unicore32: drop pte_file()-related helpers
[linux-2.6/btrfs-unstable.git] / net / rxrpc / ar-output.c
blobe1a9373e59799fd2a9cd998fbdc4399d2d021f6a
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>
17 #include <net/sock.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 kiocb *iocb,
27 struct rxrpc_sock *rx,
28 struct rxrpc_call *call,
29 struct msghdr *msg, size_t len);
32 * extract control messages from the sendmsg() control buffer
34 static int rxrpc_sendmsg_cmsg(struct rxrpc_sock *rx, struct msghdr *msg,
35 unsigned long *user_call_ID,
36 enum rxrpc_command *command,
37 u32 *abort_code,
38 bool server)
40 struct cmsghdr *cmsg;
41 int len;
43 *command = RXRPC_CMD_SEND_DATA;
45 if (msg->msg_controllen == 0)
46 return -EINVAL;
48 for_each_cmsghdr(cmsg, msg) {
49 if (!CMSG_OK(msg, cmsg))
50 return -EINVAL;
52 len = cmsg->cmsg_len - CMSG_ALIGN(sizeof(struct cmsghdr));
53 _debug("CMSG %d, %d, %d",
54 cmsg->cmsg_level, cmsg->cmsg_type, len);
56 if (cmsg->cmsg_level != SOL_RXRPC)
57 continue;
59 switch (cmsg->cmsg_type) {
60 case RXRPC_USER_CALL_ID:
61 if (msg->msg_flags & MSG_CMSG_COMPAT) {
62 if (len != sizeof(u32))
63 return -EINVAL;
64 *user_call_ID = *(u32 *) CMSG_DATA(cmsg);
65 } else {
66 if (len != sizeof(unsigned long))
67 return -EINVAL;
68 *user_call_ID = *(unsigned long *)
69 CMSG_DATA(cmsg);
71 _debug("User Call ID %lx", *user_call_ID);
72 break;
74 case RXRPC_ABORT:
75 if (*command != RXRPC_CMD_SEND_DATA)
76 return -EINVAL;
77 *command = RXRPC_CMD_SEND_ABORT;
78 if (len != sizeof(*abort_code))
79 return -EINVAL;
80 *abort_code = *(unsigned int *) CMSG_DATA(cmsg);
81 _debug("Abort %x", *abort_code);
82 if (*abort_code == 0)
83 return -EINVAL;
84 break;
86 case RXRPC_ACCEPT:
87 if (*command != RXRPC_CMD_SEND_DATA)
88 return -EINVAL;
89 *command = RXRPC_CMD_ACCEPT;
90 if (len != 0)
91 return -EINVAL;
92 if (!server)
93 return -EISCONN;
94 break;
96 default:
97 return -EINVAL;
101 _leave(" = 0");
102 return 0;
106 * abort a call, sending an ABORT packet to the peer
108 static void rxrpc_send_abort(struct rxrpc_call *call, u32 abort_code)
110 write_lock_bh(&call->state_lock);
112 if (call->state <= RXRPC_CALL_COMPLETE) {
113 call->state = RXRPC_CALL_LOCALLY_ABORTED;
114 call->abort_code = abort_code;
115 set_bit(RXRPC_CALL_ABORT, &call->events);
116 del_timer_sync(&call->resend_timer);
117 del_timer_sync(&call->ack_timer);
118 clear_bit(RXRPC_CALL_RESEND_TIMER, &call->events);
119 clear_bit(RXRPC_CALL_ACK, &call->events);
120 clear_bit(RXRPC_CALL_RUN_RTIMER, &call->flags);
121 rxrpc_queue_call(call);
124 write_unlock_bh(&call->state_lock);
128 * send a message forming part of a client call through an RxRPC socket
129 * - caller holds the socket locked
130 * - the socket may be either a client socket or a server socket
132 int rxrpc_client_sendmsg(struct kiocb *iocb, struct rxrpc_sock *rx,
133 struct rxrpc_transport *trans, struct msghdr *msg,
134 size_t len)
136 struct rxrpc_conn_bundle *bundle;
137 enum rxrpc_command cmd;
138 struct rxrpc_call *call;
139 unsigned long user_call_ID = 0;
140 struct key *key;
141 __be16 service_id;
142 u32 abort_code = 0;
143 int ret;
145 _enter("");
147 ASSERT(trans != NULL);
