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 0 // TODO: May want to transmit final ACK under some circumstances anyway
148 if (call
->state
== RXRPC_CALL_CLIENT_RECV_REPLY
) {
149 rxrpc_propose_ACK(call
, RXRPC_ACK_IDLE
, 0, serial
, true, false,
150 rxrpc_propose_ack_terminal_ack
);
151 rxrpc_send_ack_packet(call
, false, NULL
);
155 write_lock_bh(&call
->state_lock
);
157 switch (call
->state
) {
158 case RXRPC_CALL_CLIENT_RECV_REPLY
:
159 __rxrpc_call_completed(call
);
160 write_unlock_bh(&call
->state_lock
);
163 case RXRPC_CALL_SERVER_RECV_REQUEST
:
164 call
->tx_phase
= true;
165 call
->state
= RXRPC_CALL_SERVER_ACK_REQUEST
;
166 call
->expect_req_by
= jiffies
+ MAX_JIFFY_OFFSET
;
167 write_unlock_bh(&call
->state_lock
);
168 rxrpc_propose_ACK(call
, RXRPC_ACK_DELAY
, 0, serial
, false, true,
169 rxrpc_propose_ack_processing_op
);
172 write_unlock_bh(&call
->state_lock
);
178 * Discard a packet we've used up and advance the Rx window by one.
180 static void rxrpc_rotate_rx_window(struct rxrpc_call
*call
)
182 struct rxrpc_skb_priv
*sp
;
184 rxrpc_serial_t serial
;
185 rxrpc_seq_t hard_ack
, top
;
189 _enter("%d", call
->debug_id
);
191 hard_ack
= call
->rx_hard_ack
;
192 top
= smp_load_acquire(&call
->rx_top
);
193 ASSERT(before(hard_ack
, top
));
196 ix
= hard_ack
& RXRPC_RXTX_BUFF_MASK
;
197 skb
= call
->rxtx_buffer
[ix
];
198 rxrpc_see_skb(skb
, rxrpc_skb_rx_rotated
);
200 flags
= sp
->hdr
.flags
;
201 serial
= sp
->hdr
.serial
;
202 if (call
->rxtx_annotations
[ix
] & RXRPC_RX_ANNO_JUMBO
)
203 serial
+= (call
->rxtx_annotations
[ix
] & RXRPC_RX_ANNO_JUMBO
) - 1;
205 call
->rxtx_buffer
[ix
] = NULL
;
206 call
->rxtx_annotations
[ix
] = 0;
207 /* Barrier against rxrpc_input_data(). */
208 smp_store_release(&call
->rx_hard_ack
, hard_ack
);
210 rxrpc_free_skb(skb
, rxrpc_skb_rx_freed
);
212 _debug("%u,%u,%02x", hard_ack
, top
, flags
);
213 trace_rxrpc_receive(call
, rxrpc_receive_rotate
, serial
, hard_ack
);
214 if (flags
& RXRPC_LAST_PACKET
) {
215 rxrpc_end_rx_phase(call
, serial
);
217 /* Check to see if there's an ACK that needs sending. */
218 if (after_eq(hard_ack
, call
->ackr_consumed
+ 2) ||
219 after_eq(top
, call
->ackr_seen
+ 2) ||
220 (hard_ack
== top
&& after(hard_ack
, call
->ackr_consumed
)))
221 rxrpc_propose_ACK(call
, RXRPC_ACK_DELAY
, 0, serial
,
223 rxrpc_propose_ack_rotate_rx
);
224 if (call
->ackr_reason
&& call
->ackr_reason
!= RXRPC_ACK_DELAY
)
225 rxrpc_send_ack_packet(call
, false, NULL
);
230 * Decrypt and verify a (sub)packet. The packet's length may be changed due to
231 * padding, but if this is the case, the packet length will be resident in the
232 * socket buffer. Note that we can't modify the master skb info as the skb may
233 * be the home to multiple subpackets.
235 static int rxrpc_verify_packet(struct rxrpc_call
*call
, struct sk_buff
*skb
,
237 unsigned int offset
, unsigned int len
)
239 struct rxrpc_skb_priv
*sp
= rxrpc_skb(skb
);
240 rxrpc_seq_t seq
= sp
->hdr
.seq
;
241 u16 cksum
= sp
->hdr
.cksum
;
245 /* For all but the head jumbo subpacket, the security checksum is in a
246 * jumbo header immediately prior to the data.
