1 /* RxRPC packet transmission
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 #include <linux/net.h>
13 #include <linux/skbuff.h>
14 #include <linux/circ_buf.h>
16 #include <net/af_rxrpc.h>
17 #include "ar-internal.h"
19 int rxrpc_resend_timeout
= 4;
21 static int rxrpc_send_data(struct kiocb
*iocb
,
22 struct rxrpc_sock
*rx
,
23 struct rxrpc_call
*call
,
24 struct msghdr
*msg
, size_t len
);
27 * extract control messages from the sendmsg() control buffer
29 static int rxrpc_sendmsg_cmsg(struct rxrpc_sock
*rx
, struct msghdr
*msg
,
30 unsigned long *user_call_ID
,
31 enum rxrpc_command
*command
,
38 *command
= RXRPC_CMD_SEND_DATA
;
40 if (msg
->msg_controllen
== 0)
43 for (cmsg
= CMSG_FIRSTHDR(msg
); cmsg
; cmsg
= CMSG_NXTHDR(msg
, cmsg
)) {
44 if (!CMSG_OK(msg
, cmsg
))
47 len
= cmsg
->cmsg_len
- CMSG_ALIGN(sizeof(struct cmsghdr
));
48 _debug("CMSG %d, %d, %d",
49 cmsg
->cmsg_level
, cmsg
->cmsg_type
, len
);
51 if (cmsg
->cmsg_level
!= SOL_RXRPC
)
54 switch (cmsg
->cmsg_type
) {
55 case RXRPC_USER_CALL_ID
:
56 if (msg
->msg_flags
& MSG_CMSG_COMPAT
) {
57 if (len
!= sizeof(u32
))
59 *user_call_ID
= *(u32
*) CMSG_DATA(cmsg
);
61 if (len
!= sizeof(unsigned long))
63 *user_call_ID
= *(unsigned long *)
66 _debug("User Call ID %lx", *user_call_ID
);
70 if (*command
!= RXRPC_CMD_SEND_DATA
)
72 *command
= RXRPC_CMD_SEND_ABORT
;
73 if (len
!= sizeof(*abort_code
))
75 *abort_code
= *(unsigned int *) CMSG_DATA(cmsg
);
76 _debug("Abort %x", *abort_code
);
82 if (*command
!= RXRPC_CMD_SEND_DATA
)
84 *command
= RXRPC_CMD_ACCEPT
;
101 * abort a call, sending an ABORT packet to the peer
103 static void rxrpc_send_abort(struct rxrpc_call
*call
, u32 abort_code
)
105 write_lock_bh(&call
->state_lock
);
107 if (call
->state
<= RXRPC_CALL_COMPLETE
) {
108 call
->state
= RXRPC_CALL_LOCALLY_ABORTED
;
109 call
->abort_code
= abort_code
;
110 set_bit(RXRPC_CALL_ABORT
, &call
->events
);
111 del_timer_sync(&call
->resend_timer
);
112 del_timer_sync(&call
->ack_timer
);
113 clear_bit(RXRPC_CALL_RESEND_TIMER
, &call
->events
);
114 clear_bit(RXRPC_CALL_ACK
, &call
->events
);
115 clear_bit(RXRPC_CALL_RUN_RTIMER
, &call
->flags
);
116 rxrpc_queue_call(call
);
119 write_unlock_bh(&call
->state_lock
);
123 * send a message forming part of a client call through an RxRPC socket
124 * - caller holds the socket locked
125 * - the socket may be either a client socket or a server socket
127 int rxrpc_client_sendmsg(struct kiocb
*iocb
, struct rxrpc_sock
*rx
,
128 struct rxrpc_transport
*trans
, struct msghdr
*msg
,
131 struct rxrpc_conn_bundle
*bundle
;
132 enum rxrpc_command cmd
;
133 struct rxrpc_call
*call
;
134 unsigned long user_call_ID
= 0;
142 ASSERT(trans
!= NULL
);
144 ret
= rxrpc_sendmsg_cmsg(rx
, msg
, &user_call_ID
, &cmd
, &abort_code
,
151 service_id
