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/gfp.h>
14 #include <linux/skbuff.h>
15 #include <linux/circ_buf.h>
16 #include <linux/export.h>
18 #include <net/af_rxrpc.h>
19 #include "ar-internal.h"
22 * Time till packet resend (in jiffies).
24 unsigned rxrpc_resend_timeout
= 4 * HZ
;
26 static int rxrpc_send_data(struct rxrpc_sock
*rx
,
27 struct rxrpc_call
*call
,
28 struct msghdr
*msg
, size_t len
);
31 * extract control messages from the sendmsg() control buffer
33 static int rxrpc_sendmsg_cmsg(struct rxrpc_sock
*rx
, struct msghdr
*msg
,
34 unsigned long *user_call_ID
,
35 enum rxrpc_command
*command
,
42 *command
= RXRPC_CMD_SEND_DATA
;
44 if (msg
->msg_controllen
== 0)
47 for_each_cmsghdr(cmsg
, msg
) {
48 if (!CMSG_OK(msg
, cmsg
))
51 len
= cmsg
->cmsg_len
- CMSG_ALIGN(sizeof(struct cmsghdr
));
52 _debug("CMSG %d, %d, %d",
53 cmsg
->cmsg_level
, cmsg
->cmsg_type
, len
);
55 if (cmsg
->cmsg_level
!= SOL_RXRPC
)
58 switch (cmsg
->cmsg_type
) {
59 case RXRPC_USER_CALL_ID
:
60 if (msg
->msg_flags
& MSG_CMSG_COMPAT
) {
61 if (len
!= sizeof(u32
))
63 *user_call_ID
= *(u32
*) CMSG_DATA(cmsg
);
65 if (len
!= sizeof(unsigned long))
67 *user_call_ID
= *(unsigned long *)
70 _debug("User Call ID %lx", *user_call_ID
);
74 if (*command
!= RXRPC_CMD_SEND_DATA
)
76 *command
= RXRPC_CMD_SEND_ABORT
;
77 if (len
!= sizeof(*abort_code
))
79 *abort_code
= *(unsigned int *) CMSG_DATA(cmsg
);
80 _debug("Abort %x", *abort_code
);
86 if (*command
!= RXRPC_CMD_SEND_DATA
)
88 *command
= RXRPC_CMD_ACCEPT
;
105 * abort a call, sending an ABORT packet to the peer
107 static void rxrpc_send_abort(struct rxrpc_call
*call
, u32 abort_code
)
109 write_lock_bh(&call
->state_lock
);
111 if (call
->state
<= RXRPC_CALL_COMPLETE
) {
112 call
->state
= RXRPC_CALL_LOCALLY_ABORTED
;
113 call
->abort_code
= abort_code
;
114 set_bit(RXRPC_CALL_ABORT
, &call
->events
);
115 del_timer_sync(&call
->resend_timer
);
116 del_timer_sync(&call
->ack_timer
);
117 clear_bit(RXRPC_CALL_RESEND_TIMER
, &call
->events
);
118 clear_bit(RXRPC_CALL_ACK
, &call
->events
);
119 clear_bit(RXRPC_CALL_RUN_RTIMER
, &call
->flags
);
120 rxrpc_queue_call(call
);
123 write_unlock_bh(&call
->state_lock
);
127 * send a message forming part of a client call through an RxRPC socket
128 * - caller holds the socket locked
129 * - the socket may be either a client socket or a server socket
131 int rxrpc_client_sendmsg(struct rxrpc_sock
*rx
, struct rxrpc_transport
*trans
,
132 struct msghdr
*msg
, size_t len
)
134 struct rxrpc_conn_bundle
*bundle
;
135 enum rxrpc_command cmd
;
136 struct rxrpc_call
*call
;
137 unsigned long user_call_ID
= 0;
145 ASSERT(trans
!= NULL
);
147 ret
= rxrpc_sendmsg_cmsg(rx
, msg
, &user_call_ID
, &cmd
, &abort_code
,
154 service_id
= rx
->service_id
;
156 DECLARE_SOCKADDR(struct sockaddr_rxrpc
*, srx
,
158 service_id
= htons(srx
->srx_service
);
161 if (key
&& !rx
->key
->payload
.data
[0])
