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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14 #include <linux/net.h>
15 #include <linux/gfp.h>
16 #include <linux/skbuff.h>
17 #include <linux/circ_buf.h>
18 #include <linux/export.h>
20 #include <net/af_rxrpc.h>
21 #include "ar-internal.h"
24 * Time till packet resend (in jiffies).
26 unsigned int rxrpc_resend_timeout
= 4 * HZ
;
28 static int rxrpc_send_data(struct rxrpc_sock
*rx
,
29 struct rxrpc_call
*call
,
30 struct msghdr
*msg
, size_t len
);
33 * extract control messages from the sendmsg() control buffer
35 static int rxrpc_sendmsg_cmsg(struct msghdr
*msg
,
36 unsigned long *user_call_ID
,
37 enum rxrpc_command
*command
,
41 bool got_user_ID
= false;
44 *command
= RXRPC_CMD_SEND_DATA
;
46 if (msg
->msg_controllen
== 0)
49 for_each_cmsghdr(cmsg
, msg
) {
50 if (!CMSG_OK(msg
, cmsg
))
53 len
= cmsg
->cmsg_len
- CMSG_ALIGN(sizeof(struct cmsghdr
));
54 _debug("CMSG %d, %d, %d",
55 cmsg
->cmsg_level
, cmsg
->cmsg_type
, len
);
57 if (cmsg
->cmsg_level
!= SOL_RXRPC
)
60 switch (cmsg
->cmsg_type
) {
61 case RXRPC_USER_CALL_ID
:
62 if (msg
->msg_flags
& MSG_CMSG_COMPAT
) {
63 if (len
!= sizeof(u32
))
65 *user_call_ID
= *(u32
*) CMSG_DATA(cmsg
);
67 if (len
!= sizeof(unsigned long))
69 *user_call_ID
= *(unsigned long *)
72 _debug("User Call ID %lx", *user_call_ID
);
77 if (*command
!= RXRPC_CMD_SEND_DATA
)
79 *command
= RXRPC_CMD_SEND_ABORT
;
80 if (len
!= sizeof(*abort_code
))
82 *abort_code
= *(unsigned int *) CMSG_DATA(cmsg
);
83 _debug("Abort %x", *abort_code
);
89 if (*command
!= RXRPC_CMD_SEND_DATA
)
91 *command
= RXRPC_CMD_ACCEPT
;
108 * abort a call, sending an ABORT packet to the peer
110 static void rxrpc_send_abort(struct rxrpc_call
*call
, u32 abort_code
)
112 write_lock_bh(&call
->state_lock
);
114 if (call
->state
<= RXRPC_CALL_COMPLETE
) {
115 call
->state
= RXRPC_CALL_LOCALLY_ABORTED
;
116 call
->local_abort
= abort_code
;
117 set_bit(RXRPC_CALL_EV_ABORT
, &call
->events
);
118 del_timer_sync(&call
->resend_timer
);
119 del_timer_sync(&call
->ack_timer
);
120 clear_bit(RXRPC_CALL_EV_RESEND_TIMER
, &call
->events
);
121 clear_bit(RXRPC_CALL_EV_ACK
, &call
->events
);
122 clear_bit(RXRPC_CALL_RUN_RTIMER
, &call
->flags
);
123 rxrpc_queue_call(call
);
126 write_unlock_bh(&call
->state_lock
);
130 * Create a new client call for sendmsg().
