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 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,
43 *command = RXRPC_CMD_SEND_DATA;
45 if (msg->msg_controllen == 0)
48 for_each_cmsghdr(cmsg, msg) {
49 if (!CMSG_OK(msg, cmsg))
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)
59 switch (cmsg->cmsg_type) {
60 case RXRPC_USER_CALL_ID:
61 if (msg->msg_flags & MSG_CMSG_COMPAT) {
62 if (len != sizeof(u32))
64 *user_call_ID = *(u32 *) CMSG_DATA(cmsg);
66 if (len != sizeof(unsigned long))
68 *user_call_ID = *(unsigned long *)
71 _debug("User Call ID %lx", *user_call_ID);
75 if (*command != RXRPC_CMD_SEND_DATA)
77 *command = RXRPC_CMD_SEND_ABORT;
78 if (len != sizeof(*abort_code))
80 *abort_code = *(unsigned int *) CMSG_DATA(cmsg);
81 _debug("Abort %x", *abort_code);
87 if (*command != RXRPC_CMD_SEND_DATA)
89 *command = RXRPC_CMD_ACCEPT;
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,
136 struct rxrpc_conn_bundle *bundle;
137 enum rxrpc_command cmd;
138 struct rxrpc_call *call;
139 unsigned long user_call_ID = 0;
147 ASSERT(trans != NULL);
149 ret = rxrpc_sendmsg_cmsg(rx, msg, &user_call_ID, &cmd, &abort_code,
156 service_id = rx->service_id;
158 DECLARE_SOCKADDR(struct sockaddr_rxrpc *, srx,
160 service_id = htons(srx->srx_service);
163 if (key && !rx->key->payload.data)
165 bundle = rxrpc_get_bundle(rx, trans, key, service_id,
168 return PTR_ERR(bundle);
171 call = rxrpc_get_client_call(rx, trans, bundle, user_call_ID,
172 abort_code == 0, GFP_KERNEL);
174 rxrpc_put_bundle(trans, bundle);
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 */
186 } else if (cmd == RXRPC_CMD_SEND_ABORT) {
187 rxrpc_send_abort(call, abort_code);
188 } else if (cmd != RXRPC_CMD_SEND_DATA) {
190 } else if (call->state != RXRPC_CALL_CLIENT_SEND_REQUEST) {
191 /* request phase complete for this client call */
194 ret = rxrpc_send_data(iocb, rx, call, msg, len);
197 rxrpc_put_call(call);
198 _leave(" = %d", 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,
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 */
235 ret = rxrpc_send_data(NULL, call->socket, call, msg, len);
238 release_sock(&call->socket->sk);
239 _leave(" = %d", ret);
243 EXPORT_SYMBOL(rxrpc_kernel_send_data);
246 * rxrpc_kernel_abort_call - Allow a kernel service to abort a call
247 * @call: The call to be aborted
248 * @abort_code: The abort code to stick into the ABORT packet
250 * Allow a kernel service to abort a call, if it's still in an abortable state.
252 void rxrpc_kernel_abort_call(struct rxrpc_call *call, u32 abort_code)
254 _enter("{%d},%d", call->debug_id, abort_code);
256 lock_sock(&call->socket->sk);
258 _debug("CALL %d USR %lx ST %d on CONN %p",
259 call->debug_id, call->user_call_ID, call->state, call->conn);
261 if (call->state < RXRPC_CALL_COMPLETE)
262 rxrpc_send_abort(call, abort_code);
264 release_sock(&call->socket->sk);
268 EXPORT_SYMBOL(rxrpc_kernel_abort_call);
271 * send a message through a server socket
272 * - caller holds the socket locked
274 int rxrpc_server_sendmsg(struct kiocb *iocb, struct rxrpc_sock *rx,
275 struct msghdr *msg, size_t len)
277 enum rxrpc_command cmd;
278 struct rxrpc_call *call;
279 unsigned long user_call_ID = 0;
285 ret = rxrpc_sendmsg_cmsg(rx, msg, &user_call_ID, &cmd, &abort_code,
290 if (cmd == RXRPC_CMD_ACCEPT) {
291 call = rxrpc_accept_call(rx, user_call_ID);
293 return PTR_ERR(call);
294 rxrpc_put_call(call);
298 call = rxrpc_find_server_call(rx, user_call_ID);
301 if (call->state >= RXRPC_CALL_COMPLETE) {
307 case RXRPC_CMD_SEND_DATA:
308 if (call->state != RXRPC_CALL_CLIENT_SEND_REQUEST &&
309 call->state != RXRPC_CALL_SERVER_ACK_REQUEST &&
310 call->state != RXRPC_CALL_SERVER_SEND_REPLY) {
311 /* Tx phase not yet begun for this call */
316 ret = rxrpc_send_data(iocb, rx, call, msg, len);
319 case RXRPC_CMD_SEND_ABORT:
320 rxrpc_send_abort(call, abort_code);
327 rxrpc_put_call(call);
328 _leave(" = %d", ret);
333 * send a packet through the transport endpoint