248 if ((annotation
& RXRPC_RX_ANNO_JUMBO
) > 1) {
250 if (skb_copy_bits(skb
, offset
- 2, &tmp
, 2) < 0)
253 seq
+= (annotation
& RXRPC_RX_ANNO_JUMBO
) - 1;
256 return call
->conn
->security
->verify_packet(call
, skb
, offset
, len
,
261 * Locate the data within a packet. This is complicated by:
263 * (1) An skb may contain a jumbo packet - so we have to find the appropriate
266 * (2) The (sub)packets may be encrypted and, if so, the encrypted portion
267 * contains an extra header which includes the true length of the data,
268 * excluding any encrypted padding.
270 static int rxrpc_locate_data(struct rxrpc_call
*call
, struct sk_buff
*skb
,
272 unsigned int *_offset
, unsigned int *_len
)
274 unsigned int offset
= sizeof(struct rxrpc_wire_header
);
275 unsigned int len
= *_len
;
277 u8 annotation
= *_annotation
;
279 /* Locate the subpacket */
280 len
= skb
->len
- offset
;
281 if ((annotation
& RXRPC_RX_ANNO_JUMBO
) > 0) {
282 offset
+= (((annotation
& RXRPC_RX_ANNO_JUMBO
) - 1) *
283 RXRPC_JUMBO_SUBPKTLEN
);
284 len
= (annotation
& RXRPC_RX_ANNO_JLAST
) ?
285 skb
->len
- offset
: RXRPC_JUMBO_SUBPKTLEN
;
288 if (!(annotation
& RXRPC_RX_ANNO_VERIFIED
)) {
289 ret
= rxrpc_verify_packet(call
, skb
, annotation
, offset
, len
);
292 *_annotation
|= RXRPC_RX_ANNO_VERIFIED
;
297 call
->conn
->security
->locate_data(call
, skb
, _offset
, _len
);
302 * Deliver messages to a call. This keeps processing packets until the buffer
303 * is filled and we find either more DATA (returns 0) or the end of the DATA
304 * (returns 1). If more packets are required, it returns -EAGAIN.
306 static int rxrpc_recvmsg_data(struct socket
*sock
, struct rxrpc_call
*call
,
307 struct msghdr
*msg
, struct iov_iter
*iter
,
308 size_t len
, int flags
, size_t *_offset
)
310 struct rxrpc_skb_priv
*sp
;
312 rxrpc_seq_t hard_ack
, top
, seq
;
315 unsigned int rx_pkt_offset
, rx_pkt_len
;
316 int ix
, copy
, ret
= -EAGAIN
, ret2
;
318 rx_pkt_offset
= call
->rx_pkt_offset
;
319 rx_pkt_len
= call
->rx_pkt_len
;
321 if (call
->state
>= RXRPC_CALL_SERVER_ACK_REQUEST
) {
322 seq
= call
->rx_hard_ack
;
327 /* Barriers against rxrpc_input_data(). */
328 hard_ack
= call
->rx_hard_ack
;
330 while (top
= smp_load_acquire(&call
->rx_top
),
333 ix
= seq
& RXRPC_RXTX_BUFF_MASK
;
334 skb
= call
->rxtx_buffer
[ix
];
336 trace_rxrpc_recvmsg(call
, rxrpc_recvmsg_hole
, seq
,
337 rx_pkt_offset
, rx_pkt_len
, 0);
341 rxrpc_see_skb(skb
, rxrpc_skb_rx_seen
);
344 if (!(flags
& MSG_PEEK
))
345 trace_rxrpc_receive(call
, rxrpc_receive_front
,
346 sp
->hdr
.serial
, seq
);
349 sock_recv_timestamp(msg
, sock
->sk
, skb
);
351 if (rx_pkt_offset
== 0) {
352 ret2
= rxrpc_locate_data(call
, skb
,
353 &call
->rxtx_annotations
[ix
],
354 &rx_pkt_offset
, &rx_pkt_len
);
355 trace_rxrpc_recvmsg(call
, rxrpc_recvmsg_next
, seq
,
356 rx_pkt_offset
, rx_pkt_len
, ret2
);
362 trace_rxrpc_recvmsg(call
, rxrpc_recvmsg_cont
, seq
,
363 rx_pkt_offset
, rx_pkt_len
, 0);
366 /* We have to handle short, empty and used-up DATA packets. */
367 remain
= len
- *_offset
;
372 ret2
= skb_copy_datagram_iter(skb
, rx_pkt_offset
, iter
,
379 /* handle piecemeal consumption of data packets */
380 rx_pkt_offset
+= copy
;
385 if (rx_pkt_len
> 0) {
386 trace_rxrpc_recvmsg(call
, rxrpc_recvmsg_full
, seq
,
387 rx_pkt_offset
, rx_pkt_len
, 0);
388 ASSERTCMP(*_offset