163 bundle
= rxrpc_get_bundle(rx
, trans
, key
, service_id
,
166 return PTR_ERR(bundle
);
169 call
= rxrpc_get_client_call(rx
, trans
, bundle
, user_call_ID
,
170 abort_code
== 0, GFP_KERNEL
);
172 rxrpc_put_bundle(trans
, bundle
);
174 _leave(" = %ld", PTR_ERR(call
));
175 return PTR_ERR(call
);
178 _debug("CALL %d USR %lx ST %d on CONN %p",
179 call
->debug_id
, call
->user_call_ID
, call
->state
, call
->conn
);
181 if (call
->state
>= RXRPC_CALL_COMPLETE
) {
182 /* it's too late for this call */
184 } else if (cmd
== RXRPC_CMD_SEND_ABORT
) {
185 rxrpc_send_abort(call
, abort_code
);
186 } else if (cmd
!= RXRPC_CMD_SEND_DATA
) {
188 } else if (call
->state
!= RXRPC_CALL_CLIENT_SEND_REQUEST
) {
189 /* request phase complete for this client call */
192 ret
= rxrpc_send_data(rx
, call
, msg
, len
);
195 rxrpc_put_call(call
);
196 _leave(" = %d", ret
);
201 * rxrpc_kernel_send_data - Allow a kernel service to send data on a call
202 * @call: The call to send data through
203 * @msg: The data to send
204 * @len: The amount of data to send
206 * Allow a kernel service to send data on a call. The call must be in an state
207 * appropriate to sending data. No control data should be supplied in @msg,
208 * nor should an address be supplied. MSG_MORE should be flagged if there's
209 * more data to come, otherwise this data will end the transmission phase.
211 int rxrpc_kernel_send_data(struct rxrpc_call
*call
, struct msghdr
*msg
,
216 _enter("{%d,%s},", call
->debug_id
, rxrpc_call_states
[call
->state
]);
218 ASSERTCMP(msg
->msg_name
, ==, NULL
);
219 ASSERTCMP(msg
->msg_control
, ==, NULL
);
221 lock_sock(&call
->socket
->sk
);
223 _debug("CALL %d USR %lx ST %d on CONN %p",
224 call
->debug_id
, call
->user_call_ID
, call
->state
, call
->conn
);
226 if (call
->state
>= RXRPC_CALL_COMPLETE
) {
227 ret
= -ESHUTDOWN
; /* it's too late for this call */
228 } else if (call
->state
!= RXRPC_CALL_CLIENT_SEND_REQUEST
&&
229 call
->state
!= RXRPC_CALL_SERVER_ACK_REQUEST
&&
230 call
->state
!= RXRPC_CALL_SERVER_SEND_REPLY
) {
231 ret
= -EPROTO
; /* request phase complete for this client call */
233 ret
= rxrpc_send_data(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 rxrpc_sock
*rx
, struct msghdr
*msg
, size_t len
)
274 enum rxrpc_command cmd
;
275 struct rxrpc_call
*call
;
276 unsigned long user_call_ID
= 0;
282 ret
= rxrpc_sendmsg_cmsg(rx
, msg
, &user_call_ID
, &cmd
, &abort_code
,
287 if (cmd
== RXRPC_CMD_ACCEPT
) {
288 call
= rxrpc_accept_call(rx
, user_call_ID
);
290 return PTR_ERR(call
);
291 rxrpc_put_call(call
);
295 call
= rxrpc_find_server_call(rx
, user_call_ID
);
298 if (call
->state
>= RXRPC_CALL_COMPLETE
) {
304 case RXRPC_CMD_SEND_DATA
:
305 if (call
->state
!= RXRPC_CALL_CLIENT_SEND_REQUEST
&&
306 call
->state
!= RXRPC_CALL_SERVER_ACK_REQUEST
&&
307 call
->state
!= RXRPC_CALL_SERVER_SEND_REPLY
) {
308 /* Tx phase not yet begun for this call */