= rx
->service_id
;
153 struct sockaddr_rxrpc
*srx
=
154 (struct sockaddr_rxrpc
*) msg
->msg_name
;
155 service_id
= htons(srx
->srx_service
);
158 if (key
&& !rx
->key
->payload
.data
)
160 bundle
= rxrpc_get_bundle(rx
, trans
, key
, service_id
,
163 return PTR_ERR(bundle
);
166 call
= rxrpc_get_client_call(rx
, trans
, bundle
, user_call_ID
,
167 abort_code
== 0, GFP_KERNEL
);
169 rxrpc_put_bundle(trans
, bundle
);
171 _leave(" = %ld", PTR_ERR(call
));
172 return PTR_ERR(call
);
175 _debug("CALL %d USR %lx ST %d on CONN %p",
176 call
->debug_id
, call
->user_call_ID
, call
->state
, call
->conn
);
178 if (call
->state
>= RXRPC_CALL_COMPLETE
) {
179 /* it's too late for this call */
181 } else if (cmd
== RXRPC_CMD_SEND_ABORT
) {
182 rxrpc_send_abort(call
, abort_code
);
183 } else if (cmd
!= RXRPC_CMD_SEND_DATA
) {
185 } else if (call
->state
!= RXRPC_CALL_CLIENT_SEND_REQUEST
) {
186 /* request phase complete for this client call */
189 ret
= rxrpc_send_data(iocb
, rx
, call
, msg
, len
);
192 rxrpc_put_call(call
);
193 _leave(" = %d", ret
);
198 * rxrpc_kernel_send_data - Allow a kernel service to send data on a call
199 * @call: The call to send data through
200 * @msg: The data to send
201 * @len: The amount of data to send
203 * Allow a kernel service to send data on a call. The call must be in an state
204 * appropriate to sending data. No control data should be supplied in @msg,
205 * nor should an address be supplied. MSG_MORE should be flagged if there's
206 * more data to come, otherwise this data will end the transmission phase.
208 int rxrpc_kernel_send_data(struct rxrpc_call
*call
, struct msghdr
*msg
,
213 _enter("{%d,%s},", call
->debug_id
, rxrpc_call_states
[call
->state
]);
215 ASSERTCMP(msg
->msg_name
, ==, NULL
);
216 ASSERTCMP(msg
->msg_control
, ==, NULL
);
218 lock_sock(&call
->socket
->sk
);
220 _debug("CALL %d USR %lx ST %d on CONN %p",
221 call
->debug_id
, call
->user_call_ID
, call
->state
, call
->conn
);
223 if (call
->state
>= RXRPC_CALL_COMPLETE
) {
224 ret
= -ESHUTDOWN
; /* it's too late for this call */
225 } else if (call
->state
!= RXRPC_CALL_CLIENT_SEND_REQUEST
&&
226 call
->state
!= RXRPC_CALL_SERVER_ACK_REQUEST
&&
227 call
->state
!= RXRPC_CALL_SERVER_SEND_REPLY
) {
228 ret
= -EPROTO
; /* request phase complete for this client call */
230 mm_segment_t oldfs
= get_fs();
232 ret
= rxrpc_send_data(NULL
, call
->socket
, call
, msg
, len
);
236 release_sock(&call
->socket
->sk
);
237 _leave(" = %d", ret
);
241 EXPORT_SYMBOL(rxrpc_kernel_send_data
);
244 * rxrpc_kernel_abort_call - Allow a kernel service to abort a call
245 * @call: The call to be aborted
246 * @abort_code: The abort code to stick into the ABORT packet
248 * Allow a kernel service to abort a call, if it's still in an abortable state.