132 static struct rxrpc_call
*
133 rxrpc_new_client_call_for_sendmsg(struct rxrpc_sock
*rx
, struct msghdr
*msg
,
134 unsigned long user_call_ID
)
136 struct rxrpc_conn_parameters cp
;
137 struct rxrpc_conn_bundle
*bundle
;
138 struct rxrpc_transport
*trans
;
139 struct rxrpc_call
*call
;
143 DECLARE_SOCKADDR(struct sockaddr_rxrpc
*, srx
, msg
->msg_name
);
148 return ERR_PTR(-EDESTADDRREQ
);
151 if (key
&& !rx
->key
->payload
.data
[0])
154 memset(&cp
, 0, sizeof(cp
));
155 cp
.local
= rx
->local
;
157 cp
.security_level
= rx
->min_sec_level
;
158 cp
.exclusive
= test_bit(RXRPC_SOCK_EXCLUSIVE_CONN
, &rx
->flags
);
159 cp
.service_id
= srx
->srx_service
;
160 trans
= rxrpc_name_to_transport(&cp
, msg
->msg_name
, msg
->msg_namelen
,
163 ret
= PTR_ERR(trans
);
166 cp
.peer
= trans
->peer
;
168 bundle
= rxrpc_get_bundle(rx
, trans
, cp
.key
, srx
->srx_service
,
170 if (IS_ERR(bundle
)) {
171 ret
= PTR_ERR(bundle
);
175 call
= rxrpc_new_client_call(rx
, &cp
, trans
, bundle
, user_call_ID
,
177 rxrpc_put_bundle(trans
, bundle
);
178 rxrpc_put_transport(trans
);
184 _leave(" = %p\n", call
);
188 rxrpc_put_transport(trans
);
190 _leave(" = %ld", ret
);
195 * send a message forming part of a client call through an RxRPC socket
196 * - caller holds the socket locked
197 * - the socket may be either a client socket or a server socket
199 int rxrpc_do_sendmsg(struct rxrpc_sock
*rx
, struct msghdr
*msg
, size_t len
)
201 enum rxrpc_command cmd
;
202 struct rxrpc_call
*call
;
203 unsigned long user_call_ID
= 0;
209 ret
= rxrpc_sendmsg_cmsg(msg
, &user_call_ID
, &cmd
, &abort_code
);
213 if (cmd
== RXRPC_CMD_ACCEPT
) {
214 if (rx
->sk
.sk_state
!= RXRPC_SERVER_LISTENING
)
216 call
= rxrpc_accept_call(rx
, user_call_ID
);
218 return PTR_ERR(call
);
219 rxrpc_put_call(call
);
223 call
= rxrpc_find_call_by_user_ID(rx
, user_call_ID
);
225 if (cmd
!= RXRPC_CMD_SEND_DATA
)
227 call
= rxrpc_new_client_call_for_sendmsg(rx
, msg
, user_call_ID
);
229 return PTR_ERR(call
);
232 _debug("CALL %d USR %lx ST %d on CONN %p",
233 call
->debug_id
, call
->user_call_ID
, call
->state
, call
->conn
);
235 if (call
->state
>= RXRPC_CALL_COMPLETE
) {
236 /* it's too late for this call */
238 } else if (cmd
== RXRPC_CMD_SEND_ABORT
) {
239 rxrpc_send_abort(call
, abort_code
);
241 } else if (cmd
!= RXRPC_CMD_SEND_DATA
) {
243 } else if (!call
->in_clientflag
&&
244 call
->state
!= RXRPC_CALL_CLIENT_SEND_REQUEST
) {
245 /* request phase complete for this client call */
247 } else if (call
->in_clientflag
&&
248 call
->state
!= RXRPC_CALL_SERVER_ACK_REQUEST
&&
249 call
->state
!= RXRPC_CALL_SERVER_SEND_REPLY
) {
250 /* Reply phase not begun or not complete for service call. */
253 ret
= rxrpc_send_data(rx
, call
, msg
, len
);
256 rxrpc_put_call(call
);
257 _leave(" = %d", ret
);
262 * rxrpc_kernel_send_data - Allow a kernel service to send data on a call
263 * @call: The call to send data through
264 * @msg: The data to send
265 * @len: The amount of data to send
267 * Allow a kernel service to send data on a call. The call must be in an state
268 * appropriate to sending data. No control data should be supplied in @msg,
269 * nor should an address be supplied. MSG_MORE should be flagged if there's
270 * more data to come, otherwise this data will end the transmission phase.