313 ret
= rxrpc_send_data(rx
, call
, msg
, len
);
316 case RXRPC_CMD_SEND_ABORT
:
317 rxrpc_send_abort(call
, abort_code
);
324 rxrpc_put_call(call
);
325 _leave(" = %d", ret
);
330 * send a packet through the transport endpoint
332 int rxrpc_send_packet(struct rxrpc_transport
*trans
, struct sk_buff
*skb
)
338 _enter(",{%d}", skb
->len
);
340 iov
[0].iov_base
= skb
->head
;
341 iov
[0].iov_len
= skb
->len
;
343 msg
.msg_name
= &trans
->peer
->srx
.transport
.sin
;
344 msg
.msg_namelen
= sizeof(trans
->peer
->srx
.transport
.sin
);
345 msg
.msg_control
= NULL
;
346 msg
.msg_controllen
= 0;
349 /* send the packet with the don't fragment bit set if we currently
350 * think it's small enough */
351 if (skb
->len
- sizeof(struct rxrpc_header
) < trans
->peer
->maxdata
) {
352 down_read(&trans
->local
->defrag_sem
);
353 /* send the packet by UDP
354 * - returns -EMSGSIZE if UDP would have to fragment the packet
355 * to go out of the interface
356 * - in which case, we'll have processed the ICMP error
357 * message and update the peer record
359 ret
= kernel_sendmsg(trans
->local
->socket
, &msg
, iov
, 1,
362 up_read(&trans
->local
->defrag_sem
);
363 if (ret
== -EMSGSIZE
)
364 goto send_fragmentable
;
366 _leave(" = %d [%u]", ret
, trans
->peer
->maxdata
);
371 /* attempt to send this message with fragmentation enabled */
372 _debug("send fragment");
374 down_write(&trans
->local
->defrag_sem
);
375 opt
= IP_PMTUDISC_DONT
;
376 ret
= kernel_setsockopt(trans
->local
->socket
, SOL_IP
, IP_MTU_DISCOVER
,
377 (char *) &opt
, sizeof(opt
));
379 ret
= kernel_sendmsg(trans
->local
->socket
, &msg
, iov
, 1,
382 opt
= IP_PMTUDISC_DO
;
383 kernel_setsockopt(trans
->local
->socket
, SOL_IP
,
384 IP_MTU_DISCOVER
, (char *) &opt
, sizeof(opt
));
387 up_write(&trans
->local
->defrag_sem
);
388 _leave(" = %d [frag %u]", ret
, trans
->peer
->maxdata
);
393 * wait for space to appear in the transmit/ACK window
394 * - caller holds the socket locked
396 static int rxrpc_wait_for_tx_window(struct rxrpc_sock
*rx
,
397 struct rxrpc_call
*call
,
400 DECLARE_WAITQUEUE(myself
, current
);
404 CIRC_SPACE(call
->acks_head
, call
->acks_tail
, call
->acks_winsz
),
407 add_wait_queue(&call
->tx_waitq
, &myself
);
410 set_current_state(TASK_INTERRUPTIBLE
);
412 if (CIRC_SPACE(call
->acks_head
, call
->acks_tail
,
413 call
->acks_winsz
) > 0)
415 if (signal_pending(current
)) {
416 ret
= sock_intr_errno(*timeo
);
420 release_sock(&rx
->sk
);
421 *timeo
= schedule_timeout(*timeo
);
425 remove_wait_queue(&call
->tx_waitq
, &myself
);
426 set_current_state(TASK_RUNNING
);
427 _leave(" = %d", ret
);
432 * attempt to schedule an instant Tx resend
434 static inline void rxrpc_instant_resend(struct rxrpc_call
*call
)
436 read_lock_bh(&call
->state_lock
);
437 if (try_to_del_timer_sync(&call
->resend_timer
) >= 0) {
438 clear_bit(RXRPC_CALL_RUN_RTIMER
, &call
->flags
);
439 if (call
->state