149 ret = rxrpc_sendmsg_cmsg(rx, msg, &user_call_ID, &cmd, &abort_code,
150 false);
151 if (ret < 0)
152 return ret;
154 bundle = NULL;
155 if (trans) {
156 service_id = rx->service_id;
157 if (msg->msg_name) {
158 DECLARE_SOCKADDR(struct sockaddr_rxrpc *, srx,
159 msg->msg_name);
160 service_id = htons(srx->srx_service);
162 key = rx->key;
163 if (key && !rx->key->payload.data)
164 key = NULL;
165 bundle = rxrpc_get_bundle(rx, trans, key, service_id,
166 GFP_KERNEL);
167 if (IS_ERR(bundle))
168 return PTR_ERR(bundle);
171 call = rxrpc_get_client_call(rx, trans, bundle, user_call_ID,
172 abort_code == 0, GFP_KERNEL);
173 if (trans)
174 rxrpc_put_bundle(trans, bundle);
175 if (IS_ERR(call)) {
176 _leave(" = %ld", PTR_ERR(call));
177 return PTR_ERR(call);
180 _debug("CALL %d USR %lx ST %d on CONN %p",
181 call->debug_id, call->user_call_ID, call->state, call->conn);
183 if (call->state >= RXRPC_CALL_COMPLETE) {
184 /* it's too late for this call */
185 ret = -ESHUTDOWN;
186 } else if (cmd == RXRPC_CMD_SEND_ABORT) {
187 rxrpc_send_abort(call, abort_code);
188 } else if (cmd != RXRPC_CMD_SEND_DATA) {
189 ret = -EINVAL;
190 } else if (call->state != RXRPC_CALL_CLIENT_SEND_REQUEST) {
191 /* request phase complete for this client call */
192 ret = -EPROTO;
193 } else {
194 ret = rxrpc_send_data(iocb, rx, call, msg, len);
197 rxrpc_put_call(call);
198 _leave(" = %d", ret);
199 return ret;
203 * rxrpc_kernel_send_data - Allow a kernel service to send data on a call
204 * @call: The call to send data through
205 * @msg: The data to send
206 * @len: The amount of data to send
208 * Allow a kernel service to send data on a call. The call must be in an state
209 * appropriate to sending data. No control data should be supplied in @msg,
210 * nor should an address be supplied. MSG_MORE should be flagged if there's
211 * more data to come, otherwise this data will end the transmission phase.
213 int rxrpc_kernel_send_data(struct rxrpc_call *call, struct msghdr *msg,
214 size_t len)
216 int ret;
218 _enter("{%d,%s},", call->debug_id, rxrpc_call_states[call->state]);
220 ASSERTCMP(msg->msg_name, ==, NULL);
221 ASSERTCMP(msg->msg_control, ==, NULL);
223 lock_sock(&call->socket->sk);
225 _debug("CALL %d USR %lx ST %d on CONN %p",
226 call->debug_id, call->user_call_ID, call->state, call->conn);
228 if (call->state >= RXRPC_CALL_COMPLETE) {
229 ret = -ESHUTDOWN; /* it's too late for this call */
230 } else if (call->state != RXRPC_CALL_CLIENT_SEND_REQUEST &&
231 call->state != RXRPC_CALL_SERVER_ACK_REQUEST &&
232 call->state != RXRPC_CALL_SERVER_SEND_REPLY) {
233 ret = -EPROTO; /* request phase complete for this client call */
234 } else {
235 mm_segment_t oldfs = get_fs();
236 set_fs(KERNEL_DS);
237 ret = rxrpc_send_data(NULL, call->socket, call, msg, len);
238 set_fs(oldfs);
241 release_sock(&call->socket->sk);
242 _leave(" = %d", ret);
243 return ret;
246 EXPORT_SYMBOL(rxrpc_kernel_send_data);
249 * rxrpc_kernel_abort_call - Allow a kernel service to abort a call
250 * @call: The call to be aborted
251 * @abort_code: The abort code to stick into the ABORT packet
253 * Allow a kernel service to abort a call, if it's still in an abortable state.
