call->service_id;
696 mutex_unlock(&call->user_mutex);
697 _leave(" = %d [%zu,%d]", ret, iov_iter_count(iter), *_abort);
701 trace_rxrpc_rx_eproto(call, 0, tracepoint_string("short_data"));
705 trace_rxrpc_rx_eproto(call, 0, tracepoint_string("excess_data"));
709 *_abort = call->abort_code;
711 if (call->completion == RXRPC_CALL_SUCCEEDED) {
713 if (iov_iter_count(iter) > 0)
718 EXPORT_SYMBOL(rxrpc_kernel_recv_data);
721 * rxrpc_kernel_get_reply_time - Get timestamp on first reply packet
722 * @sock: The socket that the call exists on
723 * @call: The call to query
724 * @_ts: Where to put the timestamp
726 * Retrieve the timestamp from the first DATA packet of the reply if it is
727 * in the ring. Returns true if successful, false if not.
729 bool rxrpc_kernel_get_reply_time(struct socket *sock, struct rxrpc_call *call,
733 rxrpc_seq_t hard_ack, top, seq;
734 bool success = false;
736 mutex_lock(&call->user_mutex);
738 if (READ_ONCE(call->state) != RXRPC_CALL_CLIENT_RECV_REPLY)
741 hard_ack = call->rx_hard_ack;
746 top = smp_load_acquire(&call->rx_top);
750 skb = call->rxtx_buffer[seq & RXRPC_RXTX_BUFF_MASK];
754 *_ts = skb_get_ktime(skb);
758 mutex_unlock(&call->user_mutex);
761 EXPORT_SYMBOL(rxrpc_kernel_get_reply_time);