272 int rxrpc_kernel_send_data(struct rxrpc_call
*call
, struct msghdr
*msg
,
277 _enter("{%d,%s},", call
->debug_id
, rxrpc_call_states
[call
->state
]);
279 ASSERTCMP(msg
->msg_name
, ==, NULL
);
280 ASSERTCMP(msg
->msg_control
, ==, NULL
);
282 lock_sock(&call
->socket
->sk
);
284 _debug("CALL %d USR %lx ST %d on CONN %p",
285 call
->debug_id
, call
->user_call_ID
, call
->state
, call
->conn
);
287 if (call
->state
>= RXRPC_CALL_COMPLETE
) {
288 ret
= -ESHUTDOWN
; /* it's too late for this call */
289 } else if (call
->state
!= RXRPC_CALL_CLIENT_SEND_REQUEST
&&
290 call
->state
!= RXRPC_CALL_SERVER_ACK_REQUEST
&&
291 call
->state
!= RXRPC_CALL_SERVER_SEND_REPLY
) {
292 ret
= -EPROTO
; /* request phase complete for this client call */
294 ret
= rxrpc_send_data(call
->socket
, call
, msg
, len
);
297 release_sock(&call
->socket
->sk
);
298 _leave(" = %d", ret
);
302 EXPORT_SYMBOL(rxrpc_kernel_send_data
);
305 * rxrpc_kernel_abort_call - Allow a kernel service to abort a call
306 * @call: The call to be aborted
307 * @abort_code: The abort code to stick into the ABORT packet
309 * Allow a kernel service to abort a call, if it's still in an abortable state.
311 void rxrpc_kernel_abort_call(struct rxrpc_call
*call
, u32 abort_code
)
313 _enter("{%d},%d", call
->debug_id
, abort_code
);
315 lock_sock(&call
->socket
->sk
);
317 _debug("CALL %d USR %lx ST %d on CONN %p",
318 call
->debug_id
, call
->user_call_ID
, call
->state
, call
->conn
);
320 if (call
->state
< RXRPC_CALL_COMPLETE
)
321 rxrpc_send_abort(call
, abort_code
);
323 release_sock(&call
->socket
->sk
);
327 EXPORT_SYMBOL(rxrpc_kernel_abort_call
);
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_wire_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
, ACCESS_ONCE(call
->acks_tail
),
408 add_wait_queue(&call
->tx_waitq
, &myself
);
411 set_current_state(TASK_INTERRUPTIBLE
);
413 if (CIRC_SPACE(call
->acks_head
, ACCESS_ONCE(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_EV_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 }", sp
->hdr
.serial
, 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 * Convert a host-endian header into a network-endian header.
518 static void rxrpc_insert_header(struct sk_buff
*skb
)
520 struct rxrpc_wire_header whdr
;
521 struct rxrpc_skb_priv
*sp
= rxrpc_skb(skb
);
523 whdr
.epoch
= htonl(sp
->hdr
.epoch
);
524 whdr
.cid
= htonl(sp
->hdr
.cid
);
525 whdr
.callNumber
= htonl(sp
->hdr
.callNumber
);
526 whdr
.seq
= htonl(sp
->hdr
.seq
);
527 whdr
.serial
= htonl(sp
->hdr
.serial
);
528 whdr
.type
= sp
->hdr
.type
;
529 whdr
.flags
= sp
->hdr
.flags
;
530 whdr
.userStatus
= sp
->hdr
.userStatus
;
531 whdr
.securityIndex
= sp
->hdr
.securityIndex
;
532 whdr
._rsvd
= htons(sp
->hdr
._rsvd
);
533 whdr
.serviceId
= htons(sp
->hdr
.serviceId
);
535 memcpy(skb
->head
, &whdr
, sizeof(whdr
));
539 * send data through a socket
540 * - must be called in process context
541 * - caller holds the socket locked
543 static int rxrpc_send_data(struct rxrpc_sock
*rx
,
544 struct rxrpc_call
*call
,
545 struct msghdr
*msg
, size_t len
)
547 struct rxrpc_skb_priv
*sp
;
549 struct sock
*sk
= &rx
->sk
;
554 timeo
= sock_sndtimeo(sk
, msg
->msg_flags
& MSG_DONTWAIT
);
556 /* this should be in poll */
557 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE
, sk
);
559 if (sk
->sk_err
|| (sk
->sk_shutdown
& SEND_SHUTDOWN
))
562 more
= msg