255 void rxrpc_kernel_abort_call(struct rxrpc_call *call, u32 abort_code)
257 _enter("{%d},%d", call->debug_id, abort_code);
259 lock_sock(&call->socket->sk);
261 _debug("CALL %d USR %lx ST %d on CONN %p",
262 call->debug_id, call->user_call_ID, call->state, call->conn);
264 if (call->state < RXRPC_CALL_COMPLETE)
265 rxrpc_send_abort(call, abort_code);
267 release_sock(&call->socket->sk);
268 _leave("");
271 EXPORT_SYMBOL(rxrpc_kernel_abort_call);
274 * send a message through a server socket
275 * - caller holds the socket locked
277 int rxrpc_server_sendmsg(struct kiocb *iocb, struct rxrpc_sock *rx,
278 struct msghdr *msg, size_t len)
280 enum rxrpc_command cmd;
281 struct rxrpc_call *call;
282 unsigned long user_call_ID = 0;
283 u32 abort_code = 0;
284 int ret;
286 _enter("");
288 ret = rxrpc_sendmsg_cmsg(rx, msg, &user_call_ID, &cmd, &abort_code,
289 true);
290 if (ret < 0)
291 return ret;
293 if (cmd == RXRPC_CMD_ACCEPT) {
294 call = rxrpc_accept_call(rx, user_call_ID);
295 if (IS_ERR(call))
296 return PTR_ERR(call);
297 rxrpc_put_call(call);
298 return 0;
301 call = rxrpc_find_server_call(rx, user_call_ID);
302 if (!call)
303 return -EBADSLT;
304 if (call->state >= RXRPC_CALL_COMPLETE) {
305 ret = -ESHUTDOWN;
306 goto out;
309 switch (cmd) {
310 case RXRPC_CMD_SEND_DATA:
311 if (call->state != RXRPC_CALL_CLIENT_SEND_REQUEST &&
312 call->state != RXRPC_CALL_SERVER_ACK_REQUEST &&
313 call->state != RXRPC_CALL_SERVER_SEND_REPLY) {
314 /* Tx phase not yet begun for this call */
315 ret = -EPROTO;
316 break;
319 ret = rxrpc_send_data(iocb, rx, call, msg, len);
320 break;
322 case RXRPC_CMD_SEND_ABORT:
323 rxrpc_send_abort(call, abort_code);
324 break;
325 default:
326 BUG();
329 out:
330 rxrpc_put_call(call);
331 _leave(" = %d", ret);
332 return ret;
336 * send a packet through the transport endpoint
338 int rxrpc_send_packet(struct rxrpc_transport *trans, struct sk_buff *skb)
340 struct kvec iov[1];
341 struct msghdr msg;
342 int ret, opt;
344 _enter(",{%d}", skb->len);
346 iov[0].iov_base = skb->head;
347 iov[0].iov_len = skb->len;
349 msg.msg_name = &trans->peer->srx.transport.sin;
350 msg.msg_namelen = sizeof(trans->peer->srx.transport.sin);
351 msg.msg_control = NULL;
352 msg.msg_controllen = 0;
353 msg.msg_flags = 0;
355 /* send the packet with the don't fragment bit set if we currently
356 * think it's small enough */
357 if (skb->len - sizeof(struct rxrpc_header) < trans->peer->maxdata) {
358 down_read(&trans->local->defrag_sem);
359 /* send the packet by UDP
360 * - returns -EMSGSIZE if UDP would have to fragment the packet
361 * to go out of the interface
362 * - in which case, we'll have processed the ICMP error
363 * message and update the peer record
365 ret = kernel_sendmsg(trans->local->socket, &msg, iov, 1,
366 iov[0].iov_len);
368 up_read(&trans->local->defrag_sem);
369 if (ret == -EMSGSIZE)
370 goto send_fragmentable;
372 _leave(" = %d [%u]", ret, trans->peer->maxdata);
373 return ret;
376 send_fragmentable:
377 /* attempt to send this message with fragmentation enabled */