, ==, len
);
393 /* The whole packet has been transferred. */
394 last
= sp
->hdr
.flags
& RXRPC_LAST_PACKET
;
395 if (!(flags
& MSG_PEEK
))
396 rxrpc_rotate_rx_window(call
);
401 ASSERTCMP(seq
, ==, READ_ONCE(call
->rx_top
));
410 if (!(flags
& MSG_PEEK
)) {
411 call
->rx_pkt_offset
= rx_pkt_offset
;
412 call
->rx_pkt_len
= rx_pkt_len
;
415 trace_rxrpc_recvmsg(call
, rxrpc_recvmsg_data_return
, seq
,
416 rx_pkt_offset
, rx_pkt_len
, ret
);
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
);
595 trace_rxrpc_recvmsg(call
, rxrpc_recvmsg_return
, 0, 0, 0, ret
);
599 ret
= sock_intr_errno(timeo
);
601 finish_wait(sk_sleep(&rx
->sk
), &wait
);
607 * rxrpc_kernel_recv_data - Allow a kernel service to receive data/info
608 * @sock: The socket that the call exists on
609 * @call: The call to send data through
610 * @buf: The buffer to receive into
611 * @size: The size of the buffer, including data already read
612 * @_offset: The running offset into the buffer.
613 * @want_more: True if more data is expected to be read
614 * @_abort: Where the abort code is stored if -ECONNABORTED is returned
615 * @_service: Where to store the actual service ID (may be upgraded)
617 * Allow a kernel service to receive data and pick up information about the
618 * state of a call. Returns 0 if got what was asked for and there's more
619 * available, 1 if we got what was asked for and we're at the end of the data
620 * and -EAGAIN if we need more data.
622 * Note that we may return -EAGAIN to drain empty packets at the end of the
623 * data, even if we've already copied over the requested data.
625 * This function adds the amount it transfers to *_offset, so this should be
626 * precleared as appropriate. Note that the amount remaining in the buffer is
627 * taken to be size - *_offset.
629 * *_abort should also be initialised to 0.
631 int rxrpc_kernel_recv_data(struct socket
*sock
, struct rxrpc_call
*call
,
632 void *buf
, size_t size
, size_t *_offset
,
633 bool want_more
, u32
*_abort
, u16
*_service
)
635 struct iov_iter iter
;
639 _enter("{%d,%s},%zu/%zu,%d",
640 call
->debug_id
, rxrpc_call_states
[call
->state
],
641 *_offset
, size
, want_more
);
643 ASSERTCMP(*_offset
, <=, size
);
644 ASSERTCMP(call
->state
, !=, RXRPC_CALL_SERVER_ACCEPTING
);
646 iov
.iov_base
= buf
+ *_offset
;
647 iov
.iov_len
= size
- *_offset
;
648 iov_iter_kvec(&iter
, ITER_KVEC
| READ
, &iov
, 1, size
- *_offset
);
650 mutex_lock(&call
->user_mutex
);
652 switch (READ_ONCE(call
->state
)) {
653 case RXRPC_CALL_CLIENT_RECV_REPLY
:
654 case RXRPC_CALL_SERVER_RECV_REQUEST
:
655 case RXRPC_CALL_SERVER_ACK_REQUEST
:
656 ret
= rxrpc_recvmsg_data(sock
, call
, NULL
, &iter
, size
, 0,
661 /* We can only reach here with a partially full buffer if we
662 * have reached the end of the data. We must otherwise have a
663 * full buffer or have been given -EAGAIN.
669 goto read_phase_complete
;
678 case RXRPC_CALL_COMPLETE
:
690 *_service
= call
->service_id
;
691 mutex_unlock(&call
->user_mutex
);
692 _leave(" = %d [%zu,%d]", ret
, *_offset
, *_abort
);
696 trace_rxrpc_rx_eproto(call
, 0, tracepoint_string("short_data"));
700 trace_rxrpc_rx_eproto(call
, 0, tracepoint_string("excess_data"));
704 *_abort
= call
->abort_code
;
706 if (call
->completion
== RXRPC_CALL_SUCCEEDED
) {
713 EXPORT_SYMBOL(rxrpc_kernel_recv_data
);