250 void rxrpc_kernel_abort_call(struct rxrpc_call
*call
, u32 abort_code
)
252 _enter("{%d},%d", call
->debug_id
, abort_code
);
254 lock_sock(&call
->socket
->sk
);
256 _debug("CALL %d USR %lx ST %d on CONN %p",
257 call
->debug_id
, call
->user_call_ID
, call
->state
, call
->conn
);
259 if (call
->state
< RXRPC_CALL_COMPLETE
)
260 rxrpc_send_abort(call
, abort_code
);
262 release_sock(&call
->socket
->sk
);
266 EXPORT_SYMBOL(rxrpc_kernel_abort_call
);
269 * send a message through a server socket
270 * - caller holds the socket locked
272 int rxrpc_server_sendmsg(struct kiocb
*iocb
, struct rxrpc_sock
*rx
,
273 struct msghdr
*msg
, size_t len
)
275 enum rxrpc_command cmd
;
276 struct rxrpc_call
*call
;
277 unsigned long user_call_ID
= 0;
283 ret
= rxrpc_sendmsg_cmsg(rx
, msg
, &user_call_ID
, &cmd
, &abort_code
,
288 if (cmd
== RXRPC_CMD_ACCEPT
) {
289 call
= rxrpc_accept_call(rx
, user_call_ID
);
291 return PTR_ERR(call
);
292 rxrpc_put_call(call
);
296 call
= rxrpc_find_server_call(rx
, user_call_ID
);
299 if (call
->state
>= RXRPC_CALL_COMPLETE
) {
305 case RXRPC_CMD_SEND_DATA
:
306 if (call
->state
!= RXRPC_CALL_CLIENT_SEND_REQUEST
&&
307 call
->state
!= RXRPC_CALL_SERVER_ACK_REQUEST
&&
308 call
->state
!= RXRPC_CALL_SERVER_SEND_REPLY
) {
309 /* Tx phase not yet begun for this call */
314 ret
= rxrpc_send_data(iocb
, rx
, call
, msg
, len
);
317 case RXRPC_CMD_SEND_ABORT
:
318 rxrpc_send_abort(call
, abort_code
);
325 rxrpc_put_call(call
);
326 _leave(" = %d", ret
);
331 * send a packet through the transport endpoint
333 int rxrpc_send_packet(struct rxrpc_transport
*trans
, struct sk_buff
*skb
)
339 _enter(",{%d}", skb
->len
);
341 iov
[0].iov_base
= skb
->head
;
342 iov
[0].iov_len
= skb
->len
;
344 msg
.msg_name
= &trans
->peer
->srx
.transport
.sin
;
345 msg
.msg_namelen
= sizeof(trans
->peer
->srx
.transport
.sin
);
346 msg
.msg_control
= NULL
;
347 msg
.msg_controllen
= 0;
350 /* send the packet with the don't fragment bit set if we currently
351 * think it's small enough */
352 if (skb
->len
- sizeof(struct rxrpc_header
) < trans
->peer
->maxdata
) {
353 down_read(&trans
->local
->defrag_sem
);
354 /* send the packet by UDP
355 * - returns -EMSGSIZE if UDP would have to fragment the packet
356 * to go out of the interface
357 * - in which case, we'll have processed the ICMP error
358 * message and update the peer record
360 ret
= kernel_sendmsg(trans
->local
->socket
, &msg
, iov
, 1,
363 up_read(&trans
->local
->defrag_sem
);
364 if (ret
== -EMSGSIZE
)
365 goto send_fragmentable
;
367 _leave(" = %d [%u]", ret
, trans
->peer
->maxdata
);
372 /* attempt to send this message with fragmentation enabled */
373 _debug("send fragment");
375 down_write(&trans
->local
->defrag_sem
);
376 opt
= IP_PMTUDISC_DONT
;
377 ret
= kernel_setsockopt(trans
->local
->socket
, SOL_IP
, IP_MTU_DISCOVER
,
378 (char *) &opt
, sizeof(opt
));
380 ret
= kernel_sendmsg(trans
->local
->socket
, &msg
, iov
, 1,
383 opt
= IP_PMTUDISC_DO
;
384 kernel_setsockopt(trans
->local
->socket
, SOL_IP
,
385 IP_MTU_DISCOVER
, (char *) &opt
, sizeof(opt
));
388 up_write(&trans
->local
->defrag_sem
);
389 _leave(" = %d [frag %u]", ret
, trans
->peer
->maxdata
);
394 * wait for space to appear in the transmit/ACK window
395 * - caller holds the socket locked
397 static int rxrpc_wait_for_tx_window(struct rxrpc_sock
*rx
,
398 struct rxrpc_call