378 _debug("send fragment");
380 down_write(&trans->local->defrag_sem);
381 opt = IP_PMTUDISC_DONT;
382 ret = kernel_setsockopt(trans->local->socket, SOL_IP, IP_MTU_DISCOVER,
383 (char *) &opt, sizeof(opt));
384 if (ret == 0) {
385 ret = kernel_sendmsg(trans->local->socket, &msg, iov, 1,
386 iov[0].iov_len);
388 opt = IP_PMTUDISC_DO;
389 kernel_setsockopt(trans->local->socket, SOL_IP,
390 IP_MTU_DISCOVER, (char *) &opt, sizeof(opt));
393 up_write(&trans->local->defrag_sem);
394 _leave(" = %d [frag %u]", ret, trans->peer->maxdata);
395 return ret;
399 * wait for space to appear in the transmit/ACK window
400 * - caller holds the socket locked
402 static int rxrpc_wait_for_tx_window(struct rxrpc_sock *rx,
403 struct rxrpc_call *call,
404 long *timeo)
406 DECLARE_WAITQUEUE(myself, current);
407 int ret;
409 _enter(",{%d},%ld",
410 CIRC_SPACE(call->acks_head, call->acks_tail, call->acks_winsz),
411 *timeo);
413 add_wait_queue(&call->tx_waitq, &myself);
415 for (;;) {
416 set_current_state(TASK_INTERRUPTIBLE);
417 ret = 0;
418 if (CIRC_SPACE(call->acks_head, call->acks_tail,
419 call->acks_winsz) > 0)
420 break;
421 if (signal_pending(current)) {
422 ret = sock_intr_errno(*timeo);
423 break;
426 release_sock(&rx->sk);
427 *timeo = schedule_timeout(*timeo);
428 lock_sock(&rx->sk);
431 remove_wait_queue(&call->tx_waitq, &myself);
432 set_current_state(TASK_RUNNING);
433 _leave(" = %d", ret);
434 return ret;
438 * attempt to schedule an instant Tx resend
440 static inline void rxrpc_instant_resend(struct rxrpc_call *call)
442 read_lock_bh(&call->state_lock);
443 if (try_to_del_timer_sync(&call->resend_timer) >= 0) {
444 clear_bit(RXRPC_CALL_RUN_RTIMER, &call->flags);
445 if (call->state < RXRPC_CALL_COMPLETE &&
446 !test_and_set_bit(RXRPC_CALL_RESEND_TIMER, &call->events))
447 rxrpc_queue_call(call);
449 read_unlock_bh(&call->state_lock);
453 * queue a packet for transmission, set the resend timer and attempt
454 * to send the packet immediately
456 static void rxrpc_queue_packet(struct rxrpc_call *call, struct sk_buff *skb,
457 bool last)
459 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
460 int ret;
462 _net("queue skb %p [%d]", skb, call->acks_head);
464 ASSERT(call->acks_window != NULL);
465 call->acks_window[call->acks_head] = (unsigned long) skb;
466 smp_wmb();
467 call->acks_head = (call->acks_head + 1) & (call->acks_winsz - 1);
469 if (last || call->state == RXRPC_CALL_SERVER_ACK_REQUEST) {
470 _debug("________awaiting reply/ACK__________");
471 write_lock_bh(&call->state_lock);
472 switch (call->state) {
473 case RXRPC_CALL_CLIENT_SEND_REQUEST:
474 call->state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
475 break;
476 case RXRPC_CALL_SERVER_ACK_REQUEST:
477 call->state = RXRPC_CALL_SERVER_SEND_REPLY;
478 if (!last)
479 break;
480 case RXRPC_CALL_SERVER_SEND_REPLY:
481 call->state = RXRPC_CALL_SERVER_AWAIT_ACK;
482 break;
483 default:
484 break;
486 write_unlock_bh(&call->state_lock);
489 _proto("Tx DATA %%%u { #%u }",