335 int rxrpc_send_packet(struct rxrpc_transport *trans, struct sk_buff *skb)
341 _enter(",{%d}", skb->len);
343 iov[0].iov_base = skb->head;
344 iov[0].iov_len = skb->len;
346 msg.msg_name = &trans->peer->srx.transport.sin;
347 msg.msg_namelen = sizeof(trans->peer->srx.transport.sin);
348 msg.msg_control = NULL;
349 msg.msg_controllen = 0;
352 /* send the packet with the don't fragment bit set if we currently
353 * think it's small enough */
354 if (skb->len - sizeof(struct rxrpc_header) < trans->peer->maxdata) {
355 down_read(&trans->local->defrag_sem);
356 /* send the packet by UDP
357 * - returns -EMSGSIZE if UDP would have to fragment the packet
358 * to go out of the interface
359 * - in which case, we'll have processed the ICMP error
360 * message and update the peer record
362 ret = kernel_sendmsg(trans->local->socket, &msg, iov, 1,
365 up_read(&trans->local->defrag_sem);
366 if (ret == -EMSGSIZE)
367 goto send_fragmentable;
369 _leave(" = %d [%u]", ret, trans->peer->maxdata);
374 /* attempt to send this message with fragmentation enabled */
375 _debug("send fragment");
377 down_write(&trans->local->defrag_sem);
378 opt = IP_PMTUDISC_DONT;
379 ret = kernel_setsockopt(trans->local->socket, SOL_IP, IP_MTU_DISCOVER,
380 (char *) &opt, sizeof(opt));
382 ret = kernel_sendmsg(trans->local->socket, &msg, iov, 1,
385 opt = IP_PMTUDISC_DO;
386 kernel_setsockopt(trans->local->socket, SOL_IP,
387 IP_MTU_DISCOVER, (char *) &opt, sizeof(opt));
390 up_write(&trans->local->defrag_sem);
391 _leave(" = %d [frag %u]", ret, trans->peer->maxdata);
396 * wait for space to appear in the transmit/ACK window
397 * - caller holds the socket locked
399 static int rxrpc_wait_for_tx_window(struct rxrpc_sock *rx,
400 struct rxrpc_call *call,
403 DECLARE_WAITQUEUE(myself, current);
407 CIRC_SPACE(call->acks_head, call->acks_tail, call->acks_winsz),
410 add_wait_queue(&call->tx_waitq, &myself);
413 set_current_state(TASK_INTERRUPTIBLE);
415 if (CIRC_SPACE(call->acks_head, call->acks_tail,
416 call->acks_winsz) > 0)
418 if (signal_pending(current)) {
419 ret = sock_intr_errno(*timeo);
423 release_sock(&rx->sk);
424 *timeo = schedule_timeout(*timeo);
428 remove_wait_queue(&call->tx_waitq, &myself);
429 set_current_state(TASK_RUNNING);
430 _leave(" = %d", ret);
435 * attempt to schedule an instant Tx resend
437 static inline void rxrpc_instant_resend(struct rxrpc_call *call)
439 read_lock_bh(&call->state_lock);
440 if (try_to_del_timer_sync(&call->resend_timer) >= 0) {
441 clear_bit(RXRPC_CALL_RUN_RTIMER, &call->flags);
442 if (call->state < RXRPC_CALL_COMPLETE &&
443 !test_and_set_bit(RXRPC_CALL_RESEND_TIMER, &call->events))
444 rxrpc_queue_call(call);
446 read_unlock_bh(&call->state_lock);
450 * queue a packet for transmission, set the resend timer and attempt
451 * to send the packet immediately
453 static void rxrpc_queue_packet(struct rxrpc_call *call, struct sk_buff *skb,
456 struct rxrpc_skb_priv *sp = rxrpc_skb(skb);
459 _net("queue skb %p [%d]", skb, call->acks_head);
461 ASSERT(call->acks_window != NULL);
462 call->acks_window[call->acks_head] = (unsigned long) skb;
464 call->acks_head = (call->acks_head + 1) & (call->acks_winsz - 1);
466 if (last || call->state == RXRPC_CALL_SERVER_ACK_REQUEST) {
467 _debug("________awaiting reply/ACK__________");
468 write_lock_bh(&call->state_lock);
469 switch (call->state) {
470 case RXRPC_CALL_CLIENT_SEND_REQUEST:
471 call->state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
473 case RXRPC_CALL_SERVER_ACK_REQUEST:
474 call->state = RXRPC_CALL_SERVER_SEND_REPLY;
477 case RXRPC_CALL_SERVER_SEND_REPLY:
478 call->state = RXRPC_CALL_SERVER_AWAIT_ACK;
483 write_unlock_bh(&call->state_lock);
486 _proto("Tx DATA %%%u { #%u }",
487 ntohl(sp->hdr.serial), ntohl(sp->hdr.seq));
489 sp->need_resend = false;
490 sp->resend_at = jiffies + rxrpc_resend_timeout;
491 if (!test_and_set_bit(RXRPC_CALL_RUN_RTIMER, &call->flags)) {
493 call->resend_timer.expires = sp->resend_at;
494 add_timer(&call->resend_timer);