->msg_flags
& MSG_MORE
;
564 skb
= call
->tx_pending
;
565 call
->tx_pending
= NULL
;
570 size_t size
, chunk
, max
, space
;
574 if (CIRC_SPACE(call
->acks_head
,
575 ACCESS_ONCE(call
->acks_tail
),
576 call
->acks_winsz
) <= 0) {
578 if (msg
->msg_flags
& MSG_DONTWAIT
)
580 ret
= rxrpc_wait_for_tx_window(rx
, call
,
586 max
= call
->conn
->trans
->peer
->maxdata
;
587 max
-= call
->conn
->security_size
;
588 max
&= ~(call
->conn
->size_align
- 1UL);
591 if (chunk
> msg_data_left(msg
) && !more
)
592 chunk
= msg_data_left(msg
);
594 space
= chunk
+ call
->conn
->size_align
;
595 space
&= ~(call
->conn
->size_align
- 1UL);
597 size
= space
+ call
->conn
->header_size
;
599 _debug("SIZE: %zu/%zu/%zu", chunk
, space
, size
);
601 /* create a buffer that we can retain until it's ACK'd */
602 skb
= sock_alloc_send_skb(
603 sk
, size
, msg
->msg_flags
& MSG_DONTWAIT
, &ret
);
609 _debug("ALLOC SEND %p", skb
);
611 ASSERTCMP(skb
->mark
, ==, 0);
613 _debug("HS: %u", call
->conn
->header_size
);
614 skb_reserve(skb
, call
->conn
->header_size
);
615 skb
->len
+= call
->conn
->header_size
;
619 if (sp
->remain
> skb_tailroom(skb
))
620 sp
->remain
= skb_tailroom(skb
);
622 _net("skb: hr %d, tr %d, hl %d, rm %d",
628 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
634 /* append next segment of data to the current buffer */
635 if (msg_data_left(msg
) > 0) {
636 int copy
= skb_tailroom(skb
);
637 ASSERTCMP(copy
, >, 0);
638 if (copy
> msg_data_left(msg
))
639 copy
= msg_data_left(msg
);
640 if (copy
> sp
->remain
)
644 ret
= skb_add_data(skb
, &msg
->msg_iter
, copy
);
653 /* check for the far side aborting the call or a network error
655 if (call
->state
> RXRPC_CALL_COMPLETE
)
658 /* add the packet to the send queue if it's now full */
659 if (sp
->remain
<= 0 ||
660 (msg_data_left(msg
) == 0 && !more
)) {
661 struct rxrpc_connection
*conn
= call
->conn
;
665 /* pad out if we're using security */
666 if (conn
->security_ix
) {
667 pad
= conn
->security_size
+ skb
->mark
;
668 pad
= conn
->size_align
- pad
;
669 pad
&= conn
->size_align
- 1;
670 _debug("pad %zu", pad
);
672 memset(skb_put(skb
, pad
), 0, pad
);
675 seq
= atomic_inc_return(&call
->sequence
);
677 sp
->hdr
.epoch
= conn
->proto
.epoch
;
678 sp
->hdr
.cid
= call
->cid
;
679 sp
->hdr
.callNumber
= call
->call_id
;
681 sp
->hdr
.serial
= atomic_inc_return(&conn
->serial
);
682 sp
->hdr
.type
= RXRPC_PACKET_TYPE_DATA
;
683 sp
->hdr
.userStatus
= 0;
684 sp
->hdr
.securityIndex
= conn
->security_ix
;
686 sp
->hdr
.serviceId
= call
->service_id
;
688 sp
->hdr
.flags
= conn
->out_clientflag
;
689 if (msg_data_left(msg
) == 0 && !more
)
690 sp
->hdr
.flags
|= RXRPC_LAST_PACKET
;
691 else if (CIRC_SPACE(call
->acks_head
,
692 ACCESS_ONCE(call
->acks_tail
),
693 call
->acks_winsz
) > 1)
694 sp
->hdr
.flags
|= RXRPC_MORE_PACKETS
;
696 sp
->hdr
.flags
|= RXRPC_REQUEST_ACK
;
698 ret
= conn
->security
->secure_packet(
699 call
, skb
, skb
->mark
,
700 skb
->head
+ sizeof(struct rxrpc_wire_header
));
704 rxrpc_insert_header(skb
);
705 rxrpc_queue_packet(call
, skb
, !msg_data_left(msg
) && !more
);
708 } while (msg_data_left(msg
) > 0);
713 call
->tx_pending
= skb
;
714 _leave(" = %d", ret
);
719 if (call
->state
== RXRPC_CALL_NETWORK_ERROR
)
720 ret
= call
->error_report
< RXRPC_LOCAL_ERROR_OFFSET
?
722 call
->error_report
- RXRPC_LOCAL_ERROR_OFFSET
;
725 _leave(" = %d", ret
);