490 ntohl(sp->hdr.serial), ntohl(sp->hdr.seq));
492 sp->need_resend = false;
493 sp->resend_at = jiffies + rxrpc_resend_timeout;
494 if (!test_and_set_bit(RXRPC_CALL_RUN_RTIMER, &call->flags)) {
495 _debug("run timer");
496 call->resend_timer.expires = sp->resend_at;
497 add_timer(&call->resend_timer);
500 /* attempt to cancel the rx-ACK timer, deferring reply transmission if
501 * we're ACK'ing the request phase of an incoming call */
502 ret = -EAGAIN;
503 if (try_to_del_timer_sync(&call->ack_timer) >= 0) {
504 /* the packet may be freed by rxrpc_process_call() before this
505 * returns */
506 ret = rxrpc_send_packet(call->conn->trans, skb);
507 _net("sent skb %p", skb);
508 } else {
509 _debug("failed to delete ACK timer");
512 if (ret < 0) {
513 _debug("need instant resend %d", ret);
514 sp->need_resend = true;
515 rxrpc_instant_resend(call);
518 _leave("");
522 * send data through a socket
523 * - must be called in process context
524 * - caller holds the socket locked
526 static int rxrpc_send_data(struct kiocb *iocb,
527 struct rxrpc_sock *rx,
528 struct rxrpc_call *call,
529 struct msghdr *msg, size_t len)
531 struct rxrpc_skb_priv *sp;
532 unsigned char __user *from;
533 struct sk_buff *skb;
534 const struct iovec *iov;
535 struct sock *sk = &rx->sk;
536 long timeo;
537 bool more;
538 int ret, ioc, segment, copied;
540 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
542 /* this should be in poll */
543 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
545 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
546 return -EPIPE;
548 iov = msg->msg_iter.iov;
549 ioc = msg->msg_iter.nr_segs - 1;
550 from = iov->iov_base;
551 segment = iov->iov_len;
552 iov++;
553 more = msg->msg_flags & MSG_MORE;
555 skb = call->tx_pending;
556 call->tx_pending = NULL;
558 copied = 0;
559 do {
560 int copy;
562 if (segment > len)
563 segment = len;
565 _debug("SEGMENT %d @%p", segment, from);
567 if (!skb) {
568 size_t size, chunk, max, space;
570 _debug("alloc");
572 if (CIRC_SPACE(call->acks_head, call->acks_tail,
573 call->acks_winsz) <= 0) {
574 ret = -EAGAIN;
575 if (msg->msg_flags & MSG_DONTWAIT)
576 goto maybe_error;
577 ret = rxrpc_wait_for_tx_window(rx, call,
578 &timeo);
579 if (ret < 0)
580 goto maybe_error;
583 max = call->conn->trans->peer->maxdata;
584 max -= call->conn->security_size;
585 max &= ~(call->conn->size_align - 1UL);
587 chunk = max;
588 if (chunk > len && !more)
589 chunk = len;
591 space = chunk + call->conn->size_align;
592 space &= ~(call->conn->size_align - 1UL);
594 size = space + call->conn->header_size;
596 _debug("SIZE: %zu/%zu/%zu", chunk, space, size);
598 /* create a buffer that we can retain until it's ACK'd */
599 skb = sock_alloc_send_skb(
600 sk, size, msg->msg_flags & MSG_DONTWAIT, &ret);
601 if (!skb)
602 goto maybe_error;
604 rxrpc_new_skb(skb);
606 _debug("ALLOC SEND %p", skb);
608 ASSERTCMP(skb->mark, ==, 0);
610 _debug("HS: %u", call->conn->header_size);
611 skb_reserve(skb, call->conn->header_size);
612 skb->len += call->conn->header_size;