< RXRPC_CALL_COMPLETE
&&
440 !test_and_set_bit(RXRPC_CALL_RESEND_TIMER
, &call
->events
))
441 rxrpc_queue_call(call
);
443 read_unlock_bh(&call
->state_lock
);
447 * queue a packet for transmission, set the resend timer and attempt
448 * to send the packet immediately
450 static void rxrpc_queue_packet(struct rxrpc_call
*call
, struct sk_buff
*skb
,
453 struct rxrpc_skb_priv
*sp
= rxrpc_skb(skb
);
456 _net("queue skb %p [%d]", skb
, call
->acks_head
);
458 ASSERT(call
->acks_window
!= NULL
);
459 call
->acks_window
[call
->acks_head
] = (unsigned long) skb
;
461 call
->acks_head
= (call
->acks_head
+ 1) & (call
->acks_winsz
- 1);
463 if (last
|| call
->state
== RXRPC_CALL_SERVER_ACK_REQUEST
) {
464 _debug("________awaiting reply/ACK__________");
465 write_lock_bh(&call
->state_lock
);
466 switch (call
->state
) {
467 case RXRPC_CALL_CLIENT_SEND_REQUEST
:
468 call
->state
= RXRPC_CALL_CLIENT_AWAIT_REPLY
;
470 case RXRPC_CALL_SERVER_ACK_REQUEST
:
471 call
->state
= RXRPC_CALL_SERVER_SEND_REPLY
;
474 case RXRPC_CALL_SERVER_SEND_REPLY
:
475 call
->state
= RXRPC_CALL_SERVER_AWAIT_ACK
;
480 write_unlock_bh(&call
->state_lock
);
483 _proto("Tx DATA %%%u { #%u }",
484 ntohl(sp
->hdr
.serial
), ntohl(sp
->hdr
.seq
));
486 sp
->need_resend
= false;
487 sp
->resend_at
= jiffies
+ rxrpc_resend_timeout
;
488 if (!test_and_set_bit(RXRPC_CALL_RUN_RTIMER
, &call
->flags
)) {
490 call
->resend_timer
.expires
= sp
->resend_at
;
491 add_timer(&call
->resend_timer
);
494 /* attempt to cancel the rx-ACK timer, deferring reply transmission if
495 * we're ACK'ing the request phase of an incoming call */
497 if (try_to_del_timer_sync(&call
->ack_timer
) >= 0) {
498 /* the packet may be freed by rxrpc_process_call() before this
500 ret
= rxrpc_send_packet(call
->conn
->trans
, skb
);
501 _net("sent skb %p", skb
);
503 _debug("failed to delete ACK timer");
507 _debug("need instant resend %d", ret
);
508 sp
->need_resend
= true;
509 rxrpc_instant_resend(call
);
516 * send data through a socket
517 * - must be called in process context
518 * - caller holds the socket locked
520 static int rxrpc_send_data(struct rxrpc_sock
*rx
,
521 struct rxrpc_call
*call
,
522 struct msghdr
*msg
, size_t len
)
524 struct rxrpc_skb_priv
*sp
;
526 struct sock
*sk
= &rx
->sk
;
531 timeo
= sock_sndtimeo(sk
, msg
->msg_flags
& MSG_DONTWAIT
);
533 /* this should be in poll */
534 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
536 if (sk
->sk_err
|| (sk
->sk_shutdown
& SEND_SHUTDOWN
))
539 more
= msg
->msg_flags
& MSG_MORE
;
541 skb
= call
->tx_pending
;
542 call
->tx_pending
= NULL
;
547 size_t size
, chunk
, max
, space
;
551 if (CIRC_SPACE(call
->acks_head
, call
->acks_tail
,
552 call
->acks_winsz
) <= 0) {
554 if (msg
->msg_flags
& MSG_DONTWAIT
)
556 ret
= rxrpc_wait_for_tx_window(rx
, call
,
562 max
= call
->conn
->trans
->peer
->maxdata
;
563 max
-= call
->conn
->security_size
;
564 max
&= ~(call