497 /* attempt to cancel the rx-ACK timer, deferring reply transmission if
498 * we're ACK'ing the request phase of an incoming call */
500 if (try_to_del_timer_sync(&call->ack_timer) >= 0) {
501 /* the packet may be freed by rxrpc_process_call() before this
503 ret = rxrpc_send_packet(call->conn->trans, skb);
504 _net("sent skb %p", skb);
506 _debug("failed to delete ACK timer");
510 _debug("need instant resend %d", ret);
511 sp->need_resend = true;
512 rxrpc_instant_resend(call);
519 * send data through a socket
520 * - must be called in process context
521 * - caller holds the socket locked
523 static int rxrpc_send_data(struct kiocb *iocb,
524 struct rxrpc_sock *rx,
525 struct rxrpc_call *call,
526 struct msghdr *msg, size_t len)
528 struct rxrpc_skb_priv *sp;
530 struct sock *sk = &rx->sk;
535 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
537 /* this should be in poll */
538 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
540 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
543 more = msg->msg_flags & MSG_MORE;
545 skb = call->tx_pending;
546 call->tx_pending = NULL;
549 if (len > iov_iter_count(&msg->msg_iter))
550 len = iov_iter_count(&msg->msg_iter);
555 size_t size, chunk, max, space;
559 if (CIRC_SPACE(call->acks_head, call->acks_tail,
560 call->acks_winsz) <= 0) {
562 if (msg->msg_flags & MSG_DONTWAIT)
564 ret = rxrpc_wait_for_tx_window(rx, call,
570 max = call->conn->trans->peer->maxdata;
571 max -= call->conn->security_size;
572 max &= ~(call->conn->size_align - 1UL);
575 if (chunk > len && !more)
578 space = chunk + call->conn->size_align;
579 space &= ~(call->conn->size_align - 1UL);
581 size = space + call->conn->header_size;
583 _debug("SIZE: %zu/%zu/%zu", chunk, space, size);
585 /* create a buffer that we can retain until it's ACK'd */
586 skb = sock_alloc_send_skb(
587 sk, size, msg->msg_flags & MSG_DONTWAIT, &ret);
593 _debug("ALLOC SEND %p", skb);
595 ASSERTCMP(skb->mark, ==, 0);
597 _debug("HS: %u", call->conn->header_size);
598 skb_reserve(skb, call->conn->header_size);
599 skb->len += call->conn->header_size;
603 if (sp->remain > skb_tailroom(skb))
604 sp->remain = skb_tailroom(skb);
606 _net("skb: hr %d, tr %d, hl %d, rm %d",
612 skb->ip_summed = CHECKSUM_UNNECESSARY;
618 /* append next segment of data to the current buffer */
619 copy = skb_tailroom(skb);
620 ASSERTCMP(copy, >, 0);
623 if (copy > sp->remain)
627 ret = skb_add_data(skb, &msg->msg_iter, copy);
637 /* check for the far side aborting the call or a network error
639 if (call->state > RXRPC_CALL_COMPLETE)
642 /* add the packet to the send queue if it's now full */
643 if (sp->remain <= 0 || (!len && !more)) {
644 struct rxrpc_connection *conn = call->conn;
648 /* pad out if we're using security */
649 if (conn->security) {
650 pad = conn->security_size + skb->mark;
651 pad = conn->size_align - pad;
652 pad &= conn->size_align - 1;
653 _debug("pad %zu", pad);
655 memset(skb_put(skb, pad), 0, pad);
658 seq = atomic_inc_return(&call->sequence);
660 sp->hdr.epoch = conn->epoch;
661 sp->hdr.cid = call->cid;
662 sp->hdr.callNumber = call->call_id;
663 sp->hdr.seq = htonl(seq);
665 htonl(atomic_inc_return(&conn->serial));
666 sp->hdr.type = RXRPC_PACKET_TYPE_DATA;
667 sp->hdr.userStatus = 0;
668 sp->hdr.securityIndex = conn->security_ix;
670 sp->hdr.serviceId = conn->service_id;
672 sp->hdr.flags = conn->out_clientflag;
673 if (len == 0 && !more)
674 sp->hdr.flags |= RXRPC_LAST_PACKET;
675 else if (CIRC_SPACE(call->acks_head, call->acks_tail,
676 call->acks_winsz) > 1)
677 sp->hdr.flags |= RXRPC_MORE_PACKETS;
679 sp->hdr.flags |= RXRPC_REQUEST_ACK;
681 ret = rxrpc_secure_packet(
682 call, skb, skb->mark,
683 skb->head + sizeof(struct rxrpc_header));
687 memcpy(skb->head, &sp->hdr,
688 sizeof(struct rxrpc_header));
689 rxrpc_queue_packet(call, skb, !iov_iter_count(&msg->msg_iter) && !more);
697 call->tx_pending = skb;
698 _leave(" = %d", ret);
703 if (call->state == RXRPC_CALL_NETWORK_ERROR)
704 ret = call->conn->trans->peer->net_error;
707 _leave(" = %d", ret);