614 sp = rxrpc_skb(skb);
615 sp->remain = chunk;
616 if (sp->remain > skb_tailroom(skb))
617 sp->remain = skb_tailroom(skb);
619 _net("skb: hr %d, tr %d, hl %d, rm %d",
620 skb_headroom(skb),
621 skb_tailroom(skb),
622 skb_headlen(skb),
623 sp->remain);
625 skb->ip_summed = CHECKSUM_UNNECESSARY;
628 _debug("append");
629 sp = rxrpc_skb(skb);
631 /* append next segment of data to the current buffer */
632 copy = skb_tailroom(skb);
633 ASSERTCMP(copy, >, 0);
634 if (copy > segment)
635 copy = segment;
636 if (copy > sp->remain)
637 copy = sp->remain;
639 _debug("add");
640 ret = skb_add_data(skb, from, copy);
641 _debug("added");
642 if (ret < 0)
643 goto efault;
644 sp->remain -= copy;
645 skb->mark += copy;
646 copied += copy;
648 len -= copy;
649 segment -= copy;
650 from += copy;
651 while (segment == 0 && ioc > 0) {
652 from = iov->iov_base;
653 segment = iov->iov_len;
654 iov++;
655 ioc--;
657 if (len == 0) {
658 segment = 0;
659 ioc = 0;
662 /* check for the far side aborting the call or a network error
663 * occurring */
664 if (call->state > RXRPC_CALL_COMPLETE)
665 goto call_aborted;
667 /* add the packet to the send queue if it's now full */
668 if (sp->remain <= 0 || (segment == 0 && !more)) {
669 struct rxrpc_connection *conn = call->conn;
670 uint32_t seq;
671 size_t pad;
673 /* pad out if we're using security */
674 if (conn->security) {
675 pad = conn->security_size + skb->mark;
676 pad = conn->size_align - pad;
677 pad &= conn->size_align - 1;
678 _debug("pad %zu", pad);
679 if (pad)
680 memset(skb_put(skb, pad), 0, pad);
683 seq = atomic_inc_return(&call->sequence);
685 sp->hdr.epoch = conn->epoch;
686 sp->hdr.cid = call->cid;
687 sp->hdr.callNumber = call->call_id;
688 sp->hdr.seq = htonl(seq);
689 sp->hdr.serial =
690 htonl(atomic_inc_return(&conn->serial));
691 sp->hdr.type = RXRPC_PACKET_TYPE_DATA;
692 sp->hdr.userStatus = 0;
693 sp->hdr.securityIndex = conn->security_ix;
694 sp->hdr._rsvd = 0;
695 sp->hdr.serviceId = conn->service_id;
697 sp->hdr.flags = conn->out_clientflag;
698 if (len == 0 && !more)
699 sp->hdr.flags |= RXRPC_LAST_PACKET;
700 else if (CIRC_SPACE(call->acks_head, call->acks_tail,
701 call->acks_winsz) > 1)
702 sp->hdr.flags |= RXRPC_MORE_PACKETS;
703 if (more && seq & 1)
704 sp->hdr.flags |= RXRPC_REQUEST_ACK;
706 ret = rxrpc_secure_packet(
707 call, skb, skb->mark,
708 skb->head + sizeof(struct rxrpc_header));
709 if (ret < 0)
710 goto out;
712 memcpy(skb->head, &sp->hdr,
713 sizeof(struct rxrpc_header));
714 rxrpc_queue_packet(call, skb, segment == 0 && !more);
715 skb = NULL;
718 } while (segment > 0);
720 success:
721 ret = copied;
722 out:
723 call->tx_pending = skb;
724 _leave(" = %d", ret);
725 return ret;
727 call_aborted:
728 rxrpc_free_skb(skb);
729 if (call->state == RXRPC_CALL_NETWORK_ERROR)
730 ret = call->conn->trans->peer->net_error;
731 else
732 ret = -ECONNABORTED;
733 _leave(" = %d", ret);
734 return ret;
736 maybe_error:
737 if (copied)
738 goto success;
739 goto out;
741 efault:
742 ret = -EFAULT;
743 goto out;