->conn
->size_align
- 1UL);
567 if (chunk
> msg_data_left(msg
) && !more
)
568 chunk
= msg_data_left(msg
);
570 space
= chunk
+ call
->conn
->size_align
;
571 space
&= ~(call
->conn
->size_align
- 1UL);
573 size
= space
+ call
->conn
->header_size
;
575 _debug("SIZE: %zu/%zu/%zu", chunk
, space
, size
);
577 /* create a buffer that we can retain until it's ACK'd */
578 skb
= sock_alloc_send_skb(
579 sk
, size
, msg
->msg_flags
& MSG_DONTWAIT
, &ret
);
585 _debug("ALLOC SEND %p", skb
);
587 ASSERTCMP(skb
->mark
, ==, 0);
589 _debug("HS: %u", call
->conn
->header_size
);
590 skb_reserve(skb
, call
->conn
->header_size
);
591 skb
->len
+= call
->conn
->header_size
;
595 if (sp
->remain
> skb_tailroom(skb
))
596 sp
->remain
= skb_tailroom(skb
);
598 _net("skb: hr %d, tr %d, hl %d, rm %d",
604 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
610 /* append next segment of data to the current buffer */
611 if (msg_data_left(msg
) > 0) {
612 int copy
= skb_tailroom(skb
);
613 ASSERTCMP(copy
, >, 0);
614 if (copy
> msg_data_left(msg
))
615 copy
= msg_data_left(msg
);
616 if (copy
> sp
->remain
)
620 ret
= skb_add_data(skb
, &msg
->msg_iter
, copy
);
629 /* check for the far side aborting the call or a network error
631 if (call
->state
> RXRPC_CALL_COMPLETE
)
634 /* add the packet to the send queue if it's now full */
635 if (sp
->remain
<= 0 ||
636 (msg_data_left(msg
) == 0 && !more
)) {
637 struct rxrpc_connection
*conn
= call
->conn
;
641 /* pad out if we're using security */
642 if (conn
->security
) {
643 pad
= conn
->security_size
+ skb
->mark
;
644 pad
= conn
->size_align
- pad
;
645 pad
&= conn
->size_align
- 1;
646 _debug("pad %zu", pad
);
648 memset(skb_put(skb
, pad
), 0, pad
);
651 seq
= atomic_inc_return(&call
->sequence
);
653 sp
->hdr
.epoch
= conn
->epoch
;
654 sp
->hdr
.cid
= call
->cid
;
655 sp
->hdr
.callNumber
= call
->call_id
;
656 sp
->hdr
.seq
= htonl(seq
);
658 htonl(atomic_inc_return(&conn
->serial
));
659 sp
->hdr
.type
= RXRPC_PACKET_TYPE_DATA
;
660 sp
->hdr
.userStatus
= 0;
661 sp
->hdr
.securityIndex
= conn
->security_ix
;
663 sp
->hdr
.serviceId
= conn
->service_id
;
665 sp
->hdr
.flags
= conn
->out_clientflag
;
666 if (msg_data_left(msg
) == 0 && !more
)
667 sp
->hdr
.flags
|= RXRPC_LAST_PACKET
;
668 else if (CIRC_SPACE(call
->acks_head
, call
->acks_tail
,
669 call
->acks_winsz
) > 1)
670 sp
->hdr
.flags
|= RXRPC_MORE_PACKETS
;
672 sp
->hdr
.flags
|= RXRPC_REQUEST_ACK
;
674 ret
= rxrpc_secure_packet(
675 call
, skb
, skb
->mark
,
676 skb
->head
+ sizeof(struct rxrpc_header
));
680 memcpy(skb
->head
, &sp
->hdr
,
681 sizeof(struct rxrpc_header
));
682 rxrpc_queue_packet(call
, skb
, !msg_data_left(msg
) && !more
);
685 } while (msg_data_left(msg
) > 0);
690 call
->tx_pending
= skb
;
691 _leave(" = %d", ret
);
696 if (call
->state
== RXRPC_CALL_NETWORK_ERROR
)
697 ret
= call
->conn
->trans
->peer
->net_error
;
700 _leave(" = %d", ret
);