*call
,
401 DECLARE_WAITQUEUE(myself
, current
);
405 CIRC_SPACE(call
->acks_head
, call
->acks_tail
, call
->acks_winsz
),
408 add_wait_queue(&call
->tx_waitq
, &myself
);
411 set_current_state(TASK_INTERRUPTIBLE
);
413 if (CIRC_SPACE(call
->acks_head
, call
->acks_tail
,
414 call
->acks_winsz
) > 0)
416 if (signal_pending(current
)) {
417 ret
= sock_intr_errno(*timeo
);
421 release_sock(&rx
->sk
);
422 *timeo
= schedule_timeout(*timeo
);
426 remove_wait_queue(&call
->tx_waitq
, &myself
);
427 set_current_state(TASK_RUNNING
);
428 _leave(" = %d", ret
);
433 * attempt to schedule an instant Tx resend
435 static inline void rxrpc_instant_resend(struct rxrpc_call
*call
)
437 read_lock_bh(&call
->state_lock
);
438 if (try_to_del_timer_sync(&call
->resend_timer
) >= 0) {
439 clear_bit(RXRPC_CALL_RUN_RTIMER
, &call
->flags
);
440 if (call
->state
< RXRPC_CALL_COMPLETE
&&
441 !test_and_set_bit(RXRPC_CALL_RESEND_TIMER
, &call
->events
))
442 rxrpc_queue_call(call
);
444 read_unlock_bh(&call
->state_lock
);
448 * queue a packet for transmission, set the resend timer and attempt
449 * to send the packet immediately
451 static void rxrpc_queue_packet(struct rxrpc_call
*call
, struct sk_buff
*skb
,
454 struct rxrpc_skb_priv
*sp
= rxrpc_skb(skb
);
457 _net("queue skb %p [%d]", skb
, call
->acks_head
);
459 ASSERT(call
->acks_window
!= NULL
);
460 call
->acks_window
[call
->acks_head
] = (unsigned long) skb
;
462 call
->acks_head
= (call
->acks_head
+ 1) & (call
->acks_winsz
- 1);
464 if (last
|| call
->state
== RXRPC_CALL_SERVER_ACK_REQUEST
) {
465 _debug("________awaiting reply/ACK__________");
466 write_lock_bh(&call
->state_lock
);
467 switch (call
->state
) {
468 case RXRPC_CALL_CLIENT_SEND_REQUEST
:
469 call
->state
= RXRPC_CALL_CLIENT_AWAIT_REPLY
;
471 case RXRPC_CALL_SERVER_ACK_REQUEST
:
472 call
->state
= RXRPC_CALL_SERVER_SEND_REPLY
;
475 case RXRPC_CALL_SERVER_SEND_REPLY
:
476 call
->state
= RXRPC_CALL_SERVER_AWAIT_ACK
;
481 write_unlock_bh(&call
->state_lock
);
484 _proto("Tx DATA %%%u { #%u }",
485 ntohl(sp
->hdr
.serial
), ntohl(sp
->hdr
.seq
));
488 sp
->resend_at
= jiffies
+ rxrpc_resend_timeout
* HZ
;
489 if (!test_and_set_bit(RXRPC_CALL_RUN_RTIMER
, &call
->flags
)) {
491 call
->resend_timer
.expires
= sp
->resend_at
;
492 add_timer(&call
->resend_timer
);
495 /* attempt to cancel the rx-ACK timer, deferring reply transmission if
496 * we're ACK'ing the request phase of an incoming call */
498 if (try_to_del_timer_sync(&call
->ack_timer
) >= 0) {
499 /* the packet may be freed by rxrpc_process_call() before this
501 ret
= rxrpc_send_packet(call
->conn
->trans
, skb
);
502 _net("sent skb %p", skb
);
504 _debug("failed to delete ACK timer");
508 _debug("need instant resend %d", ret
);
510 rxrpc_instant_resend(call
);
517 * send data through a socket
518 * - must be called in process context
519 * - caller holds the socket locked
521 static int rxrpc_send_data(struct kiocb
*iocb
,
522 struct rxrpc_sock
*rx
,
523 struct rxrpc_call
*call
,
524 struct msghdr
*msg
, size_t len
)
526 struct rxrpc_skb_priv
*sp
;
527 unsigned char __user
*from
;
530 struct sock
*sk
= &rx
->sk
;
533 int ret
, ioc
, segment
, copied
;
535 _enter(",,,{%zu},%zu", msg
->msg_iovlen
, len
);
537 timeo
= sock_sndtimeo(sk
, msg
->msg_flags
& MSG_DONTWAIT
);
539 /* this should be in poll */
540 clear_bit(SOCK_ASYNC_NOSPACE
, &sk
->sk_socket
->flags
);
542 if (sk
->sk_err
|| (sk
->sk_shutdown
& SEND_SHUTDOWN
))
546 ioc
= msg
->msg_iovlen
- 1;
547 from
= iov
->iov_base
;
548 segment
= iov
->iov_len
;
550 more
= msg
->msg_flags
& MSG_MORE
;
552 skb
= call
->tx_pending
;
553 call
->tx_pending
= NULL
;
562 _debug("SEGMENT %d @%p", segment
, from
);
565 size_t size
, chunk
, max
, space
;
569 if (CIRC_SPACE(call
->acks_head
, call
->acks_tail
,
570 call
->acks_winsz
) <= 0) {
572 if (msg
->msg_flags
& MSG_DONTWAIT
)
574 ret
= rxrpc_wait_for_tx_window(rx
, call
,
580 max
= call
->conn
->trans
->peer
->maxdata
;
581 max
-= call
->conn
->security_size
;
582 max
&= ~(call
->conn
->size_align
- 1UL);
585 if (chunk
> len
&& !more
)
588 space
= chunk
+ call
->conn
->size_align
;
589 space
&= ~(call
->conn
->size_align
- 1UL);
591 size
= space
+ call
->conn
->header_size
;
593 _debug("SIZE: %zu/%zu/%zu", chunk
, space
, size
);
595 /* create a buffer that we can retain until it's ACK'd */
596 skb
= sock_alloc_send_skb(
597 sk
, size
, msg
->msg_flags
& MSG_DONTWAIT
, &ret
);
603 _debug("ALLOC SEND %p", skb
);
605 ASSERTCMP(skb
->mark
, ==, 0);
607 _debug("HS: %u", call
->conn
->header_size
);
608 skb_reserve(skb
, call
->conn
->header_size
);
609 skb
->len
+= call
->conn
->header_size
;
613 if (sp
->remain
> skb_tailroom(skb
))
614 sp
->remain
= skb_tailroom(skb
);
616 _net("skb: hr %d, tr %d, hl %d, rm %d",
622 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
628 /* append next segment of data to the current buffer */
629 copy
= skb_tailroom(skb
);
630 ASSERTCMP(copy
, >, 0);
633 if (copy
> sp
->remain
)
637 ret
= skb_add_data(skb
, from
, copy
);
648 while (segment
== 0 && ioc
> 0) {
649 from
= iov
->iov_base
;
650 segment
= iov
->iov_len
;
659 /* check for the far side aborting the call or a network error
661 if (call
->state
> RXRPC_CALL_COMPLETE
)
664 /* add the packet to the send queue if it's now full */
665 if (sp
->remain
<= 0 || (segment
== 0 && !more
)) {
666 struct rxrpc_connection
*conn
= call
->conn
;
669 /* pad out if we're using security */
670 if (conn
->security
) {
671 pad
= conn
->security_size
+ skb
->mark
;
672 pad
= conn
->size_align
- pad
;
673 pad
&= conn
->size_align
- 1;
674 _debug("pad %zu", pad
);
676 memset(skb_put(skb
, pad
), 0, pad
);
679 sp
->hdr
.epoch
= conn
->epoch
;
680 sp
->hdr
.cid
= call
->cid
;
681 sp
->hdr
.callNumber
= call
->call_id
;
683 htonl(atomic_inc_return(&call
->sequence
));
685 htonl(atomic_inc_return(&conn
->serial
));
686 sp
->hdr
.type
= RXRPC_PACKET_TYPE_DATA
;
687 sp
->hdr
.userStatus
= 0;
688 sp
->hdr
.securityIndex
= conn
->security_ix
;
690 sp
->hdr
.serviceId
= conn
->service_id
;
692 sp
->hdr
.flags
= conn
->out_clientflag
;
693 if (len
== 0 && !more
)
694 sp
->hdr
.flags
|= RXRPC_LAST_PACKET
;
695 else if (CIRC_SPACE(call
->acks_head
, call
->acks_tail
,
696 call
->acks_winsz
) > 1)
697 sp
->hdr
.flags
|= RXRPC_MORE_PACKETS
;
699 ret
= rxrpc_secure_packet(
700 call
, skb
, skb
->mark
,
701 skb
->head
+ sizeof(struct rxrpc_header
));
705 memcpy(skb
->head
, &sp
->hdr
,
706 sizeof(struct rxrpc_header
));
707 rxrpc_queue_packet(call
, skb
, segment
== 0 && !more
);
711 } while (segment
> 0);
716 call
->tx_pending
= skb
;
717 _leave(" = %d", ret
);
722 if (call
->state
== RXRPC_CALL_NETWORK_ERROR
)
723 ret
= call
->conn
->trans
->peer
->net_error
;
726 _leave(" = %d", ret
);